ICES Journal of Marine Science: Journal du Conseil

doi: 10.1093/icesjms/fsy060pmid: N/A

Subbey, S. Parameter estimation in stock assessment modelling: caveats with gradient-based algorithms. – ICES Journal of Marine Science, doi:10.1093/icesjms/fsy044. ICES Journal of Marine Science (2018), doi:10.1093/icesjms/fsy044 The author of the above paper wishes to notify readers that this paper published with typographical errors on the second and third pages. In the section ‘Mathematical and statistical models’ the sentence ‘The collection of the concepts and language define what is referred to a a mathematical/statistical model’, has been corrected to amend ‘a a’ to ‘as a’. In the section ‘Gradient descent algorithms’ the sentence ‘This is because the steepness from A to C is bigger than from A to C.’, has been corrected to amend the second ‘from A to C’ to ‘from A to B’. The paper has now been corrected online. © International Council for the Exploration of the Sea 2018. All rights reserved. For permissions, please email: [email protected] This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

doi: 10.1093/icesjms/fsy015pmid: N/A

Abstract Current sea louse models attempt to estimate louse burdens on wild and cultured salmon by predicting the production and distribution of lice larvae and estimating the risk of transmission. While physical characteristics of water bodies and weather can be accurately modelled, many aspects of sea lice biology require further parameterization. The aims of this review are (i) to describe current knowledge regarding the production, mortality, and infectivity of planktonic sea lice larvae and (ii) to identify gaps in knowledge and suggest research approaches to filling them. Several major gaps are identified, and those likely to have the greatest impact on infection levels are (i) egg production, viability and hatching success, (ii) predation in plankton and (iii) copepodid infectivity profiles. A key problem identified in current parameter estimates is that they originate from a number of sources and have been determined using a variety of experimental approaches. This is a barrier to the provision of “best” or consensus estimates for use in modelling. Additional and more consistent data collection and experimentation will help to fill these gaps. Furthermore, coordinated international efforts are required to generate a more complete picture of sea louse infections across all regions experiencing problems with sea lice. Introduction The parasitic copepods known as sea lice remain a key constraint to the continued growth of salmonid aquaculture industries worldwide. In the North Atlantic, Lepeophtheirus salmonis salmonis (Krøyer, 1837) is the primary species infecting cultured Atlantic salmon (Salmo salar L.), whereas in the North Pacific, Lepeophtheirus salmonis oncorhynchii (Johnson and Albright, 1991a) is prevalent in cultured salmon, although Caligus elongatus von Nordmann, 1832 also has some impact. In the southern hemisphere Caligus rogercresseyi Boxshall and Bravo 2000 is the principal pathogenic species affecting the Chilean salmon aquaculture industry. For the Norwegian salmon industry, where costs are best characterized, the economic impact of sea lice was estimated to be in excess of 3.4 billion NOK per annum (>£300M) in 2014 for 1 272 358 tonnes production (Iversen et al., 2015) with costs estimated to exceed 5 billion NOK (>£390M) in 2015 for 1 303 346 tonnes = 3836.28 NOK tonne−1 (Audun Iversen, pers. comm.). Higher estimates of 7–8 billion NOK per annum (>£540M) have also been presented (Rødseth, 2016). Using FAO statistics for global cultured Atlantic salmon production in 2015 (http://www.fao.org/fishery/statistics/global-aquaculture-production/en) for all countries that experience sea lice problems (2 332 290 tonnes) and Iversen’s estimate of cost per tonne for 2015 (3836.28 × 2 332 290) provides a rough cost estimate of ∼9 billion NOK globally for 2015 (∼£700M), with costs likely to have continued to rise since then. Current integrated pest management (IPM) strategies for sea lice control rely on a small number of licensed pesticides, few of which are effective against all stages of the parasite’s life cycle, combined with effective husbandry management tools, such as single-cohort stocking, optimized stocking densities, the use of cleaner fish in polyculture, and fallow periods (Leclercq et al., 2013; Skiftesvik et al., 2013). Physical techniques to exclude lice, such as the use of barrier nets and snorkel cages, coupled with mechanical tools, including thermal and turbulent de-licers and laser removal, also constitute an increasing component of current IPM strategies. The adoption of such an increasingly multimodal approach means that the timing of management decisions is critical to the successful control of the parasite. A central element required for the prediction of fluxes in lice populations is an understanding of the production, survival, dispersal, development, and infectivity profile of the free-swimming non-infective nauplii and infective copepodid larval stages. However, despite more than three decades of research, knowledge in this area remains extremely poor. Within the past 10 years, several models have been developed that attempt to estimate lice burdens on wild and cultured salmon by predicting the production and distribution of lice larvae from salmon farms and the subsequent risk of transmission. Although complex physical coastal processes can now be reasonably accurately modelled, aspects of larval behaviour and mortality often appear oversimplified. This knowledge gap has serious consequences as it confounds the realistic estimation of the number of lice capable of infecting wild and cultured salmonid populations. In ecological terms, sea lice can be considered r-strategists, which are characterized by small body sizes, high fecundities, and short generation times. Although offspring of r-strategists are dispersed widely, they have a low probability of survival (Cavaleiro and Santos, 2014). However, sea lice differ from many other r-strategists in that they are attached to a host, which provides a permanent food source and allows anomalies, such as a larger body size, and raises the question of whether they have a high fecundity because they experience heavy losses during the larval stages or because they have a nominally unlimited food source. The high fecundity and wide larval dispersal are key aspects of the sea louse’s life cycle that determine its overall survival and success. As a result, fecundity and larval biology should be the focus of efforts to predict lice burdens on fish. In the life cycle of the sea louse, however, the free-swimming stages are essentially a “black box” that cannot be easily observed directly from field studies. Once a copepodid has attached to a host, development is more predictable as development after infection is unaffected by copepodid age at infection (Tucker et al., 2000a; Pedersen, 2009), although at this point host factors such as host species/genotype, immunity, and site of infection intervene to affect success. Transmission is still a contentious issue with disagreement over whether lice (despite water currents) are accumulated at their source (e.g. Krkošek et al., 2005 and implied by Jansen et al., 2012) or hydrodynamically spread over large distances (e.g. Brooks, 2005; Asplin et al., 2014). Therefore, accurate data are urgently needed to inform and validate increasingly realistic models of larval dispersion and infectivity that combine physical processes with key aspects of lice biology to successfully predict larval dispersion and infection risk. Early models for predicting lice burdens rely on the relationship between gravid female lice and infective larval stages, based on factors such as fecundity, mortality, and moult timings, to predict future cohorts of lice available to infect fish (e.g. Heuch and Mo, 2001; Murray, 2002; Tucker et al., 2002). Although these models can predict louse numbers within a simple closed system, they cannot be applied to large, open systems, such as fjordic sea lochs where salmon are commonly farmed, as they do not take into account larval dispersion and exogenous sources of mortality, such as predation. Particle tracking models predict the dispersal over time of particles generated at a point source using hydrodynamic models (e.g. Corner et al., 2006), which calculate local current velocities based on local topography, fluid dynamics, and external forcing from tidal elevation, freshwater inputs, and wind-generated currents. Early attempts to predict the dispersal of sea lice larvae using a particle tracking model were made by Asplin et al. (2004), who estimated the dispersal of lice from a salmon farm in Sognefjord, Norway. Detailed currents, hydrography, and wind forcing are calculated using high-resolution, three-dimensional ocean and atmospheric models, and although a temperature-dependant larval growth model is included, there is no estimation of larval mortality or behaviour. It assumes that lice are immortal with passive behaviour, and consequently, the dispersal of lice is overestimated with larvae being spread over a distance of 100 km in just a few days (Asplin et al., 2004). In order to accurately estimate infection risk, it is clear that certain aspects of louse biology, such as survival, mortality, and development times, need to be incorporated into these types of models, and more recent models have attempted to do this. Murray and Amundrud (2007) and Amundrud and Murray (2009) present a coupled biophysical and particle tracking model of Loch Torridon, Scotland that incorporates development times as a function of temperature and a fixed mortality rate based on laboratory observations. More recent models have become increasingly complex, and Asplin et al. (2011, 2014) present a model of a Norwegian fjord comprising a number of sub-models: a coastal ocean model, an atmospheric model, a fjord model, and a salmon louse growth and advection model. While the salmon louse sub-model includes relevant parameters regarding stage timings, it only includes a few simple behavioural parameters, i.e. a diel vertical migration, limited to depths above 10 m and avoidance of salinities below 20‰; however, it does not calculate louse mortality. A further model by Stucchi et al. (2011), which models the hydrographically complex Broughton Archipelago in British Columbia, Canada, includes a comprehensive sub-model of egg production, larval development, mortality, and behaviour using data from the literature, including the effects of temperature and salinity on these parameters. In addition, a recent model similar to the one utilized by Asplin et al. (2014), which uses a mortality rate of 17%, predicts that larval behaviour potentially has significant effects on advection (Johnsen et al., 2014). Aldrin et al. (2013) and Kristoffersen et al. (2014) present a model based on a statistical network of Norwegian salmon farms. Monthly external and internal infection pressure and the risk of infection between neighbouring farms are predicted based on lice burden estimates from the previous month and the distances between neighbouring farms. The model is fitted to actual lice counts from Norwegian farms between 2003 and 2011. It uses temperature-dependent fecundity and larval demographics, although mortality rates for free-swimming larvae and chalimus stages are fixed. While these models have made significant progress in predicting larval dispersal in semi-enclosed water bodies, model validation with field data is difficult, and there are always discrepancies between the model output and field observations. For example, Salama et al. (2011) and Adams et al. (2012) found very few larval sea lice in plankton tows in areas where models had predicted high numbers. However, correlation between predicted and observed infections appear to be more accurate for the model developed by Sandvik and colleagues (Sandvik et al., 2014). Model variables are based on the best available data, and while accurate topography and hydrography data can easily be obtained, detailed information regarding the life history of sea lice is often lacking, despite over three decades of research in this area. Where models incorporate larval mortality, for instance, they use a constant mortality at each larval stage, which may be kept constant (e.g. Aldrin et al., 2013; Johnsen et al., 2014) or vary according to salinity (e.g. Amundrud and Murray, 2009; Adams et al., 2012). In reality, however, larval mortality is extremely variable according to temperature, salinity, season, moult stage, and predation in the plankton, etc. While some data are available regarding these different parameters, others are distinctly lacking, and more research is required in these areas. Acquiring experimental data on these variables will allow the more realistic parameterization of key elements relating to abundance and infectivity of free-swimming larval sea louse stages for incorporation into models that may more accurately predict the risk of infection under various environmental conditions. Some models are now considered sufficiently developed to warrant their use as components of an integrated sea louse management strategy. For example, Norwegian salmon farming from 2017 will be regulated regionally through an operational management system comprising the application of predictive models that predict louse infection intensities along the entire coastline (Asplin et al., 2014), combined with a process of continuous model validation and calibration against real-world data (Bjørn et al., 2014; Sandvik et al., 2016). The aims of this review and analysis were as follows: To analyse the available literature to determine current knowledge regarding the recruitment and survival of free-swimming nauplii and copepodid larvae and factors that affect the longevity and infectivity of copepodids. Where no specific data regarding sea lice were available, the wider literature was consulted, e.g. predator and prey selection in plankton, to inform questions regarding the fate of sea lice larvae in the ocean. To assess the remaining knowledge gaps that might be filled by experimental or field sampling studies. Additional considerations: While this review focuses primarily on Lepeophtheirus salmonis spp., observations from other species that are problematic in salmonid aquaculture are also noted where appropriate. This review also focuses principally on knowledge concerning louse larvae deriving from farmed fish due to both their greater accessibility and the fact that environmental parameters can only be sufficiently controlled or consistently measured in defined water masses. Hitherto, there has been some conflation of data arising from Atlantic and Pacific sea louse studies. Evidence for clear genomic and phenotypic differences between these subspecies has made it evident that the origin of data regarding these subspecies should be considered when interpreting the results. Larval recruitment and survival In order to accurately predict when and how many infective copepodids are available for infection, it is necessary to quantify the rate of larval production, which is based on female fecundity, and the subsequent development and survival rates of the larvae. These are influenced by a range of biotic and abiotic factors that fluctuate seasonally and can have an impact on adult lice during mating and egg production, on eggs during development and upon larvae once they have hatched. Fecundity The fecundity of sea lice varies considerably, and early observations showed that a single egg string can contain <100–700 eggs (Wootten et al., 1982). Many studies have shown that exogenous factors, such as temperature, photoperiod, salinity, and food availability, interact with endogenous factors to determine fecundity in crustaceans (e.g. Koop and Field, 1980; Williams, 1985; Johnston and Dykeman, 1987; Maranhão and Marques, 2003). Similarly, variations in the levels of sea lice infection between seasons and under different environmental conditions suggest alterations in reproductive output in response to fluctuating environmental parameters (Ritchie et al., 1993). It is clear that temperature has a strong influence on fecundity (Tully, 1989), and the number of eggs per string is positively correlated with female body size (Tully and Whelan, 1993). Heuch et al. (2000) found that adult female lice of wild origin in Norway were significantly larger than adult female lice of farm origin. Despite seasonal variations, lice of wild origin in Ireland were similarly found to be significantly larger and carried approximately twice as many eggs as lice of farm origin (Tully and Whelan, 1993). A similar pattern was reported by Pike and Wadsworth (1999), who noted that female lice of wild origin produced 965 ± 30.1 eggs per egg string pair compared to 758 ± 39.4 and 297 ± 19.1 for lice originating from untreated and treated farmed salmon, respectively, on the West Coast of Ireland. At 7.2°C, females were observed to produce a new pair of egg strings at an average of 11 days after the first pair were removed, while at 12.2°C this period was reduced to 5 days, and this continued for the reproductive life of the female, with an average of 4.95 pairs of egg strings per female under experimental conditions (Heuch et al., 2000). In this experiment, the first pair of egg strings was always significantly shorter with the mean number of eggs increasing from 152 eggs per string to 285 eggs per string for subsequent egg strings, whereas Johnson and Albright (1991b) recorded a mean number of eggs per string of 344.6 ± 79.8 in lice cultured at 10°C and 30‰ originating from wild and farmed chinook salmon (Oncorhynchus tschawytscha) and farmed Atlantic salmon. Similarly, Gravil (1996) recorded a mean of 141.09 ± 22.19 eggs per string for the first pair of egg strings, 216.4 ± 67.59 eggs per string for the second pair of egg strings, and 208.2 ± 50.97 eggs per string for the third pair of egg strings. It appears that there may be a difference in the batch size in Atlantic L. salmonis salmonis (Heuch et al., 2000; Gravil, 1996) and the Pacific L. salmonis oncorhynchi (Johnson and Albright, 1991a), which highlights the importance of discriminating between the two subspecies (Skern-Mauritzen et al., 2014). Fecundity was found to be lower in C. elongatus with the number of eggs per string being 52.62 ± 17.08 in C. elongatus compared to 206.2 ± 74.09 in L. salmonis at 14°C (Gravil, 1996). Key values for fecundity are shown in Table 1. Table 1. Key values of fecundity in L. salmonis (mean ± SD). L. salmonis salmonis L. salmonis oncorhynchi Time (d) Egg string pairs No. of eggs Time (d) Egg string pairs No. of eggs Egg string production rate  7.2°C 11a – – nd – –  12.2°C 5a – – nd – – Production capacity – 4.95a – – nd – No. egg strings per string  7.2°C – – First string 152 subsequent strings 285a – – nd  10°C – – nd – – 344.6 ± 79.8b  Firt string – – 141.09 ± 22.19c – – nd  Second string – – 216.4 ± 67.59c – – nd  Third string – – 208.2 ± 50.97c – – nd  Wild lice – – 965 ± 30.1d – – nd  Farmed untreated lice – – 758 ± 39.4d – – nd  Farmed treated lice – – 297 ± 19.1d – – nd L. salmonis salmonis L. salmonis oncorhynchi Time (d) Egg string pairs No. of eggs Time (d) Egg string pairs No. of eggs Egg string production rate  7.2°C 11a – – nd – –  12.2°C 5a – – nd – – Production capacity – 4.95a – – nd – No. egg strings per string  7.2°C – – First string 152 subsequent strings 285a – – nd  10°C – – nd – – 344.6 ± 79.8b  Firt string – – 141.09 ± 22.19c – – nd  Second string – – 216.4 ± 67.59c – – nd  Third string – – 208.2 ± 50.97c – – nd  Wild lice – – 965 ± 30.1d – – nd  Farmed untreated lice – – 758 ± 39.4d – – nd  Farmed treated lice – – 297 ± 19.1d – – nd References: (a) Heuch et al. (2000), (b) Johnson and Albright (1991b), (c) Gravil (1996), (d) Pike and Wadsworth (1999), nd = no data available. Table 1. Key values of fecundity in L. salmonis (mean ± SD). L. salmonis salmonis L. salmonis oncorhynchi Time (d) Egg string pairs No. of eggs Time (d) Egg string pairs No. of eggs Egg string production rate  7.2°C 11a – – nd – –  12.2°C 5a – – nd – – Production capacity – 4.95a – – nd – No. egg strings per string  7.2°C – – First string 152 subsequent strings 285a – – nd  10°C – – nd – – 344.6 ± 79.8b  Firt string – – 141.09 ± 22.19c – – nd  Second string – – 216.4 ± 67.59c – – nd  Third string – – 208.2 ± 50.97c – – nd  Wild lice – – 965 ± 30.1d – – nd  Farmed untreated lice – – 758 ± 39.4d – – nd  Farmed treated lice – – 297 ± 19.1d – – nd L. salmonis salmonis L. salmonis oncorhynchi Time (d) Egg string pairs No. of eggs Time (d) Egg string pairs No. of eggs Egg string production rate  7.2°C 11a – – nd – –  12.2°C 5a – – nd – – Production capacity – 4.95a – – nd – No. egg strings per string  7.2°C – – First string 152 subsequent strings 285a – – nd  10°C – – nd – – 344.6 ± 79.8b  Firt string – – 141.09 ± 22.19c – – nd  Second string – – 216.4 ± 67.59c – – nd  Third string – – 208.2 ± 50.97c – – nd  Wild lice – – 965 ± 30.1d – – nd  Farmed untreated lice – – 758 ± 39.4d – – nd  Farmed treated lice – – 297 ± 19.1d – – nd References: (a) Heuch et al. (2000), (b) Johnson and Albright (1991b), (c) Gravil (1996), (d) Pike and Wadsworth (1999), nd = no data available. Ritchie et al. (1993) and Gravil (1996) investigated the reproductive output of L. salmonis from salmon farms on the West Coast of Scotland and found that the number of eggs per string was negatively correlated with temperature, with significantly more eggs being produced in winter and early spring than in summer and autumn (Figure 1). In Ritchie et al. (1993), the mean number of eggs per string increased significantly from 147 to 246 between October and March (temperature range 12–5°C) before decreasing to 175 eggs per string in August (13°C). A similar pattern was seen by Gravil (1996), who found that the number of eggs per string ranged from 194.1 ± 66.8 in October to 286.9 ± 64 in March. There appears to be a period of lag of egg string length in response to temperature as the lowest temperature was recorded in February whereas the longest egg strings were found in March, and this lag may reflect the time required for egg strings to develop before being extruded at low temperatures. Samsing et al. (2016) found a similar trend in lice acclimatized in the laboratory at different temperatures with the number of eggs per string increasing from ∼135 ± 5 at 20°C to ∼295 ± 10 at 5°C. In the same experiment, it was found that the number of eggs per string produced at 3°C was lower (∼153 ± 10) than at the higher temperatures tested. This decrease corresponded to a decreased body size and coincided with a failure in larval development, and it was speculated that this temperature could be close to the limit of their biological tolerance, at least for the tested lice (Samsing et al., 2016). Figure 1. View largeDownload slide Relationship between water temperature and the number of eggs per egg string in Lepeophtheirus salmonis from salmon farms on the West Coast of Scotland. Redrawn from Ritchie et al. (1993). Figure 1. View largeDownload slide Relationship between water temperature and the number of eggs per egg string in Lepeophtheirus salmonis from salmon farms on the West Coast of Scotland. Redrawn from Ritchie et al. (1993). A variety of other factors may also affect lice fecundity. As an example, host condition and the use of chemotherapeutants have been proposed as possible influences on egg string length and the viability of larvae (Tully and Whelan, 1993). Likewise, fecundity may vary with host species, either as a result of diet, the physiological status of the fish, or genetic variation (Johnson and Albright, 1992; MacKinnon et al., 1995; MacKinnon, 1998). This follows from the intimate metabolic associations between hosts and parasites, which are often reflected in the evolution of their genomes (e.g. Zarowiecki and Berriman, 2015). It has also been suggested that host immune responses may modify lice fecundity. For instance, Grayson et al. (1995) found that gravid female lice on Atlantic salmon injected with extracts derived from adult L. salmonis had a significantly lower fecundity than control fish. Similarly, Nilsen (2016) has presented work suggesting that use of a recombinant vaccine to the salmon louse Ls4D8 protein, a homologue to subolesin in ticks and my32 in C. rogercresseyi, gave rise to reduction in egg strings. Host-related and abiotic conditions may not be the only factors governing salmon louse fecundity. As an example, intraspecific competition between lice on a given host is suggested to result in reduced fecundity with increasing salmon louse infection densities (Ugelvik et al., 2017). Louse fecundity is clearly the product of a number of biotic and abiotic factors, most of which remain to be fully characterized. Hatching Egg strings with non-viable eggs are sometimes extruded, and Heuch et al. (2000) found that this happened most frequently in the second and third batches of egg strings. Gravil (1996) reported that 2.1% of egg strings consisted entirely of non-viable eggs. According to Heuch et al. (2000), the number of viable eggs per string varied according to temperature, with a median of 13.3% of eggs being non-viable at 7.2°C and 7.5% being non-viable at 12.2°C. Similarly, Samsing et al. (2016) found that hatching success was strongly influenced by water temperature, with 100% success at 20°C and 15°C decreasing to 28 ± 4% success at 3°C. Conversely, Gravil (1996) found no correlation between egg viability and temperature in L. salmonis on farmed salmon on the West Coast of Scotland with a mean of 17.66 ± 23.01% non-viable eggs over 1 year. In comparison, the mean number of non-viable eggs per string in C. elongatus was 28.19 ± 24.81%, with 18.33% of egg strings entirely consisting of non-viable eggs (Gravil, 1996). Salinity has a considerable effect on hatching, and egg strings maintained at 10°C and 10‰ salinity failed to develop in Johnson and Albright’s (1991b) experiments. At salinities of 15‰ and 20‰, hatching success was 70% and 78%, respectively, but only at 20‰ were any active nauplii produced (19.8%). At salinities of 25‰ and above, hatching success was 100%, but at 25‰ only 51.1% of nauplii were active, whereas at 30‰ this figure was 65.9%. Gravil (1996) reports a similar pattern with hatching success ranging from 3.27% in freshwater to 86.36% at 30‰ salinity. The effect of photoperiod was investigated by Gravil (1996), but it had no effect on hatching period or success. Key values for hatching are shown in Table 2. Table 2. Key values of hatching in L. salmonis. L. salmonis salmonis L. salmonis oncorhynchi Proportion Time (h) Proportion Time (h) Non-viable egg strings 2.41%a – nd – Non-viable eggs per string 17.66%a – nd –  7.2°C 13.3%b – nd –  12.2°C 7.5%b – nd – Hatching period  5°C – 240a – nd  7°C – 192a – nd  10°C – 144a – 31.7 ± 17c Hatching success at 10°C  0 ppt 3.27%a – nd –  10 ppt nd – 0%c –  15 ppt nd – 70%c –  20 ppt nd – 78%c –  25 ppt nd – 100%c –  30 ppt 86.36%a – nd – Hatching success at 34 ppt  3°C 28 ± 4%d – nd –  5°C 85 ± 4%d – nd –  7°C 90 ± 4%d – nd –  10°C 87 ± 3%d – nd –  15°C 100%d – nd –  20°C 100%d – nd – Viability of nauplii  20 ppt nd – 19.8% (0–89.9) –  25 ppt nd – 51.1% (12–94.1) –  30 ppt nd – 65.9% (9.7–95) – L. salmonis salmonis L. salmonis oncorhynchi Proportion Time (h) Proportion Time (h) Non-viable egg strings 2.41%a – nd – Non-viable eggs per string 17.66%a – nd –  7.2°C 13.3%b – nd –  12.2°C 7.5%b – nd – Hatching period  5°C – 240a – nd  7°C – 192a – nd  10°C – 144a – 31.7 ± 17c Hatching success at 10°C  0 ppt 3.27%a – nd –  10 ppt nd – 0%c –  15 ppt nd – 70%c –  20 ppt nd – 78%c –  25 ppt nd – 100%c –  30 ppt 86.36%a – nd – Hatching success at 34 ppt  3°C 28 ± 4%d – nd –  5°C 85 ± 4%d – nd –  7°C 90 ± 4%d – nd –  10°C 87 ± 3%d – nd –  15°C 100%d – nd –  20°C 100%d – nd – Viability of nauplii  20 ppt nd – 19.8% (0–89.9) –  25 ppt nd – 51.1% (12–94.1) –  30 ppt nd – 65.9% (9.7–95) – References: (a) Gravil (1996), (b) Heuch et al. (2000), (c) Johnson and Albright (1991b), (d) Samsing et al. (2016), Mean ± SD, parentheses indicate ranges, nd = no data available. Table 2. Key values of hatching in L. salmonis. L. salmonis salmonis L. salmonis oncorhynchi Proportion Time (h) Proportion Time (h) Non-viable egg strings 2.41%a – nd – Non-viable eggs per string 17.66%a – nd –  7.2°C 13.3%b – nd –  12.2°C 7.5%b – nd – Hatching period  5°C – 240a – nd  7°C – 192a – nd  10°C – 144a – 31.7 ± 17c Hatching success at 10°C  0 ppt 3.27%a – nd –  10 ppt nd – 0%c –  15 ppt nd – 70%c –  20 ppt nd – 78%c –  25 ppt nd – 100%c –  30 ppt 86.36%a – nd – Hatching success at 34 ppt  3°C 28 ± 4%d – nd –  5°C 85 ± 4%d – nd –  7°C 90 ± 4%d – nd –  10°C 87 ± 3%d – nd –  15°C 100%d – nd –  20°C 100%d – nd – Viability of nauplii  20 ppt nd – 19.8% (0–89.9) –  25 ppt nd – 51.1% (12–94.1) –  30 ppt nd – 65.9% (9.7–95) – L. salmonis salmonis L. salmonis oncorhynchi Proportion Time (h) Proportion Time (h) Non-viable egg strings 2.41%a – nd – Non-viable eggs per string 17.66%a – nd –  7.2°C 13.3%b – nd –  12.2°C 7.5%b – nd – Hatching period  5°C – 240a – nd  7°C – 192a – nd  10°C – 144a – 31.7 ± 17c Hatching success at 10°C  0 ppt 3.27%a – nd –  10 ppt nd – 0%c –  15 ppt nd – 70%c –  20 ppt nd – 78%c –  25 ppt nd – 100%c –  30 ppt 86.36%a – nd – Hatching success at 34 ppt  3°C 28 ± 4%d – nd –  5°C 85 ± 4%d – nd –  7°C 90 ± 4%d – nd –  10°C 87 ± 3%d – nd –  15°C 100%d – nd –  20°C 100%d – nd – Viability of nauplii  20 ppt nd – 19.8% (0–89.9) –  25 ppt nd – 51.1% (12–94.1) –  30 ppt nd – 65.9% (9.7–95) – References: (a) Gravil (1996), (b) Heuch et al. (2000), (c) Johnson and Albright (1991b), (d) Samsing et al. (2016), Mean ± SD, parentheses indicate ranges, nd = no data available. The hatching period is variable, and Johnson and Albright (1991b) report that it ranged from 18 to 65 h, with a mean of 31.7 ± 13 h for egg strings incubated at 10°C and 30‰ salinity. The authors of the current review consider these to be at the extreme end of hatching periods observed based on personal observations, although this may represent a difference between Atlantic and Pacific L. salmonis. Stage timings Development times are highly dependent on temperature and have been addressed in various studies summarised in Table 3. Overall, the egg development time varies between 1.8 and 45.1 days for temperatures ranging between 2 and 20°C (Johnson and Albright, 1991b, Boxaspen and Næss, 2000; Samsing et al., 2016). The duration of the first nauplius stage varies between 9.2 and 52 h at temperatures ranging between 5 and 15°C, while the corresponding duration for the second nauplius stage varies between 33 and 170.3 h for temperatures ranging between 5 and 19°C (Johannessen, 1977; Wootten et al., 1982; Johnson and Albright, 1991b, Gravil, 1996). Durations of the stages seem to be comparable for Pacific and Atlantic lice, and reported ranges agree with the ranges found in publications where developmental times were reported for both naupliar stages combined (Gravil, 1996; Boxaspen and Næss, 2000; Samsing et al., 2016). While temperature has a considerable effect on egg production and larval development, photoperiod does not appear to have any significant effect (Ritchie et al., 1993; Gravil, 1996). Table 3. Key stage timings for L. salmonis (mean values). L. salmonis salmonis L. salmonis oncorhynchi Time (d) Time (h) Time (d) Time (h) Egg development time  2°C 45.1 ± 0.5e – nd –  3°C 35.2 ± 0.4e20.8 ± 1.5f – nd –  4°C 27.6 ± 0.2e – nd –  5°C 21.6 ± 0.1e13.0 ± 7.8f – 17.5a –  9°C 33–39b – nd –  9.5°C 25b – nd –  10°C 8.7 ± 0.1e4.6 ± 1.3f – 8.6a –  11.5°C 10–14b – nd –  15°C 2.88 ± 1.0f – 5.5a –  20°C 1.8 ± 0.5f – nd – Duration of first nauplius stage  5°C – nd – 52a  7.5°C – 43.25d – nd  9.2°C – 35b – nd  10°C – nd – 30.5a  12°C – 18c – nd  15°C – nd – 9.2a  15.5°C – 12b – nd Duration of second nauplius stage  5°C – nd – 170.3a  9.2°C – 77b – nd  10°C – nd – 56.9a  11°C – 63bc – nd  12°C – 46c – nd  15°C – nd – 35.6a  19°C – 33c – nd Development time to copepodid  2°C – 1644e – nd  5°C – 276f – nd  7°C – 168f – nd  10°C – 111–177.5d305e108f – nd  15°C – 36f – nd  20°C – 48f – nd L. salmonis salmonis L. salmonis oncorhynchi Time (d) Time (h) Time (d) Time (h) Egg development time  2°C 45.1 ± 0.5e – nd –  3°C 35.2 ± 0.4e20.8 ± 1.5f – nd –  4°C 27.6 ± 0.2e – nd –  5°C 21.6 ± 0.1e13.0 ± 7.8f – 17.5a –  9°C 33–39b – nd –  9.5°C 25b – nd –  10°C 8.7 ± 0.1e4.6 ± 1.3f – 8.6a –  11.5°C 10–14b – nd –  15°C 2.88 ± 1.0f – 5.5a –  20°C 1.8 ± 0.5f – nd – Duration of first nauplius stage  5°C – nd – 52a  7.5°C – 43.25d – nd  9.2°C – 35b – nd  10°C – nd – 30.5a  12°C – 18c – nd  15°C – nd – 9.2a  15.5°C – 12b – nd Duration of second nauplius stage  5°C – nd – 170.3a  9.2°C – 77b – nd  10°C – nd – 56.9a  11°C – 63bc – nd  12°C – 46c – nd  15°C – nd – 35.6a  19°C – 33c – nd Development time to copepodid  2°C – 1644e – nd  5°C – 276f – nd  7°C – 168f – nd  10°C – 111–177.5d305e108f – nd  15°C – 36f – nd  20°C – 48f – nd References: (a) Johnson and Albright (1991b), (b) Johannessen (1977), (c) Wootten et al. (1982), (d) Gravil (1996), (e) Boxaspen and Næss (2000), (f) Samsing et al. (2016), nd = no data available. Table 3. Key stage timings for L. salmonis (mean values). L. salmonis salmonis L. salmonis oncorhynchi Time (d) Time (h) Time (d) Time (h) Egg development time  2°C 45.1 ± 0.5e – nd –  3°C 35.2 ± 0.4e20.8 ± 1.5f – nd –  4°C 27.6 ± 0.2e – nd –  5°C 21.6 ± 0.1e13.0 ± 7.8f – 17.5a –  9°C 33–39b – nd –  9.5°C 25b – nd –  10°C 8.7 ± 0.1e4.6 ± 1.3f – 8.6a –  11.5°C 10–14b – nd –  15°C 2.88 ± 1.0f – 5.5a –  20°C 1.8 ± 0.5f – nd – Duration of first nauplius stage  5°C – nd – 52a  7.5°C – 43.25d – nd  9.2°C – 35b – nd  10°C – nd – 30.5a  12°C – 18c – nd  15°C – nd – 9.2a  15.5°C – 12b – nd Duration of second nauplius stage  5°C – nd – 170.3a  9.2°C – 77b – nd  10°C – nd – 56.9a  11°C – 63bc – nd  12°C – 46c – nd  15°C – nd – 35.6a  19°C – 33c – nd Development time to copepodid  2°C – 1644e – nd  5°C – 276f – nd  7°C – 168f – nd  10°C – 111–177.5d305e108f – nd  15°C – 36f – nd  20°C – 48f – nd L. salmonis salmonis L. salmonis oncorhynchi Time (d) Time (h) Time (d) Time (h) Egg development time  2°C 45.1 ± 0.5e – nd –  3°C 35.2 ± 0.4e20.8 ± 1.5f – nd –  4°C 27.6 ± 0.2e – nd –  5°C 21.6 ± 0.1e13.0 ± 7.8f – 17.5a –  9°C 33–39b – nd –  9.5°C 25b – nd –  10°C 8.7 ± 0.1e4.6 ± 1.3f – 8.6a –  11.5°C 10–14b – nd –  15°C 2.88 ± 1.0f – 5.5a –  20°C 1.8 ± 0.5f – nd – Duration of first nauplius stage  5°C – nd – 52a  7.5°C – 43.25d – nd  9.2°C – 35b – nd  10°C – nd – 30.5a  12°C – 18c – nd  15°C – nd – 9.2a  15.5°C – 12b – nd Duration of second nauplius stage  5°C – nd – 170.3a  9.2°C – 77b – nd  10°C – nd – 56.9a  11°C – 63bc – nd  12°C – 46c – nd  15°C – nd – 35.6a  19°C – 33c – nd Development time to copepodid  2°C – 1644e – nd  5°C – 276f – nd  7°C – 168f – nd  10°C – 111–177.5d305e108f – nd  15°C – 36f – nd  20°C – 48f – nd References: (a) Johnson and Albright (1991b), (b) Johannessen (1977), (c) Wootten et al. (1982), (d) Gravil (1996), (e) Boxaspen and Næss (2000), (f) Samsing et al. (2016), nd = no data available. The time required for physically moulting (exuviation) from nauplius I to nauplius II and nauplius II to copepodid are reported as 10.53 ± 4.34 min and 12.21 ± 3.87 min, respectively, and during the moult the larvae are inactive and sink through the water column (Gravil, 1996). It appears that the temperature of acclimation of adult female lice is important in determining the temperature tolerance of their eggs and larvae. Johannessen (1975) reports that in adult lice cultured at 9°C, nauplius development occurred only between 8 and 11°C, whereas acclimation at 11.5°C allowed larval development up to 22°C. In adult lice maintained at 3°C, however, nauplii failed to develop to copepodids (Samsing et al., 2016). Survival Nauplii that hatch successfully are planktonic. At this stage they do not feed, but are lecithotrophic (yolk feeding) and rely on their energy reserves until they moult to infective copepodids and find a suitable host (Johnson and Albright, 1991b). The survival of sea lice and the rate at which they deplete their energy reserves are strongly influenced by temperature and salinity. The size of larvae and their lipid stores is also dependant on season, and Gravil (1996) reports that nauplius I larvae were largest in August with a mean body width of 214.05 µm and a mean lipid reserve width of 135.84 µm compared to 197.76 µm and 112.98 µm in May for mean body width and mean lipid reserve width, respectively. It is likely that increased energy reserves will increase the longevity or compensate for a higher temperature-dependent metabolism of the non-feeding larval stages, although no data are available comparing survival at different times of year. Johnson and Albright (1991b) report that active copepodids were only obtained at salinities above 30‰ at 10°C (35.2% active), although survival was extremely variable ranging from 0 to 80.6% per egg string. Similarly, Gravil (1996) found that copepodids were only obtained at salinities greater than 25‰, and at 10°C and 35‰, 18.33% reached the infective copepodid stage with nearly 50% mortality being seen in the nauplius I stage. Samsing et al. (2016) found that sea lice larvae from Scotland did not proceed past the nauplius II stage at 5°C and 3°C, respectively, but died before moulting to copepodids, and at 7.5°C, very few copepodids were obtained (Gravil, 1996). In sea lice adapted to low temperatures, however, copepodids were obtained from 25% of egg strings reared at 2°C, 42% at 3°C, 100% at 4°C, and 75% at 5°C (Boxaspen and Næss, 2000). In C. elongatus, Pike et al. (1993) report 90% survival from the nauplius stage to the copepodid stage at 15°C with this figure decreasing to 60% at 5°C. As with all copepods, sea lice have preferred environmental conditions, which are determined by their physiological tolerances. Copepodids that were transferred from full-strength seawater to 5‰ salinity survived for just 3 h at 10°C, and those transferred to 10‰ salinity survived for less than 1 day (Johnson and Albright, 1991b). A similar experiment by Gravil (1996) found that the median survival time was 14.87 h at 0–10‰. While copepodids can osmoregulate above 16‰, their haemolymph becomes rapidly diluted below 12‰, and they are unable to regulate cell volume and die within a few hours (Hahnenkamp and Fyhn, 1985; Pike and Wadsworth, 1999). Once nauplii moult to copepodids, they need to find a suitable host before their lipid reserves are depleted, and the rate at which this occurs is also influenced by temperature and salinity. Hyperosmotic regulation is energetically costly, and an increased energy demand significantly reduces the survival time of copepodids due to their limited energy reserves (Torres et al., 2002). Johnson and Albright (1991b) report that survival was prolonged at salinities of 15–30‰ and temperatures of 5–15°C, and that mean survival times were between 2 and 8 days. Similarly, Wootten et al. (1982) report that the mean survival time of copepodids at 12°C was 4 days at an unspecified salinity. In Gravil (1996), the median survival time of copepodids was 54 h at 15‰, 67 h at 20‰, 68 h at 25‰, 55 h at 30‰, and 64 h at 35‰, which reflects the increased energy required for hyperosmotic regulation at lower salinities. Conversely, Bricknell et al. (2006) report the median survival time of L. salmonis copepodids to be 4 h at 16‰, 6 h at 19‰, 8 h at 23‰, 11 h at 26‰, 24 h at 29‰, 22 h at 33‰, and 25 h at 36‰. The reason for the differences in survival times reported in Gravil (1996) and Bricknell et al. (2006) is unknown, although Bricknell et al. used copepodids that were a few days old and cultured them with aeration whereas Gravil used unaerated containers. According to Johnson and Albright (1991b), the maximum survival time was 17 days at 10°C and 25‰ salinity, and copepodids in lower salinities (15–20‰) were generally less active than those maintained at higher salinities (25–30‰). In full strength seawater (35‰), the maximum survival time of copepodids at 10°C was 18 days (Gravil, 1996). Due to the reduced hatching success and subsequent low survival of L. salmonis in low salinities, it is likely that they may be excluded from salinities less than 15‰ (Johnson and Albright, 1991b), and survival is severely compromised at salinities below 29‰ (Tucker et al., 2000b). Although survival is reduced at lower salinities, short-term exposure to reduced salinities does not have a long-term impact on the development of surviving copepodids (Bricknell et al., 2006). Attachment to a host was not observed to improve survival at reduced salinities (Hahnenkamp and Fyhn, 1985) and these authors suggest that, unlike adult lice, the copepodid and chalimus stages are unable to use ions obtained from their host to replace those lost to a hypo-osmotic environment. However, it appears likely that due to their small size, attached larvae will receive at least some protection from reduced salinities through boundary layer effects coupled with close contact with the host/host mucus, and it is also clear that as these are feeding stages, some protection would be received from ingested host tissue. The survival time of copepodids is inversely related to temperature, and Samsing et al. (2016) report that the survival time of 80% of copepodids was 12.5 days at 7°C, 13 days at 10°C, 9.5 days at 15°C, and 6 days at 20°C; at 5°C it was reduced to 10 days. This pattern is presumably due to lower metabolism and, therefore, increased longevity of energy reserves at lower temperatures, although at very low temperatures there appear to be other factors limiting survival. Median survival times reported by Gravil (1996) were 116 h at 5°C, 90 h at 10°C, and 82 h at 15°C at full salinity (35‰), although these appear to be gross underestimations and may be due to sub-optimal culture conditions. There is, however, a seasonal investment by adult females in reproduction as nauplii are larger and have larger energy stores in summer than in winter (Gravil, 1996). At higher temperatures, metabolism is higher and larvae are more active, so their energy stores are more rapidly depleted (Gravil, 1996). It is possible that the increase in the size of larvae and their energy stores in summer may be a compensatory mechanism to account for their energy stores being depleted more rapidly than in winter, which ensures that their life expectancy is similar to that at colder winter temperatures. Further experimental work is required to confirm this. Key values for survival are shown in Table 4. Table 4. Key values of survival for L. salmonis larvae (50% survival times (LT50) are shown unless specified otherwise). L. salmonis salmonis L. salmonis oncorhynchi Width (µm) Proportion Time (h) Width (µm) Proportion Time (h) Nauplius I width  May 187.76a – – nd – –  August 214.05a – – nd – – Nauplius I lipid reserve width  May 112.98a – – nd – –  August 135.84a – – nd – – Survival to copepodid at 10°C  <25 ppt – 0%a – – nd –  <30 ppt – nd – – 0%b –  30 ppt – nd – – 35.2%b –  35 ppt – 18.33%a – – nd – Copepodid survival time at 10°C  0-10 ppt – – 15a – – nd  5 ppt – – nd – – 3b  10 ppt – – nd – – <24b  15 ppt – – 54a – – nd  16 ppt – – 4c – – nd  19 ppt – – 6c – – nd  20 ppt – – 67a – – nd  23 ppt – – 8c – – nd  25 ppt – – 68a – – Max. 17db  26 ppt – – 11c – – nd  29 ppt – – 24c – – nd  30 ppt – – 55a – – nd  33 ppt – – 22c – – nd  35 ppt – – 64 (max. 18d)a – – nd  36 ppt – – 25c – – nd Copepodid survival time at 35 ppt  5°C – – 116a – – nd 240 (LT80)e  7°C – – 300 (LT80)e – – nd  10°C – – 90a – – nd 312 (LT80)e  12°C – – 96d – – nd  15°C – – 82a – – nd 228 (LT80)e  20°C – – 144 (LT80)e – – nd L. salmonis salmonis L. salmonis oncorhynchi Width (µm) Proportion Time (h) Width (µm) Proportion Time (h) Nauplius I width  May 187.76a – – nd – –  August 214.05a – – nd – – Nauplius I lipid reserve width  May 112.98a – – nd – –  August 135.84a – – nd – – Survival to copepodid at 10°C  <25 ppt – 0%a – – nd –  <30 ppt – nd – – 0%b –  30 ppt – nd – – 35.2%b –  35 ppt – 18.33%a – – nd – Copepodid survival time at 10°C  0-10 ppt – – 15a – – nd  5 ppt – – nd – – 3b  10 ppt – – nd – – <24b  15 ppt – – 54a – – nd  16 ppt – – 4c – – nd  19 ppt – – 6c – – nd  20 ppt – – 67a – – nd  23 ppt – – 8c – – nd  25 ppt – – 68a – – Max. 17db  26 ppt – – 11c – – nd  29 ppt – – 24c – – nd  30 ppt – – 55a – – nd  33 ppt – – 22c – – nd  35 ppt – – 64 (max. 18d)a – – nd  36 ppt – – 25c – – nd Copepodid survival time at 35 ppt  5°C – – 116a – – nd 240 (LT80)e  7°C – – 300 (LT80)e – – nd  10°C – – 90a – – nd 312 (LT80)e  12°C – – 96d – – nd  15°C – – 82a – – nd 228 (LT80)e  20°C – – 144 (LT80)e – – nd References: (a) Gravil (1996), (b) Johnson and Albright (1991b), (c) Bricknell et al. (2006), (d) Wootten et al. (1982), (e) Samsing et al. (2016), nd = no data available. Table 4. Key values of survival for L. salmonis larvae (50% survival times (LT50) are shown unless specified otherwise). L. salmonis salmonis L. salmonis oncorhynchi Width (µm) Proportion Time (h) Width (µm) Proportion Time (h) Nauplius I width  May 187.76a – – nd – –  August 214.05a – – nd – – Nauplius I lipid reserve width  May 112.98a – – nd – –  August 135.84a – – nd – – Survival to copepodid at 10°C  <25 ppt – 0%a – – nd –  <30 ppt – nd – – 0%b –  30 ppt – nd – – 35.2%b –  35 ppt – 18.33%a – – nd – Copepodid survival time at 10°C  0-10 ppt – – 15a – – nd  5 ppt – – nd – – 3b  10 ppt – – nd – – <24b  15 ppt – – 54a – – nd  16 ppt – – 4c – – nd  19 ppt – – 6c – – nd  20 ppt – – 67a – – nd  23 ppt – – 8c – – nd  25 ppt – – 68a – – Max. 17db  26 ppt – – 11c – – nd  29 ppt – – 24c – – nd  30 ppt – – 55a – – nd  33 ppt – – 22c – – nd  35 ppt – – 64 (max. 18d)a – – nd  36 ppt – – 25c – – nd Copepodid survival time at 35 ppt  5°C – – 116a – – nd 240 (LT80)e  7°C – – 300 (LT80)e – – nd  10°C – – 90a – – nd 312 (LT80)e  12°C – – 96d – – nd  15°C – – 82a – – nd 228 (LT80)e  20°C – – 144 (LT80)e – – nd L. salmonis salmonis L. salmonis oncorhynchi Width (µm) Proportion Time (h) Width (µm) Proportion Time (h) Nauplius I width  May 187.76a – – nd – –  August 214.05a – – nd – – Nauplius I lipid reserve width  May 112.98a – – nd – –  August 135.84a – – nd – – Survival to copepodid at 10°C  <25 ppt – 0%a – – nd –  <30 ppt – nd – – 0%b –  30 ppt – nd – – 35.2%b –  35 ppt – 18.33%a – – nd – Copepodid survival time at 10°C  0-10 ppt – – 15a – – nd  5 ppt – – nd – – 3b  10 ppt – – nd – – <24b  15 ppt – – 54a – – nd  16 ppt – – 4c – – nd  19 ppt – – 6c – – nd  20 ppt – – 67a – – nd  23 ppt – – 8c – – nd  25 ppt – – 68a – – Max. 17db  26 ppt – – 11c – – nd  29 ppt – – 24c – – nd  30 ppt – – 55a – – nd  33 ppt – – 22c – – nd  35 ppt – – 64 (max. 18d)a – – nd  36 ppt – – 25c – – nd Copepodid survival time at 35 ppt  5°C – – 116a – – nd 240 (LT80)e  7°C – – 300 (LT80)e – – nd  10°C – – 90a – – nd 312 (LT80)e  12°C – – 96d – – nd  15°C – – 82a – – nd 228 (LT80)e  20°C – – 144 (LT80)e – – nd References: (a) Gravil (1996), (b) Johnson and Albright (1991b), (c) Bricknell et al. (2006), (d) Wootten et al. (1982), (e) Samsing et al. (2016), nd = no data available. Behaviour While both of the free-swimming larval stages are planktonic, the nauplius stages of sea lice are principally dispersal stages, whereas the copepodid stage must locate, re-establish contact with, and subsequently infect a suitable host. The larvae are subject to currents, which serve to disperse them over a wide area, and although the larvae have limited movement capabilities, their dispersal can be partially influenced by certain behaviours, e.g. aggregating at particular depths in the water column (Johnsen et al., 2014). In order to maximize their chances of survival and host interception, they must be able to respond to cues present in their environment and react to them appropriately. Their behavioural responses can be categorized according to the following activities (Bron et al., 1993): Predator avoidance Avoidance of adverse environmental conditions Movement into or maintenance within host-rich environments Host location Host contact/settlement Confirmation of host suitability Cues that may play a role in influencing the behaviour of sea lice larvae include light, chemical, pressure, temperature, and water flow/vibration. Swimming speed/activity Both nauplius and copepodid stages have been observed to actively swim upwards as they are negatively buoyant, and their movements are punctuated by periods of passive sinking (Bron, 1993; Gravil, 1996). Haury and Weihs (1976) suggest that this behaviour theoretically saves energy compared to continuous swimming at a fixed depth, which is particularly important for the lecithotrophic larvae of L. salmonis, which must conserve their limited energy reserves wherever possible. Despite their energy considerations, copepodids must maintain their position in the water column where their chances of encountering hosts are highest (Bron, 1993). However, Gravil (1996) found the activity of nauplii and copepodids to be dependent on temperature; at 5°C their movements were reduced and they aggregated at the bottom of containers, whereas at 10°C and 15°C they spent more time actively swimming than passively sinking and aggregated at the surface. However, these results may be affected by insufficient acclimation. Copepodids swim more rapidly than nauplii and have longer swimming periods and shorter rest periods (Bron, 1993). Gravil (1996) reports that the mean swimming speed of nauplii was 1.25 ± 0.16 cm s−1, whereas the mean swimming speed of copepodids was 2.14 ± 0.24 cm s−1. The mean sinking speeds were 0.09 ± 0.01 cm s−1 and 0.10 ± 0.03 cm s−1 for nauplii and copepodids, respectively. In this study, the maximum speed recorded was 10.23 cm s−1 when stimulated by vibration of the test chamber and gives an indication of the swimming ability of copepodids. A similar one-second burst speed of 9 cm s−1 was recorded by Heuch and Karlsen (1997), although a speed of 2 cm s−1 was sustained when stimulated. In comparison, the reported swimming speed of salmon is of two orders of magnitude higher (Colavecchia et al., 1998). Thus, while chemotaxis may be important in positioning the larvae in suitable water masses, the pursuit of a salmon host, as opposed to the interception of it at close range, is not a viable strategy. Current speed and host swimming speed affect the ability of infecting copepodids to make initial contact with the host and to remain attached following contact. Given the respective speeds of copepodids and salmonids, the former cannot pursue the host but must intercept it by burst swimming when detecting it in the water column. The exposure time of the copepodid to the host reduces with increasing current/host swimming speed, which in turn reduces the window of opportunity for infection. In addition, the low-flow zone (boundary layer) caused by drag at the surface of the fish, becomes thinner with increasing current/host speed, which increases the exposure of the copepodid to the ambient water flow during attachment. This means that at higher flows, the copepodid has less opportunity to make contact and is more likely to be removed from the host by the current (Bron, 1993). The greater boundary layer thickness and, hence, shelter from the ambient current offered by fin rays held perpendicular to the direction of water flow is considered to provide some explanation of the observed greater frequency of copepodid settlement on the fins of hosts (Bron, 1993; Bron et al., 1993). Similarly, the slower swimming speed of fish in tank challenges may explain the largely artefactual attachment of copepodids to the gills in such trials, an observation rarely made under field conditions (Bron et al., 1993). While larger fish swim faster, this is offset by the provision of a larger surface area for settlement and a greater boundary layer/shelter provided by larger fins. Frenzl (2014) observed declining number of attaching copepodids with increasing current speed. Following a dose of 2500 copepodids fish−1 introduced in a flume challenge, highest infection occurred at 0 cm s−1 (mean 8.4 copepodids per fish) and lowest at 32.6 cm s−1 (mean 0.2 copepodids per fish). Little is known concerning the effects of competition for space and/or resources during initial copepodid settlement. However, Frenzl (2014) has demonstrated a non-linear increase of infection numbers with challenge dose in flume challenges, possibly suggesting the increasing saturation of available settlement niches with increasing numbers available for infection. Light Copepodids of L. salmonis are highly photopositive and move toward the illuminated zone of the vessel in laboratory experiments even at low light intensities (Johannessen, 1975; Wootten et al., 1982; Bron et al., 1993; Gravil, 1996). The nauplius stages are also photopositive, but the nauplius I stage only exhibits a positive response at light intensities of 200 lux or more, whereas this value is 85 lux in the nauplius II (Gravil, 1996). Whereas nauplii exhibit increasing activity with increasing light intensity, copepodids do not (Gravil, 1996). The free-swimming larval stages of C. elongatus are also phototactic, with the copepodids showing a contrasting greater response to light than the nauplii stages (Hogans and Trudeau, 1989). In L. salmonis, a peak response was seen at a wavelength 500 nm in the nauplius II stage (Gravil, 1996) and 550 nm in the copepodid stage (Bron et al., 1993; Gravil, 1996), and this corresponds to the maximum transmitted light intensity at twilight, which may be a cue for vertical migration in copepodids as suggested for free-living copepods (Forward and Douglass, 1986). In flume challenges, Frenzl (2014) found maximum sensitivity of copepodids to light at 455 nm. In addition to the response to constant light, evidence for a response to changing light intensities/shadows (scototaxis) in adult sea lice (authors’ qualitative observations) and copepodids (Fields et al., 2017) strongly indicates a behavioural response toward moving objects obstructing or reflecting light. Heuch et al. (1995) found a strong diel vertical migration in L. salmonis copepodids where they gathered near the surface during the day and spread out into deeper layers at night. Despite the recognized photopositive behaviour of copepodid stages, a number of authors observed successful settlement or attempted settlement in darkness (Johnson and Albright, 1991b; Bron et al., 1993; Heuch et al., 2007; Frenzl, 2014), although settlement success was generally lower than when under illumination. As salmon remain in deeper waters during the day and rise to the surface at night, they swim through a population of sinking or rising copepodids every 12 h (Heuch et al., 1995). In addition, vertically migrating hosts produce stronger currents than resting fish, and pressure waves in front of swimming fish trigger a looping behaviour allowing nearby copepodids to avoid predation and attach to a host (Bron et al., 1993; Heuch and Karlsen, 1997; Heuch et al., 2007). Bron et al. (1993) and Gravil (1996) also demonstrated that copepodids are negatively geotactic, i.e. they swim toward the surface, which also suggests that they tend to aggregate in surface waters. Presumably, these experiments were conducted with illumination, and therefore, it is not known whether copepodids would be negatively geotactic in the dark when they would normally spread out into deeper water. In the study by Heuch et al. (1995), 6 m-deep mesocosm bags were suspended in the water column, and therefore, the vertical migrations of copepodids were limited by the depth of the bags. Zooplankton appear to scale their vertical migrations according to the available water depth (Young and Watt, 1993), so the relationship of experiments with constrained depths to the natural situation is uncertain. This has implications for the dispersal of lice by water currents as current velocity and direction often vary with depth. It is clear, however, that wind forcing can be a dominant component of sea lice dispersal (Murray and Amundrud, 2007; Amundrud and Murray, 2009), and therefore, improved knowledge of the diel vertical migration of copepodids between surface and deeper waters would allow the wind forcing component of sea louse dispersal to be predicted more accurately. Salinity In salinities less than 21‰, the swimming ability of nauplii and copepodids is lost, although full activity is recovered if the exposure time is short (< 5 min) (Gravil, 1996). Bricknell et al. (2006) found that copepodids actively avoided salinities lower than 27‰ by orientating themselves in a vertical sinking position and occasionally actively swimming downward. Given a choice, they will remain in full strength seawater. Energy is expended for osmoregulation and to maintain their position in the water column, as sinking rates increase with decreasing salinity due to water density changes (Bricknell et al., 2006). It is likely that copepodids avoid areas of low salinity as they require increased energy expenditure, which reduces survival time (Torres et al., 2002). As low salinities reduce the activity levels of copepodids, their ability to respond to host cues is reduced (Bricknell et al., 2006). Currents It has been proposed, although supporting evidence is lacking, that copepodids may actively migrate to river mouths where high concentrations of salmon smolts are present at certain times of year, which would increase their probability of encountering a host (Carr and Whoriskey, 2004; Costello et al., 2004; McKibben and Hay, 2004). Studies in estuarine areas in Ireland suggest that copepodids are not found near the mouths of rivers for the majority of the year (Costelloe et al., 1998a), but high concentrations coincide with the seaward migration of salmon smolts and the freshwater migration of adult salmon (Costelloe et al., 1998a; McKibben and Hay, 2004). As copepodids are capable of actively altering their position in the water column, it is possible that they may be able to use vertical positioning to compensate for lack of long distance swimming capabilities, using tidal currents to migrate toward river mouths, although no evidence has been found to support this. As copepodids have been shown to remain active in the water column (Bron et al., 1993; Heuch et al., 1995; Gravil, 1996), they are distributed within a water body according to the prevailing currents and are, thus, unlikely to directly influence their large-scale movement toward a particular location. It has been suggested that at some times of the year, a high concentration of copepodids near river mouths could result from hatching of egg strings from lice on adult salmon, which often congregate at river mouths prior to their migration upstream, particularly during periods of low river flow (Jonsson et al., 1990; Smith et al., 1994). Similarly, the absence of copepodids at river mouths during periods of high rainfall might simply be due to salmon migrating rapidly upstream when river flow is high (Costelloe et al., 1998a, b) Host location The responses of sea lice copepodids to physical cues, such as light and salinity, enable them to gather in areas where host fish are likely to be found, and mechanical cues enable them to infect a host. Chemoreception also plays an important role in host location, with copepodids employing the cues provided by kairomones, specific chemicals released by host fish, to improve the probability of host encounter. Copepodids swim with a general search pattern, but once a host odour has been detected, a host-encounter search pattern is switched on, which consists of increased duration and frequency of turning during the normal sinking and swimming behaviour (Genna, 2002). A directional component is also apparent whereby activated copepodids swim toward a suitable odour source over a distance of centimetres (Bailey et al., 2006), although a group of salmon might initiate a response over a scale of metres (Mordue Luntz and Birkett, 2009). Experiments have shown that L. salmonis copepodids are attracted to odours from salmon and sea trout, and behavioural activation and positive upstream chemotaxis occur in the presence of salmon-derived compounds (Devine et al., 2000; Genna, 2002; Ingvarsdottir et al., 2002; Bailey et al., 2006). While both light and chemoreception elicit behavioural responses in the infective copepodids, it has been shown that the effect of light on the swimming response is stronger than that of responses elicited by olfactory cues and that the two sources of sensory cues may act in combination to give stronger and more persistent responses (Fields et al., 2017). Non-host odours activate copepodids, but positive chemotactic movements are not observed, indicating that L. salmonis can discriminate between salmonid hosts and other non-host fish from their odour (Bailey et al., 2006). In comparison, C. elongatus, which is a generalist and infects many different species of fish, demonstrates behavioural changes to chemical cues from a wide range of fish, although physical cues may be more dominant in this species (Mordue Luntz and Birkett, 2009). Although the activity of copepodids appears to be affected by temperature, with reduced activity at lower temperatures (Tucker et al., 2000b), it is not known whether low temperatures affect the switch to host-seeking behaviour and the distance over which they may be able to detect host cues. Despite their avoidance of areas of low salinity, the use of haloclines by copepodids has been proposed as a host-finding mechanism, since host odours may accumulate in thin layers where a density gradient occurs. In this respect, 80% of copepodids were observed to aggregate at the confluence of a 15–30‰ step-salinity gradient in laboratory experiments (Heuch, 1995). In addition, positioning close to a halocline may increase the chance of encountering a host, as salmon have been observed to follow salinity gradients (Lyse et al., 1998; Finstad et al., 2000). Infectivity While some previous models of sea louse dispersion include a mortality factor, they do not account for variations in infectivity, i.e. the ability of a louse encountering a fish to infect it. Infection can be considered in terms of a two-phase process comprising a reversible attachment phase following contact and an irreversible settled phase during which the copepodid becomes physiologically committed and can no longer re-enter a free-swimming state. In the salmon louse, the former phase comprises initial copepodid attachment using the antennae (Bron et al., 1991) followed by manoeuvres to embed the anterior of the cephalothorax. At some point following initial attachment, the copepodid commences feeding and starts the process of metamorphosis and moulting to the chalimus I stage. Although the precise triggers and point of irreversible commitment remain to be identified, antimicrobial peptides (AMPs) have been shown to affect C. rogercresseyi frontal filament development in vitro (Núñez-Acuña et al., 2016). It is, therefore, incorrect to assume that, once the copepodid stage is reached, 100% infection will occur (Gravil, 1996). Dispersion on currents and host location behaviour bring the copepodids into the same locality as potential hosts, but the process of infection is influenced by various factors, including salinity, light, temperature, season, a range of host factors, and copepodid age. A further difficulty encountered in the literature is the somewhat nebulous concept of “infection success.” For some authors, copepodids attaching to the fish are counted directly. However, given the reversible nature of initial attachment and difficulty of capturing fish without dislodging attached copepodids, such counts may be prove less accurate, although they provide an estimate of successful contact and attachment. As an alternative, many authors only count infection success following the moult to chalimus I, at which point larvae are hard to dislodge due to the permanent frontal filament attachment. This latter approach, however, incorporates a far greater potential for the superposition of host immunity/site selection effects upon the successful completion of the copepodid instar. Age at infection As lecithotrophic larval stages are reliant on their energy reserves for swimming, moulting, and host infection, the excessive depletion of these reserves prior to infection can result in the loss of infective capability. As copepodids age, a higher proportion display reduced activity due to the depletion of energy reserves or senescence (Bron, 1993). Gravil (1996) found that the mean size of lipid vesicles in the mid-gut of copepodids was significantly reduced after 7 days, and Tucker et al. (2000a) report a significant reduction in the calorific value of L. salmonis larvae over 7 days with a sharp decline after 5 days. By measuring stored lipid volume, it is possible to determine age and viability in individual copepodids, and these can be divided into three loose categories: early copepodids with an apparent increase in lipid volume reflecting incorporation of naupliar lipids into distinct vesicles in the gut; mid-life copepodids, which show a downward trend in lipid levels and may be the most active individuals with mature infective capabilities; and late copepodids with low reserves of lipid, which may be less capable of infection (Cook et al., 2010). The depletion of energy reserves, which consist primarily of lipids, might also result in a loss of buoyancy, making swimming more energetically costly (Bron, 1993), although Gravil (1996) found no evidence to support this. Gravil (1996) observed three stages of activity: newly moulted copepodids swam in spontaneous bursts without stimulation; at 8 days at 10°C, 50% of copepodids were only active when stimulated; after 8 days, remaining copepodids only showed activity after being stimulated by a water jet from a pipette. This suggests that copepodids may adopt a strategy of energy conservation if a host is not located after a certain period of time, and that by only becoming active when stimulated, they preserve their remaining energy stores as long as possible. This reduced activity level affects infectivity, and Gravil (1996) reports that copepodid infection success at 10°C and 35‰ salinity was 22.22 ± 8.32% at 1 day old and 14 ± 8.71% at 7 days old. At 7 days old, approximately 20% of copepodids were active without stimulation and 40% were active with or without stimulation. Bron (1993) reports similar infection rates with 23.2% settlement under illuminated conditions and 18.4% settlement in the dark for 1–3-day-old copepodids, although there was no significant difference in settlement between light and dark conditions. For a cohort of copepodids hatched within 24 h, Frenzl (2014) found in flume challenges that maximal infectivity was obtained at 4 days post-moult to copepodid, with the infectivity of the cohort declining by 6 days through mortalities and lower infective capabilities. Tucker et al. (2000a) found that infection success (measured as the proportion of larvae used for infection that were found on the fish at day 5 after infection) was approximately 75% at 11°C and approximately 20% at 6.5°C in 1-day-old and 3-day-old copepodids, with infection success declining significantly in 7-day-old copepodids, although lice in this experiment were collected and cultured at 10°C before being used in experiments, which may have affected the results. The ability of copepodids to infect hosts past 7 days old is known from experiments with L. salmonis (Pedersen, 2009), but detailed temporal infectivity profiles have not been published. However, infection success is clearly linked to both the longevity and activity of the copepodid stage. Despite infection success being dependent on copepodid age, the survival of copepodids once attached to a host was not observed to differ between copepodids that infect at different ages (Tucker et al., 2000a; Pedersen, 2009), which is likely due to the commencement of feeding once attached to a host. This suggests that key determinants of variability of larval infection levels in Atlantic salmon act prior to host settlement, i.e. within the black box comprising egg production to host contact. Impacts of environmental variables on infection Host settlement success is also reduced at lower salinities, which coincides with a decrease in their energy reserves (Tucker et al., 2000a, b; Bricknell et al., 2006). It is likely that the physiological stress associated with reduced salinity rapidly depletes the energy reserves of copepodids, which causes premature senescence and results in levels of settlement success similar to those found in older copepodids (Bricknell et al., 2006). These authors report that infection levels were reduced by 45% at 26‰ (∼14% infection), 55% at 19‰ (∼10% infection), and 87.5% at 12‰ (∼1% infection) compared to full-strength seawater, which was not wholly attributable to reduced survival at these salinities. At 4‰ no copepodids were found on the fish. While settlement success is lower with reduced energy reserves, Samsing et al. (2016) used degree days to normalize copepodid energy reserves cultured at different temperatures; at 30 degree days from hatching, settlement success was 41.6 ± 2.0% at 20°C, 53.2 ± 2.3% at 10°C, and 2.1 ± 0.4% at 5°C. Key values for infectivity are shown in Table 5. Table 5. Key variables of infectivity in L. salmonis salmonis larvae. Infectivity capability Lipid reserves Proportion Copepodid age  7–10d Increasingabcd Goodabcd –  11–15d Matureabcd Decreasingabcd –  16–20d Less capableabcd Lowabcd – Infection success at 10°C and 35 ppt  1-day-old copepodids – – 22.22 ± 8.32%c  7-day-old copepodids – – 14 ± 8.71%c Infection success aged 1–3d  Illumination – – 23.2%b  No illumination – – 18.4%b Infection success at 35 ppt  5°C – – 2.1 ± 0.4%e  6.5°C – – 20%d  10°C – – 53.2 ± 2.3%e  11°C – – 75%d  20°C – – 41.6 ± 2.0%e Infection success at 12°C  12 ppt – – 1%f  19 ppt – – 10%f  26 ppt – – 14%f  34 ppt – – 31%f Infectivity capability Lipid reserves Proportion Copepodid age  7–10d Increasingabcd Goodabcd –  11–15d Matureabcd Decreasingabcd –  16–20d Less capableabcd Lowabcd – Infection success at 10°C and 35 ppt  1-day-old copepodids – – 22.22 ± 8.32%c  7-day-old copepodids – – 14 ± 8.71%c Infection success aged 1–3d  Illumination – – 23.2%b  No illumination – – 18.4%b Infection success at 35 ppt  5°C – – 2.1 ± 0.4%e  6.5°C – – 20%d  10°C – – 53.2 ± 2.3%e  11°C – – 75%d  20°C – – 41.6 ± 2.0%e Infection success at 12°C  12 ppt – – 1%f  19 ppt – – 10%f  26 ppt – – 14%f  34 ppt – – 31%f References: (a) Cook et al. (2010), (b) Bron (1993), (c) Gravil (1996), (d) Tucker et al. (2002), (e) Samsing et al. (2016), (f) Bricknell et al. (2006). No infectivity data is available for L. salmonis oncorhynchi. Table 5. Key variables of infectivity in L. salmonis salmonis larvae. Infectivity capability Lipid reserves Proportion Copepodid age  7–10d Increasingabcd Goodabcd –  11–15d Matureabcd Decreasingabcd –  16–20d Less capableabcd Lowabcd – Infection success at 10°C and 35 ppt  1-day-old copepodids – – 22.22 ± 8.32%c  7-day-old copepodids – – 14 ± 8.71%c Infection success aged 1–3d  Illumination – – 23.2%b  No illumination – – 18.4%b Infection success at 35 ppt  5°C – – 2.1 ± 0.4%e  6.5°C – – 20%d  10°C – – 53.2 ± 2.3%e  11°C – – 75%d  20°C – – 41.6 ± 2.0%e Infection success at 12°C  12 ppt – – 1%f  19 ppt – – 10%f  26 ppt – – 14%f  34 ppt – – 31%f Infectivity capability Lipid reserves Proportion Copepodid age  7–10d Increasingabcd Goodabcd –  11–15d Matureabcd Decreasingabcd –  16–20d Less capableabcd Lowabcd – Infection success at 10°C and 35 ppt  1-day-old copepodids – – 22.22 ± 8.32%c  7-day-old copepodids – – 14 ± 8.71%c Infection success aged 1–3d  Illumination – – 23.2%b  No illumination – – 18.4%b Infection success at 35 ppt  5°C – – 2.1 ± 0.4%e  6.5°C – – 20%d  10°C – – 53.2 ± 2.3%e  11°C – – 75%d  20°C – – 41.6 ± 2.0%e Infection success at 12°C  12 ppt – – 1%f  19 ppt – – 10%f  26 ppt – – 14%f  34 ppt – – 31%f References: (a) Cook et al. (2010), (b) Bron (1993), (c) Gravil (1996), (d) Tucker et al. (2002), (e) Samsing et al. (2016), (f) Bricknell et al. (2006). No infectivity data is available for L. salmonis oncorhynchi. Post-attachment variables A number of variables intervene between initial attachment of the copepodid and successful moulting to the chalimus I stage. In particular, once attached, the copepodid becomes susceptible to host defences, particularly in terms of innate host immunity, often expressed through inflammatory processes. The success of the host response in controlling infection depends upon a number of variables including the species/genotype of the host fish, its age, maturity, health and welfare/stress status, and interactions of immune capabilities with environmental parameters such as temperature. The role of the host in mediating infection success will only be covered briefly here as it has been extensively reviewed and investigated by previous authors (Skugor et al., 2008; Tadiso et al., 2011; Fast, 2014; Braden et al., 2017,inter alia). In Atlantic salmon, initial infection by the copepodid can elicit a detectable transcriptomic host response within 1 day post infection (dpi) (Tadiso et al., 2011) and some Pacific salmon species, e.g. juvenile coho, are able to mount a rapid and successful inflammatory response following infection (Johnson and Albright, 1992; Fast et al., 2002; Jones, 2011) that is capable of killing infecting copepodids within a few days. Atlantic salmon show a less developed inflammatory response and are generally considered to show a poor capacity for removing infecting copepodids (Johnson and Albright, 1992). Despite this observation, different genetic stocks or families of Atlantic salmon can show significant differences in their capacity to resist infection, although the mechanisms underlying differential resistance are currently poorly understood. Jodaa Holm et al. (2015) have suggested that differential resistance may reflect the ability of the host to avoid immunosuppression by the parasite. In a comparison of salmon family susceptibility, Gharbi et al. (2015) demonstrated a ∼60% difference in the median infection count at 7 dpi (chalimus I) for the least and most susceptible salmon families tested by copepodid infection challenge and calculated a genetic heritability of 0.3 for this trait making it a good candidate for selective breeding. The capacity of salmon to reduce infection success may also be modified by extrinsic factors such as diet and temperature. Functional feeds containing a range of active plant or bacterial extracts have, for example, been shown to have significant effects on infection success, providing infection reductions of up to 50% (Jensen et al., 2014; Jodaa Holm et al., 2016; Sutherland et al., 2017). Sea lice, like other arthropod parasites, can also suppress or redirect host immune responses by the use of a range of secretory excretory products (SEPs) including prostaglandin E-2, trypsin, peroxinectin and a range of other proteases, peroxidases, and potential defensin classes (Fast, 2014; Øvergård et al., 2016). The success of the parasite in immunomodulating the host depends on the individual host’s innate susceptibility and its state at the time of infection. Similarly, the status of the parasite can be important such that, for example, genetic family differences may affect infection success (Ljungfeldt et al., 2014) although the point at which success is mediated and the mechanisms involved remain unknown. Mortality through predation Once sea lice have attached to a host, their chances of survival are increased as they have a constant food supply and external factors affecting survival are relatively few, e.g. adverse environmental conditions, host immune response, and predation by cleaner fish. During their free-swimming planktonic stages, however, they form a part of a complex plankton food web and are subject to selective and non-selective predation by other plankton and sessile filter feeders such as bivalve molluscs. Global approximations of the partitioning of wider zooplankton mortality suggest that predation accounts for 67–75% of total mortality in the plankton (Hirst and Kiørboe, 2002). Although predation is likely to have a significant impact on sea lice survival, there are currently no estimates of sea lice predation mortality in the literature due to the difficulty in obtaining this kind of information. Some sea lice dispersion models do include a fixed mortality rate for the free-swimming stages, e.g. Amundrud and Murray (2009) used a fixed mortality rate of 0.01 h−1 for nauplii and copepodids. Providing an estimate of predation mortality is difficult as plankton assemblages vary considerably according to season and location (e.g. Daewel et al., 2014), and prey selection sizes vary amongst the different actively or passively predating species represented in the zooplankton community at any time (Hansen et al., 1994; Wirtz, 2011, 2012). As a consequence of a lack of specific data, the following discussion seeks to provide guidance based on wider knowledge of zooplankton, which may be used by researchers to formulate research questions or provide initial parameters for models. Plankton community structure In regional marine ecosystems, several processes govern the structure and dynamics of plankton communities. These processes vary according to geographical location, resulting in distinct ocean regions with their own typical plankton assemblages. Small copepods dominate inshore zooplankton with their seasonal abundance following that of the phytoplankton, and clupeid and scombrid fish are the main consumers of pelagic invertebrates (Kaiser, 2005). These broad ocean regions may further be characterized according to ocean processes in different sub-regions, e.g. the North Sea, the Norwegian Sea. The abundance of different species that are predators of sea lice larvae and the abundance of other prey will affect the mortality rate of sea lice larvae. Therefore, providing data on larval predation by different plankton assemblages and characterizing the plankton assemblage at a specific location represents an important step in predicting mortality rates due to predation. Predator selectivity The body sizes of predator and prey are fundamental in the study of aquatic food webs (Brooks and Dodson, 1965; Woodward et al., 2005). A “feeding kernel” represents a description of the probability of prey ingestion given as a function of feeding rate vs. prey size (Figure 2) (Visser and Fiksen, 2013; Wirtz, 2014). Selective grazing in the presence of a broad spectrum of prey size plays an important role in variable feeding relationships (Sommer and Stibor, 2002), and in the case of larval sea louse predation, the abundance of similar-sized prey must be considered as well as the abundance and size selectivity of predators. Figure 2. View largeDownload slide Typical ingestion (light grey area) and selection (dark grey area) feeding kernels for (a) narrow-range, selective feeders, e.g. copepods, and (b) broad-range, unselective feeders, e.g. jellyfish, where prey are abundant. Adapted and redrawn from Wirtz (2014). Figure 2. View largeDownload slide Typical ingestion (light grey area) and selection (dark grey area) feeding kernels for (a) narrow-range, selective feeders, e.g. copepods, and (b) broad-range, unselective feeders, e.g. jellyfish, where prey are abundant. Adapted and redrawn from Wirtz (2014). Although the relationship between predator and prey body sizes is the primary determinant of grazing selectivity, feeding modes can also affect the size range of plankton selected. Feeding modes can be broadly classified as passive and active ambush feeding, feeding-current feeding and cruise feeding (Kiørboe, 2011), and predators may adjust their feeding behaviour in response to the density of food items (e.g. Frost, 1972; Kiørboe and Saiz, 1995; Saiz and Kiørboe, 1995; Boenigk and Arndt, 2002; Visser et al., 2009). This behavioural plasticity shrinks the overall spectrum of potential prey toward a specific sub-range, and Wirtz (2014) describes two feeding kernels: one for ingestion, which is based on the size range of prey that can be ingested based on biomechanical principles, and one for selection, which describes the actual size range of prey selected according to the availability of prey of various sizes (Figure 2). At high prey densities, many ambush and suspension feeders, such as copepods, typically have a high selectivity resulting in a narrow selection kernel (Figure 2a), whereas many facultative, omnivorous feeders, such as jellyfish, typically have broad ingestion and selection kernels (Figure 2b) (Wirtz, 2014). Prey selection Prey size selection is determined according to the equivalent spherical diameter (ESD), which is the longest axis of the prey, i.e. length for sea lice larvae. Johnson and Albright (1991a) report that the length of the nauplius I was 0.54 ± 0.04 mm, the nauplius II was 0.56 ± 0.01 mm, and the copepodid was 0.70 ± 0.01 mm in L. salmonis oncorhynchi collected from British Columbian waters. Schram (1993) reports similar ranges for L. salmonis salmonis collected in Norway. Potential predators of sea lice larvae are likely to include obligate and facultative carnivorous zooplankton and planktivorous fish, and given their geographical distribution, predators may be represented by chaetognaths, ctenophores, scyphozoa, euphausiids, mysids and scombrid, and clupeid fish. In addition, the larval stages of most fish species rely on copepods as their principal dietary component (Kaiser, 2005). Chaetognaths, or arrow worms, are important predators of copepods and are probably major contributors to the structuring of many marine ecosystems (Steele and Frost, 1977). Chaetognaths are ambush predators, and Fulton (1984) found that active copepods, such as Acartia tonsa, decreased in abundance in the presence of Sagitta hispida, whereas inactive swimmers, such as Oithona spp. did not as encounter rates were lower. As sea lice larvae are active swimmers, it is likely that they will be predated by chaetognaths of a suitable size category. Ctenophores, or comb jellies, are found throughout the world’s oceans, and all are predatory, feeding on zooplankton (Fowler, 1911). If food is plentiful, they can eat ten times their own weight per day (Reeve et al., 1978). In laboratory experiments, copepodid I larvae of Calanus pacificus with a mean length of 0.74 mm and mean swimming speed of 0.32 mm s−1, hence similar in size to sea lice larvae, were most susceptible to predation by P. bachei, and later juvenile stages, which are larger, were less susceptible to predation (Greene et al., 1986). Scyphozoa, or jellyfish, are generally larger than many other predators in the plankton, and are seasonally common in many coastal environments including those most commonly employed for marine salmonid aquaculture (Doyle et al., 2007). Scyphozoa typically range from 2–40 cm, and their stinging or filter-feeding tentacles enable them to ingest various zooplankton taxa of different sizes, including copepods (Purcell, 1992; Purcell et al., 1994; Suchman and Sullivan, 1998). However, research has shown that scyphozoa are highly selective, and prey size has a significant impact on feeding rates (Suchman and Sullivan, 1998, 2000). As scyphozoa are neither visual nor raptorial feeders, they select prey as a consequence of prey vulnerability, and prey with faster swimming speeds and poor escape responses are most vulnerable to predation (Suchman and Sullivan, 2000). Euphausiid and mysid shrimps are two groups of arthropods that are ubiquitous throughout the world’s oceans, and due to their high abundance and position in the food chain, they are important components of marine food chains (Båmstedt and Karlson, 1998). While most are omnivorous filter feeders and feed on phytoplankton and detritus, some are carnivorous and feed on other zooplankton (Cripps and Atkinson, 2000). In the Norwegian Sea, the copepod Calanus finmarchicus (which has similar-sized juvenile stages to sea lice) is a dominant prey of euphausiid shrimp (Båmstedt and Karlson, 1998). The larval stages of many fish species rely on copepods as their principal dietary component, and although larger gadoids, such as Atlantic cod (Gadus morhua) switch to piscivory as adults, smaller species, such as Norway pout (Trisopterus esmarkii) and clupeids, such as herring (Clupea harengus) remain planktivorous throughout their lives (Daewel et al., 2014). As larval fish are active raptorial predators and rely on sight to detect prey, active prey may be more susceptible to predation. Tiselius and Jonsson (1990) and Doall et al. (1998) suggest that the high turn rates of sea lice copepodids during host-seeking behaviour may make them more attractive to predators, such as fish larvae. Some adult fish, such as scombrids and clupeids, feed on plankton throughout their lives, and switch between feeding modes depending on prey density (Janssen, 1976). Zooplankton consumption by fish in the North Sea has been estimated at 19–25 g C m−2 year−1 of which 28% of overall zooplankton consumption can be attributed to early life stages of fish (Heath, 2007). In frontal zones, fish larvae could consume up to 3–4% day−1 of the fraction of preferred zooplankton sizes (Munk et al., 1994). In addition to planktonic predators, sessile feeders, particularly bivalve molluscs and cnidarians, could also have a potential impact on larval sea louse survival. Bivalve molluscs, specifically the blue mussel Mytilis edulis, have been suggested to provide efficient clearance of mesoplankton of the same size order as sea lice larvae (Davenport et al., 2000). Only blue mussels and scallops (Placopecten magellanicus) have been specifically investigated in terms of their ability to clear larval sea lice (Molloy et al., 2011; Bartsch et al., 2013). Molloy et al. (2011) demonstrated that mussels were capable of removing copepodids from the water column under experimental conditions and this was also demonstrated by Bartsch et al. (2013) who showed that mussels and scallops could remove 18–38% of presented copepodids per hour. While it has been suggested that mussels or other bivalves might, therefore, be employed to help control sea lice on farms (Molloy et al., 2011; Bartsch et al., 2013), it has been noted (Sandra Bravo, pers. comm.) that close proximity of mussel farms and salmon farms in Chile has not served to reduce apparent levels of sea lice infections. The foregoing observations on levels of predation of zooplankton support the suggestion that the mortality of free-living sea lice stages, i.e. nauplii and copepodids, is likely to be high during the planktonic phase. Research gaps identified, recommendations, and conclusions A broad range of factors impact the levels of egg production by host-attached lice and the subsequent proportion of the initial extruded egg number that go on to successfully infect fish as copepodid larvae. Figure 3 shows the stages of the sea louse life cycle that determine the number of copepodids available for infection and their infection success and summarizes the factors reviewed in this study that may affect subsequent levels of infection. Figure 3. View largeDownload slide A conceptual model of the stages of the sea louse life cycle that determine the number of copepodids available for infection and their infection success with factors that may affect survival/infectivity at each stage. Open arrows show the life cycle and black arrows show the factors that may affect each stage of the life cycle. Figure 3. View largeDownload slide A conceptual model of the stages of the sea louse life cycle that determine the number of copepodids available for infection and their infection success with factors that may affect survival/infectivity at each stage. Open arrows show the life cycle and black arrows show the factors that may affect each stage of the life cycle. A simplified conceptual framework can be employed to summarize the findings of this review, which describes the relationships between the production and loss of free-swimming larval lice and aspects of their behaviour that together determine subsequent infection levels: S=EPhPpPdPsPeI Where S is number of successfully infecting copepodids, E is the number of eggs produced, Ph is the probability of hatching, Pp is the probability of avoiding predation, Pd is the probability of successful development from nauplii to copepodids, Ps is the probability of copepodid mortality due to senescence, Pe is the probability of encountering an appropriate host, and I is the mean infectivity of the copepodid population. The operational use of this conceptualized framework requires the estimation of the components of each of these variables, which are themselves influenced by a range of biotic (e.g. host) and abiotic (e.g. water temperature) factors and each other, i.e. they are not independent. As each component (or loss) is multiplicative, the uncertainties in each component may result in very wide error margins in S. Therefore, it is important to define and continue to refine each component through extensive data collection and parameterization to reduce the level of error. By forming a table of these variables and the observable factors that may influence them (Table 6), it is clear that there are a considerable number of permutations, each requiring observational data to allow variables to be fully defined. While a number of these variables have been previously investigated, as described in this review, a lack of data for some variables results in an incomplete dataset (Table 6). Furthermore, a lack of standardization and consistency across different studies due to various experimental conditions and the origin of experimental lice, e.g. of Atlantic or Pacific origin, farmed or wild origin, cold-adapted or not, means that many data points are not directly comparable. In addition, some studies are based on laboratory experiments conducted under controlled conditions, whereas others are based on field data. Gravil (1996) recorded the widths of nauplius I larvae and the lipid reserves from field-collected lice at different times of year, and although no other studies considered seasonal variations in their experiments per se (Table 6), seasonal variation subsumes a number of observable/observed factors, such as temperature, photoperiod and salinity, and other factors that are not considered here, such as host condition and plankton assemblages. Table 6. A summary table of parameters influencing the production, timing, and survival of sea lice larvae and observable biotic and abiotic factors that may influence them. Variable factor Parameter Origin: wild/farmed Temp. Salinity Light/photoperiod Season Reference Female size X X X a, b, c Egg string production rate X a No. of eggs X X X a, c, d, e, f, g Egg development time X h, i, j Egg development time X g Hatching period X X c, h Egg viability X X X a, c, h Hatching success X X X c, g, h Nauplius I development time X c, g, h, i, j, k Nauplius II development time X g, h, i, j, k Nauplius I width X c Nauplius I lipid reserve width X c Survival to copepodid X X c, i Copepodid survival time X X c, h, i Variable factor Parameter Origin: wild/farmed Temp. Salinity Light/photoperiod Season Reference Female size X X X a, b, c Egg string production rate X a No. of eggs X X X a, c, d, e, f, g Egg development time X h, i, j Egg development time X g Hatching period X X c, h Egg viability X X X a, c, h Hatching success X X X c, g, h Nauplius I development time X c, g, h, i, j, k Nauplius II development time X g, h, i, j, k Nauplius I width X c Nauplius I lipid reserve width X c Survival to copepodid X X c, i Copepodid survival time X X c, h, i Cells marked with an X represent areas where some data already exist and blank cells represent areas of data deficiency. References: (a) Heuch et al. (2000), (b) Tully and Whelan (1993), (c) Gravil (1996), (d) Ritchie et al. (1993), (e) Johnson and Albright (1991a), (f) Tully (1992), (g) Samsing et al. (2016), (h) Johnson and Albright (1991b), (i) Johannessen (1977), (j) Boxaspen and Næss (2000), (k) Wootten et al. (1982). Table 6. A summary table of parameters influencing the production, timing, and survival of sea lice larvae and observable biotic and abiotic factors that may influence them. Variable factor Parameter Origin: wild/farmed Temp. Salinity Light/photoperiod Season Reference Female size X X X a, b, c Egg string production rate X a No. of eggs X X X a, c, d, e, f, g Egg development time X h, i, j Egg development time X g Hatching period X X c, h Egg viability X X X a, c, h Hatching success X X X c, g, h Nauplius I development time X c, g, h, i, j, k Nauplius II development time X g, h, i, j, k Nauplius I width X c Nauplius I lipid reserve width X c Survival to copepodid X X c, i Copepodid survival time X X c, h, i Variable factor Parameter Origin: wild/farmed Temp. Salinity Light/photoperiod Season Reference Female size X X X a, b, c Egg string production rate X a No. of eggs X X X a, c, d, e, f, g Egg development time X h, i, j Egg development time X g Hatching period X X c, h Egg viability X X X a, c, h Hatching success X X X c, g, h Nauplius I development time X c, g, h, i, j, k Nauplius II development time X g, h, i, j, k Nauplius I width X c Nauplius I lipid reserve width X c Survival to copepodid X X c, i Copepodid survival time X X c, h, i Cells marked with an X represent areas where some data already exist and blank cells represent areas of data deficiency. References: (a) Heuch et al. (2000), (b) Tully and Whelan (1993), (c) Gravil (1996), (d) Ritchie et al. (1993), (e) Johnson and Albright (1991a), (f) Tully (1992), (g) Samsing et al. (2016), (h) Johnson and Albright (1991b), (i) Johannessen (1977), (j) Boxaspen and Næss (2000), (k) Wootten et al. (1982). Key gaps in knowledge identified There are a very great number of gaps in our knowledge concerning the variables affecting levels of sea louse infections. Some variables, however, are likely to have both a greater proportional/numerical impact and to be more tractable to parameterization by experimental means. These are addressed below with reference to the conceptual framework defined above. Egg production (E), egg viability, and hatching success (Ph) Previous estimates of egg production in the literature vary across more than an order of magnitude, are relatively inconsistent and are incomplete in their coverage of relevant factors. As this is the key input variable driving subsequent modelled infection levels, better estimates of production are an obvious priority. In addition to this, it is clear from the relatively sparse earlier studies that have been conducted that egg viability and hatching success are rarely, if ever, 100% and can be substantially lower than this according to a range of factors (Table 2). Egg production level is influenced by a broad range of factors including temperature (and temperature adaptation), salinity, host state (nutrition, immunity, stress, species, genotype), egg batch, and others. For this reason, it will be extremely difficult to establish realistic values through tightly controlled laboratory experiments alone. Egg production can, however, easily be established through a programme of farm sampling over a year, with counts of eggs per millimetre and the measurement of egg string lengths being conducted on-farm using a stereomicroscope or in the laboratory following sample preservation. Laboratory analysis could also employ image analysis to increase accuracy and sample throughput. During the sampling period, the recording of farm metadata, such as temperature, salinity, salmon stock, feed source, treatment regime, etc., would allow an accurate and informative predictive model to be produced. In order to give a better picture of total egg production, samples from wild salmonids would also be helpful as it is well-recognized that egg strings sourced from lice on wild fish tend to have higher numbers of eggs (Tully and Whelan, 1993; Pike and Wadsworth, 1999). Laboratory experiments could investigate controllable factors, e.g. using a range of temperatures and salinities, ideally for lice sampled from different ambient temperatures, e.g. winter, spring, and summer. The viability of eggs and hatching success are key mediators of the final number of released larvae. These parameters can be obtained by examining and hatching egg strings from challenges and/or farm samples under controlled conditions of temperature and salinity. Predation in plankton (Pp) The level of predation of larval sea lice in the plankton remains unknown. However, it is clear from other plankton studies that losses to predation are likely to be substantial. In addition, the level of predation will vary according to season, local weather conditions, and the composition of the plankton assemblage at any given time. Knowledge of predation levels will not only facilitate more accurate modelling of infection levels but could also guide coordinated treatment strategies at particular times of year. Even with good estimates of larval production, the fate of larvae in the plankton is a key mediator of numbers available to infect fish. Plankton studies are notoriously difficult and are not easily amenable to laboratory-based experiments. To achieve estimates of mortality in plankton, mesocosm studies offer the best approach, whereby in different seasons local plankton are enclosed in a mesocosm, and a known number of larval sea lice are introduced to the system. Following a period to allow for predation, the filtering of the mesocosm will allow estimations of plankton types/species present and the clearance rates of sea louse larvae. The use of molecular tools might also allow an investigation of the major predators in any given plankton sample. Using the same system with introduced “sentinel” salmonids, one could also establish the resulting infection levels, which, while not wholly realistic, would allow some estimation of both the effects of predation and also encounter rate on infection success. Infectivity profile (I) To date, there has been a tendency to equate the number of copepodids in the water column with the number of infecting individuals. From previous observations, however, it is apparent that there is a profile of infectivity, i.e. the ability of lice encountering a fish to infect it as they age, with newly moulted individuals being less infective than those having matured for 1–2 days and a subsequent decline of infectivity toward death. Infection success requires definition as not all copepodids that attach to a host may establish a successful infection; the number of copepodids developing to the chalimus I stage and developing a permanent attachment via a frontal filament may be an appropriate measure of infection success. Even under the optimal conditions of an experimental infection challenge, the infective success of maximally infective copepodids is rarely higher than 50% and is frequently lower. From the literature, few researchers have attempted to establish infection profiles for cohorts of copepodids under different conditions of, for example, temperature, salinity, and current speed, despite clear evidence that these factors will all affect infection success. Most challenging experiments employ static tanks and long exposure times, providing a totally inaccurate reflection of probabilities for real-world infection success. While the infectivity profile needs to be better established under laboratory conditions, these will not fully reflect field conditions but will tend to provide an overestimate of infection success rate. Using standard tank challenges it is possible to profile the infectivity of copepodids with age and under different temperature and salinity conditions. However, a more accurate reflection of infectivity can be achieved using flume experiments where fish are exposed to copepodids under current flow conditions more reflective of field conditions. One important source of potentially valuable data concerning losses incurred between egg hatching and the reinfection of hosts is the detailed farm louse counts already conducted in many countries. Assuming knowledge of seasonal levels of egg production and viability, which may be easily obtained, the annual profile of copepodid/chalimus counts, can, at least for some more hydrographically constrained regions, provide an indication of the proportion of hatched larvae that successfully re-establish infections on fish. Coordinated research In order to obtain the greatest benefits from modelling studies, the gaps identified need to be filled for lice and environments in all of the regions experiencing problems with L. salmonis and independently for other species, e.g. C. rogercresseyi. This means coordinating international efforts to ensure that studies are inter-comparable, and this would ideally be achieved through international agreements for matched funding by key national industry and government funders. Conclusions The estimation of lice burdens on wild and cultured fish can inform the timing of pest management decisions in salmonid aquaculture. In the life cycle of the sea louse, egg production, survival of free-swimming stages, and infectivity of survivors are key determinants of the number of lice re-establishing host infection. Despite several decades of research, however, knowledge of this area of sea louse biology is lacking, which confounds the accurate estimation of lice infections using epidemiological modelling. Even where parameters have been measured by researchers, the wide variety of data sources and experimental approaches employed, limits the possibility of providing “best” or consensus values for use in modelling. With further research of the key variables that affect the production and survival of free-swimming larval sea lice, it should be possible to more accurately model the production and dispersal of lice from cage aquaculture and wild fish, which will inform the optimum timing of pest management procedures. Furthermore, with an improved knowledge of larval sea louse mortality, it may be possible to incorporate natural processes into management decisions and to manage timing of treatments appropriately, e.g. reflecting larval predation following spring algal blooms. While many aspects of louse biology are important in determining the number of lice available for infection, care should be taken to avoid the over-parameterization of sea louse infection models. The identification of the key variables from the complex biology of sea lice that have the greatest impact on their numbers can be achieved through a sensitivity analysis of model parameters. Accurate predictions of sea lice infections are a single component of IPM protocols, and when used in conjunction with the continuous monitoring of lice populations on farmed fish and effective treatment procedures, it should be possible to minimize the environmental and economic impact of these pathogens on farmed and wild salmonids. Acknowledgements This study was funded by a Scottish Aquaculture Research Forum grant (SARF108). The authors would like to thank Dr Darren Green for guidance on mathematical modelling. References Adams T. , Black K. , MacIntyre C. , MacIntyre I. , Dean R. 2012 . Connectivity modelling and network analysis of sea lice infection in Loch Fyne, West Coast of Scotland . Aquaculture Environment Interactions , 3 : 51 – 63 . Google Scholar CrossRef Search ADS Aldrin M. , Storvik B. , Kristoffersen A. B. , Jansen P. A. 2013 . Space-time modelling of the spread of salmon lice between and within Norwegian marine salmon farms . PLoS One , 8 : e64039 – e64010 . Google Scholar CrossRef Search ADS PubMed Amundrud T. L. , Murray A. G. 2009 . Modelling sea lice dispersion under varying environmental forcing in a Scottish sea loch . Journal of Fish Diseases , 32 : 27 – 44 . Google Scholar CrossRef Search ADS PubMed Asplin L. , Boxaspen K. , Sandvik A. D. 2004 . Modelled distribution of salmon lice in a Norwegian fjord. ICES CM, 2004/P, 11. Asplin L. , Boxaspen K. K. , Sandvik A. D. 2011 . Modelling the distribution and abundance of planktonic larval stages of Lepeophtheirus salmonis in Norway. In Salmon Lice: An Integrated Approach to Understanding Parasite Abundance and Distribution , pp. 31 – 50 Ed. by Jones S. , Beamish R. . Wiley-Blackwell , Oxford . 336 pp. Asplin L. , Johnsen I. A. , Sandvik A. D. , Albretsen J. , Sundfjord V. , Aure J. , Boxaspen K. K. 2014 . Dispersion of salmon lice in the Hardangerfjord . Marine Biology Research , 10 : 216 – 225 . Google Scholar CrossRef Search ADS Bailey R. J. , Birkett M. A. , Ingvarsdóttir A. , Mordue (Luntz A. J. , Mordue W. , O’Shea B. , Pickett J. A. , et al. 2006 . The role of semiochemicals in host location and non-host avoidance by salmon louse (Lepeophtheirus salmonis) copepodids . Canadian Journal of Fisheries and Aquatic Science , 63 : 448 – 456 . Google Scholar CrossRef Search ADS Båmstedt U. , Karlson K. 1998 . Euphausiid predation on copepods in coastal waters of the Northeast Atlantic . Marine Ecology Progress Series , 172 : 149 – 168 . Google Scholar CrossRef Search ADS Bartsch A. , Robinson S. M. C. , Liutkus M. , Ang K. P. , Webb J. , Pearce C. M. 2013 . Filtration of sea louse, Lepeophtheirus salmonis, copepodids by the blue mussel, Mytilus edulis, and the Atlantic sea scallop, Placopecten magellanicus, under different flow, light and copepodid-density regimes . Journal of Fish Diseases , 36 : 361 – 370 . Google Scholar CrossRef Search ADS PubMed Bjørn P. A. , Asplin L. , Karlsen Ø. , Finstad B. , Taranger G. L. 2014 . Utvikler modelsystem for få kontroll på Lusa. (English translation: Development of a system of models to control lice). In Fisken og Havet, særnummer 1–2014. Havforskningsrapporten 2014. Boenigk J. , Arndt H. 2002 . Bacterivory by heterotrophic flagellates: community structure and feeding strategies . Antonie van Leeuwenhoek , 81 : 465 – 480 . Google Scholar CrossRef Search ADS PubMed Boxaspen K. , Næss T. 2000 . Development of eggs and the planktonic stages of salmon lice (Lepeophtheirus salmonis) at low temperatures . Contributions to Zoology , 69 : 51 – 56 . Braden L. M. , Sutherland B. J. , Koop B. F. , Jones S. R. 2017 . Enhanced transcriptomic responses in the Pacific salmon louse Lepeophtheirus salmonis oncorhynchi to the non-native Atlantic Salmon Salmo salar suggests increased parasite fitness . BMC Genomics , 18 : 1. Google Scholar CrossRef Search ADS PubMed Bricknell I. R. , Dalesman S. J. , Shea B. O. , Pert C. C. , Luntz A. J. M. 2006 . Effect of environmental salinity on sea lice Lepeophtheirus salmonis settlement success . Diseases of Aquatic Organisms , 71 : 201 – 212 . Google Scholar CrossRef Search ADS PubMed Bron J. E. 1993 . A study of the biology and behaviour of the copepodid larva of the salmon louse Lepeophtheirus salmonis (Krøyer, 1837) (Copepoda; Caligidae). PhD thesis, University of Stirling, UK. Bron J. E. , Sommerville C. , Jones M. , Rae G. H. 1991 . The settlement and attachment of early stages of the salmon louse, Lepeophtheirus salmonis (Copepoda: Caligidae) on the salmon host, Salmo salar . Journal of Zoology , 224 : 201 – 212 . Google Scholar CrossRef Search ADS Bron J. E. , Sommerville C. , Rae G. H. 1993 . Aspects of the behaviour of copepodid larvae of the salmon louse. In Pathogens of Wild and Farmed Fish: Sea Lice , 2nd edn , pp. 125 – 142 . Ed. by Boxshall G. A. , Defaye D. . Ellis Horwood Ltd ., New York . 378 pp. Brooks K. M. 2005 . The effects of water temperature, salinity, and currents on the survival and distribution of the infective copepodid stage of sea lice (Lepeophtheirus salmonis) originating on Atlantic salmon farms in the Broughton Archipelago of British Columbia, Canada . Reviews in Fisheries Science , 13 : 177 – 204 . Google Scholar CrossRef Search ADS Brooks J. L. , Dodson S. I. 1965 . Predation, body size, and composition of plankton . Science , 150 : 28 – 35 . Google Scholar CrossRef Search ADS PubMed Carr J. , Whoriskey F. 2004 . Sea lice infestation rates on wild and escaped farmed Atlantic salmon (Salmo salar L.) entering the Magaguadavic River, New Brunswick . Aquaculture Research , 35 : 723 – 729 . Google Scholar CrossRef Search ADS Cavaleiro F. I. , Santos M. J. 2014 . Egg number-egg size: an important trade-off in parasite life history strategies . International Journal for Parasitology , 44 : 173 – 182 . Google Scholar CrossRef Search ADS PubMed Colavecchia M. , Katopodis C. , Goosney R. , Scruton D. A. , McKinley R. S. 1998 . Measurement of burst swimming performance in wild Atlantic salmon (Salmo salar) using digital telemetry . Regulated Rivers: Research and Management , 14 : 41 – 51 . Google Scholar CrossRef Search ADS Cook P. F. , McBeath S. J. , Bricknell I. R. , Bron J. E. 2010 . Determining the age of individual Lepeophtheirus salmonis (Krøyer, 1837) copepodids by measuring stored lipid volume; proof of principle . Journal of Microscopy , 240 : 83 – 86 . Google Scholar CrossRef Search ADS PubMed Corner R. A. , Brooker A. J. , Telfer T. C. , Ross L. G. 2006 . A fully integrated GIS-based model of particulate waste distribution from marine fish-cage sites . Aquaculture , 258 : 299 – 311 . Google Scholar CrossRef Search ADS Costello M. J. , Burridge L. , Chang B. , Robichaud L. 2004 . Sea Lice 2003 - Proceedings of the sixth international conference on sea lice biology and control . Aquaculture Research , 35 : 711 – 712 . Google Scholar CrossRef Search ADS Costelloe M. , Costelloe J. , Coghlan N. , Donohoe G. O. , Connor B. O. 1998b . Distribution of the larval stages of Lepeophtheirus salmonis in three bays on the West Coast of Ireland . ICES Journal of Marine Science , 55 : 181 – 187 . Google Scholar CrossRef Search ADS Costelloe M. , Costelloe J. , O’Donohoe G. , Coghlan N. J. , Oonk M. , Van Der Heijden Y. 1998a . Planktonic distribution of sea lice larvae, Lepeophtheirus salmonis, in Killary Harbour, West Coast of Ireland . Journal of the Marine Biological Association of the UK , 78 : 853 – 874 . Google Scholar CrossRef Search ADS Cripps G. C. , Atkinson A. 2000 . Fatty acid composition as an indicator of carnivory in Antarctic krill, Euphausia superba . Canadian Journal of Fisheries and Aquatic Sciences , 57 : 31 – 37 . Google Scholar CrossRef Search ADS Daewel U. , Hjøllo S. S. , Huret M. , Ji R. , Maar M. , Niiranen S. , Wolfshaar K. E. V. D. 2014 . Predation control of zooplankton dynamics: a review of observations and models . ICES Journal of Marine Science , 71 : 254 – 271 . Google Scholar CrossRef Search ADS Davenport J. , Smith R. , Packer M. 2000 . Mussels Mytilus edulis: significant consumers and destroyers of mesoplankton . Marine Ecology Progress Series , 198 : 131 – 137 . Google Scholar CrossRef Search ADS Devine G. J. , Ottir A. I. , Mordue W. , Pike A. W. , Pickett J. , Duce I. A. N. , Luntz A. J. M. 2000 . Salmon lice, Lepeophtheirus salmonis, exhibit specific chemotactic responses to semiochemicals originating from the salmonid, Salmo salar . Journal of Chemical Ecology , 26 : 1833 – 1847 . Google Scholar CrossRef Search ADS Doall M. H. , Colin S. P. , Strickler J. R. , Yen J. 1998 . Locating a mate in 3D: the case of Temora longicornis . Philosophical Transactions of the Royal Society B: Biological Sciences , 353 : 681 – 689 . Google Scholar CrossRef Search ADS Doyle T. K. , Houghton J. D. R. , Buckley S. M. , Hays G. C. , Davenport J. 2007 . The broad-scale distribution of five jellyfish species across a temperate coastal environment . Hydrobiologia , 579 : 29 – 39 . Google Scholar CrossRef Search ADS Fast M. D. 2014 . Fish immune responses to parasitic copepod (namely sea lice) infection . Developmental and Comparative Immunology , 43 : 300 – 312 . Google Scholar CrossRef Search ADS PubMed Fast M. D. , Ross N. W. , Mustafa A. , Sims D. E. , Johnson S. C. , Conboy G. A. , Speare D. J. , et al. 2002 . Susceptibility of rainbow trout Oncorhynchus mykiss, Atlantic salmon Salmo salar and coho salmon Oncorhynchus kisutch to experimental infection with sea lice Lepeophtheirus salmonis . Diseases of Aquatic Organisms , 52 : 57 – 68 . Google Scholar CrossRef Search ADS PubMed Fields D. M. , Skiftesvik A. B. , Browman H. I. 2017 . Behavioural responses of infective-stage copepodids of the salmon louse (Lepeophtheirus salmonis, Copepoda: Caligidae) to host-related sensory cues . Journal of Fish Diseases , 00 : 1 – 10 . Finstad B. , Grimnes A. , Hvidsten N. A. 2000 . Laboratory and field investigations of salmon lice [Lepeophtheirus salmonis (Kroyer)] infestation on Atlantic salmon (Salmo salar L.) post-smolts . Aquaculture Research , 1 : 795 – 803 . Google Scholar CrossRef Search ADS Forward R. J. , Douglass J. 1986 . Crustacean larval visual sensitivity during diel vertical migration. In Proceedings of the 21st European Marine Biology Symposium, Gdansk, 14 September 1986, pp. 59–66. Fowler G. H. 1911 . Ctenophora. In Encyclopaedia Britannica . 7 , 11th edn , p. 593 . Ed. By Chisholm H. . Cambridge University Press , Cambridge, UK. Frenzl B. 2014 . Understanding key factors associated with the infection of farmed Atlantic Salmon by the salmon louse, Lepeophtheirus salmonis. PhD thesis, University of Stirling, UK. Frost B. W. 1972 . Effects of size and concentration of food particles on the feeding behaviour of the marine planktonic copepod Calanus pacificus . Limnology and Oceanography , 17 : 805 – 815 . Google Scholar CrossRef Search ADS Fulton R. S. 1984 . Effects of chaetognath predation and nutrient enrichment on enclosed estuarine copepod communities . Oecologia , 62 : 97 – 101 . Google Scholar CrossRef Search ADS PubMed Genna R. L. 2002 . The chemical ecology, physiology and infection dynamics of the sea louse copepodid, Lepeophtheirus salmonis Kroyer.PhD thesis, University of Aberdeen, UK. Gharbi K. , Matthews L. , Bron J. , Roberts R. , Tinch A. , Stear M. 2015 . The control of sea lice in Atlantic salmon by selective breeding . Journal of the Royal Society Interface , 12 : 20150574. Google Scholar CrossRef Search ADS Gravil H. R. 1996 . Studies on the biology and ecology of the free swimming larval stages of Lepeophtheirus salmonis (Kroyer,1838) and Caligus elongatus Nordmann, 1832 (Copepoda: Caligidae). PhD thesis, University of Stirling, UK. Grayson T. H. , John R. J. , Wadsworth S. , Greaves K. , Cox D. , Roper J. , Wrathmell A. B. , et al. 1995 . Immunization of Atlantic salmon against the salmon louse: identification of antigens and effects on louse fecundity . Journal of Fish Biology , 47 : 85 – 94 . Google Scholar CrossRef Search ADS Greene C. H. , Landry M. R. , Monger B. C. 1986 . Foraging behavior and prey selection by the ambush entangling predator Pleurobrachia bachei . Ecology , 67 : 1493 – 1501 . Google Scholar CrossRef Search ADS Hahnenkamp L. , Fyhn H. J. 1985 . The osmotic response of salmon louse, Lepeophtheirus salmonis (Copepoda: Caligidae), during the transition from sea water to fresh water . Journal of Comparative Physiology B , 155 : 357 – 365 . Google Scholar CrossRef Search ADS Hansen B. , Bjornsen P. K. , Hansen P. J. 1994 . The size ratio between planktonic predators and their prey . Limnology and Oceanography , 39 : 395 – 403 . Google Scholar CrossRef Search ADS Haury L. , Weihs D. 1976 . Energetically efficient swimming behaviour of negatively buoyant zooplankton . Limnology and Oceanography , 21 : 797 – 803 . Google Scholar CrossRef Search ADS Heath M. R. 2007 . The consumption of zooplankton by early life stages of fish in the North Sea . ICES Journal of Marine Science , 64 : 1650 – 1663 . Google Scholar CrossRef Search ADS Heuch P. A. 1995 . Experimental evidence for aggregation of salmon louse copepodids (Lepeophtheirus salmonis) in step salinity gradients . Journal of the Marine Biological Association of the UK , 75 : 927 – 939 . Google Scholar CrossRef Search ADS Heuch P. A. , Doall M. H. , Yen J. 2007 . Water flow around a fish mimic attracts a parasitic and deters a planktonic copepod . Journal of Plankton Research , 29(Suppl. 1) : i3 – i16 . Google Scholar CrossRef Search ADS Heuch P. A. , Karlsen E. 1997 . Detection of infrasonic water oscillations by copepodids of Lepeophtheirus salmonis (Copepoda Caligida) . Journal of Plankton Research , 19 : 735 – 747 . Google Scholar CrossRef Search ADS Heuch P. A. , Mo T. A. 2001 . A model of salmon louse production in Norway: effects of increasing salmon production and public management measures . Diseases of Aquatic Organisms , 45 : 145 – 152 . Google Scholar CrossRef Search ADS PubMed Heuch P. A. , Nordhagen J. R. , Schram T. A. 2000 . Egg production in the salmon louse (Lepeophtheirus salmonis (Kroyer)) in relation to origin and water temperature . Aquaculture Research , 31 : 805 – 814 . Google Scholar CrossRef Search ADS Heuch P. A. , Parsons A. , Boxaspen K. 1995 . Diel vertical migration: a possible host-finding mechanism in salmon louse (Lepeophtheirus salmonis) copepodids? Canadian Journal of Fisheries and Aquatic Science , 52 : 681 – 689 . Google Scholar CrossRef Search ADS Hirst A. , Kiørboe T. 2002 . Mortality of marine planktonic copepods: global rates and patterns . Marine Ecology Progress Series , 230 : 195 – 209 . Google Scholar CrossRef Search ADS Hogans W. E. , Trudeau D. J. 1989 . Preliminary studies on the biology of sea lice, Caligus elongatus, Caligus curtus and Lepeophthierus salmonis (Copepoda: Caligoida) parasitic on cage-cultured salmonids in the lower Bay of Fundy . Canadian Technical Report of Fisheries and Aquatic Science , 1715 : iv+14. pp. Holm H. , Santi N. , Kjøglum S. , Perisic N. , Skugor S. , Evensen Ø. 2015 . Difference in skin immune responses to infection with salmon louse (Lepeophtheirus salmonis) in Atlantic salmon (Salmo salar L.) of families selected for resistance and susceptibility . Fish and Shellfish Immunology , 42 : 384 – 394 . Google Scholar CrossRef Search ADS PubMed Ingvarsdottir A. , Birkett M. A. , Duce I. , Genna R. L. , Mordue W. , Pickett J. A. , Wadhams L. J. , et al. 2002 . Semiochemical strategies for sea louse control: host location cues . Pest Management Science , 58 : 537 – 545 . Google Scholar CrossRef Search ADS PubMed Iversen A. , Hermansen Ø. , Andreassen O. , Brandvik R. K. , Marthinussen A. , Nystøyl R. 2015 . Kostnadsdrivere i lakseoppdrett (Cost Drivers in Salmon Farming). No. 41/2015, 1–43. Nofima [in Norwegian]. Janssen J. 1976 . Feeding modes and prey size selection in the Alewife (Alosa pseudoharengus) . Journal of the Fisheries Research Board of Canada , 33 : 1972 – 1975 . Google Scholar CrossRef Search ADS Jansen P. A. , Kristoffersen A. B. , Viljugrein H. , Jimenez D. , Aldrin M. , Stien A. 2012 . Sea lice as a density-dependent constraint to salmonid farming . Proceedings of the Royal Society B , 279 : 2330 – 2338 . Google Scholar CrossRef Search ADS PubMed Jensen L. , Provan F. , Larssen E. , Bron J. , Obach A. 2014 . Reducing sea lice (Lepeophtheirus salmonis) infestation of farmed Atlantic salmon (Salmo salar L.) through functional feeds . Aquaculture Nutrition , 21 : Jodaa Holm H. , Wadsworth S. , Bjelland A. K. , Krasnov A. , Evensen Ø. , Skugor S. 2016 . Dietary phytochemicals modulate skin gene expression profiles and result in reduced lice counts after experimental infection in Atlantic salmon . Parasites and Vectors , 9 : 271. Google Scholar CrossRef Search ADS PubMed Johannessen A. 1975 . Salmon louse, Lepeophtheirus salmonis, Kroyer (Copepoda, Caligidae). Independent larval stages, growth and infection in salmon(Salmo salar L.) from breeding plants and commercial catches in west Norwegian waters 1973–1974. Thesis in Fish Biology, University of Bergen, Norway. Johannessen A. 1977 . Early stages of Lepeophtheirus salmonis (Copepoda, Caligidae) . Sarsia , 63 : 169 – 176 . Google Scholar CrossRef Search ADS Johnsen I. , Fiksen Ø. , Sandvik A. D. , Asplin L. 2014 . Vertical salmon louse behaviour as a response to environmental conditions and its influence on regional dispersion in a fjord system . Aquaculture Environment Interactions , 5 : 127 – 141 . Google Scholar CrossRef Search ADS Johnson S. , Albright L. 1991a . The developmental stages of Lepeophtheirus salmonis (Kroyer, 1837) (Copepoda: Caligidae) . Canadian Journal of Zoology , 69 : 929 – 950 . Google Scholar CrossRef Search ADS Johnson S. C. , Albright L. J. 1991b . Development, growth, and survival of Lepeophtheirus salmonis (Copepoda: Caligidae) under laboratory conditions . Canadian Journal of Zoology , 69 : 929 – 950 . Google Scholar CrossRef Search ADS Johnson S. C. , Albright L. J. 1992 . Comparative susceptibility and histopathology of the response of naïve Atlantic, Chinook and coho salmon to experimental infection with Lepeophtheirus salmonis (Copepoda: Caligidae) . Diseases of Aquatic Organisms , 14 : 179 – 193 . Google Scholar CrossRef Search ADS Johnston C. E. , Dykeman D. 1987 . Observations on body proportions and egg production in the female parasitic copepod (Salmincola salmoneus) from the gills of Atlantic salmon (Salmo salar) kelts exposed to different temperatures and photoperiods . Canadian Journal of Zoology , 65 : 415 – 419 . Google Scholar CrossRef Search ADS Jones S. R. M. 2011 . Mechanisms of resistance among salmon to the parasitic copepod Lepeophtheirus salmonis . Journal of Aquaculture Research and Development , S2 : 003 . Jonsson N. , Jonsson B. , Hansen L. P. 1990 . Partial segregation in the timing of migration of Atlantic salmon of different ages . Animal Behaviour , 40 : 313 – 321 . Google Scholar CrossRef Search ADS Kaiser M. J. (Ed.). 2005 . Marine Ecology: Processes, Systems, and Impacts . Oxford University Press , Oxford, UK. Kiørboe T. 2011 . How zooplankton feed: mechanisms, traits and trade-offs . Biological Reviews , 86 : 311 – 339 . Google Scholar CrossRef Search ADS PubMed Kiørboe T. , Saiz E. 1995 . Planktivorous feeding in calm and turbulent environments, with emphasis on copepods . Marine Ecology Progress Series , 122 : 135 – 145 . Google Scholar CrossRef Search ADS Koop K. , Field J. G. 1980 . The influence of food availability on population dynamics of a supralittoral isopod, Ligia dilatata brandt . Journal of Experimental Marine Biology and Ecology , 48 : 61 – 72 . Google Scholar CrossRef Search ADS Kristoffersen A. B. , Jimenez D. , Viljugrein H. , Grøntvedt R. , Stien A. , Jansen P. A. 2014 . Large scale modelling of salmon lice (Lepeophtheirus salmonis) infection pressure based on lice monitoring data from Norwegian salmonid farms . Epidemics , 9 : 31 – 39 . Google Scholar CrossRef Search ADS PubMed Krkošek M. , Lewis M. A. , Volpe J. P. 2005 . Transmission dynamics of parasitic sea lice from farm to wild salmon . Proceedings of the Royal Society B , 272 : 689 – 696 . Google Scholar CrossRef Search ADS PubMed Leclercq E. , Davie A. , Migaud H. 2013 . Delousing efficiency of farmed ballan wrasse (Labrus bergylta) against Lepeophtheirus salmonis infecting Atlantic salmon (Salmo salar) post-smolts . Pest Management Science , 70 : 1274 – 1282 . Google Scholar CrossRef Search ADS Ljungfeldt L. E. R. , Espedal P. G. , Nilsen F. , Skern-Mauritzen M. , Glover K. A. 2014 . A common-garden experiment to quantify evolutionary processes in copepods: the case of emamectin benzoate resistance in the parasitic sea louse Lepeophtheirus salmonis . BMC Evolutionary Biology , 14 : 108. Google Scholar CrossRef Search ADS PubMed Lyse A. A. , Stefansson S. O. , Ferno A. 1998 . Behaviour and diet of sea trout post-smolts in a Norwegian fjord system . Journal of Fish Biology , 52 : 923 – 936 . Google Scholar CrossRef Search ADS MacKinnon B. M. 1998 . Host factors important in sea lice infections . ICES Journal of Marine Science , 55 : 188 – 192 . Google Scholar CrossRef Search ADS MacKinnon B. M. , Bailey J. K. , O’Flynn F. M. , Friars G. W. 1995 . Can selection improve resistance? Northern Aquaculture , 11 : 3 . Maranhão P. , Marques J. C. 2003 . The influence of temperature and salinity on the duration of embryonic development, fecundity and growth of the amphipod Echinogammarus marinus Leach (Gammaridae) . Acta Oecologica , 24 : 5 – 13 . Google Scholar CrossRef Search ADS McKibben M. A. , Hay D. W. 2004 . Distributions of planktonic sea lice larvae Lepeophtheirus salmonis in the inter-tidal zone in Loch Torridon, Western Scotland in relation to salmon farm production cycles . Aquaculture Research , 35 : 742 – 750 . Google Scholar CrossRef Search ADS Molloy S. D. , Pietrak M. R. , Bouchard D. A. , Bricknell I. 2011 . Ingestion of Lepeophtheirus salmonis by the blue mussel Mytilus edulis . Aquaculture , 311 : 61 – 64 . Google Scholar CrossRef Search ADS Mordue Luntz A. J. , Birkett M. A. 2009 . A review of host finding behaviour in the parasitic sea louse, Lepeophtheirus salmonis (Caligidae: Copepoda) . Journal of Fish Diseases , 32 : 3 – 13 . Google Scholar CrossRef Search ADS PubMed Munk P. , Nielsen T. G. , Castle C. 1994 . Trophodynamics of the plankton community at Dogger Bank: predatory impact by larval fish . Journal of Plankton Research , 16 : 1225 – 1245 . Google Scholar CrossRef Search ADS Murray A. G. 2002 . Using observed load distributions with a simple model to analyse the epidemiology of sea lice (Lepeophtheirus salmonis) on sea trout (Salmo trutta) . Pest Management Science , 58 : 585 – 594 . Google Scholar CrossRef Search ADS PubMed Murray A. G. , Amundrud T. L. 2007 . Using coupled biophysical – particle tracking models of sea lice in Loch Torridon. In Proceedings of the International Congress on Modelling and Simulation, Christchurch, New Zealand, December 2007, pp. 2848–2853. Nilsen F. 2016 . Sea lice vaccine, pp. 1–20. Presentation available online http://www.fhf.no/media/127279/10_-_vaksine_-_grunnleggende_kunnskap_-_nilsen_-_slrc-uib.pdf (lass accessed 29 December 2017). Núñez-Acuña G. , Marambio J. P. , Valenzuela T. , Wadsworth S. , Gallardo-Escárate C. 2016 . Antimicrobial peptides from Salmon salar skin induce frontal filament development and olfactory/cuticle-related genes in the sea louse Caligus rogercresseyi . Aquaculture , 464 : 171 – 177 . Google Scholar CrossRef Search ADS Øvergård A. , Hamre L. A. , Harasimczuk E. , Dalvin S. , Nilsen F. , Grotmol S. 2016 . Exocrine glands of Lepeophtheirus salmonis (Copepoda: Caligidae): distribution, developmental appearance, and site of secretion . Journal of Morphology , 277 : 1097 – 4687 . Google Scholar CrossRef Search ADS Pedersen T. M. 2009 . Instar growth and development of Leophtheirus salmonis copepodids.MSc thesis, University of Bergen, Norway. Pike A. , Wadsworth S. 1999 . Sea lice on salmonids: their biology and control . Advances in Parasitology , 44 : 233 – 337 . Google Scholar CrossRef Search ADS PubMed Pike A. W. , Mordue (Luntz A. J. , Ritchie G. 1993 . The development of Caligus elongatus Nordmann from hatching to copepodid in relation to temperature. In Pathogens of Wild and Farmed Fish: Sea Lice , 2nd edn , pp. 51 – 60 . Ed. by Boxshall G. A. , Defaye D. . Ellis Horwood Ltd ., New York . 378 pp. Purcell J. E. 1992 . Effects of predation by the scyphomedusan Chrysaora quinquecirrha on zooplankton . Marine Ecology Progress Series , 87 : 65 – 76 . Google Scholar CrossRef Search ADS Purcell J. E. , Nemazie D. A. , Dorsey S. E. , Houde E. D. , Gamble J. C. 1994 . Predation mortality of bay anchovy Anchoa mitchilli eggs and larvae due to scyphomedusae and ctenophores in Chesapeake Bay . Marine Ecology Progress Series , 114 : 47 – 58 . Google Scholar CrossRef Search ADS Reeve M. R. , Walter M. A. , Ikeda T. 1978 . Laboratory studies of ingestion and food utilization in lobate and tentaculate ctenophores . Limnology and Oceanography , 23 : 740 – 751 . Google Scholar CrossRef Search ADS Ritchie G. , Mordue A. J. , Pike A. W. , Rae G. H. 1993 . The reproductive output of Lepeophtheirus salmonis adult females in relation to seasonal variability of temperature and photoperiod. In Pathogens of Wild and Farmed Fish: Sea Lice , 2nd edn , pp. 153 – 165 . Ed. by Boxshall G. A. , Defaye D. . Ellis Horwood Ltd ., New York . 378 pp. Rødseth G. I. 2016 . Beregninger på fem milliarder i lusekostnader er bare deler av sannheten. iLaks.no. Retrieved from https://ilaks.no/beregninger-pa-fem-milliarder-i-lusekostnader-er-bare-deler-av-sannheten/ (last accessed 10 June 2017) [in Norwegian]. Saiz E. , Kiørboe T. 1995 . Predatory and suspension feeding of the copepod Acartia tonsa in turbulent environments . Marine Ecology Progress Series , 122 : 147 – 158 . Google Scholar CrossRef Search ADS Salama N. K. G. , Murray A. G. , Rabe B. 2011 . Modelling dispersal of salmon lice in a large fjordic system: Loch Linnhe, Scotland. In 19th International Congress on Modelling and Simulation, Perth, Australia, pp. 12–16. Samsing F. , Oppedal F. , Dalvin S. , Johnsen I. , Vågseth T. , Dempster T. 2016 . Salmon lice (Lepeophtheirus salmonis) development times, body size and reproductive outputs follow universal models of temperature dependence . Canadian Journal of Fisheries and Aquatic Science , 73 : 1841 – 1851 . Google Scholar CrossRef Search ADS Sandvik A. , Bjørn P. , Ådlandsvik B. , Asplin L. , Skarðhamar J. , Johnsen I. , Myksvoll M. , et al. 2016 . Toward a model-based prediction system for salmon lice infestation pressure . Aquaculture Environment Interactions , 8 : 527 – 542 . Google Scholar CrossRef Search ADS Sandvik A. D. , Skarðhamar J. , Bjørn P. A. , Asplin L. , Llinares S. R. M. , Johnsen I. A. 2014 . The predictability of sea lice density in Norwegian Fjords. In Sea Lice 2014: Sustainable Solutions in Sea Lice Management , 31 August–5 September, Portland, Maine , p. 46 . Schram T. A. 1993 . Supplementary descriptions of the developmental stages of Lepeophtheirus salmonis (Kroyer, 1837) (Copepoda: Caligidae). In Pathogens of Wild and Farmed Fish: Sea Lice , 2nd edn , pp. 30 – 47 . Ed. by Boxshall G. A. , Defaye D. . Ellis Horwood Ltd ., New York . 378 pp. Skern-Mauritzen R. , Torrissen O. , Glover K. A. 2014 . Pacific and Atlantic Lepeophtheirus salmonis (Krøyer, 1838) are allopatric subspecies: Lepeophtheirus salmonis salmonis and L. salmonis oncorhynchi subspecies novo . BMC Genetics , 15 : 32. Google Scholar CrossRef Search ADS PubMed Skiftesvik A. B. , Bjelland R. , Durif C. M. F. , Johansen I. , Browman H. I. 2013 . Delousing of Atlantic Salmon (Salmo salar) by cultured vs. wild ballan wrasse (Labrus bergylta) . Aquaculture , 402 : 113 – 118 . Google Scholar CrossRef Search ADS Skugor S. , Glover K. A. , Nilsen F. , Krasnov A. 2008 . Local and systemic gene expression responses of Atlantic salmon (Salmo salar L.) to infection with the salmon louse (Lepeophtheirus salmonis) . BMC Genomics , 9 : 498 – 418 . Google Scholar CrossRef Search ADS PubMed Smith G. W. , Smith I. P. , Armstrong S. M. 1994 . The relationship between river flow and entry to the Aberdeenshire Dee by returning adult Atlantic salmon . Journal of Fish Biology , 45 : 953 – 960 . Google Scholar CrossRef Search ADS Sommer U. , Stibor H. 2002 . Copepoda—Cladocera—Tunicata: the role of three major mesozooplankton groups in pelagic food webs . Ecological Research , 17 : 161 – 174 . Google Scholar CrossRef Search ADS Steele J. H. , Frost B. W. 1977 . The structure of plankton communities . Proceedings of the Royal Society B , 280 : 485 – 534 . Stucchi D. J. , Guo M. , Foreman M. G. G. , Czajko P. , Galbraith M. , Mackas D. L. , Gillibrand P. A. 2011 . Modeling sea lice production and concentrations in the Broughton Archipelago, British Columbia. In Salmon Lice: An Integrated Approach to Understanding Parasite Abundance and Distribution , pp. 117 – 150 . Ed. by Jones S. , Beamish R. . Wiley-Blackwell , Oxford, UK . 336 pp. Google Scholar CrossRef Search ADS Suchman C. L. , Sullivan B. K. 1998 . Vulnerability of the copepod Acartia tonsa to predation by the scyphomedusa Chrysaora quinquecirrha: effect of prey size and behavior . Marine Biology , 132 : 237 – 245 . Google Scholar CrossRef Search ADS Suchman C. L. , Sullivan B. K. 2000 . Effect of prey size on vulnerability of copepods to predation by the scyphomedusae Aurelia aurita and Cyanea sp . Journal of Plankton Research , 22 : 2289 – 2306 . Google Scholar CrossRef Search ADS Sutherland B. J. , Covello J. M. , Friend S. E. , Poley J. D. , Koczka K. W. , Purcell S. L. , MacLeod T. L. , et al. 2017 . Host parasite transcriptomics during immunostimulant-enhanced rejection of salmon lice (Lepeophtheirus salmonis) by Atlantic salmon (Salmo salar) . FACETS , 2 : 477 – 495 . Google Scholar CrossRef Search ADS Tadiso T. M. , Krasnov A. , Skugor S. , Afanasyev S. , Hordvik I. , Nilsen F. 2011 . Gene expression analyses of immune responses in Atlantic salmon during early stages of infection by salmon louse (Lepeophtheirus salmonis) revealed bi-phasic responses coinciding with the copepod-chalimus transition . BMC Genomics , 12 : 141. Google Scholar CrossRef Search ADS PubMed Tiselius P. , Jonsson P. R. 1990 . Foraging behaviour of six calanoid copepods: observations and hydrodynamic analysis . Marine Ecology Progress Series , 1–2 : 23 – 33 . Google Scholar CrossRef Search ADS Torres G. , Gime L. , Anger K. 2002 . Effects of reduced salinity on the biochemical composition (lipid, protein) of zoea 1 decapod crustacean larvae . Journal of Experimental Marine Biology and Ecology , 277 : 43 – 60 . Google Scholar CrossRef Search ADS Tucker C. S. , Norman R. , Shinn A. P. , Bron J. E. , Sommerville C. , Wootten R. 2002 . A single cohort time delay model of the life-cycle of the salmon louse Lepeophtheirus salmonis on Atlantic salmon Salmo salar . Fish Pathology , 37 : 107 – 118 . Google Scholar CrossRef Search ADS Tucker C. , Sommerville C. , Wootten R. 2000a . An investigation into the larvel energetics and settlement of the sea louse, Lepeophtheirus salmonis, an ecotoparasite copepod of Atlantic salmon, Salmo salar . Fish Pathology , 35 : 137 – 143 . Google Scholar CrossRef Search ADS Tucker C. S. , Sommerville C. , Wootten R. 2000b . The effect of temperature and salinity on the settlement and survival of copepodids of Lepeophtheirus salmonis (Krøyer, 1837) on Atlantic salmon, Salmo salar L . Journal of Fish Diseases , 23 : 309 – 320 . Google Scholar CrossRef Search ADS Tully O. 1989 . The succession of generations and growth of the caligid copepods Caligus Elongatus and Lepeophtheirus Salmonis parasitising farmed atlantic salmon smolts (Salmo Salar L.) . Journal of the Marine Biological Association UK , 69 : 279 – 287 . Google Scholar CrossRef Search ADS Tully O. 1992 . Predicting infestation parameters and impacts of caligid copepods in wild and cultured fish populations . Invertebrate Reproduction and Development , 22 : 91 – 102 . Google Scholar CrossRef Search ADS Tully O. , Whelan K. F. 1993 . Production of nauplii of Lepeophtheirus salmonis (Kroyer) (Copepoda: Caligidae) from farmed and wild salmon and its relation to the infestation of wild sea trout (Salmo trutta L.) off the West Coast of Ireland in 1991 . Fisheries Research , 17 : 187 – 200 . Google Scholar CrossRef Search ADS Ugelvik M. S. , Skorping A. , Mennerat A. 2017 . Parasite fecundity decreases with increasing parasite load in the salmon louse Lepeophtheirus salmonis infecting Atlantic salmon Salmo salar . Journal of Fish Diseases , 40 : 671 – 678 . Google Scholar CrossRef Search ADS PubMed Visser A. W. , Fiksen Ø. 2013 . Optimal foraging in marine ecosystem models: selectivity, profitability and switching . Marine Ecology Progress Series , 473 : 91 – 101 . Google Scholar CrossRef Search ADS Visser A. W. , Mariani P. , Pigolotti S. 2009 . Swimming in turbulence: zooplankton fitness in terms of foraging efficiency and predation risk . Journal of Plankton Research , 31 : 121 – 133 . Google Scholar CrossRef Search ADS Williams J. A. 1985 . The role of photoperiod in the initiation of breeding and brood development in the amphipod Talitrus saltator Montagu . Journal of Experimental Marine Biology and Ecology , 86 : 59 – 72 . Google Scholar CrossRef Search ADS Wirtz K. W. 2011 . Non-uniform scaling in phytoplankton growth rate due to intracellular light and CO2 decline . Journal of Plankton Research , 33 : 1325 – 1341 . Google Scholar CrossRef Search ADS Wirtz K. W. 2012 . Who is eating whom? Morphology and feeding type determine the size relation between planktonic predators and their ideal prey . Marine Ecology Progress Series , 445 : 1 – 12 . Google Scholar CrossRef Search ADS Wirtz K. W. 2014 . A biomechanical and optimality-based derivation of prey-size dependencies in planktonic prey selection and ingestion rates . Marine Ecology Progress Series , 507 : 81 – 94 . Google Scholar CrossRef Search ADS Woodward G. , Speirs D. C. , Hildrew A. G. 2005 . Food webs: from connectivity to energetics . Advances in Ecological Research , 36 : 85 – 135 . Google Scholar CrossRef Search ADS Wootten R. , Smith J. W. , Needham E. A. 1982 . Aspects of the biology of the parasitic copepods Lepeophtheirus salmonis and Caligus elongatus on farmed salmonids, and their treatment . Proceedings of the Royal Society of Edinburgh B , 81 : 185 – 197 . Young S. , Watt P. 1993 . Behavioral mechanisms controlling vertical migration in Daphnia . Limnology and Oceanography , 38 : 70 – 79 . Google Scholar CrossRef Search ADS Zarowiecki M. , Berriman M. 2015 . What helminth genomes have taught us about parasite evolution . Parasitology , 142 : S85 – S97 . Google Scholar CrossRef Search ADS PubMed © International Council for the Exploration of the Sea 2018. All rights reserved. For Permissions, please email: [email protected] This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

doi: 10.1093/icesjms/fsy017pmid: N/A

Abstract Future sustainable management of fisheries will require resilience to the effects of environmental variability and climate change on stock productivity. In this study, we examined relationships between sea surface temperature (SST) in the region between Taiwan and the Sea of Japan, and annual recruitment of Pacific bluefin tuna (Thunnus orientalis: PBF) over the past 35 years. Spatial correlation maps showed that warmer SSTs south of Shikoku, in the East China Sea and in the Sea of Japan from summer to late fall were associated with above average recruitment. SST anomalies near larval and juvenile habitats were most strongly correlated with local air temperatures. Generalized Additive Models predicting annual PBF recruitment from SST fields suggested that the influence of SST on recruitment was stronger than that of spawning stock biomass. Correlations between SST and recruitment likely reflect biological processes relevant to early juvenile habitat suitability. The influence of late fall SSTs could also be a result of varying availability of age-0 fish to the troll fishery; however, the relative importance of these processes was not clear. Despite these knowledge gaps, the strong predictive power of SST on PBF recruitment can allow more proactive management of this species under varying environmental conditions. Introduction Although the stock–recruitment relationship (SRR) is a central principle of fisheries management (Hilborn and Walters, 1992), it is widely acknowledged that environmental factors can and do drive substantial variability in recruitment. In many species, recruitment appears to vary largely independent of spawning stock biomass (SSB; Szuwalski et al., 2015; Lowerre-Barbieri et al., 2017), or to change its relationship with spawning biomass over time (Megrey et al., 2005; Britten et al., 2016). Climate change is contributing to largely unprecedented conditions in some marine ecosystems (Hoegh-Guldberg and Bruno, 2010), and further directional change is likely in the future. Fisheries management processes must therefore become increasingly able to adapt to environmentally driven shifts in stock productivity and distribution (Grafton, 2010; Koehn et al., 2011). This can be achieved through several strategies, including environmentally informed reference points and indices, and development of broader ecosystem-based management frameworks (Punt et al., 2013; DePiper et al., 2017). While there is a large body of research linking environmental conditions to distribution of various life stages of fish species in the context of climate change (e.g. Cheung et al., 2010; Hazen et al., 2013), a key remaining question is how recruitment might change in the future. This knowledge is essential to assess performance of current and alternative management strategies under climate change, and to develop environmentally informed management benchmarks. There is a long history of fisheries oceanography studies linking environmental conditions to recruitment in exploited fishes (e.g. Houde, 2008; Rothschild, 2000; Planque and Frédou, 1999; Megrey et al., 2005). However, these relationships often fail when later retested with new data, or different analysis techniques (Myers, 1998). As a result, the adoption of ecosystem-based fisheries management has been somewhat hampered by the lack of robust correlates with key variables such as recruitment, as well as limited process-based understanding of how environmental variability drives recruitment in different species. Nevertheless, simulations suggest that, if a robust relationships between environmental conditions and recruitment exists, incorporation of environmental predictors into the assessment process could provide early warnings of falling stock productivity, allowing for more effective fisheries management, and higher yields (e.g. Tommasi et al., 2017a). Similar to most managed stocks, status of Pacific bluefin tuna (PBF: Thunnus orientalis) is currently assessed without explicit consideration of environmental effects. PBF range throughout the North Pacific Ocean (Fujioka et al., 2016), but their spawning grounds are primarily restricted to two small areas in the western Pacific: between the Philippines and northern Nansei Islands between April and June, and in the Sea of Japan during July and August (Tanaka and Suzuki, 2016; Ohshimo et al., 2017). Annual recruitment is estimated from a standardized index of troll fishery catches of young-of-the-year fish from around Nagasaki Prefecture in the fall–spring following spawning (Yamada et al., 2006; ISC, 2016). Recruitment was generally low prior to 1990, higher from 1994 to 2008, and has been declining since then (Figure 1). Figure 1. View largeDownload slide (a) Study area in the western North Pacific Ocean. Approximate spawning areas and larval distributions are shown, as well as the climatological path of the Kuroshio Current, and the two locations used for calculating Kuroshio Current strength, after Andres et al. (2008). The fishing grounds for the age-0 troll fishery, upon which the recruitment index is based, are shown near to Nagasaki. (b) Annual PBF recruitment with standard deviations and spawning stock biomass (SSB) from the stock assessment model, 1982–2014. Figure 1. View largeDownload slide (a) Study area in the western North Pacific Ocean. Approximate spawning areas and larval distributions are shown, as well as the climatological path of the Kuroshio Current, and the two locations used for calculating Kuroshio Current strength, after Andres et al. (2008). The fishing grounds for the age-0 troll fishery, upon which the recruitment index is based, are shown near to Nagasaki. (b) Annual PBF recruitment with standard deviations and spawning stock biomass (SSB) from the stock assessment model, 1982–2014. Some recent studies have suggested that PBF recruitment variability is linked to the ocean environment, in particular to the Pacific Decadal Oscillation (PDO; Sakuramoto, 2016; Harford et al., 2017). The PDO arises from a number of different physical processes, related to the El Nino-Southern Oscillation cycle (ENSO), random fluctuations in the Aleutian Low, oceanic thermal inertia, and decadal-scale changes in the Kuroshio–Oyashio current system (Newman et al., 2016). During positive phases of the PDO, SSTs are warmer than average along the west coast of North America, and in the tropical eastern Pacific, and cooler than average at mid-latitudes in the central and western North Pacific, including coastal Japan. This suggests that the PDO may influence PBF recruitment via its effects on SST, which was recently supported by Ishida et al. (2018). Other processes can also influence ocean conditions near to PBF early life habitats, and thus may be important for recruitment predictability. El Niño events are associated with negative SST anomalies in the western North Pacific, which are strongest along ∼40°N, but weaker negative anomalies are also present around coastal Japan during the summer and fall (Alexander et al., 2002, 2004). Other potential contributors to regional SST variability include the strength of the Asian monsoons, and the Arctic Oscillation (AO; Minobe et al., 2004; Ohshimo et al., 2009). Despite these advances, several questions remain. The PDO is a basin-scale index, and the SST measures used in Ishida et al. (2018) were spatially and temporally coarse (averaged quarterly across regions >200 000 km2). It is therefore not clear where and when SST is related to PBF recruitment, which life stages are most strongly affected, and if the effects are common to both spawning grounds. Watai et al. (2017) concluded that only PBF larvae with fast, steady growth survive to be juveniles. Temperature is well known to positively influence larval tuna growth (e.g. Kimura et al., 2010; Satoh et al., 2013, 2014), but other factors such as prey abundance and composition, and larval transport via ocean currents may also be important (Kimura et al., 2010; Satoh et al., 2014; Tanaka et al., 2014). It is generally hypothesized that survival during the larval and post-larval phases is most important in determining year class strength in pelagic fishes (e.g. Bailey and Houde, 1989; Hare and Cowen, 1997). However, Ishida et al. (2018) found stronger correlations between regional SST and PBF recruitment in summer and fall. While spawning on the northern spawning ground extends through August, by the fall all surviving PBF are well into the juvenile stage. Alternatively, SST may be influencing the PBF recruitment index (rather than recruitment itself) by changing the availability of age-0 juveniles to the troll fishery. The effects of SST on the distribution and movement of small juvenile PBF fitted with satellite tags has been noted previously (Kitagawa et al., 2006), and proposed as a potential influence on catches in the fishery (Ichinokawa et al., 2014). In this study, we aimed to address these knowledge gaps by exploring finer-scale relationships between spatiotemporal SST variability, and PBF recruitment estimates from the most recent stock assessment. We then used these findings to develop multivariate, non-linear models to define the overall predictability of annual PBF recruitment from ocean conditions around spawning and nursery grounds. Material and methods A schematic of the workflow linking the data sources and models described below is shown in Supplementary Figure S1. Biological time series We obtained annual estimates of female SSB and recruitment from the 2016 PBF stock assessment report (ISC, 2016). Population dynamics were estimated for the assessment using a fully integrated age-structured model (Stock Synthesis v3.24f), which was fit to catch, size-composition and catch-per-unit-effort (CPUE) data from 1952 to 2015. However, we only used spawning biomass and recruitment values since 1982 in this study, to match the period when both SST and recruitment estimates were likely most reliable (Ishida et al., 2018). We used values from the base-case model only, which assumes that there is essentially no SRR for PBF, and thus that the curvature of the SRR relationship (steepness) is 0.999 (ISC, 2016; Nakatsuka et al., 2017). Temperature time series We defined the study area as including locations where (i) adult PBF are known or likely to spawn, or (ii) where larvae or small juvenile PBF have been collected previously (Fujioka et al., 2016; Tanaka and Suzuki, 2016; Ohshimo et al., 2017; Figure 1). Monthly SSTs were extracted on a 1 × 1 degree grid, for all months between April and (the following) January each year between 1982 and 2014, and were obtained from the National Oceanographic and Atmospheric Administration (NOAA) Optimum Interpolation Sea Surface Temperature (OISST) analysis version 2 (Reynolds et al., 2002). Spatiotemporal correlations Environmental variables can influence recruitment in different ways across the geographic extent of a species reproductive range (e.g. Planque and Frédou, 1999). We therefore calculated the correlation between annual recruitment and SST at each 1 × 1° grid point, in each month, across all years from 1982 to 2014. We evaluated both linear and simple non-linear correlations using 2nd order polynomials, as relationships between temperature and both spawning activity and larval survival have previously been reported to be non-linear for PBF (e.g. Kimura et al., 2010; Ohshimo et al., 2017). These were fit using the “lm” routine in R 3.3.2. Results were visualized by plotting values at each grid point by month, using Surfer 9 (Golden Software). Months and locations with the highest correlations were then selected for further examination, to define the nature of the relationship between recruitment and SST in each area. Neighbouring grid points with stronger correlations were grouped together to form “Areas of Interest” (AOIs). Temporal autocorrelation can reduce the degrees of freedom of the sample correlation between time series. To assess the importance of temporal autocorrelation when comparing recruitment to mean SST within our AOIs, we calculated the strength and significance of the autocorrelation function (ACF) for SST within each AOI at lag 1 year. Drivers of regional temperature variability SST has been highlighted as a potential driver of PBF recruitment, but is temporally and spatially autocorrelated (Ishida et al., 2018). We therefore used Principal Components Analysis (PCA) to summarize the spatiotemporal variability. This technique reduces the dimensionality of multivariate data to principle modes of variation, and calculates a set of linearly uncorrelated variables (principle components), which summarize progressively less of the variance (Legendre and Legendre, 1998). We first calculated SST anomalies against longitude/latitude/month means from 1982 to 2014, to remove the seasonal signal. We then applied PCA to the full SST anomaly dataset using the “prcomp” routine in the “stats” package in R 3.3.2 (R Core Team, 2016). We obtained time-series of climate indices from the NOAA Earth System Research Laboratory (Niño 3.4, PDO, AO) and the Japan Meteorological Agency (summer monsoon) at monthly resolution, and collapsed these into June–December means for each year from 1982 to 2014 (see Results for rationale behind selection of these months). Correlation coefficients were then calculated between mean values of the first four PCs, and means of each of the four climate indices. The significance of each correlation was assessed using “lm” in R 3.3.2, with a Durbin–Watson test for temporal autocorrelation of residuals non-significant at p < 0.05 unless otherwise stated in the text. We also included two local environmental indices: air temperature anomalies, and an index of Kuroshio Current transport. Air temperatures were obtained from the National Centers for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) Reanalysis at 2.5 latitude × longitude resolution (Kalnay et al., 1996). We used values at 700 hPa, as near-surface air temperatures can be influenced by feedbacks from the surface ocean, and we were most interested in processes which drive SST. Anomalies of mean monthly values (1982–2014) were calculated for a box defined by 122.5–135°E, and 22.5–37.5°N. The Kuroshio Current transport index was calculated after Andres et al. (2008), who showed that the difference in sea surface height anomaly (SSHa) between two points near the Nansei Islands was strongly and linearly correlated with observed transport (Figure 1). SSHa values were extracted at the same points as in Andres et al. (2008), but we obtained them from the DUACS 2014 delayed time 0.25° gridded analysis provided by CMEMS, from 1993 to 2014, instead of using along-track measurements. However, results were very similar to those in Andres et al. (2008). The two local indices and four climate indices were not independent from each other, and shared common forcing mechanisms. However, we used them in an exploratory manner to show the potential influence of various processes on observed SST variability within the study area. Predictive recruitment models The effects of spatiotemporal SST variability across the early life habitats of PBF on recruitment were summarized using Generalized Additive Models (GAMs: Wood, 2006), which were built using the “mgcv” package in R 3.3.2. GAMs are similar to Generalized Linear Models, but incorporate smoothing functions of predictors, to allow non-linear relationships between predictors and response variables. They have been previously shown to perform well in estimating recruitment–environment relationships, when compared with linear models (Megrey et al., 2005; Tommasi et al., 2015). We compared two approaches for predicting annual recruitment from regional SST. First, we derived SST time series from selected AOIs where SST correlated most strongly with recruitment, and used these as predictors in the GAM. This approach was conceptually simple and easy to interpret, but raised the potential problem of multicollinearity among predictors, as SST within the study area was highly auto-correlated in space and time. Secondly, we investigated using the first several PC axes as predictor variables in the GAM. Using PCs as predictors had the advantage of both summarizing SST conditions throughout the spawning season, and also providing variables that were uncorrelated to each other. However, this may come at the cost of increased difficulties with model interpretation. Each GAM also included annual SSB as a predictor. SSB was modelled to allow for a non-linear response (and hence density dependent effects) between SSB and recruitment. Annual estimated recruitment from the stock assessment report was used as the response variable. GAMs were built using a quasi-Poisson error distribution with a log link function. We restricted the number of “knots” for polynomial smoothers, to ensure biologically realistic response curves, and, and we used out-of-model validation to select the optimal value. This was done by training each GAM on the first 23 years of the time series (1982–2004), and then testing its accuracy on the last 10 years (2005–2014). Both time periods included both high and low recruitment years. The process was repeated for GAMs using a maximum of 3, 4, and 5 knots for smoothers, and the best model selected using the R2 value on the unseen, out-of-model data. We used the Shapiro–Wilk test and visual examination to check that model residuals were roughly normally distributed, and showed no serious heterogeneity. Temporal autocorrelation of residuals was tested for using the Durbin–Watson test. To provide predictions of potential PBF recruitment in years beyond the current stock assessment (2015 and 2016), OISST fields for those years were also extracted separately, and scored through both GAMs. SSB for 2015 and 2016 was held constant at 2014 levels for these predictions. Results Spatiotemporal correlations Across all months and locations, the increase in skill gained by using 2nd order polynomial correlations over linear relationships was modest, with a mean R2 increase of 0.02. The R2 statistic improved by > 0.1 in only 2.6% of month/location combinations, and so we proceeded using linear relationships for the correlation maps. These showed that SST was positively associated with recruitment off the southern coast of Shikoku from June to October, in the East China Sea in July–October, and then again in December, and in the Sea of Japan during August–December. Nearly all correlations were positive, with the exception of some very weak negative relationships in April and May. The strongest correlations (ρ > 0.6) were found south of Shikoku in June and July (Figure 2). Figure 2. View largeDownload slide Linear correlations between annual PBF recruitment and monthly sea surface temperatures, April to (following) January, 1982–2014. The locations of three Areas of Interest where SST was strongly correlated with recruitment are overlaid on the April map. Figure 2. View largeDownload slide Linear correlations between annual PBF recruitment and monthly sea surface temperatures, April to (following) January, 1982–2014. The locations of three Areas of Interest where SST was strongly correlated with recruitment are overlaid on the April map. We selected three AOIs based on the spatial patterns identified in Figure 2. The first was the area south of Shikoku, which is a juvenile nursery area for larvae spawned on the southern spawning ground (Figure 1). The second AOI was the southwest Sea of Japan, which encompasses larval and juvenile habitats for the northern spawning ground (Figure 1). The third AOI covered the northern East China Sea. Small juveniles from both spawning grounds may recruit to this habitat, and are targeted by troll fisheries based out of Nagasaki Prefecture (Figure 1). Correlations among time series of SST anomalies in the three AOIs within each month were moderate to high (ρ > 0.5), particularly between the East China Sea and south of Shikoku AOIs, with ρ > 0.7 within all months between July and December. This strong spatial autocorrelation is also evident in Figure 2, with areas of high recruitment predictability covering large portions of the study region in July–October. In contrast, the temporal persistence of SST anomalies was not particularly strong in the summer and fall. Correlations between mean SST anomalies within each AOI were <ρ = 0.5 for any lead time of >2 months, except for the Sea of Japan in June and July (Supplementary Figure S2). In July–December, it is therefore not possible to show precisely at which location SST was most important for recruitment. It is clear, however, that the positive relationships in fall were not a result of the temporal persistence of summer SST anomalies. For example, warmer conditions in the East China Sea in both July and December were associated with higher PBF recruitment. However, the correlation between July and December SST anomalies in this AOI across all years was only 0.12. SST anomalies averaged across fourteen AOI/month combinations could all predict PBF recruitment with ρ > 0.4 (Figure 2). However, the strong spatial, and to a lesser extent temporal, autocorrelation in SST fields resulted in high multicollinearity among these indices. An hierarchical cluster analysis (“hclust” in R 3.3.2) highlighted four groups of covarying indices with cross-correlations among groups of ρ < 0.6. These were therefore averaged to provide spatiotemporally aggregated SST predictors of PBF recruitment. These were the Shikoku AOI averaged from June to August (“Shikoku Summer”), the East China Sea in July–August and the Sea of Japan in August–September (“Marginal Seas Summer”), the East China Sea during December (“East China Sea Winter”), and a mean of Shikoku and the East China Sea in September–October, and the Sea of Japan in October and November (“All AOIs Fall”). Warmer temperatures in all of these area/month groups resulted in significantly higher PBF recruitment (p < 0.002). These relationships were strongest south of Shikoku in June–August (R2 = 0.42), and weakest in the East China Sea in December (R2 = 0.29). ACFs at lag 1 (year) were positive for the fall and winter AOIs (0.26 and 0.18, respectively), negative for the Marginal Seas Summer AOI (−0.23), and near zero for the Shikoku Summer AOI (−0.06). However, none was statistically significant at p < 0.05. The Shikoku Summer index was most relevant to larvae spawned on the southern spawning ground, whereas the Marginal Seas Summer index was more relevant to the northern spawning ground. The 10 years of the time series with the lowest recruitment were all associated with cooler anomalies for one (e.g. 2012) or both (e.g. 1993) of these two indices (Figure 3). Conversely, the ten highest recruitment years were associated with average to warm conditions in both areas. Figure 3. View largeDownload slide Mean SST in the Shikoku area of interest (x-axis) vs. mean SST in the Marginal Seas area of interest (y-axis) during summer (months defined in the text). The 10 years of highest recruitment from 1982 to 2014, and the 10 years of lowest recruitment, are also shown. The grey lines represent mean SST values (1982–2014). Figure 3. View largeDownload slide Mean SST in the Shikoku area of interest (x-axis) vs. mean SST in the Marginal Seas area of interest (y-axis) during summer (months defined in the text). The 10 years of highest recruitment from 1982 to 2014, and the 10 years of lowest recruitment, are also shown. The grey lines represent mean SST values (1982–2014). As SSTs during April, May and the following January were not strongly correlated with recruitment (Figure 2), we computed the PCA on SST anomalies from June to December. Approximately 65% of the variation in SST across the study area was captured by the first 5 PC axes. Mean loadings along all PCs primarily showed interannual variability, rather than any strong trend through time (Supplementary Figure S3). Correlations between PC1 and SST anomalies were positive for all four SST indices, suggesting that years with positive loadings along PC1 were warmer throughout the study region (Figure 4). Years with positive loading along PC2 were cooler during summer, but warmer in fall and winter, whereas years with positive loadings along PC3 were cooler in summer and fall, particularly in the marginal seas (Figure 4). PC 4 explained only 8.7% of the overall variability in SST anomalies. However, positive loadings along PC4 were associated with warmer conditions in the East China Sea during December, but cooler conditions during fall (Figure 4). Spatial correlations of each PC with SST anomalies within each month are shown for the whole study area in Supplementary Figures S4 and S5. Overall, annual values of PC1 (ρ = 0.47) and PC3 (ρ = −0.52) were more strongly correlated with annual PBF recruitment than PC5 (ρ = 0.26) or PCs 2 and 4 (ρ < 0.1). Figure 4. View largeDownload slide Linear correlations between each PC and SST anomalies from within the AOIs shown in Figure 2. Figure 4. View largeDownload slide Linear correlations between each PC and SST anomalies from within the AOIs shown in Figure 2. Positive values along PC1 were associated with warmer air temperatures at 700 hPa, and negative values of the PDO and Nino 3.4 index (Table 1). In addition, positive loadings along PC2 were negatively associated with the AO, whereas values of PC3 were positively associated with the Kuroshio Current index, and the PDO. Positive values of PC4 were negatively associated with the monsoon index, and positively correlated with the Nino 3.4 index. However, whereas all these linear correlations were significant at p < 0.05, only the correlation between PC1 and air temperature remained so after a Bonferroni correction for multiple comparisons (Table 1). Table 1. Linear correlations between June and December means of four climate variables, and two local environmental variables, and mean values of each of the first four Principal Component (PC) axes, 1982–2014. PC1 PC2 PC3 PC4 Air temperature 0.64a −0.05 −0.14 0.11 Arctic Oscillation −0.07 −0.37* −0.23 −0.02 Kuroshio Index 0.15 −0.01 0.50* 0.12 Monsoon Index 0.32 −0.16 −0.22 −0.44* Nino 3.4 −0.42* 0.20 −0.11 0.43* Pacific Decadal Oscillation −0.45* 0.10 0.47* 0.23 PC1 PC2 PC3 PC4 Air temperature 0.64a −0.05 −0.14 0.11 Arctic Oscillation −0.07 −0.37* −0.23 −0.02 Kuroshio Index 0.15 −0.01 0.50* 0.12 Monsoon Index 0.32 −0.16 −0.22 −0.44* Nino 3.4 −0.42* 0.20 −0.11 0.43* Pacific Decadal Oscillation −0.45* 0.10 0.47* 0.23 * Correlations significant at p < 0.05. a Those which remained so after a Bonferonni correction for multiple comparisons. Table 1. Linear correlations between June and December means of four climate variables, and two local environmental variables, and mean values of each of the first four Principal Component (PC) axes, 1982–2014. PC1 PC2 PC3 PC4 Air temperature 0.64a −0.05 −0.14 0.11 Arctic Oscillation −0.07 −0.37* −0.23 −0.02 Kuroshio Index 0.15 −0.01 0.50* 0.12 Monsoon Index 0.32 −0.16 −0.22 −0.44* Nino 3.4 −0.42* 0.20 −0.11 0.43* Pacific Decadal Oscillation −0.45* 0.10 0.47* 0.23 PC1 PC2 PC3 PC4 Air temperature 0.64a −0.05 −0.14 0.11 Arctic Oscillation −0.07 −0.37* −0.23 −0.02 Kuroshio Index 0.15 −0.01 0.50* 0.12 Monsoon Index 0.32 −0.16 −0.22 −0.44* Nino 3.4 −0.42* 0.20 −0.11 0.43* Pacific Decadal Oscillation −0.45* 0.10 0.47* 0.23 * Correlations significant at p < 0.05. a Those which remained so after a Bonferonni correction for multiple comparisons. Results shown in Table 1 suggested that both air temperature and Kuroshio Current transport could influence regional SST through local forcing. To explore this further, we calculated the correlation coefficient between June-December mean air temperature anomalies averaged across the region, and mean June–December SST anomaly at each grid point. The same exercise was then repeated to compare SST anomalies to the Kuroshio Current index. Results suggested that air temperature was a strong driver of SST anomalies, except in the Sea of Japan and east of Taiwan (Figure 5). A stronger Kuroshio Current appeared to result in weak warm SST anomalies in the southern part of the study area, and stronger cool anomalies in the eastern Sea of Japan. Correlations between area-averaged SST anomalies and area-averaged air temperature anomalies by month showed that the strongest relationships were nearly always at zero lead times (i.e. June SST anomaly was best predicted by June air temperature anomalies, rather than April or May air temperatures). Figure 5. View largeDownload slide Spatial correlation between SST anomalies at each grid point and top: April–December mean 700 hPa air temperature anomalies across the study region, and bottom: the Kuroshio Current index, 1982–2014. Figure 5. View largeDownload slide Spatial correlation between SST anomalies at each grid point and top: April–December mean 700 hPa air temperature anomalies across the study region, and bottom: the Kuroshio Current index, 1982–2014. Predictive recruitment models The first GAM (“SST GAM”) was built using the four SST indices described above as predictors, along with SSB. Using k = 3 gave the highest R2 (0.51), and lowest RMSE (4211.8) on the out-of-model validation data (years 2005–2014). The second GAM (“PC GAM”) was built using the first five PC axes as predictors, as well as SSB. This model also showed the highest validation skill (R2 = 0.47), and lowest RMSE when k was equal to 3. The final GAMs were thus built with k = 3. Whereas PC1, PC3, and SSB were significant to the PC GAM at p < 0.05, PC2, PC4, and PC5 were not (Table 2). Excluding PC5 resulted in no loss of model skill (as determined by the –REML score and the % deviance explained by the model), but excluding PC2 and PC4 did. The final model was therefore built with all variables retained except for PC5. Similarly, excluding East China Sea December SST did not degrade the performance of the SST GAM, and so this variable was excluded from the final model (Table 2). However, excluding SSB did slightly degrade skill, and so was left in, even though the relationship was weak. Although conceptually simpler, the SST GAM performed slightly better on the out-of-model validation years (R2 = 0.51 vs. R2 = 0.47). The PC GAM provided a better fit to the training data (R2 = 0.82 vs. R2 = 0.72). Table 2. Results of two GAMs (SST GAM and PC GAM) predicting annual PBF recruitment from surface temperatures. Variable p-Value Model R2 SST GAM Shikoku Summer 0.007 Training Testing All years Marginal Seas Summer 0.063 0.72 0.51 0.65 All AOIs Fall 0.073 East China Sea December Removed SSB 0.092 PC GAM PC1 0.006 Training Testing All years PC2 0.082 0.82 0.47 0.74 PC3 0.0001 PC4 0.127 PC5 Removed SSB 0.011 Variable p-Value Model R2 SST GAM Shikoku Summer 0.007 Training Testing All years Marginal Seas Summer 0.063 0.72 0.51 0.65 All AOIs Fall 0.073 East China Sea December Removed SSB 0.092 PC GAM PC1 0.006 Training Testing All years PC2 0.082 0.82 0.47 0.74 PC3 0.0001 PC4 0.127 PC5 Removed SSB 0.011 SSB denotes spawning stock biomass from the most recent stock assessment model. The R2 for each model is shown for the training dataset (1982–2004), the out-of-model testing dataset (2005–2014), and for all years together. Table 2. Results of two GAMs (SST GAM and PC GAM) predicting annual PBF recruitment from surface temperatures. Variable p-Value Model R2 SST GAM Shikoku Summer 0.007 Training Testing All years Marginal Seas Summer 0.063 0.72 0.51 0.65 All AOIs Fall 0.073 East China Sea December Removed SSB 0.092 PC GAM PC1 0.006 Training Testing All years PC2 0.082 0.82 0.47 0.74 PC3 0.0001 PC4 0.127 PC5 Removed SSB 0.011 Variable p-Value Model R2 SST GAM Shikoku Summer 0.007 Training Testing All years Marginal Seas Summer 0.063 0.72 0.51 0.65 All AOIs Fall 0.073 East China Sea December Removed SSB 0.092 PC GAM PC1 0.006 Training Testing All years PC2 0.082 0.82 0.47 0.74 PC3 0.0001 PC4 0.127 PC5 Removed SSB 0.011 SSB denotes spawning stock biomass from the most recent stock assessment model. The R2 for each model is shown for the training dataset (1982–2004), the out-of-model testing dataset (2005–2014), and for all years together. Results from the SST GAM confirmed that high PBF recruitment was associated with warm SSTs south of Shikoku and in the marginal seas during summer and fall. The partial plot for Shikoku summer SST was somewhat non-linear, however all other SST indices showed largely linear relationships with recruitment. These results corresponded closely with those from the PC GAM, where moderately high values of PC1, low values of PC3, and high values of SSB were positively associated with recruitment (Supplementary Figure S6). Both models showed positive relationships between SSB and recruitment, although this was weaker and non-linear in the SST GAM. While both GAMs showed that years of higher recruitment generally coincided with years of higher SSB, the relationship was inconsistent through time (Figure 6). In particular, recruitment was higher than might be expected from SSB in years after 2005. If both GAMs were re-run without SSB, the SST GAM still explained 70.3% of the deviance in recruitment for years 1982–2004 (vs. 74.7% with SSB included), and the PC GAM still explained 73.5% (vs. 81.2% with SSB included). In contrast, running a GAM with SSB as the only predictor explained 29.7%, with a positive linear relationship. This suggests that SST had a generally greater ability to explain interannual variability in recruitment than did SSB. In addition, there was no strong evidence of density dependence (which would be indicated by a consistently non-linear partial response of SSB in the GAMs). However, the PBF stock was at low biomass levels for all years between 1982 and 2015 (depletion ratios ranging from 1.8 to 9.6%: ISC, 2016), and so there may not have been enough contrast to show density dependence. The deviance in annual recruitment explained by both GAMs was quite similar, suggesting that a large portion of the temperature effect was associated with the AOIs defined in Figure 2. However, the PC GAM did noticeably better at several points in the time series, particularly the recruitment peaks in 1994 and 2004 (Figure 6). Both GAMs were able to replicate the overall lower levels of recruitment in the 1980s, with higher values since 1994, but both models over-predicted in the last 3 years of the time series, particularly during 2013 (Figure 6). Figure 6. View largeDownload slide Time series of observed and predicted PBF recruitment from both GAMs (left: PCA GAM, right: SST GAM), with spawning stock biomass also shown. Standard errors are shown in thin lines for each GAM. Vertical lines divide the time series into years used for GAM training (1982–2004), validation (2005–2014), and predictions of 2015 and 2016 recruitment using OISST fields, and 2014 levels of SSB. Figure 6. View largeDownload slide Time series of observed and predicted PBF recruitment from both GAMs (left: PCA GAM, right: SST GAM), with spawning stock biomass also shown. Standard errors are shown in thin lines for each GAM. Vertical lines divide the time series into years used for GAM training (1982–2004), validation (2005–2014), and predictions of 2015 and 2016 recruitment using OISST fields, and 2014 levels of SSB. When the two GAMs were applied to OISST fields from 2015 and 2016 (with SSB kept at 2014 levels), both models predicted that recruitment in 2015 would be similar to what it was in 2014, and 2016 recruitment would be more favourable. The SST GAM was considerably more optimistic than the PC GAM, however. In both years, SST in the AOIs shown in Figure 2 was at least >0.5°C above normal in 2016. In contrast, temperatures in the marginal seas during summer in 2015 were >1.0°C cooler than average. Higher predicted recruitment in 2016 was therefore a result of warm anomalies across the region in summer to fall, whereas 2015 had a cool summer and fall, resulting in predictions of below average recruitment. Discussion Temperature effects on PBF recruitment Results from this study suggest a spatially and temporally variable temperature effect on interannual PBF recruitment. The effect was most marked off the Pacific coast of Shikoku, in the northern East China Sea and in the southwest Sea of Japan. The strongest year classes were associated with warm conditions from summer to early winter, in all AOIs. The highest correlations were present south of Shikoku, a nursery area for small juveniles advected northwards in the Kuroshio Current from the southern spawning ground (Kitagawa et al., 2010). PBF arriving off Shikoku in June–August occupy warm waters (mixed layer > 27°C) during their first summer (Furukawa et al., 2017), and remain in the area through winter before migrating northwards the following spring. In contrast, the Sea of Japan in August and September is primarily larval and juvenile habitat for the northern spawning ground. Cooling temperatures in the Sea of Japan during fall result in southward migration of juvenile PBF towards Tsushima Island and the East China Sea (Kitagawa et al., 2006). By fall, small juveniles from both spawning grounds mix in the East China Sea area (Fujioka et al., 2016), and so correlations between SST and recruitment in the combined marginal seas region are not specific to either one of the two spawning grounds. Some juveniles from the southern spawning ground arrive in the East China Sea as early as July, around the same time as spawning commences in the Sea of Japan. The positive effect of temperature on recruitment in the marginal seas in summer could therefore apply primarily to larvae in the Sea of Japan, to juveniles in the East China Sea, or both. In contrast, there were no strong correlations observed between recruitment and temperature near Taiwan and the southern Nansei Islands in April and May, with some weak positive correlations in June. This suggests that, for the southern spawning ground, temperature-driven survival of larval PBF is not driving recruitment variability. Recent studies suggest that PBF survival to the juvenile phase may depend primarily on growth-dependent survival of larvae (Tanaka et al., 2006, 2014; Satoh et al., 2013; Watai et al., 2017). Whereas studies on cultured PBF larvae growth show strong positive relationships with temperature (Kimura et al., 2010), results have been more complex for field collected larvae. Satoh et al. (2013, 2014) showed that temperature and prey density interacted to positively influenced PBF larval growth. Conversely, Tanaka et al. (2006) found little effect of temperature on larval PBF growth, across a range of ∼25–29°C. At <14 days post hatch, PBF larvae are primarily zooplanktivorous (Tanaka et al., 2014). Tanaka et al. (2008) showed that larval PBF have a very low tolerance to starvation, and so prey densities may be more important for determining survival than temperature for larvae spawned on the warmer southern spawning ground. Mortality rates on pelagic early life stages of fishes are often assumed to be highest at younger ages, with recruitment strength determined by high and variable mortality at the larval stage (Bailey and Houde, 1989; Pepin and Myers, 1991). However, studies on captive PBF have demonstrated that mortality on small juveniles can also be substantial. This is often associated with stress from handling, transfer, and collisions with tank edges, but Tsuda et al. (2012) also showed a temperature effect on 5–6 month old juvenile PBF in sea cages. Decreases in ambient water temperature below ∼20°C resulted in decreased daily survival, with cooling below 15–17°C causing particularly strong spikes in mortality. Colder temperatures in juvenile PBF habitats over their first year of life, before they are completely endothermic (Kubo et al., 2008), may therefore result in higher mortality, particularly during fall and winter. Steady and fast growth in early larval phases, whether mediated by temperature or prey availability, may determine the likelihood of survival to the juvenile stage, but overall recruitment of age-0 fish may also depend on temperature-driven survival in juveniles. Direct mortality from cold temperatures is most likely to occur in late fall and winter, however unfavourable temperatures may also decrease survival through indirect mechanisms (e.g. Ottersen and Loeng, 2000). These may include slower juvenile growth causing higher predation mortality, or temperature-driven impacts on prey fields. Such complex regulation of recruitment by differing processes over multiple life stages has been shown previously for some species, which are benthic as juveniles and adults (e.g. Duffy-Anderson et al., 2005), but not for pelagic species such as tunas. Another possibility is that temperature is influencing the accessibility of age-0 fish to troll fisheries, and thus the recruitment index. Unlike the captive PBF studied by Tsuda et al. (2012), wild juveniles can move away from unsuitable thermal habitat. Mean SST near to the main troll fishing grounds in Nagasaki Prefecture begins to cool below 20°C in December, and may fall below 16°C from January to April (Ichinokawa et al., 2014; this study). Most fishing vessels targeting age-0 PBF have limited range, and so temperature-driven southward movement of age-0 PBF towards the Kuroshio Current boundary (Kitagawa et al., 2006), could result in less fish being available to the troll fisheries working in the Tsushima Strait. This could bias the recruitment index derived from catches in these fisheries. Alternatively, cooler temperatures could lead to slower growth and smaller sizes in age-0 juveniles, reducing their vulnerability to gears employed by the troll fishery. Examination of both of these possibilities in future studies would provide more information on the representativeness of the current troll-based recruitment index, and help to better define mechanistic links between temperature and recruitment strength. Climate drivers of variability in larval habitats Our results showed that SST had strong predictive skill for PBF recruitment, and so the underlying drivers of ocean temperature variability in the region are important to understand. SST within the AOIs varied widely across the 33 years of data. For example, three of the four SST indices varied by >2°C interannually, and the Sea of Japan during August varied by > 5°C. PCA of the SST fields suggested that dynamics in the Sea of Japan were often distinct from those in the open Pacific, or in the southern East China Sea, especially in June and July. While the importance of the Kuroshio Current and associated eddies to PBF recruitment has been highlighted previously (Kimura et al., 2010; Kitagawa et al., 2010), the strong SST variability shown in this study was mostly atmospheric in origin. Years, which were warmer across most of the study region (and thus loaded positively along PC1), were associated with above average air temperatures, and negative values of ENSO and the PDO. PC3 appeared to incorporate some of the variability from the PDO, and Kuroshio Current flow, whereas PC4 had some relation to the monsoon index, and ENSO. El Niño events result in negative SST anomalies in the western North Pacific, as atmospheric teleconnections drive changes in air temperatures, wind patterns, and humidity (Alexander et al., 2002, 2004), and so their link to regional SST variability is expected. The correlation between area-averaged SST anomalies and a June–December mean of the Nino 3.4 index was stronger than that between mean 700 hPa air temperature anomalies and the same Nino 3.4 index (ρ = −0.44 vs. ρ = −0.21). This appears to confirm the importance of variables other than air temperature in linking ENSO to SST anomalies, such as cloud cover (Alexander et al., 2004). While air temperature explained more of the variability in SST than did Kuroshio Current flow, a stronger Kuroshio Current was associated with positive SST anomalies near to the Nansei Islands, and negative anomalies in the southern Sea of Japan. As the Kuroshio strengthens, its path meanders less in the East China Sea, and the Tsushima Current (which transports warm water northwards through the Tsushima Strait) weakens. This leads to negative temperature anomalies in the southwest Sea of Japan (Gordon and Giulivi, 2004). While positive phases of the PDO are associated with negative SST anomalies near to Japan, the mechanisms vary seasonally (Newman et al., 2016). Winter temperature anomalies can become decoupled from the surface as the water column stratifies during summer. These anomalies may then “reemerge” as stratification breaks down the following fall, leading to time series of winter SST that are correlated year-to-year, lengthening the overall timescale of SST anomalies (Alexander et al., 1999). In contrast, warm season SST anomalies in the North Pacific are primarily related to air–sea heat fluxes, and local forcing (Newman et al., 2016). Some studies have linked the strength of the Kuroshio Current south of Japan to the PDO at zero lag (e.g. Gordon and Giulivi, 2004; Andres et al., 2009). However, Soeyanto et al. (2014) found that the correlation described by Andres et al. (2009) weakened substantially when extended to 2012. The correlations between the PDO and PBF recruitment noted previously (Sakuramoto, 2016; Harford et al., 2017; Ishida et al., 2018) thus appear likely to result from connections between the PDO and local SST anomalies. The higher correlations between the PDO and recruitment in fall noted by Harford et al. (2017) and Ishida et al. (2018) likely result from the stronger effect of the PDO on regional SSTs during the cooler months, rather than a stronger effect of environmental variability on recruitment during fall. As shown in this study, temperature effects on recruitment are also likely to be present during summer, but SST during this season is not as well described by the PDO. Management implications SST anomalies in the marginal seas vs. those in the open Pacific did not always correspond strongly in space and time, with the result that conditions could be potentially favourable in some months and locations, but not others. Itoh (2009) used daily otolith increments to show comparatively high contribution of the Sea of Japan spawning ground to recruitment in 1994, low contribution in 1993, and moderate contribution in 1995–1997. This is consistent with the marginal seas being warm in 1994, cold in 1993, and moderate in 1995–1997 during summer, as shown in the present study (Figure 3). As smaller adults spawn in the Sea of Japan, and larger fish further south towards Taiwan (Itoh, 2006), this could result in substantial interannual variability in reproductive output by both spawning ground, and by adult size class. Although the current recruitment index is spatially aggregated, future work could consider estimating recruitment by spawning ground, by expanding work that assigns juveniles to spawning ground based on back-calculated hatch dates. This would enhance understanding of how environmental conditions, including temperature, drive variability in recruitment from each of the two spawning grounds. The current stock assessment model for PBF assumes that SSB does not drive substantial variation in recruitment, but there has been significant discussion around this point (e.g. ISC, 2016; Nakatsuka et al., 2017). The importance of SSB to the predictive GAMs was relatively weak compared with the effects of SST, with both models retaining much of their skill if SSB was removed. The relationship between SSB and recruitment was non-stationary, even if we followed Nakatsuka et al. (2017), and used a 5-year moving mean of both variables (results not shown). The SRR for PBF may therefore not be well represented by traditional models, and environmental effects may be substantially confounding it, a situation likely common to many fish stocks (e.g. Szuwalski et al., 2015; Lowerre-Barbieri et al., 2017). Future work could test the development of an environmentally explicit SRR model, where the potential influence of environmental on density-dependence could be included. However, the mechanistic processes linking SSB and environment to recruitment require further investigation. Temperature-driven changes in survival, growth, or distribution of early life stages are all plausible, and should be examined more closely using field and laboratory studies. In response to increasing concern about the status of the PBF stock, the Western and Central Pacific Fisheries Commission (WCPFC) recommended the adoption of an emergency rule in the event of “drastic drops in recruitment” (WCPFC, 2015). While the specifics of this rule are still under development, the relationship between SST and recruitment highlighted in this study could contribute to this type of measure, with anomalously cold SSTs providing an early warning to managers that recruitment may be low in that year. This can potentially be incorporated into a more formal decision rule, which could increase the adaptive management potential for the stock. However, the potential benefit of such a rule to the long-term sustainability of the stock, its rebuilding potential, and future harvest would need to be quantitatively assessed through a management strategy evaluation, before inclusion in the formal management process. Alternatively, if future research can show that SST is significantly influencing availability of age-0 fish to the troll fisheries whose catches form the basis of the recruitment index, a temperature-based standardization could be explored. The strong relationship between PBF recruitment and SST also raises the possibility of future prediction capabilities, either on a seasonal or multi-annual time scale (Tommasi et al., 2017b). However, whereas seasonal prediction of SST anomalies in the Sea of Japan region is strong at a 6–12 month lead time in winter and spring, prediction of summer SST anomalies is only moderate (Stock et al., 2015). Multiyear prediction skill of annual SST upper and lower terciles for the Sea of Japan and Pacific coastal Japan is relatively strong up to 10 years in advance (Tommasi et al., 2017c). However, predictive capabilities during summer and at the smaller spatial scale relevant to PBF recruitment may be lower due to the inability of current prediction systems to represent fine scale shelf processes and the stronger importance of local atmospheric variability, rather than predictable basin-scale SST variations, in modulating summer SST. This presents a challenge for the seasonal–decadal prediction of PBF recruitment with current models. On longer timescales, climate change is likely to drive substantial warming on PBF spawning grounds in the coming decades. The southern PBF spawning area is projected to increase in temperature by 1.5–2°C by the end of the 21st century, while the Sea of Japan may warm by more than 3°C (Woodworth-Jefcoats et al., 2017). This warming may benefit PBF recruitment, if we assume that the relationships defined in this study continue to hold. However, increasing incidence of very warm (>29°C) temperatures may be deleterious. Kimura et al. (2007) showed that increasing ambient temperature from 26 to 29°C over a short period of time resulted in increased mortality in cultured PBF larvae. In the field, this temperature increase would happen over a much longer time scale, and may allow some adaptation to new conditions. Nevertheless, the emergence of conditions outside the range of historical measurements, in terms of temperature, regional current systems, prey fields, metabolic demands, and other variables, may challenge the adaptation of PBF across life stages. Kimura et al. (2010) highlighted the seemingly delicate balance between SSTs on the southern spawning ground, transport of larvae northwards in the Kuroshio Current, and arrival of post-larvae on nursery grounds around coastal Japan. The projected strengthening of the Kuroshio Current (Sakamoto et al., 2005), and the different warming rates projected for the two spawning grounds may require complex adaptation responses, on a relatively short evolutionary timescale. How long-term climate change will impact PBF is thus complex, and difficult to predict. Conclusions Results from this study highlighted strong correlations between spatiotemporal variability of spring—early winter SSTs around coastal Japan, and annual PBF recruitment. Warmer temperatures appeared to be most advantageous, with near linear positive relationships in most cases. Interannual variability in SSTs was most strongly associated with air temperature anomalies, with some weaker connections to the ENSO cycle, the PDO, and Kuroshio Current transport. Predictive models for recruitment using the main modes of SSTs variability may allow more adaptive management of PBF, across environmental regime shifts, and under future climate change conditions. Acknowledgements NCEP Reanalysis Derived data and NOAA_OI_SST_V2 data were provided by the NOAA/OAR/ESRL PSD, Boulder, Colorado, USA, from their web site at http://www.esrl.noaa.gov/psd/. We also thank H. Dewar and three anonymous reviewers for comments, which significantly improved the manuscript. This study was funded by the NOAA National Marine Fisheries Service. Supplementary data Supplementary material is available at the ICESJMS online version of the manuscript. References Alexander M. A. , Deser C. , Timlin M. S. 1999 . The reemergence of SST anomalies in the North Pacific Ocean . Journal of Climate , 12 : 2419 – 2433 . Google Scholar CrossRef Search ADS Alexander M. A. , Blade I. , Newman M. , Lanzante J. R. , Lau N-C. , Scott J. D. 2002 . The atmospheric bridge: the influence of ENSO teleconnections on air–sea interaction over the global oceans . Journal of Climate , 15 : 2205 – 2231 . Google Scholar CrossRef Search ADS Alexander M. A. , Lau N-C. , Scott J. D. 2004 . Broadening the atmospheric bridge paradigm: ENSO teleconnections to the North Pacific in summer and to the tropical west Pacific-Indian Oceans over the seasonal cycle. In Earth Climate: The Ocean-Atmosphere Interaction , pp. 85 – 104 . Eds. by Wang C. , Xie S-P. , Carton J. , AGU Monograph . Andres M. , Park J. H. , Wimbush M. , Zhu X. H. , Chang K. I. , Ichikawa H. 2008 . Study of the Kuroshio/Ryukyu Current system based on satellite-altimeter and in situ measurements . Journal of Oceanography , 64 : 937 – 950 . Google Scholar CrossRef Search ADS Andres M. , Park J. H. , Wimbush M. , Zhu X. H. , Nakamura H. , Kim K. , Chang K. I. 2009 . Manifestation of the pacific decadal oscillation in the Kuroshio . Geophysical Research Letters , 36 . doi:10.1029/2009GL039216. Bailey K. M. , Houde E. D. 1989 . Predation on eggs and larvae of marine fishes and the recruitment problem . Advances in Marine Biology , 25 : 1 – 83 . Google Scholar CrossRef Search ADS Britten G. L. , Dowd M. , Worm B. 2016 . Changing recruitment capacity in global fish stocks . Proceedings of the National Academy of Sciences , 113 : 134 – 139 . Google Scholar CrossRef Search ADS Cheung W. W. , Lam V. W. , Sarmiento J. L. , Kearney K. , Watson R. E. G. , Zeller D. , Pauly D. 2010 . Large-scale redistribution of maximum fisheries catch potential in the global ocean under climate change . Global Change Biology , 16 : 24 – 35 . Google Scholar CrossRef Search ADS DePiper G. S. , Gaichas S. K. , Lucey S. M. , Pinto da Silva P. , Anderson M. R. , Breeze H. , Bundy A. , et al. 2017 . Operationalizing integrated ecosystem assessments within a multidisciplinary team: lessons learned from a worked example . ICES Journal of Marine Science , 74 : 2076 – 2086 . Google Scholar CrossRef Search ADS Duffy-Anderson J. T. , Bailey K. , Ciannelli L. , Cury P. , Belgrano A. , Stenseth N. C. 2005 . Phase transitions in marine fish recruitment processes . Ecological Complexity , 2 : 205 – 218 . Google Scholar CrossRef Search ADS Fujioka K. , Masujima M. , Boustany A. M. , Kitagawa T. 2016 . Horizontal movements of Pacific bluefin tuna. In Biology and Ecology of Bluefin Tuna , pp. 101 – 122 . Eds. by Kitagawa T. , Kimura S. . CRC Press , Boca Raton, FL . Furukawa S. , Fujioka K. , Fukuda H. , Suzuki N. , Tei Y. , Ohshimo S. 2017 . Archival tagging reveals swimming depth and ambient and peritoneal cavity temperature in age-0 Pacific bluefin tuna, Thunnus orientalis, off the southern coast of Japan . Environmental Biology of Fishes , 100 : 35 – 48 . Google Scholar CrossRef Search ADS Gordon A. L. , Giulivi C. F. 2004 . Pacific decadal oscillation and sea level in the Japan/East Sea . Deep Sea Research Part I: Oceanographic Research Papers , 51 : 653 – 663 . Google Scholar CrossRef Search ADS Grafton R. Q. 2010 . Adaptation to climate change in marine capture fisheries . Marine Policy , 34 : 606 – 615 . Google Scholar CrossRef Search ADS Hare J. A. , Cowen R. K. 1997 . Size, growth, development, and survival of the planktonic larvae of Pomatomus saltatrix (Pisces: pomatomidae) . Ecology , 78 : 2415 – 2431 . Google Scholar CrossRef Search ADS Harford W. J. , Karnauskas M. , Walter J. F. , Liu H. 2017 . Non-parametric modelling reveals environmental effects on bluefin tuna recruitment in Atlantic, Pacific, and Southern Oceans . Fisheries Oceanography , 26 : 396 – 412 . Google Scholar CrossRef Search ADS Hazen E. L. , Jorgensen S. , Rykaczewski R. R. , Bograd S. J. , Foley D. G. , Jonsen I. D. , Shaffer S. A. , et al. 2013 . Predicted habitat shifts of Pacific top predators in a changing climate . Nature Climate Change , 3 : 234 – 238 . Google Scholar CrossRef Search ADS Hilborn R. , Walters C. J. 1992 . Quantitative fisheries stock assessment: choice, dynamics and uncertainty . Reviews in Fish Biology and Fisheries , 2 : 177 – 178 . Google Scholar CrossRef Search ADS Hoegh-Guldberg O. , Bruno J. F. 2010 . The impact of climate change on the world’s marine ecosystems . Science , 328 : 1523 – 1528 . Google Scholar CrossRef Search ADS PubMed Houde E. D. 2008 . Emerging from Hjort’s shadow . Journal of Northwest Atlantic Fishery Science , 41 : 53 – 70 . Google Scholar CrossRef Search ADS Ichinokawa M. , Okamura H. , Oshima K. , Yokawa K. , Takeuchi Y. 2014 . Spatiotemporal catch distribution of age-0 Pacific bluefin tuna Thunnus orientalis caught by the Japanese troll fishery in relation to surface sea temperature and seasonal migration . Fisheries Science , 80 : 1181 – 1191 . Google Scholar CrossRef Search ADS ISC 2016 Pacific Bluefin Tuna Stock Assessment, 2016. Ishida Y. , Fukuda H. , Fujioka K. , Sakai O. , Hiraoka Y. , Oshima K. , Nakatsuka S. , et al. 2018 . Long-term changes in recruitment of age-0 Pacific bluefin tuna (Thunnus orientalis) and environmental conditions around Japan . Fisheries Oceanography , 27 : 41 – 48 . Google Scholar CrossRef Search ADS Itoh T. 2006 . Sizes of adult bluefin tuna Thunnus orientalis in different areas of the western Pacific Ocean . Fisheries Science , 72 : 53 – 62 . Google Scholar CrossRef Search ADS Itoh T. 2009 . Contributions of different spawning seasons to the stock of Pacific bluefin tuna Thunnus orientalis estimated from otolith daily increments and catch-at-length data of age-0 fish . Nippon Suisan Gakkaishi (Japanese Edition) , 75 : 412 – 418 . Google Scholar CrossRef Search ADS Kalnay E. , Kanamitsu M. , Kistler R. , Collins W. , Deaven D. , Gandin L. , Iredell L. , et al. 1996 . The NCEP/NCAR 40-year reanalysis project . Bulletin of the American Meteorological Society , 77 : 437 – 471 . Google Scholar CrossRef Search ADS Kimura S. , Kitagawa T. , Kato Y. , Yamaoka N. 2007 . Fluctuation in spawning environment of bluefin tuna and Japanese eel associated with global warming and their ecological response . Kaiyo Monthly , 39 : 317 – 322 (in Japanese). Kimura S. , Kato Y. , Kitagawa T. , Yamaoka N. 2010 . Impacts of environmental variability and global warming scenario on Pacific bluefin tuna (Thunnus orientalis) spawning grounds and recruitment habitat . Progress in Oceanography , 86 : 39 – 44 . Google Scholar CrossRef Search ADS Kitagawa T. , Sartimbul A. , Nakata H. , Kimura S. , Yamada H. , Nitta A. 2006 . The effect of water temperature on habitat use of young Pacific bluefin tuna Thunnus orientalis in the East China Sea . Fisheries Science , 72 : 1166 – 1176 . Google Scholar CrossRef Search ADS Kitagawa T. , Kato Y. , Miller M. J. , Sasai Y. , Sasaki H. , Kimura S. 2010 . The restricted spawning area and season of Pacific bluefin tuna facilitate use of nursery areas: a modelling approach to larval and juvenile dispersal processes . Journal of Experimental Biology and Ecology , 393 : 23 – 31 . Google Scholar CrossRef Search ADS Koehn J. D. , Hobday A. J. , Pratchett M. S. , Gillanders B. M. 2011 . Climate change and Australian marine and freshwater environments, fishes and fisheries: synthesis and options for adaptation . Marine and Freshwater Research , 62 : 1148 – 1164 . Google Scholar CrossRef Search ADS Kubo T. , Sakamoto W. , Murata O. , Kumai H. 2008 . Whole‐body heat transfer coefficient and body temperature change of juvenile Pacific bluefin tuna Thunnus orientalis according to growth . Fisheries Science , 74 : 995 – 1004 . Google Scholar CrossRef Search ADS Legendre P. , Legendre L. , 1998 . Numerical Ecology . Elsevier Science , Oxford, U.K . 840 pp. Lowerre-Barbieri S. , DeCelles G. , Pepin P. , Catalán I. A. , Muhling B. , Erisman B. , Cadrin S. X. , et al. 2017 . Reproductive resilience: a paradigm shift in understanding spawner‐recruit systems in exploited marine fish . Fish and Fisheries , 18 : 285 – 312 . Google Scholar CrossRef Search ADS Megrey B. A. , Lee Y. W. , Macklin S. A. 2005 . Comparative analysis of statistical tools to identify recruitment–environment relationships and forecast recruitment strength . ICES Journal of Marine Science , 62 : 1256 – 1269 . Google Scholar CrossRef Search ADS Minobe S. , Sako A. , Nakamura M. 2004 . Interannual to interdecadal variability in the Japan Sea based on a new gridded upper water temperature dataset . Journal of Physical Oceanography , 34 : 2382 – 2397 . Google Scholar CrossRef Search ADS Myers R. A. 1998 . When do environment–recruitment correlations work? Reviews in Fish Biology and Fisheries , 8 : 285 – 305 . Google Scholar CrossRef Search ADS Nakatsuka S. , Ishida Y. , Fukuda H. , Akita T. 2017 . A limit reference point to prevent recruitment overfishing of Pacific bluefin tuna . Marine Policy , 78 : 107 – 113 . Google Scholar CrossRef Search ADS Newman M. , Alexander M. A. , Ault T. R. , Cobb K. M. , Deser C. , Di Lorenzo E. , Mantua N. J. , et al. 2016 . The Pacific decadal oscillation, revisited . Journal of Climate , 29 : 4399 – 4427 . Google Scholar CrossRef Search ADS Ohshimo S. , Tanaka H. , Hiyama Y. 2009 . Long-term stock assessment and growth changes of the Japanese sardine (Sardinops melanostictus) in the Sea of Japan and East China Sea from 1953 to 2006 . Fisheries Oceanography , 18 : 346 – 358 . Google Scholar CrossRef Search ADS Ohshimo S. , Tawa A. , Ota T. , Nishimoto S. , Ishihara T. , Watai M. , Satoh K. , et al. 2017 . Horizontal distribution and habitat of Pacific bluefin tuna, Thunnus orientalis, larvae in the waters around Japan . Bulletin of Marine Science , 9 : 769 – 787 . Google Scholar CrossRef Search ADS Ottersen G. , Loeng H. 2000 . Covariability in early growth and year-class strength of Barents Sea cod, haddock, and herring: the environmental link . ICES Journal of Marine Science , 57 : 339 – 348 . Google Scholar CrossRef Search ADS Pepin P. , Myers R. A. 1991 . Significance of egg and larval size to recruitment variability of temperate marine fish . Canadian Journal of Fisheries and Aquatic Sciences , 48 : 1820 – 1828 . Google Scholar CrossRef Search ADS Planque B. , Frédou T. 1999 . Temperature and the recruitment of Atlantic cod (Gadus morhua) . Canadian Journal of Fisheries and Aquatic Sciences , 56 : 2069 – 2077 . Google Scholar CrossRef Search ADS Punt A. E. , A'mar T. , Bond N. A. , Butterworth D. S. , de Moor C. L. , De Oliveira J. A. , Haltuch M. A. , et al. 2013 . Fisheries management under climate and environmental uncertainty: control rules and performance simulation . ICES Journal of Marine Science , 71 : 2208 – 2220 . Google Scholar CrossRef Search ADS R Core Team 2016 . R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing , Vienna, Austria . http://www.R-project.org/ (last accessed 19 February 2018). Reynolds R. W. , Rayner N. A. , Smith T. M. , Stokes D. C. , Wang W. 2002 . An improved in situ and satellite SST analysis for climate . Journal of Climate , 15 : 1609 – 1625 . Google Scholar CrossRef Search ADS Rothschild B. J. 2000 . “Fish stocks and recruitment”: the past thirty years . ICES Journal of Marine Science , 57 : 191 – 201 . Google Scholar CrossRef Search ADS Sakamoto T. T. , Hasumi H. , Ishii M. , Emori S. , Suzuki T. , Nishimura T. , Sumi A. 2005 . Responses of the Kuroshio and the Kuroshio Extension to global warming in a high‐resolution climate model . Geophysical Research Letters , 32 . doi:10.1029/2005GL023384. Sakuramoto K. 2016 . Case study: a simulation model of the spawning stock biomass of Pacific bluefin tuna and evaluation of fisheries regulations . American Journal of Climate Change , 5 : 245 – 260 . Google Scholar CrossRef Search ADS Satoh K. , Tanaka Y. , Masujima M. , Okazaki M. , Kato Y. , Shono H. , Suzuki K. 2013 . Relationship between the growth and survival of larval Pacific bluefin tuna, Thunnus orientalis . Marine Biology , 160 : 691 – 702 . Google Scholar CrossRef Search ADS Satoh K. , Masujima M. , Tanaka Y. , Okazaki M. , Kato Y. , Shono H. 2014 . Transport, distribution and growth of larval patches of Pacific bluefin tuna (Thunnus orientalis) in the northwestern Pacific Ocean . Bulletin of the Fisheries Research Agency of Japan , 38 : 81 – 86 . Soeyanto E. , Guo X. , Ono J. , Miyazawa Y. 2014 . Interannual variations of Kuroshio transport in the East China Sea and its relation to the Pacific Decadal Oscillation and mesoscale eddies . Journal of Geophysical Research: Oceans , 119 : 3595 – 3616 . Google Scholar CrossRef Search ADS Stock C. A. , Pegion K. , Vecchi G. A. , Alexander M. A. , Tommasi D. , Bond N. A. , Fratantoni P. S. , et al. 2015 . Seasonal sea surface temperature anomaly prediction for coastal ecosystems . Progress in Oceanography , 137 : 219 – 236 . Google Scholar CrossRef Search ADS Szuwalski C. S. , Vert‐Pre K. A. , Punt A. E. , Branch T. A. , Hilborn R. 2015 . Examining common assumptions about recruitment: a meta-analysis of recruitment dynamics for worldwide marine fisheries . Fish and Fisheries , 16 : 633 – 648 . Google Scholar CrossRef Search ADS Tanaka Y. , Suzuki N. 2016 . Early life history. In Biology and Ecology of Bluefin Tuna , pp 19 – 46 . Eds. by Kitagawa T. , Kimura S. . CRC Press , Boca Raton, FL . Tanaka Y. , Satoh K. , Iwahashi M. , Yamada H. 2006 . Growth-dependent recruitment of Pacific bluefin tuna Thunnus orientalis in the northwestern Pacific Ocean . Marine Ecology Progress Series , 319 : 225 – 235 . Google Scholar CrossRef Search ADS Tanaka Y. , Satoh K. , Yamada H. , Takebe T. , Nikaido H. , Shiozawa S. 2008 . Assessment of the nutritional status of field-caught larval Pacific bluefin tuna by RNA/DNA ratio based on a starvation experiment of hatchery-reared fish . Journal of Experimental Marine Biology and Ecology , 354 : 56 – 64 . Google Scholar CrossRef Search ADS Tanaka Y. , Minami H. , Ishihi Y. , Kumon K. , Higuchi K. , Eba T. , Nishi A. , et al. 2014 . Relationship between prey utilization and growth variation in hatchery-reared Pacific bluefin tuna, Thunnus orientalis (Temminck et Schlegel), larvae estimated using nitrogen stable isotope analysis . Aquaculture Research , 45 : 537 – 545 . Google Scholar CrossRef Search ADS Tommasi D. , Nye J. , Stock C. , Hare J. A. , Alexander M. , Drew K. 2015 . Effect of environmental conditions on juvenile recruitment of alewife (Alosa pseudoharengus) and blueback herring (Alosa aestivalis) in fresh water: a coastwide perspective . Canadian Journal of Fisheries and Aquatic Sciences , 72 : 1037 – 1047 . Google Scholar CrossRef Search ADS Tommasi D. , Stock C. A. , Pegion K. , Vecchi G. A. , Methot R. D. , Alexander M. A. , Checkley D. M. 2017a . Improved management of small pelagic fisheries through seasonal climate prediction . Ecological Applications , 27 : 378 – 388 . Google Scholar CrossRef Search ADS Tommasi D. , Stock C. A. , Hobday A. J. , Methot R. , Kaplan I. C. , Eveson J. P. , Holsman K. , et al. 2017b . Managing living marine resources in a dynamic environment: the role of seasonal to decadal climate forecasts . Progress in Oceanography , 152 : 15 – 49 . Google Scholar CrossRef Search ADS Tommasi D. , Stock C. A. , Alexander M. A. , Yang X. , Rosati A. , Vecchi G. A. 2017c . Multi-annual climate predictions for fisheries: an assessment of skill of sea surface temperature forecasts for large marine ecosystems . Frontiers in Marine Science , 4 . doi:10.3389/fmars.2017.00201. Tsuda Y. , Sakamoto W. , Yamamoto S. , Murata O. 2012 . Effect of environmental fluctuations on mortality of juvenile Pacific bluefin tuna, Thunnus orientalis, in closed life-cycle aquaculture . Aquaculture , 330–333 : 142 – 147 . Google Scholar CrossRef Search ADS Watai M. , Ishihara T. , Abe O. , Ohshimo S. , Strussmann C. A. 2017 . Evaluation of growth-dependent survival during early stages of Pacific bluefin tuna using otolith microstructure analysis . Marine and Freshwater Research , 68 : 2008 – 2017 . Google Scholar CrossRef Search ADS WCPFC . 2015 . Conservation and management measure to establish a multi-annual rebuilding plan for Pacific bluefin tuna. Commission Twelfth regular session, Bali, Indonesia, December 2014, 3 p. Wood S. 2006 . An Introduction to Generalized Additive Models with R . Chapman , Boca Raton . Woodworth-Jefcoats P. A. , Polovina J. J. , Drazen J. C. 2017 . Climate change is projected to reduce carrying capacity and redistribute species richness in North Pacific pelagic marine ecosystems . Global Change Biology , 23 : 1000 – 1008 . Google Scholar CrossRef Search ADS PubMed Yamada H. , Takagi N. , Nishimura D. 2006 . Recruitment abundance index of Pacific bluefin tuna using fisheries data on juveniles . Fisheries Science , 72 : 333 – 341 . Google Scholar CrossRef Search ADS © International Council for the Exploration of the Sea 2018. All rights reserved. For Permissions, please email: [email protected] This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

doi: 10.1093/icesjms/fsx237pmid: N/A

Abstract Coastal ecosystems are ecologically, culturally, and economically important, and hence are under pressure from diverse human activities. We reviewed the literature for existing evidence of effects of human-induced habitat changes on exploited fish utilizing coastal habitats. We focused on fish species of the Northeast Atlantic for which fisheries advice is provided by International Council for the Exploration of the Sea (ICES) and which utilize coastal habitats for at least one life-history stage (LHS). We found that 92% of these species are impacted by human activity in at least one LHS while utilizing coastal habitat and 38% in multiple stages. Anthropogenic pressures most commonly shown to impact these fish species were toxicants and pollutants (75% of species). Eutrophication and anoxia, invasive species, and physical coastal development affected about half of the species (58, 54, and 42% of species, respectively), while indirect fishing impacts affected a minority (17% of species). Moreover, 71% of the ICES advice species that utilize coastal habitats face impacts from more than one pressure, implying cumulative effects. Given that three-fourths of the commercial landings come from fish species utilizing coastal habitats, there is an obvious need for a better understanding of the impacts that human activities cause in these habitats for the development of ecosystem-based fisheries management. Introduction Coastal habitats (as defined in Seitz et al., 2014) are valuable for numerous fish and invertebrate species, functioning as spawning grounds, juvenile growth areas, foraging areas, and migration corridors (Beck et al., 2001; Elliott and Hemingway, 2002). For example, 44% of the commercially important species for which advice is provided by the International Council for the Exploration of the Sea (ICES) in 2010 in the Northeast Atlantic have been reported to utilize coastal habitats for at least one life-history stage (LHS) for one of the above functions (Seitz et al., 2014). While coastal and estuarine areas are widely acknowledged to be of high ecological and economic value (Costanza et al., 1997; de Groot et al., 2012), they are also highly vulnerable and impacted by multiple human activities (Halpern et al., 2007, 2009; Crain et al., 2008; Batista et al., 2014; Vasconcelos et al., 2017). In temperate regions, coastal habitats such as rocky intertidal and subtidal reefs, mudflats, seagrass meadows, kelp forests, and salt marshes are exposed to high levels of anthropogenic pressures (Lotze et al., 2005; Airoldi and Beck, 2007; Halpern et al., 2008). Diverse activities - whether urban, industrial, agricultural, land reclamation, or direct exploitation of resources in the estuarine and coastal realm - often impose several pressures with cumulative impacts on coastal fish habitats (Vasconcelos et al., 2007). With human populations continuously increasing and aggregating around coastal areas worldwide (Airoldi and Beck, 2007; Kummu et al., 2016), space for human settlement and activities is often gained through land reclamation, i.e. implying the loss and fragmentation of shallow-water aquatic coastal habitats. In addition, loss, modification, and fragmentation of aquatic coastal habitats is also caused by changes to hydrological regimes, novel artificial coastal defence structures, substrate extraction (e.g. mining or dredging for maintenance of navigation canals), and disposal (e.g. coastal nourishment) (Borja et al., 2010; Peterson et al., 2014). In terms of physical perturbations, destructive fishing methods such as bottom trawling also disrupt important fish habitats (Hiddink et al., 2006). Simultaneously, degradation of coastal fish habitats can also be brought about through eutrophication and subsequent macroalgal blooms (Rabalais, 2015; Le Luherne et al., 2016) or anoxic events (Cloern, 2001) derived from nutrient input associated with urban and agricultural activities. Other terrestrial and coastal activities (e.g. industry and mining) introduce a variety of xenobiotics, which impact regular physiological processes of fish and other organisms across different trophic levels (Davis, 1999). Concurrently, the transport of people, goods, and animals often lead to the introduction of non-native species and subsequent ecological disruption (Molnar et al., 2008). Since the onset of industrialization, human activities in coastal seas and estuaries have caused successive changes in habitat quantity and quality for many key fish species (Lotze et al., 2006; Vasconcelos et al., 2007). Resulting ecological effects can be extended to the provisioning of ecosystem services, e.g. effects on the availability of fish for viable commercial and recreational fisheries or of food and habitat for protected species (Holmlund and Hammer, 1999; Lotze et al., 2006; Worm et al., 2006). In the Northeast Atlantic, coastal ecosystems are bordered by dense human populations and thus are severely affected by cumulative anthropogenic pressures (Airoldi and Beck, 2007; Halpern et al., 2009). In this region, a wide acknowledgement of the overexploitation of many commercially important species has led to the coordination of fisheries management actions at national and European levels (Lagares and Ordaz, 2014). Yet, the effects of other human-driven pressures on fish in coastal habitats of this region have been poorly collectively evaluated or insufficiently taken into consideration in species conservation or marine resource management plans (Kempf, 2010; Jennings and Rice, 2011). These effects must be estimated, according to the European Marine Strategy Framework Directive (MSFD; 2008/56/EC), so that measures can be taken to achieve “good environmental status”, by 2020. In this study, we aim to make a critical assessment of the impacts and perturbations in coastal habitats on different life-history stages of commercially important fish. We focus on fish populations in the Northeast Atlantic for which fisheries advice is provided by ICES (henceforth referred to as ICES advice species). To achieve our aim, we build upon the seminal paper by Seitz et al. (2014) by reviewing existing literature to find evidence of impacts (negative or positive) on commercially important fish species from coastal habitat changes caused by human activities. We discuss our results in relation to ongoing improvements in quantifying human impacts on coastal fish. We then make suggestions for future research and propose potential avenues for incorporating fish habitat considerations into ecosystem-based fisheries management. Methods We searched existing primary literature to assess the extent to which ICES advice fish species are impacted by human activities in coastal habitats. The aim here was to find evidence of impacts (i) across species, (ii) by coastal LHS (i.e. juvenile, feeding, spawning, migration; Figure 1), and (iii) across different sources of anthropogenic pressures. Species investigated were those for which ICES provides advice. The coastal use of different life-history stages was taken from Seitz et al. (2014) and updated using their methodology (see Supplementary Material). To differentiate between anthropogenic pressures, we placed the most commonly occurring pressures into five categories (Table 1). The “Physical Coastal Development” category deals with both physical changes to the aquatic environment (e.g. human-induced changes in surface sediment properties) and loss of area (e.g. marina construction or land reclamation). The “Eutrophication and Anoxia” category deals with changes in nutrient and oxygen concentrations. The “Toxicants and Pollutants” category deals with toxic substances and xenobiotics present or entering the coastal habitat as a result of human activity. The “Invasive Species” category refers to introduced, non-native species that become abundant and alter ecosystem structures and functions. The “Indirect Fishing Impacts” category excludes direct fishing mortality but deals with effects such as physical disturbance of the seabed, destruction of reefs, or changes in community structure through the removal of key species. Table 1. Keywords used to define habitat degradation in review of human impacts on fish using coastal habitats in different life-history stages. Anthropogenic pressure Relevant keywords used in search Toxicants and Pollutants “xenobiotics” or “toxic* or “sewage” or “contaminant” or “pollution” Eutrophication and Anoxia “eutrophication” or “hypoxia” Invasive Species “invasi*” or “outbreak” or “proliferation” Physical Coastal Development “land reclamation” or “sediment” or “habitat loss” or “extraction” or “depth change” Indirect Fishing Impacts “fishing” or “trawling” or “dredging” Anthropogenic pressure Relevant keywords used in search Toxicants and Pollutants “xenobiotics” or “toxic* or “sewage” or “contaminant” or “pollution” Eutrophication and Anoxia “eutrophication” or “hypoxia” Invasive Species “invasi*” or “outbreak” or “proliferation” Physical Coastal Development “land reclamation” or “sediment” or “habitat loss” or “extraction” or “depth change” Indirect Fishing Impacts “fishing” or “trawling” or “dredging” Table 1. Keywords used to define habitat degradation in review of human impacts on fish using coastal habitats in different life-history stages. Anthropogenic pressure Relevant keywords used in search Toxicants and Pollutants “xenobiotics” or “toxic* or “sewage” or “contaminant” or “pollution” Eutrophication and Anoxia “eutrophication” or “hypoxia” Invasive Species “invasi*” or “outbreak” or “proliferation” Physical Coastal Development “land reclamation” or “sediment” or “habitat loss” or “extraction” or “depth change” Indirect Fishing Impacts “fishing” or “trawling” or “dredging” Anthropogenic pressure Relevant keywords used in search Toxicants and Pollutants “xenobiotics” or “toxic* or “sewage” or “contaminant” or “pollution” Eutrophication and Anoxia “eutrophication” or “hypoxia” Invasive Species “invasi*” or “outbreak” or “proliferation” Physical Coastal Development “land reclamation” or “sediment” or “habitat loss” or “extraction” or “depth change” Indirect Fishing Impacts “fishing” or “trawling” or “dredging” Figure 1. View largeDownload slide Conceptual diagram of common life-history stages of fish in coastal habitats: S = mature adults during spawning, J = immature juveniles, and F = feeding adults not in spawning. Arrows represent migrations (M). Figure 1. View largeDownload slide Conceptual diagram of common life-history stages of fish in coastal habitats: S = mature adults during spawning, J = immature juveniles, and F = feeding adults not in spawning. Arrows represent migrations (M). We compiled relevant scientific literature linking habitat degradation to coastal life-history stages. A database search initially utilized Google Scholar on 21 July 2015 combining species’ name (both binomial and common) with keywords relevant to different categories of habitat degradation (Table 1). This list was updated, and the collection increased using searches of the same format later in 2015 and from authors’ own literature databases. We evaluated the results of this search to compile a three-dimensional inventory of evidence of impacts for each species, each LHS, and each pressure (24 × 4 × 5 = 480, respectively). Once evidence of impact was found for a given inventory position, the search moved on to the next position. The amount of research found and the magnitude of impact effects were not quantified; hence, the result was a binary table (evidence or no evidence). From this inventory of evidence, we calculated the relative proportions of impacted species and life-history stages and of the respective category of anthropogenic pressure involved. Results Evidence of human activities impacting ICES advice species in coastal habitats was collated by category of anthropogenic pressure and LHS (Table 2). Considering all 5 impact categories, 4 life-history stages, and 24 species utilizing coastal habitat, a total of 58 occurrences of anthropogenic impact were found (Table 2). Of these 24 fish species, 92% (22) were impacted by at least one category of human activity in one or more life-history stages in coastal habitats. The category of impacts most commonly linked to fish in coastal habitats was “Toxicants and Pollutants”, with 75% (18/24) of species having evidence of being impacted in at least one LHS (Figure 2). “Eutrophication and Anoxia” and “Invasive Species” both affected over half of species with coastal habitat use with 58% (14/24) and 54% (13/24), respectively. “Physical Coastal Development” impacted 42% (11/24) of species, while least commonly documented was the “Indirect Fishing Impacts” category where 17% (4/24) of ICES advice species utilizing the coast were shown to be linked to these types of impacts. Moreover, 71% (17/24) of species face impacts from more than one pressure (Table 2). Table 2. Anthropogenic pressures impacting commercially important fish species in coastal habitats for which ICES provides advice. Species Common name Anthropogenic pressure Toxicants and Pollutants Eutrophication and Anoxia Invasive Species Physical Coastal Development Indirect Fishing Impacts References Engraulis encrasicolus Anchovy S J F Tsikhon-Lykanina and Reznitchenko (1991), Kideys (1994), Niermann et al. (1994), Shiganova et al. (2001), Drake et al. (2007), Oguz et al. (2008), Ruiz et al. (2009), Ekau et al. (2010) Cetorhinus maximus Basking shark F Fossi et al. (2014) Scophthalmus rhombus Brill J Kostecki et al., (2011) Mallotus villosus Capelin S Frantzen et al. (2012) Gadus morhua Cod F J F F F J Isaksson et al. (1994), Lindholm et al. (1999), Hall-Spencer et al. (2003), Johansen et al. (2006, 2011), Österblom et al. (2007), Patel et al. (2007), Teschner et al. (2010), Thurstan et al. (2010), Malovic et al. (2010), Bratberg et al. (2013), Reubens et al. (2013), Johannessen (2014) Limanda limanda Dab F J J Berghahn et al. (1992), Houlihan et al. (1994), Petersen and Pihl (1995), Power et al. (2000) Anguilla anguilla Eel M M M Feunteun (2002), Palstra et al. (2006), Maes et al. (2007) Dicentrarchus labrax European sea bass J J Laffaille et al. (2000), Reynolds et al. (2003), Kerambrun et al. (2012b) Platichthys flesus Flounder J J J Carl et al. (2008), Amara et al. (2009), Kostecki et al. (2011), Jokinen et al. (2016) Clupea harengus Herring S S S S Johnston and Wildish (1981), Winters et al. (1986), Aneer (1987), Costello and Gamble (1992), Kornilovs (1993), Lappalainen and Pesonen (2000), Gorokhova et al. (2004, 2005), McIntosh et al. (2010), Frantzen et al. (2015) Scomber scombrus Mackerel S J Longwell et al. (1992), Öztürk (2006) Trisopterus esmarkii Norway pout Pleuronectes platessa Plaice J F J J J J Petersen and Pihl (1995), Secombes et al. (1995), Pihl et al. (2005), van de Wolfshaar et al. (2011) Pollachius pollachius Pollack J Johannessen (2014) Pollachius virens Saithe J J J J Bordehore et al. (2003), Kamenos et al. (2004), Olsen et al. (2010), Falk-Petersen et al. (2011), Støttrup et al. (2014) Salmo salar Salmon M M Alabaster et al. (1991), Tentelier and Piou (2011), Martignac et al. (2013) Ammodytes marinus Sandeel Sardina pilchardus Sardine F F Canli and Atli (2003), Goren and Galil (2005) Salmo trutta Sea trout F M F M Olsson et al. (2001), Meland et al. (2010), Ilarri et al. (2014), Taranger et al. (2015) Solea solea Sole J F J J J Via et al. (1998), Lefrançois and Claireaux (2003), Le Pape et al. (2004), Gilliers et al. (2006), Amara et al. (2007), Davoodi and Claireaux (2007), Jenkinson et al. (2007), Rochette et al. (2010), Kostecki et al. (2011), Jebali et al. (2013) Sprattus sprattus Sprat S J J F S J Parmanne et al. (1994), Cameron and Von Westernhagen (1997), Shiganova and Bulgakova (2000), Peters et al. (2001), Österblom et al. (2007) Mullus surmuletus Striped red mullet F F Levi and Francour (2004), Bianchi et al. (2014), Scopelliti et al. (2015) Scophtalmus maximus Turbot J Kerambrun et al. (2012a, 2012b) Merlangius merlangus Whiting S F J Westernhagen et al. (1989), Shiganova and Bulgakova (2000), Turnpenny and Taylor (2000), Ogut and Palm (2005), Pihl et al. (2006) Percentage of species 75% 58% 54% 42% 17% Species Common name Anthropogenic pressure Toxicants and Pollutants Eutrophication and Anoxia Invasive Species Physical Coastal Development Indirect Fishing Impacts References Engraulis encrasicolus Anchovy S J F Tsikhon-Lykanina and Reznitchenko (1991), Kideys (1994), Niermann et al. (1994), Shiganova et al. (2001), Drake et al. (2007), Oguz et al. (2008), Ruiz et al. (2009), Ekau et al. (2010) Cetorhinus maximus Basking shark F Fossi et al. (2014) Scophthalmus rhombus Brill J Kostecki et al., (2011) Mallotus villosus Capelin S Frantzen et al. (2012) Gadus morhua Cod F J F F F J Isaksson et al. (1994), Lindholm et al. (1999), Hall-Spencer et al. (2003), Johansen et al. (2006, 2011), Österblom et al. (2007), Patel et al. (2007), Teschner et al. (2010), Thurstan et al. (2010), Malovic et al. (2010), Bratberg et al. (2013), Reubens et al. (2013), Johannessen (2014) Limanda limanda Dab F J J Berghahn et al. (1992), Houlihan et al. (1994), Petersen and Pihl (1995), Power et al. (2000) Anguilla anguilla Eel M M M Feunteun (2002), Palstra et al. (2006), Maes et al. (2007) Dicentrarchus labrax European sea bass J J Laffaille et al. (2000), Reynolds et al. (2003), Kerambrun et al. (2012b) Platichthys flesus Flounder J J J Carl et al. (2008), Amara et al. (2009), Kostecki et al. (2011), Jokinen et al. (2016) Clupea harengus Herring S S S S Johnston and Wildish (1981), Winters et al. (1986), Aneer (1987), Costello and Gamble (1992), Kornilovs (1993), Lappalainen and Pesonen (2000), Gorokhova et al. (2004, 2005), McIntosh et al. (2010), Frantzen et al. (2015) Scomber scombrus Mackerel S J Longwell et al. (1992), Öztürk (2006) Trisopterus esmarkii Norway pout Pleuronectes platessa Plaice J F J J J J Petersen and Pihl (1995), Secombes et al. (1995), Pihl et al. (2005), van de Wolfshaar et al. (2011) Pollachius pollachius Pollack J Johannessen (2014) Pollachius virens Saithe J J J J Bordehore et al. (2003), Kamenos et al. (2004), Olsen et al. (2010), Falk-Petersen et al. (2011), Støttrup et al. (2014) Salmo salar Salmon M M Alabaster et al. (1991), Tentelier and Piou (2011), Martignac et al. (2013) Ammodytes marinus Sandeel Sardina pilchardus Sardine F F Canli and Atli (2003), Goren and Galil (2005) Salmo trutta Sea trout F M F M Olsson et al. (2001), Meland et al. (2010), Ilarri et al. (2014), Taranger et al. (2015) Solea solea Sole J F J J J Via et al. (1998), Lefrançois and Claireaux (2003), Le Pape et al. (2004), Gilliers et al. (2006), Amara et al. (2007), Davoodi and Claireaux (2007), Jenkinson et al. (2007), Rochette et al. (2010), Kostecki et al. (2011), Jebali et al. (2013) Sprattus sprattus Sprat S J J F S J Parmanne et al. (1994), Cameron and Von Westernhagen (1997), Shiganova and Bulgakova (2000), Peters et al. (2001), Österblom et al. (2007) Mullus surmuletus Striped red mullet F F Levi and Francour (2004), Bianchi et al. (2014), Scopelliti et al. (2015) Scophtalmus maximus Turbot J Kerambrun et al. (2012a, 2012b) Merlangius merlangus Whiting S F J Westernhagen et al. (1989), Shiganova and Bulgakova (2000), Turnpenny and Taylor (2000), Ogut and Palm (2005), Pihl et al. (2006) Percentage of species 75% 58% 54% 42% 17% Evidence of impacts at different life-history stages is indicated by J (juvenile), F (feeding), S (spawning), and M (migration). Table 2. Anthropogenic pressures impacting commercially important fish species in coastal habitats for which ICES provides advice. Species Common name Anthropogenic pressure Toxicants and Pollutants Eutrophication and Anoxia Invasive Species Physical Coastal Development Indirect Fishing Impacts References Engraulis encrasicolus Anchovy S J F Tsikhon-Lykanina and Reznitchenko (1991), Kideys (1994), Niermann et al. (1994), Shiganova et al. (2001), Drake et al. (2007), Oguz et al. (2008), Ruiz et al. (2009), Ekau et al. (2010) Cetorhinus maximus Basking shark F Fossi et al. (2014) Scophthalmus rhombus Brill J Kostecki et al., (2011) Mallotus villosus Capelin S Frantzen et al. (2012) Gadus morhua Cod F J F F F J Isaksson et al. (1994), Lindholm et al. (1999), Hall-Spencer et al. (2003), Johansen et al. (2006, 2011), Österblom et al. (2007), Patel et al. (2007), Teschner et al. (2010), Thurstan et al. (2010), Malovic et al. (2010), Bratberg et al. (2013), Reubens et al. (2013), Johannessen (2014) Limanda limanda Dab F J J Berghahn et al. (1992), Houlihan et al. (1994), Petersen and Pihl (1995), Power et al. (2000) Anguilla anguilla Eel M M M Feunteun (2002), Palstra et al. (2006), Maes et al. (2007) Dicentrarchus labrax European sea bass J J Laffaille et al. (2000), Reynolds et al. (2003), Kerambrun et al. (2012b) Platichthys flesus Flounder J J J Carl et al. (2008), Amara et al. (2009), Kostecki et al. (2011), Jokinen et al. (2016) Clupea harengus Herring S S S S Johnston and Wildish (1981), Winters et al. (1986), Aneer (1987), Costello and Gamble (1992), Kornilovs (1993), Lappalainen and Pesonen (2000), Gorokhova et al. (2004, 2005), McIntosh et al. (2010), Frantzen et al. (2015) Scomber scombrus Mackerel S J Longwell et al. (1992), Öztürk (2006) Trisopterus esmarkii Norway pout Pleuronectes platessa Plaice J F J J J J Petersen and Pihl (1995), Secombes et al. (1995), Pihl et al. (2005), van de Wolfshaar et al. (2011) Pollachius pollachius Pollack J Johannessen (2014) Pollachius virens Saithe J J J J Bordehore et al. (2003), Kamenos et al. (2004), Olsen et al. (2010), Falk-Petersen et al. (2011), Støttrup et al. (2014) Salmo salar Salmon M M Alabaster et al. (1991), Tentelier and Piou (2011), Martignac et al. (2013) Ammodytes marinus Sandeel Sardina pilchardus Sardine F F Canli and Atli (2003), Goren and Galil (2005) Salmo trutta Sea trout F M F M Olsson et al. (2001), Meland et al. (2010), Ilarri et al. (2014), Taranger et al. (2015) Solea solea Sole J F J J J Via et al. (1998), Lefrançois and Claireaux (2003), Le Pape et al. (2004), Gilliers et al. (2006), Amara et al. (2007), Davoodi and Claireaux (2007), Jenkinson et al. (2007), Rochette et al. (2010), Kostecki et al. (2011), Jebali et al. (2013) Sprattus sprattus Sprat S J J F S J Parmanne et al. (1994), Cameron and Von Westernhagen (1997), Shiganova and Bulgakova (2000), Peters et al. (2001), Österblom et al. (2007) Mullus surmuletus Striped red mullet F F Levi and Francour (2004), Bianchi et al. (2014), Scopelliti et al. (2015) Scophtalmus maximus Turbot J Kerambrun et al. (2012a, 2012b) Merlangius merlangus Whiting S F J Westernhagen et al. (1989), Shiganova and Bulgakova (2000), Turnpenny and Taylor (2000), Ogut and Palm (2005), Pihl et al. (2006) Percentage of species 75% 58% 54% 42% 17% Species Common name Anthropogenic pressure Toxicants and Pollutants Eutrophication and Anoxia Invasive Species Physical Coastal Development Indirect Fishing Impacts References Engraulis encrasicolus Anchovy S J F Tsikhon-Lykanina and Reznitchenko (1991), Kideys (1994), Niermann et al. (1994), Shiganova et al. (2001), Drake et al. (2007), Oguz et al. (2008), Ruiz et al. (2009), Ekau et al. (2010) Cetorhinus maximus Basking shark F Fossi et al. (2014) Scophthalmus rhombus Brill J Kostecki et al., (2011) Mallotus villosus Capelin S Frantzen et al. (2012) Gadus morhua Cod F J F F F J Isaksson et al. (1994), Lindholm et al. (1999), Hall-Spencer et al. (2003), Johansen et al. (2006, 2011), Österblom et al. (2007), Patel et al. (2007), Teschner et al. (2010), Thurstan et al. (2010), Malovic et al. (2010), Bratberg et al. (2013), Reubens et al. (2013), Johannessen (2014) Limanda limanda Dab F J J Berghahn et al. (1992), Houlihan et al. (1994), Petersen and Pihl (1995), Power et al. (2000) Anguilla anguilla Eel M M M Feunteun (2002), Palstra et al. (2006), Maes et al. (2007) Dicentrarchus labrax European sea bass J J Laffaille et al. (2000), Reynolds et al. (2003), Kerambrun et al. (2012b) Platichthys flesus Flounder J J J Carl et al. (2008), Amara et al. (2009), Kostecki et al. (2011), Jokinen et al. (2016) Clupea harengus Herring S S S S Johnston and Wildish (1981), Winters et al. (1986), Aneer (1987), Costello and Gamble (1992), Kornilovs (1993), Lappalainen and Pesonen (2000), Gorokhova et al. (2004, 2005), McIntosh et al. (2010), Frantzen et al. (2015) Scomber scombrus Mackerel S J Longwell et al. (1992), Öztürk (2006) Trisopterus esmarkii Norway pout Pleuronectes platessa Plaice J F J J J J Petersen and Pihl (1995), Secombes et al. (1995), Pihl et al. (2005), van de Wolfshaar et al. (2011) Pollachius pollachius Pollack J Johannessen (2014) Pollachius virens Saithe J J J J Bordehore et al. (2003), Kamenos et al. (2004), Olsen et al. (2010), Falk-Petersen et al. (2011), Støttrup et al. (2014) Salmo salar Salmon M M Alabaster et al. (1991), Tentelier and Piou (2011), Martignac et al. (2013) Ammodytes marinus Sandeel Sardina pilchardus Sardine F F Canli and Atli (2003), Goren and Galil (2005) Salmo trutta Sea trout F M F M Olsson et al. (2001), Meland et al. (2010), Ilarri et al. (2014), Taranger et al. (2015) Solea solea Sole J F J J J Via et al. (1998), Lefrançois and Claireaux (2003), Le Pape et al. (2004), Gilliers et al. (2006), Amara et al. (2007), Davoodi and Claireaux (2007), Jenkinson et al. (2007), Rochette et al. (2010), Kostecki et al. (2011), Jebali et al. (2013) Sprattus sprattus Sprat S J J F S J Parmanne et al. (1994), Cameron and Von Westernhagen (1997), Shiganova and Bulgakova (2000), Peters et al. (2001), Österblom et al. (2007) Mullus surmuletus Striped red mullet F F Levi and Francour (2004), Bianchi et al. (2014), Scopelliti et al. (2015) Scophtalmus maximus Turbot J Kerambrun et al. (2012a, 2012b) Merlangius merlangus Whiting S F J Westernhagen et al. (1989), Shiganova and Bulgakova (2000), Turnpenny and Taylor (2000), Ogut and Palm (2005), Pihl et al. (2006) Percentage of species 75% 58% 54% 42% 17% Evidence of impacts at different life-history stages is indicated by J (juvenile), F (feeding), S (spawning), and M (migration). Figure 2. View largeDownload slide Percentage (by number) of ICES advice species that utilize the coastal zone and have evidence of being impacted by anthropogenic pressures separated by pressure category. Figure 2. View largeDownload slide Percentage (by number) of ICES advice species that utilize the coastal zone and have evidence of being impacted by anthropogenic pressures separated by pressure category. When considering individual life-history stages (Table 3), 54% (13/24) of fish species utilizing coastal habitats were impacted in only one LHS, while 38% (9/24) had evidence of two or more life-history stages being impacted. If we consider only those species utilizing coastal habitats for two or more life-history stages (15/24), then 60% (9/15) of these had evidence of being impacted. For two species, Norway pout (Trisopterus esmarkii) and sandeel (Ammodytes spp.), no evidence of anthropogenic impacts was found in the literature. Table 3. ICES advice species showing coastal habitat use. Life history stage Common name Binomial classification Juvenile Feeding Spawning Migration Anchovy Engraulis encrasicolus X X X Basking shark Cetorhinus maximus X Brill Scophthalmus rhombus X Capelin Mallotus villosus X Cod Gadus morhua X X Dab Limanda limanda X X Eel Anguilla anguilla X European sea bass Dicentrarchus labrax X Flounder Platichthys flesus X Herring Clupea harengus X Mackerel Scomber scombrus X X Norway pout Trisopterus esmarkii Plaice Pleuronectes platessa X X Pollock Pollachius pollachius X Saithe Pollachius virens X Atlantic salmon Salmo salar X Sandeel Hyperoplus spp./Ammodytes spp. Sardine Sardina pilchardus X Sea trout Salmo trutta X X Sole Solea solea X X Sprat Sprattus sprattus X X X Striped red mullet Mullus surmuletus X Turbot Scophthalmus maximus X Whiting Merlangus merlangus X X X Percentage of species with evidence of impact for life-history stages that utilize coastal habitat 78% 69% 67% 50% Life history stage Common name Binomial classification Juvenile Feeding Spawning Migration Anchovy Engraulis encrasicolus X X X Basking shark Cetorhinus maximus X Brill Scophthalmus rhombus X Capelin Mallotus villosus X Cod Gadus morhua X X Dab Limanda limanda X X Eel Anguilla anguilla X European sea bass Dicentrarchus labrax X Flounder Platichthys flesus X Herring Clupea harengus X Mackerel Scomber scombrus X X Norway pout Trisopterus esmarkii Plaice Pleuronectes platessa X X Pollock Pollachius pollachius X Saithe Pollachius virens X Atlantic salmon Salmo salar X Sandeel Hyperoplus spp./Ammodytes spp. Sardine Sardina pilchardus X Sea trout Salmo trutta X X Sole Solea solea X X Sprat Sprattus sprattus X X X Striped red mullet Mullus surmuletus X Turbot Scophthalmus maximus X Whiting Merlangus merlangus X X X Percentage of species with evidence of impact for life-history stages that utilize coastal habitat 78% 69% 67% 50% Updated from Seitz et al., 2014 (see Supplementary Material) in different life-history stages (shaded cells) and those life-history stages found (this study) to be impacted upon by human activities (marked by X) Table 3. ICES advice species showing coastal habitat use. Life history stage Common name Binomial classification Juvenile Feeding Spawning Migration Anchovy Engraulis encrasicolus X X X Basking shark Cetorhinus maximus X Brill Scophthalmus rhombus X Capelin Mallotus villosus X Cod Gadus morhua X X Dab Limanda limanda X X Eel Anguilla anguilla X European sea bass Dicentrarchus labrax X Flounder Platichthys flesus X Herring Clupea harengus X Mackerel Scomber scombrus X X Norway pout Trisopterus esmarkii Plaice Pleuronectes platessa X X Pollock Pollachius pollachius X Saithe Pollachius virens X Atlantic salmon Salmo salar X Sandeel Hyperoplus spp./Ammodytes spp. Sardine Sardina pilchardus X Sea trout Salmo trutta X X Sole Solea solea X X Sprat Sprattus sprattus X X X Striped red mullet Mullus surmuletus X Turbot Scophthalmus maximus X Whiting Merlangus merlangus X X X Percentage of species with evidence of impact for life-history stages that utilize coastal habitat 78% 69% 67% 50% Life history stage Common name Binomial classification Juvenile Feeding Spawning Migration Anchovy Engraulis encrasicolus X X X Basking shark Cetorhinus maximus X Brill Scophthalmus rhombus X Capelin Mallotus villosus X Cod Gadus morhua X X Dab Limanda limanda X X Eel Anguilla anguilla X European sea bass Dicentrarchus labrax X Flounder Platichthys flesus X Herring Clupea harengus X Mackerel Scomber scombrus X X Norway pout Trisopterus esmarkii Plaice Pleuronectes platessa X X Pollock Pollachius pollachius X Saithe Pollachius virens X Atlantic salmon Salmo salar X Sandeel Hyperoplus spp./Ammodytes spp. Sardine Sardina pilchardus X Sea trout Salmo trutta X X Sole Solea solea X X Sprat Sprattus sprattus X X X Striped red mullet Mullus surmuletus X Turbot Scophthalmus maximus X Whiting Merlangus merlangus X X X Percentage of species with evidence of impact for life-history stages that utilize coastal habitat 78% 69% 67% 50% Updated from Seitz et al., 2014 (see Supplementary Material) in different life-history stages (shaded cells) and those life-history stages found (this study) to be impacted upon by human activities (marked by X) We found that 69% (34/49, i.e. total no. crosses/total no. shaded cells in Table 3) of life-history stages utilizing coastal habitats have documented evidence of being impacted by human activity. Within each LHS utilizing coastal habitat (Figure 3), the juvenile LHS had the most evidence of being impacted both by proportion (78%) and absolute number (14/18; Table 3). The feeding LHS was impacted at a slightly lower rate of 69% (11/16), while two-thirds (6/9) of species utilizing coastal habitats for spawning had evidence of human impacts. The migration LHS had the fewest species with evidence of impacts from human activity, where 50% (3/6) of the occasions of coastal utilization were impacted. Figure 3. View largeDownload slide Conceptual diagram of common life-history stages of fish in coastal habitats displaying the percentage of impacted species at each LHS (S = mature adults during spawning, J = immature juveniles, and F = feeding adults not in spawning). Arrows represent migrations (M). Figure 3. View largeDownload slide Conceptual diagram of common life-history stages of fish in coastal habitats displaying the percentage of impacted species at each LHS (S = mature adults during spawning, J = immature juveniles, and F = feeding adults not in spawning). Arrows represent migrations (M). Discussion Coastal habitats experience a large variety and extent of anthropogenic pressures (Lotze et al., 2006; Airoldi and Beck, 2007; Halpern et al., 2008; Vasconcelos et al., 2017); hence, the many fish species utilizing these habitats across different parts of their LHS may also be impacted by human activities. Based on the occurrence of reported impacts, this study highlights the current evidence of such impacts across the Northeast Atlantic. A major conclusion of this review is that there is a large body of evidence linking a variety of human activities and habitat degradation to impacts on commercially important fishes utilizing coastal habitats. In fact, 92% of ICES fish species utilizing coastal habitats were found to be impacted by at least one pressure in these habitats. Species with certain life-history stages exhibiting a strong dependence on specific habitat types are especially at risk of habitat loss and degradation, which is usually the case for early life-history stages (i.e. juvenile LHS in the present study; Mumby et al., 2004; Seitz et al., 2014; Sundblad et al., 2014). While our assessment targets only the presence or absence of impacts for each species and LHS, some commonalities are to be mentioned concerning the effects and mechanisms of the different impact categories on fish utilizing coastal habitats. Eutrophication, for example, was found to impair spawning and recruitment via periodic increases in primary production from certain algal species (Isaksson et al., 1994; Carl et al., 2008). Although mild eutrophication was also found to increase the productivity of some systems (Parmanne et al., 1994; Österblom et al., 2007), higher nutrient loads may negatively impact fish populations by decreasing the cover of canopy forming vegetation (Cloern, 2001) and increasing the frequency of anoxic events (Rabalais, 2015), in turn causing increases in the mortality of early life-history stages and reducing growth rates in fish (Kornilovs, 1993; Petersen and Pihl, 1995; Maes et al., 2007; Teschner et al., 2010). In addition to the widely known direct effects of fishing on fish species, this study found that fishing often indirectly impacted fish through the destruction of biogenic structures which provide shelter (Hall-Spencer et al., 2003; Kamenos et al., 2004). Furthermore, trawling also causes physical damage to abiogenic bottom habitats causing changes in benthic invertebrate communities (Sciberras and Hiddink, 2014; van Denderen et al., 2014; Rijnsdorp et al., 2016), which can be expected to affect fish populations as they are important prey for fish (Henriques et al., 2014). Invasive species were mostly noted to either directly exploit or compete for food with resident species (Öztürk, 2006; Oguz et al., 2008) or otherwise alter the structure of the habitat (Pihl et al., 2005). Physical development was reported to lead to reduced productivity via dams altering terrestrial water discharge regimes in nursery areas (Drake et al., 2002) and the direct removal of habitat via dredging, diking, and harbour extensions (Rochette et al., 2010). However, other physical developments may lead to habitat creation due to artificial hard substrates providing shelter and feeding opportunities (Reubens et al., 2013). Impacts from xenobiotics ranged from egg mortality through decreased growth rates to reduced migration success (Houlihan et al., 1994; Feunteun, 2002; Kerambrun et al., 2012a), all of which lead to lowered fish survival (Gilliers et al., 2006; Le Pape et al., 2007) and density (Courrat et al., 2009; Delpech et al., 2010). In reality, the impacts that fish experience in coastal habitats are not acting in isolation, but interact to form a complex combination of cumulative impacts (Lotze et al., 2006; Halpern et al., 2007; Vasconcelos et al., 2007). In coastal habitats, fish can face cumulative impacts both in terms of different anthropogenic pressures and in terms of exposure during more than one LHS. In this study, 71% of species had evidence of exposure to multiple impact categories. A good example of this is the plaice (Pleuronectes platessa), where the juvenile LHS had a reported impact from all five impact categories examined in this study. Similarly, sprat (Sprattus sprattus) juveniles are faced with impacts from three impact categories: “Eutrophication and Anoxia”, “Invasive Species”, and “Physical Coastal Development”. Furthermore, there is evidence of human impacts affecting sprat across three of the four life-history stages illustrating additional cumulative impacts over time. Similar cumulative impacts acting across different life-history stages are faced by 38% of the species, without considering human impacts experienced in offshore habitats. Impacts imposed at a given LHS may continue to have an effect throughout the remainder of the life cycle (Schmidt et al., 2012). For example, exposure to xenobiotics at the juvenile LHS, affecting juvenile growth and survival (Gilliers et al., 2006), also impacts adult fitness (Jonsson and Jonsson, 2014). The cumulative impacts from these multiple stressors can be additive, synergistic, or antagonistic (for definitions, see Crain et al., 2008; Piggott et al., 2015). In marine ecosystems, cumulative impacts are often synergistic (Crain et al., 2008). In coastal marine systems, more than half of combined impacts are derived from non-additive interactions, meaning simple additive approaches are often insufficient to adequately describe or predict impacts (Teichert et al., 2016). A simple illustration of synergistic cumulative impacts is the multiplicative effect of habitat loss and the simultaneous degradation of habitat quality in residual habitats (Archambault et al., 2015). So-called positively synergistic stressors, where multiple pressures amplify the stress on species, will have the largest impact, potentially altering entire assemblages (Tomczak et al., 2013). The identification and mitigation of such combinations of anthropogenic pressures and effects along successive life-history stages will also have the largest scope for restoration benefits (Piggott et al., 2015). The results of the current work provide some guidance with regard to species, different life-history stages, and the types of pressures to be considered in cumulative impact studies. Knowledge of these impacts and their interrelated effects on fish populations needs to be quantified to be useful in management. An integrated population model (IPM) that quantifies key demographic rates across different life-history stages has proven successful in describing population changes and trends related to environmental drivers (Deegan, 1990; Fodrie et al., 2009). Like all modelling approaches, IPMs can only create good approximations of reality when informed with accurate parameters, of which IPMs require many (Koons et al., 2017). Because of the complexity involved in linking many submodels together, many constructions have excluded important drivers (Klanjšček and Legović, 2007), used explicit assumptions (Levin and Stunz, 2005), or failed to consider the assumptions of underlying models (Anderson, 2005), thus reducing the applicability to scenario-based predictions. This is not to say that such endeavours are not warranted; to the contrary, they can act as informed thought experiments in probing ecological questions (Levin and Stunz, 2005). However, such approaches should be iterative (Rochette et al., 2013; Archambault et al., 2015, 2016) and considered with an appropriate level of criticism until research in the system they aim to describe matures and provides ample and accurate parameters to inform the model (Meynecke and Richards, 2014; Archambault et al., 2016). In the short term, approaches that reduce the number of parameters in a model can provide better predictions of population changes (Ruiz et al., 2009). However, in complex systems with multiple pressures, the exclusion of certain drivers will neglect both their direct impact and any cumulative impacts on the population. Studies that include more and more accurately parameterized drivers in constructing IPMs can provide better insight into how cumulative impacts affect populations and hence better inform management of these stocks and resources (van de Wolfshaar et al., 2011). Taking this approach a step further involves applying IPMs from mature research areas to spatially explicit contexts (e.g. Stelzenmüller et al., 2011; van de Wolfshaar et al., 2011; Meynecke and Richards, 2014; Archambault et al., 2015; Rahikainen et al., 2017). To properly parameterize and to make these approaches more accurate and ubiquitous, research focus and investment should be made in empirical studies to link drivers to population demographic rates at specific life-history stages. With knowledge of the population-level effects of cumulative anthropogenic pressures, trade-offs of impacts under alternative scenarios of use can be investigated. Information on the sensitivity of different habitats and fish populations across their entire life history to individual pressures may then be represented by impact scores and mapped, while cumulative effects are subsequently considered (Halpern et al., 2008; Foden et al., 2011; Andersen et al., 2015). Such methods of estimating cumulative impacts on ecosystems are currently being developed for marine management and spatial planning purposes (Goodsir et al., 2015; Knights et al., 2015). Providing authorities with fish habitat maps together with quantitative information on the effects of human pressures would be an important step towards securing long-term sustainability of these coastal habitats and the ecosystem services they provide. While the development of these mapping methods continues, there is a parallel need to develop ways to integrate this new knowledge with fisheries management and maritime spatial planning (EU Directive MSP 2014/89/EU). Our study indicates a high variety of how and where, within a species’ life history, anthropogenic pressures in coastal habitats may impact commercially important fish. It also discusses the development of methods to investigate population effects of such impacts and highlights knowledge gaps in this area of research. Considering that the majority of commercial landings come from fish utilizing coastal habitats (ca. 75%) (Champbers, 1991—cited in Fodrie and Mendoza, 2006; Seitz et al., 2014; Supplementary Material), it is of great importance that the impacts of human activities in coastal areas are understood and accounted for in the context of ecosystem-based fisheries management. Supplementary data Supplementary material is available at the ICESJMS online version of the manuscript. Acknowledgements This work was developed within the context of the ICES Working Group on the Value of Coastal Habitats for Exploited Species (WGVHES). We thank both ICES and all participants of the Working Group during 2014–2017. Contributors to this research were independently funded and supported: EJB was funded by the Danish Recreational Fishers Fund - Marine Fiskepleje and the Otto Mønsted Fund; RPV was financed through Fundação para a Ciência e a Tecnologia (FCT) via Investigador FCT Programme 2013 (IF/00058/2013), project PTDC/AAG-GLO/5849/2014 and project UID/MAR/04292/, and through European Commission via National Programme for Biological Sampling (PNAB) integrated in the Data Collection Framework, and KvdW was supported by Wageningen Marine Research. The authors also greatly appreciate detailed and constructive feedback provided by the anonymous reviewers. The views and opinions expressed in this publication are the sole responsibility of the authors and do not necessarily reflect the views of their institutions or funding agencies. References Airoldi L. , Beck M. W. 2007 . Loss, status and trends for coastal marine habitats of Europe . Oceanography and Marine Biology: An Annual Review 45 : 345 – 405 . Alabaster J. S. , Gough P. J. , Brooker W. J. 1991 . The environmental requirements of Atlantic salmon, Salmo salar L., during their passage through the Thames Estuary, 1982-1989 . Journal of Fish Biology 38 : 741 – 762 . Google Scholar CrossRef Search ADS Amara R. , Meziane T. , Gilliers C. , Hermel G. , Laffargue P. 2007 . Growth and condition indices in juvenile sole Solea solea measured to assess the quality of essential fish habitat . Marine Ecology Progress Series 351 : 201 – 208 . Google Scholar CrossRef Search ADS Amara R. , Selleslagh J. , Billon G. , Minier C. 2009 . Growth and condition of 0-group European flounder, Platichthys flesus as indicator of estuarine habitat quality . Hydrobiologia 627 : 87 – 98 . Google Scholar CrossRef Search ADS Andersen J. H. , Halpern B. S. , Korpinen S. , Murray C. , Reker J. 2015 . Baltic Sea biodiversity status vs. cumulative human pressures . Estuarine, Coastal and Shelf Science 161 : 88 – 92 . Google Scholar CrossRef Search ADS Anderson T. R. 2005 . Plankton functional type modelling: running before we can walk? Journal of Plankton Research 27 : 1073 – 1081 . Google Scholar CrossRef Search ADS Aneer G. 1987 . High natural mortality of Baltic herring (Clupea harengus) eggs caused by algal exudates? Marine Biology 94 : 163 – 169 . Google Scholar CrossRef Search ADS Archambault B. , Le Pape O. , Baulier L. , Vermard Y. , Véron M. , Rivot E. 2016 . Adult-mediated connectivity affects inferences on population dynamics and stock assessment of nursery-dependent fish populations . Fisheries Research 181 : 198 – 213 . Google Scholar CrossRef Search ADS Archambault B. , Rivot E. , Savina M. , Le Pape O. 2015 . Using a spatially structured life cycle model to assess the influence of multiple stressors on an exploited coastal-nursery-dependent population . Estuarine, Coastal and Shelf Science : 1 – 10 . http://dx.doi.org/10.1016/j.ecss.2015.12.009. Batista M. I. , Henriques S. , Pais M. P. , Cabral H. N. 2014 . Assessment of cumulative human pressures on a coastal area: integrating information for MPA planning and management . Ocean and Coastal Management 102 : 248 – 257 . Google Scholar CrossRef Search ADS Beck M. W. , Heck K. L. , Able K. W. , Childers D. L. , Eggleston D. B. , Gillanders B. M. , Halpern B. S. 2001 . The identification, conservation, and management of estuarine and marine nurseries for fish and invertebrates . BioScience 51 : 633. Google Scholar CrossRef Search ADS Berghahn R. , Waltemath M. , Rijnsdorp A. D. 1992 . Mortality of fish from the by-catch of shrimp vessels in the North Sea . Journal of Applied Ichthyology 8 : 293 – 306 . Google Scholar CrossRef Search ADS Bianchi C. N. , Corsini-Foka M. , Morri C. , Zenetos A. 2014 . Thirty years after: dramatic change in the coastal marine ecosystems of Kos Island (Greece), 1981-2013 . Mediterranean Marine Science 15 : 482 – 497 . Google Scholar CrossRef Search ADS Bordehore C. , Ramos-Espla A. A. , Riosmena-Rodriguez R. 2003 . Comparative study of two maerl beds with different otter trawling history, southeast Iberian Peninsula . Aquatic Conservation: Marine and Freshwater Ecosystems 13 : S43 – S54 . Google Scholar CrossRef Search ADS Borja Á. , Dauer D. M. , Elliott M. , Simenstad C. A. 2010 . Medium and long-term recovery of estuarine and coastal ecosystems: patterns, rates and restoration effectiveness . Estuaries and Coasts 33 : 1249 – 1260 . Google Scholar CrossRef Search ADS Bratberg M. , Olsvik P. A. , Edvardsen R. B. , Brekken H. K. , Vadla R. , Meier S. 2013 . Effects of oil pollution and persistent organic pollutants (POPs) on glycerophospholipids in liver and brain of male Atlantic cod (Gadus morhua) . Chemosphere 90 : 2157 – 2171 . Google Scholar CrossRef Search ADS PubMed Cameron P. , Von Westernhagen H. 1997 . Malformation rates in embryos of North Sea fishes in 1991 and 1992 . Marine Pollution Bulletin 34 : 129 – 134 . Google Scholar CrossRef Search ADS Canli M. , Atli G. 2003 . The relationships between heavy metal (Cd, Cr, Cu, Fe, Pb, Zn) levels and the size of six Mediterranean fish species . Environmental Pollution 121 : 129 – 136 . Google Scholar CrossRef Search ADS PubMed Carl J. D. , Sparrevohn C. R. , Nicolajsen H. , Stttrup J. G. 2008 . Substratum selection by juvenile flounder Platichthys flesus (L.): effect of ephemeral filamentous macroalgae . Journal of Fish Biology 72 : 2570 – 2578 . Google Scholar CrossRef Search ADS Cloern J. E. 2001 . Our evolving conceptual model of the coastal eutrophication problem . Marine Ecology Progress Series 210 : 223 – 253 . Google Scholar CrossRef Search ADS Costanza R. , d'Arge R. , de Groot R. , Farber S. , Grasso M. , Hannon B. , Limburg K. et al. 1997 . The value of the world’s ecosystem services and natural capital . Nature 387 : 253 – 260 . Google Scholar CrossRef Search ADS Costello M. J. , Gamble J. C. 1992 . Effects of sewage sludge on marine fish embryos and larvae . Marine Environmental Research 33 : 49 – 74 . Google Scholar CrossRef Search ADS Courrat A. , Lobry J. , Nicolas D. , Laffargue P. , Amara R. , Lepage M. , Girardin M. et al. 2009 . Anthropogenic disturbance on nursery function of estuarine areas for marine species . Estuarine, Coastal and Shelf Science 81 : 179 – 190 . Google Scholar CrossRef Search ADS Crain C. M. , Kroeker K. , Halpern B. S. 2008 . Interactive and cumulative effects of multiple human stressors in marine systems . Ecology Letters 11 : 1304 – 1315 . Google Scholar CrossRef Search ADS PubMed Davis J. F. 1999 . Fate of environmental pollutants . Water Environment Research 71 : 1070 – 1078 . Google Scholar CrossRef Search ADS Davoodi F. , Claireaux G. 2007 . Effects of exposure to petroleum hydrocarbons upon the metabolism of the common sole Solea solea . Marine Pollution Bulletin 54 : 928 – 934 . Google Scholar CrossRef Search ADS PubMed de Groot R. , Brander L. , van der Ploeg S. , Costanza R. , Bernard F. , Braat L. , Christie M. et al. 2012 . Global estimates of the value of ecosystems and their services in monetary units . Ecosystem Services 1 : 50 – 61 . Google Scholar CrossRef Search ADS Deegan L. A. 1990 . Effects of estuarine environmental conditions on population dynamics of young-of-the-year Gulf menhaden . Marine Ecology Progress Series 68 : 195 – 205 . Google Scholar CrossRef Search ADS Delpech C. , Courrat A. , Pasquaud S. , Lobry J. , Le Pape O. , Nicolas D. , Boët P. , Girardin M. et al. 2010 . Development of a fish-based index to assess the ecological quality of transitional waters: the case of French estuaries . Marine Pollution Bulletin 60 : 908 – 918 . Google Scholar CrossRef Search ADS PubMed Drake P. , Arias A. M. , Baldó F. , Cuesta J. A. , Rodríguez A. , Silva-Garcia A. , Sobrino I. 2002 . Spatial and temporal variation of the nekton and hyperbenthos from a temperate European estuary with regulated freshwater inflow . Estuaries 25 : 451 – 468 . Google Scholar CrossRef Search ADS Drake P. , Borlán A. , González-Ortegón E. , Baldó F. , Vilas C. , Fernández-Delgado C. 2007 . Spatio-temporal distribution of early life stages of the European anchovy Engraulis encrasicolus L. within a European temperate estuary with regulated freshwater inflow: effects of environmental variables . Journal of Fish Biology 70 : 1689 – 1709 . Google Scholar CrossRef Search ADS Ekau W. , Auel H. , Pörtner H.-O. , Gilbert D. 2010 . Impacts of hypoxia on the structure and processes in pelagic communities (zooplankton, macro-invertebrates and fish) . Biogeosciences 7 : 1669 – 1699 . Google Scholar CrossRef Search ADS Elliott M. , Hemingway K. 2002 . Fishes in estuaries. In Fishes in Estuaries , pp. i–xx. Ed. by Elliott M. , Hemingway K. . Blackwell Science , Oxford . Google Scholar CrossRef Search ADS Falk-Petersen J. , Renaud P. , Anisimova N. 2011 . Establishment and ecosystem effects of the alien invasive red king crab (Paralithodes camtschaticus) in the Barents Sea—a review . ICES Journal of Marine Science 68 : 479 – 488 . Google Scholar CrossRef Search ADS Feunteun E. 2002 . Management and restoration of European eel population (Anguilla anguilla): an impossible bargain . Ecological Engineering 18 : 575 – 591 . Google Scholar CrossRef Search ADS Foden J. , Rogers S. I. , Jones A. P. 2011 . Human pressures on UK seabed habitats: a cumulative impact assessment . Marine Ecology Progress Series 428 : 33 – 47 . Google Scholar CrossRef Search ADS Fodrie F. J. , Mendoza G. 2006 . Availability, usage and expected contribution of potential nursery habitats for the California halibut . Estuarine, Coastal and Shelf Science 68 : 149 – 164 . Google Scholar CrossRef Search ADS Fodrie F. J. , Levin L. A. , Lucas A. 2009 . Use of population fitness to evaluate the nursery function of juvenile habitats . Marine Ecology Progress Series 385 : 39 – 49 . Google Scholar CrossRef Search ADS Fossi M. C. , Coppola D. , Baini M. , Giannetti M. , Guerranti C. , Marsili L. , Panti C. et al. 2014 . Large filter feeding marine organisms as indicators of microplastic in the pelagic environment: the case studies of the Mediterranean basking shark (Cetorhinus maximus) and fin whale (Balaenoptera physalus) . Marine Environmental Research 100 : 17 – 24 . Google Scholar CrossRef Search ADS PubMed Frantzen M. , Falk-Petersen I. B. , Nahrgang J. , Smith T. J. , Olsen G. H. , Hangstad T. A. , Camus L. 2012 . Toxicity of crude oil and pyrene to the embryos of beach spawning capelin (Mallotus villosus) . Aquatic Toxicology 108 : 42 – 52 . Google Scholar CrossRef Search ADS PubMed Frantzen S. , Maage A. , Duinker A. , Julshamn K. , Iversen S. A. 2015 . A baseline study of metals in herring (Clupea harengus) from the Norwegian Sea, with focus on mercury, cadmium, arsenic and lead . Chemosphere 127 : 164 – 170 . Google Scholar CrossRef Search ADS PubMed Gilliers C. , Le Pape O. , Désaunay Y. , Morin J. , Guérault D. , Amara R. 2006 . Are growth and density quantitative indicators of essential fish habitat quality? An application to the common sole Solea solea nursery grounds . Estuarine, Coastal and Shelf Science 69 : 96 – 106 . Google Scholar CrossRef Search ADS Goodsir F. , Bloomfield H. J. , Judd A. D. , Kral F. , Robinson L. A. , Knights A. M. 2015 . A spatially resolved pressure-based approach to evaluate combined effects of human activities and management in marine ecosystems . ICES Journal of Marine Science 72 : 2245 – 2256 . Google Scholar CrossRef Search ADS Goren M. , Galil B. S. 2005 . A review of changes in the fish assemblages of Levantine inland and marine ecosystems following the introduction of non-native fishes . Journal of Applied Ichthyology 21 : 364 – 370 . Google Scholar CrossRef Search ADS Gorokhova E. , Fagerberg T. , Hansson S. 2004 . Predation by herring (Clupea harengus) and sprat (Sprattus sprattus) on Cercopagis pengoi in a western Baltic Sea bay . ICES Journal of Marine Science 61 : 959 – 965 . Google Scholar CrossRef Search ADS Gorokhova E. , Hansson S. , Höglander H. , Andersen C. M. 2005 . Stable isotopes show food web changes after invasion by the predatory cladoceran Cercopagis pengoi in a Baltic Sea bay . Oecologia 143 : 251 – 259 . Google Scholar CrossRef Search ADS PubMed Hall-Spencer J. M. , Grall J. , Moore P. G. , Atkinson R. J. A. 2003 . Bivalve fishing and maerl-bed conservation in France and the UK—retrospect and prospect . Aquatic Conservation: Marine and Freshwater Ecosystems 13 : S33. Google Scholar CrossRef Search ADS Halpern B. S. , Ebert C. M. , Kappel C. V. , Madin E. M. P. , Micheli F. , Perry M. , Selkoe K. A. et al. 2009 . Global priority areas for incorporating land–sea connections in marine conservation . Conservation Letters 2 : 1–8. Halpern B. S. , Selkoe K. A. , Micheli F. , Kappel C. V. 2007 . Evaluating and ranking the vulnerability of global marine ecosystems to anthropogenic threats . Conservation Biology 21 : 1301 – 1315 . Google Scholar CrossRef Search ADS PubMed Halpern B. S. , Walbridge S. , Selkoe K. A. , Kappel C. V. , Micheli F. , D'Agrosa C. , Bruno J. F. et al. 2008 . A global map of human impact on marine ecosystems . Science 319 : 948 – 952 . Google Scholar CrossRef Search ADS PubMed Henriques S. , Pais M. P. , Vasconcelos R. P. , Murta A. , Azevedo M. , Costa M. J. , Cabral H. N. et al. 2014 . Structural and functional trends indicate fishing pressure on marine fish assemblages . Journal of Applied Ecology 51 : 623 – 631 . Google Scholar CrossRef Search ADS Hiddink J. G. , Jennings S. , Kaiser M. J. , Queirós A. M. , Duplisea D. E. , Piet G. J. 2006 . Cumulative impacts of seabed trawl disturbance on benthic biomass, production, and species richness in different habitats . Canadian Journal of Fisheries and Aquatic Sciences 63 : 721 – 736 . Google Scholar CrossRef Search ADS Holmlund C. M. , Hammer M. 1999 . Ecosystem services generated by fish populations . Ecological Economics 29 : 253 – 268 . Google Scholar CrossRef Search ADS Houlihan D. F. , Costello M. J. , Secombes C. J. , Stagg R. , Brechin J. 1994 . Effects of sewage sludge exposure on growth, feeding and protein synthesis of dab (Limanda limanda (L.)) . Marine Environmental Research 37 : 331 – 353 . Google Scholar CrossRef Search ADS Ilarri M. I. , Souza A. T. , Antunes C. , Guilhermino L. , Sousa R. 2014 . Influence of the invasive Asian clam Corbicula fluminea (Bivalvia: Corbiculidae) on estuarine epibenthic assemblages . Estuarine, Coastal and Shelf Science 143 : 12 – 19 http://dx.doi.org/10.1016/j.ecss.2014.03.017 . Google Scholar CrossRef Search ADS Isaksson I. , Pihl L. , van Montfrans J. 1994 . Eutrophication-related changes in macrovegetation and foraging of young cod (Gadus morhua L.): a mesocosm experiment . Journal of Experimental Marine Biology and Ecology 177 : 203 – 217 . Google Scholar CrossRef Search ADS Jebali J. , Sabbagh M. , Banni M. , Kamel N. , Ben-Khedher S. , M’hamdi N. , Boussetta H. 2013 . Multiple biomarkers of pollution effects in Solea solea fish on the Tunisia coastline . Environmental Science and Pollution Research 20 : 3812 – 3821 . Google Scholar CrossRef Search ADS PubMed Jenkinson I. R. , Claireaux G. , Gentien P. 2007 . Biorheological properties of intertidal organic fluff on mud flats and its modification of gill ventilation in buried sole Solea solea . Marine Biology 150 : 471 – 485 . Google Scholar CrossRef Search ADS Jennings S. , Rice J. 2011 . Towards an ecosystem approach to fisheries in Europe: a perspective on existing progress and future directions . Fish and Fisheries 12 : 125 – 137 . Google Scholar CrossRef Search ADS Johannessen T. 2014 . Repeated incidents of abrupt and persistent recruitment failures in gadoids in relation to increasing eutrophication, 1919–2001. In From an Antagonistic to a Synergistic Predator Prey Perspective , pp. 11 – 37 . Ed. by Johannessen T. . Elsevier , London . Google Scholar CrossRef Search ADS Johansen J. L. , Herbert N. A. , Steffensen J. F. 2006 . The behavioural and physiological response of Atlantic cod Gadus morhua L. to short-term acute hypoxia . Journal of Fish Biology 68 : 1918 – 1924 . Google Scholar CrossRef Search ADS Johansen L. , Jensen I. , Mikkelsen H. , Bjørn P. , Jansen P. A. , Bergh Ø. 2011 . Disease interaction and pathogens exchange between wild and farmed fish populations with special reference to Norway . Aquaculture 315 : 167 – 186 . Google Scholar CrossRef Search ADS Johnston D. W. , Wildish D. J. 1981 . Avoidance of dredge spoil by herring (Clupea harengus harengus) . Bulletin of Environmental Contamination and Toxicology 26 : 307 – 314 . Google Scholar CrossRef Search ADS PubMed Jokinen H. , Wennhage H. , Ollus V. , Aro E. , Norkko A. 2016 . Juvenile flatfish in the northern Baltic Sea—long-term decline and potential links to habitat characteristics . Journal of Sea Research 107 : 67 – 75 . Google Scholar CrossRef Search ADS Jonsson B. , Jonsson N. 2014 . Early environment influences later performance in fishes . Journal of Fish Biology 85 : 151 – 188 . Google Scholar CrossRef Search ADS PubMed Kamenos N. A. , Moore P. G. , Hall-Spencer J. M. 2004 . Small-scale distribution of juvenile gadoids in shallow inshore waters; what role does maerl play? ICES Journal of Marine Science 61 : 422 – 429 . Google Scholar CrossRef Search ADS Kempf A. 2010 . Ecosystem approach to fisheries in the European context—history and future challenges . Journal of Applied Ichthyology 26 : 102 – 109 . Google Scholar CrossRef Search ADS Kerambrun E. , Henry F. , Courcot L. , Gevaert F. , Amara R. 2012 . Biological responses of caged juvenile sea bass (Dicentrarchus labrax) and turbot (Scophtalmus maximus) in a polluted harbour . Ecological Indicators 19 : 161 – 171 . Google Scholar CrossRef Search ADS Kerambrun E. , Henry F. , Perrichon P. , Courcot L. , Meziane T. , Spilmont N. , Amara R. 2012a . Growth and condition indices of juvenile turbot, Scophthalmus maximus, exposed to contaminated sediments: effects of metallic and organic compounds . Aquatic Toxicology 108 : 130 – 140 . Google Scholar CrossRef Search ADS Kideys A. 1994 . Recent dramatic changes in the Black Sea ecosystem: the reason for the sharp decline in Turkish anchovy fisheries . Journal of Marine Systems 5 : 171 – 181 . Google Scholar CrossRef Search ADS Klanjšček J. , Legović T. 2007 . Is anchovy (Engraulis encrasicolus, L.) overfished in the Adriatic Sea? Ecological Modelling 201 : 312 – 316 . Google Scholar CrossRef Search ADS Knights A. M. , Piet G. J. , Jongbloed R. H. , Tamis J. E. , White L. , Akoglu E. , Boicenco L. et al. 2015 . An exposure-effect approach for evaluating ecosystem-wide risks from human activities . ICES Journal of Marine Science 72 : 1105 – 1115 . Google Scholar CrossRef Search ADS Koons D. N. , Arnold T. W. , Schaub M. 2017 . Understanding the demographic drivers of realized population growth rates . Ecological Applications 27 : 2102 – 2115 . Google Scholar CrossRef Search ADS PubMed Kornilovs G. 1993 . The modern state of Baltic herring spawning grounds in the Gulf of Riga. ICES Document CM 1993/J: 26. 6 pp. Kostecki C. , Rochette S. , Girardin R. , Blanchard M. , Desroy N. , Le Pape O. 2011 . Reduction of flatfish habitat as a consequence of the proliferation of an invasive mollusc . Estuarine, Coastal and Shelf Science 92 : 154 – 160 . Google Scholar CrossRef Search ADS Kummu M. , de Moel H. , Salvucci G. , Viviroli D. , Ward P. J. , Varis O. 2016 . Over the hills and further away from coast: global geospatial patterns of human and environment over the 20th–21st centuries . Environmental Research Letters 11 : 15. . Google Scholar CrossRef Search ADS Laffaille P. , Lefeuvre J.-C. , Feunteun E. 2000 . Impact of sheep grazing on juvenile sea bass, Dicentrarchus labrax L., in tidal salt marshes . Biological Conservation 96 : 271 – 277 . Google Scholar CrossRef Search ADS Lagares E. C. , Ordaz F. G. 2014 . Fisheries structural policy in the European Union: a critical analysis of a subsidised sector . Ocean and Coastal Management 102 : 200 – 211 . Google Scholar CrossRef Search ADS Lappalainen A. , Pesonen L. 2000 . Changes in fish community structure after cessation of waste water discharge in a coastal bay area west of Helsinki, northern Baltic Sea . Archive of Fishery and Marine Research Archiv fur Fischerei und Meeresforschung 48 : 226 – 241 http://apps.isiknowledge.com/full_record.do? product=WOS&search_mode=GeneralSearch&qid=30&SID=N1AnkHbaLm3 @p84keBB&page=1&doc=1. Lefrançois C. , Claireaux G. 2003 . Influence of ambient oxygenation and temperature on metabolic scope and scope for heart rate in the common sole Solea solea . Marine Ecology Progress Series 259 : 273 – 284 . Google Scholar CrossRef Search ADS Le Luherne E. , Réveillac E. , Ponsero A. , Sturbois A. , Ballu S. , Perdriau M. , Le Pape O. 2016 . Fish community responses to green tides in shallow estuarine and coastal areas . Estuarine, Coastal and Shelf Science 175 : 79 – 92 . Google Scholar CrossRef Search ADS Le Pape O. , Baulier L. , Cloarec A. , Martin J. , Le Loc'h F. , Désaunay Y. 2007 . Habitat suitability for juvenile common sole (Solea solea, L.) in the Bay of Biscay (France): a quantitative description using indicators based on epibenthic fauna . Journal of Sea Research 57 : 126 – 136 . Google Scholar CrossRef Search ADS Le Pape O. , Guérault D. , Désaunay Y. 2004 . Effect of an invasive mollusc, American slipper limpet Crepidula fornicata, on habitat suitability for juvenile common sole Solea solea in the Bay of Biscay . Marine Ecology Progress Series 277 : 107 – 115 . Google Scholar CrossRef Search ADS Levi F. , Francour P. 2004 . Behavioural response of Mullus surmuletus to habitat modification by the invasive macroalga Caulerpa taxifolia . Journal of Fish Biology 64 : 55 – 64 . Google Scholar CrossRef Search ADS Levin P. S. , Stunz G. W. 2005 . Habitat triage for exploited fishes: can we identify essential ‘Essential Fish Habitat?’ . Estuarine, Coastal and Shelf Science 64 : 70 – 78 . Google Scholar CrossRef Search ADS Lindholm J. B. , Auster P. J. , Kaufman L. S. 1999 . Habitat-mediated survivorship of juvenile (0-year) Atlantic cod Gadus morhua . Marine Ecology Progress Series 180 : 247 – 255 . Google Scholar CrossRef Search ADS Longwell A. C. , Chang S. , Hebert A. , Hughes J. B. , Perry D. 1992 . Pollution and developmental abnormalities of Atlantic fishes . Environmental Biology of Fishes 35 : 1 – 21 . Google Scholar CrossRef Search ADS Lotze H. K. , Lenihan H. S. , Bourque B. J. , Bradbury R. H. , Cooke R. G. , Kay M. C. , Kidwell S. M. et al. 2006 . Depletion, degradation, and recovery potential of estuaries and coastal seas . Science 312 : 1806 – 1809 . Google Scholar CrossRef Search ADS PubMed Lotze H. K. , Reise K. , Worm B. , van Beusekom J. , Busch M. , Ehlers A. , Heinrich D. et al. 2005 . Human transformations of the Wadden Sea ecosystem through time: a synthesis . Helgoland Marine Research 59 : 84 – 95 . Google Scholar CrossRef Search ADS Maes J. , Stevens M. , Breine J. 2007 . Modelling the migration opportunities of diadromous fish species along a gradient of dissolved oxygen concentration in a European tidal watershed . Estuarine, Coastal and Shelf Science 75 : 151 – 162 . Google Scholar CrossRef Search ADS Malovic I. , Hemmingsen W. , MacKenzie K. 2010 . Trypanosome infections of marine fish in the southern Barents Sea and the invasive red king crab Paralithodes camtschaticus . Marine Pollution Bulletin 60 : 2257 – 2262 . Google Scholar CrossRef Search ADS PubMed Martignac F. , Baglinière J. L. , Thieulle L. , Ombredane D. , Guillard J. 2013 . Influences of a dam on Atlantic salmon (Salmo salar) upstream migration in the Couesnon River (Mont Saint Michel Bay) using hydroacoustics . Estuarine, Coastal and Shelf Science 134 : 181 – 187 . Google Scholar CrossRef Search ADS McIntosh S. , King T. , Wu D. , Hodson P. V. 2010 . Toxicity of dispersed weathered crude oil to early life stages of Atlantic herring (Clupea harengus) . Environmental Toxicology and Chemistry 29 : 1160 – 1167 . Google Scholar PubMed Meland S. , Salbu B. , Rosseland B. O. 2010 . Ecotoxicological impact of highway runoff using brown trout (Salmo trutta L.) as an indicator model . Journal of Environmental Monitoring 12 : 654 – 664 . Google Scholar CrossRef Search ADS PubMed Meynecke J.-O. , Richards R. G. 2014 . A full life cycle and spatially explicit individual-based model for the giant mud crab (Scylla serrata): a case study from a marine protected area . ICES Journal of Marine Science 71 : 484 – 498 . Google Scholar CrossRef Search ADS Molnar J. L. , Gamboa R. L. , Revenga C. , Spalding M. D. 2008 . Assessing the global threat of invasive species to marine biodiversity . Frontiers in Ecology and the Environment 6 : 485 – 492 . Google Scholar CrossRef Search ADS Mumby P. J. , Edwards A. J. , Arias-González J. E. , Lindeman K. C. , Blackwell P. G. , Gall A. , Gorczynska M. I. et al. 2004 . Mangroves enhance the biomass of coral reef fish communities in the Caribbean . Nature 427 : 533 – 536 . Google Scholar CrossRef Search ADS PubMed Niermann U. , Bingel F. , Gorban A. , Gordina A. D. , Gücü A. C. , Kideys A. E. , Konsulov A. 1994 . Distribution of anchovy eggs and larvae (Engraulis encrasicolus Cuv.) in the Black Sea in 1991-1992 . ICES Journal of Marine Science 51 : 395 – 406 . Google Scholar CrossRef Search ADS Ogut H. , Palm H. W. 2005 . Seasonal dynamics of Trichodina spp. on whiting (Merlangius merlangus) in relation to organic pollution on the eastern Black Sea coast of Turkey . Parasitology Research 96 : 149 – 153 . Google Scholar CrossRef Search ADS PubMed Oguz T. , Fach B. , Salihoglu B. 2008 . Invasion dynamics of the alien ctenophore Mnemiopsis leidyi and its impact on anchovy collapse in the Black Sea . Journal of Plankton Research 30 : 1385 – 1397 . Google Scholar CrossRef Search ADS Olsen E. , Aanes S. , Mehl S. , Holst J. C. , Aglen A. , Gjosaeter H. 2010 . Cod, haddock, saithe, herring, and capelin in the Barents Sea and adjacent waters: a review of the biological value of the area . ICES Journal of Marine Science 67 : 87 – 101 . Google Scholar CrossRef Search ADS Olsson I. C. , Greenberg L. A. , Eklöv A. G. 2001 . Effect of an artificial pond on migrating brown trout smolts . North American Journal of Fisheries Management 21 : 498 – 506 . Google Scholar CrossRef Search ADS Österblom H. , Hansson S. , Larsson U. , Hjerne O. , Wulff F. , Elmgren R. , Folke C. 2007 . Human-induced trophic cascades and ecological regime shifts in the Baltic Sea . Ecosystems 10 : 877 – 889 . Google Scholar CrossRef Search ADS Öztürk B. 2006 . Invasive species of the Turkish Straits. In The Turkish Straits: Maritime Safety, Legal and Environmental Aspects , pp. 96 – 105 . Ed. by Oral N. , Öztürk B. . Turkish Marine Research Foundation , İstanbul . Publication Number 25. 160 pp. Palstra A. P. , Van Ginneken V. J. T. , Murk A. J. , Van Den Thillart G. E. E. J. M. 2006 . Are dioxin-like contaminants responsible for the eel (Anguilla anguilla) drama? Naturwissenschaften 93 : 145 – 148 . Google Scholar CrossRef Search ADS PubMed Parmanne R. , Rechlin O. , Sjöstrand B. 1994 . Status and future of herring and sprat stocks in the Baltic Sea . Dana 10 : 29 – 59 . Patel S. , Korsnes K. , Bergh Ø. , Vik-Mo F. , Pedersen J. , Nerland A. 2007 . Nodavirus in farmed Atlantic cod Gadus morhua in Norway . Diseases of Aquatic Organisms 77 : 169 – 173 . Google Scholar CrossRef Search ADS PubMed Peters L. D. , Porte C. , Livingstone D. R. 2001 . Variation of antioxidant enzyme activities of sprat (Sprattus sprattus) larvae and organic contaminant levels in mixed zooplankton from the southern North Sea . Marine Pollution Bulletin 42 : 1087 – 1095 . Google Scholar CrossRef Search ADS PubMed Petersen J. K. , Pihl L. 1995 . Responses to hypoxia of plaice, Pleuronectes platessa, and dab, Limanda limanda, in the south-east Kattegat: distribution and growth . Environmental Biology of Fishes 43 : 311 – 321 . Google Scholar CrossRef Search ADS Peterson C. H. , Bishop M. J. , D'Anna L. M. , Johnson G. A. 2014 . Multi-year persistence of beach habitat degradation from nourishment using coarse shelly sediments . Science of the Total Environment 487 : 481 – 492 . Google Scholar CrossRef Search ADS PubMed Piggott J. J. , Townsend C. R. , Matthaei C. D. 2015 . Reconceptualizing synergism and antagonism among multiple stressors . Ecology and Evolution 5 : 1538 – 1547 . Google Scholar CrossRef Search ADS PubMed Pihl L. , Modin J. , Wennhage H. 2005 . Recruitment to deteriorating habitat quality: effects of macroalgal blooms in coastal nursery grounds . Canadian Journal of Fisheries and Aquatic Sciences 62 : 1184 – 1193 . Google Scholar CrossRef Search ADS Pihl L. , Baden S. , Kautsky N. , Rönnbäck P. , Söderqvist T. , Troell M. , Wennhage H. 2006 . Shift in fish assemblage structure due to loss of seagrass Zostera marina habitats in Sweden . Estuarine, Coastal and Shelf Science 67 : 123 – 132 . Google Scholar CrossRef Search ADS Power M. , Attrill M. J. , Thomas R. M. 2000 . Environmental factors and interactions affecting the temporal abundance of juvenile flatfish in the Thames Estuary . Journal of Sea Research 43 : 135 – 149 . Google Scholar CrossRef Search ADS Rabalais N. N. 2015 . Human impacts on fisheries across the land–sea interface . Proceedings of the National Academy of Sciences 112 : 7892 – 7893 . Google Scholar CrossRef Search ADS Rahikainen M. , Hoviniemi K.-M. , Mäntyniemi S. , Vanhatalo J. , Helle I. , Lehtiniemi M. , Pönni J. et al. 2017 . Impacts of eutrophication and oil spills on the Gulf of Finland herring stock . Canadian Journal of Fisheries and Aquatic Sciences 74 : 1218 – 1232 . Google Scholar CrossRef Search ADS Reubens J. T. , Braeckman U. , Vanaverbeke J. , Van Colen C. , Degraer S. , Vincx M. 2013 . Aggregation at windmill artificial reefs: CPUE of Atlantic cod (Gadus morhua) and pouting (Trisopterus luscus) at different habitats in the Belgian part of the North Sea . Fisheries Research 139 : 28 – 34 . Google Scholar CrossRef Search ADS Reynolds W. J. , Lancaster J. E. , Pawson M. G. 2003 . Patterns of spawning and recruitment of sea bass to Bristol Channel nurseries in relation to the 1996 ‘Sea Empress’ oil spill . Journal of the Marine Biological Association of the UK 83 : 1163 – 1170 . Google Scholar CrossRef Search ADS Rijnsdorp A. D. , Bastardie F. , Bolam S. G. , Buhl-Mortensen L. , Eigaard O. R. , Hamon K. G. , Hiddink J. G. et al. 2016 . Towards a framework for the quantitative assessment of trawling impact on the seabed and benthic ecosystem . ICES Journal of Marine Science 73 : i127 – i138 . Google Scholar CrossRef Search ADS Rochette S. , Le Pape O. , Vigneau J. , Rivot E. 2013 . A hierarchical Bayesian model for embedding larval drift and habitat models in integrated life cycles for exploited fish . Ecological Applications 23 : 1659 – 1676 . Google Scholar CrossRef Search ADS PubMed Rochette S. , Rivot E. , Morin J. , Mackinson S. , Riou P. , Le Pape O. 2010 . Effect of nursery habitat degradation on flatfish population: application to Solea solea in the Eastern Channel (Western Europe) . Journal of Sea Research 64 : 34 – 44 . Google Scholar CrossRef Search ADS Ruiz J. , González-Quirós R. , Prieto L. , Navarro G. 2009 . A Bayesian model for anchovy (Engraulis encrasicolus): the combined forcing of man and environment . Fisheries Oceanography 18 : 62 – 76 . Google Scholar CrossRef Search ADS Schmidt B. R. , Hödl W. , Schaub M. 2012 . From metamorphosis to maturity in complex life cycles: equal performance of different juvenile life history pathways . Ecology 93 : 657 – 667 . Google Scholar CrossRef Search ADS PubMed Sciberras M. , Hiddink J. G. 2014 . Predicting the effect of trawling based on biological traits of organisms and functional correlates of these traits to predict which functions may be disproportionally affected. Benthic Ecosystem Fisheries Impact Study, IMARES, IJmuiden, The Netherlands, Deliverable 4.3. 125 pp. http://www.benthis.eu/web/file? uuid=ac0db45d-135d-4b84-806d-9a9408bf2b6e&owner=fd9fa22c-6bf7-42dc-ad64-ad4cbd966f98 (Accessed 11 April 2016). Scopelliti G. , Di Leonardo R. , Tramati C. D. , Mazzola A. , Vizzini S. 2015 . Premature aging in bone of fish from a highly polluted marine area . Marine Pollution Bulletin 97 : 333 – 341 . Google Scholar CrossRef Search ADS PubMed Secombes C. J. , White A. , Fletcher T. C. , Stagg R. , Houlihan D. F. 1995 . Immune parameters of plaice, Pleuronectes platessa, L. along a sewage sludge gradient in the Firth of Clyde, Scotland . Ecotoxicology 4 : 329 – 340 . Google Scholar CrossRef Search ADS PubMed Seitz R. D. , Wennhage H. , Bergström U. , Lipcius R. N. , Ysebaert T. 2014 . Ecological value of coastal habitats for commercially and ecologically important species . ICES Journal of Marine Science 71 : 648 – 655 (Accessed 1 December 2014). Google Scholar CrossRef Search ADS Shiganova T. A. , Bulgakova Y. V. 2000 . Effects of gelatinous plankton on Black Sea and Sea of Azov fish and their food resources . ICES Journal of Marine Science 57 : 641 – 648 . Google Scholar CrossRef Search ADS Shiganova T. A. , Mirzoyan Z. A. , Studenikin E. A. , Volovik S. P. , Siokou-Frangou I. , Zervoudaki S. , Christou E. D. , Skirta A. Y. , Dumont H. J. 2001 . Population development of the invader ctenophore Mnemiopsis leidyi, in the Black Sea and in other seas of the Mediterranean basin . Marine Biology 139 : 431 – 445 . Google Scholar CrossRef Search ADS Stelzenmüller V. , Schulze T. , Fock H. O. , Berkenhagen J. 2011 . Integrated modelling tools to support risk-based decision-making in marine spatial management . Marine Ecology Progress Series 441 : 197 – 212 . Google Scholar CrossRef Search ADS Støttrup J. G. , Stenberg C. , Dahl K. , Kristensen L. D. , Richardson K. 2014 . Restoration of a temperate reef: effects on the fish community . Open Journal of Ecology 04 : 4: 1045 – 1059 . Google Scholar CrossRef Search ADS Sundblad G. , Bergström U. , Sandstrom A. , Eklov P. 2014 . Nursery habitat availability limits adult stock sizes of predatory coastal fish . ICES Journal of Marine Science 71 : 672 – 680 . Google Scholar CrossRef Search ADS Taranger G. L. , Karlsen Ø. , Bannister R. J. , Glover K. A. , Husa V. , Karlsbakk E. , Kvamme B. O. et al. 2015 . Risk assessment of the environmental impact of Norwegian Atlantic salmon farming . ICES Journal of Marine Science 72 : 997 – 1021 . Google Scholar CrossRef Search ADS Teichert N. , Borja A. , Chust G. , Uriarte A. , Lepage M. 2016 . Restoring fish ecological quality in estuaries: implication of interactive and cumulative effects among anthropogenic stressors . Science of the Total Environment 542 : 383 – 393 . Google Scholar CrossRef Search ADS PubMed Tentelier C. , Piou C. 2011 . Obstacles to migration constrain nest distribution of Atlantic salmon . Ecology of Freshwater Fish 20 : 400 – 408 . Google Scholar CrossRef Search ADS Teschner E. C. , Kraus G. , Neuenfeldt S. , Voss R. , Hinrichsen H. H. , Köster F. W. 2010 . Impact of hypoxia on consumption of Baltic cod in a multispecies stock assessment context . Journal of Applied Ichthyology 26 : 836 – 842 . Google Scholar CrossRef Search ADS Thurstan R. H. , Roberts C. M. , Unsworth R. K. F. 2010 . Ecological meltdown in the Firth of Clyde, Scotland: two centuries of change in a coastal marine ecosystem . PLoS One 5 : e11767. Google Scholar CrossRef Search ADS PubMed Tomczak M. T. , Heymans J. J. , Yletyinen J. , Niiranen S. , Otto S. A. , Blenckner T. , MacKenzie B. R. 2013 . Ecological network indicators of ecosystem status and change in the Baltic Sea . PLoS ONE 8 : 1 – 11 . Google Scholar CrossRef Search ADS Tsikhon-Lykanina E. A. , Reznitchenko O. G. 1991 . Feeding peculiarities of different size species of ctenophore Mnemiopsis in the Black Sea . Oceanology 31 : 442 – 446 . Turnpenny A. W. , Taylor C. J. 2000 . An assessment of the effect of the Sizewell power stations on fish populations . Hydroécologie Appliquée 12 : 87 – 134 . Google Scholar CrossRef Search ADS van Denderen P. D. , Hintzen N. T. , Rijnsdorp A. D. , Ruardij P. , van Kooten T. 2014 . Habitat-specific effects of fishing disturbance on benthic species richness in marine soft sediments . Ecosystems 17 : 1216 – 1226 . Google Scholar CrossRef Search ADS van de Wolfshaar K. E. , HilleRisLambers R. , Gårdmark A. 2011 . Effect of habitat productivity and exploitation on populations with complex life cycles . Marine Ecology Progress Series 438 : 175 – 184 . Google Scholar CrossRef Search ADS Vasconcelos R. P. , Batista M. I. , Henriques S. 2017 . Current limitations of global conservation to protect higher vulnerability and lower resilience fish species . Scientific Reports 7 : 7702. . Google Scholar CrossRef Search ADS PubMed Vasconcelos R. P. , Reis-Santos P. , Fonseca V. , Maia A. , Ruano M. , França S. , Vinagre C. et al. 2007 . Assessing anthropogenic pressures on estuarine fish nurseries along the Portuguese coast: a multi-metric index and conceptual approach . Science of the Total Environment 374 : 199 – 215 . Google Scholar CrossRef Search ADS PubMed Via J. D. , Van den Thillart G. , Cattani O. , Cortesi P. 1998 . Behavioural responses and biochemical correlates in Solea solea to gradual hypoxic exposure . Canadian Journal of Zoology 76 : 2108 – 2113 . Google Scholar CrossRef Search ADS Westernhagen H. , Cameron P. , Dethlefsen V. , Janssen D. 1989 . Chlorinated hydrocarbons in North Sea whiting (Merlangius merlangus L), and effects on reproduction. I. Tissue burden and hatching success . Helgoländer Meeresuntersuchungen 43 : 45 – 60 . Google Scholar CrossRef Search ADS Winters G. H. , Wheeler J. P. , Dalley E. L. 1986 . Survival of a herring stock subjected to a catastrophic event and fluctuating environmental conditions . Journal du Conseil International pour l'Exploration de la Mer 43 : 26 – 42 . Google Scholar CrossRef Search ADS Worm B. , Barbier E. B. , Beaumont N. , Duffy J. E. , Folke C. , Halpern B. S. , Jackson J. B. C. et al. 2006 . Impacts of biodiversity loss on ocean ecosystem services . Science 314 : 787 – 790 . Google Scholar CrossRef Search ADS PubMed © International Council for the Exploration of the Sea 2018. All rights reserved. For Permissions, please email: [email protected] This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

Lea, James S E;Wetherbee, Bradley M;Sousa, Lara L;Aming, Choy;Burnie, Neil;Humphries, Nicolas E;Queiroz, Nuno;Harvey, Guy M;Sims, David W;Shivji, Mahmood S

doi: 10.1093/icesjms/fsx238pmid:

doi: 10.1093/icesjms/fsy009pmid: N/A

Abstract Fish school descriptors extracted from omnidirectional multi-beam data are biased due to beam width-related effects, and echotraces are distorted in a range-dependent manner that is a function of transducer intrinsic properties, as well as fish school characteristics. This work investigates a simulation approach that models the three-dimensional insonification of fish schools by an omnidirectional fishery sonar in order to assess the bias in measuring two key morphometric and energetic descriptors, namely the horizontal cross-sectional area of schools and their mean volume backscattering strength. Simulated fish schools of different sizes and backscattering properties were insonified at various ranges from the multi-beam transducer, outputting volume backscattering strength echograms. The simulated data were used to develop empirical models that correct the examined descriptors using only information extracted from the observed echotraces. Depending on the difference between the observed mean volume backscattering strength of a school and the echogram processing threshold, mean absolute percentage errors in measured area and volume backscatter reduced from 100.7% and 79.5% to 5.2% and 6.4%, respectively. The mean volume backscattering strength of a school is a key parameter for obtaining fish density estimates, and the results highlight the need for descriptor corrections to better interpret the multi-beam data. Introduction Scientific echosounders and multi-beam sonars are the primary underwater observation tools for acquiring fishery-independent data on the abundance and distribution of fish stocks (Misund, 1997; Simmonds and MacLennan, 2005). A common analysis procedure of the acoustic data is echotrace detection and classification (Reid, 2000), i.e. the processing of echograms with specialised algorithms (e.g. Weill et al., 1993; Coetzee, 2000; Reid et al., 2000) that extract morphometric, energetic, and positional descriptors at the fish school level. Often complemented with auxiliary information (McClatchie et al., 2000), fish school descriptors from downwards-looking echosounders find multiple uses in partitioning the acoustic data into species or target groups (Haralabous and Georgakarakos, 1996; Lawson et al., 2001; Petitgas et al., 2003; Fernandes, 2009), and in assessing the diel variability (Fréon et al., 1996; Zwolinski et al., 2007; Tsagarakis et al., 2012), clustering (Petitgas et al., 2001; Petitgas, 2003) and spatial structure of fish school biomass (Scalabrin and Massé, 1993; Bahri and Fréon, 2000; Castillo and Robotham, 2004). Fish school behavioural aspects such as swimming speed (Peraltilla and Bertrand, 2014), migration (Hafsteinsson and Misund, 1995; Kvamme et al., 2003), and reaction to vessels or fishing and other shipboard operations (Soria et al., 2003; Peña et al., 2013; Stockwell et al., 2013) are typically investigated with multi-beam sonars, due to the larger sampling volume of these instruments and their ability for concurrent insonification of multiple fish schools (Trygonis et al., 2016) near the sea surface (Misund et al., 1996). The multi-beam data per ping can also be visualized as a two-dimensional image (hereafter referred to as a “multi-beam echogram”), by arranging the acoustic samples per beam in a sonar-centred polar grid, whose across-beam resolution increases with range. School detection techniques are also applicable in multi-beam echograms, and range from image processing of the sonar display (Brehmer et al., 2006; Uranga et al., 2017) to direct manipulations of the acoustic data samples in order to isolate and measure the echotraces of interest (Gerlotto and Paramo, 2003; Trenkel et al., 2009; Trygonis et al., 2009). Regardless of the acoustic platform, an inherent restriction that underlies the school detection methodologies is the impact of the insonification procedure on observed fish school properties. It has been early recognized that fish school descriptors extracted from echograms are biased by the beam pattern effect (Olsen, 1969; Johannesson and Losse, 1977), and simulation approaches have provided the means for corrections in single beam echograms (Diner, 2001, 2007). The beam width-related distortion of school echotraces and the threshold dependency of effective detections (Aglen, 1983) are also present in horizontal multi-beam measurements (Misund, 1990; Misund et al., 1995), with increased complexities due to the larger ranges typically used, the angular resolution of echograms, as well as the overlap of neighbouring beams and the oblique angles under which fish are insonified (Cutter and Demer, 2007). As demonstrated by recent works (Holmin et al., 2012; Vatnehol et al., 2017), simulations of the multi-beam insonification procedure can greatly enhance the interpretation of echograms and provide more accurate estimates of fish school metrics. This paper presents a fisheries acoustic simulator that models the insonification of fish schools by omnidirectional multi-beam sonar in an environment without noise or reverberation. Taking into account the range-dependent distortion of echotraces induced by the polar geometry of the echogram, the beam pattern effect, and the overlap between neighbouring beams, the study objectives are to: (a) simulate data from an omnidirectional fishery sonar that represent the insonification of fish schools of different size and density classes, observed at varying distances from the transducer; (b) assess the errors in measuring two key morphometric and energetic descriptors of fish schools, namely their horizontal cross-sectional area and mean volume backscattering strength; (c) develop empirical models which use only information extracted from the observed echotrace to correct these descriptors. The empirical models are also applied to real sonar data to examine how the predicted corrections modify the acoustic descriptors of real fish schools. Methods General description of the simulation The simulator (developed in MATLAB®) implements a model of the transducer’s two-way beam pattern and simulates echoes from fish schools, outputting volume backscattering strength echograms that are comparable to those produced by the real acoustic device being modelled. The virtual fish schools insonified can be customized into different sizes and backscattering properties, and can be positioned at various distances and depths relative to the transducer. Sonar operational settings such as transducer depth, observation range, and tilt angle of beams are decoupled from the insonified targets and can be modified on a ping-by-ping basis. The main assumptions of this simulation approach are that: Target echoes are incoherent, i.e. their phases are unrelated (Simmonds and MacLennan, 2005). Acoustic extinction and multiple scattering are negligible. The insonified school is the only source of backscatter. Sound speed is constant throughout the propagation path. There is no noise or reverberation, nor seafloor or sea surface reflections. The simulator was configured to implement the Simrad SP90 (Simrad, 2007), which is an omnidirectional multi-beam fishery sonar that operates at a frequency of 20–30 kHz (for research applications of the SP90, see Brehmer et al., 2007, 2012; Stockwell et al., 2013; Trygonis et al., 2009, 2016). The sonar can be operated in two configurations, namely the “horizontal” and “vertical” transmit modes. In the horizontal mode, the SP90 provides full 360° coverage around the cylindrical transducer and all beams share the same tilt angle relative to the sea surface, forming an umbrella-shaped omnidirectional fan; echoes are received by 64 beams. The vertical mode provides a 60° wide vertical slice of the water column in a single transmission. Only the horizontal (omnidirectional) configuration was simulated in the current study, using the beam pattern corresponding to the 26 kHz continuous wave (CW) normal mode of the sonar (in this mode, the SP90 operates at a fixed signal frequency; pulse duration: 10 ms; nominal beam width: 11.25° for horizontal reception, 9° vertical). Notations The subscripts annotating the various quantities are as follows: 0: True property of a simulated fish school. obs: Observed descriptor of an echotrace, as extracted from the multi-beam echogram. CF: Correction factor of an observed descriptor. c: Corrected descriptor. The “bar” and “dot” accents in descriptors (e.g. Ŝv and Ṡv) denote the mean and maximum, respectively. Fish school model The virtual schools subjected to insonification (Figure 1) are modelled as a three-dimensional structure of adjacent cubic voxels k (volume elements) that have the same volume Vvox but varying volume backscattering coefficients (MacLennan et al., 2002) sv,k. For the simulations described herein, the shape of the school and the sv,k value per voxel k remain fixed across insonifications, and each voxel is assumed to have an omnidirectional response. Figure 1. View largeDownload slide Example voxel school that consists of Nk = 10 720 cubic voxels with identical volume (Vvox = 8 m3) but varying volume backscattering coefficients sv,k that derive from an exponential distribution with <μ> = 10−39/10. The actual mean of the 10 720 voxel backscattering coefficients is s̄v,0 = 1.2676 × 10−4 m−1, corresponding to the true mean volume backscattering strength of the school Ŝv,0 = 10 log10(s̄v,0) = –38.970 dB. The true maximum length along the x-, y-, z-axes is (Ŀx,0, Ŀy,0, Ŀz,0) = (72, 72, 32) m, which yields a true cross-sectional area A0 = 4048 m2 on the x–y plane at z = 0. To simplify notations in the main text, this school is referred to as A0 = 4000 m2, Ŝv,0 = –39 dB. Figure 1. View largeDownload slide Example voxel school that consists of Nk = 10 720 cubic voxels with identical volume (Vvox = 8 m3) but varying volume backscattering coefficients sv,k that derive from an exponential distribution with <μ> = 10−39/10. The actual mean of the 10 720 voxel backscattering coefficients is s̄v,0 = 1.2676 × 10−4 m−1, corresponding to the true mean volume backscattering strength of the school Ŝv,0 = 10 log10(s̄v,0) = –38.970 dB. The true maximum length along the x-, y-, z-axes is (Ŀx,0, Ŀy,0, Ŀz,0) = (72, 72, 32) m, which yields a true cross-sectional area A0 = 4048 m2 on the x–y plane at z = 0. To simplify notations in the main text, this school is referred to as A0 = 4000 m2, Ŝv,0 = –39 dB. The geometric properties of a simulated school are determined by the user-defined voxel volume Vvox, and the nominal equatorial (α) and polar (a’ < α) radii of an oblate spheroid that represents the general school shape. To construct the school, the entire volume of the spheroid is converted into an equivalent discrete space that consists of cubic voxels of volume Vvox within a locally defined system of reference. This process automatically determines the total number of voxels Nk comprising the school, as well as the position of each voxel in space relative to the school-centred coordinate system. The school’s true cross-sectional area A0 (m2) is defined as the area of the largest (equatorial) horizontal cross-section of the voxel-based structure. In terms of acoustic characteristics, each voxel is allocated a different volume backscattering coefficient sv,k that is pseudo-randomly drawn from the exponential distribution with mean parameter <μ> = 10<M>/10 (m−1), where <Μ> in dB is the nominal (expected) mean volume backscattering strength of the school, as defined by the user. The mean of these Nk exponentially distributed sv,k values assigned to the voxels is the true mean s̄v,0 (m−1) of the school, and corresponds to the true mean volume backscattering strength Ŝv,0 = 10 log10(s̄v,0) dB. Note that: the actual Ŝv,0 of a school will slightly differ from the nominal value <Μ>, due to the pseudo-random sampling involved; the school’s true cross-sectional area A0 on the equatorial (x–y) plane will slightly differ from the nominal value πα2, due to discrete nature of the school model. To simplify notations in the remainder of this text, voxel schools will be referred to using their nominal properties <Μ> and πα2; the actual true values will be used for all calculations. Simulated measurement of school backscatter Consider a school voxel k that includes nk fish individuals and has a known true volume backscattering coefficient sv,k. Each ith individual contained in the voxel has a different backscattering cross-section σbs,i,k and, assuming random phases and negligible acoustic extinction, the voxel’s total backscatter is the sum of all nk fish contributions: sv,k=Σi=1nkσbs,i,kVvox, (1) where Vvox is the voxel’s volume, which is identical for all Nk voxels that comprise the simulated school. The effective sampling volume of an acoustic sample s located at range rs from the transducer is Vs,bm=13((rs+Δr/2)3−(rs−Δr/2)3)ψbm, (2) where the subscript bm denotes the beam index, ψbm is the equivalent beam angle in steradians, and Δr is the along-beam sample size in meters; i.e. Vs,bm is a spherical cone shell that extends Δr in the along-beam direction and across the face of the beam to include the equivalent beam angle. This volume may contain Nk,s,bm voxels during an insonification, and all individuals within this volume will contribute to the volume backscattering coefficient of sample s sv,s,bm=Σk=1Nk,s,bmΣi=1nkσbs,i,k bbm2(θi,k,bm , φi,k,bm)Vs,bm, (3) where bbm2(θi,k,bm , φi,k,bm) is the two-way beam pattern regarding the ith individual within voxel k. The external sum in Equation (3) adds across the Nk,s,bm voxels that can contribute backscatter to sample s, while the internal one sums the backscattering cross-sections within these voxels, weighted by the beam pattern. Here, the beam angles θi,k,bm   and  φi,k,bm are referenced to the acoustic axis and operate on the orthogonal directions that correspond to the vertical (elevation) and horizontal (azimuth) aspect of the beam, respectively. Assuming that the beam angle differences for all individuals within a specific voxel are negligible, θi,k,bm and φi,k,bm can be replaced with θk,bm and φk,bm, respectively, i.e. the elevation and azimuth angle of the voxel’s geometric centre relative to the acoustic axis of beam bm. Combining Equations (1), (2), and (3) results in the following expression that describes the volume backscattering coefficient measured by the simulator: sv,s,bm=VvoxΣk=1Nk,s,bmsv,k bbm2(θk,bm, φk,bm)13((rs+Δr/2)3− (rs−Δr/2)3)ψbm; (4) i.e. for each insonification, the volume backscattering strength Sv,s,bm = 10 log10(sv,s,bm) of each sample s in the simulated echogram is calculated as the weighted sum of the (known) volume backscattering coefficients sv,k of the Nk,s,bm voxels that effectively contribute backscatter to Vs,bm. In order to acquire Nk,s,bm, the simulator selects the voxels whose radial position rk is (rs – Δr/2 < rk ≤ rs + Δr/2), and, to reduce unnecessary computations on voxels whose backscatter is practically nulled by the beam pattern, it filters out any voxels that have angular position θk,bm and/or φk,bm greater than five times the full beam width bwθ and bwϕ, respectively. If a voxel k contributes to more than one range-ring of the echogram (i.e. it spans across multiple samples in the along-beam direction), Equation (4) is modified for that k by replacing Vvox with vk,s, where vk,s is the approximate portion of the voxel’s volume that falls within sample s (0 < vk,s < Vvox); to perform the latter calculations, it is temporarily assumed that the voxel’s face is normal to the acoustic axis and the samples have a planar (instead of a spherical) face in the along-beam direction. For each simulated insonification, the position of school voxels relative to the sonar beams is assumed to be constant between transmission and reception. Transducer model The beam pattern of the SP90 transducer was manually digitized from the sonar manufacturer diagrams (26 kHz, normal mode, see Brehmer et al., 2007 and sources therein) with a 0.5° step over the [–25°, +25°] domain centred to the acoustic axis. Digitization of the beam pattern was performed separately for the horizontal (bh) and vertical (bv) aspect, and piecewise cubic hermite interpolating polynomials (PCHIP) (Fritsch and Carlson, 1980) were used to model bh and bv as a function of the beam angles φk and θk, respectively. The shape of only one beam was mapped as described, and its directivity was applied to all beams of the simulated omnidirectional fan. Note that, when operating in the horizontal mode simulated here, the sonar is directive in the horizontal aspect during reception, but the overall transmitted wave is omnidirectional when viewed as a projection on a horizontal plane. Therefore, for each beam, the two-way beam pattern used by the simulator in the horizontal (azimuth) aspect is bh(φk)=bhTx(φk) bhRx(φk), (5) where bhTx(φk)=1 ∀φk, and the two-way directivity in the vertical (elevation) aspect is bv(θk)=bvTx(θk) bvRx(θk), (6) where bvTx(θk)=bvRx(θk). Superscripts Tx and Rx in Equations (5) and (6) denote transmission and reception, respectively, while the parameters bhRx(φk) and bvTx(θk) are the output of the PCHIP model. According to these definitions, the two-way beam pattern used in Equation (4) is bbm2(θk, φk)=bv(θk) bh(φk) (7) and is assumed identical for all beams bm. The equivalent beam angle ψbm was estimated from the integral of the entire beam pattern over the hemisphere in front of the beam (Simmonds and MacLennan, 2005). Simulated scenarios The stationary transducer was configured to simulate the SP90 sonar and was positioned at (x, y, z) = (0, 0, –4 m) in the global Cartesian coordinate system. Sonar observation range was set to 900 m, and beam tilt angle was –5° below the horizon. The sonar was subsequently rotated about the vertical so that, when projected on the x–y plane, the acoustic axis of a beam was aligned with the +x-axis (starboard side). The transducer placement and insonification settings remained fixed throughout the simulations. Ten school size classes were created, characterized by their true maximum horizontal cross-sectional area (A0) that ranged from 1000 to 10 000 m2. Ten different true mean volume backscattering strength (Ŝv,0) classes were allocated per school size class, producing 100 simulated schools in total (Table 1); voxel volume Vvox was 8 m3 (2 × 2 × 2 m) for all schools. During the simulations, the geometric centre of a school was incrementally positioned at (x, y, z) = (XG,0,n, 0, ZG,0) in the global Cartesian coordinate system, where the subscript n denotes the nth insonification and ZG,0 is the school depth. Depending on school size class, ZG,0 varied between –25 and –35 m to accommodate the school’s width in the vertical dimension. Considering this positioning setup, the XG,0,n coordinate practically governs the school’s distance from the transducer, and was set to XG,0,n = 100 + 20(n – 1) m, where n ∈ [1, 2, …, 36]. Only one school was present in the simulated world per insonification, and ZG,0 remained constant as a specific school was translated (with no rotation) away from the transducer. Table 1. Summary of simulated scenarios, listing the stationary transducer settings, the true properties of the simulated schools, and their position on the global Cartesian coordinate system. Transducer properties Range Position Tilt 900 m (x, y, z) = (0, 0, –4) m –5° School properties A0 (m2) Ŝv,0 (dB) Ŀx,0 (m) Ŀy,0 (m) Ŀz,0 (m) ZG,0 (m) 1000 [–42, …, –33]a 36 36 18 –25 2000 [–42, …, –33]a 48 48 24 –25 3000 [–42, …, –33]a 60 60 26 –28 4000 [–42, …, –33]a 72 72 32 –28 5000 [–42, …, –33]a 80 80 36 –30 6000 [–42, …, –33]a 88 88 38 –30 7000 [–42, …, –33]a 96 96 40 –32 8000 [–42, …, –33]a 100 100 42 –32 9000 [–42, …, –33]a 108 108 44 –35 10 000 [–42, …, –33]a 112 112 46 –35 School position (x, y, z) = (XG,0, 0, ZG,0), where XG,0 = [100, …, 800] in steps of 20 m Transducer properties Range Position Tilt 900 m (x, y, z) = (0, 0, –4) m –5° School properties A0 (m2) Ŝv,0 (dB) Ŀx,0 (m) Ŀy,0 (m) Ŀz,0 (m) ZG,0 (m) 1000 [–42, …, –33]a 36 36 18 –25 2000 [–42, …, –33]a 48 48 24 –25 3000 [–42, …, –33]a 60 60 26 –28 4000 [–42, …, –33]a 72 72 32 –28 5000 [–42, …, –33]a 80 80 36 –30 6000 [–42, …, –33]a 88 88 38 –30 7000 [–42, …, –33]a 96 96 40 –32 8000 [–42, …, –33]a 100 100 42 –32 9000 [–42, …, –33]a 108 108 44 –35 10 000 [–42, …, –33]a 112 112 46 –35 School position (x, y, z) = (XG,0, 0, ZG,0), where XG,0 = [100, …, 800] in steps of 20 m Voxel backscattering coefficients are exponentially distributed and average at Ŝv,0 when expressed in dB. Ŀx,0, Ŀy,0, Ŀz,0 represent the maximum school width along the x-, y-, z-axis, respectively; ZG,0 is the depth of the geometric centre of the school; A0 is the school area at the largest horizontal cross-section, i.e. on the x–y plane at z = ZG,0. a In steps of 1 dB. Table 1. Summary of simulated scenarios, listing the stationary transducer settings, the true properties of the simulated schools, and their position on the global Cartesian coordinate system. Transducer properties Range Position Tilt 900 m (x, y, z) = (0, 0, –4) m –5° School properties A0 (m2) Ŝv,0 (dB) Ŀx,0 (m) Ŀy,0 (m) Ŀz,0 (m) ZG,0 (m) 1000 [–42, …, –33]a 36 36 18 –25 2000 [–42, …, –33]a 48 48 24 –25 3000 [–42, …, –33]a 60 60 26 –28 4000 [–42, …, –33]a 72 72 32 –28 5000 [–42, …, –33]a 80 80 36 –30 6000 [–42, …, –33]a 88 88 38 –30 7000 [–42, …, –33]a 96 96 40 –32 8000 [–42, …, –33]a 100 100 42 –32 9000 [–42, …, –33]a 108 108 44 –35 10 000 [–42, …, –33]a 112 112 46 –35 School position (x, y, z) = (XG,0, 0, ZG,0), where XG,0 = [100, …, 800] in steps of 20 m Transducer properties Range Position Tilt 900 m (x, y, z) = (0, 0, –4) m –5° School properties A0 (m2) Ŝv,0 (dB) Ŀx,0 (m) Ŀy,0 (m) Ŀz,0 (m) ZG,0 (m) 1000 [–42, …, –33]a 36 36 18 –25 2000 [–42, …, –33]a 48 48 24 –25 3000 [–42, …, –33]a 60 60 26 –28 4000 [–42, …, –33]a 72 72 32 –28 5000 [–42, …, –33]a 80 80 36 –30 6000 [–42, …, –33]a 88 88 38 –30 7000 [–42, …, –33]a 96 96 40 –32 8000 [–42, …, –33]a 100 100 42 –32 9000 [–42, …, –33]a 108 108 44 –35 10 000 [–42, …, –33]a 112 112 46 –35 School position (x, y, z) = (XG,0, 0, ZG,0), where XG,0 = [100, …, 800] in steps of 20 m Voxel backscattering coefficients are exponentially distributed and average at Ŝv,0 when expressed in dB. Ŀx,0, Ŀy,0, Ŀz,0 represent the maximum school width along the x-, y-, z-axis, respectively; ZG,0 is the depth of the geometric centre of the school; A0 is the school area at the largest horizontal cross-section, i.e. on the x–y plane at z = ZG,0. a In steps of 1 dB. Extraction of simulated fish school descriptors A range of cut-off thresholds T (–53 to –47 dB, in steps of 1 dB) were applied on each simulated echogram, and a series of morphometric, energetic, and positional echotrace descriptors were extracted per threshold level, using the multi-beam school detection algorithm described in Trygonis et al. (2009). Echotraces consisting of ≤ 10 echogram samples were excluded from all further analyses. Based on these simulation results, where both the true properties of the schools and the observed echotrace descriptors were known, (observed) correction factors were calculated per threshold level T to assess the systematic bias on school horizontal area and mean volume backscattering strength. Specifically, the area correction factor ACF was defined as the dimensionless number that the observed echotrace area must be multiplied with, in order to match the true value: A0 = Aobs × ACF. For the logarithmic mean volume backscattering strength of the school, the correction factor Ŝv,CF was defined as the difference (in dB) between the observed and the true value: Ŝv,CF = Ŝv,obs – Ŝv,0. Accordingly, ACF < 1 denotes overestimation of the true school area, while Ŝv,CF < 0 corresponds to underestimation of the true school density. Development of correction models After calculating the observed correction factors, multiple linear regression models were developed in order to investigate the ability to predict the true mean Sv and true area of a school (i.e. predict the correction factors) using only descriptors extracted from the school’s echotrace. The underlying hypothesis is that, once all sources of bias that are intrinsic to the system are in effect, including the systematic distortions over the nominal echogram geometry, the resulting echotrace retains enough information to statistically infer the true properties of the school with a sufficient level of confidence. To this end, separate models were built for Ŝv,CF and ACF, in which the correction factor was the dependent variable and predictors were the school range (RG,obs, m), the number of school echotrace samples (nsobs), the area (Aobs, m2), the maximum along-beam (Ŀwobs, m) and across-beam width (Ċwobs, m), the ratio Ŀwobs/RG,obs, the mean (Ŝv,obs, dB), max (Ṡv,obs,dB) and sum (ΣSv,obs, dB) volume backscattering strength, and the log standard deviation of the volume backscattering coefficients of school echotrace samples [log10(σsv,obs), m−1]; see Trygonis et al. (2009) for details on the calculations of school descriptors. Note that Ŝv,obs and ΣSv,obs are computed using the linear volume backscattering coefficients of an observed echotrace, and then transformed to dB (MacLennan et al., 2002). Statistically non-significant (p > 0.05) predictors were removed from each regression model, and the latter were rebuilt. The Ŝv,CF and ACF prediction models were developed on training data and validated against unknown (testing) ones. Note that, in order to assess the limitations of this descriptor correction approach in echotraces of low Ŝv,obs, several candidate multiple linear regression models were initially built using training data with varying ΔŜT = Ŝv,obs – T, i.e. with varying difference (in dB) between the observed mean volume backscattering strength of a school and the cut-off threshold. Specifically, all available simulated data (i.e. 100 voxel schools of varying A0 and Ŝv,0, each insonified at different distances XG,0 and incrementally processed with seven thresholds T) were pooled, and then filtered by increasing levels of ΔŜT (≥ 0, 1, 2, 3, 4, 5, and 6 dB). Within each ΔŜT pool, stratified random sampling was performed with a 50–50% partitioning scheme: for each unique combination of [Ŝv,0, A0, T], the insonification distance XG,0 was split into [140–240, 260–360, …, 620–720, 740–800 m] bins, and 50% of echotraces falling within each XG,0 bin were randomly selected and moved to the training data set, while the other 50% were moved to the testing one. This produced seven preliminary training/testing data sets (one per ΔŜT level), within which the relative proportion of Ŝv,0, A0, XG,0, and T instances was maintained, but without any common cases between training and testing data. One Ŝv,CF and one ACF candidate prediction model was then built per ΔŜT training data set, and was applied to the respective testing data. The candidate models built from the lowest ΔŜT, which produced consistent prediction performance with all higher ΔŜT models were selected as the final correction models presented herein. Note that the two smallest XG,0 = [100, 120 m] classes were omitted from all models (and subsequent corrections). This was based on a preliminary analysis, which showed that, due to the fact that the largest portion of the school—or its entirety—was outside the tilted acoustic beams at such short distances from the transducer, the inclusion of these few, insufficiently sampled by the sonar, echotraces reduced the performance of all candidate Ŝv,CF and ACF models. Results The insonification of the 100 simulated schools across 36 different positions relative to the stationary transducer (Table 1) produced 3600 cases per threshold level T, yielding a total of 25 200 simulated pings that helped investigate the variability in the observed descriptors. To facilitate the presentation, detailed results are initially shown for the simulation scenarios processed with T = –51 dB; the analysis of the total simulation output is presented further below in this section. Simulation scenarios processed with threshold –51 dB Figure 2a and e shows the simulated echograms of two schools that have identical true geometric characteristics, but different true mean volume backscattering strength, insonified at various distances from the stationary transducer. Selected geometric and energetic descriptors of the observed echotraces and their relation to the respective true values are plotted in Figure 2c–d and g–h. The simulation examples of Figure 2 demonstrate the angular and range-dependent distortion of echotraces, i.e. the smearing of the echotrace of a school over the span of all beams that fully or partially contain it (see also Figure 2 in Vatnehol et al., 2017 for a schematic illustration of this distortion). The number of beams that effectively sample the school at a given range also varies, and depends on beam directivity, the processing threshold, the density of the school, and its position within the acoustic sampling volume. This smearing effect has small impact on the along-beam width of the school (Lwobs), but it typically leads to the overestimation of the across-beam width (Cwobs, about 50–100% at 200 m from the transducer), which is the primary contributor to the respective overestimation in the observed school area. Conversely, the average density of the school is systematically underestimated (Figure 2d and h). Low Sv samples at the periphery of the school lower the overall computed mean, while, as in single-beam echosounding (Diner, 2001), the bias further increases with range, as the school fills up a smaller proportion of the—progressively larger—acoustic sampling volume; this also applies to the central beam(s) that insonify the school, depending on the beam tilt angle and the school’s true depth and size. Moreover, schools at very short distances to the transducer are severely under-sampled. Figure 2. View largeDownload slide (a) Simulated echotraces of a school with true area A0 = 5000 m2 and true mean volume backscattering strength Ŝv,0 = –36 dB insonified at various distances from the stationary transducer, and processed with a threshold T = –51 dB; the intermediate insonifications at XG,0 = [120, 140, 160, 180, 220, …, 780 m] are not shown. At each insonification position, the projection of the school’s geometric centre on the x–y plane is denoted with a closed marker, and a circle marks the school’s outer boundary. (b) x–z view of the simulated scenario. The sonar beams are tilted at –5°, the transducer depth is –4 m, and the depth of the school’s geometric centre is constant at ZG,0 = –30 m; the grey patch represents the acoustic beam that is aligned with the +x-axis. (c) Observed geometric echotrace descriptors vs. the respective true properties. (d) Observed maximum (Ṡv,obs) and mean (Ŝv,obs) volume backscattering strength; the horizontal line at –36 dB denotes the true mean. (e–h) As in panels a–d, but for a school with the same true area and lower Ŝv,0 (= –41 dB). The dashed ellipse in panel e and the dashed lines with open markers in panels g and h denote echotraces with ΔŜT < 2 dB (i.e. Ŝv,obs < –49 dB), which could not be corrected with the final Ŝv,CF and ACF and prediction models. The “x” markers and dotted lines in panels c, d, g, and h show the observations at XG,0 = [100, 120 m], which were omitted from all analyses. Figure 2. View largeDownload slide (a) Simulated echotraces of a school with true area A0 = 5000 m2 and true mean volume backscattering strength Ŝv,0 = –36 dB insonified at various distances from the stationary transducer, and processed with a threshold T = –51 dB; the intermediate insonifications at XG,0 = [120, 140, 160, 180, 220, …, 780 m] are not shown. At each insonification position, the projection of the school’s geometric centre on the x–y plane is denoted with a closed marker, and a circle marks the school’s outer boundary. (b) x–z view of the simulated scenario. The sonar beams are tilted at –5°, the transducer depth is –4 m, and the depth of the school’s geometric centre is constant at ZG,0 = –30 m; the grey patch represents the acoustic beam that is aligned with the +x-axis. (c) Observed geometric echotrace descriptors vs. the respective true properties. (d) Observed maximum (Ṡv,obs) and mean (Ŝv,obs) volume backscattering strength; the horizontal line at –36 dB denotes the true mean. (e–h) As in panels a–d, but for a school with the same true area and lower Ŝv,0 (= –41 dB). The dashed ellipse in panel e and the dashed lines with open markers in panels g and h denote echotraces with ΔŜT < 2 dB (i.e. Ŝv,obs < –49 dB), which could not be corrected with the final Ŝv,CF and ACF and prediction models. The “x” markers and dotted lines in panels c, d, g, and h show the observations at XG,0 = [100, 120 m], which were omitted from all analyses. Summarizing the scenarios processed with threshold T = –51 dB, Figure 3 shows the observed mean volume backscattering strength (Ŝv,obs) and area (Aobs) of simulated echotraces; the same values expressed as relative percentage errors δ are plotted in Supplementary Figure S1. The results show that the Ŝv,obs extracted from the echogram always underestimates the true value Ŝv,0, and does so by a magnitude of 4–16 dB (off-axis insonifications included), depending on the range, the true size, and the true density of the school. Overall, smaller schools fall below the detection threshold at closer ranges to the transducer, and for the same Ŝv,0 class, they are susceptible to a stronger underestimation of their true density and larger relative errors in their observed area (Aobs). For the tilted omnidirectional scenarios considered herein, the variability of Aobs is also greater in smaller schools, as the same target that maintains a constant depth is observed at different distances from the transducer (Supplementary Figure S1e–h). It can be suggested that for these insonification conditions, a favourable school observation range is about 150–300 m. Figure 3. View largeDownload slide (a–d) Observed mean volume backscattering strength Ŝv,obs and (e–h) area Aobs of simulated schools vs. their distance from the transducer (training and testing data pooled), tabulated by true Ŝv,0 and A0. Threshold T is –51 dB. The Ŝv,0 = [–41, –40, –38, –37, –35, –34 dB] and A0 = [7000, 9000 m2] classes are not plotted to reduce clutter. Dashed lines with open markers denote echotraces with ΔŜT < 2 dB (i.e. Ŝv,obs < –49 dB, marked by the horizontal line in panels a–d), which could not be corrected with the final Ŝv,CF and ACF and prediction models. The “x” markers and dashed lines show the observations at XG,0 = [100, 120 m], which were omitted from all analyses (i.e. XG,0 < 140 m, marked by the vertical line in panels a–d). Figure 3. View largeDownload slide (a–d) Observed mean volume backscattering strength Ŝv,obs and (e–h) area Aobs of simulated schools vs. their distance from the transducer (training and testing data pooled), tabulated by true Ŝv,0 and A0. Threshold T is –51 dB. The Ŝv,0 = [–41, –40, –38, –37, –35, –34 dB] and A0 = [7000, 9000 m2] classes are not plotted to reduce clutter. Dashed lines with open markers denote echotraces with ΔŜT < 2 dB (i.e. Ŝv,obs < –49 dB, marked by the horizontal line in panels a–d), which could not be corrected with the final Ŝv,CF and ACF and prediction models. The “x” markers and dashed lines show the observations at XG,0 = [100, 120 m], which were omitted from all analyses (i.e. XG,0 < 140 m, marked by the vertical line in panels a–d). For each combination of (Ŝv,0, A0, distance from transducer) plotted in Figure 3, the deviation between the observed and the known true value was used to compute an observed correction factor for the mean volume backscattering strength and area (Ŝv,CF and ACF, respectively). Note that echotraces with ΔŜT < 2 dB (i.e. with Ŝv,obs < –49 dB, shown with open markers in Figure 3) were maintained in the data set, but were eventually excluded from the final, model-based corrections. This was based on the output of all simulations analysed (see the respective section below), which showed that, for all examined processing thresholds T, echotraces with Ŝv,obs comparable to T and/or Ċwobs comparable to the beam width (examples shown in Figure 2e) could not be sufficiently corrected with the final Ŝv,CF and ACF and prediction models. Similarly, under-sampled schools that were detected at very short distances from the transducer (XG,0 = [100, 120 m], corresponding to RG,obs of about < 130 m) were omitted from all analyses. The observed correction factors Ŝv,CF and ACF are plotted in Figure 4, and represent the exact corrections that must be applied to the respective echotrace descriptors when the true properties of the school are known. In practical applications, however, the true properties are not known, and the only information available about the insonified school is its acoustic image displayed in the multi-beam echogram. Figure 4. View largeDownload slide Observed correction factors for the (a) mean volume backscattering strength and (b) area of simulated echotraces (training and testing data pooled) vs. the school distance from the transducer. Threshold T is –51 dB. Each observed correction factor corresponds to a data point plotted in Figure 3, with the addition of all Ŝv,0 and A0 classes that were not shown therein, and is marked respectively: closed markers denote echotraces with ΔŜT ≥ 2 dB (N = 2708), open markers correspond to echotraces with ΔŜT < 2 dB (N = 318), and “x” markers (N = 196) show the observations at XG,0 = [100, 120 m], which were omitted from all analyses. The overlaid mean and its confidence interval per distance category refer to the ΔŜT ≥ 2 dB data points only, and were estimated after bootstrapping (α = 0.05, 10 000 resamples with replacement); error bars show the standard deviation. Figure 4. View largeDownload slide Observed correction factors for the (a) mean volume backscattering strength and (b) area of simulated echotraces (training and testing data pooled) vs. the school distance from the transducer. Threshold T is –51 dB. Each observed correction factor corresponds to a data point plotted in Figure 3, with the addition of all Ŝv,0 and A0 classes that were not shown therein, and is marked respectively: closed markers denote echotraces with ΔŜT ≥ 2 dB (N = 2708), open markers correspond to echotraces with ΔŜT < 2 dB (N = 318), and “x” markers (N = 196) show the observations at XG,0 = [100, 120 m], which were omitted from all analyses. The overlaid mean and its confidence interval per distance category refer to the ΔŜT ≥ 2 dB data points only, and were estimated after bootstrapping (α = 0.05, 10 000 resamples with replacement); error bars show the standard deviation. Analysis of the entire simulation Following the same procedures described above for T = –51 dB, all simulation scenarios listed in Table 1 were processed with a range of additional cut-off thresholds T (–53, –52, –50, –49, –48, and –47 dB), and the school detector was run for each threshold level; the echotraces were then tabulated according to their ΔŜT (≥ 0, 1, 2, 3, 4, 5, and 6 dB) and split into stratified training/testing data sets. Separate Ŝv,CF and ACF candidate prediction models were built per ΔŜT training set, in order to investigate the range of ΔŜT values at which the correction method should not be used. The results are summarized in Table 2, shown as adjusted R2 values of the candidate models and as mean absolute percentage errors |δ| (mean ± SD) of the corresponding corrections performed on the respective (i.e. same ΔŜT) testing data; the errors |δ| for mean volume backscatter were computed using the linear coefficients (|δs̄v,c| = |s̄v,c – s̄v,0|/s̄v,0 × 100). The results show that the two lowest ΔŜT data sets (≥ 0 and ≥ 1 dB) had poor correction performance; their mean percentage errors were almost double for |δs̄v,c| than all other ΔŜT cases (e.g. 9.4 ± 7.9% for ΔŜT ≥ 0 dB and 6.4 ± 6.0% for ΔŜT ≥ 2 dB), and were almost triple for |δAc|, reaching up to 13.6 ± 12.0% for ΔŜT ≥ 0 dB. Table 2. Performance of 14 candidate multiple linear regression models for predicting the mean volume backscattering strength and area correction factors (Ŝv,CF and ACF, respectively). Ŝv,CF ACF ΔŜT (dB) Adj. R2 |δs̄v,c| % Adj. R2 |δAc| % N ≥0 0.951 9.4 ± 7.9 0.905 13.6 ± 12.0 10031 ≥1 0.952 9.2 ± 7.9 0.882 13.4 ± 11.6 9910 ≥2a 0.974 6.4 ± 6.0 0.966 5.2 ± 4.8 8601 ≥3 0.975 5.8 ± 5.3 0.971 4.4 ± 3.9 7555 ≥4 0.975 5.4 ± 5.0 0.973 4.0 ± 3.3 6484 ≥5 0.975 5.0 ± 4.7 0.974 3.6 ± 3.1 5321 ≥6 0.977 4.6 ± 4.4 0.976 3.3 ± 2.7 4137 Ŝv,CF ACF ΔŜT (dB) Adj. R2 |δs̄v,c| % Adj. R2 |δAc| % N ≥0 0.951 9.4 ± 7.9 0.905 13.6 ± 12.0 10031 ≥1 0.952 9.2 ± 7.9 0.882 13.4 ± 11.6 9910 ≥2a 0.974 6.4 ± 6.0 0.966 5.2 ± 4.8 8601 ≥3 0.975 5.8 ± 5.3 0.971 4.4 ± 3.9 7555 ≥4 0.975 5.4 ± 5.0 0.973 4.0 ± 3.3 6484 ≥5 0.975 5.0 ± 4.7 0.974 3.6 ± 3.1 5321 ≥6 0.977 4.6 ± 4.4 0.976 3.3 ± 2.7 4137 Each candidate model is built from training data that encompass simulated echotraces that fulfil the specific ΔŜT criterion. The mean absolute percentage errors |δ| (mean ± SD) refer to the performance of models on the corresponding testing data. Volume backscatter errors |δ| were computed using the linear (s̄v) coefficients; N shows the echotrace count per testing data set, p < 0.001 for all models (see also Supplementary Table S1 for the performance of preliminary candidate models that include the two smallest XG,0 = [100, 120 m] classes, which were omitted from all analyses presented herein). a Models selected as “final” and listed in Table 3. Table 2. Performance of 14 candidate multiple linear regression models for predicting the mean volume backscattering strength and area correction factors (Ŝv,CF and ACF, respectively). Ŝv,CF ACF ΔŜT (dB) Adj. R2 |δs̄v,c| % Adj. R2 |δAc| % N ≥0 0.951 9.4 ± 7.9 0.905 13.6 ± 12.0 10031 ≥1 0.952 9.2 ± 7.9 0.882 13.4 ± 11.6 9910 ≥2a 0.974 6.4 ± 6.0 0.966 5.2 ± 4.8 8601 ≥3 0.975 5.8 ± 5.3 0.971 4.4 ± 3.9 7555 ≥4 0.975 5.4 ± 5.0 0.973 4.0 ± 3.3 6484 ≥5 0.975 5.0 ± 4.7 0.974 3.6 ± 3.1 5321 ≥6 0.977 4.6 ± 4.4 0.976 3.3 ± 2.7 4137 Ŝv,CF ACF ΔŜT (dB) Adj. R2 |δs̄v,c| % Adj. R2 |δAc| % N ≥0 0.951 9.4 ± 7.9 0.905 13.6 ± 12.0 10031 ≥1 0.952 9.2 ± 7.9 0.882 13.4 ± 11.6 9910 ≥2a 0.974 6.4 ± 6.0 0.966 5.2 ± 4.8 8601 ≥3 0.975 5.8 ± 5.3 0.971 4.4 ± 3.9 7555 ≥4 0.975 5.4 ± 5.0 0.973 4.0 ± 3.3 6484 ≥5 0.975 5.0 ± 4.7 0.974 3.6 ± 3.1 5321 ≥6 0.977 4.6 ± 4.4 0.976 3.3 ± 2.7 4137 Each candidate model is built from training data that encompass simulated echotraces that fulfil the specific ΔŜT criterion. The mean absolute percentage errors |δ| (mean ± SD) refer to the performance of models on the corresponding testing data. Volume backscatter errors |δ| were computed using the linear (s̄v) coefficients; N shows the echotrace count per testing data set, p < 0.001 for all models (see also Supplementary Table S1 for the performance of preliminary candidate models that include the two smallest XG,0 = [100, 120 m] classes, which were omitted from all analyses presented herein). a Models selected as “final” and listed in Table 3. Overall, Table 2 shows that filtering-out echotraces with low Ŝv,obs (i.e. keeping only the high ΔŜT cases) significantly improved the correction performance. For example, the correction models built from the ΔŜT ≥ 6 dB testing echotraces achieved a mean |δs̄v,c| and |δAc| error of 4.6 ± 4.4% and 3.3 ± 2.7%, respectively, when applied to the corresponding testing data. The higher the ΔŜT filter, however, the more echotraces are removed out of the usable data set. While the selection of a ΔŜT criterion other than ≥ 0 and ≥ 1 is largely an ad hoc decision, a compromise appears to be ΔŜT ≥ 2 dB; here, a large pool of simulated data is maintained to build the correction models (thus avoiding over-fitting the higher ΔŜT subsets), the variability of |δs̄v,c| and |δAc| errors drops significantly when compared to the two lower ΔŜT sets, and is the lowest ΔŜT at which the mean errors |δ| and adjusted R2 values appear to stabilize. Based on these criteria, the Ŝv,CF and ACF prediction models built from, and applicable to, echotraces with ΔŜT ≥ 2 dB were selected as the final ones, and are listed in Table 3. Both models use only descriptors extracted from the observed echotraces as predictive variables, and they explain more than 96% of the variance in their training data set (N = 8618 simulated echotraces) with a standard error of 0.379 dB for Ŝv,CF and 0.040 for ACF. Table 3. Final multiple linear regression models for predicting the mean volume backscattering strength and area correction factors (Ŝv,CF and ACF, respectively), built from the ΔŜT ≥ 2 dB training data set (N = 8618 simulated echotraces). Dependent Adjusted R2 S.E. of the estimate Dependent Adjusted R2 S.E. of the estimate Ŝv,CF 0.974 0.379 ACF 0.966 0.040 Predictors B S.E. B b S.E. b Predictors B S.E. B b S.E. b (constant) –23.351291 0.217 (constant) –1.465251 0.028 ΣSv,obs 2.060 0.026 1.113656 0.014 log10(σsv,obs) –3.711 0.033 –2.218302 0.020 log10(σsv,obs) –1.579 0.016 –10.183501 0.101 Ṡv,obs 1.797 0.051 0.125210 0.004 RG,obs –1.327 0.006 –0.017038 0.000 ΣSv,obs 1.255 0.030 0.062886 0.002 Ŀwobs/RG,obs –0.627 0.008 –11.629761 0.157 nsobs –1.015 0.010 –0.005403 0.000 Ŀwobs 0.521 0.011 0.057721 0.001 Ŀwobs/RG,obs 0.991 0.010 1.702406 0.017 nsobs –0.496 0.009 –0.028484 0.000 RG,obs –0.845 0.007 –0.001006 0.000 Ŝv,obs –0.092 0.013 –0.081974 0.011 Aobs 0.650 0.011 0.000025 0.000 Aobs 0.074 0.009 0.000030 0.000 Ŝv,obs 0.416 0.025 0.034432 0.002 Ċwobs –0.051 0.006 –0.002091 0.000 Ċwobs 0.238 0.007 0.000909 0.000 Ŀwobs –0.199 0.013 –0.002044 0.000 Dependent Adjusted R2 S.E. of the estimate Dependent Adjusted R2 S.E. of the estimate Ŝv,CF 0.974 0.379 ACF 0.966 0.040 Predictors B S.E. B b S.E. b Predictors B S.E. B b S.E. b (constant) –23.351291 0.217 (constant) –1.465251 0.028 ΣSv,obs 2.060 0.026 1.113656 0.014 log10(σsv,obs) –3.711 0.033 –2.218302 0.020 log10(σsv,obs) –1.579 0.016 –10.183501 0.101 Ṡv,obs 1.797 0.051 0.125210 0.004 RG,obs –1.327 0.006 –0.017038 0.000 ΣSv,obs 1.255 0.030 0.062886 0.002 Ŀwobs/RG,obs –0.627 0.008 –11.629761 0.157 nsobs –1.015 0.010 –0.005403 0.000 Ŀwobs 0.521 0.011 0.057721 0.001 Ŀwobs/RG,obs 0.991 0.010 1.702406 0.017 nsobs –0.496 0.009 –0.028484 0.000 RG,obs –0.845 0.007 –0.001006 0.000 Ŝv,obs –0.092 0.013 –0.081974 0.011 Aobs 0.650 0.011 0.000025 0.000 Aobs 0.074 0.009 0.000030 0.000 Ŝv,obs 0.416 0.025 0.034432 0.002 Ċwobs –0.051 0.006 –0.002091 0.000 Ċwobs 0.238 0.007 0.000909 0.000 Ŀwobs –0.199 0.013 –0.002044 0.000 Excluding b, values are rounded to the third decimal. B and b are the standardized and raw regression coefficients, respectively; p < 0.001 for both models. Table 3. Final multiple linear regression models for predicting the mean volume backscattering strength and area correction factors (Ŝv,CF and ACF, respectively), built from the ΔŜT ≥ 2 dB training data set (N = 8618 simulated echotraces). Dependent Adjusted R2 S.E. of the estimate Dependent Adjusted R2 S.E. of the estimate Ŝv,CF 0.974 0.379 ACF 0.966 0.040 Predictors B S.E. B b S.E. b Predictors B S.E. B b S.E. b (constant) –23.351291 0.217 (constant) –1.465251 0.028 ΣSv,obs 2.060 0.026 1.113656 0.014 log10(σsv,obs) –3.711 0.033 –2.218302 0.020 log10(σsv,obs) –1.579 0.016 –10.183501 0.101 Ṡv,obs 1.797 0.051 0.125210 0.004 RG,obs –1.327 0.006 –0.017038 0.000 ΣSv,obs 1.255 0.030 0.062886 0.002 Ŀwobs/RG,obs –0.627 0.008 –11.629761 0.157 nsobs –1.015 0.010 –0.005403 0.000 Ŀwobs 0.521 0.011 0.057721 0.001 Ŀwobs/RG,obs 0.991 0.010 1.702406 0.017 nsobs –0.496 0.009 –0.028484 0.000 RG,obs –0.845 0.007 –0.001006 0.000 Ŝv,obs –0.092 0.013 –0.081974 0.011 Aobs 0.650 0.011 0.000025 0.000 Aobs 0.074 0.009 0.000030 0.000 Ŝv,obs 0.416 0.025 0.034432 0.002 Ċwobs –0.051 0.006 –0.002091 0.000 Ċwobs 0.238 0.007 0.000909 0.000 Ŀwobs –0.199 0.013 –0.002044 0.000 Dependent Adjusted R2 S.E. of the estimate Dependent Adjusted R2 S.E. of the estimate Ŝv,CF 0.974 0.379 ACF 0.966 0.040 Predictors B S.E. B b S.E. b Predictors B S.E. B b S.E. b (constant) –23.351291 0.217 (constant) –1.465251 0.028 ΣSv,obs 2.060 0.026 1.113656 0.014 log10(σsv,obs) –3.711 0.033 –2.218302 0.020 log10(σsv,obs) –1.579 0.016 –10.183501 0.101 Ṡv,obs 1.797 0.051 0.125210 0.004 RG,obs –1.327 0.006 –0.017038 0.000 ΣSv,obs 1.255 0.030 0.062886 0.002 Ŀwobs/RG,obs –0.627 0.008 –11.629761 0.157 nsobs –1.015 0.010 –0.005403 0.000 Ŀwobs 0.521 0.011 0.057721 0.001 Ŀwobs/RG,obs 0.991 0.010 1.702406 0.017 nsobs –0.496 0.009 –0.028484 0.000 RG,obs –0.845 0.007 –0.001006 0.000 Ŝv,obs –0.092 0.013 –0.081974 0.011 Aobs 0.650 0.011 0.000025 0.000 Aobs 0.074 0.009 0.000030 0.000 Ŝv,obs 0.416 0.025 0.034432 0.002 Ċwobs –0.051 0.006 –0.002091 0.000 Ċwobs 0.238 0.007 0.000909 0.000 Ŀwobs –0.199 0.013 –0.002044 0.000 Excluding b, values are rounded to the third decimal. B and b are the standardized and raw regression coefficients, respectively; p < 0.001 for both models. When the models of Table 3 were applied to the ΔŜT ≥ 2 dB testing data (N = 8601), relative percentage errors δ in the observed mean volume backscatter and area of simulated echotraces decreased significantly, and were normally distributed around 0% (Figure 5). In terms of mean absolute percentage errors, |δs̄v,obs| averaged at 79.5 ± 9.6% (percentiles P5%–P95%: 63.1–93.4%) and dropped to |δs̄v,c| = 6.4 ± 6.0% (Table 2) after correction (P5%–P95%: 0.5–19.1%); the respective |δs̄v,c| errors in the ΔŜT ≥ 2 dB training data (N = 8618) were 6.3 ± 5.9%. With respect to area, errors in the testing data were |δAobs| = 100.7 ± 65.2% (P5%–P95%: 10.9–220.0%) and reduced to |δAc| = 5.2 ± 4.8% (P5%–P95%: 0.4–13.7%) after corrections; in the ΔŜT ≥ 2 dB training data, |δAc| was 5.1 ± 4.7%. Figure 5. View largeDownload slide Distribution of relative percentage errors δ in the mean volume backscatter (s̄v) and area (A) of all simulated echotraces in the ΔŜT ≥ 2 dB testing data set (N = 8601), (a–b) before and (c–d) after correction with the models of Table 3. Figure 5. View largeDownload slide Distribution of relative percentage errors δ in the mean volume backscatter (s̄v) and area (A) of all simulated echotraces in the ΔŜT ≥ 2 dB testing data set (N = 8601), (a–b) before and (c–d) after correction with the models of Table 3. To further investigate the variability of corrections in the testing data set across different true school properties and processing thresholds, Table 4 lists the tabulation of |δs̄v,c| and |δAc| by selected thresholds T and Ŝv,0 and A0 classes. The results show that the mean difference between the corrected and true Ŝv was 0.1–0.5 dB for most cases, while area mean absolute percentage errors typically averaged around 4–8% for most true size and density classes examined. Errors |δ| of the corrected descriptors were generally similar across increasing school sizes, with the exception of the smallest schools (A0 = 1000 m2) for which the corrections did not perform so adequately; |δs̄v,c| and |δAc| of these echotraces was 11.8 ± 8.0% and 10.6 ± 9.5%, respectively (Table 4, “All Ŝv,0” row). It is also worth noting that increasing the processing threshold for large schools of low true density degrades the corrections, especially those of volume backscatter, but improves them, especially the corrections of area, for denser schools of the same size class (see the A0 = 10 000 m2 column in Table 4 for the Ŝv,0 = –42 and –33 dB cases, processed with different thresholds). Table 4. Mean absolute percentage errors |δ| (mean ± SD) of the corrected mean volume backscatter (s̄v,c) and area (Ac) of simulated echotraces in the ΔŜT ≥ 2 dB testing data set, tabulated by true school area (A0), true mean volume backscattering strength (Ŝv,0), and processing threshold (T). A0 (m2) 1000 5000 10 000 1000 5000 10 000 Ŝv,0 (dB) T (dB) |δs̄v,c| % |δAc| % –42 –53 10.5 ± 4.0 (3) 6.3 ± 4.2 (11) 5.0 ± 7.5 (17) 7.0 ± 2.5 3.8 ± 1.9 2.4 ± 2.0 –51 — 9.6 ± 3.8 (9) 6.5 ± 4.2 (13) — 2.2 ± 1.6 4.2 ± 1.5 –49 — 8.1 ± 3.3 (6) 8.7 ± 5.9 (9) — 2.4 ± 1.4 4.5 ± 6.3 –47 — — 20.6 ± 1.5 (3) — — 8.0 ± 1.8 All Ta 13.4 ± 5.3 (5) 7.7 ± 4.1 (44) 7.9 ± 7.1 (75) 6.0 ± 2.3 3.0 ± 2.1 4.2 ± 4.2 –38 –53 8.7 ± 3.8 (8) 4.2 ± 4.2 (17) 3.9 ± 3.5 (17) 6.7 ± 4.7 4.7 ± 2.9 3.9 ± 3.0 –51 6.9 ± 6.1 (6) 7.8 ± 8.0 (16) 4.3 ± 4.4 (17) 8.4 ± 5.0 4.2 ± 2.7 3.6 ± 2.5 –49 9.1 ± 4.6 (4) 6.7 ± 6.7 (12) 8.1 ± 9.7 (17) 6.9 ± 2.5 4.6 ± 3.8 3.6 ± 2.6 –47 — 7.1 ± 2.6 (9) 8.6 ± 5.7 (14) — 3.9 ± 2.2 4.2 ± 3.1 All Ta 7.8 ± 4.6 (31) 6.2 ± 5.5 (94) 5.7 ± 6.1 (114) 7.5 ± 4.9 4.5 ± 2.9 3.6 ± 2.6 –33 –53 15.8 ± 8.5 (17) 4.6 ± 3.8 (17) 6.4 ± 4.6 (17) 31.4 ± 21.2 8.3 ± 4.8 15.0 ± 11.7 –51 11.9 ± 9.4 (13) 3.3 ± 6.6 (17) 6.6 ± 6.6 (17) 13.4 ± 7.8 7.9 ± 2.9 9.4 ± 8.2 –49 10.2 ± 7.1 (9) 4.2 ± 4.2 (17) 4.2 ± 4.3 (17) 4.7 ± 3.1 4.3 ± 2.7 3.2 ± 2.7 –47 6.2 ± 5.1 (8) 6.3 ± 3.1 (17) 3.1 ± 2.5 (17) 3.3 ± 3.3 4.6 ± 2.6 1.8 ± 1.7 All Ta 12.2 ± 8.4 (81) 4.3 ± 4.0 (119) 5.0 ± 4.9 (119) 14.9 ± 15.1 6.1 ± 3.9 7.1 ± 8.2 All Ŝv,0b 11.8 ± 8.0 (396) 5.1 ± 4.5 (918) 5.4 ± 5.4 (1072) 10.6 ± 9.5 4.9 ± 3.1 4.6 ± 4.8 Totalc 6.4 ± 6.0 (8601) 5.2 ± 4.8 A0 (m2) 1000 5000 10 000 1000 5000 10 000 Ŝv,0 (dB) T (dB) |δs̄v,c| % |δAc| % –42 –53 10.5 ± 4.0 (3) 6.3 ± 4.2 (11) 5.0 ± 7.5 (17) 7.0 ± 2.5 3.8 ± 1.9 2.4 ± 2.0 –51 — 9.6 ± 3.8 (9) 6.5 ± 4.2 (13) — 2.2 ± 1.6 4.2 ± 1.5 –49 — 8.1 ± 3.3 (6) 8.7 ± 5.9 (9) — 2.4 ± 1.4 4.5 ± 6.3 –47 — — 20.6 ± 1.5 (3) — — 8.0 ± 1.8 All Ta 13.4 ± 5.3 (5) 7.7 ± 4.1 (44) 7.9 ± 7.1 (75) 6.0 ± 2.3 3.0 ± 2.1 4.2 ± 4.2 –38 –53 8.7 ± 3.8 (8) 4.2 ± 4.2 (17) 3.9 ± 3.5 (17) 6.7 ± 4.7 4.7 ± 2.9 3.9 ± 3.0 –51 6.9 ± 6.1 (6) 7.8 ± 8.0 (16) 4.3 ± 4.4 (17) 8.4 ± 5.0 4.2 ± 2.7 3.6 ± 2.5 –49 9.1 ± 4.6 (4) 6.7 ± 6.7 (12) 8.1 ± 9.7 (17) 6.9 ± 2.5 4.6 ± 3.8 3.6 ± 2.6 –47 — 7.1 ± 2.6 (9) 8.6 ± 5.7 (14) — 3.9 ± 2.2 4.2 ± 3.1 All Ta 7.8 ± 4.6 (31) 6.2 ± 5.5 (94) 5.7 ± 6.1 (114) 7.5 ± 4.9 4.5 ± 2.9 3.6 ± 2.6 –33 –53 15.8 ± 8.5 (17) 4.6 ± 3.8 (17) 6.4 ± 4.6 (17) 31.4 ± 21.2 8.3 ± 4.8 15.0 ± 11.7 –51 11.9 ± 9.4 (13) 3.3 ± 6.6 (17) 6.6 ± 6.6 (17) 13.4 ± 7.8 7.9 ± 2.9 9.4 ± 8.2 –49 10.2 ± 7.1 (9) 4.2 ± 4.2 (17) 4.2 ± 4.3 (17) 4.7 ± 3.1 4.3 ± 2.7 3.2 ± 2.7 –47 6.2 ± 5.1 (8) 6.3 ± 3.1 (17) 3.1 ± 2.5 (17) 3.3 ± 3.3 4.6 ± 2.6 1.8 ± 1.7 All Ta 12.2 ± 8.4 (81) 4.3 ± 4.0 (119) 5.0 ± 4.9 (119) 14.9 ± 15.1 6.1 ± 3.9 7.1 ± 8.2 All Ŝv,0b 11.8 ± 8.0 (396) 5.1 ± 4.5 (918) 5.4 ± 5.4 (1072) 10.6 ± 9.5 4.9 ± 3.1 4.6 ± 4.8 Totalc 6.4 ± 6.0 (8601) 5.2 ± 4.8 All corrections derive from the models listed in Table 3; numbers in parentheses show the echotrace count (N) per category. Note that only selected thresholds T and Ŝv,0 and A0 classes are shown to reduce information density. a Including the T = [–52, –50, –48 dB] levels not shown in this table. b Including the Ŝv,0 = [–41, –40, –39, –37, –36, –35, –34 dB] classes not shown in this table. c Including the Ŝv,0 and A0 = [2000, 3000, 4000, 6000, 7000, 8000, 9000 m2] classes not shown in this table. Table 4. Mean absolute percentage errors |δ| (mean ± SD) of the corrected mean volume backscatter (s̄v,c) and area (Ac) of simulated echotraces in the ΔŜT ≥ 2 dB testing data set, tabulated by true school area (A0), true mean volume backscattering strength (Ŝv,0), and processing threshold (T). A0 (m2) 1000 5000 10 000 1000 5000 10 000 Ŝv,0 (dB) T (dB) |δs̄v,c| % |δAc| % –42 –53 10.5 ± 4.0 (3) 6.3 ± 4.2 (11) 5.0 ± 7.5 (17) 7.0 ± 2.5 3.8 ± 1.9 2.4 ± 2.0 –51 — 9.6 ± 3.8 (9) 6.5 ± 4.2 (13) — 2.2 ± 1.6 4.2 ± 1.5 –49 — 8.1 ± 3.3 (6) 8.7 ± 5.9 (9) — 2.4 ± 1.4 4.5 ± 6.3 –47 — — 20.6 ± 1.5 (3) — — 8.0 ± 1.8 All Ta 13.4 ± 5.3 (5) 7.7 ± 4.1 (44) 7.9 ± 7.1 (75) 6.0 ± 2.3 3.0 ± 2.1 4.2 ± 4.2 –38 –53 8.7 ± 3.8 (8) 4.2 ± 4.2 (17) 3.9 ± 3.5 (17) 6.7 ± 4.7 4.7 ± 2.9 3.9 ± 3.0 –51 6.9 ± 6.1 (6) 7.8 ± 8.0 (16) 4.3 ± 4.4 (17) 8.4 ± 5.0 4.2 ± 2.7 3.6 ± 2.5 –49 9.1 ± 4.6 (4) 6.7 ± 6.7 (12) 8.1 ± 9.7 (17) 6.9 ± 2.5 4.6 ± 3.8 3.6 ± 2.6 –47 — 7.1 ± 2.6 (9) 8.6 ± 5.7 (14) — 3.9 ± 2.2 4.2 ± 3.1 All Ta 7.8 ± 4.6 (31) 6.2 ± 5.5 (94) 5.7 ± 6.1 (114) 7.5 ± 4.9 4.5 ± 2.9 3.6 ± 2.6 –33 –53 15.8 ± 8.5 (17) 4.6 ± 3.8 (17) 6.4 ± 4.6 (17) 31.4 ± 21.2 8.3 ± 4.8 15.0 ± 11.7 –51 11.9 ± 9.4 (13) 3.3 ± 6.6 (17) 6.6 ± 6.6 (17) 13.4 ± 7.8 7.9 ± 2.9 9.4 ± 8.2 –49 10.2 ± 7.1 (9) 4.2 ± 4.2 (17) 4.2 ± 4.3 (17) 4.7 ± 3.1 4.3 ± 2.7 3.2 ± 2.7 –47 6.2 ± 5.1 (8) 6.3 ± 3.1 (17) 3.1 ± 2.5 (17) 3.3 ± 3.3 4.6 ± 2.6 1.8 ± 1.7 All Ta 12.2 ± 8.4 (81) 4.3 ± 4.0 (119) 5.0 ± 4.9 (119) 14.9 ± 15.1 6.1 ± 3.9 7.1 ± 8.2 All Ŝv,0b 11.8 ± 8.0 (396) 5.1 ± 4.5 (918) 5.4 ± 5.4 (1072) 10.6 ± 9.5 4.9 ± 3.1 4.6 ± 4.8 Totalc 6.4 ± 6.0 (8601) 5.2 ± 4.8 A0 (m2) 1000 5000 10 000 1000 5000 10 000 Ŝv,0 (dB) T (dB) |δs̄v,c| % |δAc| % –42 –53 10.5 ± 4.0 (3) 6.3 ± 4.2 (11) 5.0 ± 7.5 (17) 7.0 ± 2.5 3.8 ± 1.9 2.4 ± 2.0 –51 — 9.6 ± 3.8 (9) 6.5 ± 4.2 (13) — 2.2 ± 1.6 4.2 ± 1.5 –49 — 8.1 ± 3.3 (6) 8.7 ± 5.9 (9) — 2.4 ± 1.4 4.5 ± 6.3 –47 — — 20.6 ± 1.5 (3) — — 8.0 ± 1.8 All Ta 13.4 ± 5.3 (5) 7.7 ± 4.1 (44) 7.9 ± 7.1 (75) 6.0 ± 2.3 3.0 ± 2.1 4.2 ± 4.2 –38 –53 8.7 ± 3.8 (8) 4.2 ± 4.2 (17) 3.9 ± 3.5 (17) 6.7 ± 4.7 4.7 ± 2.9 3.9 ± 3.0 –51 6.9 ± 6.1 (6) 7.8 ± 8.0 (16) 4.3 ± 4.4 (17) 8.4 ± 5.0 4.2 ± 2.7 3.6 ± 2.5 –49 9.1 ± 4.6 (4) 6.7 ± 6.7 (12) 8.1 ± 9.7 (17) 6.9 ± 2.5 4.6 ± 3.8 3.6 ± 2.6 –47 — 7.1 ± 2.6 (9) 8.6 ± 5.7 (14) — 3.9 ± 2.2 4.2 ± 3.1 All Ta 7.8 ± 4.6 (31) 6.2 ± 5.5 (94) 5.7 ± 6.1 (114) 7.5 ± 4.9 4.5 ± 2.9 3.6 ± 2.6 –33 –53 15.8 ± 8.5 (17) 4.6 ± 3.8 (17) 6.4 ± 4.6 (17) 31.4 ± 21.2 8.3 ± 4.8 15.0 ± 11.7 –51 11.9 ± 9.4 (13) 3.3 ± 6.6 (17) 6.6 ± 6.6 (17) 13.4 ± 7.8 7.9 ± 2.9 9.4 ± 8.2 –49 10.2 ± 7.1 (9) 4.2 ± 4.2 (17) 4.2 ± 4.3 (17) 4.7 ± 3.1 4.3 ± 2.7 3.2 ± 2.7 –47 6.2 ± 5.1 (8) 6.3 ± 3.1 (17) 3.1 ± 2.5 (17) 3.3 ± 3.3 4.6 ± 2.6 1.8 ± 1.7 All Ta 12.2 ± 8.4 (81) 4.3 ± 4.0 (119) 5.0 ± 4.9 (119) 14.9 ± 15.1 6.1 ± 3.9 7.1 ± 8.2 All Ŝv,0b 11.8 ± 8.0 (396) 5.1 ± 4.5 (918) 5.4 ± 5.4 (1072) 10.6 ± 9.5 4.9 ± 3.1 4.6 ± 4.8 Totalc 6.4 ± 6.0 (8601) 5.2 ± 4.8 All corrections derive from the models listed in Table 3; numbers in parentheses show the echotrace count (N) per category. Note that only selected thresholds T and Ŝv,0 and A0 classes are shown to reduce information density. a Including the T = [–52, –50, –48 dB] levels not shown in this table. b Including the Ŝv,0 = [–41, –40, –39, –37, –36, –35, –34 dB] classes not shown in this table. c Including the Ŝv,0 and A0 = [2000, 3000, 4000, 6000, 7000, 8000, 9000 m2] classes not shown in this table. Application to real sonar data The descriptor correction approach investigated in this study was applied to real SP90 data recorded around a drifting fish aggregating device (FAD) in the western Indian Ocean (see Trygonis et al., 2016 and sources therein for details on data collection methods). The data originate from a recording session carried out with similar sonar settings as to those used in the simulated scenarios (horizontal omnidirectional mode at 26 kHz; observation range: 900 m; beam tilt: –3°; automatic gain control filter: off), and regard the consecutive observations of a single school of large pelagic (tuna) fish with compact echotraces that also fell within the simulated envelop (2–5-min gaps exist in the records due to vessel manoeuvres around the FAD). The scientific output of the uncalibrated sonar was analysed with custom software (Trygonis et al., 2009) that implements the same school detector used to process the simulated echograms. The cut-off threshold T during school detection was set to –51 dB, and the correction models listed in Table 3 were applied to real echotraces with ΔŜT ≥ 2 dB. The results are summarized in Figure 6, and show that, on an average, the corrections increased the estimated mean volume backscattering strength of real schools by 11.3 dB (mean Ŝv,obs = –45.5 dB, mean Ŝv,c = –34.2 dB; Figure 6e) and reduced their estimated mean area by 68.7% (mean Aobs = 3766.0 m2, mean Ac = 1180.4 m2; Figure 6f). When plotted against their distance from the transducer (Figure 7), the observed and corrected real data show similar patterns as those produced by the simulator (Supplementary Figure S2). Specifically, the sonar systematically under-samples the low density schools with increasing range (i.e. only the denser instances are detected at large distances from the transducer), and, simultaneously with this sampling bias, the observed Ŝv of a detected school is an underestimate of its respective true value. Moreover, assuming that the true school area is independent from range, the results suggest that the corrections compensate for the range-dependent overestimation of school area (Figure 7b). Figure 6. View largeDownload slide (a–c) Comparison of observed descriptors extracted from real and simulated echotraces; processing threshold T is –51 dB for both data categories. Open markers denote observations with ΔŜT < 2 dB. Simulated data include both training and testing observations and are only shown at selected distances from the transducer (140, 200, 260, …, 800 m) to reduce clutter. (d) Example real echotraces of the same school observed at different ranges (T = –51 dB). The echotraces are shown at their actual range, but for illustration purposes, their angular (across-beam) position has been rotated to align with the +x-axis. (e) Distribution of mean volume backscattering strength of real echotraces (ΔŜT ≥ 2 dB, N = 213 detections), before (Ŝv,obs) and after correction (Ŝv,c); only real data plotted with closed markers in panels a–c were included in the corrections, using the models of Table 3. (f) As in panel e, but for echotrace area. Figure 6. View largeDownload slide (a–c) Comparison of observed descriptors extracted from real and simulated echotraces; processing threshold T is –51 dB for both data categories. Open markers denote observations with ΔŜT < 2 dB. Simulated data include both training and testing observations and are only shown at selected distances from the transducer (140, 200, 260, …, 800 m) to reduce clutter. (d) Example real echotraces of the same school observed at different ranges (T = –51 dB). The echotraces are shown at their actual range, but for illustration purposes, their angular (across-beam) position has been rotated to align with the +x-axis. (e) Distribution of mean volume backscattering strength of real echotraces (ΔŜT ≥ 2 dB, N = 213 detections), before (Ŝv,obs) and after correction (Ŝv,c); only real data plotted with closed markers in panels a–c were included in the corrections, using the models of Table 3. (f) As in panel e, but for echotrace area. Figure 7. View largeDownload slide Scatter plots of observed and corrected (a) mean volume backscattering strength and (b) area of real echotraces with ΔŜT ≥ 2 dB vs. their distance from the transducer (see also Supplementary Figure S2 for corresponding plots based on simulated data). Figure 7. View largeDownload slide Scatter plots of observed and corrected (a) mean volume backscattering strength and (b) area of real echotraces with ΔŜT ≥ 2 dB vs. their distance from the transducer (see also Supplementary Figure S2 for corresponding plots based on simulated data). Discussion The acoustic image observed on the echogram depends both on transducer and fish school characteristics, as well as on the processing threshold used to analyse the data. The acoustic beam is directive, thereby schools with a higher volume backscattering strength are sampled with an effectively wider beam angle. The same school can produce different echotraces when observed at different ranges due to beam spreading, and at any given range, the echotrace only represents an approximation of the true school dimensions, distorted to the discrete resolution of the echogram (Figure 2a and e). These issues are intrinsic to the acoustic measurement (Reid, 2000; Reid et al., 2000; Diner, 2001) and must be accounted for in order to reliably infer the true shape, size, and density of schools from their observed echotraces; see, for example, Johannesson and Losse (1977), Kieser et al. (1993), Reid and Simmonds (1993), Reid et al. (2000), and Diner (2001, 2007) for related discussions or correction approaches with emphasis on vertical beaming. Focusing on horizontal multi-beam observations, Misund (1990) proposed a formula for correcting the across-beam width of the school (Cw) based on the nominal beam width of the sonar. Vatnehol et al. (2017) used simulated multi-beam data and modified Misund’s (1990) geometric model to also account for long-range distortions of Cw; a respective correction formula for school height was also produced by simulating the vertical insonification mode of the sonar. Applied to multi-beam horizontal area measurements (AR) of the same school observed at different ranges (R), Misund et al. (1995) used the linear regression of √AR against R, and extrapolated it to the origin of the sonar beams to estimate the true value. This area correction approach makes no assumptions about the nominal beam width, but cannot handle a number of cases that often occur in horizontal multi-beam sampling. For example, the regression yields incorrect results if the proportion of the school that is being insonified changes with range, and is not applicable if the school area measured near the transducer is larger than the area measured at an outer position. The simulation exercise investigated herein is an attempt to build sonar device-specific correction models that use the descriptors extracted from the observed echotrace as the only predictors of the true morphometric and energetic values of the insonified school. This implicitly assumes that, although the multi-beam sonar induces substantial intrinsic distortions to echotraces, the latter contain sufficient information to statistically infer the true properties of the school that is being sampled. The simulated scenarios included a broad range of school sizes and mean densities (1000 to 10000 m2 and –42 to –33 dB, respectively) to increase variability in the source data, which, when combined with a tilted omnidirectional fan and constant school depth, ensured that partial (off-axis) insonification of schools was also included in the simulation. Overall, the corrections significantly reduced the errors in the measured area and mean volume backscattering strength of echotraces, over the entire range of processing thresholds examined. The applicability limits of the method were empirically outlined via ΔŜT, which is also a quantity that can be computed from the observed echotrace, and is similar to the corresponding metric used in Diner’s (2001) simulations. While the correction performance (see candidate models of Table 2) was superior when only echotraces with high signal-to-noise ratio were processed (meaning that their observed mean volume backscattering strength was at least 5–6 dB above the processing threshold), the analysis showed that a lower ΔŜT criterion was sufficient for consistent results. Specifically, when echotraces with ΔŜT < 2 dB were filtered out, the corrections produced mean absolute percentage errors of 5.2% and 6.4% for Ac and s̄v,c, respectively. These area correction errors are very similar to those reported by Vatnehol et al. (2017) for the closely related (see Figure 2c and g) Cw descriptor measured with a Simrad SX90 sonar (Simrad, 2015a), although the two studies used different simulator and correction model designs to account for the systematic measurement errors. Corrections could not be applied to echotraces representing fish schools that were located near the transducer, and were, as a result, severely under-sampled. Directing a tilted omnidirectional fan at fish schools that occupy a depth layer is a characteristic that underpins the sampling with horizontal sonar (Misund and Coetzee, 2000), but it inherently creates an asymmetrical sampling with range, unless a single school is selected and tracked by actively manipulating the tilt angle (Misund et al., 1998). With respect to our simulations, this would correspond to a design that translated the schools away from the transducer while keeping them always on (or near) the acoustic axis. This sampling mode was partially represented in the more general simulation scenarios examined herein, where the centre of schools intersected the acoustic axis at a range of about 250–300 m from the transducer. While specific simulations are needed to conclude with confidence, the overall good performance of the corrections across a challenging mixture of on- and off-axis school insonifications suggests that the method investigated herein would also be applicable to sampling modes that actively modify the beam tilt angle to keep the tracked school on-axis. In its current design, the simulator accounted for some of the systematic measurement errors involved in horizontal sonar sampling, i.e. those related to the beam pattern effect, the overlap of neighbouring beams, the polar geometry of the multi-beam echogram, and the spatial configuration of the tilted omnidirectional fan in relation to the location of fish schools. While the overall behaviour of the simulator was convincing and the systematic issues were reproduced well, the study did not consider environmental sources of variability or fish school behavioural aspects. For example, the simplified school model ignored fish polarization and its substantial effects on the recorded Sv (Boswell et al., 2009; Holmin et al., 2012), while the allocation of voxel backscatter had no spatial correlation within the school, thus dense regions or vacuoles that emerge in schooling fish (Fréon et al., 1992; Gerlotto et al., 2006; Guillard et al., 2011) were not represented in the simulations. Furthermore, this internal structure remained static across all insonification positions, and the general shape of the school was fixed to an idealized oblate spheroid. Fish schools are in fact dynamic structures (Misund, 1993) that can present substantial temporal fluctuations in their cross-sectional area (Trygonis et al., 2016), while their vertical or horizontal dynamics introduce additional uncertainties due to the variable position of the schools’ gravity centre relative to the acoustic axis (Brehmer et al., 2006). In addition, the actual in situ sampling schemes used to collect the sonar data may also impact the observed fish school descriptors. For example, Brehmer et al. (2002, 2006) proposed that school area can be approximated through the along-beam width (Lw) of the school, by assuming a spherical school shape that has a diameter equal to Lw. Yet, a different variant of Lw was appropriate (i.e. the mean or maximum Lw) depending on whether data were recorded from a free-drifting vessel or from one that moved along a transect line or adapted its course to actively follow a school. Given the enormous complexity of modelling all dynamic and behavioural aspects in a single simulation, several simplifications were made to both transducer and school model properties. The systematic sources of bias, however, that were integrated into the analysis were not trivial, and by themselves, contributed to mean absolute percentage errors in the order of 100% and 80% in the observed area and mean volume backscatter, respectively, as per the simulated ΔŜT ≥ 2 dB testing data set. Of course, simplifications increase discrepancies between the simulated and real system, and indeed, Figure 6a shows that, for the same observed mean volume backscattering strength, the echotraces currently generated by the simulator have smaller variance than the real ones. However, overall, the side-by-side comparison of simulated vs. real echotrace descriptors revealed no anomalous behaviour in the former, increasing confidence in the validity of the simulation engine. The experimental application of corrections to real SP90 data also yielded a pattern in the corrected descriptors that was concordant to the one expected by the simulations, but there was no means to ground truth the true properties of real fish schools. Therefore, it is not possible to assess the errors in the corrected real data. Using a calibrated sonar (Foote et al., 2005; Demer et al., 2015; Macaulay et al., 2016) for the comparisons and implementing its actual directivity in the simulator is necessary to obtain more precise results. Moreover, incorporating dynamic aspects in the internal structure and motion of schools, and expanding the simulation scenarios to cover multiple sonar settings and school sizes will increase the fidelity of the simulated system and the applicability of generated correction models to a broader range of real data. This is of particular importance, given that the correction models established herein are based on observed descriptors, which in turn depend on the properties of the simulated schools and transducer. In addition, the models significantly reduce the observed errors through a linear combination of the distorted echotrace descriptors, but provide no insight on the relative importance of the sources of bias that actually caused that distortion. Therefore, and as in all simulation attempts, the resulting correction models are applicable only to data from a similar sonar and fish school size/density properties as those used in the simulations that produced them. New fishery sonars like the Simrad SX90 or SU90 (Simrad, 2015b) have identical frequency range and similar spatial configuration of beams with the SP90 (i.e. cylindrical transducer and 64 beams arranged in an omnidirectional fan), but narrower beam widths. The methodological framework of the correction models suggests that a similar approach can be applied to these sonars, and future work should expand the simulations on these devices. Notwithstanding complexities pertaining to the target strength of schooling fish in the lateral aspect (Cutter and Demer, 2007; Tang et al., 2009), this will help establish more general solutions to the multi-parametric problem of debiasing fish school descriptors, in order to convert the corrected mean volume backscattering strength of schools to absolute fish density estimates. Supplementary data Supplementary material is available at the ICESJMS online version of the article. Acknowledgements We thank Stratis Georgakarakos for his advice during development of the simulator. Two anonymous reviewers and Patrice Brehmer provided very helpful comments on this manuscript. References Aglen A. 1983 . Echo integrator threshold and fish density distribution . FAO Fisheries Reports , 300 : 35 – 44 . Bahri T. , Fréon P. 2000 . Spatial structure of coastal pelagic schools descriptors in the Mediterranean Sea . Fisheries Research , 48 : 157 – 166 . Google Scholar CrossRef Search ADS Boswell K. M. , Roth B. M. , Cowan J. H. Jr 2009 . Simulating the effects of side-aspect fish orientation on acoustic biomass estimates . ICES Journal of Marine Science , 66 : 1398 – 1403 . Google Scholar CrossRef Search ADS Brehmer P. , Georgakarakos S. , Josse E. , Trygonis V. , Dalen J. 2007 . Adaptation of fisheries sonar for monitoring schools of large pelagic fish: dependence of schooling behaviour on fish finding efficiency . Aquatic Living Resources , 20 : 377 – 384 . Google Scholar CrossRef Search ADS Brehmer P. , Gerlotto F. , Rouault A. 2002 . Inter-standardisation of acoustic devices: an integrated fish school database for behavioural studies . Acta Acoustica , 88 : 730 – 734 . Brehmer P. , Josse E. , Nøttestad L. 2012 . Evidence that whales (Balaenoptera borealis) visit drifting fish aggregating devices: do their presence affect the processes underlying fish aggregation? Marine Ecology , 33 : 176 – 182 . Google Scholar CrossRef Search ADS Brehmer P. , Lafont T. , Georgakarakos S. , Josse E. , Gerlotto F. , Collet C. 2006 . Omnidirectional multibeam sonar monitoring: applications in fisheries science . Fish and Fisheries , 7 : 165 – 179 . Google Scholar CrossRef Search ADS Castillo J. , Robotham H. 2004 . Spatial structure and geometry of schools of sardine (Sardinops sagax) in relation to abundance, fishing effort, and catch in northern Chile . ICES Journal of Marine Science , 61 : 1113 – 1119 . Google Scholar CrossRef Search ADS Coetzee J. 2000 . Use of a shoal analysis and patch estimation system (SHAPES) to characterise sardine schools . Aquatic Living Resources , 13 : 1 – 10 . Google Scholar CrossRef Search ADS Cutter G. R. , Demer D. A. 2007 . Accounting for scattering directivity and fish behaviour in multibeam-echosounder surveys . ICES Journal of Marine Science , 64 : 1664 – 1674 . Google Scholar CrossRef Search ADS Demer D. A. , Berger L. , Bernasconi M. , Bethke E. , Boswell K. , Chu D. , Domokos R. , et al. 2015 . Calibration of acoustic instruments . ICES Cooperative Research , Report No. 326. 133 pp. Diner N. 2001 . Correction on school geometry and density: approach based on acoustic image simulation . Aquatic Living Resources , 14 : 211 – 222 . Google Scholar CrossRef Search ADS Diner N. 2007 . Evaluating uncertainty in measurements of fish shoal aggregate backscattering cross-section caused by small shoal size relative to beam width . Aquatic Living Resources , 20 : 117 – 121 . Google Scholar CrossRef Search ADS Fernandes P. G. 2009 . Classification trees for species identification of fish-school echotraces . ICES Journal of Marine Science , 66 : 1073 – 1080 . Google Scholar CrossRef Search ADS Foote K. G. , Chu D. , Hammar T. R. , Baldwin K. C. , Mayer L. A. , Hufnagle L. C. , Jech J. M. 2005 . Protocols for calibrating multibeam sonar . Journal of the Acoustical Society of America , 117 : 2013 – 2027 . Google Scholar CrossRef Search ADS PubMed Fréon P. , Gerlotto F. , Soria M. 1992 . Changes in school structure according to external stimuli: description and influence on acoustic assessment . Fisheries Research , 15 : 45 – 66 . Google Scholar CrossRef Search ADS Fréon P. , Gerlotto F. , Soria M. 1996 . Diel variability of school structure with special reference to transition periods . ICES Journal of Marine Science , 53 : 459 – 464 . Google Scholar CrossRef Search ADS Fritsch F. N. , Carlson R. E. 1980 . Monotone piecewise cubic interpolation . SIAM Journal on Numerical Analysis , 17 : 238 – 246 . Google Scholar CrossRef Search ADS Gerlotto F. , Bertrand S. , Bez N. , Gutierrez M. 2006 . Waves of agitation inside anchovy schools observed with multibeam sonar: a way to transmit information in response to predation . ICES Journal of Marine Science , 63 : 1405 – 1417 . Google Scholar CrossRef Search ADS Gerlotto F. , Paramo J. 2003 . The three-dimensional morphology and internal structure of clupeid schools as observed using vertical scanning multibeam sonar . Aquatic Living Resources , 16 : 113 – 122 . Google Scholar CrossRef Search ADS Guillard J. , Fernandes P. , Laloë T. , Brehmer P. 2011 . Three-dimensional internal spatial structure of young-of-the-year pelagic freshwater fish provides evidence for the identification of fish school species . Limnology and Oceanography: Methods , 9 : 322 – 328 . Google Scholar CrossRef Search ADS Hafsteinsson M. T. , Misund O. A. 1995 . Recording the migration behaviour of fish schools by multi-beam sonar during conventional acoustic surveys . ICES Journal of Marine Science , 52 : 915 – 924 . Google Scholar CrossRef Search ADS Haralabous J. , Georgakarakos S. 1996 . Artificial neural networks as a tool for species identification of fish schools . ICES Journal of Marine Science , 53 : 173 – 180 . Google Scholar CrossRef Search ADS Holmin A. J. , Handegard N. O. , Korneliussen R. J. , Tjøstheim D. 2012 . Simulations of multi-beam sonar echos from schooling individual fish in a quiet environment . Journal of the Acoustical Society of America , 132 : 3720 – 3734 . Google Scholar CrossRef Search ADS PubMed Johannesson K. A. , Losse G. F. 1977 . Methodology of acoustic estimations of fish abundance in some UNDP/FAO resource survey projects . Rapports Et Procès-Verbaux Des Réunions Du Conseil International Pour L’Exploration De La Mer , 170 : 296 – 318 . Kieser R. , Mulligan T. J. , Richards L. J. , Leaman B. M. 1993 . Bias correction of rockfish school cross section widths from digitized echo sounder data . Canadian Journal of Fisheries and Aquatic Sciences , 50 : 1801 – 1811 . Google Scholar CrossRef Search ADS Kvamme C. , Nøttestad L. , Fernö A. , Misund O. A. , Dommasnes A. , Axelsen B. E. , Dalpadado P. , et al. 2003 . Migration patterns in Norwegian spring-spawning herring: why young fish swim away from the wintering area in late summer . Marine Ecology Progress Series , 247 : 197 – 210 . Google Scholar CrossRef Search ADS Lawson G. L. , Barange M. , Fréon P. 2001 . Species identification of pelagic fish schools on the South African continental shelf using acoustic descriptors and ancillary information . ICES Journal of Marine Science , 58 : 275 – 287 . Google Scholar CrossRef Search ADS Macaulay G. J. , Vatnehol S. , Gammelsæter O. B. , Peña H. , Ona E. 2016 . Practical calibration of ship-mounted omni-directional fisheries sonars . Methods in Oceanography , 17 : 206 – 220 . Google Scholar CrossRef Search ADS MacLennan D. N. , Fernandes P. G. , Dalen J. 2002 . A consistent approach to definitions and symbols in fisheries acoustics . ICES Journal of Marine Science , 59 : 365 – 369 . Google Scholar CrossRef Search ADS McClatchie S. , Thorne R. E. , Grimes P. , Hanchet S. 2000 . Ground truth and target identification for fisheries acoustics . Fisheries Research , 47 : 173 – 191 . Google Scholar CrossRef Search ADS Misund O. A. 1990 . Sonar observations of schooling herring: school dimensions, swimming behaviour, and avoidance of vessel and purse seine . Rapports Et Procès-Verbaux Des Réunions Du Conseil International Pour L’Exploration De La Mer , 189 : 135 – 146 . Misund O. A. 1993 . Dynamics of moving masses: variability in packing density, shape, and size among herring, sprat, and saithe schools . ICES Journal of Marine Science , 50 : 145 – 160 . Google Scholar CrossRef Search ADS Misund O. A. 1997 . Underwater acoustics in marine fisheries and fisheries research . Reviews in Fish Biology and Fisheries , 7 : 1 – 34 . Google Scholar CrossRef Search ADS Misund O. A. , Aglen A. , Frønæs E. 1995 . Mapping the shape, size, and density of fish schools by echo integration and a high-resolution sonar . ICES Journal of Marine Science , 52 : 11 – 20 . Google Scholar CrossRef Search ADS Misund O. A. , Aglen A. , Hamre J. , Ona E. , Røttingen I. , Skagen D. , Valdemarsen J. W. 1996 . Improved mapping of schooling fish near the surface: comparison of abundance estimates obtained by sonar and echo integration . ICES Journal of Marine Science , 53 : 383 – 388 . Google Scholar CrossRef Search ADS Misund O. A. , Coetzee J. 2000 . Recording fish schools by multi-beam sonar: potential for validating and supplementing echo integration recordings of schooling fish . Fisheries Research , 47 : 149 – 159 . Google Scholar CrossRef Search ADS Misund O. A. , Fernö A. , Pitcher T. , Totland B. 1998 . Tracking herring schools with a high resolution sonar. Variations in horizontal area and relative echo intensity . ICES Journal of Marine Science , 55 : 58 – 66 . Google Scholar CrossRef Search ADS Olsen S. 1969 . A note on estimating school size from echo-traces . FAO Fisheries Reports , 78 : 37 – 48 . Peña H. , Handegard N. O. , Ona E. 2013 . Feeding herring schools do not react to seismic airgun surveys . ICES Journal of Marine Science , 70 : 1174 – 1180 . Google Scholar CrossRef Search ADS Peraltilla S. , Bertrand S. 2014 . In situ measurements of the speed of Peruvian anchovy schools . Fisheries Research , 149 : 92 – 94 . Google Scholar CrossRef Search ADS Petitgas P. 2003 . A method for the identification and characterization of clusters of schools along the transect lines of fisheries-acoustic surveys . ICES Journal of Marine Science , 60 : 872 – 884 . Google Scholar CrossRef Search ADS Petitgas P. , Massé J. , Beillois P. , Lebarbier E. , Le Cann A. 2003 . Sampling variance of species identification in fisheries-acoustic surveys based on automated procedures associating acoustic images and trawl hauls . ICES Journal of Marine Science , 60 : 437 – 445 . Google Scholar CrossRef Search ADS Petitgas P. , Reid D. G. , Carrera P. , Iglesias M. , Georgakarakos S. , Liorzou B. , Massé J. 2001 . On the relation between schools, clusters of schools, and abundance in pelagic fish stocks . ICES Journal of Marine Science , 58 : 1150 – 1160 . Google Scholar CrossRef Search ADS Reid D. G. 2000 . Report on echo trace classification . ICES Cooperative Research , Report No. 238. 107 pp. Reid D. G. , Scalabrin C. , Petitgas P. , Massé J. , Aukland R. , Carrera P. , Georgakarakos S. 2000 . Standard protocols for the analysis of school based data from echo sounder surveys . Fisheries Research , 47 : 125 – 136 . Google Scholar CrossRef Search ADS Reid D. G. , Simmonds E. J. 1993 . Image analysis techniques for the study of fish school structure from acoustic survey data . Canadian Journal of Fisheries and Aquatic Sciences , 50 : 886 – 893 . Google Scholar CrossRef Search ADS Scalabrin C. , Massé J. 1993 . Acoustical detection of the spatial and temporal distribution of fish shoals in the Bay of Biscay . Aquatic Living Resources , 6 : 269 – 283 . Google Scholar CrossRef Search ADS Simmonds J. , MacLennan D. 2005 . Fisheries Acoustics , 2nd edn. Chapman & Hall , London . Google Scholar CrossRef Search ADS Simrad . 2007 . Simrad SP90 low frequency long range fishery sonar; Operator manual. 850-164511/Rev.F, Horten, Norway. Simrad . 2015a . Simrad SX90; Operator manual. 307672/Rev.G, Horten, Norway. Simrad. 2015b . Simrad SU90; Operator manual. 381291/Rev.B, Horten, Norway. Soria M. , Bahri T. , Gerlotto F. 2003 . Effect of external factors (environment and survey vessel) on fish school characteristics observed by echosounder and multibeam sonar in the Mediterranean Sea . Aquatic Living Resources , 16 : 145 – 157 . Google Scholar CrossRef Search ADS Stockwell J. D. , Weber T. C. , Baukus A. J. , Jech J. M. 2013 . On the use of omnidirectional sonars and downwards-looking echosounders to assess pelagic fish distributions during and after midwater trawling . ICES Journal of Marine Science , 70 : 196 – 203 . Google Scholar CrossRef Search ADS Tang Y. , Nishimori Y. , Furusawa M. 2009 . The average three-dimensional target strength of fish by spheroid model for sonar surveys . ICES Journal of Marine Science , 66 : 1176 – 1183 . Google Scholar CrossRef Search ADS Trenkel V. M. , Berger L. , Bourguignon S. , Doray M. , Fablet R. , Massé J. , Mazauric V. , et al. 2009 . Overview of recent progress in fisheries acoustics made by Ifremer with examples from the Bay of Biscay . Aquatic Living Resources , 22 : 433 – 445 . Google Scholar CrossRef Search ADS Trygonis V. , Georgakarakos S. , Dagorn L. , Brehmer P. 2016 . Spatiotemporal distribution of fish schools around drifting fish aggregating devices . Fisheries Research , 177 : 39 – 49 . Google Scholar CrossRef Search ADS Trygonis V. , Georgakarakos S. , Simmonds E. J. 2009 . An operational system for automatic school identification on multibeam sonar echoes . ICES Journal of Marine Science , 66 : 935 – 949 . Google Scholar CrossRef Search ADS Tsagarakis K. , Giannoulaki M. , Somarakis S. , Machias A. 2012 . Variability in positional, energetic and morphometric descriptors of European anchovy Engraulis encrasicolus schools related to patterns of diurnal vertical migration . Marine Ecology Progress Series , 446 : 243 – 258 . Google Scholar CrossRef Search ADS Uranga J. , Arrizabalaga H. , Boyra G. , Hernandez M. C. , Goñi N. , Arregui I. , Fernandes J. A. , et al. 2017 . Detecting the presence-absence of blue fin tuna by automated analysis of medium-range sonars on fishing vessels . PLoS One , 12 : e0171382 . Google Scholar CrossRef Search ADS PubMed Vatnehol S. , Peña H. , Ona E. 2017 . Estimating the volumes of fish schools from observations with multi-beam sonars . ICES Journal of Marine Science , 74 : 813 – 821 . Weill A. , Scalabrin C. , Diner N. 1993 . MOVIES-B: an acoustic detection description software. Application to shoal species' classification . Aquatic Living Resources , 6 : 255 – 267 . Google Scholar CrossRef Search ADS Zwolinski J. , Morais A. , Marques V. , Stratoudakis Y. , Fernandes P. G. 2007 . Diel variation in the vertical distribution and schooling behaviour of sardine (Sardina pilchardus) off Portugal . ICES Journal of Marine Science , 64 : 963 – 972 . Google Scholar CrossRef Search ADS © International Council for the Exploration of the Sea 2018. All rights reserved. For Permissions, please email: [email protected] This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

doi: 10.1093/icesjms/fsx246pmid: N/A

Abstract Due to strong spatial interplays between intraspecific interactions and environmental forcing, both density-dependent and density-independent processes can affect spatio-temporal dynamics of fish populations in a spatially explicit fashion. To this end, this study investigated the underlying mechanisms of spatio-temporal dynamics of Atlantic cod (Gadus morhua) in the Gulf of Maine (GoM). Based on the data from the Northeast Fisheries Science Center (NEFSC) bottom-trawl surveys in spring and fall from 1982 to 2013, empirical cumulative distribution function (ECDF) curves and geographic distribution indices were used to examine the species–environment and abundance–occupancy relationship, respectively. Then, a variable-coefficient generalized additive model was constructed to quantify the simultaneous effects of environmental variables and population size on the spatio-temporal dynamics of cod distribution. Area occupied remained relatively high through the late 1990s, but underwent a pronounced contraction into the western GoM (WGoM) for the reminder of the time-series. The model results suggest that the spatio-temporal dynamics of GoM cod have been driven by complex interactions of density-dependent and density-independent factors over the past three decades. Better knowledge of these dynamics can improve our understanding of the causality of abundance–occupancy and species–environment relationships and help to reduce error estimates for survey-based indices. Introduction Understanding spatio-temporal dynamics of fish distributions is crucial for their conservation and management (Anderson and Gregory, 2000; Bartolino et al., 2017), especially for rebuilding-depleted populations (Tamdrari et al., 2010). However, studies of population dynamics have typically focused on temporal variability, leading to a limited understanding of how spatial variability affects the dynamics of species abundance (Ciannelli et al., 2007, 2008). Recent studies have shown that spatial dynamics of species are the result of complex interactions between density-independent (i.e. environment) and density-dependent (i.e. demography) sources of variability (Bartolino et al., 2017). The spatial variability of fish populations could be strongly influenced by density-independent factors, including temperature (Spencer, 2008; Ciannelli et al., 2012), depth (Gröger et al., 2007), or oxygen concentration (Craig et al., 2005; Youcef et al., 2013). However, the common approach has been to focus solely on either density-independent or density-dependent mechanisms underlying changes in fish distributions, which may lead to an incomplete interpretation of spatial dynamics of populations (Spencer, 2008). Atlantic cod (Gadus morhua) is an economically, ecologically, and culturally important species once prevalent in many of the world’s marine ecosystems (Drinkwater, 2005; Link et al., 2009). Recently, the Gulf of Maine (GoM) cod stock reached historically low levels, largely due to overfishing and rapid warming (Mills et al., 2013; Pershing et al., 2015). In the GoM, sea surface temperatures have increased by an average of 0.026 °C per year since 1982, with a tenfold acceleration in rate after 2004 (Mills et al., 2013). For the GoM cod stock, four subpopulations can be identified (Western, Midcoast, Eastern, and Bay of Fundy) based on the historical spawning components, with the last three subpopulations having been depleted or collapsed by the late 1940s (Ames, 2004; Ames and Lichter, 2013). Concurrently, the species have become increasingly concentrated in the western coastal area in the GoM (Churchill et al., 2011), reflecting a density-dependent habitat selection (Swain and Wade, 1993; Chang et al., 2010). The rapid changes in environmental conditions and the intraspecific abundance–occupancy (or density-distribution) relationships (Gaston et al., 2000; Päivinen et al., 2005) can typically be interpreted as evidence for the potential of both density-independent and density-dependent processes to affect distribution. Although studies have found temporal associations between density-independent factors and population dynamics of Atlantic cod in the GoM (Pershing et al., 2015; Guan et al., 2017), no analytical investigation has been performed on the ways in which both density-dependent and density-independent factors influence the long-term changes in the spatial distribution of Atlantic cod. Spatially explicit (or variant) species distribution models prove to be useful tools for studying spatial heterogeneity in population abundance (Dormann et al., 2007), especially in species with highly variable dynamics. For example, variable coefficient generalized additive models (vc-GAM) have been successfully applied to describe the spatially explicit forces of density-dependent and density-independent processes on spatial and temporal dynamics of fish (Bacheler et al., 2009; Bartolino et al., 2011; Ciannelli et al., 2012), cephalopods (Puerta et al., 2015), and plankton (Llope et al., 2009, 2012). The utilization of these models, especially for exploited and depleted fish populations, could improve our understanding of the effects of habitat loss and climate change, and advance the management of fisheries resources (Bacheler et al., 2009; Bartolino et al., 2017). Atlantic cod in the GoM provide a unique opportunity to study the influence of density-dependent and density-independent factors on spatial distributions as a result of recent changes in population size and the linkages between environmental variability and population dynamics (Ames, 2004; NEFSC, 2015). The purpose of this study was to examine changes in cod distribution as a function of density-independent (environmental variables) and density-dependent factors (population size) in the GoM. Significant associations of cod density occurring with environmental variables (e.g. temperature and depth) in spring and fall since 1982 were examined. The geographic distribution and relationships between population size of cod in spring and fall from 1982 to 2013 were then characterized. A spatially explicit statistical model was developed (Ciannelli et al., 2008) to detect spatially variable density-dependent and density-independent relationships for significant changes in both cod population size and environmental conditions in the GoM. Material and methods The GoM Coastal Current (GMCC) has two principal branches (Townsend et al., 2001; Pettigrew et al., 2005): the Eastern Maine Coastal Current (EMCC) extends along the eastern coast of Maine to mid-coast Maine at Jeffreys Bank, and the Western Maine Coastal Current (WMCC) extends westward from mid-coast Maine to Stellwagen Bank. Both of these principal branches are often centred near the 100 m isobath around basins and increase their transport in spring and summer due to freshwater river runoff (Townsend et al., 2001; Pettigrew et al., 2005; Balch et al., 2008). Considering the oceanographic characteristics described earlier, we defined the WGoM (see Figure 1a) according to the fishing statistical area in the Greater Atlantic Region (Halliday and Pinhorn, 1990). The slope water can flow in the GoM through the Northeast Channel and into the three main basins (Georges, Jordan, and Wilkinson basins, see Figure 1b), which are the major source of dissolved inorganic nutrients (Townsend et al., 2015). Figure 1. View largeDownload slide Study area (solid black line). (a) The NEFSC bottom survey strata (thin black line) and WGoM (shaded area); Hague Line separating the United States and Canada is shown. (b) Bottom topography of the GoM. Figure 1. View largeDownload slide Study area (solid black line). (a) The NEFSC bottom survey strata (thin black line) and WGoM (shaded area); Hague Line separating the United States and Canada is shown. (b) Bottom topography of the GoM. Data sources The Northeast Fisheries Science Center (NEFSC) bottom-trawl surveys were carried out on the Northeast United States continental shelf in spring and fall (Overholtz and Friedland, 2002). The survey employs a stratified random design, with stations allocated proportionally to depth-region stratum area (Nye et al., 2009). We used data from spring and fall (1982–2013) bottom-trawl surveys in the GoM (Strata 01260–01300, 01340–01400, 03560–03660, see Figure 1a) with depths from 12 to 371 m (Table 1). Spatio-temporal information (i.e. beginning and end of geographic coordinates, date/time) and the number of cod caught were collected for all of the 4601 tows, of which 15 had no depth recorded and 673 had no sea bottom temperature recorded. Records of bottom salinity began in the fall of 1996 and only 2606 tows had sea bottom salinity records (Table 1). Successful tows were for 30 min at a speed of 3.5 knots before 2009 and 20 min at a speed of 3.0 knots since 2009, with area swept changing from 0.038 to 0.024 km2. The survey density data was standardized by calibration factors for changes in vessel (“Delaware II”, “Albatross IV”, and “Henry B Bigelow”), door configurations (Bergens Mekaniske Versteder door, polyvalent, and polyice oval) and survey to maintain a consistent time-series of catch rates (NEFSC, 2013). The standardized survey data are continuous and non-negative, characterized by left-skewed distributions (Supplementary Figure S1), with a high proportion of zero catches (48.84% in spring and 55.48% in fall) and a few observations with high catch rates (Figure 2). Table 1. The main dataset attributes and the number of missing values (NA) Attribute Description Spring Fall Domain NA Domain NA Year Year {1982: 2013} {1982: 2013} Longitude Beginning longitude of each tow (decimal degree) [–70.83, –65.67] [–70.84, –65.68] Latitude Beginning latitude of each tow (decimal degree) [41.68, 44.43] [41.71, 44.43] Temperature Bottom water temperatures (°C) [2.1, 11.3] 332 [4.6, 15.7] 341 Salinity Bottom salinity (mainly in 1982–1996) [30.51, 35.31] 1 058 [31.19, 35.27] 1 037 Depth Maximum depth of survey site (m) [19, 371] 9 [12, 368] 6 Density Survey catch density (no. tow−1) [0, 913] [0, 487] Attribute Description Spring Fall Domain NA Domain NA Year Year {1982: 2013} {1982: 2013} Longitude Beginning longitude of each tow (decimal degree) [–70.83, –65.67] [–70.84, –65.68] Latitude Beginning latitude of each tow (decimal degree) [41.68, 44.43] [41.71, 44.43] Temperature Bottom water temperatures (°C) [2.1, 11.3] 332 [4.6, 15.7] 341 Salinity Bottom salinity (mainly in 1982–1996) [30.51, 35.31] 1 058 [31.19, 35.27] 1 037 Depth Maximum depth of survey site (m) [19, 371] 9 [12, 368] 6 Density Survey catch density (no. tow−1) [0, 913] [0, 487] Note: A total of 4601 bottom-trawl samples with 2328 in spring (March–May) and 2273 in fall (October–November). Table 1. The main dataset attributes and the number of missing values (NA) Attribute Description Spring Fall Domain NA Domain NA Year Year {1982: 2013} {1982: 2013} Longitude Beginning longitude of each tow (decimal degree) [–70.83, –65.67] [–70.84, –65.68] Latitude Beginning latitude of each tow (decimal degree) [41.68, 44.43] [41.71, 44.43] Temperature Bottom water temperatures (°C) [2.1, 11.3] 332 [4.6, 15.7] 341 Salinity Bottom salinity (mainly in 1982–1996) [30.51, 35.31] 1 058 [31.19, 35.27] 1 037 Depth Maximum depth of survey site (m) [19, 371] 9 [12, 368] 6 Density Survey catch density (no. tow−1) [0, 913] [0, 487] Attribute Description Spring Fall Domain NA Domain NA Year Year {1982: 2013} {1982: 2013} Longitude Beginning longitude of each tow (decimal degree) [–70.83, –65.67] [–70.84, –65.68] Latitude Beginning latitude of each tow (decimal degree) [41.68, 44.43] [41.71, 44.43] Temperature Bottom water temperatures (°C) [2.1, 11.3] 332 [4.6, 15.7] 341 Salinity Bottom salinity (mainly in 1982–1996) [30.51, 35.31] 1 058 [31.19, 35.27] 1 037 Depth Maximum depth of survey site (m) [19, 371] 9 [12, 368] 6 Density Survey catch density (no. tow−1) [0, 913] [0, 487] Note: A total of 4601 bottom-trawl samples with 2328 in spring (March–May) and 2273 in fall (October–November). Figure 2. View largeDownload slide The distribution of the NEFSC bottom-trawl survey stations with observed Atlantic cod density (no. tow−1) in spring and fall over the period 1982–1998 and 1999–2013. The size and color of points represent the calibrated catch rate. Figure 2. View largeDownload slide The distribution of the NEFSC bottom-trawl survey stations with observed Atlantic cod density (no. tow−1) in spring and fall over the period 1982–1998 and 1999–2013. The size and color of points represent the calibrated catch rate. Considering the large number of missing values for environmental variables in the NEFSC bottom-trawl surveys (e.g. bottom salinity, see Table 1), the hourly hindcast from the Northeast Coastal Ocean Forecast System (NECOFS) was used to substitute the missing values in order to minimize information loss. NECOFS is an integrated atmosphere–ocean circulation model forecast system designed for the northeast U.S. coastal region and has been validated by hindcast experiments since 1978 (http://fvcom.smast.umassd.edu/necofs/). For each environmental variable (e.g. depth, temperature, and salinity), estimated values at each station were calculated by averaging the hourly hindcast of its nearest three neighbour stations. Then, the estimates were linearly regressed against the complete records to evaluate congruency between observed and estimated data (Guan et al., 2017). Regression coefficients showed that NECOFS predictions were almost unbiased for depth, bottom temperature, and bottom salinity (Supplementary Figure S2). As a preliminary step in the statistical analysis, the variance inflation factor (VIF) was used to test and remove collinearity between explanatory variables (Dormann et al., 2013). The result showed that salinity should be removed in the following analysis because its VIF value was larger than the cut-off value of 4 in both spring and fall (Supplementary Figure S3). Habitat associations and geographic distribution Associations between cod catch and environmental variables (depth and temperature) were quantified using the ECDF curve, f(t), and the catch-weighted ECDF curve, g(t) (Perry and Smith, 1994). The median environmental variables favoured by cod as well as the 25th and 75th percentiles were calculated using f(t) and g(t). The test statistic D is the maximum absolute difference between f(t) and g(t) curves, similar to the Kolmogorov–Smirnov test for comparing ECDFs. Bootstrapping was used to assess the significance of difference between the observed D and distribution of values obtained from 2999 bootstraps of the data (Kreiner et al., 2015). A distribution index of geographic range (area occupied) was calculated in spring and fall. First, the catch-weighted ECDF curve F(c) and cumulative area G(c) of cod catch were calculated (Swain and Sinclair, 1994) for each year. A metric of concentration D90 (Swain and Sinclair, 1994; Reuchlin-Hugenholtz et al., 2015) was then evaluated, with the minimum area containing 90% of cod catch as a proportion of the total area. Given the recent increase in relative concentration of cod in the WGoM, we calculated the proportion of cod distribution that occurred in the WGoM over 1982–2013 (PWGoM; Figure 1a). A detailed description of f(t), g(t), F(t), G(t), and PWGoM is presented in the Supplementary Material. Regression models The regression models were developed to study both environmentally driven processes and density-dependent dynamics. Two different formulations were used: one spatially invariant and one spatially variant coefficient models (Bartolino et al., 2012). For our modelling effort, the term “cod density” or “cod catch” refers to the standardized catch rates (no. tow−1) of cod at a particular location (longitude and latitude), while “cod abundance” refers to the annual estimates of the stratified mean abundance in a given year. In all formulations, the response variable y was the cod density and Y- is the cod abundance. The two models were expressed as follows: M1:y=α+Y-+gtemp+gdepth+slon, lat+ε M2:y=α+Y-+g(temp)+g(depth)+slon, lat+slon, lat·Y-+slon, lat·temp+ε where α is the year-specific intercept, ε is random error term, g and s are one-dimensional thin plate regression spline and two-dimensional soap film smoothing functions (Miller and Wood, 2014; Wood et al., 2016), respectively. The model M1 represents fully additive models (Ciannelli et al., 2012), where the linear effect of cod population size is assumed to be spatially invariant by the term Y- (Youcef et al., 2013; Bartolino et al., 2017). The model M2 is a spatially explicit variable coefficient model (Bacheler et al., 2009), which assumes that the coefficients of regression are allowed to change smoothly in relation to the geographical position (Bacheler et al., 2009; Bartolino et al., 2011, 2012). This model was used to test the presence of spatially variable effects between density-independent (e.g. temperature: s(lon, lat)·temp) and density-dependent variables (e.g. abundance; s(lon, lat)·Y-). A Tweedie regression model with a log-link was selected for this study, which is the most common approach for overdispersed continuous data (Miller et al., 2013; Víkingsson et al., 2015). The best-fitting M1 and M2 were identified using a backward selection procedure (removing insignificant terms successively) based on genuine cross validation (gCV) score (Ciannelli et al., 2007) and Akaike information criterion (AIC) scores and model diagnostics. In our study, tenfold cross validation procedures (a total of 10 × 10 = 100 experiments) were used for each model, and the gCV was calculated as the average squared prediction error. Smooth terms in regression models were centred around 0 to assure model identifiability, and a constant ( α) was added to the predictions to scale back to original values (Bacheler et al., 2009; Bartolino et al., 2011). The estimation method was the restricted maximum likelihood (REML). Considering the large fluctuation in survey abundance indices (Supplementary Figure S4), these two regression models were developed without the data from 2000, 2007, and 2008 in spring to 1982 and 1988 in fall. A hexagonal grid (3864 hexagons, 20.77 km2 per hexagon) was created for model prediction over our study area, within which modelled cod densities and annual cod abundance were estimated using the median survey date in spring (21 April) and fall (25 October) from 1982 to 2013. All analyses were done in the R programing environment (R Core Team, 2016). The following R packages were used to implement the statistical analyses and visualization; mgcv (version 1.8-17; Wood, 2001), ggplot2 (Wickham, 2009), and latticeExtra (Sarkar and Andrews, 2012). Results Habitat associations Significant associations were identified between cod catch and the bottom temperature and depth variables in spring and fall during 1982–2013 (Figure 3). The curves of the raw cumulative frequency distribution f(t) were significantly smaller than the catch-weighted ECDFs, g(t), over the range of each environmental variable (p < 0.05) and significantly larger for bottom temperature for fall (Figure 3c). The optimal temperature ranges were found between 5.34°C and 6.26 °C in spring (Figure 3a) and 8.43–9.08 °C in fall (Figure 3c). For depth, the optimal ranges were between 123 and 141 m (Figure 3b) and 139–156 m correspondingly in spring and fall (Figure 3d). The maximum absolute difference, D(t), indicating the strongest seasonal association, was found at 5.44 °C and 8.69 °C for bottom temperature, 33.08 and 33.49 for salinity, and 135 m and 148 m for depth for spring and fall, respectively. Figure 3. View largeDownload slide The empirical relationship between seasonal Atlantic cod abundance (spring [a and b] and fall [c and d]: 1983–2013) and two environmental variables (bottom water temperature and depth). The solid lines represent the ECDF curves. The short dotted lines represent the catch-weighted ECDF curves. The long dotted lines represent the difference between the two curves. The black dashed lines indicate the maximum D(t), and the shaded areas indicate the 95% confidence intervals. The histogram in each panel shows the frequency distribution of survey effort (no. of tows) within the observed ranges of two environmental variables. Figure 3. View largeDownload slide The empirical relationship between seasonal Atlantic cod abundance (spring [a and b] and fall [c and d]: 1983–2013) and two environmental variables (bottom water temperature and depth). The solid lines represent the ECDF curves. The short dotted lines represent the catch-weighted ECDF curves. The long dotted lines represent the difference between the two curves. The black dashed lines indicate the maximum D(t), and the shaded areas indicate the 95% confidence intervals. The histogram in each panel shows the frequency distribution of survey effort (no. of tows) within the observed ranges of two environmental variables. Geographic distribution The estimated geographic range (D90) of cod gradually declined throughout the time-series during both seasons (Figure 4), reflecting a migration to shallow waters (Stellwagen Bank: Supplementary Figure S5). In most of the years after 1998, both spring and fall cod concentrations were higher in the WGoM (Figure 4a and c). For the years 1982–1998, the average PWGoM was 51.35% in spring and 57.29% in fall. For 1999–2013, the average PWGoM was 73.13% in spring and 76.09% in fall. A significant negative relationship (p < 0.05) was found between D90 and PWGoM in spring (Figure 4b) and fall (Figure 4d). Figure 4. View largeDownload slide Time-series and scatterplot of the minimum area containing 90% of cod catch (D90) and the proportion of cod occurring in WGoM (PWGoM) in spring (a and b) and fall (c and d). The heavy dashed lines (panels a and c) are the means of PWGoM for the 1982–1998 and 1999–2013 periods. Linear regression lines and equations (* p < 0.05; panels b and d) are shown. D95 and PWGoM is loge and arcsine transformed, respectively. WGoM: western Gulf of Maine. Figure 4. View largeDownload slide Time-series and scatterplot of the minimum area containing 90% of cod catch (D90) and the proportion of cod occurring in WGoM (PWGoM) in spring (a and b) and fall (c and d). The heavy dashed lines (panels a and c) are the means of PWGoM for the 1982–1998 and 1999–2013 periods. Linear regression lines and equations (* p < 0.05; panels b and d) are shown. D95 and PWGoM is loge and arcsine transformed, respectively. WGoM: western Gulf of Maine. Correlation analysis showed that non-significant associations between D90 and cod abundance indices were marginally negative during both spring (Figure 5a, r2 = 0.108) and fall (Figure 5c, r2 = 0.028). However, PWGoM showed significant positive associations with abundance indices for both seasons (spring: Figure 5b, r2 = 0.124, p < 0.05; fall: Figure 5d, r2 = 0.137, p < 0.05). Figure 5. View largeDownload slide Relationships between D90 (loge transformed) or PWGoM (arcsine transformed) and Atlantic cod abundance index (loge transformed) in spring (panels a and b) and fall (panels c and d). Linear regression lines and equations (* p < 0.05) are shown. Numbers within each panel denote year. Figure 5. View largeDownload slide Relationships between D90 (loge transformed) or PWGoM (arcsine transformed) and Atlantic cod abundance index (loge transformed) in spring (panels a and b) and fall (panels c and d). Linear regression lines and equations (* p < 0.05) are shown. Numbers within each panel denote year. Regression models The model-selection procedure (Table 2) showed that the spatially variant coefficient models (M2) were statistically superior to the fully additive models (M1). The best-fitting spring M2 model showed that depth has a non-linear effect, which remained positive for depth ranges shallower than ca. 196 m (Figure 6a). The interactive spatial effect on cod densities was highest in eastern and western waters (Figure 6b). Abundance had a negative effect on cod densities near Wilkinson and Jordan basins (Figures 1b and 6c). Offshore waters were predominantly characterized by a negative bottom temperature effect on cod densities (Figure 6d). In fall, bottom temperature had a non-linear effect on cod densities, with peak positive effects between 6.51°C and 10.42 °C (Figure 7a), and depth exhibited a non-linear effect, which is positive at depths <163 m (Figure 7b). The negative interactive spatial effect on cod densities was observed in northeastern coastal waters and the southwestern edge of the study area (Figure 7c). Cod abundance had a positive effect on cod densities in the most northern areas and a negative effect in the southeast (Figure 7d). Table 2. Parsimonious generalized additive models for the two formulations (M1 and M2) implemented in the analysis of cod spatial distribution in the GoM Spring Fall M1 M2 M1 M2 Y- 0.357** 0.417** g (temp) 0.772 4.412** 4.416** g (depth) 4.574** 4.781** 3.796** 3.247** s (lon., lat.) 49.169** 33.835** 46.778** 45.728** s (lon., lat.): Y- 10.329** 21.033** s (lon., lat.): temp 24.801** R2 (adj.) 0.185 0.199 0.326 0.366 Dev. expl. 0.535 0.562 0.582 0.602 AIC 8 415.504 8 354.692 7 498.588 7 463.905 gCV 15.755 15.714 10.850 10.581 Spring Fall M1 M2 M1 M2 Y- 0.357** 0.417** g (temp) 0.772 4.412** 4.416** g (depth) 4.574** 4.781** 3.796** 3.247** s (lon., lat.) 49.169** 33.835** 46.778** 45.728** s (lon., lat.): Y- 10.329** 21.033** s (lon., lat.): temp 24.801** R2 (adj.) 0.185 0.199 0.326 0.366 Dev. expl. 0.535 0.562 0.582 0.602 AIC 8 415.504 8 354.692 7 498.588 7 463.905 gCV 15.755 15.714 10.850 10.581 Estimated degrees of freedom (or linear coefficient in the case of parametric terms) and statistical significance are shown for each term (**p < 0.01), as well as the adjusted R2, deviance explained (Dev. expl.), AIC, and gCV score. Table 2. Parsimonious generalized additive models for the two formulations (M1 and M2) implemented in the analysis of cod spatial distribution in the GoM Spring Fall M1 M2 M1 M2 Y- 0.357** 0.417** g (temp) 0.772 4.412** 4.416** g (depth) 4.574** 4.781** 3.796** 3.247** s (lon., lat.) 49.169** 33.835** 46.778** 45.728** s (lon., lat.): Y- 10.329** 21.033** s (lon., lat.): temp 24.801** R2 (adj.) 0.185 0.199 0.326 0.366 Dev. expl. 0.535 0.562 0.582 0.602 AIC 8 415.504 8 354.692 7 498.588 7 463.905 gCV 15.755 15.714 10.850 10.581 Spring Fall M1 M2 M1 M2 Y- 0.357** 0.417** g (temp) 0.772 4.412** 4.416** g (depth) 4.574** 4.781** 3.796** 3.247** s (lon., lat.) 49.169** 33.835** 46.778** 45.728** s (lon., lat.): Y- 10.329** 21.033** s (lon., lat.): temp 24.801** R2 (adj.) 0.185 0.199 0.326 0.366 Dev. expl. 0.535 0.562 0.582 0.602 AIC 8 415.504 8 354.692 7 498.588 7 463.905 gCV 15.755 15.714 10.850 10.581 Estimated degrees of freedom (or linear coefficient in the case of parametric terms) and statistical significance are shown for each term (**p < 0.01), as well as the adjusted R2, deviance explained (Dev. expl.), AIC, and gCV score. Figure 6. View largeDownload slide Parsimonious generalized additive model results. (a) Partial effect of depth, (b) partial effect of spatial position, (c) spatial effect of abundance, and (d) spatial effect of temperature on cod density in spring. Decimal values in the title represent the estimated degrees of freedom, and the dotted lines indicate the zero value. Shaded areas indicate the 95% confidence intervals, and the rugplots on the x-axis indicate the distribution of observations for single covariates. Figure 6. View largeDownload slide Parsimonious generalized additive model results. (a) Partial effect of depth, (b) partial effect of spatial position, (c) spatial effect of abundance, and (d) spatial effect of temperature on cod density in spring. Decimal values in the title represent the estimated degrees of freedom, and the dotted lines indicate the zero value. Shaded areas indicate the 95% confidence intervals, and the rugplots on the x-axis indicate the distribution of observations for single covariates. Figure 7. View largeDownload slide Parsimonious generalized additive model results. (a) Partial effect of bottom water temperature, (b) partial effect of depth, (c) partial effect of spatial position, and (d) spatial effect of abundance on cod density in fall. See Figure 6 for more details. Figure 7. View largeDownload slide Parsimonious generalized additive model results. (a) Partial effect of bottom water temperature, (b) partial effect of depth, (c) partial effect of spatial position, and (d) spatial effect of abundance on cod density in fall. See Figure 6 for more details. The average bottom temperature remained relatively stable between 1982 and 2007, but showed an abrupt increase in fall 2009 and spring 2010 (Figure 8a). The interannual variance of average bottom temperature after 1999 was greater (5.61–7.66 °C) than that of previous years (5.73–6.72 °C) in spring. Predicted seasonal cod distribution for 1983, 1994, 2002, and 2013 based on the best-fitting M2 model showed cod abundance declined in the 1980s, increased slightly from 1999 to 2002, and decreased from 2010 to the lowest level in 2013 (Figure 8b). Average abundance between 2009 and 2013 was about the same as in the mid-1990s. Over the study period, modelled cod abundance was mainly concentrated around Stellwagen Bank in the WGoM, with low-density, small-scale aggregations (i.e. patches) in spring and relatively high-density patches in fall in the northern and eastern waters (e.g. Jeffreys Bank and German Bank). The decline in cod abundance since 2010 was accompanied by the disappearance of patches in northern waters and a further contraction near Stellwagen Bank. High cod density was rarely predicted in deep waters, especially in three basins: Wilkinson Basin, Jordan Basin, and Georges Basin (Figure 8c). Figure 8. View largeDownload slide Temporal trend of (a) average bottom water temperature and (b) modelled annual Atlantic cod abundance index; and (c) predicted cod distributions at different abundance levels (1983, 1994, 2002, and 2013) during spring and fall. For modelled annual abundance index (b), the error bars indicate their respective uncertainty (±1 s.e.). Figure 8. View largeDownload slide Temporal trend of (a) average bottom water temperature and (b) modelled annual Atlantic cod abundance index; and (c) predicted cod distributions at different abundance levels (1983, 1994, 2002, and 2013) during spring and fall. For modelled annual abundance index (b), the error bars indicate their respective uncertainty (±1 s.e.). Discussion Our study identified changes in habitat associations and geographic distribution of Atlantic cod in the GoM in both spring and fall during 1982–2013. The area occupied (D90) remained relatively high through the late 1990s, but shifted to a pronounced contraction into the WGoM area for the reminder of our time-series. The model results suggest that spatio-temporal distribution of GoM cod has been driven by both density-dependent and density-independent effects during the past three decades. The abrupt warming of the GoM has reduced recruitment and increased mortality of the GoM cod stock (Mills et al., 2013), reducing the species’ capacity to rebound from overfishing (Pershing et al., 2015). Our results have shown that the average temperature in the GoM increased steadily and rapidly after 2007 during both spring and fall (Figure 8a). Strong associations were identified between cod catch and bottom water temperature in spring and fall during 1982–2013 (Figure 3a and c). Colder waters during spring could lead to an increase in the availability of suitable cod habitat, while colder waters during fall could result in less area occupied (i.e. lower D90). Simultaneously, cod mostly appeared in shallower waters in spring (Figure 3b; 123–141 m) compared with fall (Figure 3d; 139–156 m). The seasonal difference in habitat associations likely reflects the seasonal migrations regulated by changes in bottom temperature and local bathymetric features (Freitas et al., 2015; Zemeckis et al., 2017). GoM cod spawn in winter and spring (Huret et al., 2007; Howell et al., 2008; Kovach et al., 2010), with peak spawning activity tending to occur in May (Churchill et al., 2011). The spring-spawning GoM cod may exhibit an intrinsic homing behaviour, with little migration out of the spawning area in the inshore WGoM (Howell et al., 2008). The offshore GoM fishing grounds could be inhabited by cod that migrate offshore after spawning along the coast, likely moving in pursuit of prey (Ames and Lichter, 2013). Habitat occupation is a key ecological indicator and its quantification is essential for an ecosystem-based management approach (Bartolino et al., 2017). The recent decline in the GoM cod biomass has resulted in shrinkage of the geographic aggregation and may increase the risk of local depletion because fishing effort might be forced to relocate to areas with higher habitat suitability (Jennings, 2000). In fact, annual catch of the GoM cod declined from 20 978 t (the highest since the early 1900s) in 1991 to 3078 t in 1999, fluctuated between 4200 and 8400 t before 2011, and then declined sharply to 1715 t (the lowest since the early 1900s) in 2013 (NEFSC, 2014). This difference between survey abundance and total landings of cod is likely due to higher concentration and subsequent patches having a positive feedback effect on overall abundance (Blanchard et al., 2008) derived from the NEFSC bottom-trawl surveys. Compared with total landings, the NEFSC bottom-trawl surveys based on the depth-stratified random design have limited coverage of inshore waters <20 m. For marine fish populations, the abundance–occupancy relationship can be used to examine density-dependent habitat selection with spatially and temporally extensive fishery-independent data (Blanchard et al., 2005; Bacheler et al., 2012). Knowledge of the abundance–occupancy relationship has implications for fisheries management and for the recovery of depleted fish populations. According to the ideal free distribution (IFD) theory (Fretwell and Calver, 1969), species that are locally abundant tend to be more widespread than species that are locally rare, leading to a positive abundance–occupancy relationship (Gaston et al., 1997, 2000) commonly found in many marine fish population (Atkinson et al., 1997; Blanchard et al., 2005; Swain and Benoît, 2006; Youcef et al., 2013). However, evidence for the negative relationship is also found in some marine fish populations (Sagarese et al., 2014; Thorson et al., 2016). Previous studies on cod (age group 2) in the North Sea have shown that the abundance–occupancy relationship may be negative in winter (Quinn et al., 2009) and positive in summer (Blanchard et al., 2005). In this study, the abundance–occupancy relationship for cod was found to be negative during both spring and fall (Figure 6). Both artefactual (e.g. sampling artefact) and ecological (e.g. niche breadthand and metapopulation theory) mechanisms can be used to explain the abundance–occupancy relationship (Päivinen et al., 2005). According to the niche breadthand, the GoM cod may be habitat-specific species, although the species could transform to generalist foragers in poorer environments (Ljungberg et al., 2012). Cod prefer sand lance (Ammodytes spp.), crabs (Cancer spp.), and herring (Clupea harengus) (Link and Garrison, 2002), and the GoM cod have been aggregating in a small area (e.g. Stellwagen Bank) because of an increase in sand lance abundance (Richardson et al., 2014). According to the metapopulation theory, the GoM cod may have relatively low dispersal ability, and only the largest and best quality patches (e.g. Stellwagen Bank in WGoM) will be occupied (Blanchard et al., 2005). In addition, the abundance–occupancy relationship can also be affected by fishing activities around the patches, which can cause individual fish to avoid the margins and restrict their movements to the interior (Dean et al., 2014). This study found that the spatially variant coefficient M2 models were statistically superior to the fully additive M1 models, showing that density-dependent (abundance) and density-independent (temperature) variables had spatially variant effects on the spring cod densities, while only abundance had a spatial effect in fall (Table 2). The results revealed that the response of cod density to abundance and temperature has a season-specific spatial structure and relies on a number of ecological and fisheries implications (Bacheler et al., 2009). From an ecological perspective, variable coefficient models can improve our understanding of the non-stationary and non-linear effects (Ciannelli et al., 2004) of both density-dependent and density-independent variables on the local densities of cod. This spatially variant information can underscore the value of spatially explicit management as a means of designing the marine-protected areas and fostering the recovery of cod population. Our model showed a larger decline in cod density in the western and northern coastal waters than in the rest of the GoM during spring. In fall, cod density decreased more in widespread coastal waters. Concurrently, cod density during spring in the narrow coastal waters increased more than in the rest of the GoM with the rapid increase in temperature in the late 1990s, resulting in a spatial contraction of the cod population after the late 1990s into the WGoM area. This contraction into WGoM could have also been related to the loss of spawning subcomponents along the northeast coast of GoM (Zemeckis et al., 2014). The spatial overlap between abundance and temperature effects suggest that density-independent factors can mediate the influence of density-dependent processes in spring, as reported for other species in previous studies (Ciannelli et al., 2004; Bacheler et al., 2009). However, the rapid increase in temperature could influence cod density uniformly over all areas during fall, probably because cod spatial distributions are less sensitive to temperature then than during spring. The partial effects of our model also revealed that cod density was higher at depths <196 m in spring and 163 m in fall, indicating these depths as the preferred bathymetry for cod at the local scale. Moreover, many of other hydrographic conditions (e.g. substrate type, water column stratification) and intraspecific interactions (e.g. prey abundance, predation risk) probably influence cod density at much smaller scales than temperature, which tends to integrate more small-scale patchiness for fish distributions (Blanchard et al., 2005). In this study, the survey density data are characterized by left-skewed distributions (Supplementary Figure S1), with many zero catches and few extremely high catches (Figure 2), which are very similar to the highly aggregated or patchy distributions encountered during ichthyoplankton or plankton surveys (Li et al., 2017). The Tweedie distribution with a log-link was selected to these overdispersed continuous data to avoid the multiplicative structure of two-stage models (Foster and Bravington, 2013; Li et al., 2017). Moreover, a GAM with threshold interaction (Ciannelli et al., 2004) may be an alternative model, because the responses of cod to climate change seem to be discrete in the GoM, which may need further study. Long-term, spatially resolved data on fishing pressure are not currently available and could not be included in our analyses, but if available in the future, they should be taken into account in further analyses on the GoM cod spatial distribution to understand the fishery effect on cod at different spatial scales. In conclusion, our study suggests that both density-dependent and density-independent processes can affect spatio-temporal dynamics of fish populations in a spatially explicit fashion because of the strong spatial component in the interplay between intraspecific interactions and environmental forcing (Bartolino et al., 2017). Future studies of population dynamics should not only consider variables that directly influence the abundance of cod, but also how those variables affect the spatial distribution of cod. Better knowledge of the spatio-temporal dynamics of cod distributions can improve our understanding of the causality of abundance–occupancy and species–environment relationships and help us to reduce error estimates for survey-based indices (Ciannelli et al., 2012). Moreover, understanding and predicting the spatial dynamics has important implications for fisheries management and biological conservation of this cod stock. This study in the GoM area highlights the need to incorporate both density-dependent and density-independent variables into fishery resource management. Supplementary data Supplementary material is available at the ICESJMS online version of the manuscript. Acknowledgements This study was supported by the Specialized Research Fund for the Doctoral Program of Higher Education (20120132130001) and the Fundamental Research Funds for the Central Universities (201562030). The senior author’s study in Yong Chen’s laboratory in the School of Marine Sciences, University of Maine was supported by the China Scholarship Council, Ocean University of China, and University of Maine. References Ames E. P. 2004 . Atlantic cod stock structure in the Gulf of Maine . Fisheries , 29 : 10 – 28 . Google Scholar CrossRef Search ADS Ames E. P. , Lichter J. 2013 . Gadids and alewives: structure within complexity in the Gulf of Maine . Fisheries Research , 141 : 70 – 78 . Google Scholar CrossRef Search ADS Anderson J. T. , Gregory R. S. 2000 . Factors regulating survival of northern cod (NAFO 2J3KL) during their first 3 years of life . ICES Journal of Marine Science , 57 : 349 – 359 . Google Scholar CrossRef Search ADS Atkinson D. B. , Rose G. A. , Murphy E. F. , Bishop C. A. 1997 . Distribution changes and abundance of northern cod (Gadus morhua), 1981–1993 . Canadian Journal of Fisheries and Aquatic Sciences , 54 : 132 – 138 . Bacheler N. M. , Bailey K. M. , Ciannelli L. , Bartolino V. , Chan K.-S. 2009 . Density-dependent, landscape, and climate effects on spawning distribution of walleye pollock Theragra chalcogramma . Marine Ecology Progress Series , 391 : 1 – 12 . Google Scholar CrossRef Search ADS Bacheler N. M. , Buckel J. A. , Paramore L. M. , Rochet M.-J. 2012 . Density-dependent habitat use and growth of an estuarine fish . Canadian Journal of Fisheries and Aquatic Sciences , 69 : 1734 – 1747 . Google Scholar CrossRef Search ADS Balch W. M. , Drapeau D. T. , Bowler B. C. , Booth E. S. , Windecker L. A. , Ashe A. 2008 . Space–time variability of carbon standing stocks and fixation rates in the Gulf of Maine, along the GNATS transect between Portland, ME, USA, and Yarmouth, Nova Scotia, Canada . Journal of Plankton Research , 30 : 119 – 139 . Google Scholar CrossRef Search ADS Bartolino V. , Ciannelli L. , Bacheler N. M. , Chan K.-S. 2011 . Ontogenetic and sex-specific differences in density-dependent habitat selection of a marine fish population . Ecology , 92 : 189 – 200 . Google Scholar CrossRef Search ADS PubMed Bartolino V. , Ciannelli L. , Spencer P. , Wilderbuer T. , Chan K. 2012 . Scale-dependent detection of the effects of harvesting a marine fish population . Marine Ecology Progress Series , 444 : 251 – 261 . Google Scholar CrossRef Search ADS Bartolino V. , Tian H. , Bergström U. , Jounela P. , Aro E. , Dieterich C. , Meier H. E. M. , et al. 2017 . Spatio-temporal dynamics of a fish predator: density-dependent and hydrographic effects on Baltic Sea cod population . PLoS One , 12 : e0172004. Google Scholar CrossRef Search ADS PubMed Blanchard J. L. , Maxwell D. L. , Jennings S. 2008 . Power of monitoring surveys to detect abundance trends in depleted populations: the effects of density-dependent habitat use, patchiness, and climate change . ICES Journal of Marine Science , 65 : 111 – 120 . Google Scholar CrossRef Search ADS Blanchard J. L. , Mills C. , Jennings S. , Fox C. J. , Rackham B. D. , Eastwood P. D. , O'Brien C. M. 2005 . Distribution abundance relationships for North Sea Atlantic cod (Gadus morhua): observation versus theory . Canadian Journal of Fisheries and Aquatic Sciences , 62 : 2001 – 2009 . Google Scholar CrossRef Search ADS Chang J.-H. , Chen Y. , Holland D. , Grabowski J. 2010 . Estimating spatial distribution of American lobster Homarus americanus using habitat variables . Marine Ecology Progress Series , 420 : 145 – 156 . Google Scholar CrossRef Search ADS Churchill J. H. , Runge J. , Chen C. 2011 . Processes controlling retention of spring-spawned Atlantic cod (Gadus morhua) in the western Gulf of Maine and their relationship to an index of recruitment success . Fisheries Oceanography , 20 : 32 – 46 . Google Scholar CrossRef Search ADS Ciannelli L. , Bartolino V. , Chan K.-S. 2012 . Non-additive and non-stationary properties in the spatial distribution of a large marine fish population . Proceedings of the Royal Society B: Biological Sciences , 279 : 3635 – 3642 . Google Scholar CrossRef Search ADS Ciannelli L. , Chan K.-S. , Bailey K. M. , Stenseth N. C. 2004 . Nonadditive effects of the environment on the survival of a large marine fish population . Ecology , 85 : 3418 – 3427 . Google Scholar CrossRef Search ADS Ciannelli L. , Dingsør G. E. , Bogstad B. , Ottersen G. , Chan K-s. , Gjøsæter H. , Stiansen J. E. , et al. 2007 . Spatial anatomy of species survival: effects of predation and climate-driven environmental variability . Ecology , 88 : 635 – 646 . Google Scholar CrossRef Search ADS PubMed Ciannelli L. , Fauchald P. , Chan K. S. , Agostini V. N. , Dingsør G. E. 2008 . Spatial fisheries ecology: recent progress and future prospects . Journal of Marine Systems , 71 : 223 – 236 . Google Scholar CrossRef Search ADS Craig J. K. , Crowder L. B. , Henwood T. A. 2005 . Spatial distribution of brown shrimp (Farfantepenaeus aztecus) on the northwestern Gulf of Mexico shelf: effects of abundance and hypoxia . Canadian Journal of Fisheries and Aquatic Sciences , 62 : 1295 – 1308 . Google Scholar CrossRef Search ADS Dean M. J. , Hoffman W. S. , Zemeckis D. R. , Armstrong M. P. 2014 . Fine-scale diel and gender-based patterns in behaviour of Atlantic cod (Gadus morhua) on a spawning ground in the western Gulf of Maine . ICES Journal of Marine Science , 71 : 1474 – 1489 . Google Scholar CrossRef Search ADS Dormann C. F. , Elith J. , Bacher S. , Buchmann C. , Carl G. , Carré G. , Marquéz J. R. G. , et al. 2013 . Collinearity: a review of methods to deal with it and a simulation study evaluating their performance . Ecography , 36 : 27 – 46 . Google Scholar CrossRef Search ADS Dormann F. C. , McPherson M. J. , Araújo B. M. , Bivand R. , Bolliger J. , Carl G. , Davies G. R. , et al. 2007 . Methods to account for spatial autocorrelation in the analysis of species distributional data: a review . Ecography , 30 : 609 – 628 . Google Scholar CrossRef Search ADS Drinkwater K. F. 2005 . The response of Atlantic cod (Gadus morhua) to future climate change . ICES Journal of Marine Science , 62 : 1327 – 1337 . Google Scholar CrossRef Search ADS Foster S. D. , Bravington M. V. 2013 . A Poisson–Gamma model for analysis of ecological non-negative continuous data . Environmental and Ecological Statistics , 20 : 533 – 552 . Google Scholar CrossRef Search ADS Freitas C. , Olsen E. M. , Moland E. , Ciannelli L. , Knutsen H. 2015 . Behavioral responses of Atlantic cod to sea temperature changes . Ecology and Evolution , 5 : 2070 – 2083 . Google Scholar CrossRef Search ADS PubMed Fretwell S. D. , Calver J. S. 1969 . On territorial behavior and other factors influencing habitat distribution in birds . Acta Biotheoretica , 19 : 37 – 44 . Google Scholar CrossRef Search ADS Gaston K. J. , Blackburn T. M. , Greenwood J. J. D. , Gregory R. D. , Quinn R. M. , Lawton J. H. 2000 . Abundance–occupancy relationships . Journal of Applied Ecology , 37 : 39 – 59 . Google Scholar CrossRef Search ADS Gaston K. J. , Blackburn T. M. , Lawton J. H. 1997 . Interspecific abundance-range size relationships: an appraisal of mechanisms . Journal of Animal Ecology , 66 : 579 – 601 . Google Scholar CrossRef Search ADS Gröger J. P. , Rountree R. A. , Thygesen U. H. , Jones D. , Martins D. , Xu Q. , Rothschild B. J. 2007 . Geolocation of Atlantic cod (Gadus morhua) movements in the Gulf of Maine using tidal information . Fisheries Oceanography , 16 : 317 – 335 . Google Scholar CrossRef Search ADS Guan L. , Chen Y. , Wilson J. A. 2017 . Evaluating spatio-temporal variability in the habitat quality of Atlantic cod (Gadus morhua) in the Gulf of Maine . Fisheries Oceanography , 26 : 83 – 96 . Google Scholar CrossRef Search ADS Halliday R. G. , Pinhorn A. T. 1990 . The delimitation of fishing areas in the Northwest Atlantic . Journal of Northwest Atlantic Fishery Science , 10 : 1 – 51 . Google Scholar CrossRef Search ADS Howell W. H. , Morin M. , Rennels N. , Goethel D. 2008 . Residency of adult Atlantic cod (Gadus morhua) in the western Gulf of Maine . Fisheries Research , 91 : 123 – 132 . Google Scholar CrossRef Search ADS Huret M. , Runge J. A. , Chen C. , Cowles G. , Xu Q. , Pringle J. M. 2007 . Dispersal modeling of fish early life stages: sensitivity with application to Atlantic cod in the western Gulf of Maine . Marine Ecology Progress Series , 347 : 261 – 274 . Google Scholar CrossRef Search ADS Jennings S. 2000 . Patterns and prediction of population recovery in marine reserves . Reviews in Fish Biology and Fisheries , 10 : 209 – 231 . Google Scholar CrossRef Search ADS Kovach A. I. , Breton T. S. , Berlinsky D. L. , Maceda L. , Wirgin I. 2010 . Fine-scale spatial and temporal genetic structure of Atlantic cod off the Atlantic coast of the USA . Marine Ecology Progress Series , 410 : 177 – 195 . Google Scholar CrossRef Search ADS Kreiner A. , Yemane D. , Stenevik E. K. 2015 . Spawning habitats of Cape horse mackerel (Trachurus capensis) in the northern Benguela upwelling region . Fisheries Oceanography , 24 : 46 – 55 . Google Scholar CrossRef Search ADS Li Z. , Wan R. , Ye Z. , Chen Y. , Ren Y. , Liu H. , Jiang Y. 2017 . Use of random forests and support vector machines to improve annual egg production estimation . Fisheries Science , 83 : 1 – 11 . Google Scholar CrossRef Search ADS Link J. S. , Bogstad B. , Sparholt H. , Lilly G. R. 2009 . Trophic role of Atlantic cod in the ecosystem . Fish and Fisheries , 10 : 58 – 87 . Google Scholar CrossRef Search ADS Link J. S. , Garrison L. P. 2002 . Trophic ecology of Atlantic cod Gadus morhua on the northeast US continental shelf . Marine Ecology Progress Series , 227 : 109 – 124 . Google Scholar CrossRef Search ADS Ljungberg P. , Nilsson P. A. , Persson A. 2012 . Prey selectivity by juvenile Atlantic cod Gadus morhua in three coastal habitat types . Marine Ecology Progress Series , 466 : 215 – 223 . Google Scholar CrossRef Search ADS Llope M. , Chan K.-S. , Ciannelli L. , Reid P. C. , Stige L. C. , Stenseth N. C. 2009 . Effects of environmental conditions on the seasonal distribution of phytoplankton biomass in the North Sea . Limnology and Oceanography , 54 : 512 – 524 . Google Scholar CrossRef Search ADS Llope M. , Licandro P. , Chan K. S. , Stenseth N. C. 2012 . Spatial variability of the plankton trophic interaction in the North Sea: a new feature after the early 1970s . Global Change Biology , 18 : 106 – 117 . Google Scholar CrossRef Search ADS Miller D. L. , Burt M. L. , Rexstad E. A. , Thomas L. , Gimenez O. 2013 . Spatial models for distance sampling data: rcent developments and future directions . Methods in Ecology and Evolution , 4 : 1001 – 1010 . Google Scholar CrossRef Search ADS Miller D. L. , Wood S. N. 2014 . Finite area smoothing with generalized distance splines . Environmental and Ecological Statistics , 21 : 715 – 731 . Google Scholar CrossRef Search ADS Mills K. , Pershing A. , Brown C. , Chen Y. , Chiang F.-S. , Holland D. , Lehuta S. , et al. 2013 . Fisheries management in a changing climate lessons from the 2012 ocean heat wave in the Northwest Atlantic . Oceanography , 26 : 191 – 195 . Google Scholar CrossRef Search ADS NEFSC . 2013 . 55th Northeast regional stock assessment workshop (55th SAW) assessment report. National Oceanic and Atmospheric Administration, National Marine Fisheries Service, NEFSC Reference Document 13-11. NEFSC . 2014 . Gulf of Maine Atlantic Cod 2014 assessment update report. Draft working paper for peer review only. National Oceanic and Atmospheric Administration, National Marine Fisheries Service, Woods Hole, MA. NEFSC . 2015 . Gulf of Maine Atlantic Cod 2015 assessment update report. Draft working paper for peer review only. National Oceanic and Atmospheric Administration, National Marine Fisheries Service, Woods Hole, MA. Nye J. A. , Link J. S. , Hare J. A. , Overholtz W. J. 2009 . Changing spatial distribution of fish stocks in relation to climate and population size on the Northeast United States continental shelf . Marine Ecology Progress Series , 393 : 111 – 129 . Google Scholar CrossRef Search ADS Overholtz W. J. , Friedland K. D. 2002 . Recovery of the Gulf of Maine–Georges Bank Atlantic herring (Clupea harengus) complex: perspectives based on bottom trawl survey data . Fishery Bulletin , 100 : 593 – 608 . Päivinen J. , Grapputo A. , Kaitala V. , Komonen A. , Kotiaho J. S. , Saarinen K. , Wahlberg N. 2005 . Negative density-distribution relationship in butterflies . BMC Biology , 3 : 5. Google Scholar CrossRef Search ADS PubMed Perry R. I. , Smith S. J. 1994 . Identifying habitat associations of marine fishes using survey data: an application to the Northwest Atlantic . Canadian Journal of Fisheries and Aquatic Sciences , 51 : 589 – 602 . Google Scholar CrossRef Search ADS Pershing A. J. , Alexander M. A. , Hernandez C. M. , Kerr L. A. , Bris A. L. , Mills K. E. , Nye J. A. , et al. 2015 . Slow adaptation in the face of rapid warming leads to collapse of the Gulf of Maine cod fishery . Science , 350 : 809 – 812 . Google Scholar CrossRef Search ADS PubMed Pettigrew N. R. , Churchill J. H. , Janzen C. D. , Mangum L. J. , Signell R. P. , Thomas A. C. , Townsend D. W. , et al. 2005 . The kinematic and hydrographic structure of the Gulf of Maine Coastal Current . Deep Sea Research Part II: Topical Studies in Oceanography , 52 : 2369 – 2391 . Google Scholar CrossRef Search ADS Puerta P. , Hunsicker M. E. , Quetglas A. , Álvarez-Berastegui D. , Esteban A. , González M. , Hidalgo M. 2015 . Spatially explicit modeling reveals cephalopod distributions match contrasting trophic pathways in the Western Mediterranean Sea . PLoS One , 10 : e0133439. Google Scholar CrossRef Search ADS PubMed Quinn T. , Lewy P. , Kristensen K. 2009 . Modelling the distribution of fish accounting for spatial correlation and overdispersion . Canadian Journal of Fisheries and Aquatic Sciences , 66 : 1809 – 1820 . Google Scholar CrossRef Search ADS R Core Team . 2016 . R: A Language and Environment for Statistical Computing. Vienna, Austria: R Foundation for Statistical Computing. Reuchlin-Hugenholtz E. , Shackell N. L. , Hutchings J. A. , Stergiou K. I. 2015 . The potential for spatial distribution indices to signal thresholds in marine fish biomass . PLoS One , 10 : e0120500. Google Scholar CrossRef Search ADS PubMed Richardson D. E. , Palmer M. C. , Smith B. E. , Cooper A. 2014 . The influence of forage fish abundance on the aggregation of Gulf of Maine Atlantic cod (Gadus morhua) and their catchability in the fishery . Canadian Journal of Fisheries and Aquatic Sciences , 71 : 1349 – 1362 . Google Scholar CrossRef Search ADS Sagarese S. R. , Frisk M. G. , Miller T. J. , Sosebee K. A. , Musick J. A. , Rago P. J. , Rosenfeld J. 2014 . Influence of environmental, spatial, and ontogenetic variables on habitat selection and management of spiny dogfish in the Northeast (US) shelf large marine ecosystem . Canadian Journal of Fisheries and Aquatic Sciences , 71 : 567 – 580 . Google Scholar CrossRef Search ADS Sarkar D. , Andrews F. 2012 . Latticeextra: extra graphical utilities based on lattice. R package v.0.6-28. Spencer P. D. 2008 . Density-independent and density-dependent factors affecting temporal changes in spatial distributions of eastern Bering Sea flatfish . Fisheries Oceanography , 17 : 396 – 410 . Google Scholar CrossRef Search ADS Swain D. P. , Benoît H. P. 2006 . Change in habitat associations and geographic distribution of thorny skate (Amblyraja radiata) in the southern Gulf of St Lawrence: density-dependent habitat selection or response to environmental change? Fisheries Oceanography , 15 : 166 – 182 . Google Scholar CrossRef Search ADS Swain D. P. , Sinclair A. F. 1994 . Fish distribution and catchability: what is the appropriate measure of distribution? Canadian Journal of Fisheries and Aquatic Sciences , 51 : 1046 – 1054 . Google Scholar CrossRef Search ADS Swain D. P. , Wade E. J. 1993 . Density-dependent geographic distribution of Atlantic cod (Gadus morhua) in the southern Gulf of St. Lawrence . Canadian Journal of Fisheries and Aquatic Sciences , 50 : 725 – 733 . Google Scholar CrossRef Search ADS Tamdrari H. , Castonguay M. , Brêthes J.-C. , Duplisea D. 2010 . Density-independent and -dependent habitat selection of Atlantic cod (Gadus morhua) based on geostatistical aggregation curves in the northern Gulf of St Lawrence . ICES Journal of Marine Science , 67 : 1676 – 1686 . Google Scholar CrossRef Search ADS Thorson J. T. , Rindorf A. , Gao J. , Hanselman D. H. , Winker H. 2016 . Density-dependent changes in effective area occupied for sea-bottom-associated marine fishes . Proceedings of the Royal Society B , 283 : 20161853. Google Scholar CrossRef Search ADS PubMed Townsend D. W. , Pettigrew N. R. , Thomas A. C. 2001 . Offshore blooms of the red tide dinoflagellate, Alexandrium sp., in the Gulf of Maine . Continental Shelf Research , 21 : 347 – 369 . Google Scholar CrossRef Search ADS Townsend D. W. , Pettigrew N. R. , Thomas M. A. , Neary M. G. , McGillicuddy D. J. Jr , O'Donnell J. 2015 . Water masses and nutrient sources to the Gulf of Maine . Journal of Marine Research , 73 : 93 – 122 . Google Scholar CrossRef Search ADS PubMed Víkingsson G. A. , Pike D. G. , Valdimarsson H. , Schleimer A. , Gunnlaugsson T. , Silva T. , Elvarsson B. Þ. 2015 . Distribution, abundance, and feeding ecology of baleen whales in Icelandic waters: have recent environmental changes had an effect? Frontiers in Ecology and Evolution , 3 : 1 – 18 . Wickham H. 2009 . ggplot2: Elegant Graphics for Data Analysis . Springer , New York . 213 pp. Wood S. N. 2001 . mgcv: GAMs and generalized ridge regression for R . R News , 1 : 20 – 25 . Wood S. N. , Pya N. , Säfken B. 2016 . Smoothing parameter and model selection for general smooth models . Journal of the American Statistical Association , 111 : 1548 – 1563 . Google Scholar CrossRef Search ADS Youcef W. A. , Lambert Y. , Audet C. 2013 . Spatial distribution of Greenland halibut Reinhardtius hippoglossoides in relation to abundance and hypoxia in the estuary and Gulf of St. Lawrence . Fisheries Oceanography , 22 : 41 – 60 . Google Scholar CrossRef Search ADS Zemeckis D. R. , Dean M. J. , Cadrin S. X. 2014 . Spawning dynamics and associated management implications for Atlantic cod . North American Journal of Fisheries Management , 34 : 424 – 442 . Google Scholar CrossRef Search ADS Zemeckis D. R. , Liu C. , Cowles G. W. , Dean M. J. , Hoffman W. S. , Martins D. , Cadrin S. X. 2017 . Seasonal movements and connectivity of an Atlantic cod (Gadus morhua) spawning component in the western Gulf of Maine . ICES Journal of Marine Science , 74 : 1780 – 1796 . © International Council for the Exploration of the Sea 2018. All rights reserved. For Permissions, please email: [email protected] This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

doi: 10.1093/icesjms/fsy024pmid: N/A

Abstract Estimating spatial distribution of a species is traditionally achieved using global regression models with the assumption of spatial stationarity of relationships between species and environmental variables. However, species abundance and environmental variables are often spatially correlated and the strength of environmental effects may exhibit spatial non-stationarity on the species distribution. We applied local models, such as season-, sex-, and size-specific geographically weighted regression (GWR) models, on American lobster to explore non-stationary environmental effects on the presence and density of lobsters in the inshore Gulf of Maine (GOM). This species and its fishery have undergone a dramatic increase in abundance over the past two decades. Model results showed that the strength of the estimated relationships in the western GOM were different with the relationships in the eastern GOM during 2000–2014. Bottom water temperature had a more significant positive impact on the increase of lobsters in the eastern GOM, while the influence of temperature was less significant in the west and the more distinguishable drivers of distribution needed to be identified. The estimation of locally varied relationships can further improve regionally informed management plans. The modeling approach can be widely applied to many other species or study areas. Introduction Understanding the effects of environmental changes on species distribution is required for many aspects of resource management and environmental research (Franklin and Miller, 2009). There is a growing body of literature suggesting that changes in the spatial structure of a population may be caused by shifts in distribution in response to variations in environmental conditions (Ciannelli et al., 2012; Kotwicki and Lauth, 2013; Szuwalski and Hollowed, 2016). The distribution of a species is influenced by many abiotic (e.g. temperature and salinity) and biotic (e.g. predator, prey, and disease) drivers that operate simultaneously with different strengths at different spatial scales or locations, and may change over time. The dominant drivers of the distribution pattern may depend on the spatial scale and vary by subarea when the spatial scale changes. Therefore, it is difficult to disentangle the effects of multiple environmental variables on the presence or density of a species spatially (Ciannelli et al., 2012). This would require a study being conducted at multiple finer spatial scales. Identification of the spatial non-stationary environmental effects on the distribution of a species can improve our understanding of the species spatial dynamics at finer scales (Windle et al., 2010). The strength of an environmental effect on a species changes with its life stage or along the species’ range if there are sharp hydrographic or biogeographical gradients such as differences in local food availability (Frank et al., 2006). For example, the distribution of American lobster is regulated by both abiotic and biotic drivers but primarily driven by water temperature in the Gulf of Maine (GOM, Boudreau et al., 2015; Tanaka and Chen, 2016). The effect of predation on the lobster population increases and dominates at the cold and warm extremes of the thermal range (Boudreau et al., 2015; Le Bris et al., 2018). Post-settlement natural mortality rates have increased dramatically with the onset of disease at the southern end of the species’ range (Wahle et al., 2009). Previous studies have also indicated that lobster recruitment is derived from local sources in the GOM and the stock–recruitment relationships differ between eastern and western GOM (Incze et al., 2010). This is a result of the variation in primary production that is affected by different temperatures and circulation patterns across these two areas (Chang et al., 2015). Given the differences in the eastern and western GOM, it is important to evaluate the spatial non-stationary environmental effects on the lobster distribution in this area. There is a regional decrease in the degree of vertical mixing from the eastern to the western GOM lobsters can be exposed to a wide range of bottom temperatures. In the western GOM, the bottom water temperature is below the lower bound of ideal temperature range (i.e. 12–18°C) for lobsters whereas the temperature is within the ideal range in the eastern GOM at same depth stratum besides very shallow nearshore areas in spring and fall (Crossin et al., 1998; Kleisner et al., 2016). Therefore, the depth-wise distribution of lobsters may be expected to vary from east to west during the year because of seasonal differences in the degree of vertical mixing. Other environmental variables, such as salinity and sediment type, may further influence the coastal distribution and density of lobster, and may interact with temperature effects on lobster density (Jury et al., 1994). Previous studies suggest that recent increases of lobsters in the GOM are related to warming bottom temperature (Chang et al., 2010; Tanaka and Chen, 2016; Le Bris et al., 2018), but these increases have not been uniform in space (ASMFC, 2015). The annual mean density of lobsters from the Maine-New Hampshire Bottom Trawl Survey showed that lobsters have increased considerably more in the eastern GOM than in the west (Sherman et al., 2015). The magnitude of environmental effects from two different temperature conditions may result in this density difference. It is important to understand how densities of lobsters change over space under different environmental conditions. Descriptions of the varied temperature-presence and temperature-density relationships over space would provide managers and fishermen greater insight into the role of temperature on the expansion of lobsters into the eastern GOM. Regression models are the most common technique used to evaluate statistical relationships between species abundance and environmental variables (Windle et al., 2010; Tseng et al., 2013; Ward et al., 2015). However, global regression models (e.g. generalized linear regression model) estimate only a single relationship between environmental variables and species with the assumption of spatial stationarity over large spatial scales. This creates a challenge in understanding complex local patterns (Cadenasso et al., 2007; Hoeting, 2009; Windle et al., 2010). Given the reality that the impact of environmental variables on the distribution and abundance of a species may differ in intensity in different areas of the species range, local models may better characterize spatially varying relationships between abundance and environmental variables (Brunsdon et al., 1998; Fotheringham and Brunsdon, 1999; Franklin and Miller, 2009; Windle et al., 2010; Tseng et al., 2013; Runge et al., 2014). There are several approaches which can improve our understanding of non-stationary effects of environmental variables on distribution of a species, such as linear mixed models and geographically weighted regression (GWR) models. Both linear mixed models and GWR models are able to provide geographical varied intercepts and species-environment relationships (Franklin and Miller, 2009). Linear mixed models incorporate random parameters and model the variability detected for a given phenomenon among different locations (Thorson et al., 2015). The GWR model estimates intercepts and coefficients of each predictor variable at each observation point (Fotheringham and Brunsdon, 1999). It shows promise in verifying spatial variability of environmental effects on species distribution and identifying dominant environmental driver at potential subarea. Previous studies also suggested that the GWR could make relatively good predictions (Zhang and Gove, 2005). For example, Windle et al. (2010) compared several models in predicting presence of cod in the North Atlantic Ocean. The results showed that the GWR produced more accurate predictions and less spatial pattern in the residuals compared with global logistic regression and generalized additive model (GAM). In this study, we conduct a comprehensive GWR model framework to analyse the local relationships between environmental variables and presence and density of American lobster in the GOM. To detect locally varied relationships between lobsters and environmental variables in the study area, we developed a two-stage season-, sex-, and size-specific model implementing the GWR approach to explore presence and density distribution of lobsters. We also conducted a simulation approach to examine whether the GWR model under- or overestimates lobster density with spatial patterns. The developed analytical framework is suitable for testing the non-stationary environmental effects on the distribution of other species. Methods Study area and data sources The lobster density data were collected from the Maine-New Hampshire Inshore Bottom Trawl Survey. This biannual survey has been conducted in the coastal waters of Maine and New Hampshire since fall 2000. The survey area includes 16 001 km2 of coastal shelf from Downeast Maine to southern Maine and New Hampshire (Figure 1; Sherman et al., 2005). This stratified random survey has a target number of 115 stations for each survey resulting in a sampling density of one station for every 137 km2 (Sherman et al., 2005). However, the actual number of surveyed stations is smaller than 115 for various reasons (Chen et al., 2006). The target tow duration is 20 minutes covering a length of approximately 1.48 km. Data from 261 928 individual lobsters were included in this study. Figure 1. View largeDownload slide Catch density (log number of lobsters/0.016 km2; top panel) of the Maine-New Hampshire Inshore Bottom Trawl Survey and observed bottom water temperatures (bottom panel) from 2000 to 2014. The star symbol in the US map shows the location of the GOM. Figure 1. View largeDownload slide Catch density (log number of lobsters/0.016 km2; top panel) of the Maine-New Hampshire Inshore Bottom Trawl Survey and observed bottom water temperatures (bottom panel) from 2000 to 2014. The star symbol in the US map shows the location of the GOM. Lobster biological data were collected from 2000 to 2014. At each survey station, the carapace length (CL), sex, and weight of each individual lobster were measured. Lobster catch per tow was standardized according to tow distance by dividing catch quantity with tow distance then multiplying target tow distance. The standardized lobster densities, measured as the number of lobster per 0.016 km2 (ASMFC, 2015), were grouped by seasons (i.e. fall and spring), sexes (i.e. female and male), and two size classes (i.e. juvenile and adult). Juveniles were lobsters <50 mm CL; this classification is based on the differences in activity patterns (Lawton and Lavalli, 1995). Lobsters <50 mm CL show limited movement, whereas adult lobsters show more active seasonal movement (Lawton and Lavalli, 1995). A total of eight groups (two seasons × two sexes × two sizes) of data were modelled independently. Environmental data and spatial information, such as bottom water temperature (°C, Figure 1), bottom water salinity, depth (m), latitude (decimal degree), and longitude (decimal degree), were collected during the survey. Distance offshore (km), which is defined as shortest distance from the sampling station to the shore, was calculated by using the gdistance package in R (Etten, 2015). Sediment mean grain size (phi; -log of grain size) was obtained from the US Geographical Survey East-Coast Sediment Texture Database (McMullen et al., 2014). The average value of grain size in units of phi was estimated for each survey station using ArcGIS interpolation with kriging. Grain size decreases with increasing phi values, and the range of grain size was from 5.93 to 10.62 phi with a resolution of 0.01 phi. Model development Two-stage GWR models were used to evaluate the relationships between lobster density and environmental variables. The first stage GWR was used to estimate the probability of lobster presence (p) as a function of environmental variables with a binomial error distribution: GWRI: logitpi=β0Xi, Yi+β1Xi, YiTi+β2Xi, YiSi+β3Xi, YiDei+β4Xi, YiDOi+β5Xi, YiSei. The second stage GWR estimated the log-transformed lobster density (d) with a Gaussian error distribution: GWRII:lndi=β0Xi, Yi+β1Xi, YiTi+β2Xi, YiSi+β3(Xi, Yi)Dei+β4(Xi, Yi)DOi+β5(Xi, Yi)Sei, where p is the probability of presence of lobster at location i, d is the density of lobster at location i, β0 is the intercept specific to location i, and ( Xi, Yi) is the coordinate of the ith location. β1Xi, Yi to β5(Xi, Yi) are coefficients of independent variables varying conditional on location i. Ti, Si, Dei, DOi, and Sei, are bottom water temperature (°C), salinity, depth (m), distance offshore (km), and mean grain size of sediment (phi) at location i, respectively. A preliminary variance inflation factor (VIF) analysis was conducted to remove variables with multicollinearity. We excluded environmental variables with VIFs that exceeded 3 for each model (Sagarese et al., 2014). Based on VIF results and p-value (<0.05) from the generalized linear regression model using all the data, temperature, distance offshore, and sediment size were included for all the GWR models. Salinity was included in the first stage male adult and spring male juvenile GWR models. Depth was only included for second stage fall female juvenile GWR model. The input variables were centred on zero because the interpretation of regression coefficients is often sensitive to the scale of the input variables (Franklin and Miller, 2009). The GWR models were conducted applying a generalized linear regression model at each survey station using data from nearest stations with a defined weight. With the moving generalized linear regression model going through each station, location specific intercepts, coefficients, and significances of each explanatory variable (p < 0.05) were determined. The weighting matrix was calculated based on type of distance, kernel function, and bandwidth (Gollini et al., 2014; Nakaya, 2014). The final weighting matrix was determined based on model optimization criteria such as the smallest corrected Akaike’s Information Criterion (AICc) value. The great circle distance between spatial coordinates was used in this study to calculate shortest distance between two points with consideration of the curvature of the Earth (Gollini et al., 2014). A bi-square kernel function was applied by giving a unit weight to each sample point, but null weights to observations with a distance greater than the bandwidth (Gollini et al., 2014). Adaptive bandwidth with a fixed number of points was selected to establish the weight matrix after model optimization (Nakaya, 2014). Golden section search, which is an efficient optimization tool for locating the maximum or minimum of a function by searching between potential bandwidth intervals, was used in order to find optimal bandwidth: wij=(1-dij2/bi(k)2)2 if dij<bi(k)0  otherwise, where wij is the weight value of the observation at station j for estimating the coefficient at station i, dij is the great circle distance between stations i and j, and bi(k) is the adaptive bandwidth size defined as the kth nearest neighbour point. The bandwidth of the model that produced the smallest corrected AICc value was determined as the optimal bandwidth. The GWR analysis was conducted using GWR 4.0 software. Model fitting and validation Area under the receiver operating characteristic curve (AUC) was used to verify the abilities of the first stage GWR models to fit presence of lobsters. The AUC values range from 0 to 1, and a high AUC value implies that model has a high probability in fitting the presence of lobsters correctly (Zou et al., 2007). Root mean square error (RMSE) and Moran’s I were used to assess the mode fitting performance of the second stage GWR models. RMSE was calculated to quantify the discrepancy between observed and fitted densities and a value close to 0 indicates better model fit (Stow et al., 2009). Moran’s I statistic evaluates whether the pattern of model residuals expressed is clustered, dispersed, or random (Lu et al., 2014). A Moran’s I value near 1 indicates strong positive autocorrelation of the residuals, and a value near −1 indicates strong negative autocorrelation (Windle et al., 2010). The eight groups’ data for each model were divided into training and testing data to calibrate the model and validate the predictions. Partitioning of training and testing data varies between different models. The proportion of testing data for each model was 1/(1+p-1), where p is the number of predictor variables (Franklin and Miller, 2009). Presence and density of lobsters at locations of testing data were predicted based on the model developed using training data. The AUC values were used to evaluate the discrepancy between observed and predicted presence. The adjusted R2s from linear regression model were used to measure the similarity between observed and predicted density. We repeated the cross validation 100 times for each GWR model and averaged the estimated performance measures. Results Model performance and validation The bi-square kernel bandwidth adapted itself in size depending on the original data density (Table 1). For example, the presence of adults showed a larger spatial coverage in the GOM than juveniles. With few absence values, adults required a larger bandwidth size than juveniles to develop the GWRI model; therefore, the GWRI models for juveniles showed better performances than models for adults with higher deviance explained and AUC values (Table 1). The performance results from GWRII models indicated that the models for adults fitted better in explaining the variation of the density because of the lower RMSE values (Table 1). The mean of 100 cross validation results from the simple linear regression analysis showed that the models had reasonable prediction skill because the fitting lines were close to the 1:1 line (Supplementary Figure S1). The global Moran’s I ranged from −0.04 to 0.09 and indicated weak autocorrelation of model residuals for all models (Supplementary Figure S2). The estimated coefficients were not highly correlated with observed environmental variables (Supplementary Table S3). Table 1. Summary of optimal bandwidth and model perfomance for the GWR models. Model Sample Size Bandwidth Deviance (%) AUC CV_AUC ±SD FLFJ I 1059 161 42.77 0.92 0.90 ± 0.05 FLFA I 1059 377 34.10 0.90 0.87 ± 0.10 FLMJ I 1059 131 40.67 0.91 0.89 ± 0.05 FLMA I 1059 495 37.42 0.92 0.67 ± 0.13 SPFJ I 1406 156 34.19 0.90 0.88 ± 0.05 SPFA I 1406 445 20.71 0.83 0.80 ± 0.14 SPMJ I 1406 162 38.26 0.92 0.75 ± 0.07 SPMA I 1406 412 34.84 0.91 0.66 ± 0.12 Model Sample Size Bandwidth RMSE R2 CV_R2 ± SD FLFJ II 561 64 0.37 0.43 0.39 ± 0.12 FLFA II 976 48 0.85 0.54 0.58 ± 0.12 FLMJ II 572 53 0.47 0.41 0.42 ± 0.11 FLMA II 957 48 1.04 0.51 0.60 ± 0.11 SPFJ II 676 58 0.63 0.34 0.37 ± 0.11 SPFA II 1316 62 0.70 0.43 0.57 ± 0.10 SPMJ II 670 65 0.58 0.30 0.30 ± 0.12 SPMA II 1259 65 0.76 0.43 0.55 ± 0.10 Model Sample Size Bandwidth Deviance (%) AUC CV_AUC ±SD FLFJ I 1059 161 42.77 0.92 0.90 ± 0.05 FLFA I 1059 377 34.10 0.90 0.87 ± 0.10 FLMJ I 1059 131 40.67 0.91 0.89 ± 0.05 FLMA I 1059 495 37.42 0.92 0.67 ± 0.13 SPFJ I 1406 156 34.19 0.90 0.88 ± 0.05 SPFA I 1406 445 20.71 0.83 0.80 ± 0.14 SPMJ I 1406 162 38.26 0.92 0.75 ± 0.07 SPMA I 1406 412 34.84 0.91 0.66 ± 0.12 Model Sample Size Bandwidth RMSE R2 CV_R2 ± SD FLFJ II 561 64 0.37 0.43 0.39 ± 0.12 FLFA II 976 48 0.85 0.54 0.58 ± 0.12 FLMJ II 572 53 0.47 0.41 0.42 ± 0.11 FLMA II 957 48 1.04 0.51 0.60 ± 0.11 SPFJ II 676 58 0.63 0.34 0.37 ± 0.11 SPFA II 1316 62 0.70 0.43 0.57 ± 0.10 SPMJ II 670 65 0.58 0.30 0.30 ± 0.12 SPMA II 1259 65 0.76 0.43 0.55 ± 0.10 The unit of bandwidth is number of points. Table 1. Summary of optimal bandwidth and model perfomance for the GWR models. Model Sample Size Bandwidth Deviance (%) AUC CV_AUC ±SD FLFJ I 1059 161 42.77 0.92 0.90 ± 0.05 FLFA I 1059 377 34.10 0.90 0.87 ± 0.10 FLMJ I 1059 131 40.67 0.91 0.89 ± 0.05 FLMA I 1059 495 37.42 0.92 0.67 ± 0.13 SPFJ I 1406 156 34.19 0.90 0.88 ± 0.05 SPFA I 1406 445 20.71 0.83 0.80 ± 0.14 SPMJ I 1406 162 38.26 0.92 0.75 ± 0.07 SPMA I 1406 412 34.84 0.91 0.66 ± 0.12 Model Sample Size Bandwidth RMSE R2 CV_R2 ± SD FLFJ II 561 64 0.37 0.43 0.39 ± 0.12 FLFA II 976 48 0.85 0.54 0.58 ± 0.12 FLMJ II 572 53 0.47 0.41 0.42 ± 0.11 FLMA II 957 48 1.04 0.51 0.60 ± 0.11 SPFJ II 676 58 0.63 0.34 0.37 ± 0.11 SPFA II 1316 62 0.70 0.43 0.57 ± 0.10 SPMJ II 670 65 0.58 0.30 0.30 ± 0.12 SPMA II 1259 65 0.76 0.43 0.55 ± 0.10 Model Sample Size Bandwidth Deviance (%) AUC CV_AUC ±SD FLFJ I 1059 161 42.77 0.92 0.90 ± 0.05 FLFA I 1059 377 34.10 0.90 0.87 ± 0.10 FLMJ I 1059 131 40.67 0.91 0.89 ± 0.05 FLMA I 1059 495 37.42 0.92 0.67 ± 0.13 SPFJ I 1406 156 34.19 0.90 0.88 ± 0.05 SPFA I 1406 445 20.71 0.83 0.80 ± 0.14 SPMJ I 1406 162 38.26 0.92 0.75 ± 0.07 SPMA I 1406 412 34.84 0.91 0.66 ± 0.12 Model Sample Size Bandwidth RMSE R2 CV_R2 ± SD FLFJ II 561 64 0.37 0.43 0.39 ± 0.12 FLFA II 976 48 0.85 0.54 0.58 ± 0.12 FLMJ II 572 53 0.47 0.41 0.42 ± 0.11 FLMA II 957 48 1.04 0.51 0.60 ± 0.11 SPFJ II 676 58 0.63 0.34 0.37 ± 0.11 SPFA II 1316 62 0.70 0.43 0.57 ± 0.10 SPMJ II 670 65 0.58 0.30 0.30 ± 0.12 SPMA II 1259 65 0.76 0.43 0.55 ± 0.10 The unit of bandwidth is number of points. Environmental and spatial variables Non-stationary environmental effects on distribution of lobsters were visually explored by mapping the local coefficient estimates of each predictor variable. Overall, there was a positive relationship between bottom water temperature and presence of lobsters. The coefficients between presence of lobsters and bottom water temperature varied from −0.44 to 2.97. Positive relationships between presence and bottom water temperature were significant in the eastern GOM, while the western GOM showed non-significant relationships for most of the modelling groups (Figure 2). Near the Mid-Coast region of the GOM, presence of spring female adults showed significant positive relationships with bottom water temperature. However, this relationship was not significant in fall. The significant positive relationships between lobster density and bottom water temperature were more spread throughout the study area (Figure 2). The coefficients for the relationship between lobster density and bottom water temperature ranged from −0.42 to 1.30. When considering different seasons, sex and size of lobsters, there was no large difference in coefficients in GWRII relationship patterns. Figure 2. View largeDownload slide Local coefficient estimates derived from the GWR models for bottom water temperature. The circles with dots denoted that the bottom water temperature were not significant to the models at these locations. FLFJ I: FL denotes fall, F denotes female, J denotes juvenile, and I denotes first stage model. SPFA II: SP denotes spring, F denotes female, A denotes adult, and II denotes second stage model. Figure 2. View largeDownload slide Local coefficient estimates derived from the GWR models for bottom water temperature. The circles with dots denoted that the bottom water temperature were not significant to the models at these locations. FLFJ I: FL denotes fall, F denotes female, J denotes juvenile, and I denotes first stage model. SPFA II: SP denotes spring, F denotes female, A denotes adult, and II denotes second stage model. The estimated coefficients of the distance offshore varied from −0.81 to 0.34 and −0.64 to 0.88 for GWRI and GWRII, respectively. The negative relationships between the presence of lobster and distance offshore were stronger for juveniles than for adults (Figure 3). The mean sediment size (phi) displayed significant negative relationships with presence of lobsters at most stations. However, there was a cluster of significantly positive relationships between spring adult lobster and sediment size in the Downeast Maine region. The positive relationships indicated that the probability of presence of lobster increases with finer sediment since large phi values indicate finer sediment. In the relationships between presence of lobster and sediment size, the magnitude of the coefficients ranged from −2.80 to 1.99. Lobster density showed patterns of non-significant relationships with sediment size and the range of relationship coefficients was from −5.03 to 2.24 (Supplementary Figure S4). There was no large difference in sediment coefficients among seasons and sexes. The estimated intercepts were the mean of the response when all predictors were zeros and they were also varied over space with ranges of −2.8 to 8.81 and −0.28 to 13.36 for GWRI and GWRII respectively (Supplementary Figure S5). The presence of lobsters showed a lower probability of occurrence outside of the Penobscot Bay compared with the eastern and western GOM and the density of lobsters slightly decreased with increased distance offshore. Figure 3. View largeDownload slide Boxplot of coefficients from distance offshore (top panel) and sediment (bottom panel) at different regions. The regions from 1 to 5 represent New Hampshire and Southern Maine, Mid-coast, Penobscot Bay, Mt Desert Area, and Downeast Maine region respectively. SPFJ I denots presence model with spring female juveniles and SPFA denotes presence model with spring female adults. Figure 3. View largeDownload slide Boxplot of coefficients from distance offshore (top panel) and sediment (bottom panel) at different regions. The regions from 1 to 5 represent New Hampshire and Southern Maine, Mid-coast, Penobscot Bay, Mt Desert Area, and Downeast Maine region respectively. SPFJ I denots presence model with spring female juveniles and SPFA denotes presence model with spring female adults. Salinity and depth were only included in a few models. The relationships between presence of lobster and salinity were significant in the western GOM in fall and in the eastern GOM in spring. Most of the salinity coefficients were negative and as large as −5.06 (Figure 4). In spring, juvenile lobster presence showed a similar pattern with adult lobster presence. Depth showed a weak relationship (−0.07 to 0.08) with the density of female juveniles in fall (Figure 5). The most significant positive relationship between lobster density and depth was in Mid-Coast Maine, and a negative relationship appeared in Mt Desert outer inshore areas. Figure 4. View largeDownload slide Local coefficient estimates derived from the GWR models for salinity. The circles with dots denoted that the salinity were not significant to the models at these locations. FLMA denotes fall male adult, SPMJ denotes spring male juvenile, SPMA denotes spring male adult, and I denotes first stage of model. Figure 4. View largeDownload slide Local coefficient estimates derived from the GWR models for salinity. The circles with dots denoted that the salinity were not significant to the models at these locations. FLMA denotes fall male adult, SPMJ denotes spring male juvenile, SPMA denotes spring male adult, and I denotes first stage of model. Figure 5. View largeDownload slide Local coefficient estimates derived from the GWR models for depth. The circles with dots denoted that the salinity were not significant to the models at these locations. FLFJ I denotes first stage of model for fall female juveniles. Figure 5. View largeDownload slide Local coefficient estimates derived from the GWR models for depth. The circles with dots denoted that the salinity were not significant to the models at these locations. FLFJ I denotes first stage of model for fall female juveniles. Discussion We developed a modelling approach for understanding the local relationships between environmental variables that influence season-, size-, and sex-specific distribution of American lobster in the GOM. Bottom water temperature, distance offshore, and sediment size were the key variables that affect the spatial distribution of American lobster. The relationships between lobster distribution and these environmental variables varied locally over the GOM. One implication of rapidly changing coefficients in space is that the strength of environmental effects on lobster distribution was non-stationary because of the spatial variations of the environmental variables, showing the importance of spatial scale in studying interactions between lobster distribution and environmental variables. In a previous study, the temperature has long been recognized to be a key determinant of the distribution and abundance of American lobster using global models (Chang et al., 2010; Tanaka and Chen, 2016). In this study, the GWR models with finer spatial scale showed that the temperature had a significantly positive effect on lobster distribution but the strength of temperature effects varied spatially in the GOM. The relationships were significantly positive in the eastern GOM while most of the area in the western GOM showed non-significant relationships. In addition to the non-significant relationships found in this study, the influence of temperature on lobster presence and density may be obscured by other factors in the western GOM. It is possible the estimated relationships differed between the western and eastern GOM because there is limited water in the favourable temperature range in the western GOM. A thermal front separates the cold, vertically stratified water in the western GOM from the warm, tidally mixed, bottom water in the eastern GOM (Townsend et al., 2006). Thus, except for the shallowest, nearshore trawl sites, most of the bottom water temperatures in the western GOM were below the ideal temperature range for lobsters, whereas the temperatures at sites in the eastern GOM were above the range. Although the warm stratified layer is not well established in the west at time of the spring trawl survey, it is more conspicuous in the fall survey, when it included depth strata shallower than 37 m (Supplementary Figures S6 and S7). With limited favourable temperature in the western GOM at the seasons of sampling, there was no statistically significant positive relationship between lobster and temperature. The fluctuations of water temperature showed a negligible effect on the presence and density of lobsters. The non-significant relationship between temperature and presence and density of lobsters indicates that other factors (e.g. predators) or a combination of temperature and other factors may influence the lobster population in this region. The relative influence of the predators in regulating the lobster population may intensify at thermal range boundaries (Boudreau et al., 2015). Distance offshore might be one of the useful variables to predict the lobster distribution in the western GOM. In the western GOM, some relationships between distance offshore and lobster were significantly negative. The rate of decrease in juvenile presence was faster than that of adults given the same value of distance offshore. Juveniles exhibit less mobility than adults (Lawton and Lavalli, 1995; Wahle et al., 2013a); thus, the distance offshore is a better predictor for juveniles than adults, because the limited movement ability of juveniles restricts them inshore where they have been found to settle at highest density (Wahle et al., 2013a). In addition to distance offshore, sediment size was also an important variable showing spatial varying relationships with presence and density of lobsters. The presence of adult lobster in spring showed both positive and negative relationships with sediment size in different regions. The sediment size has no large distance offshore-wise gradient in the Downeast Maine region according to the data from the US Geographical Survey East-Coast Sediment Texture Database (McMullen et al., 2014). The sediment size in other areas gradually changes from coarse to fine with increasing distance offshore. With more options in terms of sediment types, the presence and density of lobsters decreased with increasing sediment size in regions other than the Downeast Maine region. This pattern was not detected in previous studies (Chang et al., 2010) and suggests that the adult lobster is less restricted to the coarse, rocky, cobble sediment found in the Downeast Maine region. Alternatively, coarse substrate such as boulders and cobble may limit trawl sampling efficiency. The complex relationships between lobster and sediment size require further exploration as catchability of lobsters by the trawl survey may explain some of the spatial variability in apparent lobster density (Somerton et al., 2013). Lobsters often prefer boulder and rocky substrates (Steneck and Wilson, 2001), but the trawl survey generally showed higher lobster catches in clay and silt areas. The trawl survey in general has poor catch efficiency at those rocky substrates (Chang et al., 2010); therefore, the model might underestimate the importance of sediment related variables. In addition to sediment, the variation in estimated temperature coefficients might be induced by varied catchability at different temperatures. The contrast of temperatures in the western and eastern GOM might cause differences in catchability and therefore, likely affects the explanatory power of temperature in the models from these regions. All the environmental variables examined in this study are abiotic, and we have not investigated any biotic variables that can impact the distribution of American lobster. Boudreau et al. (2015) suggested that both biotic and abiotic effects could affect the abundance of American lobster in the northwest Atlantic Ocean by altering lobster interactions with predators and abiotic environmental variables. Given the non-significant relationship between lobster and temperature in the western GOM, incorporating biotic variables such as the abundance of cod, a predator of the American lobster, into the GWR model may further explain the spatial dynamics in the western GOM. The abundance of cod is higher in the western GOM than in the eastern GOM, which is the opposite pattern of the density of American lobster (NEFSC, 2013; Wahle et al., 2013b). The model validation results suggest low spatial autocorrelation of model residuals, which is consistent with the results of previous GWR model studies (Zhang and Gove, 2005; Windle et al., 2010; Lu et al., 2014). Although global models often assume independence between observations and model residuals, the GWR model assumes and quantifies spatial dependence between observed variables. The lack of spatial patterns in model residuals suggests relative good performance of the GWR models. The GWR model is particularly useful when the target species tends to have a patchy distribution which requires consideration of spatial dependence. Although using a GWR model to predict the spatial distribution of a marine species has many advantages, GWR models have limitations. Coordinates were the only information required by the GWR model for coefficients prediction at unobserved locations. Thus, the estimated parameters are not suitable to predict future distribution of a species if there are substantial changes in ocean conditions. Like Chang et al. (2010), we assumed that the lobster behavioural response to environmental gradients did not vary much between years. Year-to-year variations in lobster density could be used to set up time blocks to explore the distribution under different density conditions. However, data could not be divided into several time blocks since the lobster density from the trawl survey did not have an obvious shift within the study period (2000–2014). Shifts in estimated relationships can be assumed if low- and high-density periods exist and if there is a density dependent behaviour of lobsters. In addition to the difficulty in adding year as a factor into the models, the GWR model may need to be carefully examined if it is used to conduct annual analyses. The AUC values from the model validation results were lower than the fitting results with an average decrease of 10.83%. This decline is a result of the model being developed with training data, which has fewer data points than the full dataset; therefore, a model developed with a single year’s data may not be sufficient to produce ecologically interpretable results. GWR model, compared with global models (e.g. GAM), may have less flexibility in forecasting presence or density of species if there is a sudden change in the environment. The large data quantity required to estimate locally varied relationships constrain the application of the GWR model. Windle et al. (2010) and Lu et al. (2014) suggested using the GWR model and global regression model together to better understand how the distribution of a species varies with different factors in a large ecosystem. The GWR model has the ability of producing results that are as accurate as the more frequently used global regression models (Zhang et al., 2008). In addition to the similar performance in prediction, the GWR model can explore spatial non-stationarity of environmental influences at various scales. The GWR can be used as an identifier to determine the spatial scale relationships between species and environmental variables become stationary (Segovia et al., 2016). Thus, the GWR model can first be used to detect the non-stationarity scale, and this defined scale could be used to divide the study area into subareas. Using both GWR and global models can further improve the hindcast or forecast of the species distribution outside of the temporal range of existing data. Furthermore, the estimated coefficients maps can clearly reveal heterogeneity of the relationships throughout the study area and facilitate interpretation of model results (Segovia et al., 2016). The GWR model is also able to predict coefficients/relationships at unsampled locations without additional measurements. Non-stationary relationships between species and environmental variables have several important implications for management. First, a dominant driver in different subareas can be identified by comparing the magnitudes of the estimated relationships. With a careful examination of the biological mechanism underlying the estimated relationship, a more specific monitoring program can be developed by focussing on sampling data for the identified dominant drivers (e.g. predators). In addition to identify dominant drivers at subareas, this analysis will facilitate derivation of spatially explicit species abundance indices if the species is managed under multiple management areas. Furthermore, the abundance indices can be simulated under various climate scenarios to inform temperature-explicit conservation plans. Improved conservation of the American lobster requires locally informed management plans to better serve the region’s economically and culturally important coastal fishing communities. Fishing communities along the coast may be at a risk if the local lobster density starts to decline with a change in environment. In addition, variation in timing and location of fishing may lead to an increase of variation in the temporal trends of lobster density in different areas. Regarding the different trends of lobster density, each of the seven Maine lobster management zones needs to develop a local management plan to respond to the different challenges it may face. For example, the management zones in the eastern GOM may have laxer management rules because the fishery is likely to be more abundant if temperature continues to rise within the favourable temperature range of lobsters. On the other hand, zones in the western GOM may need more information on the impact of predation on the fishery before relaxing their management rules. In conclusion, this modelling approach can be applied to a wide suite of species and has important management implications. The developed GWR models in this study provided details of the relationships between presence or density of a species and other environmental variables. Coefficient maps enable the importance and significance of each environmental variable at a specific location to be easily assessed. The improved performance of using local models compared with global models highlights the limitations of using only a global model to study the distribution of a species in a large marine ecosystem and provides insights in managing a species by subareas. Future studies need to explore the influences of interactions among abiotic and biotic variables (e.g. the density of predators) on the non-stationary environmental effects on the distribution of a species to achieve better estimation of the lobster population. Acknowledgements This work was funded by Research Reinvestment Fund Graduate Assistantship from the University of Maine. We thank Mr Carl Wilson, Ms Kathleen Reardon, Ms Katherine Thompson, and Ms Sally Sherman from the Maine Department of Marine Resources for providing Maine-New Hampshire Inshore Bottom Trawl Survey data and valuable comments. Also, we would like to thank Kevin Staples, Kisei Tanaka, and Dongyan Han for their help and discussions in interpreting American lobster distribution and environmental data. Supplementary data Supplementary material is available at the ICESJMS online version of the manuscript. References Atlantic States Marine Fisheries Commission (ASMFC), 2015. American lobster stock assessment peer review report. Stock Assessment Report No. 09-01 (Supplement). Woods Hole, Massachusetts. 493 pp. http://www.asmfc.org/uploads/file//55d61d73AmLobsterStockAssmt_PeerReviewReport_Aug2015_red2.pdf (last accessed 10 December 2017). Boudreau S. A. , Anderson S. C. , Worm B. 2015 . Top-down and bottom-up forces interact at thermal range extremes on American lobster . Journal of Animal Ecology , 84 : 840 – 850 . Google Scholar CrossRef Search ADS PubMed Brunsdon C. , Fotheringham S. , Charlton M. 1998 . Geographically weighted regression modelling spatial non-stationarity . Journal of the Royal Statistical Society , 47 : 431 – 443 . Cadenasso M. L. , Pickett S. T. , Schwarz K. 2007 . Spatial heterogeneity in urban ecosystems: reconceptualizing land cover and a framework for classification . Frontier in Ecology and the Environment , 5 : 80 – 88 . Google Scholar CrossRef Search ADS Chang J.-H. , Chen Y. , Holland D. , Grabowski J. 2010 . Estimating spatial distribution of American lobster Homarus americanus using habitat variables . Marine Ecology Progress Series , 420 : 145 – 156 . Google Scholar CrossRef Search ADS Chang J.-H. , Chen Y. , Halteman W. , Wilson C. 2015 . Roles of spatial scale in quantifying stock-recruiment relatinships for American lobsters in the inshore Gulf of Maine . Canadian Journal of Fisheries and Aquatic Sciences , 73 : 885 – 909 . Google Scholar CrossRef Search ADS Chen Y. , Sherman S. , Wilson C. , Sowles J. , Minoru K. 2006 . A comparison of two-fishery-independent survey programs used to define the population structure of American lobster (Homarus americanus) in the Gulf of Maine . Fishery Bulletin , 104 : 247 – 255 . Ciannelli L. , Bartolino V. , Chan K. S. 2012 . Non-additive and non-stationary properties in the spatial distribution of a large marine fish population . Proceedings of the Royal Society B , 279 : 3635 – 3642 . Google Scholar CrossRef Search ADS PubMed Crossin G. , Al-Ayoub S. , Jury S. , Howell W. H. , Watton W. III . 1998 . Behavioral thermaoregulation in the American lobster Hormarus americanus . Journal of Experimental Biology , 201 : 365 – 374 . Google Scholar PubMed Etten J. V. 2015 . Gdistance: distances and routes on geographical grids. R package version 1.1-9. http://CRAN.R-project.org/package=gdistance (last accessed 10 December 2017). Fotheringham A. S. , Brunsdon C. 2010 . Local forms of spatial analysis . Geographical Analysis , 31 : 340 – 358 . Google Scholar CrossRef Search ADS Frank K. T. , Petrie B. , Shackell N. I. , Choi J. S. 2006 . Reconciling differences in trophic control in mid-latitude marine ecosystems . Ecology Letters , 9 : 1096 – 1150 . Google Scholar CrossRef Search ADS PubMed Franklin J. , Miller J. , 2009 . Mapping Species Distributions: Spatial Inference and Prediction . Cambridge University Press , Cambridge . 212 pp. Gollini I. , Lu B. , Charlton M. , Brunsdon C. , Harris P. , 2014 . GWmodel: an R package for exploring spatial heterogeneity using georaphically weighted regression models. Journal of Statistical Software, 63: 1–50. Hoeting J. A. 2009 . The importance of accounting for spatial and temporal correlation in analyses of ecological data . Ecological Applications , 19 : 574 – 577 . Google Scholar CrossRef Search ADS PubMed Incze L. , Xue H. , Wolff N. , Xu D. , Wilson C. , Steneck R. , Wahle R. , Lawton P. , Pettigrew N. , Chen Y. 2010 . Connectivity of lobster populations in the coastal Gulf of Maine. Part II. Coupled biophysical dynamics . Fisheries Oceanography , 19 : 1 – 20 . Google Scholar CrossRef Search ADS Jury S. H. , Kinnison M. T. , Howell W. H. , Watson W. H. 1994 . The behavior of lobsters in response to reduced salinity . Journal of Experimental Marine Biology and Ecology , 180 : 23 – 37 . Google Scholar CrossRef Search ADS Kleisner K. M. , Fogarty M. J. , McGee S. , Barnett A. , Fratantoni P. , Greene J. , Hare J. A. , Lucey S. M. , McGuire C. , Odell J. , Saba V. S. , Smith L. , Weaver K. J. , Pinsky M. L. 2016 . The effect of sub-regional climate velocity on the distribution and spatial extent of marine species assemblages . PloS One , 11 : e0149220. Google Scholar CrossRef Search ADS PubMed Kotwicki S. , Lauth R. R. 2013 . Detecting temporal trends and environmentally-driven changes in the spatial distribution of groundfishes and crabs on the eastern Bering Sea Shelf . Deep Sea Research II: Topical Studies in Oceanography , 94 : 231 – 243 . Google Scholar CrossRef Search ADS Lawton P. , Lavalli K. L. , 1995 . Postlarval, juvenile, adolescent and adult ecology. In Biology of the Lobster, Homarus Americanus , pp. 120 – 122 . Ed. by Factor J. R. Academic Press, Inc . Le Bris A. , Mills K. E. , Wahle R. A. , Chen Y. , Alexander M. A. , Allyn A. J. , Schuetz J. G. , Scott J. D. , Pershing A. J. 2018 . Climate vulnerability and resilience in the most valuable North American fishery . Proceedings of the National Academy of Sciences of the United States of America , 115 : 1831 – 1831-6 . Google Scholar CrossRef Search ADS PubMed Lu B. , Charlton M. , Harris P. , Fotheringham A. S. 2014 . Geographically Weighted Regression with a non-euclidean distance metric: a case study using hedonic house price data . International Journal of Geographical Information Science , 28 : 660 – 681 . Google Scholar CrossRef Search ADS McMullen K. Y. , Paskevich V. F. , Poppe L. J. , 2014 . GIS data catalog (ver. 3.0, November 2014). In USGS east-coast sediment analysis: Procedures, database, and GIS data, U.S. Geological Survey Open-File Report 2005-1001. Ed. by Poppe, L. J., McMullen, K. Y., Williams, S. J., and Paskevich, V. F. http://pubs.usgs.gov/of/2005/1001/ (last accessed 10 December 2017). Nakaya T. 2014 . GWR4 User Manual: Windows application for Geographically Weighted Refression Modelling. 40 pp. http://gwrtools.github.io (last accessed 10 December 2017). Northeast Fisheries Science Center (NEFSC). 2013. 55th Northeast Regional Stock Assessment Workshop (55th SAW) Assessment Report. US Dept Commer, Northeast Fish Sci Cent Ref Doc. 13-11; 845 p. Available from: National Marine Fisheries Service, 166 Water Street, Woods Hole, MA 02543-1026, or online at http://www.nefsc.noaa.gov/publications/ (last accessed 10 December 2017). Runge C. A. , Martin T. G. , Possingham H. P. , Willis S. G. , Fuller R. A. 2014 . Conserving mobile species . Frontiers in Ecology and the Environment , 12 : 395 – 402 . Google Scholar CrossRef Search ADS Sagarese S. R. , Frisk M. G. , Cerrato R. M. , Sosebee K. A. , Musick J. A. , Rago P. J. 2014 . Application of generalized additive models to examine ontogenetic and seasonal distributions of spiny dogfish (Squalus acanthias) in the Northeast (US) shelf large marine ecosystem . Canadian Journal of Fisheries & Aquatic Sciences , 71 : 847 – 877 . Google Scholar CrossRef Search ADS Segovia M. Q. , Ruiz S. C. , Drapela K. 2016 . Comparison of height-diameter models based on geographically weighted regressions and linear mixed modelling applied to large scale forest inventory data . 25 : 076 . Sherman S. A. , Stepanek K. , Sowles J. , 2005 . Maine – New Hampshire Inshore Groundfish Trawl Survey Procedures and Protocols. 42 pp. http://www.maine.gov/dmr/science-research/projects/trawlsurvey/reports/index.html (last accessed 10 December 2017). Sherman S. A. , Stepanek K. , Pierce F. , Tetrault R. , O’Donnell C. , 2015 . Annual report on the Maine-New Hampshire Inshore Trawl Survey. 128 pp. Somerton, D., Weinberg, K., and Goodman, S. 2013. Catchability of Snow Crab (Chionoecetes opilio) by the eastern Bering Sea bottom trawl survey estimated using a catch comparison experiment. Canadian Journal of Fisheries and Aquatic Sciences, 70: 1699–1708. Steneck R. S. , Wilson C. J. 2001 . Large-scale and long-term spatial and temporal patterns in demography and landings of the American lobster, Homarus Americanus, in Maine . Fisheries Bulletin , 52 : 1303 – 1320 . Stow C. A. , Jolliff J. , McGillicuddy D. J. , Doney S. C. , Allen J. I. , Friedrichs M. A. M. , Rose K. A. , Wallhead P. 2009 . Skill assessment for coupled biological/physical models of marine systems . Journal of Marine Systems , 76 : 4 – 15 . Google Scholar CrossRef Search ADS PubMed Szuwalski C. S. , Hollowed A. B. 2016 . Climate change and non-stationary population processes in fisheries management . ICES Journal of Marine Science , 73 : 1297 – 1305 . Google Scholar CrossRef Search ADS Tanaka K. R. , Chen Y. 2016 . Modeling spatiotemporal variability of the bioclimate envolope of Homarus americanus in the coastal waters of Maine and New Hampshire . Fisheries Research , 177 : 137 – 152 . Google Scholar CrossRef Search ADS Thorson J. T. , Shelton A. O. , Ward E. J. , Skaug H. J. 2015 . Geostatistical delta-generalized linear mixed models improve precision for estimated abundance indices for West Coast groundfishes . ICES Journal of Marine Science , 72: 1297–1310. Townsend D. W. , Thomas A. C. , Mayer L. M. , Thomas M. , Quinlan J. , 2006 . Oceanography of the Northwest Atlantic Continental Shelf (1, W). In The Sea: The Global Coastal Ocean: Interdisciplinary Regional Studies and Syntheses ., Ed. by Robinson A. R. , Brink K. H. , Vol. 14 , pp. 119 – 168 . Harvard University Press , Cambridge, USA . Tseng C.-T. , Su N.-J. , Sun C.-L. , Punt A. E. , Yeh S.-Z. , Liu D.-C. , Su W.-C. 2013 . Spatial and temporal variability of the Pacific saury (Cololabis saira) distribution in the northwestern Pacific Ocean . ICES Journal of Marine Science , 70 : 991 – 999 . Google Scholar CrossRef Search ADS Ward E. J. , Jannot J. E. , Lee Y.-W. , Ono K. , Shelton A. O. , Thorson J. 2015 . Using spatiotemporal species distribution models to identify temporally evolving hotspots of species co-occurance . Ecological Applications , 25 : 2198 – 2209 . Google Scholar CrossRef Search ADS PubMed Wahle R. A. , Gibson M. , Fogarty M. J. 2009 . Distinguishing disease impacts from larval supply effects in a lobster fishery collapse . Marine Ecology Progress Series , 376 : 185 – 192 . Google Scholar CrossRef Search ADS Wahle R. A. , Bergeron C. , Tremblay J. , Wilson C. , Burdett-Coutts V. , Comeau M. , Gibson M. 2013a . The geography and bathymetry of American lobster benthic recruitment as measured by diver-based suction sampling and passive collectors . Marine Biology Research , 9 : 42 – 58 . Google Scholar CrossRef Search ADS Wahle R. A. , Battison A. , Bernatchez L. , Boudreau S. , Castro K. , Grabowski J. H. , Wilson J. 2013b . The American lobster in a changing ecosystem: a US–Canada science symposium, 27–30 November 2012, Portland, Maine . Canadian Journal of Fisheries and Aquatic Sciences , 70 : 1571 – 1575 . Google Scholar CrossRef Search ADS Windle M. J. S. , Rose G. A. , Devillers R. , Fortin M.-J. 2010 . Exploring spatial non-stationarity of fisheries survey data using geographically weighted regression (GWR): an example from the Northwest Atlantic . ICES Journal of Marine Sceince , 67 : 145 – 154 . Google Scholar CrossRef Search ADS Zhang L. , Gove J. H. 2005 . Spatial assessment of model errors from four regression techniques . Forest Sciences , 51 : 334 – 346 . Zhang L. , Ma Z. , Guo L. 2008 . Spatially assessing model errors of four regression techniques for three types of forest stands . Forestry , 81 : 209 – 225 . Google Scholar CrossRef Search ADS Zou K. H. , O’Malley A. J. , Mauri L. 2007 . Receiver-operating characteristic analysis for evaluting diagnostic tests and perdictive models . Circulation , 115 : 654 – 657 . Google Scholar CrossRef Search ADS PubMed © International Council for the Exploration of the Sea 2018. All rights reserved. For permissions, please email: [email protected] This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

doi: 10.1093/icesjms/fsx251pmid: N/A

Abstract In marine fisheries, considerable development has occurred in capture technology. Yet, some of the current fishing methods impact the environment by large greenhouse gas emission, harmful effects to benthic communities, and/or high bycatch of juvenile and unwanted species. It is proposed that for some fish species these deficiencies could be mitigated by classical conditioning using sound and food reward to concentrate wild fish before capture with environmentally friendly fishing gear. Atlantic cod (Gadus morhua), which globally is among the fish species with the highest landed value, can be acoustically trained. In a sea cage, it takes about a week to train a group of naïve cod to associate low frequency (250 Hz) sound with food, whereas the training of a group of naïve cod accompanied with one trained cod takes less than a day. In inshore areas, it takes a few weeks to attract thousands of cod to stations where food is regularly delivered. These conditioned cod wait at the stations for their meals and do not mingle much with the unconditioned cod which hunt for wild prey. It is suggested that by calling acoustically conditioned fish between stations, a much larger number of naïve fish can be gathered. This so-called fish aggregating sound technique (FAST) may thus facilitate the accumulation of wild fish and expedite their capture with a purse seine or a trap in a way that minimizes fuel consumption and mortality of juveniles and unwanted species. The operation of FAST requires exclusive rights of a designated fishing area. The exclusivity makes it possible to on-grow the fish in free-ranging schools and sea cages for several months to increase their size and food quality before capture. Introduction Globally, the annual marine fish catches have been relatively stable around 80 million tons for almost three decades (www.fao.org). Many catching methods are being used and most of them are under continuous development (Gabriel et al., 2005). The five most effective fishing gears during the years 1950–2000 account for more than 90% of the global catch, seine (30–40%), midwater trawl (10–20%), bottom trawl (20%), gillnet (10–20%), and hook and line (10%) (Watson et al., 2006). During these five decades, the relative catch has increased for midwater trawl but decreased for gillnet. Some of the capture methods impact the environment in up to three ways. First is the large fuel consumption and greenhouse gas emission (carbon footprint) required to catch fish, especially with trawls that have to be dragged long distances through water in areas where fish densities are low (Suuronen et al., 2012); second is the harmful effects of some fishing gears to benthic communities, gears like dredges and bottom trawls that crush many species of benthic animals (Løkkeborg, 2005; Grieve et al., 2014; Eigaard et al., 2016; Rijnsdorp et al., 2016); third is the bycatch of undersized, unwanted and in some cases endangered species, which in many cases have low survival chances once they enter the fishing gear, e.g. longline and trawl (Hall et al., 2000; Davies et al., 2009). It is proposed that for some fish species, these deficiencies could be mitigated by classical conditioning using sound and food reward to concentrate wild fish before capture with environmentally friendly fishing gear. One of these species may be Atlantic cod (Gadus morhua). The annual global landings of cod have usually ranged between 1 and 3 million tons (Björnsson et al., 2010c) and in terms of revenue it is considered the most valuable fish species in the North Atlantic (Íslandsbanki, 2013; FAO, 2016). Historically, the cod stocks in Canada were among the largest. They collapsed in 1990s but are recently making a comeback (Rose and Rowe, 2015). In recent years, the largest cod stock has been the Northeast Arctic cod stock with landings commonly ranging between 0.5 and 1.0 million tons (www.ices.dk). The main feeding area of this stock is in the Barents Sea, the catch being shared by Norway and Russia. In recent years, the annual catch of the Icelandic cod stock has ranged between 0.2 and 0.3 million tons and it has been mainly fished with a bottom trawl (40–50%), longline (30%), and gillnet (10%) (Björnsson et al., 2015). Mammals, birds, and fish use social learning to assist in finding food (Zion et al., 2007; Brown and Laland, 2011). As an example, a few seabirds feeding on fish offal from a boat quickly attract large numbers of birds. From a distance of up to several kilometres, using visual and sometimes acoustic stimuli, birds are able to infer from the behaviour of other birds that they have found food (Silverman et al., 2004; Thiebault et al., 2014). Below the sea surface, fish can detect visual stimuli only up to a distance of a few metres to some tens of metres, depending on the light level and clarity of the water (Rose, 1993; Meager and Batty, 2007). Chemical stimuli may have a longer range, e.g. the Atlantic cod has been found to react to a line with baited hooks from a distance of up to 700 m (Løkkeborg, 1998). Acoustic stimuli, especially low-frequency sound signals, can travel longer distances in water than in air (Hawkins, 1993). Fish have directional hearing (Schuijf and Buwalda, 1975; Hawkins and Sand, 1977), are most sensitive to low-frequency sounds within the range of 100–300 Hz (Popper and Carlson, 1998), and can detect loud sounds from a distance of at least several kilometres (Hawkins, 1993). Many species of fish have been trained to associate sound signals with fear (Hawkins and Sand, 1977) or food (Abbott, 1972; Midling et al., 1987; Levin and Levin, 1994). The utility of acoustic training was demonstrated for the first time with fish, when conditioned juvenile red sea bream (Pagurus major) were conditioned to associate sound signals with food before they were released into a small bay in Japan. Three months later, a large fraction of the conditioned fish responded to the signals and some wild red sea bream were attracted as well (Fujiya et al., 1980), indicating social learning. Acoustic training has also been successfully applied in cod ranching in Norwegian and Icelandic fjords (Midling et al., 1987; Björnsson, 1999b, 2002), and in ranching of common carp (Cyprinus carpio) in a fresh water reservoir in Israel (Zion et al., 2012). The presence of trained fish facilitates the acoustic training of naïve fish by social learning (Zion et al., 2007). In three ranching studies, a substantial percentage of the fish attracted to the feeding site was of wild origin; 16% of the red sea bream in a Japanese bay (Fujiya et al., 1980); 25% of the cod in a Norwegian inlet (Midling et al., 1987); and in a fresh water reservoir in Israel, the total number of carp captured exceeded the number of stocked carp (154%) (Zion et al., 2012). Two ranching studies in Iceland showed that a considerable number of wild cod could be attracted to feeding stations, where fresh or frozen feed was deployed regularly (Björnsson, 1999b, 2011). The fish “herds” that were formed consisted almost exclusively of relatively large cod (>40 cm), suggesting that juvenile cod and most other fish species stay away to reduce their risk of predation or cannibalism (Björnsson, 2011). Thus, the unwanted bycatch commonly plaguing long-liners in nearshore waters (Björnsson et al., 2015) was not a problem in catches from the herds (using a lift net). The cod learned to associate specific sounds with feed and they could be lured to the surface using acoustics (Björnsson, 1999b, 2011), which facilitates their capture with a purse seine and reduces their risk of barotrauma during capture (Korsøen et al., 2010; Kristiansen et al., 2011). The first ranching study, carried out in a 7-km long fjord in East-Iceland, attracted thousands of cod to three feeding stations and their growth rate and condition factor increased considerably (Björnsson, 1999b). During the first year of experimental ranching, the weight gain was three times greater for the conditioned than unconditioned cod (Björnsson, 2002). On the basis of these results, it was proposed that fisheries yield could be enhanced by large-scale feeding of the Icelandic cod stock in its natural environment (Björnsson, 2001). The second ranching study, carried out in a 40-km long fjord in Northwest-Iceland, attracted tens of thousands of cod to four feeding stations in a few weeks, but this was only a small proportion of the cod in the whole fjord (Björnsson, 2011). Tagging and recaptures of cod at and outside the feeding stations showed that there was limited mingling between conditioned and wild cod (Björnsson et al., 2010b). Stomach samples confirmed that cod at the feeding stations consumed almost exclusively the food provided, whereas cod elsewhere in the fjord consumed only living prey. Fish samples taken in October 2005 in the herds with a lift net (n = 171) and in 22 other places of the fjord with a shrimp trawl (n = 592) showed a large difference in mean stomach fullness, 11.7 and 0.6% of gutted weight of herd and wild cod, respectively (unpublished results). It thus appeared that the anthropogenic feeding attracted mainly fish in close proximity of the feeding stations and majority of the conditioned fish did not leave the stations while waiting for their meals. In an attempt to understand how classical conditioning and social learning can be used to attract wild cod to a feeding station, an experiment with adult cod was carried out in a sea cage. It took about a week to acoustically train 20 naïve fish to come to a feeding platform, but less than a day to train 19 naïve fish accompanied with one trained conspecific (Björnsson et al., 2010a). This demonstrates the ability of social learning in adult cod as has been found for several species of fish (Zion et al., 2007; Brown and Laland, 2011). The aim of this paper is to postulate and discuss from published research and own experience how cod can be aggregated in open water to facilitate fishing and ranching. The discussion includes the technical details required to control the behaviour of free-roaming fish and how they can be conditioned, fed, and harvested optimally. The hope is that in the future this method, called fish aggregating sound technique (FAST), can be used to harvest cod and several other fish species more economically, with less carbon footprint and less bycatch of juveniles and unwanted species than is possible in the traditional fisheries. Acoustic aggregation of wild cod Cod can be easily trained acoustically to come to a feeding location within a sea cage (Björnsson et al., 2010a), but it is more difficult to train fish in open water. Cod in the wild are not aware of the meaning of a new sound and may at first be repelled by it (own observation). It will take some time to make the connection between sound and food because, for effective learning, the fish have to be present close to the feeding station at the time of feeding and sound emission. However, once the cod have learned the acoustic feeding cue they will remember it for an extended period of time without reinforcement (Björnsson, 1999b; Nilsson et al., 2008; Tlusty et al., 2008). There are ways to expedite the learning of the feeding cues among fish in nature by social learning, which most likely depends on visual and/or auditory cues requiring relatively close contact between the teacher and the student (Brown and Laland, 2011). Cod that have been conditioned to take food at a given feeding station in open water will persistently wait for the next meal at the station (Björnsson, 2011). As a limited number of wild unconditioned fish may be close enough to the conditioned fish at the station, when a sound signal is emitted and feed given, a limited number of naïve fish will be trained. On the other hand, if the trained fish would be systematically called between feeding stations with the familiar sound signal, followed by food reward, they are more likely to come into contact with naïve fish on their way between stations. This will give the naïve fish a greater chance to learn the meaning of the sound signal from the trained fish. Only subtle changes in the behaviour of the teachers are required to inform the naïve fish about their intention to feed (Ioannou et al., 2011). What makes this information transfer so effective is that only a minority of individuals need to be informed (Reebs, 2000; Swaney et al., 2001; Ward et al., 2008). As group size increases, the required proportion of informed individuals to affect the group decreases (Couzin et al., 2005). Once the fish have been aggregated, they can be captured either with a purse seine or a trap, fishing gears that under ideal conditions require little fuel consumption (Tyedmers, 2004; Driscoll and Tyedmers, 2010). Both these gears can be operated in such a way as to minimize mortalities of juveniles and unwanted species (He and Inoue, 2010; Suuronen et al., 2012). The largest fish can be immediately slaughtered, the fish of intermediate size put in sea cages for on-growing to better adjust the supply to seasonally variable demand (Gunnarsson and Björnsson, 2015) and the smallest fish released for continued training and aggregation of wild fish. Rigid sorting grids will allow the small fish to swim out of the trap or purse seine before harvest (Larsen and Isaksen, 1993). When the operation surpasses certain volume, it would be ideal to use well-boats with fish pumps to size-grade and transport the live catch to sea cages and slaughtering plants, a common practice in the salmon farming industry (Gatica et al., 2008; Lines and Spence, 2014). The training period could be either short (several days), with the sole purpose of catching wild fish without large feed expenses, or long (up to several months), with the additional purpose of on-growing free-ranging fish (Björnsson, 2011). The selected length of the ranching period would to a large extent depend on the feed cost relative to the additional quantity, quality and price of fish gained by on-growing. A feasibility study indicated that cod ranching in Northwest Iceland, using forage fish as feed, could be profitable (Halldórsson et al., 2012). Virtual cage and sea ranching Zion and Barki (2012) introduced the useful concept of virtual cage, where acoustic ranching is carried out without confining the fish in a sea cage, i.e. where behavioural control is used instead of netting to keep the conditioned fish within the hearing distance of the conditioning sound source. In their review of acoustic ranching, Zion and Barki (2012) assume that hatchery reared juveniles are conditioned and released into the environment without substantial feeding and thus their growth would mainly be on the basis of consumption of wild prey. Their idea of acoustic ranching is to maintain contact with the fish by periodic signalling and rewarding and subsequent harvesting. In a ranching experiment with small St. Peter’s fish (Sarotherodon galilaeus) in a reservoir in Israel, the target fish did not come to the trap whereas large-sized fish of other species learned to come for the feed that was given at the time of the acoustic signal (Zion and Barki, 2012). Probably, the small fish were trying to minimize their predatory risk by avoiding the meeting place of the big fish. Subsequent experiments in the same reservoir with larger common carp resulted in high returns of both reared and wild carp. Sea ranching, on the basis of releasing hatchery reared juvenile animals into the marine environment for increased harvest, has been carried out in several countries for more than a century with variable success (Kitada, 1999; Kitada and Kishino, 2006; Bell et al., 2008). Recruitment limitation is the theoretical basis for stock enhancement (Doherty, 1999) and ranching of fish without substantial food provisioning is not going to be viable in areas where survival and growth rate is food limited. This was the case in the large-scale sea ranching project in coastal areas in Norway in 1980s and 1990s (Otterå et al., 1999; Svåsand et al., 2000), where more than a million tagged cod juveniles (mostly 20–40 cm) were released in 16 separate studies without significant increase in cod production and catches. The carrying capacity for a given fish population can be defined as the biomass level, which can be supported by the available food resources in the absence of harvesting. Maximum payoff per released individual can only be achieved if the fish stock being targeted for enhancement has a biomass considerably lower than the carrying capacity of the stock (Svåsand et al., 2000). Sea ranching of hatchery reared Spanish mackerel (Scomberomorus niphonius) in the Seto Inland Sea in Japan reduced the growth rate of wild conspecifics when the stock size exceeded the carrying capacity of the environment (Nakajima et al., 2013). Therefore, to acoustically ranch fish in the wild environment, it may, in most cases, be necessary to provide the feed required for maximum growth and survival. Feed given The ideal feed for aggregating and ranching predatory fish, such as cod, would be inexpensive, frozen or fresh forage fish not suitable for direct human consumption; several trials in Canada, Norway, and Iceland to wean half grown wild cod onto dry feed have not been successful (Gunnarsson and Björnsson, 2011). About 33 million tons or 36% of the total world fisheries catch were destined for non-food uses in 2006, either targeted for reduction into fish meal and fish oil used in animal feeds or used directly as animal feed in fresh, frozen, or wet processed form, with the aquaculture sector being the largest consumer of non-food catches (Tacon and Metian, 2009). In Norway and Iceland, two of the countries which may be suitable for cod ranching, the main supply of local industrial fish is capelin (Mallotus villosus), a small oily fish with an average capture weight of about 20 g (Vilhjálmsson, 1994). During the years 1973–2015, the average annual catch of capelin was equal both in Icelandic waters and the Barents Sea, 0.7 million tons in each area, but years with annual catches >0.3 million tons were more common in the former (81%) than in the latter area (51%) (Marine Research Institute, 2016; www.ices.dk). For much of the time, a cod ranching company would have to rely on frozen feed because fresh forage fish is usually only available for a limited part of the year. Two methods have been used successfully to deliver feed to conditioned wild cod. Both tried to minimize the amount of feed that fell to the sea floor, where numerous scavengers, such as crabs, amphipods, gastropods, and starfish, are ready to take any feed that appears. The first method is based on flushing thawed out forage fish through a feeding hose down to a certain depth well above the bottom (Björnsson, 1999b). A rope with a buoy was used to keep the hose at the selected depth, and an underwater video camera was attached to the end of the hose for observing onboard the feeding activity of the fish (Figure 1a). Figure 1. View largeDownload slide Two feeding methods that have been used in cod ranching in Iceland: (a) thawed out forage fish is flushed through a feeding hose down to selected depth and a video camera used to monitor the feeding behaviour of the fish; (b) blocks of frozen forage fish are put in feed bags made of trawl netting. Figure 1. View largeDownload slide Two feeding methods that have been used in cod ranching in Iceland: (a) thawed out forage fish is flushed through a feeding hose down to selected depth and a video camera used to monitor the feeding behaviour of the fish; (b) blocks of frozen forage fish are put in feed bags made of trawl netting. The second method is based on leaving feed bags at feeding stations (Figure 1b). The bags, made of trawl netting, were filled with blocks of frozen forage fish and moored at a suitable depth. The netting was made of a thick double twine to reduce the risk of entanglement by the fish. This feeding method is much less labour intensive than the former because thawing of the feed is not required and it takes relatively little time to fill the bags, load them onto the boat and deploy at each feeding station (Björnsson, 2011). Only stations with empty bags were stocked with more feed. In the bags, the feed gradually thaws and becomes edible by the cod surrounding them. In the majority of feedings, all the food was consumed, and it was very uncommon to find a cod entangled in the net. Video observation at the feeding stations indicated that feed waste was minimal and that the herds consisted almost entirely of relatively large cod; a few haddock (Melanogrammus aeglefinus) were sometimes seen in the periphery but they were rarely caught by the lift net used for sampling (Björnsson, 2011). In these feeding trials, it was not possible to measure biomass of fish at each station, a prerequisite for estimating accurately the food requirement of the fish. Equipping the feed boat with sonar would be a feasible option for this task. Routine sonar measurements of the herds will also aid in assessing the progress of the fish aggregation and help to organize the harvesting of the fish. In remote areas, it may not be feasible to attend the feeding stations daily because of the long sailing time. In spite of the additional investment required, some type of interactive feeders stocked with feed would be preferred, ones that could be activated according to amounts of fish near the feeding stations, either automatically or interactively by the fisher/rancher. The amount and type of conditioned fish could be estimated with echo sounders and video cameras attached to a feeding platform. For fish species that will accept dry feed, a simple feeding system can be developed. For cod and other fish that only accept forage fish, it may be possible, although more complicated, to design a suitable feeder, because a sufficient bulk of frozen forage fish can be stored intact in an insulated container for several days (the time it takes to warm up the feed from –20 to 0°C). Fish aggregating sound technique The proposed method, called FAST, uses low-frequency sound (200–300 Hz) to call trained fish between feeding stations, where they get their food reward. On their way, they are likely to meet some naïve fish that will hear the sound and follow the keen looking and knowledgeable fish to the feeding station where they can be captured (Figure 2). The simplest aggregating system would be two feeding stations, initially located close enough to each other to make it easy to call fish from one station to the other by alternating the sound transmission and feeding between the two stations. Later, the stations could be moved further apart to let the fish swim longer distances and thus be able to sweep larger areas occupied by naïve fish. Also, one or both stations could be moved stepwise towards an area containing high density of wild fish. Figure 2. View largeDownload slide How low-frequency sound could be used in fishing. An illustration of how acoustically trained fish (those marked with flags) can be used to lead wild fish to a trap or a location, where they can be encircled with a purse seine. A sound signal and food is given at alternate stations to drive the trained fish between locations to influence large numbers of wild fish with the purpose to aggregate and capture them. Figure 2. View largeDownload slide How low-frequency sound could be used in fishing. An illustration of how acoustically trained fish (those marked with flags) can be used to lead wild fish to a trap or a location, where they can be encircled with a purse seine. A sound signal and food is given at alternate stations to drive the trained fish between locations to influence large numbers of wild fish with the purpose to aggregate and capture them. The feeding stations would consist of moorings with feeders and sound equipment (transducers). One type of a practical feeder would be a net bag with frozen feed (as described earlier). There is available commercially, a small, robust, programmable, autonomous, and economical low-frequency sound buoy with an output power of 150 dB/1 μPa/1 m (www.star-oddi.com, FishCall). The transmitting range for cod with this buoy may be approximately 1.5 km at 250 Hz, assuming a detection threshold of 80 dB at this frequency (Popper and Carlson, 1998). Acoustic signals have been detected by fish at distances up to 10 km (Zion et al., 2011), but the distances are species and site specific. Even though the fish can hear the feeding cue they may be reluctant to move beyond certain distance from their preferred location. In a ranching experiment with cod in an Icelandic fjord, using a feeding boat with a submersible feeding hose and a sound transmitter, the fish usually did not follow the boat for a longer distance than a few hundred metres (Björnsson, 1999b) but on one occasion up to 500 m (Björnsson and Reynisson, 2013). The average cruising speed of cod is about 0.6 bl s−1 (Björnsson, 1993). Thus, it would take a 70-cm cod swimming at 0.42 m s−1 about 20 min to travel 500 m. For the following reasons, it would be beneficial for a FAST fishing and ranching company to operate not only two but a large number of feeding stations simultaneously within the designated ranching area (Figure 3), although the amount of fish per station may decline somewhat: More fish is aggregated from a larger area (economies of size). The marginal cost of each station is low because the capital cost of each station is low and one boat can service a large number of feeding stations (Björnsson, 2011). Fish have easier access to a feeder if the fish density is not too high. Economical to harvest the whole aggregation of fish at each station with one purse seine cast and therefore a large number of stations must be in operation simultaneously to provide a regular supply of market size fish. Mortality during harvesting may be less if the catch is not too large, allowing the not fully grown fish to be released unharmed. The local organic load will be less. Figure 3. View largeDownload slide FAST: (a) initially attracting fish from a small local area using classical conditioning with low-frequency sound and food reward; (b, c) later attracting wild fish from a larger area by calling trained fish between feeding stations. The sound transmitted from a feeding station is indicated with concentric circles. Figure 3. View largeDownload slide FAST: (a) initially attracting fish from a small local area using classical conditioning with low-frequency sound and food reward; (b, c) later attracting wild fish from a larger area by calling trained fish between feeding stations. The sound transmitted from a feeding station is indicated with concentric circles. The user of FAST must have exclusive rights to fish in a given area (Björnsson, 2011). Otherwise any other fisher could capitalize on the aggregations formed with the acoustic training. During the initial developmental phase of FAST, authorities could temporarily prohibit fishing in the designated area, as has been done previously with experimental ranching of cod (Björnsson, 1999b, 2011). Later on, based on the trial period, general rules and regulations must be implemented. Only repeated trials will tell the viability of the method and how extensively it can be used. Enhancement of landed value Aggregating activity that only lasts for several days and results in insignificant growth may be referred to as FAST fishing, whereas aggregating activity that continues for weeks or months and results in significant growth may be referred to as FAST ranching (Figure 4). For inshore cod, mainly half-grown (0.5–4 kg) fish will be affected. To stabilize supply of fresh fish throughout the year, medium-sized fish (2–3 kg) could be brought to sea cages for on-growing, all of those from FAST fishing and some of those from FAST ranching (Figure 4). Figure 4. View largeDownload slide FAST fishing and FAST ranching of cod. Some of the fish of intermediate size can be brought to sea cages for on-growing to stabilize supply of fresh fish though out the year. Figure 4. View largeDownload slide FAST fishing and FAST ranching of cod. Some of the fish of intermediate size can be brought to sea cages for on-growing to stabilize supply of fresh fish though out the year. One decade ago there was great interest in full-cycle farming of cod in sea cages using juveniles produced in land based hatcheries (Björnsson et al., 2010c). It was prognosed that within 15–20 years, the cod production could reach levels similar to those of salmon (Rosenlund and Skretting, 2006). These plans have not materialized partly because of the difficulty and cost of producing juveniles of marine fish species. A feasibility study indicated that ranching wild cod, starting with half-grown fish (1–2 kg), is more economical than full-cycle cod farming, mainly because the rancher does not have to pay for the production of juveniles and the feed required to produce half-grown cod (Halldórsson et al., 2012). In the marine environment, it is common that recruitment of juvenile fish surpasses the carrying capacity of the stock, making it possible with additional food input to increase the fish production of the stock. There is usually a high density of young fish in the nearshore areas of Iceland, which poses a big problem in the longline fishery for cod and haddock (Björnsson et al., 2015). Frequently, a large proportion of the catch in these areas is undersized fish, which is either landed or discarded at sea, the majority moribund (Milliken et al., 1999). Temporary area closures have been adopted to alleviate this problem, but with limited success (Björnsson et al., 2015). Generally, the inshore cod are food limited (Björnsson, 1999a) and often in a poor nutritional condition (Björnsson, 1999b). The small and lean inshore fish is of relatively low commercial value and not preferred by fish processors. If these fish are offered enough feed, their size, condition factor, liver index, and food quality increase greatly with only a few months of on-growing in free-ranging herds (Björnsson, 2011) or sea cages (Gunnarsson and Björnsson, 2011, 2015). Thus, there is marketing advantage in ranching and on-growing inshore cod. There are many inshore areas in Iceland suitable for cod ranching with FAST. Instead of killing large numbers of half-grown cod, as is the present procedure, they could be reared to larger and more economical sizes to enhance the landed value of the stock. It is hard to predict with accuracy what sorts of quantities of cod could realistically be ranched each year around Iceland. With the crude technology used in the ranching study in Arnarfjördur 2005–2006 about 100 tons of cod were attracted to the four feeding stations, but it was estimated that the total biomass of cod in the 40 × 7 km (285 km2) fjord was approximately 2000 tons of cod >40 cm. It seems realistic to condition at least half of these cod to eat forage fish given sufficient numbers of feeding stations. To get some idea about the amounts of cod that can be ranched in inshore waters of Iceland, the figures for Arnarfjördur were extrapolated purely based on the size of the potential ranching areas (km2) within the fjords and bays West, Northwest, North, and East of Iceland. The estimated amount is approximately 25 000 tons per year, i.e. about 1/10th of the annual catch of cod in Icelandic waters. The average weight of inshore cod entering the herds (1–2 kg) can be doubled with ample feeding with forage fish for 6 months, requiring approximately 3 kg of energy rich forage fish to produce each additional kilogram of cod (Björnsson et al., 2001; Björnsson, 2011; Gunnarsson and Björnsson, 2011). Thus, it seems possible that every year, 25 000 tons of cod could be acoustically ranched and partially reared in sea cages for marketing reasons. Ranching the cod for 6 months would double their biomass to 50 000 tons, requiring approximately 75 000 tons of forage fish. As FAST will only affect a small fraction of the fish stock, the consequences for reproduction and other life history traits of the stock will most likely be small. Regular feeding of the targeted fish, either in free-ranging herds or in sea cages, may reduce their predation on commercially valuable stocks, such as shrimp and juvenile fish (Björnsson, 2001; Björnsson et al., 2011), which may in time increase their Total allowable catch. Increased availability of wild prey may also increase the growth rate of the part of the targeted fish stock, which was left unconditioned. The feeding of the relevant species may thus in time increase the productivity of the prey and/or the unconditioned predator to further augment the overall macro economy of the feeding operation. Limitations of FAST It is important to address some of the limitations of FAST: There is considerable cost involved in fishing, freezing, storing, transporting, and deploying the forage fish used to aggregate the target species; long-term ranching can only be viable if the price of feed is low. FAST is only applicable to valuable, locally abundant and food limited carnivorous fish species, such as cod in Icelandic waters. Straying of fish from the feeding area could be a problem, but it may be minimized with careful feeding and timely harvesting. FAST may not be suitable in offshore areas where rough weather and great depth could make it difficult to feed and harvest the fish. FAST may not be suitable far from ports and freezer plants (logistics). The total amount of feed used must not exceed the acceptable organic load of the designated area (Stigebrandt et al., 2004; Tett et al., 2011). In some cases, non-target species-, predator-, and pathogen-control may be required. The ownership of the fish stocks (utilization right) and fisheries management rules, such as the quota system and harvest control rules, may vary between countries. It is beyond the scope of the paper to discuss all these issues, but straying and predator control will be discussed below. Straying (escapement) Results have shown that with careful feeding and behavioural control, majority of the conditioned cod can be kept within the virtual cage during the main growing period. However, as the conditioned cod become sexually mature they lose their appetite and begin to stray away from the feeding area (Björnsson et al., 2010b; Björnsson, 2011); thus, the maturing fish must be timely harvested and some of them could be stocked in sea cages to avoid escapement. The economic consequences of escapement depend on how much has been invested in feeding and training of the fish. If only a small amount of food has been used in attracting fish, some straying of fish out of the feeding area may have little effect on the economy of the operation, whereas the escapement of a large number of conditioned fish that have been fed and grown for a long time may have a large effect on the profitability. Thus, the escapement problem may be less for FAST fishing than FAST ranching. Many aquaculture people do not like the idea that the ranched fish are able to roam freely out of the feeding area to places where they can be captured by the commercial fleet and would rather like to put the fish in sea cages for on-growing as soon as possible. However, ranching of free-ranging herds is more economical than on-growing in sea cages if the escapement is low (Halldórsson et al., 2012) or if the ownership of the strayed fish can be managed. In sheep- and cattle-ranching, the animals are tagged and later at the end of the growing season they are gathered and sorted according to ownership. However, tagging of ranched fish does not seem a viable option because of the high cost of tagging and the considerable enforcement required. Results have shown that after spawning many of the conditioned cod return to the feeding area (Björnsson, 1999b), most likely a much larger proportion than what the commercial fishing fleet manages to catch. According to the Harvest control rule used for managing the Icelandic cod stock, the fishing fleet is allowed to take annually 20% of the biomass of the fishable part of the stock. Conditioned cod that leave the feeding stations for 6 months may thus have approximately 10% chance of being caught each year. Some straying may even be advantageous for the ranching operation because the returning fish may bring some naïve fish to the feeding stations. As the biomass of the fish stock increases due to increased food intake, it would be reasonable to modify the quota according to the amount of feed deployed. For example, if FAST would be used solely to aggregate and catch fish without any biomass increase, the quota used would be equal to the weight of the catch. However, if the fish would be ranched in herds and/or grown in sea cages to result in, for instance, doubling of the initial mean weight of the fish, the required quota would be 50% of the weight of the harvest. In other words, the original quota would give twice as much fish. This is a big incentive to ranch, but the size structure of the herds must be carefully monitored to allow an accurate estimate of the biomass increase and quota gain to be made. In cases when straying is a big problem, communal feeding may be more suitable than an exclusive right of a private company to fish in a given area. Communal feeding is when all quota holders pay for the feeding of the fish and benefit equally from the increased growth of the stock and subsequent addition of quota. FAST would still be a useful way of aggregating and feeding the fish but the existing commercial fleet would be allowed to harvest the fish. The stock could be fed communally on a grand scale as proposed by Björnsson (2001) and the feeding areas opened to the fleet at certain times after a sufficiently long feeding and growing period. With this approach straying would become a non-issue. Non-target species-, predator-, and pathogen-control One might expect that many non-target species would inevitably come for the free food. However, in the two reasonably long cod ranching experiments in Iceland (17 and 21 months), there was no evidence for this to be the case (Björnsson, 1999b, 2011). As mentioned previously, the herds consisted almost solely of large cod (>40 cm) and a few haddock. Although haddock is a non-target species, it is a commercial species of a similar market value per kilogram as cod and thus the feed it consumes is not wasted if the haddock is harvested. Ranching other target species elsewhere in the world may, however, attract several non-target species. If the non-target species are not as commercially valuable as the target species, it may be necessary to periodically harvest the herd and sort from it the non-target species and either send the catch to the appropriate market or use it as forage fish for the target species, if it is of sufficiently low commercial value. In the two cod ranching experiments in Iceland, there was no evidence for a predator attacking the cod herds. For example, seals or whales, potential predators of cod, were never seen. If marine mammals would become a problem, there are ways to keep them away, such as by Acoustic harassment devices (Fjälling et al., 2006; Graham et al., 2009). Concerning pathogen control, diseases are less likely to occur in free-ranging herds than in sea cages because of lower densities and less chance of abrasion when no cage is used. In the two experiments mentioned earlier, there was no evidence of a disease, but if it would occur, the herd could be captured, vaccinated and medically treated on board a well-boat before the ranching is resumed. FAST and FAD FAST is analogous to the well-known Fish Aggregating Device (FAD), which is either drifting or anchored floating objects that attract sparsely distributed fish schools of pelagic predatory fish, such as tuna, to make purse seining more efficient. Today, approximately half of the global tuna catch comes from purse seiners using drifting FADs (Fonteneau et al., 2013; Davies et al., 2014). Purse seiners use much less energy to catch each ton of fish (Tyedmers and Parker, 2012) and have four to five times lower discard ratios (by weight) than pelagic long liners (Kelleher, 2005). Also purse seines can be equipped with sorting grids to release juvenile fish (Dagorn et al., 2013). The main difference between FAD and FAST is that the latter involves conditioning with sound and food reward, whereas the former uses only the natural affinity of fish for floating objects. More advanced types of FADs, already in use, include echo sounders, GPS and GSM capabilities so that the operator can remotely contact them via satellite to obtain information about the amount of fish before deciding which FADs to visit (Davies et al., 2014). By adding low-frequency sound source and feeders to the more advanced FADs, they are likely to aggregate and keep more fish than the existing FAD technology. Both methods (FAD and FAST) depend strongly on exclusive rights to fish in a specific area because of the cost required to aggregate fish. Traditionally, individual fishers or a group of fishers have put out FADs at their own expense. The fishers believe that once they put out a FAD, only they should fish around it and are often willing to protect this right by force without legal support (Christy, 1992; FAO, 2007; Satria and Adhuri, 2010). More recently, governments and other organizations have deployed large arrays of moored public FADs that are not associated with exclusive rights, in an attempt to make the technology more widely available. Public FADs may, however, be exploited less efficiently and also give rise to new conflicts related to crowding, misuse, and possibly overfishing (Sidman et al., 2014). Thus, the lesson learned from the mismanagement of FADs is to legalize from the start the exclusive rights of the FAST fishers, i.e. by leasing the area where fish are going to be trained, aggregated, captured, or ranched. However, the granting of these exclusive rights must be done according to transparent and reasonable rules, e.g. by public auctioning. It is likely that the most efficient companies would in time lease most of the coastal areas to maximize the profit from the fish stocks. Certain restrictions about maximum ownership could be adopted to maintain healthy competition as has been done in the Individual Transferable Quota system (Kroetz and Sanchirico, 2010). Implementation of FAST A sufficiently long trial period is necessary to study and fine tune the methodology required to acoustically attract, ranch, and harvest each target species, although the experience with one species may facilitate the procedure of the next one. The first step in implementing FAST may be for the pioneer to contract a research institute to carry out a pilot project. It must be done in close cooperation between the authorities, fishing industry, scientists, managers, and other stakeholders, as has been suggested for the development of other fuel-efficient fishing technologies (Suuronen et al., 2012). Also, it seems feasible for the pioneer and research institute to cooperate with a fishing company, which has a suitable fishing boat, other required equipment and the necessary quota. The study area must contain sufficient biomass of a target species to make sure that the fish catches are lucrative enough. On the basis of experience, the study area must also be large enough to minimize the risk of poaching (Björnsson, 2011). The scientists and fishermen would design jointly the acoustic training and capturing system, collect data, and make a report, which can be used to evaluate the ecological and environmental attributes of the method, such as mortality of undersized fish and fuel efficiency, and assess its economic potential. If a pilot project indicates that FAST is a more environmentally friendly and a more economical method than the conventional fishery in the study area, special laws and regulations about this method must be implemented by the authorities. Once the general rules and regulations about FAST have been set out, e.g. those concerning the required data collection, official inspection and enforcement, the development of the method would be in the hands of the private fishing companies, which pay for exclusive rights to fish in the designated areas. Stages in implementing FAST: Obtain fish quota and exclusive fishing rights to a coastal area with high abundance of the target species. Set-up several feeding stations in places where the density of the target species is high. Attract fish from a local area around the feeding stations using classical conditioning with sound and feed. These fish will form the stock of teachers required for the next step. Attract naïve fish from a larger area by calling fish schools between feeding stations (let the teachers swim across area occupied with the targeted species). When ranching is feasible, increase the amount of feed according to increasing biomass of the herds (the fattening period). Harvest the biggest fish, transport the medium-sized fish to sea cages for on-growing and year-round marketing, and release juveniles and unwanted species for continued aggregation and ranching. Generally, the fishing industry is resistant to change (Eayrs et al., 2015). Therefore, the biggest obstacle to implement such a radical change in harvesting the fish stocks is undoubtedly the opposition of fishers and fish quota holders to the idea of exclusive fishing rights, implying that only one quota holder is allowed to fish within a given area. However, as long as the rules of obtaining these exclusive rights are reasonable, transparent, fair, and apply equally to all citizens, the stakeholders might be willing to comply, if they are convinced that this type of fishing (FAST) is competitive with the current fishing methods. The best way to find out about the economic aspects of the method is to operate a pilot project for a sufficient length of time. It seems that the remaining technical aspects of the method would be relatively easy to solve with modern technology, such as sound transmitters, freezer plants, feeders, echo sounders, sonars, sea cages, purse seines, fish pumps, well-boats, etc. The FAST contracts must be sufficiently long lasting to promote investment in research and development. During the last few decades, great development in aquaculture has taken place and worldwide its production has reached similar volumes as the world’s fisheries production. It is likely that in the future there will be greater overlap between aquaculture and fisheries. More knowledge about the behaviour of the main commercial fish species and how it can be controlled can assist in increasing the economic output from the fisheries, e.g. by various FADs and provisioning of wild stocks with food. Hopefully, this paper will be of some help in promoting and planning future studies of acoustic ranching. Acknowledgements The author would like to thank Jón Sólmundsson and James Kennedy, colleagues at the Marine and Freshwater Research Institute, for critically reading the manuscript and making several suggestions of how to improve it. References Abbott R. R. 1972 . Induced aggregation of pond-reared rainbow trout (Salmo gairdneri) through acoustic conditioning . Transactions of the American Fisheries Society , 101 : 35 – 43 . Google Scholar CrossRef Search ADS Bell J. D. , Leber K. M. , Blankenship H. L. , Loneragan N. R. , Masuda R. 2008 . A new era for restocking, stock enhancement and sea ranching of coastal fisheries resources . Reviews in Fisheries Science , 16 : 1 – 9 . Google Scholar CrossRef Search ADS Björnsson B. 1993 . Swimming speed and swimming metabolism of Atlantic cod (Gadus morhua) in relation to available food: a laboratory study . Canadian Journal of Fisheries and Aquatic Sciences , 50 : 2542 – 2551 . Google Scholar CrossRef Search ADS Björnsson B. 1999a . Is the growth rate of Icelandic cod (Gadus morhua L.) food-limited? A comparison between pen-reared cod and wild cod living under similar thermal conditions . Rit Fiskideildar , 16 : 271 – 279 . Björnsson B. 1999b . Fjord-ranching of wild cod in an Icelandic fjord: effects of feeding on nutritional condition, growth rate and behaviour. In Stock Enhancement and Sea Ranching , pp. 243 – 256 . Ed. by Howell B. R. , Moksness E. , Svåsand T. . Fishing News Books, Blackwell Science , Oxford . Björnsson B. 2001 . Can fisheries yield be enhanced by large-scale feeding of a predatory fish stock? A case study of the Icelandic cod stock . Canadian Journal of Fisheries and Aquatic Sciences , 58 : 2091 – 2104 . Google Scholar CrossRef Search ADS Björnsson B. 2002 . Effects of anthropogenic feeding on the growth rate, nutritional status and migratory behaviour of free-ranging cod in an Icelandic fjord . ICES Journal of Marine Science , 59 : 1248 – 1255 . Google Scholar CrossRef Search ADS Björnsson B. 2011 . Ranching of wild cod in ‘herds’ formed with anthropogenic feeding . Aquaculture , 312 : 43 – 51 . Google Scholar CrossRef Search ADS Björnsson B. , Karlsson H. , Gudbjörnsson S. 2010a . The presence of experienced cod (Gadus morhua) facilitates the acoustic training of naïve conspecifics. ICES CM 2010/I: 09. Björnsson B. , Karlsson H. , Thorsteinsson V. 2010b . Effects of anthropogenic feeding on the migratory behaviour of coastal cod (Gadus morhua) in Northwest Iceland . Fisheries Research , 106 : 81 – 92 . Google Scholar CrossRef Search ADS Björnsson B. , Litvak M. , Trippel E. A. , Suquet M. 2010c . The codfishes (Family: Gadidae). In Finfish Aquaculture Diversification , pp. 290 – 322 . Ed. by Le François N. R. , Jobling M. , Carter C. , Blier P.U. . CABI , Wallingford, UK . Google Scholar CrossRef Search ADS Björnsson B. , Reynisson P. , Solmundsson J. , Valdimarsson H. 2011 . Seasonal changes in migratory and predatory activity of two species of gadoid preying on inshore northern shrimp Pandalus borealis . Journal of Fish Biology , 78 : 1110 – 1131 . Google Scholar CrossRef Search ADS PubMed Björnsson B. , Reynisson P. 2013 . Synchronous and vertically undulating swimming behaviour of Atlantic cod Gadus morhua . Aquatic Biology , 19 : 13 – 18 . Google Scholar CrossRef Search ADS Björnsson B. , Sólmundsson J. , Pálsson Ó. K. 2015 . Can permanent closures of nearshore areas reduce the proportions of undersized fish in the Icelandic longline fishery? ICES Journal of Marine Science , 72 : 841 – 850 . Google Scholar CrossRef Search ADS Björnsson B. , Steinarsson A. , Oddgeirsson M. 2001 . Optimal temperature for growth and feed conversion of immature cod (Gadus morhua L.) . ICES Journal of Marine Science , 58 : 29 – 38 . Google Scholar CrossRef Search ADS Brown C. , Laland K. 2011 . Social learning in fishes. In Fish Cognition and Behavior , 2nd edn , pp. 240 – 257 . Ed. by Brown C. , Laland K. , Krause J. . Blackwell Publishing , Oxford, UK . Google Scholar CrossRef Search ADS Christy F. T. 1992 . Territorial use rights in marine fisheries: definitions and conditions. FAO Fisheries Technical Paper 227, FAO, Rome. Couzin I. D. , Krause J. , Franks N. R. , Levin S. A. 2005 . Effective leadership and decision-making in animal groups on the move . Nature , 433 : 513 – 516 . Google Scholar CrossRef Search ADS PubMed Dagorn L. , Holland K. N. , Restrepo V. , Moreno G. 2013 . Is it good or bad to fish with FADs? What are the real impacts of the use of drifting FADs on pelagic marine ecosystems? Fish and Fisheries , 14 : 391 – 415 . Google Scholar CrossRef Search ADS Davies R. W. D. , Cripps S. J. , Nickson A. , Porter G. 2009 . Defining and estimating global marine fisheries bycatch . Marine Policy , 33 : 661 – 672 . Google Scholar CrossRef Search ADS Davies T. K. , Mees C. C. , Milner-Gulland E. J. 2014 . The past, present and future use of drifting fish aggregating devices (FADs) in the Indian Ocean . Marine Policy , 45 : 163 – 170 . Google Scholar CrossRef Search ADS Doherty P. J. 1999 . Recruitment limitation is the theoretical basis for stock enhancement in marine populations. In Stock Enhancement and Sea Ranching , pp. 9 – 21 . Ed. by Howell B. R. , Moksness E. , Svåsand T. . Fishing News Books, Blackwell Science , Oxford . Driscoll J. , Tyedmers P. 2010 . Fuel use and greenhouse gas emission implications of fisheries management: the case of the New England Atlantic herring fishery . Marine Policy , 34 : 353 – 359 . Google Scholar CrossRef Search ADS Eigaard O. R. , Bastardie F. , Breen M. , Dinesen G. E. , Hintzen N. T. , Laffargue P. , Mortensen L. O. , et al. 2016 . Estimating seabed pressure from demersal trawls, seines, and dredges based on gear design and dimensions . ICES Journal of Marine Science , 73 : 27 – 43 . Google Scholar CrossRef Search ADS Eayrs S. , Cadrin S. X. , Glass C. W. 2015 . Managing change in fisheries: a missing key to fishery-dependent data collection? ICES Journal of Marine Science , 72 : 1152 – 1158 . Google Scholar CrossRef Search ADS FAO 2007 . Report of the second meeting of the WECAFC ad hoc working group on the development of sustainable moored fish aggregating device fishing in the Lesser Antilles. FAO Fisheries Report 797, FAO Rome. FAO 2016 . The state of the world fisheries and aquaculture 2016. Contributing to food security and nutrition for all. Rome. 200 pp. Fjälling A. , Wahlberg M. , Westerberg H. 2006 . Acoustic harassment devices reduce seal interaction in the Baltic salmon-trap, net fishery . ICES Journal of Marine Science , 63 : 1751 – 1758 . Google Scholar CrossRef Search ADS Fonteneau A. , Chassot E. , Bodin N. 2013 . Global spatio-temporal patterns in tropical tuna purse sein fisheries on drifting fish aggregating devices (DFADs): taking a historical perspective to inform current challenges . Aquatic Living Resources , 26 : 37 – 48 . Google Scholar CrossRef Search ADS Fujiya M. , Sakaguchi S. , Fukuhara O. 1980 . Training of fishes applied to ranching of red sea bream in Japan. In Fish Behavior and Its Use in the Capture and Culture of Fishes , pp. 200 – 209 . Ed. by Bardach J. E. , Magnuson J. J. , May R. C. , Reinhart J. M. . ICLARM , Manila, Philippines . Gabriel O. , Lange K. , Dahm E. , Wendt T. 2005 . Fish Catching Methods of the World . 4th edn , Blackwell Publishing , Oxford, UK . Google Scholar CrossRef Search ADS Gatica M. C. , Monti G. , Gallo C. , Knowles T. G. , Warriss P. D. 2008 . Effects of well-boat transportation on the muscle pH and onset of rigor mortis in Atlantic salmon . Veterinary Record , 163 : 111 – 116 . Google Scholar CrossRef Search ADS PubMed Graham I. M. , Harris R. N. , Denny B. , Fowden D. , Pullan D. 2009 . Testing the effectiveness of an acoustic deterrent device for excluding seals from Atlantic salmon rivers in Scotland . ICES Journal of Marine Science , 66 : 860 – 864 . Google Scholar CrossRef Search ADS Grieve C. , Brady D. C. , Polet H. 2014 . Best practices for managing, measuring and mitigating the benthic impacts of fishing – Part 1 . Marine Stewardship Council Science Series , 2 : 18 – 88 . Gunnarsson V. I. , Björnsson B. 2011 . Feed and feeding of wild farmed cod . Marine Research in Iceland , 157 : 21 – 87 (in Icelandic). Gunnarsson V. I. , Björnsson B. 2015 . Cod quota for on-growing: summary for the years 2002-2014 . Marine Research in Iceland , 184 : 102 pp. (in Icelandic). Hall M. A. , Alverson D. L. , Metuzals K. I. 2000 . By-catch: problems and solutions . Marine Pollution Bulletin , 41 : 204 – 219 . Google Scholar CrossRef Search ADS Halldórsson J. E. , Björnsson B. , Gunnlaugsson S. B. 2012 . Feasibility of ranching coastal cod (Gadus morhua) compared with on-growing, full-cycle farming and fishing . Marine Policy , 36 : 11 – 17 . Google Scholar CrossRef Search ADS Hawkins A. D. 1993 . Underwater sound and fish behaviour. In Behaviour of Teleost Fishes . Ed. by Pitcher T. A. , pp. 129 – 169 . Chapman & Hall , London . Google Scholar CrossRef Search ADS Hawkins A. D. , Sand O. 1977 . Directional hearing in the median vertical plane by the cod . Journal of Comparative Physiology , 122 : 1 – 8 . Google Scholar CrossRef Search ADS He P. , Inoue Y. , 2010 . Large-scale fish traps: gear design, fish behavior, and conservation challenges. In Behaviour of Marine Fishes: Capture Processes and Conservation Challenges , pp. 1 – 181 . Ed. by He P. . Wiley-Blackwell , Iowa, USA . Google Scholar CrossRef Search ADS Ioannou C. C. , Couzin I. D. , James R. , Croft D. P. , Krause J. 2011 . Social organization and information transfer in schooling fish. In Fish Cognition and Behavior , 2nd edn , pp. 217 – 239 . Ed. by Brown C. , Laland K. , Krause J. . Blackwell Publishing , Oxford, UK . Google Scholar CrossRef Search ADS Íslandsbanki . 2013 . North Atlantic Seafood Market Report. Reykjavík. 22 pp. www.islandsbanki.is/Library/Skrar/Seafood-Reports/northAtlantic_FINAL_low.pdf (last accessed 8 January 2018). Kelleher K. 2005 . Discarding in the world’s marine fisheries: an update. FAO Fisheries Technical Paper 470. FAO, Rome. Kitada S. 1999 . Effectiveness of Japan’s stock enhancement programmes: current perspectives. In Stock Enhancement and Sea Ranching , pp. 103 – 131 . Ed. by Howell B. R. , Moksness E. , Svåsand T. . Fishing News Books, Blackwell Science , Oxford . Kitada S. , Kishino H. 2006 . Lessons learned from Japanese marine finfish stock enhancement programmes . Fisheries Research , 80 : 101 – 112 . Google Scholar CrossRef Search ADS Korsøen Ø. J. , Dempster T. , Fosseidengen J. E. , Fernö A. , Heegaard E. , Kristiansen T. S. 2010 . Behavioural responses to pressure changes in cultured Atlantic cod (Gadus morhua): defining practical limits for submerging and lifting sea-cages . Aquaculture , 308 : 106 – 115 . Google Scholar CrossRef Search ADS Kristiansen T. S. , Stien L. H. , Fosseidengen J. E. , Strand E. , Juell J.-E. 2011 . Voluntary reponses and limits of tolerance to pressure reduction and swimbladder expansion in farmed Atlantic cod . Aquaculture International , 19 : 625 – 636 . Google Scholar CrossRef Search ADS Kroetz K. , Sanchirico J. N. 2010 . Economic insights into the cost of design restrictions in ITQ programs. RFF report, Resources for the Future. 46 pp. Larsen R. B. , Isaksen B. 1993 . Size selectivity of rigid sorting grids in bottom trawls for Atlantic cod (Gadus morhua) and haddock (Melanogrammus aeglefinus) . ICES Marine Science Symposia , 196 : 178 – 182 . Levin L. E. , Levin A. F. 1994 . Conditioning as an aid to fish harvest . Aquaculture Engineering , 13 : 201 – 210 . Google Scholar CrossRef Search ADS Lines J. A. , Spence J. 2014 . Humane harvesting and slaughter of farmed fish . Scientific and Technical Review of the Office International des Epizooties , 33 : 255 – 264 . Google Scholar CrossRef Search ADS Løkkeborg S. 1998 . Feeding behaviour of cod, Gadus morhua: activity rhythm and chemically mediated food search . Animal Behaviour , 56 : 371 – 378 . Google Scholar CrossRef Search ADS PubMed Løkkeborg S. 2005 . Impacts of trawling and scallop dredging on benthic habitats and communities. FAO Fisheries Technical Paper 472. FAO, Rome. 58 pp. Marine Research Institute . 2016 . State of marine stocks in Icelandic waters 2015/2016 and prospects for the quota year 2016/2017. Marine Research in Iceland 185. 188 pp. https://www.hafogvatn.is/static/research/files/fjolrit_185pdf (last accessed 8 January 2018). Meager J. J. , Batty R. S. 2007 . Effects of turbidity on the spontaneous and prey-searching activity of juvenile Atlantic cod (Gadus morhua) . Philosophical Transactions of the Royal Society B , 362 : 2123 – 2130 . Google Scholar CrossRef Search ADS Midling K. Ø. , Kristiansen T. S. , Ona E. , Øiestad V. 1987 . Fjordranching with conditioned cod. ICES CM 1987/F: 29. Milliken H. O. , Carr H. A. , Farrington M. , Lent E. 1999 . Survival of Atlantic cod (Gadus morhua) in the Northwest Atlantic longline fishery . Marine Technology Society Journal , 33 : 19 – 24 . Google Scholar CrossRef Search ADS Nakajima K. , Kitada S. , Yamazaki H. , Takemori H. , Obata Y. , Iwamoto A. , Hamasaki K. 2013 . Ecological interactions between hatchery and wild fish: a case study based on the highly piscivorous Japanese Spanish mackerel . Aquaculture Environment Interactions , 3 : 231 – 243 . Google Scholar CrossRef Search ADS Nilsson J. , Kristiansen T. S. , Fosseidengen J. E. , Fernö A. , van den Bos R. 2008 . Learning in cod (Gadus morhua): long trace interval retention . Animal Cognition , 11 : 215 – 222 . Google Scholar CrossRef Search ADS PubMed Otterå H. , Kristiansen T. S. , Svåsand T. , Nordeide J. T. , Nævdal G. , Borge A. , Pedersen J. P. 1999 . Enhancement studies of Atlantic cod (Gadus morhua L.) in an exposed area in western Norway. In Stock Enhancement and Sea Ranching , pp. 257 – 276 . Ed. by Howell B.R. , Moksness E. , Svåsand T. . Fishing News Books, Blackwell Science , Oxford . Popper A. N. , Carlson T. J. 1998 . Application of sound and other stimuli to control fish behavior . Transactions of the American Fisheries Society , 127 : 673 – 707 . Google Scholar CrossRef Search ADS Reebs S. G. 2000 . Can a minority of informed leaders determine the foraging movements of a fish shoal? Animal Behaviour , 59 : 403 – 409 . Google Scholar CrossRef Search ADS PubMed Rijnsdorp A. D. , Bastardie F. , Bolam S. G. , Buhl-Mortensen L. , Eigaard O. R. , Hamon K. G. , Hiddink J. G. , et al. 2016 . Towards a framework for the quantitative assessment of trawling impact on the seabed and benthic ecosystem . ICES Journal of Marine Science , 73(Suppl. 1) : 127 – 138 . Google Scholar CrossRef Search ADS Rose G. A. 1993 . Cod spawning on a migration highway in the north-west Atlantic . Nature , 366 : 458 – 461 . Google Scholar CrossRef Search ADS Rose G. A. , Rowe S. 2015 . Northern cod comeback . Canadian Journal of Fisheries and Aquatic Sciences , 72 : 1789 – 1798 . Google Scholar CrossRef Search ADS Rosenlund G. , Skretting M. 2006 . Worldwide status and perspective on gadoid culture . ICES Journal of Marine Science , 63 : 194 – 197 . Google Scholar CrossRef Search ADS Satria A. , Adhuri D. S. 2010 . Pre-existing fisheries management systems in Indonesia, focusing on Lombok and Maluku, pp. 31 – 55 . In Managing Coastal and Inland Waters: Pre-existing Aquatic Management Systems in Southeast Asia . Ed. by Ruddle K. , Satria A. . Springer Science , Netherlands . Schuijf A. , Buwalda R. J. A. 1975 . On the mechanism of directional hearing in cod (Gadus morhua L.) . Journal of Comparative Physiology , 98 : 333 – 343 . Google Scholar CrossRef Search ADS Sidman C. , Lorenzen K. , Magloire A. , Sebastien R. , Hazell J. , Masters J. , Johnson H. 2014 . Piloting an engagement strategy to support co-management of the Caribbean FAD fishery. GCFI, 66: 119–124. Proceedings of the 66th Gulf and Caribbean Fisheries Institute, November 4–8, 2013, Corpus Christi, Texas, USA. Silverman E. D. , Veit R. R. , Nevitt G. A. 2004 . Nearest neighbors as foraging cues: information transfer in a patchy environment . Marine Ecology Progress Series , 277 : 25 – 35 . Google Scholar CrossRef Search ADS Stigebrandt A. , Aure J. , Ervik A. , Hansen P. K. 2004 . Regulating the local environmental impact of intensive marine fish farming. III. A model for estimation of the holding capacity in the Modelling – Ongrowing fish farm – Monitoring system . Aquaculture , 234 : 239 – 261 . Google Scholar CrossRef Search ADS Suuronen P. , Chopin F. , Glass C. , Løkkeborg S. , Matsushita Y. , Queirolo D. , Rihan D. 2012 . Low impact and fuel efficient fishing – looking beyond the horizon . Fisheries Research , 119–120 : 135 – 146 . Google Scholar CrossRef Search ADS Svåsand T. , Kristiansen T. S. , Pedersen T. , Salvanes A. G. V. , Engelsen R. , Naevdal G. , Nødtvedt M. 2000 . The enhancement of cod stocks . Fish and Fisheries , 1 : 173 – 205 . Google Scholar CrossRef Search ADS Swaney W. , Kendal J. , Capon H. , Brown C. , Laland K. N. 2001 . Familiarity facilitates social learning of foraging behaviour in the gubby . Animal Behaviour , 62 : 591 – 598 . Google Scholar CrossRef Search ADS Tacon A. G. J. , Metian M. 2009 . Fishing for aquaculture: non-food use of small pelagic forage fish–a global perspective . Reviews in Fisheries Science , 17 : 305 – 317 . Google Scholar CrossRef Search ADS Tett P. , Portilla E. , Gillibrand P. A. , Inall M. 2011 . Carrying and assimilative capacities: the ACExR-LESV model for sea-loch aquaculture . Aquaculture Research , 42 : 51 – 67 . Special Issue: Proceedings of the International Symposium, Scottish Aquaculture: A Sustainable Future, Supplement S1. Google Scholar CrossRef Search ADS Thiebault A. , Mullers R. H. E. , Pistorius P. A. , Tremblay Y. 2014 . Local enhancement in a seabird: reaction distances and foraging consequence of predator aggregations . Behavioral Ecology , 25 : 1302 – 1310 . Google Scholar CrossRef Search ADS Tlusty M. F. , Andrew J. , Baldwin K. , Bradley T. M. 2008 . Acoustic conditioning for recall/recapture of escaped Atlantic salmon and rainbow trout . Aquaculture , 274 : 57 – 64 . Google Scholar CrossRef Search ADS Tyedmers P. 2004 . Fisheries and energy use. In Encyclopedia of Energy , vol. 2 , pp. 683 – 693 . Ed. by Cleveland C.J. . Elsevier , San Diego . Google Scholar CrossRef Search ADS Tyedmers P. , Parker R. 2012 . Fuel consumption and greenhouse gas emissions from global tuna fisheries: A preliminary assessment. ISSF Technical Report 2012-03. International Seafood Sustainability Foundation, McLean, Virginia, USA. Vilhjálmsson H. 1994 . The Icelandic capelin stock. Capelin (Mallotus villosus Müller) in the Iceland-Greenland-Jan Mayen area . Rit Fiskideildar , 13 : 1 – 281 . Ward A. J. W. , Sumpter D. J. T. , Couzin I. D. , Hart P. J. B. , Krause J. 2008 . Quorum decision-making facilitates information transfer in fish shoals . Proceedings of the National Academy of Sciences , 105 : 6948 – 6953 . Google Scholar CrossRef Search ADS Watson R. , Revenga C. , Kura Y. 2006 . Fishing gear associated with global catches. I. Database development . Fisheries Research , 79 : 97 – 102 . Google Scholar CrossRef Search ADS Zion B. , Barki A. 2012 . Ranching fish using acoustic conditioning: has it reached a dead end? Aquaculture , 344-349 : 3 – 11 . Google Scholar CrossRef Search ADS Zion B. , Barki A. , Grinshpon J. , Rosenfeld L. , Karplus I. 2007 . Social facilitation of acoustic training in the common carp Cyprinus carpio (L.) . Behaviour , 144 : 611 – 630 . Google Scholar CrossRef Search ADS Zion B. , Barki A. , Grinshpon J. , Rosenfeld L. , Karplus I. 2011 . An automatic fishing machine based on acoustic conditioning . Aquacultural Engineering , 45 : 87 – 91 . Google Scholar CrossRef Search ADS Zion B. , Karplus I. , Barki A. 2012 . Ranching acoustically conditioned fish using an automatic fishing machine . Aquaculture , 330–333 : 136 – 141 . Google Scholar CrossRef Search ADS © International Council for the Exploration of the Sea 2018. All rights reserved. For Permissions, please email: [email protected] This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

doi: 10.1093/icesjms/fsx250pmid: N/A

Abstract This paper follows my journey from childhood in Missouri, where I saw my first jellyfish, to the oceans of the world. Pelagic cnidarians and ctenophores (“jellies”) have been the focus of my career. I think my work has been relevant to the broader scientific community because jellies are predators and potential competitors of fish. In my early research, I quantitatively estimated the predation effects of jellies on zooplankton and ichthyoplankton. I found that most jellies are selective predators, with a few species having diets of only fish larvae or soft-bodied prey. As I learned more about the physical environment that jellies encounter, my early reductionist approach evolved into a more holistic approach. I thought the asexual multiplication from the attached polyp stage would be fundamental in determining jellyfish population size and that the effects of environmental variables could be tested experimentally. It also seemed that humans have changed the natural environment in ways favoring jellies over fish and jelly populations may have increased in developed, eutrophic, hypoxic, overfished, and warming coastal waters. Many opportunities were available that gave me a global perspective. I have persisted despite some difficulties because I love to learn and I am still having fun! Landlocked (Missouri, 1954–1972) I have always loved animals and nature. As a child, I spent my free time outdoors. I would have filled up our house with animals of all sorts, had my mother allowed it. As it was, I had various insects, injured squirrels and birds, and the more traditional dogs, aquarium fish, and pet birds. I was an observer and scientist even when I was young. As an only child, I probably had more time alone than others and, when friends were unavailable to play outside, I poured over identification books for all kinds of animals, my favourites being the sea creatures. I was always drawn to water, which involved as much time in the swimming pool during the summer as I could manage. I travelled little in my youth, which probably led to my desire to see the world. One time, my father took my mother and me with him on a business trip to Chicago, where I had a memorable visit to the Shedd Aquarium. Our summer vacations were to rivers and lakes in the Ozark Mountains to catch fish, where I clearly remember seeing my first jellyfish when I was 4 years old. Men were important positive influences in my early life. Foremost was my father, who enthusiastically supported me in everything. He told me bedtime stories, my favourites being the adventures of a hermit crab and shark feeding frenzies. Jacques Cousteau was also very inspirational; I was transfixed by the amazing creatures that he filmed in the sea. By high school, I knew that I wanted to become a marine biologist. My dad thought I was going to save the world by studying the ocean. Beyond engendering a love of the sea creatures, my father also instilled strong beliefs in fairness, honesty, and justice. Even with my supportive family, I struggled against insecurity, which took me years to overcome. My advanced biology teacher in high school told me one day that he knew I wanted to be a marine biologist, but he did not think I had any special talent for it. I was completely crushed! My parents suggested I talk with my previous biology teacher, who said, “you can do whatever you set your mind to.” I decided to believe this teacher and disprove the other. I knew that achieving a career in marine biology would be difficult. I looked for opportunities to distinguish myself from other students in order to be accepted at a top college. An National Science Foundation (NSF) program at the Scripps Institution of Oceanography designed to give high school students research experience provided my first hands-on exposure to the wonders of the ocean and cemented my direction. From this and later experiences, I believe that hands-on exposure of young students to research is one key to stimulating their interest. College (California, 1972–1976) I headed to the west coast for college at Stanford University, wanting a school with a marine laboratory that offered undergraduate classes. I introduced myself to a professor, Chuck Baxter, at the Hopkins Marine Station (HMS) in Monterey, who would become a friend and mentor for my adult life. As a freshman, I took classes from Chuck and later in full-time study in residence at HMS. As a new researcher, I solved a long-standing mystery (how clonal Metridium senile sea anemones use “catch tentacles”) and excitedly told Chuck that they fight among different clones (Purcell, 1977b). He chuckled and said, “I thought that might be it.” I asked why he hadn’t said so earlier and he replied, “I wanted you to figure it out for yourself,” thus giving me the greatest gift from any teacher—the joy of discovery. I received another great gift from a different advisor there, Don Abbott, who taught me how to write a scientific paper. We went over and over and over my first paper. He wanted it to be “perfect,” emphasizing that publication makes it permanent. I just wanted to be finished. The editors accepted the submitted manuscript “as is”; that remains my only paper accepted without revision (Purcell, 1977a). As I prepared my research for publication in graduate school, I gave each draft to the same three committee members. Their responses were, frustratingly, consistently inconsistent: one would say (my paraphrasing), “this is perfect, submit it,” another would say, “this is terrible, start over,” and the third always had constructive suggestions. I learned difficult lessons from these reviewers: to accept criticism, that all reviewers have different opinions, and that all criticisms should be considered carefully because what was written may not be clear to all readers. Every paper I write or edit, I think of how lucky I was to have had someone who taught me so carefully about writing. I try to pass that gift along to students and non-native English writers who seek my editing advice. I realize how helpful are some of the basic suggestions for writing that often are ignored for lack of time: outlines at the beginning and throughout the writing process to promote good organization; an outside reader and re-reading your own paper after at least a week to improve understandability. Writing takes a lot of practice; it is essential to learn and not be discouraged. As an undergraduate, I took four intensive courses that permitted me to focus on one research class without the distractions of multiple classes and to get to know the professors. Three of the classes were at the HMS marine laboratory, which I found to be a small, friendly subset of the larger university. Throughout my career, I have sought to teach in intensive courses, which I find to be rewarding for the faculty and the students, and to be in marine lab environments. I now realize I have been trying to find the same atmosphere I had at HMS. Graduate school (California, 1976–1981) For graduate school at the University of California, Santa Barbara, I began in the lab of a neurobiologist, having been convinced by others that I should pursue the “recognition of self” that I had found in the sea anemone with neuro- or developmental biology. That field and I were mismatched. It revealed to me one of my weaknesses: I am generally distrustful of machines, upon which I was completely dependent in neurobiology. After a year of making no progress toward a thesis topic and being generally unhappy, I concluded I did not want a career in neurobiology and told that advisor I was leaving his lab. Fortunately, an exciting new professor, Alice Alldredge, who worked on the behaviour and ecology of gelatinous zooplankton, had joined the faculty. I saw her present a seminar and thought her research was suited to my interests. She agreed that I could join her group, in which we subdivided the gelatinous taxa so as not to intrude in each other’s research: Alice worked on appendicularians, the other graduate student had ctenophores, I chose siphonophores, which are in the same phylum as sea anemones. In three weeks, I had a thesis plan and was on my way. I was in the water with my eyes and nothing else except collecting jars. Thus, I happened into “jellyfish ecology” by accident, not by design. But it seemed favourable, being a new and uncrowded field in the late 1970s. I feel lucky to have experienced “blue water diving,” which perhaps gave me a better understanding of life as a gelatinous zooplankter. I think the perceptions of organisms experienced only from cultures, net collections, or test tubes do not reflect nature. Thus, I recommend to everyone to know your animals in situ. “Learning from nature” has much to teach us (Able, 2016). Unfortunately, blue water diving is seldom possible now due to the unfundability of “natural history” and increasing concerns of institutions over liability. I also learned experimental approaches from faculty working in the more mature fields of limnology and benthic ecology. I saw the value of combining quantitative research in situ and experiments for gelatinous species. Experience with jellies, which generally do poorly in the laboratory, taught me that nothing done in a laboratory experiment is natural. I also have worked with mesocosms of various designs, which have the same problems plus others. Containers of any size are useful for carefully designed experiments to test hypotheses (e.g. de Lafontaine and Leggett, 1987), but not to reproduce nature. Therefore, I have tried to use in situ data as much as possible throughout my career. I also learned about “Occam’s razor,” paraphrased “that the simplest answer is often correct,” which resonated with me; consequently, I have always looked for the most direct way to answer questions. Three methods of estimating ingestion by jellies were being used: from metabolic rates, from feeding experiments in laboratory containers, and from in situ gut contents and digestion times (Purcell and Kremer, 1983). The last method seemed to be the most direct with the fewest artefacts. Thus, my doctoral thesis focused on the trophic ecology of siphonophores. Because they are fragile colonies and, to be complete or healthy, specimens must be collected in jars, my doctoral work remains some of the only research on their ecology (Purcell, 1980, 1981a, b, c, 1982). Although few people have studied living siphonophores, their small, sturdy nectophores (“swimming bells”) that are preserved in standard zooplankton tows provide some of the best data available on abundances and distributions of gelatinous predators (reviewed by Mackie et al., 1987; Mapstone, 2015). I believe that luck and determination played important parts in my early career. We scuba-dived weekly and I collected siphonophores to quantify their diets. Although most specimens I found had crustacean exoskeletons in their gastrozooids (“stomachs” or “guts”), I found only unrecognizable black balls in the gastrozooids of one species, Rhizophysa eysenhardti. But I kept looking and one day found a tiny fish head. That discovery set the course of my future research for many years. Additional research showed that species examined in that suborder of siphonophores (Rhizostomae), which includes Physalia, the Portuguese man of war, eat almost exclusively fish larvae (Purcell, 1981a, b, 1984 b). That seemed remarkable to me, especially because copepods outnumber fish larvae by more than 100 to 1 in the oceans (e.g. de Lafontaine and Leggett, 1987; Purcell et al., 1994) and that the siphonophores do not see or swim to catch prey. This selectivity reflects the types of nematocysts (“stinging capsules”) in their tentacles, which lack the adhesive properties of nematocysts in siphonophores that eat mostly crustaceans (Purcell, 1984a) and may require appropriate chemical stimuli to discharge their nematocysts (Purcell and Anderson, 1995). Differences in nematocysts also contribute to differences in feeding among hydromedusae (Purcell and Mills, 1988; Corrales-Ugalde et al., 2017) and other medusae (Carrette et al., 2002). With increasing blooms of stinging jellyfish and improved methods of nematocyst isolation and venom extraction, more is being learned about their functions in prey capture and effects on humans (e.g. Badré, 2014 and references therein). During graduate school, I attended every departmental seminar in the belief that I never knew what I might learn. That has remained my philosophy throughout my career and I recommend that students attend seminars. I also took at least one seminar class per term and found them to be an excellent way to experience a variety of topics. I have enjoyed teaching seminar classes, which permit interaction between students and professors. I have always encouraged “lunch-time seminars” to provide an informal setting where students can practice speaking and receive constructive feedback on their work. Postdoctoral fellowships (East and West Coasts, 1981–1984) Subsequently, I did two postdocs, during which I began traditions that I continue today. First, throughout my career, I have tried to do things that had never been done before. Thus, in my first postdoc at the Woods Hole Oceanographic Institution on research cruises with Larry Madin, I made direct measurements of assimilation by siphonophores, which hold individual prey in a gastrozooid for digestion. Those measurements showed that siphonophores digest everything except the chitin exoskeleton of the crustaceans and the eye lenses and pigments of fish larvae, with assimilation of > 90% (Purcell, 1983). Because empty exoskeletons predominate in digestion experiments on other gelatinous taxa (Purcell et al., 1991), I believe that the high assimilation rates of siphonophores are characteristic of gelatinous predators feeding in nature, although those given high-density prey in the laboratory show lower assimilation rates (e.g. Reeve et al., 1989). Second, I thought that fish and fisheries scientists would care about what ate the baby fish. Therefore, wherever I went, I connected with, learned from, and collaborated with the experts on the local fish (e.g. Greg Lough, Doug Hay). My third tradition started with an invitation by my second postdoctoral advisor, George Mackie at the University of Victoria in British Columbia, Canada, to join with him to write a review on siphonophores, which became my first review paper (Mackie et al., 1987). This second postdoc was on jelly predation on ichthyoplankton, particularly herring larvae, because the spawning populations in North Puget Sound were fished commercially and had decreased dramatically (Purcell, 1989, 1990). I had collected so much information on jelly predation on fish eggs and larvae that it seemed easy to write another review to inform others about the subject (Purcell, 1985). The postdoc years were the best ones for me because I was free to focus on research without degree requirements, faculty duties, and bureaucracy. Assistant professor (West Coast, 1984–1986) In the early 1980s, there seemed to be two kinds of academic positions—soft money (salary from grants) research positions and hard money (salary from the school) faculty positions with a heavy dose of teaching. Although I was not averse to teaching, my real love and strength was in research. The ideal job for me seemed to be a tenure-track faculty position that offered the security of a long-term commitment, with a mix of teaching and research. My college and postdoc years had not prepared me completely for a position in academia. Although I had learned to do research, to write scientific papers and grant proposals, and to teach by assisting in laboratory courses and “learning by example,” I knew nothing about interpersonal and institutional politics, which are important everywhere, even in academia. I thought doing “cutting-edge” research, publishing in top journals, getting funding for a research program, and teaching was what was required; however, that was not all that I needed to know to succeed. I idealistically expected the same supportive working environment I had experienced previously. My first professorship was tenure track, had a light graduate teaching load, and provided an increasing percentage of salary support over the years. I was so enthusiastic that I accepted what was offered for “start-up” funds and salary. I knew nothing about negotiation. It was at a top oceanographic institution and I eagerly settled into begin my professional career, writing up research from my postdocs, and applying for grants from Sea Grant and the NSF. I was working hard, providing about half of my own salary, although not the desired 85%, writing proposals, publishing papers, teaching, and doing what I considered to be “cutting-edge” research—directly estimating competition between fish larvae and jellies (Purcell and Grover, 1990) and estimating growth rates of jellies with RNA/DNA ratios. Nevertheless, ten months after I began, the dean, who was not the biologist who hired me, called me into his office to say that he was not renewing my contract. I was speechless! I thought I had 7 years to prove myself, the typical duration before tenure review. After his pronouncement, the dean asked me what I did for research. I did not know what, if anything, I had done wrong. I had not considered that this new dean, a physical oceanographer, did not know me and might not realize the importance of my work. In hindsight, I realize it is important to ensure that the administrators know you and your work. I had not even been there long enough to be reviewed by the department. I started looking at job advertisements. In conversation after a seminar I gave to the group “The Association of Women in Science,” they recommended that I forgo a complaint – “It will ruin your career.” Their lessons were that such problems are not rare in academia and that it is better not to complain. Meanwhile, two of the proposals I had submitted were awarded. One of those was thanks to an invitation from Charlie Miller to study grazing by salps in the subarctic Pacific (Purcell and Madin, 1991; Madin and Purcell, 1992; Madin et al., 1997). Although I had never thought of myself as a role model, one of my proposals was funded by an NSF program Research Opportunities for Women (ROW), which emphasized that seeing women faculty encouraged women students. This was a new perspective for me as a faculty member. Following the awards announcements, the dean came to my office to say he was tearing up the letter of termination. Thus, I learned the supreme importance of Funding. While I recovered emotionally and continued working, I was offered two positions for which I had applied. Although I had my original job back, because of the apparent emphasis of the administrators on funding, I did not think I had a future there. I was not interested in spending all of my time writing proposals. I accepted one of the positions, which seemed too good to be true—a hard money, mostly research professorship. When I interviewed, I saw that the scientists covered the various trophic levels and I was told they were “famous for working together.” That sounded great to me! I had realized that one person cannot be an expert on everything and that improved understanding results from collaborations among researchers. This unexpected stimulus to look for employment fortuitously resulted in a much better position for me. I also got married then. I thought that the busy schedules of two academic scientists might cause too much stress in a family and married someone with a permanent job, no after-hours work, and no travel. My husband has provided the stable home environment that was necessary given my often intense work and many research expeditions. Assistant to full professor (East Coast, 1987–2000) I began my second tenure-track position, where I remained through full professor. My background was organismal in focus and I learned much more about oceanography and ecosystem function than I had known. The state Sea Grant office funded me to join with ongoing local research. At the beginning of my time there, I was invited to go on day-cruises for ongoing research funded by NSF, Sea Grant, and by local state agencies. I really appreciate the help those professors gave me. I learned about the local environment and research topics from participating in their trips. I also connected with “fish people”; Ed Houde and I directly estimated mortality of fish eggs and larvae (21% and 41%, respectively) attributable to jellyfish and ctenophores (Purcell et al., 1994). I also attended numerous scientific conferences where my philosophy was to learn as much as possible. At the questions after a talk on Alaskan walleye pollock, one person in the audience asked what jellyfish did in that system. The speaker, Dave Duffy, said he did not know, but they had been unsuccessful in finding someone to determine that. After the talk, I ran up to him and said, “I can do jellyfish for you.” That began some substantial funding with the Apex Predators Ecosystem eXperiment (APEX) and trips to Prince William Sound, Alaska, one of the most beautiful environments I have seen. Thus, I learned the importance of recognizing a good opportunity and following through. Not only did I enjoy the Alaskan trips, but I was able to estimate feeding by jellyfish in situ (Purcell, 2003), dietary overlap among jellyfish and forage fish (Purcell and Sturdevant, 2001) and to quantify aerial and underwater video sampling of jellyfish aggregations (Purcell et al., 2000). I am grateful to those coauthors that were willing to share their data with me. During this part of my career, I became convinced of the importance of communicating science to children and the public. That was, in part, because the institution had an environmental education centre and also because I saw how passionate my own small children were about nature and learning, and how passionate parents are about their children. By teaching children about the environment, it seemed possible to influence whole families when they heard what had been learned. My younger son had quickly influenced our family to become more careful with recycling waste and eating healthy foods. Therefore, I have volunteered to teach about jellyfish to pre-, grade-, and high-schoolers, and accepted every request to speak to the public and to be interviewed by the media. I think jellies are excellent teaching tools for all ages! Early in my career, a director told me that the responsibility of having students is second only to that of having children. I have remembered that advice and given students and my children top priority. One model of a research program that I had observed and did not want to implement was for the professor to manage numerous students to conduct their research. By contrast, I believe it is important for students to develop their own research ideas, not to follow a specified plan. Therefore, I have hired technicians to help with my research activities and mentored few students relative to some faculty. Over time at this institution, I noticed both subtle and overt actions by some faculty indicating to me that I was being excluded from internal funding and external proposals. This was unexpected, based on their previous statement about “being famous for working together” and my own expectations. Other faculty said I was not mistaken or alone—a small group of male professors controlled everything. Two examples of overt actions follow. I had a co-PI who told me they would not do the research they were funded to do; instead, my group had to do their work. Having contributed the jellyfish section to another proposal that was funded, I was told by the lead-PI, “There is no money for you.” Again, I expected equitable treatment from others. Such experiences made me angry and unhappy. I now understand that when “one is contributing to a proposal, it is prudent to be very clear on how funding is to be allocated.” After waiting for another funding cycle and writing a full proposal, I received money from NSF to conduct the jellyfish part of the research. Although earlier I had been told I could ruin my career by complaining, I went to my superiors, who said “those things happen all the time” and that I should not to talk to anyone outside the institution about this, presumably to protect its reputation. I never heard anything more about these conversations. This behaviour from other faculty and the leadership was not what I expected, given my upbringing with fair and equal treatment of everyone and my lack of political skills. In the spring of 1997, I began communicating with a compliance officer of the university’s Office of Human Relations (OHR), who urged me to work with them to investigate and resolve my concerns. I did not pursue the OHR investigation because, at exactly that time, I was diagnosed with cancer. I think that the stress of the difficulties at work, in addition to stresses of striving for tenure and raising a family, contributed to my cancer (e.g. Reiche et al., 2004). I spent six months undergoing medical treatments. During the one week each month I was too sick from chemotherapy to work, I directed my thinking to winning my health battle. I believe I am a stronger person psychologically from this experience. I consider myself cured (now 20 years post-treatment). During my treatment and recovery from cancer, with the help of students (Christine Baier, Karla Heidelberg, and Xiping Ma), co-PIs (Denise Breitberg and Mary Beth Decker), and assistants (Kim Black, Rob Condon, Margaret Leonard, and Dave Nemazie), work on the above projects continued. The First International Symposium on Jellyfish Blooms in Gulf Shores, Alabama presented the opportunity to write review papers (Purcell and Arai, 2001; Purcell et al., 2001b). I also began to study climate effects with an NSF grant on the temperature, salinity, and food effects on jellyfish polyps (Purcell et al., 1999 b), although my treatment and recovery slowed publication down. The effects of the environment on jellyfish populations became a new focus of my career. Also during that time, I was asked by a retired NSF Division director to investigate how to increase the participation of minorities and women in marine science. I spent a year half working half time interviewing people and attending conferences. One conference (Women in Science and Engineering: Choices for Success) featured notable women scientists, some of whom were Nobel laureates, who had been at the original conference 25 years earlier at which the problems of women in science were explored. The consensus and report of the conference was, depressingly, that things had not improved much for women since 1973 (Selby, 1999). They concluded that discrimination against women in 1998 was just more subtle than in 1973; specifically, groping and lewd jokes were replaced by subtle behaviours like I had experienced that can reduce women’s opportunities, salary, and advancement. I had explored this topic during my recovery in part to see if I was wrong in my perceptions. I was not. Three years into my cancer recovery, I called personnel at my institution and learned that I had 18 years until I could retire. I was afraid I could not face the same environment daily for 18 more years without the cancer returning. So I relinquished my hard money, full professorship and my husband and I moved where we wanted to live. Life on soft money (West Coast, 2001–2014) My husband and I relocated where we both could work. The director of the lab I chose had a reputation for helping people succeed. Therefore, I moved with only a promise of space and the ability to write proposals, with hopes that I would be incorporated into the existing faculty. Again, in retrospect, I was unrealistically optimistic. I began to involve myself in teaching and research in my new environment. I felt like a life-saving ring had been thrown to me when was awarded a NSF grant designed to help mid-career women in science (ADVANCE). The grant involved undergraduate and graduate students in studies of the effects of climate factors on population dynamics of jellyfish. My graduate student, Amanda Winans, showed that asexual reproduction by Aurelia labiata polyps was not affected by low pH, as would occur as a result of ocean acidification (Winans and Purcell, 2010). Another graduate student, Rich Hoover, and an assistant, Nate Schwarck, helped to make possible an in situ study of A. labiata polyps over 3 years (Purcell et al., 2009). That study showed greater asexual production of jellyfish (strobilation) in years with high temperature, salinity, and light. I believe that sunlight is necessary to stimulate a light-sensitive hormone (like melatonin, which is present in all single- and multicelled organisms), to synchronize the timing of strobilation in Aurelia spp., and possibly reproduction in other jellies, with the seasonal production cycle (Purcell, 2007). A recent molecular study found that decreased temperature in darkness triggered strobilation in Aurelia aurita, suggesting that light was not necessary (Fuchs et al., 2014). While preparing a seminar, I realized that most previous long-term studies of jellyfish showed that blooms were related to high temperatures and salinities (Purcell, 2005). Since that review, more species have shown increased asexual reproduction with higher temperature (Purcell et al., 2012b), but all species have not (e.g. Widmer et al., 2016). Preliminary analysis of the data from the 5-year-long monthly sampling for jellies, zooplankton, temperature, and salinity showed that Aequorea victoria medusa populations also increased significantly with increasing spring temperatures (Purcell, 2017). The ADVANCE program began because NSF identified a mass disappearance of women from academia at mid-career, as well as at postdoctoral and employment transitions in science (e.g. NSF, 2003; Monroe et al., 2008). Unfortunately, the objective of that program later changed, because it was impossible to insist that the institutions hire those women, which proved to be true for me. After 7 years of working hard to be part of the teaching and research, and repeated reminders to the director that I wanted and needed financial support and to be part of the faculty, I still had only support from grants. When the ADVANCE grant ran out of money, I spent the next year writing eight proposals and getting rejections from various agencies. I knew that in order to survive on soft money, many scientists must apply for funding on other topics; however, I still believed that jellyfish needed my full attention and did not want to change fields after 20 years working on them. I concluded that I could not survive on 100% soft money at my third institution. There were no faculty members with related interests with whom I could write collaborative proposals. In order to obtain substantial salary, I needed more than one NSF grant, which seemed impossible to get. Not only had I repeatedly failed, but I was disheartened by what I observed by serving on an NSF proposal review panel. We were told that over half of the original 400 proposals were discarded without review. Of those reviewed by the panel, only the four best-prepared proposals received funding, not necessarily those most scientifically compelling. Two of the four were deep-sea submersible projects that were very expensive, either of which could have funded my research for my entire career. I thought that I had better things to do with my life than spend it writing proposals that would not be funded. I realized that there were no hopes of salary or a position at my home institution. The next step for many academics is in administration. Although I had no administrative experience, I thought I could direct a department and applied for a few positions, for which I was not chosen. In hindsight, I realized that I probably would be a terrible director because personnel and financial problems are very stressful for me. Nevertheless, I recommend that management skills should be developed during every career (Goodall, 2009). Fortuitously, Wen-tseng (James) Lo invited me to be a visiting professor for a year at the National Sun Yat-Sen University in Kaohsiung, Taiwan. I took a leave of absence from my home institution, assuring the director at my home institution that I intended to return to work there. Although the year in Taiwan was a necessary financial and morale boost, it took me away from my young family. During that year, I continued to think about the damage that humans inflict on the environment. In my career to that point, I had encountered the effects of humans on the oceans that might benefit jelly populations—first with overfishing of herring, which removes zooplanktivores that could compete with jellies, then eutrophication and hypoxia, which may favour jellies over fish (Purcell et al., 1999a, 2001a; Condon et al., 2001), then an oil spill, and global warming (Purcell et al., 1999 b). That year I wrote one of my most-cited papers, which includes a dramatic demonstration of aquaculture enhancing a jellyfish population in Taiwan (Purcell et al., 2007). I also worked closely with a student on her master’s research to test the combined effects of temperature and light on asexual reproduction by A. aurita polyps (Liu et al., 2009). I returned to my home institution and reiterated to the director that I considered it my base and wanted to work there. Nevertheless, new opportunities to work overseas appeared. First, Tom Doyle and Graham Hays invited me for two simultaneous projects in 2008–2009: GillPath on the gill pathologies in finfish in Ireland and EcoJel on jellyfish ecology in the Irish Sea (Purcell et al., 2012a). I learned a lot about the problems jellyfish cause for aquaculture (Purcell et al., 2013). In 2008, I was invited to give a presentation in Barcelona, Spain, which led to the most rewarding collaboration of my career with Verónica Fuentes and her team, all of whom became like family, at the Institute of Marine Sciences in Barcelona. There I learned about the various Mediterranean jellyfish during many visits and projects over the next decade (e.g. Purcell et al., 2014). Sun Song, Director of the Chinese Academy of Sciences, Institute of Oceanology in Qingdao, China recognized the great importance of jellyfish and invited me to assist his team during 2012 on their project “The Key Processes, Mechanisms and Ecological Consequences of Jellyfish Blooms in China Coastal Waters.” Here I learned first-hand about the giant jellyfish, Nemopilema nomurai (e.g. Dong et al., 2014) and aquaculture production of the edible jellyfish (Rhopilema esculentum) for the Chinese fishery. Stefano Piraino had invited me to be on the Research Advisory Board of the European project VECTORS (Vectors of Change in Oceans and Seas Marine Life, Impact on Economic Sectors), which led to my employment at the University of Salento in Lecce, Italy in 2013–2015. My education on the ecology and problems of northern and southern European jellyfish continued (e.g. Purcell et al., 2015). Thus, since 2006 I have been an itinerant scientist. I have felt much more appreciated in other countries than I have in the United States. Although I have not directly sponsored any students working on jellyfish since 2009 due to my itinerant situation, I have enjoyed working with many students in several countries, some of whom continue as colleagues (e.g. Martell et al., in press). I think that my wide travels have given me a broader view of jellies and research on them than if I had remained at one institution. I have worked in waters from the Arctic to the tropics, from the surface to the deep sea, on the North Pacific coasts from Alaska, United States to the Gulf of California, Mexico, and in China and in the Atlantic Ocean from the open ocean to estuaries, in European waters from Norway to the Mediterranean, and in the Benguela Current, Namibia, and on many different gelatinous taxa (siphonophores, hydrozoans, scyphozoans, ctenophores, and salps), their food populations, and environmental conditions. I recommend that everyone see different environments. If you think you know what you will find from place to place or from year to year, you will be proven wrong. Life without funding (West Coast, 2015–present) Working in other scientists’ programs ultimately became unsatisfying for me, because I was unable to direct any of the research. I returned to the United States in 2015, thinking from prior conversations with the director that my position at my home institution would be intact, enabling me to write proposals and have students. That summer, I was contacted by Christopher Krembs at the Washington Department of Ecology (WDE), who flew over Puget Sound monthly to take aerial pictures of harmful algal blooms visible at the surface (EOPS, 2017). They had noticed large whitish patches that were determined to be aggregations of jellyfish (A. labiata). He planned day trips on a small boat and hoped to estimate the numbers of jellyfish in aggregations while aerial photographs were taken. That would accomplish something I had advocated for years (Purcell, 2009); specifically, to sample in situ simultaneously with aerial photographic sampling, which had been accomplished only once to date (Graham et al., 2003). The spectacular images and video of the jellyfish aggregations attracted attention from local TV News (Morrow, 2015), National Geographic, Nature Magazine (Hamilton, 2016) and, secondarily, European public stations PBS (PBS, 2016) and Deutschlandfunke (Röhrlich, 2017). I contacted the research office at my university to submit proposals for funding from the partners involved (WDE and NOAA); however, to my surprise, I no longer had a faculty appointment. Subsequently, the new director of my previous laboratory denied my application for appointment saying there was no place for me. Again, I encountered a new director who did not know me or my work. I was given an appointment in another department that does not give me space, permit me to submit proposals, do research, teach, or have students. I have only e-mail and library access. I do not know the reasons behind this decision. The administrators (politicians) to whom I have spoken have not explained their actions. Meanwhile, I received a Fulbright Scholarship to learn to estimate respiration enzymatically, which could advance the study of trophic ecology in all marine organisms (Purcell et al., 2010; Packard, 2017). My goal is to teach this method to the next generation of marine ecologists. Unfortunately, at the time of this publication, I have no ability to do that. The current academic environment The rules of proper scientific ethics were written decades ago and recently updated (AGU, 2017). The problems I have written about are not limited to male perpetrators or female victims. Two of my women “friends” took my ideas, which I had suggested we work on together, and submitted proposals without me. I know men and women who have experienced worse injustices than I have. Why are such abuses typically not reported? I think the victims fear retaliation and damage to their careers; the battle probably seems impossible to win and no one wants to take the risk. Silence makes this possible, so perhaps it is time to speak out. One could argue that learning about unprofessional behaviour from academic scientists could discourage young scientists from continuing in the field; however, such actions apparently are present in most work environments. My view has always been that “knowledge is power.” If I had known what to expect, perhaps I could have been prepared and would not have been so badly affected. I strongly recommend that young scientists educate themselves about these issues. In my experience, students often do not think there is any problem; indeed, I did not notice a problem until I had a faculty position. It is 2017, nearly 50 years after that first conference on women in science. The inequality between genders is still a problem (Cohen and Duberley, 2017). My essay is intended only to relate my own experiences, not to be a comprehensive review on inequality in science, on which extensive information exits (e.g. NSF, 2003, 2017; NAS, 2007; Monroe et al., 2008; Ceci et al., 2014; Williams et al., 2014; European Union, 2017). Although women continue to disappear disproportionately from the academic “pipe-line,” the latest NSF study (NSF, 2017) suggests that there is some increase in the numbers of women compared to previous years (NSF, 2003). Hopefully, these next generations of scientists will be more inclusive. The economic environment has changed and may have made the academic environment even more difficult. The financial “collapses” in the past two decades in many countries have caused funding budgets to shrink. Thus, more people are competing for fewer resources, perhaps making the research environment less collegial and discouraging others from entering it or continuing. Likewise, institutions have suffered funding cuts. At my own institution, the total cut was 30% of the public funding in 2009 and more cuts followed (Long, 2011). Ways that institutions have chosen to offset such losses are to discontinue programs or departments and increase tuition rates and the numbers of students (Long, 2011). The result for the faculty is more teaching and higher expectations to perform. Academic institutions may have always been run like businesses and I just did not notice. But what I have seen is that security in academic positions may become rare. For example, legislation in two states in the United States proposes to eliminate tenure at public colleges and universities (Flaherty, 2017). I have observed that scientists with doctorates are hired to teach on annual contracts without long-term benefits or security. They provide the institutions with a low-cost, low-risk, high-quality work force at the expense of academic freedom (Braben, 2008). Fish and jellyfish For decades, the goal of fisheries managers around the world has been to predict the harvest of the following years. When I was an early career scientist, fisheries managers focused on the size of individual stocks (e.g. herring) to predict harvest and much effort was directed toward understanding variability in recruitment (e.g. Houde, 2009). Those methods have repeatedly proven unsuccessful; hence, thousands of scientists in various organizations realize that single-stock management is inadequate. The new trend toward the ecosystem approach to fisheries (EAF) or ecosystem-based management (EBM) sounds more reasonable. The transition has been very slow, however, with limited success (papers in Link and Browman, 2014, 2017). After six decades in the fisheries field, Brian Rothschild wrote, “there has been little progress towards the goal of predicting stock variability and recruitment” (Rothschild, 2015). In papers I have read, fisheries researchers mostly continue to consider only fish as consumers and do not mention jellies, even though they have been demonstrated for decades to be important zooplankton consumers whose diets overlap with fish larvae and schooling forage fish, as well as consuming fish eggs and larvae (e.g. Purcell, 2017). Jellyfish are also key components in ecosystem analyses featuring fish (e.g. Samhouri et al., 2009; Coll and Libralato, 2012; Ruzicka et al., 2012, 2016; Robinson et al., 2014). There is a glimmer of hope that jellyfish might be included in the thinking of some fisheries researchers; for example they appear in the annual Bering Sea Report Card (Zador et al., 2017). A few fisheries scientists have recognized the importance of jellies and made invaluable contributions to promoting their study. Daniel Pauly was notable for alerting all of us to “fishing down the food web” to create “jellyfish soup” (Pauly et al., 1998). He included jellyfish experts, myself included, in Ecopath workshops to populate the models with the best information on gelatinous species available around North America, beginning with the Strait of Georgia in 1995. Ric Brodeur has directed programs at two National Marine Fisheries Centers in the northwestern United States that have produced two of the longest-term semi-quantitative datasets on jellyfish that exist (e.g. Brodeur et al., 2008a, b). He recently urged fisheries researchers to include jellyfish bycatch in their data, which could be accomplished with little additional effort or cost (Brodeur et al., 2016). Another champion, Hermes Mianzan (deceased 9 July 2014) at the National Institution of Fisheries, Argentina promoted the study of jellyfish by fisheries scientists on both coasts of South America (e.g. Mianzan et al., 2014; Quiñones et al., 2015). Those of us who began studying jellies 40 or more years ago have always said that they are important. Only now that jellyfish seem to be increasing and causing more problems for humans have they become popular research topics. In terms of knowledge about jellies, this increased attention is very good news. Because so much is unknown about jellies, we need many strong, determined scientists to study them. I have continued to study jellies despite so many difficulties because I love to keep learning new things. My next adventure will be at the Federal University of Rio Grande in Brazil. I am excited by the many new things I will learn! Acknowledgements I especially want to thank all those who aided me during my career and broadened my experiences and perspectives: my gradate and postdoctoral advisors, those who taught me about ichthyoplankton, those who invited me to work with them or go on cruises, and many students and helpers in the field and lab. I thank the U.S. National Sea Grant Program and the U.S. NSF for support and promoting research opportunities for young people and gender equity in science. I also thank Howard Browman for the invitation to present this essay and for his and the reviewers’ comprehensive and constructive comments for improvement. I dedicate this paper to my friend and colleague, Mary N. Arai (deceased 6 September 2017), who was a pioneer for women scientists and the interactions between jellyfish and fish. Footnotes Food for Thought articles are essays in which the author provides their perspective on a research area, topic, or issue. They are intended to provide contributors with a forum through which to air their own views and experiences, with few of the constraints that govern standard research articles. This Food for Thought article is one in a series solicited from leading figures in the fisheries and aquatic sciences community. The objective is to offer lessons and insights from their careers in an accessible and pedagogical form from which the community, and particularly early career scientists, will benefit. The International Council for the Exploration of the Sea (ICES) and Oxford University Press are pleased to make these Food for Thought articles immediately available as free access documents. References Able K. W. 2016 . Natural history: an approach whose time has come, passed, and needs to be resurrected . ICES Journal of Marine Science , 73 : 2150 – 2155 . Google Scholar CrossRef Search ADS AGU (American Geophysical Union) . 2017 . Scientific integrity and professional ethics. https://ethics.agu.org/files/2013/03/Scientific-Integrity-and-Professional-Ethics.pdf. Badré S. 2014 . Bioactive toxins from stinging jellyfish . Toxicon , 91 : 114 – 125 . Google Scholar CrossRef Search ADS PubMed Braben D. W. 2008 . Scientific Freedom: The Elixir of Civilization . Wiley Press , Hoboken . 200 pp. Google Scholar CrossRef Search ADS Brodeur R. D. , Decker M. B. , Ciannelli L. , Purcell J. E. , Bond N. A. , Stabeno P. J. , Acuna E. , et al. 2008a . The rise and fall of jellyfish in the Bering Sea in relation to climate regime shifts . Progress in Oceanography , 77 : 103 – 111 . Google Scholar CrossRef Search ADS Brodeur R. D. , Link J. S. , Smith B. E. , Ford M. D. , Kobayashi D. R. , Jones T. T. 2016 . Ecological and economic consequences of ignoring jellyfish: a plea for increased monitoring of ecosystems . Fisheries , 41 : 623 – 629 . Google Scholar CrossRef Search ADS Brodeur R. D. , Suchman C. L. , Reese D. , Miller T. , Daly E. 2008b . Spatial overlap and trophic interactions between fish and large jellyfish in the Northern California Current . Marine Biology , 154 : 649 – 659 . Google Scholar CrossRef Search ADS Carrette T. , Alderslade P. , Seymour J. 2002 . Nematocyst ratio and prey in two Australian cubomedusans, Chironex fleckeri and Chiropsalmus sp . Toxicon , 40 : 1547 – 1551 . Google Scholar CrossRef Search ADS PubMed Ceci S. , Ginther D. K. , Kahn S. , Williams W. M. 2014 . Women in academic science: a changing landscape . Psychological Science in the Public Interest , 15 : 75 – 141 . Google Scholar CrossRef Search ADS PubMed Cohen L. , Duberley J. 2017 . Gender equality: universities are still all talk and too many trousers. www.timeshighereducation.com/comment/gender-equality-universities-are-still-all-talk-and-too-many-trousers. Coll M. , Libralato S. 2012 . Contributions of food-web modelling for an ecosystem approach of marine resource management in the Mediterranean Sea . Fish and Fisheries , 13 : 60 – 88 . Google Scholar CrossRef Search ADS Condon R. H. , Decker M. B. , Purcell J. E. 2001 . Effects of low dissolved oxygen on survival and asexual reproduction of scyphozoan polyps (Chrysaora quinquecirrha) . Hydrobiologia , 451 : 89 – 95 . Google Scholar CrossRef Search ADS Corrales-Ugalde M. , Colin S. P. , Sutherland K. R. 2017 . Nematocyst distribution corresponds to prey capture location in hydromedusae with different predation modes . Marine Ecology Progress Series , 568 : 101 – 110 . Google Scholar CrossRef Search ADS de Lafontaine Y. , Leggett W. C. 1987 . Effect of container size on estimates of mortality and predation rates in experiments with macrozooplankton and larval fish . Canadian Journal of Fisheries and Aquatic Sciences , 44 : 1534 – 1543 . Google Scholar CrossRef Search ADS Dong J. , Sun M. , Purcell J. E. , Chai Y. , Zhao Y. , Wang A. 2015 . Effect of salinity and light intensity on somatic growth and podocyst production in the giant jellyfish Nemopilema nomurai (Scyphozoa: Rhizostomeae) . Hydrobiologia , 754 : 75 – 83 . Google Scholar CrossRef Search ADS EOPS (Eyes Over Puget Sound) . 2017 . Marine water quality monitoring. http://www.ecy.wa.gov/programs/eap/mar_wat/surface.html. European Union . 2017 . 2017 report on equality between women and men in the European Union. 68 pp, doi: 10.2838/52591. Flaherty C. 2017 . Killing tenure. https://www.insidehighered.com/news/2017/01/13/legislation-two-states-seeks-eliminate-tenure-public-higher-education. Fuchs B. , Wang W. , Graspeuntner S. , Li Y. , Insua S. , Herbst E-M. , Dirksen P. , et al. 2014 . Regulation of polyp-to-jellyfish transition in Aurelia aurita . Current Biology , 24 : 263 – 273 . Google Scholar CrossRef Search ADS PubMed Goodall A. H. 2009 . Socrates in the Boardroom Why Research Universities Should Be Led by Top Scholars . Princeton University Press , Princeton. 200 pp. Graham W. M. , Martin D. L. , Felder D. L. , Asper V. L. , Perry H. M. 2003 . Ecological and economic implications of the tropical jellyfish invader, Phyllorhiza punctata von Lendenfeld, in the northern Gulf of Mexico . Biological Invasions , 5 : 53 – 69 . Google Scholar CrossRef Search ADS Hamilton G. 2016 . The secret lives of jellyfish . Nature , 531 : 432 – 434 . Google Scholar CrossRef Search ADS PubMed Houde E. D. 2009 . Recruitment variability. In Reproductive Biology of Fishes: Implications for Assessment and Management , pp. 91 – 171 . Ed. by Jakobsen T. , Fogarty M. , Megrey B. , Moksness E. . Wiley-Blackwell, Chichester , West Sussex . xi + 424 pp. Link J. S. , Browman H. I. 2014 . Integrating what? Levels of marine ecosystem-based assessment and management . ICES Journal of Marine Science , 71 : 1170 – 1173 . Google Scholar CrossRef Search ADS Link J. S. , Browman H. I. 2017 . Operationalizing and implementing ecosystem-based management . ICES Journal of Marine Science , 74 : 379 – 381 . Google Scholar CrossRef Search ADS Liu W-C. , Lo W-T. , Purcell J. E. , Chang H-H. 2009 . Effects of temperature and light intensity on asexual reproduction of the scyphozoan, Aurelia aurita (L.) in Taiwan . Hydrobiologia , 616 : 247 – 258 . Google Scholar CrossRef Search ADS Long K. 2011 . WWU redefines itself in light of state cuts. www.seattletimes.com/seattle-news/wwu-redefines-itself-in-light-of-state-cuts/. Mackie G. O. , Pugh P. R. , Purcell J. E. 1987 . Siphonophore biology . Advances in Marine Biology , 24 : 97 – 262 . Google Scholar CrossRef Search ADS Madin L. P. , Purcell J. E. 1992 . Feeding, metabolism and growth of Cyclosalpa bakeri in the subarctic Pacific . Limnology and Oceanography , 37 : 1236 – 1251 . Google Scholar CrossRef Search ADS Madin L. P. , Purcell J. E. , Miller C. B. 1997 . Abundance and grazing effects of Cyclosalpa bakeri in the subarctic Pacific . Marine Ecology Progress Series , 157 : 175 – 183 . Google Scholar CrossRef Search ADS Mapstone G. 2015 . Correction: Global diversity and review of Siphonophorae (Cnidaria: Hydrozoa) . PLoS ONE , 10 : e0118381 . Google Scholar CrossRef Search ADS PubMed Martell L. F. , Bracale R. , Carrion S. A. , Purcell J. E. , Lezzi M. , Gravili C. , Piraino S. , et al. Successional dynamics of marine fouling Hydrozoa (Cnidaria) at a finfish aquaculture facility in central Mediterranean Sea . PLoS One , in press. Mianzan H. , Quiñones J. , Palma S. , Schiariti A. , Acha E. M. , Robinson K. L. , Graham W. M. 2014 . Chrysaora plocamia: a poorly understood jellyfish from South American waters In Jellyfish Blooms , pp. 219 – 236 . Ed. by Pitt K. A. , Lucas C. H. . Springer Science+Business Media , Dordrecht . xi + 304 pp. Google Scholar CrossRef Search ADS Monroe K. , Ozyurt S. , Wrigley T. , Alexander A. 2008 . Gender equality in academia: bad news from the trenches, and some possible solutions . Perspectives on Politics , 6 ( 2 ): 215 . Google Scholar CrossRef Search ADS Morrow A. 2015 . Jellyfish boom prompts research by plane and boat. http://king5.com/story/tech/science/environment/2015/08/05/jellyfish-ecology-noaa-puget-sound/31194035/ NAS (National Academy of Science) . 2007 . Beyond Bias and Barriers: Fulfilling the Potential of Women in Academic Science and Engineering . The National Academies Press , Washington, DC . xxvii + 317 pp. NSF (National Science Foundation, Division of Science Resources Statistics) . 2003 . Gender differences in the careers of academic scientists and engineers: a literature review, NSF03-322, Project Officer, Rapport, A. I., Arlington, VA. NSF (National Science Foundation, National Center for Science and Engineering Statistics) . 2017 . Women, minorities, and persons with disabilities in science and engineering: 2017. Special Report NSF 17-310. Arlington, VA. Available at www.nsf.gov/statistics/wmpd/. Packard T. T. 2017 . From Thoreau’s woods to the Canary Islands: exploring ocean biogeochemistry through enzymology . ICES Journal of Marine Science , 75 : 912 – 992 . Google Scholar CrossRef Search ADS Pauly D. , Christensen V. , Dalsgaard J. , Froese R. , Torres F. Jr. 1998 . Fishing down marine food webs . Science , 279 : 860 – 863 . Google Scholar CrossRef Search ADS PubMed PBS (Public Broadcasting Station) . 2016 . In the increasingly damaged sea, one animal is thriving. http://www.pbs.org/newshour/bb/increasingly-damaged-sea-one-animal-thriving/ Purcell J. E. 1977a . Diet of large and small individuals of the sea anemone Metridium senile (Coelenterata, Actiniaria) . Bulletin of the Southern California Academy of Science , 76 : 168 – 172 . Purcell J. E. 1977b . Aggressive function and induced development of catch tentacles in the sea anemone Metridium senile (Coelenterata, Actiniaria) . Biological Bulletin , 153 : 355 – 368 . Google Scholar CrossRef Search ADS Purcell J. E. 1980 . Influence of siphonophore behavior upon their natural diets: evidence for aggressive mimicry . Science , 209 : 1045 – 1047 . Google Scholar CrossRef Search ADS PubMed Purcell J. E. 1981a . Dietary composition and diel feeding patterns of epipelagic siphonophores . Marine Biology , 65 : 83 – 90 . Google Scholar CrossRef Search ADS Purcell J. E. 1981b . Feeding ecology of Rhizophysa eysenhardti, a siphonophore predator of fish larvae . Limnology and Oceanography , 26 : 424 – 432 . Google Scholar CrossRef Search ADS Purcell J. E. 1981c . Selective predation and caloric consumption by the siphonophore Rosacea cymbiformis in nature . Marine Biology , 63 : 283 – 294 . Google Scholar CrossRef Search ADS Purcell J. E. 1982 . Feeding and growth in the siphonophore Muggiaea atlantica . Journal of Experimental Marine Biology and Ecology , 62 : 39 – 54 . Google Scholar CrossRef Search ADS Purcell J. E. 1983 . Digestion rates and assimilation efficiencies of siphonophores fed zooplankton prey . Marine Biology , 73 : 257 – 261 . Google Scholar CrossRef Search ADS Purcell J. E. 1984a . The functions of nematocysts in prey capture by epipelagic siphonophores (Coelenterata, Hydrozoa) . Biological Bulletin , 166 : 310 – 327 . Google Scholar CrossRef Search ADS Purcell J. E. 1984b . Predation on fish larvae by Physalia physalis, the Portuguese man of war . Marine Ecology Progress Series , 19 : 189 – 191 . Google Scholar CrossRef Search ADS Purcell J. E. 1985 . Predation on fish eggs and larvae by pelagic cnidarians and ctenophores . Bulletin Marine Science , 37 : 739 – 755 . Purcell J. E. 1989 . Predation by the hydromedusa Aequorea victoria on fish larvae and eggs at a herring spawning ground in British Columbia . Canadian Journal of Fisheries and Aquatic Sciences , 46 : 1415 – 1427 . Google Scholar CrossRef Search ADS Purcell J. E. 1990 . Soft-bodied zooplankton predators and competitors of larval herring (Clupea harengus pallasi) at herring spawning grounds in British Columbia . Canadian Journal of Fisheries and Aquatic Sciences , 47 : 505 – 515 . Google Scholar CrossRef Search ADS Purcell J. E. 2003 . Predation on zooplankton by large jellyfish (Aurelia labiata, Cyanea capillata, Aequorea aequorea) in Prince William Sound, Alaska . Marine Ecology Progress Series , 246 : 137 – 152 . Google Scholar CrossRef Search ADS Purcell J. E. 2005 . Climate effects on formation of jellyfish and ctenophore blooms: a review . Journal of the Marine Biological Association of the United Kingdom , 85 : 461 – 476 . Google Scholar CrossRef Search ADS Purcell J. E. 2007 . Environmental effects on asexual reproduction rates of the scyphozoan, Aurelia labiata . Marine Ecology Progress Series , 348 : 183 – 196 . Google Scholar CrossRef Search ADS Purcell J. E. 2009 . Extension of methods for jellyfish and ctenophore trophic ecology to large-scale research . Hydrobiologia , 616 : 23 – 50 . Google Scholar CrossRef Search ADS Purcell J. E. 2017 . Successes and challenges in jellyfish ecology: examples from Aequorea spp . Marine Ecology Progress Series , AdvanceView: 1. doi: 10.3354/meps12213. Purcell J. E. , Anderson P. A. V. 1995 . Electrical responses to water-soluble components of fish mucus recorded from the cnidocytes of a fish predator, Physalia physalis . Marine and Freshwater Behaviour and Physiology , 26 : 149 – 162 . Google Scholar CrossRef Search ADS Purcell J. E. , Arai M. N. 2001 . Interactions of pelagic cnidarians and ctenophores with fishes: a review . Hydrobiologia , 451 : 27 – 44 . Google Scholar CrossRef Search ADS Purcell J. E. , Baxter E. J. , Fuentes V. L. 2013 . Jellyfish as products and problems of aquaculture. Chapter 13. In Advances in Aquaculture Hatchery Technology , pp. 404 – 443 . Ed. by Allan G. , Burnell G. . Woodhead Publishing, Cambridge. Series in Food Science, Technology and Nutrition , Number 242. Google Scholar CrossRef Search ADS Purcell J. E. , Breitburg D. L. , Decker M. B. , Graham W. M. , Youngbluth M. J. , Raskoff K. A. 2001a . Pelagic cnidarians and ctenophores in low dissolved oxygen environments: a review. In Coastal Hypoxia: Consequences for Living Resources and Ecosystems , pp. 77 – 100 . Ed. by Rabalais N. N. , Turner R. E. . American Geophysical Union , Washington, DC . Coastal and Estuarine Studies 58. Google Scholar CrossRef Search ADS Purcell J. E. , Brown E. D. , Stokesbury K. D. E. , Haldorson L. H. , Shirley T. C. 2000 . Aggregations of the jellyfish Aurelia labiata: abundance, distribution, association with age-0 walleye pollock, and behaviors promoting aggregation in Prince William Sound, Alaska, USA . Marine Ecology Progress Series , 195 : 145 – 158 . Google Scholar CrossRef Search ADS Purcell J. E. , Clarkin E. , Doyle T. K. 2012a . Foods of Velella velella (Cnidaria: Hydrozoa) in algal rafts and its distribution in Irish seas . Hydrobiologia , 690 : 47 – 55 . Google Scholar CrossRef Search ADS Purcell J. E. , Cresswell F. P. , Cargo D. G. , Kennedy V. S. 1991 . Differential ingestion and digestion of bivalve larvae by the scyphozoan Chrysaora quinquecirrha and by the ctenophore Mnemiopsis leidyi . Biological Bulletin , 180 : 103 – 111 . Google Scholar CrossRef Search ADS PubMed Purcell J. E. , Fuentes V. , Atienza D. , Olariaga A. , Tilves U. , Colahan C. , Gili J-M. 2012b . Temperature effects on asexual reproduction rates of scyphozoan polyps from the NW Mediterranean Sea . Hydrobiologia , 690 : 169 – 180 . Google Scholar CrossRef Search ADS Purcell J. E. , Fuentes V. , Atienza D. , Tilves U. , Astorga D. , Kawahara M. , Hays G. C. 2010 . Use of respiration rates of scyphozoan jellyfish to estimate their effects on the food web . Hydrobiologia , 645 : 135 – 152 . Google Scholar CrossRef Search ADS Purcell J. E. , Grover J. J. 1990 . Predation and food limitation as causes of mortality in larval herring at a spawning ground in British Columbia . Marine Ecology Progress Series , 59 : 55 – 67 . Google Scholar CrossRef Search ADS Purcell J. E. , Hoover R. A. , Schwarck N. T. 2009 . Interannual variation of strobilation of the scyphozoan Aurelia labiata in relation to polyp density, temperature, salinity, and light conditions . Marine Ecology Progress Series , 375 : 139 – 149 . Google Scholar CrossRef Search ADS Purcell J. E. , Kremer P. 1983 . Feeding and metabolism of the siphonophore Sphaeronectes gracilis . Journal of Plankton Research , 5 : 95 – 106 . Google Scholar CrossRef Search ADS Purcell J. E. , Madin L. P. 1991 . Diel patterns of migration, feeding, and spawning by salps in the Subarctic Pacific . Marine Ecology Progress Series , 73 : 211 – 217 . Google Scholar CrossRef Search ADS Purcell J. E. , Malej A. , Benović A. 1999a . Potential links of jellyfish to eutrophication and fisheries. In Ecosystems at the Land-Sea Margin: Drainage Basin to Coastal Sea , pp. 241 – 263 . Ed. by Malone T. C. , Malej A. , Harding L. W. Jr. , Smodlaka N. , Turner R. E. . American Geophysical Union , Washington, DC . Coastal and Estuarine Studies 55. Google Scholar CrossRef Search ADS Purcell J. E. , Milisenda G. , Rizzo A. , Carrion S. A. , Zampardi S. , Airoldi S. , Zagami G. , et al. 2015 . Digestion and predation rates of zooplankton by the pleustonic hydrozoan Velella velella and widespread blooms in 2013 and 2014 . Journal of Plankton Research , 37 : 1056 – 1067 . Google Scholar CrossRef Search ADS Purcell J. E. , Mills C. E. 1988 . The correlation between nematocyst types and diets in pelagic hydrozoa. In The Biology of Nematocysts , pp. 463 – 485 . Ed. by Hessinger D. A. , Lenhoff H. M. , Academic Press , San Diego . xii + 600 pp. Google Scholar CrossRef Search ADS Purcell J. E. , Nemazie D. A. , Dorsey S. E. , Houde E. D. , Gamble J. C. 1994 . Predation mortality of bay anchovy (Anchoa mitchilli) eggs and larvae due to scyphomedusae and ctenophores in Chesapeake Bay . Marine Ecology Progress Series , 114 : 47 – 58 . Google Scholar CrossRef Search ADS Purcell J. E. , Shiganova T. A. , Decker M. B. , Houde E. D. 2001b . The ctenophore Mnemiopsis in native and exotic habitats: U.S. estuaries versus the Black Sea basin . Hydrobiologia , 451 : 145 – 176 . Google Scholar CrossRef Search ADS Purcell J. E. , Sturdevant M. V. 2001 . Prey selection and dietary overlap among zooplanktivorous jellyfish and juvenile fishes in Prince William Sound, Alaska . Marine Ecology Progress Series , 210 : 67 – 83 . Google Scholar CrossRef Search ADS Purcell J. E. , Tilves U. , Fuentes V. , Milisenda G. , Olariaga A. , Sabatés A. 2014 . Digestion times and predation potentials of Pelagia noctiluca eating fish larvae and copepods in the NW Mediterranean Sea . Marine Ecology Progress Series , 510 : 201 – 213 . Google Scholar CrossRef Search ADS Purcell J. E. , Uye S-I. , Lo W-T. 2007 . Anthropogenic causes of jellyfish blooms and direct consequences for humans: a review . Marine Ecology Progress Series , 350 : 153 – 174 . Google Scholar CrossRef Search ADS Purcell J. E. , White J. R. , Nemazie D. A. , Wright D. A. 1999b . Temperature, salinity and food effects on asexual reproduction and abundance of the scyphozoan Chrysaora quinquecirrha . Marine Ecology Progress Series , 180 : 187 – 196 . Google Scholar CrossRef Search ADS Quiñones J. , Mianzan H. , Purca S. , Robinson K. L. , Adams G. D. , Acha E. M. 2015 . Climate-driven population size fluctuations of jellyfish (Chrysaora plocamia) off Peru . Marine Biology , 162 , doi: 10.1007/s00227-015-2751-4. Reeve M. R. , Syms M. A. , Kremer P. 1989 . Growth dynamics of a ctenophore (Mnemiopsis) in relation to variable food supply. I. Carbon biomass, feeding, egg production, growth and assimilation efficiency . Journal of Plankton Research , 11 : 535 – 552 . Google Scholar CrossRef Search ADS Reiche E. M. V. , Nunes S. O. V. , Morimoto H. K. 2004 . Stress, depression, the immune system, and cancer . Lancet Oncology , 5 : 617 – 625 . Google Scholar CrossRef Search ADS PubMed Robinson K. L. , Ruzicka J. J. , Decker M. B. , Brodeur R. D. , Hernandez F. J. , Quiñones J. , Acha E. M. 2014 . Jellyfish, forage fish, and the world’s major fisheries . Oceanography , 27 : 104 – 115 . Google Scholar CrossRef Search ADS Röhrlich D. 2017 . Begegnung mit dem Schwarm. http://www.deutschlandfunk.de/unter-quallen-begegnung-mit-dem-schwarm.740.de.html? dram: article_id=395466 Rothschild B. J. 2015 . Food for Thought: On the birth and death of ideas in marine science . ICES Journal of Marine Science , 72 : 1237 – 1244 . Google Scholar CrossRef Search ADS Ruzicka J. J. , Brodeur R. D. , Emmett R. L. , Steele J. H. , Zamon J. E. , Morgan C. A. , Thomas A. C. , et al. 2012 . Interannual variability in the northern California Current food web structure: changes in energy flow pathways and the role of forage fish, euphausiids, and jellyfish . Progress in Oceanography , 102 : 19 – 41 . Google Scholar CrossRef Search ADS Ruzicka J. J. , Daly E. A. , Brodeur R. D. , Peters D. P. C. 2016 . Evidence that summer jellyfish blooms impact Pacific Northwest salmon production . Ecosphere , 7 : e01324. Google Scholar CrossRef Search ADS Samhouri J. F. , Levin P. S. , Harvey C. J. 2009 . Quantitative evaluation of marine ecosystem indicator performance using food web models . Ecosystems , 12 : 1283 – 1298 . Google Scholar CrossRef Search ADS Selby C. C. 1999 . Women in Science and Engineering: choices for Success . Annals of the New York Academy of Sciences , 869 : 1 – 261 . Google Scholar CrossRef Search ADS PubMed Widmer C. L. , Fox C. J. , Brierley A. S. 2016 . Effects of temperature and salinity on four species of northeastern Atlantic scyphistomae (Cnidaria: Scyphozoa) . Marine Ecology Progress Series , 559 : 73 – 88 . Google Scholar CrossRef Search ADS Williams J. C. , Phillips K. W. , Hall E. V. 2014 . Double jeopardy? Gender bias against women of color in science. www.worklifelaw.org. Winans A. K. , Purcell J. E. 2010 . Effects of pH and temperature on asexual reproduction and statolith formation of the scyphozoan, Aurelia labiata . Hydrobiologia , 645 : 39 – 52 . Google Scholar CrossRef Search ADS Zador S. G. , Holsman K. K. , Aydin K. Y. , Gaichas S. K. 2017 . Ecosystem considerations in Alaska: the value of qualitative assessments . ICES Journal of Marine Science , 74 : 421 – 430 . Google Scholar CrossRef Search ADS © International Council for the Exploration of the Sea 2018. All rights reserved. For Permissions, please email: [email protected] This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)