Evaluating the effectiveness of real-time closures for reducing susceptibility of small fish to capture

Evaluating the effectiveness of real-time closures for reducing susceptibility of small fish to... Abstract Real-time spatial management in fisheries, a type of dynamic ocean management, uses nearly real-time data collection and dissemination to reduce susceptibility of certain species or age classes to being caught in mixed fisheries. However, as with many fisheries regulations, it is difficult to assess whether such a regulation can produce tangible results on population dynamics. In this study, we take advantage of a rare opportunity in which data regarding real-time closures (RTCs) are available for 1990–2014 alongside annual estimates of fishing mortality for three species (Atlantic cod, haddock, and herring) and catch for four species (all plus saithe) in Icelandic fisheries management. We use time series analyses to assess whether RTCs work as expected and yield a lower susceptibility of small fish to being caught, indicated by lower catch levels and selectivities (as estimated from fishing mortalities) in years with more closures. Results indicate that haddock and herring followed this pattern, but only under conditions of generally high fishing mortality. This study represents the first time evidence has been presented that real-time fishery closures can have a beneficial effect on population dynamics, but also suggests that results differ among species. Introduction Spatial management has long been used in fisheries to reduce discards and bycatch, thereby increasing fishery selectivity or protecting important habitat (Dunn et al., 2011). When designed appropriately, fishery closures that are static in time and space can be effective management tools (O’Keefe et al., 2014). However, they have also been criticized as ineffective, often due to the mismatch between the spatial and temporal scales on which closures are drawn vs. the scales on which resource users and ecological processes operate (Bailey et al., 2010; O’Keefe et al., 2014; Maxwell et al., 2015; Dunn et al., 2016). Dynamic ocean management (DOM) is a promising solution to this mismatch, as it is “management that uses near real-time data to guide the spatial distribution of commercial activities” (Lewison et al., 2015, p. 486). It can theoretically respond faster to changes in the ecosystem than can typical static management schemes, which generally assimilate data on an annual basis (Dunn et al., 2014, 2016; Kraak et al., 2014). As technological innovation speeds and eases the burdens of data collection and sharing, DOM has become an increasingly viable option to facilitate adaptive ecosystem-based management (Lewison et al., 2015; Maxwell et al., 2015; Dunn et al., 2016). DOM includes mechanisms such as spatial real-time fishery closures (RTCs), move-on rules (Dunn et al., 2014, 2016; Bjorkland et al., 2015), risk pools (Holland, 2010; Holland and Jannot, 2012), and real-time incentive-based management (Kraak et al., 2014, 2015). The actual implementation of such mechanisms vary widely, from top-down government-sponsored regulations to industry-sponsored voluntary programs, as well as co-management arrangements that fall in-between (Little et al., 2015). RTCs have gained popularity in the past decade as a fisheries management tool for reducing discards or avoiding hotspots of unwanted species (O’Keefe et al., 2014; Lewison et al., 2015; Little et al., 2015). Real-time oceanographic or catch data may be used to fit spatial models aimed at predicting the occurrence of species to avoid, or in other instances, relatively simple data are needed (Lewison et al., 2015). For example, RTCs are often event-triggered, such that a single reported high catch of a bycatch species or age class will induce a no-fishing zone for several weeks. Similarly, move-on rules require that fishers move a certain distance before continuing operations (Dunn et al., 2014, 2016; Bjorkland et al., 2015). In top-down orchestrated closures, reported catch data are generally gathered from observer coverage or inspections; in voluntary situations, personal catch data are exchanged for access to larger pools of collective catch data and/or quota (Little et al., 2015). Although DOM has the benefit of reacting quicker to stock dynamics and is a clear example of adaptive management, it has not been widely implemented because it poses a variety of challenges for success (Hobday et al., 2014; Lewison et al., 2015). First, data must be collected, potentially processed, and finally disseminated on fast time-scales, which may be relatively labour-intensive and costly for both industry participants and managers (Lewison et al., 2015; Little et al., 2015). Second, it often relies on stakeholders and managers to work together to set up a system in which resource use data are shared, requiring negotiation and agreement on the terms of use (e.g. Holmes et al., 2011; O’Keefe and DeCelles, 2013; Wallace et al., 2015), within a context where laws and policies support such use (Hobday et al., 2014). Third, the correct incentive structure must be in place in order for the benefits to appear great enough for both initiation and compliance (Holmes et al., 2011; Hobday et al., 2014; Lewison et al., 2015). Finally, although it has a greater potential to match spatiotemporal ecosystem dynamics than static annual closures, some knowledge regarding ecosystem dynamics (e.g. site fidelity or movement patterns) is needed to optimally design the closure system for the purpose at hand (O’Keefe et al., 2014; Kraak et al., 2015; Dunn et al., 2016). Therefore, it is useful to know whether DOM is effective, and under what terms of implementation. Overcoming the first three challenges indicates that a program can be initiated and is at least operating as intended. Clearly, voluntary programs yield incentives for negotiation and participation (second and third challenges) or else they would not exist. For example, participants of risk pools use more of the high-value quotas without using more constraining species quota, which would risk an early fishery-wide closure (Labrum and Oberhoff, 2013; De Alessi et al., 2014). However, evaluating the fourth challenge requires linking the programme function with intended ecological effects. For example, some programmes can be deemed ecologically effective through demonstrated avoidance of habitat or bycatch, while simultaneously beneficial to fishers by enabling retention of catch that would have been foregone, avoidance of nuisances in catches, or avoidance of penalties (Dunn et al., 2014, 2016; Bjorkland et al., 2015; Lewison et al., 2015; O’Keefe et al., 2014; Dunn et al., 2016). In the cases of long-standing government-sponsored regulations, there is often less impetus to evaluate whether a regulation is ecologically effective, as long as it is perceived to be beneficial with no negative side effects (i.e. “if it ain’t broke don't fix it”). Evidence for effectiveness in this form may be simply that the system is at least operating as intended, without analysing ecological consequences. For example, pilot projects indicate that fishers do indeed move from areas of higher to lower concentrations of the bycatch species when given catch information (Needle and Catarino, 2011; Bethoney et al., 2013). Evidence for effectiveness in reducing fishing mortality or other environmental impacts, on the other hand, is more difficult to trace due to the lack of a controlled experiment: typically no fishing occurs within closures to test the closure’s effectiveness (Bailey et al., 2010). Closures may be ineffective if fishers move straight back into closures after re-opening or the closure length is too short (Needle and Catarino, 2011). Furthermore, evaluating the effectiveness of any regulation in fisheries is often difficult because of confounding factors, such as simultaneously introduced regulations or economic and environmental changes. In Norway, a long-standing system of RTCs, along with effort reduction and favourable environmental conditions, are thought to have contributed to the recovery of cod in the Norwegian and Barents Seas, but these three factors cannot be disentangled (Johnsen and Eliasen, 2014; Eliasen, 2014). Nonetheless, there exists some limited evidence for reductions in catch of certain species or age classes (Bailey et al., 2010; Holmes et al., 2011; O’Keefe and DeCelles, 2013; O’Keefe et al., 2014), and a theoretical study indicates that real-time management may be even more effective at controlling harvest rates than traditional annual harvest control rules (Kraak et al., 2015). In no cases, however, has an RTC system been linked to population dynamics, thereby yielding direct evidence for fulfilling its ultimate mandate. In this study, we attempt to do just this by testing whether there is a correlation between usage of the Icelandic RTC system and fishing selectivity or catch of young fish. RTCs have occurred in Icelandic fisheries as far back as the late 1960s for the Atlantic cod Gadus morhua and 1970s for the Atlantic herring Clupea harengus, although other species generally have closures beginning in the late 1980s. Today, the Icelandic RTC system applies to nine demersal groundfish species, four pelagic fish species, Norwegian lobster (Nephrops norvegicus), and shrimp (Pandalus borealis) (Schopka, 2007). In most cases, to trigger a closure, catch of a certain size range of the species exceeds a certain percentage of the total catch (e.g. 25% under 55 cm for cod). Closures are generally detected by onboard observers, who subsample and measure the catch in 15–20% of the cod fishery (Little et al., 2015), although other mechanisms are possible. The Icelandic Marine Research Institute (MRI) then has the authority to define a closure area, which must be advertised on radio by the coastguard and the broadcast by the government radio station before the closure takes effect. The closure can take effect within less than an hour from advertisement, and lasts for 14 days, but may be extended for another 7 days if necessary. To properly assess whether the RTC system works (i.e. whether it is having ecological effects), it would be best to test differences in mortality before and after implementation of a regulation, or between stocks or areas that differ in a specific regulation, while controlling for all other covariates. However, the introduction of a regulation as a controlled experiment is exceedingly rare in management systems in fisheries. As a result, the task often relies on correlations, as is the case in this study. Luckily, extensive time series of records regarding Icelandic RTCs allow for enough data to test for correlations. In most other locations, RTCs have only been implemented in the past decade; only RTCs in the Barents and Norwegian Sea cod and haddock fishery have been implemented for a comparable length of time (Little et al., 2015). As a result, the Icelandic system yields an unprecedented opportunity to analyse a unique dataset. Here, we add to the literature surrounding RTCs by providing the first analysis of a direct effect of RTCs on catch and/or selectivity of four species in the Icelandic fishery: Atlantic cod, haddock (Melanogrammus aeglefinus), herring, and saithe (Pollachius virens). These species were chosen because they all have a consistent record of relatively high numbers of closures within the last 30 years, as well as data-rich stock assessments that yield time series of fishing mortality and biomass estimates. We test for a relationship between the numbers of RTCs and catch levels (adjusted by biomass) or selectivity-at-age estimates (calculated from fishing mortalities) for the youngest age class possible of each species. A negative relationship would support the idea that RTCs reduce susceptibility of small fish to being caught. We also include two covariates, the yearly effect of fishing mortality (which we term “relative total effort”) and adult biomass, in an attempt to control for possible confounding factors associated with changes in fishing pressure and the general state of the fishery. Methods Data sources Databases at the MRI were used to obtain data on the number of RTCs (Rt,s,g) by calendar year (t), species (s), and gear (g). Trawls, longlines, and gillnets caught the vast majority of cod, haddock and saithe, and comprise roughly <1% of those data. Bottom trawls, mid-water trawls, and demersal seine were summed to form the “trawl” RTC category; only longline composed the “longline” RTC category. Handline RTCs were excluded. RTC data for herring were not split by gear as they generally included only herring-specific or mid-water gears (i.e. herring trawls, herring nets, herring or non-specific mid-water trawls, and purse seines). Not enough gillnet RTCs occurred for any species for analysis, due to very low selectivity for small fish. Survey abundance indices at age (a) were taken from acoustic surveys for herring and spring groundfish surveys for all other species (It,a,s). Catches from all gears ( Ct,a,s), weights (Wt,a,s), and fishing mortality estimates (Ft,a,s) were obtained from stock assessments completed by the MRI. These data were also available in national annual stock assessment and ICES reports (Anonymous, 2015; ICES, 2015). Survey abundance indices were transformed into reference biomass indices after multiplying by weight-at-age and summing over the appropriate age range of available survey data (Bt,4+,s for cod, saithe, and herring, and Bt,3+,s for haddock). Fishing mortality estimates were decomposed into an age-separable component of fishing mortality (here referred to as selectivity, St,a,s) and a yearly scaling component of fishing mortality (here referred to as relative total effort, Et,s) within each species, by dividing by the selectivity within a year via the equation Et,s = Ft,a,s/St,a,s. The selectivity of the fully selected age(s) within each year then equalled 1, and relative total effort for that year was the maximum fishing mortality value. Relative total effort therefore scales overall fishing mortalities across years, partially indicating differences in fishing effort as estimated in the stock assessment models, although it is not a true measure of fishing effort. Similarly, age-specific selectivity varies across years and reflects the degree to which a certain age class was susceptible to being caught relative to other age classes within that year, due to gear, fish vs. fishing distributions, or other factors. Background on Icelandic fisheries In Iceland, herring was the first species to have ITQs established in 1979 (Jakobsson and Stefánsson, 1999), although most species have been managed under an individual transferable quota (ITQ) system since 1991 (Arnason, 2005). Fishers are required to keep detailed logbooks for each haul/set. There appears to be little unreported catch as landings generally agree with logbook-based catches, although there are still low levels of discards in the form of high-grading for the largest stocks (0–5%, ICES, 2015; Pálsson et al., 2015;). Annual stock assessments are completed in conjunction with the International Council for Exploration of the Seas (ICES), resulting in total allowable catches. There is a full discard ban, and there are technically no minimum landing sizes, but gear restrictions and the RTC system are intended to reduce selectivity on small-sized fish (ICES, 2015). The number of RTCs has greatly increased over the time period analysed (Figures 1–4, panels a). Permanent time/area closures are likewise utilized to protect spawning or juvenile aggregations, and a series of consecutive RTCs may allow the Minister of Fisheries and Agriculture to establish a longer closed area (Schopka, 2007). Further details can be found in Supplementary Material Section S1.1. Figure 1. View largeDownload slide Time series of numbers of RTCs by gear (a), selectivity estimates for ages of recruitment (3) and entry (4) (b), relative total efforts (left axis) and reference biomass indices (right dotted axis) (c), and catch numbers by gear (d) are shown for cod. Figure 1. View largeDownload slide Time series of numbers of RTCs by gear (a), selectivity estimates for ages of recruitment (3) and entry (4) (b), relative total efforts (left axis) and reference biomass indices (right dotted axis) (c), and catch numbers by gear (d) are shown for cod. Figure 2. View largeDownload slide Time series of numbers of RTCs by gear (a), selectivity estimates for ages of recruitment (3) and entry (4) (b), relative total efforts (left axis) and reference biomass indices (right dotted axis) (c), and catch numbers by gear (d) are shown for haddock. Figure 2. View largeDownload slide Time series of numbers of RTCs by gear (a), selectivity estimates for ages of recruitment (3) and entry (4) (b), relative total efforts (left axis) and reference biomass indices (right dotted axis) (c), and catch numbers by gear (d) are shown for haddock. Figure 3. View largeDownload slide Time series of numbers of RTCs by gear (a), selectivity estimates for ages of recruitment (3) and entry (4) (b), relative total efforts (left axis) and reference biomass indices (right dotted axis) (c), and catch numbers by gear (d) are shown for herring. Survey data are missing from 1995. Figure 3. View largeDownload slide Time series of numbers of RTCs by gear (a), selectivity estimates for ages of recruitment (3) and entry (4) (b), relative total efforts (left axis) and reference biomass indices (right dotted axis) (c), and catch numbers by gear (d) are shown for herring. Survey data are missing from 1995. Figure 4. View largeDownload slide Time series of numbers of RTCs by gear (a), selectivity estimates for ages of recruitment (3) and entry (4) (b), relative total efforts (left axis) and reference biomass indices (right dotted axis) (c), and catch numbers by gear (d) are shown for saithe. Figure 4. View largeDownload slide Time series of numbers of RTCs by gear (a), selectivity estimates for ages of recruitment (3) and entry (4) (b), relative total efforts (left axis) and reference biomass indices (right dotted axis) (c), and catch numbers by gear (d) are shown for saithe. Cod Substantial effort reductions began in the 1990s (ICES, 2015) and a harvest control rule was introduced in the 1995–1996 fishing season. Subsequent modifications to the harvest control rule decreased effort slowly over the following decades (Figure 1b and c). Biomass levels have been increasing, especially after 2000 (Figure 1c and d), during which longline catch has increased as trawl and gillnet catches have declined (Figure 1d, ICES, 2015). RTCs have been implemented since the late 1960s, but the longline closures have increased as trawl closures have decreased as a reflection of recent gear shifts (Figure 1a and d). RTCs are currently triggered by catch numbers consisting of 25% or greater cod that are <55 cm. Haddock Landings of Icelandic haddock are currently low (Figure 2d) due to excessive fishing mortality and low survival have caused decreases in the stock until 2011. Since then, reduced fishing mortality has helped to slow decreases (ICES, 2015). Similar to cod, longline catch has increased while trawl and gillnet catches have decreased (Figure 2d, ICES, 2015). As a result, longline RTCs have increased while trawl RTCs have decreased (Figure 2a). Haddock RTCs are triggered by catches composed of more than 30% of haddock <45 cm. Herring Prior to 2008 fishing mortalities were steadily higher than target fishing mortalities, but high stock levels persisted until an Ichthyophonous infection reduced stock levels during the following 2 years (Figure 3, ICES, 2015). Most catch originates from pelagic trawls or purse seines (ICES, 2015). Other regulations are in place to reduce selectivity of herring, including certain gear and bycatch limitations, as well as RTCs that have been implemented since the 1970s (ICES, 2015). RTCs are currently triggered when >25% of a catch is composed of herring <27 cm. Saithe The harvest rate for saithe peaked in the 1990s, but since then has stabilized around a similar rate as intended by a newly introduced harvest control rule (20% of the reference biomass, ICES, 2015). Supporting trends can be seen in effort and catch data (Figure 4b and d). Saithe selectivity estimates are fixed as constant in the stock assessment within three periods: one occurring before 1996, from 1997 to 2003, from 2004 to present, leading to artificially large changes between periods (Figure 4b). Therefore, we did not analyse saithe selectivity estimates, only catch. We also included a dummy variable to reflect changes in catch from the periods 1990–2003 to 2004–2014 as a process unrelated to RTCs, as the stock assessment model indicates a large increase in selectivity around this time (Figure 4b). Small-sized saithe have recently been targeted to a greater extent than in the past, partially as a result of the reduction of gillnets that target larger fish (ICES, 2015). RTCs are triggered when catch of saithe <55 cm compose at least 30% of a catch in numbers. Data analyses Detecting susceptibility of small fish to being caught RTCs are designed in general to reduce susceptibility of small fish to being caught. Therefore, if RTCs are successfully implemented, there should be more RTCs in years when catches with large proportions of small fish are more frequent (positive relationship). However, if RTCs are furthermore effective at reducing the susceptibility of small fish to being caught (i.e. they are working as intended), then this relationship should reverse: there should be lower values of selectivity and catch for small fish in years when there are a high number of closures, relative to other years (negative relationship). We expect model results to be quite similar among selectivity and catch, but we use both because there are pros and cons for using each source as data. Although changes in selectivity are ultimately what we are trying to detect, selectivity estimates come from stock assessment outputs, which means that their accuracies rely on model assumptions and any restrictions put on the model. For example, saithe selectivities were constrained to be constant within three time periods in the saithe stock assessment, so these were excluded from analyses. On the other hand, cod selectivities were estimated as random effects in statistical catch-at-age models, so although they are not true data, they are thought to be reliable as they are not constrained to take a specified shape (e.g. a dome or sinusoidal shape). Herring and haddock selectivities were estimated in a statistical virtual population analysis (VPA). Although some formulations of statistical VPAs can impose an asymptotic selectivity curve, the oldest age class selectivity values are generally the ones that are constrained, while younger age classes, such as those analysed here, should be less affected (Butterworth and Rademeyer, 2008). In contrast, catch-at-age data are not stock assessment output, despite being extrapolated from total catches and sub-sampled age distributions. However, catch is positively related to fishing effort and biomass of that age class, so it may only be used as an indicator of selectivity if changes if effort and biomass are accounted for. Effort was included as a covariate in main data analyses (see following section), whereas the effect of biomass of the same age class was removed a priori by first regressing catch over the same-age biomass index. Catch residuals ( C˙t,a,s) were then used as the dependent variable instead of raw catch to reflect catch level relative to stock status. Statistical framework Four further statistical issues were apparent from the outset. First, although RTCs are meant to reduce susceptibility of recruits and younger fish to being caught, precision of selectivity estimates from recruitment-aged fish, indicated in stock assessments, are quite low due to generally low mortality rates relative to other age classes. Therefore, we focused on predicting selectivity and catch of the age class above recruitment, which we refer to as “entry-aged” and are generally the first year of inclusion in reference biomass (i.e. age 4 for cod, saithe, and herring, or 3 for haddock). In general, entry-aged selectivity estimates had higher precision than recruitment-aged estimates and were correlated with recruitment-aged estimates due to overlapping size ranges between recruit and entry-aged fish (Figures 1b–4b). As a result, there is likely to be a direct effect of RTCs on this age class as well. Second, throughout the history of any fishery, stock biomasses and fishing effort levels will fluctuate in response to environmental or economic changes. Such fluctuations can affect the regulatory environment or stock age structure, as well as the behaviour of fishermen (e.g. through changes in gear or location), and consequently can also affect observed selectivity patterns. It is therefore useful to include variables reflecting the general state of the fishery to avoid confounding these changes with the effect of RTC closures. We included both relative total fishing effort and reference biomass in the analysis as covariates for this purpose. Third, as these are time series, autocorrelation may need to be accounted for. To do this, we fit ARIMA time-series models with a matrix of external regressors (also known as ARIMAX models, Equations 1–3), and searched for the best combination of external regressors and autocorrelation parameters to explain patterns in selectivity or catch. This search included all possible combinations of predictors (i.e. the number of trawl, long-line, or herring-gear RTCs, and the covariates relative total effort, reference biomass, and a dummy variable for saithe reflecting a change in conditions from 2003 to 2004) and their interactions with the constraint that main effects were included when interactions composed of them were included. Both trawl and long-line RTCs were included for cod and haddock, only trawl RTCs were included for saithe, and only herring-gear RTCs were included for herring. All analyses were restricted to 1990–2014, which coincides with the broad establishment of the ITQ system (Arnason, 2005). All variables were log-transformed to increase normality except effort variables. Unity was added to guard against zeros causing failure of definition. ARIMAX model selection For each possible combination of predictors, all possible combinations of 5 or fewer auto-regressive (ARp, max(p) = 5) or moving average (MAq, max(q) = 5) parameters were tested, as well as the order of differencing (d) needed to obtain independent, stationary residuals. For cod, we restricted the search to include only d = 1, as preliminary analyses indicated that single-lag differencing could avoid the inclusion of a dummy variable used to reflect a large change in selectivity due to the introduction of the harvest control rule (Figure 1), which would extend the exhaustive best-model search to an infeasible level. AR and MA parameters quantify autocorrelation within the error of an initial regression (Equations 1 and 2), with random error quantified as normal and independently distributed (Equation 3). In such a regression, the predictions at time t ( yt') are determined by a set of k predictors ( xt,k'), their regression coefficients ( β0… βk), and autocorrelated error ( nt'). The prime notation ( yt', xt', nt') indicates that model fitting may occur on data that were differenced, the order d of which reflects how many times single-lag differencing was needed for residuals to become stationary according to a Kwiatkowski–Phillips–Schmidt–Shin (KPSS) test. For example, the single-order differencing used for cod (d = 1) indicates yt' =  yt-yt-1 (Hyndman and Athanasopoulos, 2013). Model fitting was implemented in R (R Core Team, 2015), mainly using the “auto.arima” function in the “forecast” package (Hyndman and Khandakar, 2008). The best model was chosen as that which minimized Aikaike Information Criteria corrected for small sample sizes (AICc). The maximum and minimum residual from final models were tested as outliers using a Grubbs test in R package “outliers” (Komsta, 2011), which tests the difference between the residual and the mean over all residuals, divided by the standard deviation, against a t-distribution. The statistic U is the ratio of residual variances with and without the tested residual included. We use null and residual deviances, standard outputs of generalized linear models but not ARIMA models, to report a convenient standard measure of model fit assuming normally distributed error: percent variation explained ([null deviance−residual deviance]/null deviance). We also show t-statistics (coefficients divided by their standard errors), as tests indicate a difference from 0, and corresponding P-values with degrees of freedom (n – d – p where n is the number of years of data and p is the total of linear and autocorrelation coefficients).   yt'=β0+ β1x1,t'+…+βkxk,t'+nt' (1)  nt'=c+AR1nt-1'+…+ARpnt-p'+MA1et-1+…+MAqet-q+et (2)  et∼N(0,σ2) (3) Spurious correlations in time series The last statistical issue noticeable from the outset was that, even after including covariates that reflect the general state of the fishery, any relationship we find between the number of RTCs and selectivity or catch could be due to spurious correlations with historical unobserved effects. For example, if observer coverage is correlated with the number of RTCs, perhaps greater observer coverage is the main mechanism reducing small-fish susceptibility to being caught rather than RTCs themselves. To rule out as many spurious correlations as possible, we repeat the presented analysis in Supplementary Material Section S2, but after removing the effects of a number of additional effects from the predictor (number of RTCs) before using it as a predictor in main analyses. In particular, we removed the effects of annual variation in observer coverage, recruitment, and gear-specific effort on number of RTCs, and correlations among number of RTCs from different gear types. However, approaching the problem as we did in Supplementary Material did not produce sufficiently different model results to justify presenting the more complex results here. We instead highlighted the main differences in “Results” section. Results Detecting susceptibility of small fish to being caught For all species, entry-aged catch was regressed over entry-aged biomass with positive effects, so catch was adjusted for biomass levels by taking residuals (Table 1). Table 1. Regression results from best models chosen by AICc to predict entry-aged catch numbers (Ct,a,s) using entry-aged biomass (Bt,a,s). Species (s)  Predictors  Estimate  S.E.  t  P  % Expl.  Null df  Res. df  Residual formed  Cod  Intercept  0.0000  0.1697  0.0000  1.0000  31.00  24  23  C˙t,a=4,cod    Bt,4,cod  0.5568  0.1732  3.2145  0.0038          Haddock  Intercept  0.0000  0.1108  0.0000  1.0000  70.58  24  23  C˙t,a=3,haddock    Bt,3,haddock  0.8401  0.1131  7.4284  <0.0001          Herring  Intercept  0.0000  0.1704  0.0000  1.0000  30.41  24  23  C˙t,a=4,herring    Bt,4,herring  0.5514  0.1739  3.1702  0.0043          Saithe  Intercept  0.0000  0.1674  0.0000  1.0000  32.88  24  23  C˙t,a=4,saithe    Bt,4,saithe  0.5734  0.1708  3.3568  0.0027          Species (s)  Predictors  Estimate  S.E.  t  P  % Expl.  Null df  Res. df  Residual formed  Cod  Intercept  0.0000  0.1697  0.0000  1.0000  31.00  24  23  C˙t,a=4,cod    Bt,4,cod  0.5568  0.1732  3.2145  0.0038          Haddock  Intercept  0.0000  0.1108  0.0000  1.0000  70.58  24  23  C˙t,a=3,haddock    Bt,3,haddock  0.8401  0.1131  7.4284  <0.0001          Herring  Intercept  0.0000  0.1704  0.0000  1.0000  30.41  24  23  C˙t,a=4,herring    Bt,4,herring  0.5514  0.1739  3.1702  0.0043          Saithe  Intercept  0.0000  0.1674  0.0000  1.0000  32.88  24  23  C˙t,a=4,saithe    Bt,4,saithe  0.5734  0.1708  3.3568  0.0027          Model parameter estimates, standard errors (S.E.), t-statistics (t), P-values (p), percent variation explained as calculated from null and residual deviances (% Expl.), and deviance degrees of freedom (df). Residuals formed ( C˙t,a,s) were predicted in ARIMA models. ARIMAX model selection Using our model selection method, we found two practical problems. First, because the level of differencing is chosen after selecting the set of regression variables (using “auto.arima”), a peculiar situation can arise in which the level of differencing varies depending on the regression variables included. Every time a dataset is differenced, the time series is shortened by one data point, so that AICc values are no longer comparable: the shorter dataset automatically decreases the magnitude of the log likelihood. In these situations, there may be incomparable competing models at different levels of differencing. We chose to report the best model as that with more differencing (i.e. fewer data points), as these can be thought of as more conservative: by differencing the data, stationarity is removed a priori rather than as a result of correlations with regression parameters. However, in practice, competing models with different levels of differencing had very similar results. Second, our model selection method also often tended to select overfit models. Therefore, we added selection criteria to exclude overfit models: the best model must include fewer than seven parameters, including autocorrelation parameters. In practice, the best models left after applying this criterion included far fewer parameters (Table 2). Table 2. Results of the best ARIMA models chosen by AICc to predict entry-aged selectivity (St,a,s) or catch residuals ( C˙t,a,s) by each species s using (1) relative effort (Et,s,), (2) number of gear-specific RTCs (Rt,s,g), and (3) reference biomasses summed over appropriate age ranges (Bt,a,s). Species  Dependent variable  d  Predictors  Estimate  S.E.  t  P  % Expl.  Null df  Res. df  Cod  St,a=4,cod  1  AR1  0.4342  0.1877  2.3130  0.0309  40.05  23  21        Rt,cod,trawl  0.1955  0.0876  2.2328  0.0366              Et,cod  −0.5224  0.1770  −2.9509  0.0076          C˙t,a=4,cod  1  AR1  −0.8515  0.1749  −4.8696  <0.0001  64.49  23  20        AR2  −0.4923  0.1743  −2.8240  0.0105              Rt,cod,trawl  0.4404  0.0905  4.8642  <0.0001              Bt,4+,cod  −0.2425  0.1013  −2.3950  0.0265        Haddock  St,a=3,haddock  0  Rt,haddock,longline  −0.4170  0.1489  −2.8003  0.0102  50.00  24  23        Bt,3+,haddock  −0.4551  0.1489  −3.0562  0.0056          C˙t,a=3,haddock  0  Rt,haddock,longline  0.2602  0.1296  2.0086  0.0576  73.79  24  21        Et,haddock  0.4303  0.1048  4.1053  0.0005              Bt,3+,haddock  −0.4288  0.1128  −3.8020  0.0010              Rt,haddock,longline x Et,haddock  −0.5593  0.1348  −4.1502  0.0005        Herring  St,a=4,herring  0  Rt,herring,herring-gear  0.3534  0.1283  2.7548  0.0116  65.84  23  21        Et,herring  −0.5432  0.1180  −4.6029  0.0001              Rt,herring,herring-gear x Et,herring  −0.3784  0.1328  −2.8500  0.0093          C˙t,a=4,herring  0  Rt,herring,herring-gear  0.4769  0.1342  3.5550  0.0019  65.96  24  21        Et,herring  0.3963  0.1307  3.0327  0.0063              Bt,4+,herring  −0.2905  0.1374  −2.1143  0.0019              Rt,herring,herring-gear x Et,herring  −0.4283  0.1394  −3.0722  0.0063        Saithe  C˙t,a=4,saithe  1  AR  −0.4308  0.1851  −2.3273  0.0295  24.80  23  22        Et,saithe  0.4911  0.2107  2.3306  0.0293        Species  Dependent variable  d  Predictors  Estimate  S.E.  t  P  % Expl.  Null df  Res. df  Cod  St,a=4,cod  1  AR1  0.4342  0.1877  2.3130  0.0309  40.05  23  21        Rt,cod,trawl  0.1955  0.0876  2.2328  0.0366              Et,cod  −0.5224  0.1770  −2.9509  0.0076          C˙t,a=4,cod  1  AR1  −0.8515  0.1749  −4.8696  <0.0001  64.49  23  20        AR2  −0.4923  0.1743  −2.8240  0.0105              Rt,cod,trawl  0.4404  0.0905  4.8642  <0.0001              Bt,4+,cod  −0.2425  0.1013  −2.3950  0.0265        Haddock  St,a=3,haddock  0  Rt,haddock,longline  −0.4170  0.1489  −2.8003  0.0102  50.00  24  23        Bt,3+,haddock  −0.4551  0.1489  −3.0562  0.0056          C˙t,a=3,haddock  0  Rt,haddock,longline  0.2602  0.1296  2.0086  0.0576  73.79  24  21        Et,haddock  0.4303  0.1048  4.1053  0.0005              Bt,3+,haddock  −0.4288  0.1128  −3.8020  0.0010              Rt,haddock,longline x Et,haddock  −0.5593  0.1348  −4.1502  0.0005        Herring  St,a=4,herring  0  Rt,herring,herring-gear  0.3534  0.1283  2.7548  0.0116  65.84  23  21        Et,herring  −0.5432  0.1180  −4.6029  0.0001              Rt,herring,herring-gear x Et,herring  −0.3784  0.1328  −2.8500  0.0093          C˙t,a=4,herring  0  Rt,herring,herring-gear  0.4769  0.1342  3.5550  0.0019  65.96  24  21        Et,herring  0.3963  0.1307  3.0327  0.0063              Bt,4+,herring  −0.2905  0.1374  −2.1143  0.0019              Rt,herring,herring-gear x Et,herring  −0.4283  0.1394  −3.0722  0.0063        Saithe  C˙t,a=4,saithe  1  AR  −0.4308  0.1851  −2.3273  0.0295  24.80  23  22        Et,saithe  0.4911  0.2107  2.3306  0.0293        Both trawl and longline RTCs were tested for haddock and cod; only herring-gear RTCs were tested for herring, and only trawl RTCs were tested for saithe. Autoregression parameters (AR), moving average parameters (MA), and the number of single-lag differences (d) are also shown, as well as linear model parameter estimates, standard errors (S.E.), t-statistics (t), P-values (P), percent variation explained as calculated from null and residual deviances (% Expl.), and deviance degrees of freedom (df). For all models, we interpret only statistically significant results as the presence of an effect. For herring, 2013 selectivity was removed due to a positive outlier test showing extremely low residual selectivity that year in a preliminary model (U = 0.6545, P = 0.0251). All species with models predicting entry-aged catch had a positive effect (haddock, saithe, herring, P ≤ 0.05) or no effect (cod, P > 0.05) of relative total effort (Figure 5b, d, f, and h, Table 2). These results are unsurprising because high effort levels lead to greater catch in general. In contrast, models predicting selectivity included a negative effect of relative total effort (cod and herring, P ≤ 0.05, Figure 5a and e, Table 2) or no effect (haddock, P > 0.05, Figure 5c), indicating that years of high relative total effort were generally more likely to produce low susceptibility of small fish to being caught. Surprisingly, although biomass of the same age class was accounted for a priori, there was also a negative effect of reference biomass in models of cod, haddock, and herring catch, as well as haddock selectivity. High biomass years therefore also led to generally low catches of entry-aged fish. Results indicated that using ARIMAX models were helpful to counter some nuisances in fitting time series data, such as dealing with non-stationarity through differencing and accounting for autocorrelation in error by including autoregressive or moving average parameters (Table 2). Figure 5. View large Download slide Contours represent predictions of entry-aged selectivity estimates and catch residuals resulting from the best-estimated ARIMA models presented in Supplementary Table S2. Main effects of predictors (excluding dummy variables and reference biomass) can be seen as linear changes in the prediction surface along the y-axis (the number of RTCs) or x-axis (relative total effort). Biomass was held constant at the mean value. Vertical gradients from dark colors at high values of the y-axis toward light colors at the low values of the y-axis correspond with our expectation, that selectivity is lower in years with more RTCs. The nonlinear contours found in panels d–f reflect effects of interactions, resulting in saddle shapes with areas of low selectivity estimates and catch values in upper right and lower left panel corners. Black points show the data used to estimate models. If the number of RTCs was not included as an effect, data were plotted along a horizontal 0 line, whereas data were plotted along a vertical 0 line if relative total effort was not included as an effect. Figure 5. View large Download slide Contours represent predictions of entry-aged selectivity estimates and catch residuals resulting from the best-estimated ARIMA models presented in Supplementary Table S2. Main effects of predictors (excluding dummy variables and reference biomass) can be seen as linear changes in the prediction surface along the y-axis (the number of RTCs) or x-axis (relative total effort). Biomass was held constant at the mean value. Vertical gradients from dark colors at high values of the y-axis toward light colors at the low values of the y-axis correspond with our expectation, that selectivity is lower in years with more RTCs. The nonlinear contours found in panels d–f reflect effects of interactions, resulting in saddle shapes with areas of low selectivity estimates and catch values in upper right and lower left panel corners. Black points show the data used to estimate models. If the number of RTCs was not included as an effect, data were plotted along a horizontal 0 line, whereas data were plotted along a vertical 0 line if relative total effort was not included as an effect. The number of trawl RTCs (but not longline RTCs) was positively correlated with the susceptibility of small cod to being caught (i.e. selectivity estimates, Figure 5a) as well as cod catch (Figure 5b, Table 2). In contrast, the number of longline RTCs was negatively correlated with susceptibility of small haddock (Figure 5c), and had a negative interaction with relative total effort in the model of haddock catch (Figure 5d). This interaction resulted in a saddle-shaped contour, and indicated a reduction in small-haddock catch with greater numbers of longline RTCs only in years with relatively high effort (Figure 5d). Although the number of trawl RTCs was not selected in the haddock models as a significant effect, analyses in Supplementary Material show it to be correlated with longline RTCs (Supplementary Table S3). In both herring models, the inclusion of a positive main effect of the number of herring-gear RTCs alongside a negative interaction with effort caused a similar saddle shape as that found for haddock catch. Both the susceptibility of small herring to being caught and catch levels were reduced in years with a high number of RTCs, but only under high effort levels (Figure 5e). No effect of the number of RTCs on catch was found for saithe (Figure 5h). Residuals of all ARIMA models presented here were generally normally distributed according to QQ and residual plots (Supplementary Figures S1 and S2 in Supplementary Material Section S1.2). No extreme residuals were found to be outliers using a Grubbs test. Autocorrelations and partial autocorrelations calculated from ARIMA residuals also did not differ from zero at any lag (Supplementary Figures S3 and S4). Spurious correlations in time series Results differed slightly from those presented in Supplementary Material where it was attempted to control for spurious correlations by removing effects of covariates from the number of RTCs a priori (Supplementary Material Section S2). The major differences shown in model results given in Supplementary Material include (1) an interaction between effort and the number of RTCs when predicting haddock selectivity, resulting in similar saddle shape as the haddock catch model presented here, (2) a three-way interaction for herring catch among the number of RTCs, relative total effort, and reference biomass, rather than the saddle-shaped effect presented here, and (3) a reduction in significance of RTCs on cod. The difference in herring results is attributed to the removal of a strong positive effect of recruitment on the number of herring-gear RTCs in Supplementary Material analyses. The difference in cod results appears to be related to the removal of a positive contribution of trawl-specific effort or recruitment to observed cod selectivity and catch (Supplementary Table S2). Qualitatively, model interpretations are similar to those presented here. Discussion Here we analysed whether an effect of the Icelandic RTC system could be detected in either selectivity estimates, reflecting the susceptibility of small fish to being caught, or catch levels, relative to biomass, of young fish over time. An increase in selectivity estimates or entry-aged catch with number of closures would indicate greater fishing mortality, relative to other age classes, in years when RTCs are triggered more often. Such a positive effect therefore would affirm that the RTC system was being triggered as expected: more closures occur in years when more entry-aged fish are being caught in relation to other age classes. However, positive effects do not lend credence to the idea that the fishery closures are ecologically effective; that is, that they reduce susceptibility of small fish to being caught. To have the desired effect on population dynamics, any positive correlation must be overwhelmed by negative effect of closures on susceptibility to being caught; that is, RTCs should cause a reduction in fishing mortality, and possibly catch as well, of small fish. Although the entire RTC system includes far more species than the four we analyse in this study (Schopka, 2007), we restricted the analysis to species with more reliable stock assessments that could yield selectivity estimates, and biomass estimates used to adjust catch data. We found that two of the four species, haddock and herring, showed evidence of ecological effectiveness through a reduction in selectivity estimates and catch of entry-aged fish in years with greater numbers of RTCs. However, this effect was only detectable as an interaction with effort in both herring models and the haddock catch model, indicating that the reduction in small-fish susceptibility and catch is only detectable in high-effort years. Cod, on the other hand, only showed a positive relationship, affirming that the RTCs were being triggered properly. No effect was detected for saithe. We can only speculate why certain species exhibit effects of the RTC system while others do not. There are a variety of potential biological mechanisms that can lead to differences in how effective an RTC system may be in reducing mortality. How aggregated a species age class is, or whether it schools, can cause a species age class to be more or less susceptible to fishing gear, and therefore how sensitive selectivity patterns are to slightly different fishing practices. Similarly, if a species is more likely to mix with other species or age classes, the probability of initiating triggers may differ among species, and their effectiveness may be reduced. As herring are pelagic and form highly aggregated size-specific schools, it is not surprising that they are one of the two species with effective RTCs. However, of the species analysed here saithe is the second-most prone to strong size-specific schooling behaviour and spends more time in the water column than the other gadoids (Scott and Scott, 1988; Neilson et al., 2002, 2003), but did not show an effect of RTCs. Perhaps because saithe are also fast swimmers, patchily distributed, and generally elusive to find (Pol et al., 2016), aggregations may exit a closed area before any protection against mortality can be afforded. Haddock and cod are more similar as they spend more time searching for food close to the seabed rather than in the water column (Anders et al., 2017). However, RTCs were only ecologically effective for haddock, not cod. Perhaps the system is not effective enough to reverse the cod selectivity and catch trends, supporting previous suppositions made based on preliminary analyses that the RTC system likely has no effect on cod dynamics (ICES, 2011; Needle and Catarino, 2011). Besides differences in ecology, differences in the behaviour of fishers targeting each species could likewise affect RTC system effectiveness. For example, fishers may already actively avoid small sizes of some species so that the system becomes redundant. Conversely, if smaller fish fetch a higher price or are otherwise desirable, then fishers have an incentive to target small sizes, thereby counterbalancing any reductions in selectivity yielded by the RTC system. For example, if the closures are not of a sufficient length, movement of fishers back into closed areas can negate any effect of the closures (Needle and Catarino, 2011). Alternatively, fishers could avoid triggering the RTCs while observers are on board by fishing in other locations. For saithe, recent years are marked by a rise in selectivity of small fish (Figure 4b), possibly due to direct targeting (Condie et al., 2014), but this is difficult to distinguish from unrelated changes in gear (i.e. a reduction in gillnet usage, Figure 4d). Furthermore, concurrent regulations may also reduce the incentive to avoid small-sized fish at the same time as RTCs increase it. For example, Condie et al. (2014) point out that despite the presence of a discard ban and RTC system in Iceland, there is very little evidence to suggest that these regulations have changed selectivity patterns of fishers, especially as discards still occur and there are still a large number of small-sized fish landed under a 5% overquota provision. In addition, mandatory sorting grid regulations were implemented in 2000 but reduced in 2006 and 2013, thereby changing selectivity patterns both directly and by making certain fishing locations more accessible. Despite an inability to pinpoint causes for differing results among species, this study at least serves as proof-of-concept: it is the first to show that RTCs aimed at reducing susceptibility of small-sized fish to being caught are likely effective for some species. Haddock and herring, both of which are highly profitable fisheries in Iceland, exhibited reduced selectivity of small fish under high effort conditions, and therefore likely produce economic benefits, given that changes in selectivity on small fish will cause increased yield at older ages. Although Supplementary Material analyses indicate that the effect of RTCs on small herring is confounded with herring recruitment, we are not aware of any other specific mechanism that would reduce small-herring susceptibility in high-recruitment years, so it would be premature to attribute herring results to a spurious correlation. Nonetheless, this study also suggests there is room for improvement. For example, the closures are for the same length of time for all species, even though life histories are quite varied. Although species-specific, it may be possible to refine rules for triggering RTCs using data on movement and site fidelity (Dunn et al., 2014). Similarly, efficiency may be improved through analyses of catch composition data to optimally design trigger thresholds or fine-tune closures according to time, place, or environmental conditions. Finally, if the greatest benefits are reaped when fishing effort is high, then effort or overall fishing mortality could theoretically be incorporated into the trigger. In this case, analysing the relationship between selectivity patterns and effort levels, as well as how they differ by gear or change through fishing seasons, should be given priority for assessing when and how RTCs could be made most effective. This would be a good starting point, for example, to understand why RTCs appear not to be effective for cod, followed by analyses regarding other regulations or fishing behaviour that may counter-balance RTCs. Given that cod dominates the demersal fisheries in Iceland by far, potentially large gains could be made by small modifications to the current system. A cost-benefit analysis would be a logical next step in evaluating by how current implementation of RTCs can be improved (e.g. through move-on rules, Dunn et al., 2016). While examining the benefits of RTCs, two additional non-monetary benefits should likely be considered. First, a greater awareness among both the public and industry participants can develop regarding the importance of avoiding recruitment overfishing, as closures are advertised widely in public media when triggered. In conjunction with a discard ban, size-based closures can incentivize technological innovation and voluntary use of selective fishing gear, as well as promote the idea that discarding is wasteful and morally wrong. This was the experience in Norway when implementing RTCs in the 1980s in the Norwegian and Barents Sea cod and haddock fishery (Graham et al., 2007; Condie et al., 2014; Gullestad et al., 2015). In Iceland, fishing captains can voluntarily trigger a closure by self-reporting (as verified by three independent fishing captains). Sharing responsibility or co-designing and co-managing the system with industry partners can also increase a sense of stewardship and communication among industry participants (O’Keefe and DeCelles, 2013). Often, motivations for co-management, gear innovations, or innovative pilot projects have arisen out of an impending regulations that threaten to reduce fishery profitability, such as proposed permanent closures (e.g. see Needle and Catarino, 2011; Bethoney et al., 2013; Condie et al., 2014; Eliasen, 2014; O’Keefe et al., 2014; Lewison et al., 2015; Little et al., 2015). Real-time spatial management can therefore reduce conflict between management and industry by minimizing regulatory burden and the mismatch between management goals and economic outcomes, thereby potentially increasing trust and compliance (Kraak et al., 2015; Maxwell et al., 2015; Dunn et al., 2016). Second, although reductions in selectivity may be possible through more permanent time-area closures (Dunn et al., 2011), as Eliasen (2014) suggests, in many cases the modelling necessary to justify and define such closures requires a longer time series of data and cannot yet be accomplished, in part because permanent closures are often not supported by the industry and become politically contentious issue (O’Keefe et al., 2014; Lewison et al., 2015; Dunn et al., 2016). In the meantime, RTCs can allow for immediate response to a known, but incompletely defined problem. As is the experience in Iceland, RTCs can also aid the development of more permanent closures by gradually accumulating necessary data (Schopka, 2007; ICES, 2011). Finally, this study points out the weaknesses of analysing fisheries regulations without an experimental design. Therefore, if there is a desire to implement RTCs in other locations, then there is also an opportunity to design its implementation such that causal effects of the regulation may be more rigorously detected. Supplementary data Supplementary material is available at the ICESJMS online version of the manuscript. Acknowledgements We wish to thank two anonymous reviewers who suggested improvements on an earlier version of the manuscript. Funding This material is mostly based upon work supported by the U.S. National Science Foundation under Award No. OCE-1323991, and has received additional funding from the European Commission’s Horizon 2020 Research and Innovation Programme under Grant Agreement No. 634495 for the project Science, Technology, and Society Initiative to minimize Unwanted Catches in European Fisheries (MINOUW). References Anders N., Fernö A., Humborstad O.-B., Løkkeborg S., Utne-Palm A. C. 2017. Species specific behavior and catchability of gadoid fish to floated and bottom set pots. ICES Journal of Marine Science , 74: 769– 779. Anonymous. 2015. Nytjastofnar sjávar 2014/2015. Aflahorfur fiskveiðiárið 2012/2013. State of Marine Stocks in Icelandic Waters 2014/2015. Prospects for the Quota Year 2015/2016. Report 182, Hafrannsóknastofnunin (Marine Research Institute), Reykjavík, Iceland. 217 pp. [In Icelandic with English summary]. Arnason R. 2005. Property rights in fisheries: Iceland's experience with ITQs. Reviews in Fish Biology and Fisheries , 15: 243– 264. Google Scholar CrossRef Search ADS   Bailey N., Campbell N., Holmes S., Needle C., Wright P. 2010. Real time closures of fisheries. European Parliament, Brussels. 50 pp. URL: http://www.europarl.europa.eu/studies. Bethoney N. D., Schondelmeier B. P., Stokesbury K. D. E., Hoffman W. S. 2013. Developing a fine scale system to address river herring (Alosa pseudoharengus, A. aestivalis) and American shad (A. sapidissima) bycatch in the U.S. Northwest Atlantic mid-water trawl fishery. Fisheries Research , 141: 79– 87. Google Scholar CrossRef Search ADS   Bjorkland R., Dunn D. C., McClure M., Jannot J., Bellman M. A., Gleason M., Schiffers K. 2015. Spatiotemporal patterns of rockfish bycatch in U.S. west coast groundfish fisheries: opportunities for reducing incidental catch of depleted species. Canadian Journal of Fisheries And Aquatic Science , 72: 1835– 1846. Google Scholar CrossRef Search ADS   Butterworth D. S., Rademeyer R. A. 2008. Statistical catch-at-age analysis vs. ADAPT-VPA: the case of Gulf of Maine cod. ICES Journal of Marine Sciences , 65: 1717– 1732. Google Scholar CrossRef Search ADS   Condie H. M., Grant A., Catchpole T. L. 2014. Incentivizing selective fishing under a policy to ban discards; lessons from European and global fisheries. Marine Policy , 45: 287– 292. Google Scholar CrossRef Search ADS   De Alessi M., Sullivan J. M., Hilborn R. 2014. The legal, regulatory, and institutional evolution of fishing cooperatives in Alaska and the West coast of the United States. Marine Policy , 43: 217– 225. Google Scholar CrossRef Search ADS   Dunn D. C., Boustany A. M., Halpin P. N. 2011. Spatio-temporal management of fisheries to reduce by-catch and increase fishing selectivity. Fish and Fisheries , 12: 110– 119. Google Scholar CrossRef Search ADS   Dunn D. C., Boustany A. M., Roberts J. J., Brazer E., Sanderson M., Gardner B., Halpin P. N. 2014. Empirical move-on rules to inform fishing strategies: a New England case study. Fish and Fisheries , 15: 359– 375. Google Scholar CrossRef Search ADS   Dunn D. C., Maxwell S. M., Boustany A. M., Halpin P. N. 2016. Dynamic ocean management increases the efficiency and efficacy of fisheries management. Proceedings of the National Academy of Sciences , 113: 668– 673. Google Scholar CrossRef Search ADS   Eliasen S. Q. 2014. Cod avoidance by area regulations in Kattegat—experiences for the implementation of a discard ban in the EU. Marine Policy , 45: 108– 113. Google Scholar CrossRef Search ADS   Graham N., Ferro R. S. T., Karp W. A., MacMullen P. 2007. Fishing practice, gear design, and the ecosystem approach—three case studies demonstrating the effect of management strategy on gear selectivity and discards. ICES Journal of Marine Science , 64: 744– 750. Google Scholar CrossRef Search ADS   Gullestad P., Blom G., Bakke G., Bogstad B. 2015. The “Discard Ban Package”: experiences in efforts to improve the exploitation patterns in Norwegian fisheries. Marine Policy , 54: 1– 9. Google Scholar CrossRef Search ADS   Hobday A. J., Maxwell S., Forgie J., McDonald J., Darby M., Seto K., Bailey H., et al.   2014. Dynamic ocean management: integrating scientific and technological capacity with law, policy and management. Stanford Environmental Law Journal , 33: 125– 165. Holland D. S. 2010. Markets, pooling and insurance for managing bycatch in fisheries. Ecological Economics , 70: 121– 133. Google Scholar CrossRef Search ADS   Holland D. S., Jannot J. E. 2012. Bycatch risk pools for the US West Coast groundfish fishery. Ecolgical Econmics , 78: 132– 147. Google Scholar CrossRef Search ADS   Holmes S. J., Bailey N., Campbell N., Catarino R., Barratt K., Gibb A., Fernandes P. G. 2011. Using fishery dependent data to inform the development and operation of a co-management initiative to reduce cod mortality and cut discards. ICES Journal of Marine Science , 68: 1679– 1688. Google Scholar CrossRef Search ADS   Hyndman R. J., Athanasopoulos G. 2013. Forecasting: Principles and Practice . OTexts, Melbourne. http://otexts.com/fpp (last accessed 13 February 2017). Hyndman R. J., Khandakar Y. 2008. Automatic time series forecasting: the forecast package for R. Journal of Statistical Software , 26: 1– 22. Google Scholar PubMed  ICES. 2011. Report of the ICES Advisory Committee, 2011. ICES Advice, 2011. Book 2, section 2.4.2, pp. 9– 15. ICES. 2015. Report of the North-Western Working Group (NWWG), 28 April–5 May, ICES HQ, Copenhagen, Denmark. ICES CM 2015/ACOM:07. 717 pp. Jakobsson J., Stefánsson G. 1999. Management of summer-spawning herring off Iceland. ICES Journal of Marine Science , 56: 827– 833. Google Scholar CrossRef Search ADS   Johnsen J. P., Eliasen S. 2011. Solving complex fisheries management problems: what the EU can learn from the Nordic experiences of reduction of discards. Marine Policy , 35: 130– 139. Google Scholar CrossRef Search ADS   Komsta L. 2011. outliers: tests for outliers. R package version 0.14. http://CRAN.R-project.org/package=outliers. Kraak S. B. M., Reid D. G., Codling E. A. 2014. Exploring the RTI (real-time incentive) tariff-based approach to single-species fisheries management. Fisheries Research , 155: 90– 102. Google Scholar CrossRef Search ADS   Kraak S. B. M., Reid D. G., Bal G., Barkai A., Codling E. A., Kelly C. J., Rogan E. 2015. RTI (“Real-Time Incentives”) outperforms traditional management in a simulated mixed fishery and cases incorporating protection of vulnerable species and areas. Fisheries Research , 172: 209– 224. Google Scholar CrossRef Search ADS   Labrum K., Oberhoff D. 2013. Fort Bragg—Central Coast risk pool. Annual summary report 2012, 17 pp. Lewison R., Hobday A. J., Maxwell S., Hazen E., Hartog J. R., Dunn D. C., Briscoe D., et al.   2015. Dynamic ocean management: identifying the critical ingredients of dynamic approaches to ocean resource management. BioScience , 65: 486– 498. Google Scholar CrossRef Search ADS   Little A. S., Needle C. L., Hilborn R., Holland D. S., Marshall C. T. 2015. Real-time spatial management approaches to reduce bycatch and discards: experiences from Europe and the United States. Fish and Fisheries , 16: 576– 602. Google Scholar CrossRef Search ADS   Maxwell S. M., Hazen E. L., Lewison R. L., Dunn D. C., Bailey H., Bograd S. J., Briscoe D. K., et al.   2015. Dynamic ocean management: defining and conceptualizing real-time management of the ocean. Marine Policy , 58: 42– 50. Google Scholar CrossRef Search ADS   Needle C. L., Catarino R. 2011. Evaluating the effect of real-time closures on cod targeting. ICES Journal of Marine Science , 68: 1647– 1655. Google Scholar CrossRef Search ADS   Neilson J. D., Annis L., Perley P., Clay A., O’Connor M. 2002. Seasonal aggregations of Canadian east coast Pollock as inferred from the commercial fishery and hydroacoustic observations. Journal of Fish Biology , 61: 1067– 1084. Google Scholar CrossRef Search ADS   Neilson J. D., Clark D., Melvin G. D., Perley P., Stevens S. 2003. The diel vertical distribution and characteristics of pre-spawning aggregations of pollock (Pollachius virens) as inferred from hydroacoustic observations: the implications for survey design. ICES Journal of Marine Science , 60: 860– 871. Google Scholar CrossRef Search ADS   O’Keefe C. E., DeCelles G. R. 2013. Forming a partnership to avoid bycatch. Fisheries , 38: 434– 444. Google Scholar CrossRef Search ADS   O’Keefe C. E., Cadrin S. X., Stokesbury K. D. E. 2014. Evaluating effectiveness of time/area closures, quotas/caps, and fleet communications to reduce fisheries bycatch. ICES Journal of Marine Science , 71: 1286– 1297. Google Scholar CrossRef Search ADS   Pálsson O. K., Björnsson H., Guðmundsson S., Otteson Þ. 2015. Mælingar á brottkasti þorsks og ýsu 2013. Discard measurements of cod and haddock 2013. Report 183, Hafrannsóknastofnunin (Marine Research Institute), Reykjavík, Iceland. 11 pp. [in Icelandic] Pol M. V., Herrmann B., Rillahan C., He P. 2016. Selectivity and retention of Pollock Pollachius virens in a Gulf of Maine trawl fishery. Fisheries Research , 184: 47– 53. Google Scholar CrossRef Search ADS   R Core Team. 2015. R: A Language and Environment for Statistical Computing . R Foundation for Statistical Computing, Vienna. https://www.R-project.org/ (last accessed 13 February 2017). Schopka S. A. 2007. Friðunarsvæði við Ísland og skyndilokanir á Íslandsmiðum. Sögulegt yfirlit. Area closures in Icelandic waters and the real-time closure system. A historical review. Report 133, Hafrannsóknastofnunin (Marine Research Institute), Reykjavík, Iceland. 86 pp. [in Icelandic]. Scott W. B., Scott M. G. 1988. Atlantic fishes of Canada. Canadian Bulletin of Fisheries and Aquatic Sciences , 219: 286– 288. Wallace S., Turris B., Driscoll J., Bodtker K., Mose B., Munro G. 2015. Canada’s Pacific groundfish trawl habitat agreement: a global first in an ecosystem approach to bottom trawl impacts. Marine Policy , 60: 240– 248. Google Scholar CrossRef Search ADS   © International Council for the Exploration of the Sea 2017. All rights reserved. For Permissions, please email: journals.permissions@oup.com http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png ICES Journal of Marine Science Oxford University Press

Evaluating the effectiveness of real-time closures for reducing susceptibility of small fish to capture

Loading next page...
 
/lp/ou_press/evaluating-the-effectiveness-of-real-time-closures-for-reducing-V07t3pKetB
Publisher
Oxford University Press
Copyright
© International Council for the Exploration of the Sea 2017. All rights reserved. For Permissions, please email: journals.permissions@oup.com
ISSN
1054-3139
eISSN
1095-9289
D.O.I.
10.1093/icesjms/fsx152
Publisher site
See Article on Publisher Site

Abstract

Abstract Real-time spatial management in fisheries, a type of dynamic ocean management, uses nearly real-time data collection and dissemination to reduce susceptibility of certain species or age classes to being caught in mixed fisheries. However, as with many fisheries regulations, it is difficult to assess whether such a regulation can produce tangible results on population dynamics. In this study, we take advantage of a rare opportunity in which data regarding real-time closures (RTCs) are available for 1990–2014 alongside annual estimates of fishing mortality for three species (Atlantic cod, haddock, and herring) and catch for four species (all plus saithe) in Icelandic fisheries management. We use time series analyses to assess whether RTCs work as expected and yield a lower susceptibility of small fish to being caught, indicated by lower catch levels and selectivities (as estimated from fishing mortalities) in years with more closures. Results indicate that haddock and herring followed this pattern, but only under conditions of generally high fishing mortality. This study represents the first time evidence has been presented that real-time fishery closures can have a beneficial effect on population dynamics, but also suggests that results differ among species. Introduction Spatial management has long been used in fisheries to reduce discards and bycatch, thereby increasing fishery selectivity or protecting important habitat (Dunn et al., 2011). When designed appropriately, fishery closures that are static in time and space can be effective management tools (O’Keefe et al., 2014). However, they have also been criticized as ineffective, often due to the mismatch between the spatial and temporal scales on which closures are drawn vs. the scales on which resource users and ecological processes operate (Bailey et al., 2010; O’Keefe et al., 2014; Maxwell et al., 2015; Dunn et al., 2016). Dynamic ocean management (DOM) is a promising solution to this mismatch, as it is “management that uses near real-time data to guide the spatial distribution of commercial activities” (Lewison et al., 2015, p. 486). It can theoretically respond faster to changes in the ecosystem than can typical static management schemes, which generally assimilate data on an annual basis (Dunn et al., 2014, 2016; Kraak et al., 2014). As technological innovation speeds and eases the burdens of data collection and sharing, DOM has become an increasingly viable option to facilitate adaptive ecosystem-based management (Lewison et al., 2015; Maxwell et al., 2015; Dunn et al., 2016). DOM includes mechanisms such as spatial real-time fishery closures (RTCs), move-on rules (Dunn et al., 2014, 2016; Bjorkland et al., 2015), risk pools (Holland, 2010; Holland and Jannot, 2012), and real-time incentive-based management (Kraak et al., 2014, 2015). The actual implementation of such mechanisms vary widely, from top-down government-sponsored regulations to industry-sponsored voluntary programs, as well as co-management arrangements that fall in-between (Little et al., 2015). RTCs have gained popularity in the past decade as a fisheries management tool for reducing discards or avoiding hotspots of unwanted species (O’Keefe et al., 2014; Lewison et al., 2015; Little et al., 2015). Real-time oceanographic or catch data may be used to fit spatial models aimed at predicting the occurrence of species to avoid, or in other instances, relatively simple data are needed (Lewison et al., 2015). For example, RTCs are often event-triggered, such that a single reported high catch of a bycatch species or age class will induce a no-fishing zone for several weeks. Similarly, move-on rules require that fishers move a certain distance before continuing operations (Dunn et al., 2014, 2016; Bjorkland et al., 2015). In top-down orchestrated closures, reported catch data are generally gathered from observer coverage or inspections; in voluntary situations, personal catch data are exchanged for access to larger pools of collective catch data and/or quota (Little et al., 2015). Although DOM has the benefit of reacting quicker to stock dynamics and is a clear example of adaptive management, it has not been widely implemented because it poses a variety of challenges for success (Hobday et al., 2014; Lewison et al., 2015). First, data must be collected, potentially processed, and finally disseminated on fast time-scales, which may be relatively labour-intensive and costly for both industry participants and managers (Lewison et al., 2015; Little et al., 2015). Second, it often relies on stakeholders and managers to work together to set up a system in which resource use data are shared, requiring negotiation and agreement on the terms of use (e.g. Holmes et al., 2011; O’Keefe and DeCelles, 2013; Wallace et al., 2015), within a context where laws and policies support such use (Hobday et al., 2014). Third, the correct incentive structure must be in place in order for the benefits to appear great enough for both initiation and compliance (Holmes et al., 2011; Hobday et al., 2014; Lewison et al., 2015). Finally, although it has a greater potential to match spatiotemporal ecosystem dynamics than static annual closures, some knowledge regarding ecosystem dynamics (e.g. site fidelity or movement patterns) is needed to optimally design the closure system for the purpose at hand (O’Keefe et al., 2014; Kraak et al., 2015; Dunn et al., 2016). Therefore, it is useful to know whether DOM is effective, and under what terms of implementation. Overcoming the first three challenges indicates that a program can be initiated and is at least operating as intended. Clearly, voluntary programs yield incentives for negotiation and participation (second and third challenges) or else they would not exist. For example, participants of risk pools use more of the high-value quotas without using more constraining species quota, which would risk an early fishery-wide closure (Labrum and Oberhoff, 2013; De Alessi et al., 2014). However, evaluating the fourth challenge requires linking the programme function with intended ecological effects. For example, some programmes can be deemed ecologically effective through demonstrated avoidance of habitat or bycatch, while simultaneously beneficial to fishers by enabling retention of catch that would have been foregone, avoidance of nuisances in catches, or avoidance of penalties (Dunn et al., 2014, 2016; Bjorkland et al., 2015; Lewison et al., 2015; O’Keefe et al., 2014; Dunn et al., 2016). In the cases of long-standing government-sponsored regulations, there is often less impetus to evaluate whether a regulation is ecologically effective, as long as it is perceived to be beneficial with no negative side effects (i.e. “if it ain’t broke don't fix it”). Evidence for effectiveness in this form may be simply that the system is at least operating as intended, without analysing ecological consequences. For example, pilot projects indicate that fishers do indeed move from areas of higher to lower concentrations of the bycatch species when given catch information (Needle and Catarino, 2011; Bethoney et al., 2013). Evidence for effectiveness in reducing fishing mortality or other environmental impacts, on the other hand, is more difficult to trace due to the lack of a controlled experiment: typically no fishing occurs within closures to test the closure’s effectiveness (Bailey et al., 2010). Closures may be ineffective if fishers move straight back into closures after re-opening or the closure length is too short (Needle and Catarino, 2011). Furthermore, evaluating the effectiveness of any regulation in fisheries is often difficult because of confounding factors, such as simultaneously introduced regulations or economic and environmental changes. In Norway, a long-standing system of RTCs, along with effort reduction and favourable environmental conditions, are thought to have contributed to the recovery of cod in the Norwegian and Barents Seas, but these three factors cannot be disentangled (Johnsen and Eliasen, 2014; Eliasen, 2014). Nonetheless, there exists some limited evidence for reductions in catch of certain species or age classes (Bailey et al., 2010; Holmes et al., 2011; O’Keefe and DeCelles, 2013; O’Keefe et al., 2014), and a theoretical study indicates that real-time management may be even more effective at controlling harvest rates than traditional annual harvest control rules (Kraak et al., 2015). In no cases, however, has an RTC system been linked to population dynamics, thereby yielding direct evidence for fulfilling its ultimate mandate. In this study, we attempt to do just this by testing whether there is a correlation between usage of the Icelandic RTC system and fishing selectivity or catch of young fish. RTCs have occurred in Icelandic fisheries as far back as the late 1960s for the Atlantic cod Gadus morhua and 1970s for the Atlantic herring Clupea harengus, although other species generally have closures beginning in the late 1980s. Today, the Icelandic RTC system applies to nine demersal groundfish species, four pelagic fish species, Norwegian lobster (Nephrops norvegicus), and shrimp (Pandalus borealis) (Schopka, 2007). In most cases, to trigger a closure, catch of a certain size range of the species exceeds a certain percentage of the total catch (e.g. 25% under 55 cm for cod). Closures are generally detected by onboard observers, who subsample and measure the catch in 15–20% of the cod fishery (Little et al., 2015), although other mechanisms are possible. The Icelandic Marine Research Institute (MRI) then has the authority to define a closure area, which must be advertised on radio by the coastguard and the broadcast by the government radio station before the closure takes effect. The closure can take effect within less than an hour from advertisement, and lasts for 14 days, but may be extended for another 7 days if necessary. To properly assess whether the RTC system works (i.e. whether it is having ecological effects), it would be best to test differences in mortality before and after implementation of a regulation, or between stocks or areas that differ in a specific regulation, while controlling for all other covariates. However, the introduction of a regulation as a controlled experiment is exceedingly rare in management systems in fisheries. As a result, the task often relies on correlations, as is the case in this study. Luckily, extensive time series of records regarding Icelandic RTCs allow for enough data to test for correlations. In most other locations, RTCs have only been implemented in the past decade; only RTCs in the Barents and Norwegian Sea cod and haddock fishery have been implemented for a comparable length of time (Little et al., 2015). As a result, the Icelandic system yields an unprecedented opportunity to analyse a unique dataset. Here, we add to the literature surrounding RTCs by providing the first analysis of a direct effect of RTCs on catch and/or selectivity of four species in the Icelandic fishery: Atlantic cod, haddock (Melanogrammus aeglefinus), herring, and saithe (Pollachius virens). These species were chosen because they all have a consistent record of relatively high numbers of closures within the last 30 years, as well as data-rich stock assessments that yield time series of fishing mortality and biomass estimates. We test for a relationship between the numbers of RTCs and catch levels (adjusted by biomass) or selectivity-at-age estimates (calculated from fishing mortalities) for the youngest age class possible of each species. A negative relationship would support the idea that RTCs reduce susceptibility of small fish to being caught. We also include two covariates, the yearly effect of fishing mortality (which we term “relative total effort”) and adult biomass, in an attempt to control for possible confounding factors associated with changes in fishing pressure and the general state of the fishery. Methods Data sources Databases at the MRI were used to obtain data on the number of RTCs (Rt,s,g) by calendar year (t), species (s), and gear (g). Trawls, longlines, and gillnets caught the vast majority of cod, haddock and saithe, and comprise roughly <1% of those data. Bottom trawls, mid-water trawls, and demersal seine were summed to form the “trawl” RTC category; only longline composed the “longline” RTC category. Handline RTCs were excluded. RTC data for herring were not split by gear as they generally included only herring-specific or mid-water gears (i.e. herring trawls, herring nets, herring or non-specific mid-water trawls, and purse seines). Not enough gillnet RTCs occurred for any species for analysis, due to very low selectivity for small fish. Survey abundance indices at age (a) were taken from acoustic surveys for herring and spring groundfish surveys for all other species (It,a,s). Catches from all gears ( Ct,a,s), weights (Wt,a,s), and fishing mortality estimates (Ft,a,s) were obtained from stock assessments completed by the MRI. These data were also available in national annual stock assessment and ICES reports (Anonymous, 2015; ICES, 2015). Survey abundance indices were transformed into reference biomass indices after multiplying by weight-at-age and summing over the appropriate age range of available survey data (Bt,4+,s for cod, saithe, and herring, and Bt,3+,s for haddock). Fishing mortality estimates were decomposed into an age-separable component of fishing mortality (here referred to as selectivity, St,a,s) and a yearly scaling component of fishing mortality (here referred to as relative total effort, Et,s) within each species, by dividing by the selectivity within a year via the equation Et,s = Ft,a,s/St,a,s. The selectivity of the fully selected age(s) within each year then equalled 1, and relative total effort for that year was the maximum fishing mortality value. Relative total effort therefore scales overall fishing mortalities across years, partially indicating differences in fishing effort as estimated in the stock assessment models, although it is not a true measure of fishing effort. Similarly, age-specific selectivity varies across years and reflects the degree to which a certain age class was susceptible to being caught relative to other age classes within that year, due to gear, fish vs. fishing distributions, or other factors. Background on Icelandic fisheries In Iceland, herring was the first species to have ITQs established in 1979 (Jakobsson and Stefánsson, 1999), although most species have been managed under an individual transferable quota (ITQ) system since 1991 (Arnason, 2005). Fishers are required to keep detailed logbooks for each haul/set. There appears to be little unreported catch as landings generally agree with logbook-based catches, although there are still low levels of discards in the form of high-grading for the largest stocks (0–5%, ICES, 2015; Pálsson et al., 2015;). Annual stock assessments are completed in conjunction with the International Council for Exploration of the Seas (ICES), resulting in total allowable catches. There is a full discard ban, and there are technically no minimum landing sizes, but gear restrictions and the RTC system are intended to reduce selectivity on small-sized fish (ICES, 2015). The number of RTCs has greatly increased over the time period analysed (Figures 1–4, panels a). Permanent time/area closures are likewise utilized to protect spawning or juvenile aggregations, and a series of consecutive RTCs may allow the Minister of Fisheries and Agriculture to establish a longer closed area (Schopka, 2007). Further details can be found in Supplementary Material Section S1.1. Figure 1. View largeDownload slide Time series of numbers of RTCs by gear (a), selectivity estimates for ages of recruitment (3) and entry (4) (b), relative total efforts (left axis) and reference biomass indices (right dotted axis) (c), and catch numbers by gear (d) are shown for cod. Figure 1. View largeDownload slide Time series of numbers of RTCs by gear (a), selectivity estimates for ages of recruitment (3) and entry (4) (b), relative total efforts (left axis) and reference biomass indices (right dotted axis) (c), and catch numbers by gear (d) are shown for cod. Figure 2. View largeDownload slide Time series of numbers of RTCs by gear (a), selectivity estimates for ages of recruitment (3) and entry (4) (b), relative total efforts (left axis) and reference biomass indices (right dotted axis) (c), and catch numbers by gear (d) are shown for haddock. Figure 2. View largeDownload slide Time series of numbers of RTCs by gear (a), selectivity estimates for ages of recruitment (3) and entry (4) (b), relative total efforts (left axis) and reference biomass indices (right dotted axis) (c), and catch numbers by gear (d) are shown for haddock. Figure 3. View largeDownload slide Time series of numbers of RTCs by gear (a), selectivity estimates for ages of recruitment (3) and entry (4) (b), relative total efforts (left axis) and reference biomass indices (right dotted axis) (c), and catch numbers by gear (d) are shown for herring. Survey data are missing from 1995. Figure 3. View largeDownload slide Time series of numbers of RTCs by gear (a), selectivity estimates for ages of recruitment (3) and entry (4) (b), relative total efforts (left axis) and reference biomass indices (right dotted axis) (c), and catch numbers by gear (d) are shown for herring. Survey data are missing from 1995. Figure 4. View largeDownload slide Time series of numbers of RTCs by gear (a), selectivity estimates for ages of recruitment (3) and entry (4) (b), relative total efforts (left axis) and reference biomass indices (right dotted axis) (c), and catch numbers by gear (d) are shown for saithe. Figure 4. View largeDownload slide Time series of numbers of RTCs by gear (a), selectivity estimates for ages of recruitment (3) and entry (4) (b), relative total efforts (left axis) and reference biomass indices (right dotted axis) (c), and catch numbers by gear (d) are shown for saithe. Cod Substantial effort reductions began in the 1990s (ICES, 2015) and a harvest control rule was introduced in the 1995–1996 fishing season. Subsequent modifications to the harvest control rule decreased effort slowly over the following decades (Figure 1b and c). Biomass levels have been increasing, especially after 2000 (Figure 1c and d), during which longline catch has increased as trawl and gillnet catches have declined (Figure 1d, ICES, 2015). RTCs have been implemented since the late 1960s, but the longline closures have increased as trawl closures have decreased as a reflection of recent gear shifts (Figure 1a and d). RTCs are currently triggered by catch numbers consisting of 25% or greater cod that are <55 cm. Haddock Landings of Icelandic haddock are currently low (Figure 2d) due to excessive fishing mortality and low survival have caused decreases in the stock until 2011. Since then, reduced fishing mortality has helped to slow decreases (ICES, 2015). Similar to cod, longline catch has increased while trawl and gillnet catches have decreased (Figure 2d, ICES, 2015). As a result, longline RTCs have increased while trawl RTCs have decreased (Figure 2a). Haddock RTCs are triggered by catches composed of more than 30% of haddock <45 cm. Herring Prior to 2008 fishing mortalities were steadily higher than target fishing mortalities, but high stock levels persisted until an Ichthyophonous infection reduced stock levels during the following 2 years (Figure 3, ICES, 2015). Most catch originates from pelagic trawls or purse seines (ICES, 2015). Other regulations are in place to reduce selectivity of herring, including certain gear and bycatch limitations, as well as RTCs that have been implemented since the 1970s (ICES, 2015). RTCs are currently triggered when >25% of a catch is composed of herring <27 cm. Saithe The harvest rate for saithe peaked in the 1990s, but since then has stabilized around a similar rate as intended by a newly introduced harvest control rule (20% of the reference biomass, ICES, 2015). Supporting trends can be seen in effort and catch data (Figure 4b and d). Saithe selectivity estimates are fixed as constant in the stock assessment within three periods: one occurring before 1996, from 1997 to 2003, from 2004 to present, leading to artificially large changes between periods (Figure 4b). Therefore, we did not analyse saithe selectivity estimates, only catch. We also included a dummy variable to reflect changes in catch from the periods 1990–2003 to 2004–2014 as a process unrelated to RTCs, as the stock assessment model indicates a large increase in selectivity around this time (Figure 4b). Small-sized saithe have recently been targeted to a greater extent than in the past, partially as a result of the reduction of gillnets that target larger fish (ICES, 2015). RTCs are triggered when catch of saithe <55 cm compose at least 30% of a catch in numbers. Data analyses Detecting susceptibility of small fish to being caught RTCs are designed in general to reduce susceptibility of small fish to being caught. Therefore, if RTCs are successfully implemented, there should be more RTCs in years when catches with large proportions of small fish are more frequent (positive relationship). However, if RTCs are furthermore effective at reducing the susceptibility of small fish to being caught (i.e. they are working as intended), then this relationship should reverse: there should be lower values of selectivity and catch for small fish in years when there are a high number of closures, relative to other years (negative relationship). We expect model results to be quite similar among selectivity and catch, but we use both because there are pros and cons for using each source as data. Although changes in selectivity are ultimately what we are trying to detect, selectivity estimates come from stock assessment outputs, which means that their accuracies rely on model assumptions and any restrictions put on the model. For example, saithe selectivities were constrained to be constant within three time periods in the saithe stock assessment, so these were excluded from analyses. On the other hand, cod selectivities were estimated as random effects in statistical catch-at-age models, so although they are not true data, they are thought to be reliable as they are not constrained to take a specified shape (e.g. a dome or sinusoidal shape). Herring and haddock selectivities were estimated in a statistical virtual population analysis (VPA). Although some formulations of statistical VPAs can impose an asymptotic selectivity curve, the oldest age class selectivity values are generally the ones that are constrained, while younger age classes, such as those analysed here, should be less affected (Butterworth and Rademeyer, 2008). In contrast, catch-at-age data are not stock assessment output, despite being extrapolated from total catches and sub-sampled age distributions. However, catch is positively related to fishing effort and biomass of that age class, so it may only be used as an indicator of selectivity if changes if effort and biomass are accounted for. Effort was included as a covariate in main data analyses (see following section), whereas the effect of biomass of the same age class was removed a priori by first regressing catch over the same-age biomass index. Catch residuals ( C˙t,a,s) were then used as the dependent variable instead of raw catch to reflect catch level relative to stock status. Statistical framework Four further statistical issues were apparent from the outset. First, although RTCs are meant to reduce susceptibility of recruits and younger fish to being caught, precision of selectivity estimates from recruitment-aged fish, indicated in stock assessments, are quite low due to generally low mortality rates relative to other age classes. Therefore, we focused on predicting selectivity and catch of the age class above recruitment, which we refer to as “entry-aged” and are generally the first year of inclusion in reference biomass (i.e. age 4 for cod, saithe, and herring, or 3 for haddock). In general, entry-aged selectivity estimates had higher precision than recruitment-aged estimates and were correlated with recruitment-aged estimates due to overlapping size ranges between recruit and entry-aged fish (Figures 1b–4b). As a result, there is likely to be a direct effect of RTCs on this age class as well. Second, throughout the history of any fishery, stock biomasses and fishing effort levels will fluctuate in response to environmental or economic changes. Such fluctuations can affect the regulatory environment or stock age structure, as well as the behaviour of fishermen (e.g. through changes in gear or location), and consequently can also affect observed selectivity patterns. It is therefore useful to include variables reflecting the general state of the fishery to avoid confounding these changes with the effect of RTC closures. We included both relative total fishing effort and reference biomass in the analysis as covariates for this purpose. Third, as these are time series, autocorrelation may need to be accounted for. To do this, we fit ARIMA time-series models with a matrix of external regressors (also known as ARIMAX models, Equations 1–3), and searched for the best combination of external regressors and autocorrelation parameters to explain patterns in selectivity or catch. This search included all possible combinations of predictors (i.e. the number of trawl, long-line, or herring-gear RTCs, and the covariates relative total effort, reference biomass, and a dummy variable for saithe reflecting a change in conditions from 2003 to 2004) and their interactions with the constraint that main effects were included when interactions composed of them were included. Both trawl and long-line RTCs were included for cod and haddock, only trawl RTCs were included for saithe, and only herring-gear RTCs were included for herring. All analyses were restricted to 1990–2014, which coincides with the broad establishment of the ITQ system (Arnason, 2005). All variables were log-transformed to increase normality except effort variables. Unity was added to guard against zeros causing failure of definition. ARIMAX model selection For each possible combination of predictors, all possible combinations of 5 or fewer auto-regressive (ARp, max(p) = 5) or moving average (MAq, max(q) = 5) parameters were tested, as well as the order of differencing (d) needed to obtain independent, stationary residuals. For cod, we restricted the search to include only d = 1, as preliminary analyses indicated that single-lag differencing could avoid the inclusion of a dummy variable used to reflect a large change in selectivity due to the introduction of the harvest control rule (Figure 1), which would extend the exhaustive best-model search to an infeasible level. AR and MA parameters quantify autocorrelation within the error of an initial regression (Equations 1 and 2), with random error quantified as normal and independently distributed (Equation 3). In such a regression, the predictions at time t ( yt') are determined by a set of k predictors ( xt,k'), their regression coefficients ( β0… βk), and autocorrelated error ( nt'). The prime notation ( yt', xt', nt') indicates that model fitting may occur on data that were differenced, the order d of which reflects how many times single-lag differencing was needed for residuals to become stationary according to a Kwiatkowski–Phillips–Schmidt–Shin (KPSS) test. For example, the single-order differencing used for cod (d = 1) indicates yt' =  yt-yt-1 (Hyndman and Athanasopoulos, 2013). Model fitting was implemented in R (R Core Team, 2015), mainly using the “auto.arima” function in the “forecast” package (Hyndman and Khandakar, 2008). The best model was chosen as that which minimized Aikaike Information Criteria corrected for small sample sizes (AICc). The maximum and minimum residual from final models were tested as outliers using a Grubbs test in R package “outliers” (Komsta, 2011), which tests the difference between the residual and the mean over all residuals, divided by the standard deviation, against a t-distribution. The statistic U is the ratio of residual variances with and without the tested residual included. We use null and residual deviances, standard outputs of generalized linear models but not ARIMA models, to report a convenient standard measure of model fit assuming normally distributed error: percent variation explained ([null deviance−residual deviance]/null deviance). We also show t-statistics (coefficients divided by their standard errors), as tests indicate a difference from 0, and corresponding P-values with degrees of freedom (n – d – p where n is the number of years of data and p is the total of linear and autocorrelation coefficients).   yt'=β0+ β1x1,t'+…+βkxk,t'+nt' (1)  nt'=c+AR1nt-1'+…+ARpnt-p'+MA1et-1+…+MAqet-q+et (2)  et∼N(0,σ2) (3) Spurious correlations in time series The last statistical issue noticeable from the outset was that, even after including covariates that reflect the general state of the fishery, any relationship we find between the number of RTCs and selectivity or catch could be due to spurious correlations with historical unobserved effects. For example, if observer coverage is correlated with the number of RTCs, perhaps greater observer coverage is the main mechanism reducing small-fish susceptibility to being caught rather than RTCs themselves. To rule out as many spurious correlations as possible, we repeat the presented analysis in Supplementary Material Section S2, but after removing the effects of a number of additional effects from the predictor (number of RTCs) before using it as a predictor in main analyses. In particular, we removed the effects of annual variation in observer coverage, recruitment, and gear-specific effort on number of RTCs, and correlations among number of RTCs from different gear types. However, approaching the problem as we did in Supplementary Material did not produce sufficiently different model results to justify presenting the more complex results here. We instead highlighted the main differences in “Results” section. Results Detecting susceptibility of small fish to being caught For all species, entry-aged catch was regressed over entry-aged biomass with positive effects, so catch was adjusted for biomass levels by taking residuals (Table 1). Table 1. Regression results from best models chosen by AICc to predict entry-aged catch numbers (Ct,a,s) using entry-aged biomass (Bt,a,s). Species (s)  Predictors  Estimate  S.E.  t  P  % Expl.  Null df  Res. df  Residual formed  Cod  Intercept  0.0000  0.1697  0.0000  1.0000  31.00  24  23  C˙t,a=4,cod    Bt,4,cod  0.5568  0.1732  3.2145  0.0038          Haddock  Intercept  0.0000  0.1108  0.0000  1.0000  70.58  24  23  C˙t,a=3,haddock    Bt,3,haddock  0.8401  0.1131  7.4284  <0.0001          Herring  Intercept  0.0000  0.1704  0.0000  1.0000  30.41  24  23  C˙t,a=4,herring    Bt,4,herring  0.5514  0.1739  3.1702  0.0043          Saithe  Intercept  0.0000  0.1674  0.0000  1.0000  32.88  24  23  C˙t,a=4,saithe    Bt,4,saithe  0.5734  0.1708  3.3568  0.0027          Species (s)  Predictors  Estimate  S.E.  t  P  % Expl.  Null df  Res. df  Residual formed  Cod  Intercept  0.0000  0.1697  0.0000  1.0000  31.00  24  23  C˙t,a=4,cod    Bt,4,cod  0.5568  0.1732  3.2145  0.0038          Haddock  Intercept  0.0000  0.1108  0.0000  1.0000  70.58  24  23  C˙t,a=3,haddock    Bt,3,haddock  0.8401  0.1131  7.4284  <0.0001          Herring  Intercept  0.0000  0.1704  0.0000  1.0000  30.41  24  23  C˙t,a=4,herring    Bt,4,herring  0.5514  0.1739  3.1702  0.0043          Saithe  Intercept  0.0000  0.1674  0.0000  1.0000  32.88  24  23  C˙t,a=4,saithe    Bt,4,saithe  0.5734  0.1708  3.3568  0.0027          Model parameter estimates, standard errors (S.E.), t-statistics (t), P-values (p), percent variation explained as calculated from null and residual deviances (% Expl.), and deviance degrees of freedom (df). Residuals formed ( C˙t,a,s) were predicted in ARIMA models. ARIMAX model selection Using our model selection method, we found two practical problems. First, because the level of differencing is chosen after selecting the set of regression variables (using “auto.arima”), a peculiar situation can arise in which the level of differencing varies depending on the regression variables included. Every time a dataset is differenced, the time series is shortened by one data point, so that AICc values are no longer comparable: the shorter dataset automatically decreases the magnitude of the log likelihood. In these situations, there may be incomparable competing models at different levels of differencing. We chose to report the best model as that with more differencing (i.e. fewer data points), as these can be thought of as more conservative: by differencing the data, stationarity is removed a priori rather than as a result of correlations with regression parameters. However, in practice, competing models with different levels of differencing had very similar results. Second, our model selection method also often tended to select overfit models. Therefore, we added selection criteria to exclude overfit models: the best model must include fewer than seven parameters, including autocorrelation parameters. In practice, the best models left after applying this criterion included far fewer parameters (Table 2). Table 2. Results of the best ARIMA models chosen by AICc to predict entry-aged selectivity (St,a,s) or catch residuals ( C˙t,a,s) by each species s using (1) relative effort (Et,s,), (2) number of gear-specific RTCs (Rt,s,g), and (3) reference biomasses summed over appropriate age ranges (Bt,a,s). Species  Dependent variable  d  Predictors  Estimate  S.E.  t  P  % Expl.  Null df  Res. df  Cod  St,a=4,cod  1  AR1  0.4342  0.1877  2.3130  0.0309  40.05  23  21        Rt,cod,trawl  0.1955  0.0876  2.2328  0.0366              Et,cod  −0.5224  0.1770  −2.9509  0.0076          C˙t,a=4,cod  1  AR1  −0.8515  0.1749  −4.8696  <0.0001  64.49  23  20        AR2  −0.4923  0.1743  −2.8240  0.0105              Rt,cod,trawl  0.4404  0.0905  4.8642  <0.0001              Bt,4+,cod  −0.2425  0.1013  −2.3950  0.0265        Haddock  St,a=3,haddock  0  Rt,haddock,longline  −0.4170  0.1489  −2.8003  0.0102  50.00  24  23        Bt,3+,haddock  −0.4551  0.1489  −3.0562  0.0056          C˙t,a=3,haddock  0  Rt,haddock,longline  0.2602  0.1296  2.0086  0.0576  73.79  24  21        Et,haddock  0.4303  0.1048  4.1053  0.0005              Bt,3+,haddock  −0.4288  0.1128  −3.8020  0.0010              Rt,haddock,longline x Et,haddock  −0.5593  0.1348  −4.1502  0.0005        Herring  St,a=4,herring  0  Rt,herring,herring-gear  0.3534  0.1283  2.7548  0.0116  65.84  23  21        Et,herring  −0.5432  0.1180  −4.6029  0.0001              Rt,herring,herring-gear x Et,herring  −0.3784  0.1328  −2.8500  0.0093          C˙t,a=4,herring  0  Rt,herring,herring-gear  0.4769  0.1342  3.5550  0.0019  65.96  24  21        Et,herring  0.3963  0.1307  3.0327  0.0063              Bt,4+,herring  −0.2905  0.1374  −2.1143  0.0019              Rt,herring,herring-gear x Et,herring  −0.4283  0.1394  −3.0722  0.0063        Saithe  C˙t,a=4,saithe  1  AR  −0.4308  0.1851  −2.3273  0.0295  24.80  23  22        Et,saithe  0.4911  0.2107  2.3306  0.0293        Species  Dependent variable  d  Predictors  Estimate  S.E.  t  P  % Expl.  Null df  Res. df  Cod  St,a=4,cod  1  AR1  0.4342  0.1877  2.3130  0.0309  40.05  23  21        Rt,cod,trawl  0.1955  0.0876  2.2328  0.0366              Et,cod  −0.5224  0.1770  −2.9509  0.0076          C˙t,a=4,cod  1  AR1  −0.8515  0.1749  −4.8696  <0.0001  64.49  23  20        AR2  −0.4923  0.1743  −2.8240  0.0105              Rt,cod,trawl  0.4404  0.0905  4.8642  <0.0001              Bt,4+,cod  −0.2425  0.1013  −2.3950  0.0265        Haddock  St,a=3,haddock  0  Rt,haddock,longline  −0.4170  0.1489  −2.8003  0.0102  50.00  24  23        Bt,3+,haddock  −0.4551  0.1489  −3.0562  0.0056          C˙t,a=3,haddock  0  Rt,haddock,longline  0.2602  0.1296  2.0086  0.0576  73.79  24  21        Et,haddock  0.4303  0.1048  4.1053  0.0005              Bt,3+,haddock  −0.4288  0.1128  −3.8020  0.0010              Rt,haddock,longline x Et,haddock  −0.5593  0.1348  −4.1502  0.0005        Herring  St,a=4,herring  0  Rt,herring,herring-gear  0.3534  0.1283  2.7548  0.0116  65.84  23  21        Et,herring  −0.5432  0.1180  −4.6029  0.0001              Rt,herring,herring-gear x Et,herring  −0.3784  0.1328  −2.8500  0.0093          C˙t,a=4,herring  0  Rt,herring,herring-gear  0.4769  0.1342  3.5550  0.0019  65.96  24  21        Et,herring  0.3963  0.1307  3.0327  0.0063              Bt,4+,herring  −0.2905  0.1374  −2.1143  0.0019              Rt,herring,herring-gear x Et,herring  −0.4283  0.1394  −3.0722  0.0063        Saithe  C˙t,a=4,saithe  1  AR  −0.4308  0.1851  −2.3273  0.0295  24.80  23  22        Et,saithe  0.4911  0.2107  2.3306  0.0293        Both trawl and longline RTCs were tested for haddock and cod; only herring-gear RTCs were tested for herring, and only trawl RTCs were tested for saithe. Autoregression parameters (AR), moving average parameters (MA), and the number of single-lag differences (d) are also shown, as well as linear model parameter estimates, standard errors (S.E.), t-statistics (t), P-values (P), percent variation explained as calculated from null and residual deviances (% Expl.), and deviance degrees of freedom (df). For all models, we interpret only statistically significant results as the presence of an effect. For herring, 2013 selectivity was removed due to a positive outlier test showing extremely low residual selectivity that year in a preliminary model (U = 0.6545, P = 0.0251). All species with models predicting entry-aged catch had a positive effect (haddock, saithe, herring, P ≤ 0.05) or no effect (cod, P > 0.05) of relative total effort (Figure 5b, d, f, and h, Table 2). These results are unsurprising because high effort levels lead to greater catch in general. In contrast, models predicting selectivity included a negative effect of relative total effort (cod and herring, P ≤ 0.05, Figure 5a and e, Table 2) or no effect (haddock, P > 0.05, Figure 5c), indicating that years of high relative total effort were generally more likely to produce low susceptibility of small fish to being caught. Surprisingly, although biomass of the same age class was accounted for a priori, there was also a negative effect of reference biomass in models of cod, haddock, and herring catch, as well as haddock selectivity. High biomass years therefore also led to generally low catches of entry-aged fish. Results indicated that using ARIMAX models were helpful to counter some nuisances in fitting time series data, such as dealing with non-stationarity through differencing and accounting for autocorrelation in error by including autoregressive or moving average parameters (Table 2). Figure 5. View large Download slide Contours represent predictions of entry-aged selectivity estimates and catch residuals resulting from the best-estimated ARIMA models presented in Supplementary Table S2. Main effects of predictors (excluding dummy variables and reference biomass) can be seen as linear changes in the prediction surface along the y-axis (the number of RTCs) or x-axis (relative total effort). Biomass was held constant at the mean value. Vertical gradients from dark colors at high values of the y-axis toward light colors at the low values of the y-axis correspond with our expectation, that selectivity is lower in years with more RTCs. The nonlinear contours found in panels d–f reflect effects of interactions, resulting in saddle shapes with areas of low selectivity estimates and catch values in upper right and lower left panel corners. Black points show the data used to estimate models. If the number of RTCs was not included as an effect, data were plotted along a horizontal 0 line, whereas data were plotted along a vertical 0 line if relative total effort was not included as an effect. Figure 5. View large Download slide Contours represent predictions of entry-aged selectivity estimates and catch residuals resulting from the best-estimated ARIMA models presented in Supplementary Table S2. Main effects of predictors (excluding dummy variables and reference biomass) can be seen as linear changes in the prediction surface along the y-axis (the number of RTCs) or x-axis (relative total effort). Biomass was held constant at the mean value. Vertical gradients from dark colors at high values of the y-axis toward light colors at the low values of the y-axis correspond with our expectation, that selectivity is lower in years with more RTCs. The nonlinear contours found in panels d–f reflect effects of interactions, resulting in saddle shapes with areas of low selectivity estimates and catch values in upper right and lower left panel corners. Black points show the data used to estimate models. If the number of RTCs was not included as an effect, data were plotted along a horizontal 0 line, whereas data were plotted along a vertical 0 line if relative total effort was not included as an effect. The number of trawl RTCs (but not longline RTCs) was positively correlated with the susceptibility of small cod to being caught (i.e. selectivity estimates, Figure 5a) as well as cod catch (Figure 5b, Table 2). In contrast, the number of longline RTCs was negatively correlated with susceptibility of small haddock (Figure 5c), and had a negative interaction with relative total effort in the model of haddock catch (Figure 5d). This interaction resulted in a saddle-shaped contour, and indicated a reduction in small-haddock catch with greater numbers of longline RTCs only in years with relatively high effort (Figure 5d). Although the number of trawl RTCs was not selected in the haddock models as a significant effect, analyses in Supplementary Material show it to be correlated with longline RTCs (Supplementary Table S3). In both herring models, the inclusion of a positive main effect of the number of herring-gear RTCs alongside a negative interaction with effort caused a similar saddle shape as that found for haddock catch. Both the susceptibility of small herring to being caught and catch levels were reduced in years with a high number of RTCs, but only under high effort levels (Figure 5e). No effect of the number of RTCs on catch was found for saithe (Figure 5h). Residuals of all ARIMA models presented here were generally normally distributed according to QQ and residual plots (Supplementary Figures S1 and S2 in Supplementary Material Section S1.2). No extreme residuals were found to be outliers using a Grubbs test. Autocorrelations and partial autocorrelations calculated from ARIMA residuals also did not differ from zero at any lag (Supplementary Figures S3 and S4). Spurious correlations in time series Results differed slightly from those presented in Supplementary Material where it was attempted to control for spurious correlations by removing effects of covariates from the number of RTCs a priori (Supplementary Material Section S2). The major differences shown in model results given in Supplementary Material include (1) an interaction between effort and the number of RTCs when predicting haddock selectivity, resulting in similar saddle shape as the haddock catch model presented here, (2) a three-way interaction for herring catch among the number of RTCs, relative total effort, and reference biomass, rather than the saddle-shaped effect presented here, and (3) a reduction in significance of RTCs on cod. The difference in herring results is attributed to the removal of a strong positive effect of recruitment on the number of herring-gear RTCs in Supplementary Material analyses. The difference in cod results appears to be related to the removal of a positive contribution of trawl-specific effort or recruitment to observed cod selectivity and catch (Supplementary Table S2). Qualitatively, model interpretations are similar to those presented here. Discussion Here we analysed whether an effect of the Icelandic RTC system could be detected in either selectivity estimates, reflecting the susceptibility of small fish to being caught, or catch levels, relative to biomass, of young fish over time. An increase in selectivity estimates or entry-aged catch with number of closures would indicate greater fishing mortality, relative to other age classes, in years when RTCs are triggered more often. Such a positive effect therefore would affirm that the RTC system was being triggered as expected: more closures occur in years when more entry-aged fish are being caught in relation to other age classes. However, positive effects do not lend credence to the idea that the fishery closures are ecologically effective; that is, that they reduce susceptibility of small fish to being caught. To have the desired effect on population dynamics, any positive correlation must be overwhelmed by negative effect of closures on susceptibility to being caught; that is, RTCs should cause a reduction in fishing mortality, and possibly catch as well, of small fish. Although the entire RTC system includes far more species than the four we analyse in this study (Schopka, 2007), we restricted the analysis to species with more reliable stock assessments that could yield selectivity estimates, and biomass estimates used to adjust catch data. We found that two of the four species, haddock and herring, showed evidence of ecological effectiveness through a reduction in selectivity estimates and catch of entry-aged fish in years with greater numbers of RTCs. However, this effect was only detectable as an interaction with effort in both herring models and the haddock catch model, indicating that the reduction in small-fish susceptibility and catch is only detectable in high-effort years. Cod, on the other hand, only showed a positive relationship, affirming that the RTCs were being triggered properly. No effect was detected for saithe. We can only speculate why certain species exhibit effects of the RTC system while others do not. There are a variety of potential biological mechanisms that can lead to differences in how effective an RTC system may be in reducing mortality. How aggregated a species age class is, or whether it schools, can cause a species age class to be more or less susceptible to fishing gear, and therefore how sensitive selectivity patterns are to slightly different fishing practices. Similarly, if a species is more likely to mix with other species or age classes, the probability of initiating triggers may differ among species, and their effectiveness may be reduced. As herring are pelagic and form highly aggregated size-specific schools, it is not surprising that they are one of the two species with effective RTCs. However, of the species analysed here saithe is the second-most prone to strong size-specific schooling behaviour and spends more time in the water column than the other gadoids (Scott and Scott, 1988; Neilson et al., 2002, 2003), but did not show an effect of RTCs. Perhaps because saithe are also fast swimmers, patchily distributed, and generally elusive to find (Pol et al., 2016), aggregations may exit a closed area before any protection against mortality can be afforded. Haddock and cod are more similar as they spend more time searching for food close to the seabed rather than in the water column (Anders et al., 2017). However, RTCs were only ecologically effective for haddock, not cod. Perhaps the system is not effective enough to reverse the cod selectivity and catch trends, supporting previous suppositions made based on preliminary analyses that the RTC system likely has no effect on cod dynamics (ICES, 2011; Needle and Catarino, 2011). Besides differences in ecology, differences in the behaviour of fishers targeting each species could likewise affect RTC system effectiveness. For example, fishers may already actively avoid small sizes of some species so that the system becomes redundant. Conversely, if smaller fish fetch a higher price or are otherwise desirable, then fishers have an incentive to target small sizes, thereby counterbalancing any reductions in selectivity yielded by the RTC system. For example, if the closures are not of a sufficient length, movement of fishers back into closed areas can negate any effect of the closures (Needle and Catarino, 2011). Alternatively, fishers could avoid triggering the RTCs while observers are on board by fishing in other locations. For saithe, recent years are marked by a rise in selectivity of small fish (Figure 4b), possibly due to direct targeting (Condie et al., 2014), but this is difficult to distinguish from unrelated changes in gear (i.e. a reduction in gillnet usage, Figure 4d). Furthermore, concurrent regulations may also reduce the incentive to avoid small-sized fish at the same time as RTCs increase it. For example, Condie et al. (2014) point out that despite the presence of a discard ban and RTC system in Iceland, there is very little evidence to suggest that these regulations have changed selectivity patterns of fishers, especially as discards still occur and there are still a large number of small-sized fish landed under a 5% overquota provision. In addition, mandatory sorting grid regulations were implemented in 2000 but reduced in 2006 and 2013, thereby changing selectivity patterns both directly and by making certain fishing locations more accessible. Despite an inability to pinpoint causes for differing results among species, this study at least serves as proof-of-concept: it is the first to show that RTCs aimed at reducing susceptibility of small-sized fish to being caught are likely effective for some species. Haddock and herring, both of which are highly profitable fisheries in Iceland, exhibited reduced selectivity of small fish under high effort conditions, and therefore likely produce economic benefits, given that changes in selectivity on small fish will cause increased yield at older ages. Although Supplementary Material analyses indicate that the effect of RTCs on small herring is confounded with herring recruitment, we are not aware of any other specific mechanism that would reduce small-herring susceptibility in high-recruitment years, so it would be premature to attribute herring results to a spurious correlation. Nonetheless, this study also suggests there is room for improvement. For example, the closures are for the same length of time for all species, even though life histories are quite varied. Although species-specific, it may be possible to refine rules for triggering RTCs using data on movement and site fidelity (Dunn et al., 2014). Similarly, efficiency may be improved through analyses of catch composition data to optimally design trigger thresholds or fine-tune closures according to time, place, or environmental conditions. Finally, if the greatest benefits are reaped when fishing effort is high, then effort or overall fishing mortality could theoretically be incorporated into the trigger. In this case, analysing the relationship between selectivity patterns and effort levels, as well as how they differ by gear or change through fishing seasons, should be given priority for assessing when and how RTCs could be made most effective. This would be a good starting point, for example, to understand why RTCs appear not to be effective for cod, followed by analyses regarding other regulations or fishing behaviour that may counter-balance RTCs. Given that cod dominates the demersal fisheries in Iceland by far, potentially large gains could be made by small modifications to the current system. A cost-benefit analysis would be a logical next step in evaluating by how current implementation of RTCs can be improved (e.g. through move-on rules, Dunn et al., 2016). While examining the benefits of RTCs, two additional non-monetary benefits should likely be considered. First, a greater awareness among both the public and industry participants can develop regarding the importance of avoiding recruitment overfishing, as closures are advertised widely in public media when triggered. In conjunction with a discard ban, size-based closures can incentivize technological innovation and voluntary use of selective fishing gear, as well as promote the idea that discarding is wasteful and morally wrong. This was the experience in Norway when implementing RTCs in the 1980s in the Norwegian and Barents Sea cod and haddock fishery (Graham et al., 2007; Condie et al., 2014; Gullestad et al., 2015). In Iceland, fishing captains can voluntarily trigger a closure by self-reporting (as verified by three independent fishing captains). Sharing responsibility or co-designing and co-managing the system with industry partners can also increase a sense of stewardship and communication among industry participants (O’Keefe and DeCelles, 2013). Often, motivations for co-management, gear innovations, or innovative pilot projects have arisen out of an impending regulations that threaten to reduce fishery profitability, such as proposed permanent closures (e.g. see Needle and Catarino, 2011; Bethoney et al., 2013; Condie et al., 2014; Eliasen, 2014; O’Keefe et al., 2014; Lewison et al., 2015; Little et al., 2015). Real-time spatial management can therefore reduce conflict between management and industry by minimizing regulatory burden and the mismatch between management goals and economic outcomes, thereby potentially increasing trust and compliance (Kraak et al., 2015; Maxwell et al., 2015; Dunn et al., 2016). Second, although reductions in selectivity may be possible through more permanent time-area closures (Dunn et al., 2011), as Eliasen (2014) suggests, in many cases the modelling necessary to justify and define such closures requires a longer time series of data and cannot yet be accomplished, in part because permanent closures are often not supported by the industry and become politically contentious issue (O’Keefe et al., 2014; Lewison et al., 2015; Dunn et al., 2016). In the meantime, RTCs can allow for immediate response to a known, but incompletely defined problem. As is the experience in Iceland, RTCs can also aid the development of more permanent closures by gradually accumulating necessary data (Schopka, 2007; ICES, 2011). Finally, this study points out the weaknesses of analysing fisheries regulations without an experimental design. Therefore, if there is a desire to implement RTCs in other locations, then there is also an opportunity to design its implementation such that causal effects of the regulation may be more rigorously detected. Supplementary data Supplementary material is available at the ICESJMS online version of the manuscript. Acknowledgements We wish to thank two anonymous reviewers who suggested improvements on an earlier version of the manuscript. Funding This material is mostly based upon work supported by the U.S. National Science Foundation under Award No. OCE-1323991, and has received additional funding from the European Commission’s Horizon 2020 Research and Innovation Programme under Grant Agreement No. 634495 for the project Science, Technology, and Society Initiative to minimize Unwanted Catches in European Fisheries (MINOUW). References Anders N., Fernö A., Humborstad O.-B., Løkkeborg S., Utne-Palm A. C. 2017. Species specific behavior and catchability of gadoid fish to floated and bottom set pots. ICES Journal of Marine Science , 74: 769– 779. Anonymous. 2015. Nytjastofnar sjávar 2014/2015. Aflahorfur fiskveiðiárið 2012/2013. State of Marine Stocks in Icelandic Waters 2014/2015. Prospects for the Quota Year 2015/2016. Report 182, Hafrannsóknastofnunin (Marine Research Institute), Reykjavík, Iceland. 217 pp. [In Icelandic with English summary]. Arnason R. 2005. Property rights in fisheries: Iceland's experience with ITQs. Reviews in Fish Biology and Fisheries , 15: 243– 264. Google Scholar CrossRef Search ADS   Bailey N., Campbell N., Holmes S., Needle C., Wright P. 2010. Real time closures of fisheries. European Parliament, Brussels. 50 pp. URL: http://www.europarl.europa.eu/studies. Bethoney N. D., Schondelmeier B. P., Stokesbury K. D. E., Hoffman W. S. 2013. Developing a fine scale system to address river herring (Alosa pseudoharengus, A. aestivalis) and American shad (A. sapidissima) bycatch in the U.S. Northwest Atlantic mid-water trawl fishery. Fisheries Research , 141: 79– 87. Google Scholar CrossRef Search ADS   Bjorkland R., Dunn D. C., McClure M., Jannot J., Bellman M. A., Gleason M., Schiffers K. 2015. Spatiotemporal patterns of rockfish bycatch in U.S. west coast groundfish fisheries: opportunities for reducing incidental catch of depleted species. Canadian Journal of Fisheries And Aquatic Science , 72: 1835– 1846. Google Scholar CrossRef Search ADS   Butterworth D. S., Rademeyer R. A. 2008. Statistical catch-at-age analysis vs. ADAPT-VPA: the case of Gulf of Maine cod. ICES Journal of Marine Sciences , 65: 1717– 1732. Google Scholar CrossRef Search ADS   Condie H. M., Grant A., Catchpole T. L. 2014. Incentivizing selective fishing under a policy to ban discards; lessons from European and global fisheries. Marine Policy , 45: 287– 292. Google Scholar CrossRef Search ADS   De Alessi M., Sullivan J. M., Hilborn R. 2014. The legal, regulatory, and institutional evolution of fishing cooperatives in Alaska and the West coast of the United States. Marine Policy , 43: 217– 225. Google Scholar CrossRef Search ADS   Dunn D. C., Boustany A. M., Halpin P. N. 2011. Spatio-temporal management of fisheries to reduce by-catch and increase fishing selectivity. Fish and Fisheries , 12: 110– 119. Google Scholar CrossRef Search ADS   Dunn D. C., Boustany A. M., Roberts J. J., Brazer E., Sanderson M., Gardner B., Halpin P. N. 2014. Empirical move-on rules to inform fishing strategies: a New England case study. Fish and Fisheries , 15: 359– 375. Google Scholar CrossRef Search ADS   Dunn D. C., Maxwell S. M., Boustany A. M., Halpin P. N. 2016. Dynamic ocean management increases the efficiency and efficacy of fisheries management. Proceedings of the National Academy of Sciences , 113: 668– 673. Google Scholar CrossRef Search ADS   Eliasen S. Q. 2014. Cod avoidance by area regulations in Kattegat—experiences for the implementation of a discard ban in the EU. Marine Policy , 45: 108– 113. Google Scholar CrossRef Search ADS   Graham N., Ferro R. S. T., Karp W. A., MacMullen P. 2007. Fishing practice, gear design, and the ecosystem approach—three case studies demonstrating the effect of management strategy on gear selectivity and discards. ICES Journal of Marine Science , 64: 744– 750. Google Scholar CrossRef Search ADS   Gullestad P., Blom G., Bakke G., Bogstad B. 2015. The “Discard Ban Package”: experiences in efforts to improve the exploitation patterns in Norwegian fisheries. Marine Policy , 54: 1– 9. Google Scholar CrossRef Search ADS   Hobday A. J., Maxwell S., Forgie J., McDonald J., Darby M., Seto K., Bailey H., et al.   2014. Dynamic ocean management: integrating scientific and technological capacity with law, policy and management. Stanford Environmental Law Journal , 33: 125– 165. Holland D. S. 2010. Markets, pooling and insurance for managing bycatch in fisheries. Ecological Economics , 70: 121– 133. Google Scholar CrossRef Search ADS   Holland D. S., Jannot J. E. 2012. Bycatch risk pools for the US West Coast groundfish fishery. Ecolgical Econmics , 78: 132– 147. Google Scholar CrossRef Search ADS   Holmes S. J., Bailey N., Campbell N., Catarino R., Barratt K., Gibb A., Fernandes P. G. 2011. Using fishery dependent data to inform the development and operation of a co-management initiative to reduce cod mortality and cut discards. ICES Journal of Marine Science , 68: 1679– 1688. Google Scholar CrossRef Search ADS   Hyndman R. J., Athanasopoulos G. 2013. Forecasting: Principles and Practice . OTexts, Melbourne. http://otexts.com/fpp (last accessed 13 February 2017). Hyndman R. J., Khandakar Y. 2008. Automatic time series forecasting: the forecast package for R. Journal of Statistical Software , 26: 1– 22. Google Scholar PubMed  ICES. 2011. Report of the ICES Advisory Committee, 2011. ICES Advice, 2011. Book 2, section 2.4.2, pp. 9– 15. ICES. 2015. Report of the North-Western Working Group (NWWG), 28 April–5 May, ICES HQ, Copenhagen, Denmark. ICES CM 2015/ACOM:07. 717 pp. Jakobsson J., Stefánsson G. 1999. Management of summer-spawning herring off Iceland. ICES Journal of Marine Science , 56: 827– 833. Google Scholar CrossRef Search ADS   Johnsen J. P., Eliasen S. 2011. Solving complex fisheries management problems: what the EU can learn from the Nordic experiences of reduction of discards. Marine Policy , 35: 130– 139. Google Scholar CrossRef Search ADS   Komsta L. 2011. outliers: tests for outliers. R package version 0.14. http://CRAN.R-project.org/package=outliers. Kraak S. B. M., Reid D. G., Codling E. A. 2014. Exploring the RTI (real-time incentive) tariff-based approach to single-species fisheries management. Fisheries Research , 155: 90– 102. Google Scholar CrossRef Search ADS   Kraak S. B. M., Reid D. G., Bal G., Barkai A., Codling E. A., Kelly C. J., Rogan E. 2015. RTI (“Real-Time Incentives”) outperforms traditional management in a simulated mixed fishery and cases incorporating protection of vulnerable species and areas. Fisheries Research , 172: 209– 224. Google Scholar CrossRef Search ADS   Labrum K., Oberhoff D. 2013. Fort Bragg—Central Coast risk pool. Annual summary report 2012, 17 pp. Lewison R., Hobday A. J., Maxwell S., Hazen E., Hartog J. R., Dunn D. C., Briscoe D., et al.   2015. Dynamic ocean management: identifying the critical ingredients of dynamic approaches to ocean resource management. BioScience , 65: 486– 498. Google Scholar CrossRef Search ADS   Little A. S., Needle C. L., Hilborn R., Holland D. S., Marshall C. T. 2015. Real-time spatial management approaches to reduce bycatch and discards: experiences from Europe and the United States. Fish and Fisheries , 16: 576– 602. Google Scholar CrossRef Search ADS   Maxwell S. M., Hazen E. L., Lewison R. L., Dunn D. C., Bailey H., Bograd S. J., Briscoe D. K., et al.   2015. Dynamic ocean management: defining and conceptualizing real-time management of the ocean. Marine Policy , 58: 42– 50. Google Scholar CrossRef Search ADS   Needle C. L., Catarino R. 2011. Evaluating the effect of real-time closures on cod targeting. ICES Journal of Marine Science , 68: 1647– 1655. Google Scholar CrossRef Search ADS   Neilson J. D., Annis L., Perley P., Clay A., O’Connor M. 2002. Seasonal aggregations of Canadian east coast Pollock as inferred from the commercial fishery and hydroacoustic observations. Journal of Fish Biology , 61: 1067– 1084. Google Scholar CrossRef Search ADS   Neilson J. D., Clark D., Melvin G. D., Perley P., Stevens S. 2003. The diel vertical distribution and characteristics of pre-spawning aggregations of pollock (Pollachius virens) as inferred from hydroacoustic observations: the implications for survey design. ICES Journal of Marine Science , 60: 860– 871. Google Scholar CrossRef Search ADS   O’Keefe C. E., DeCelles G. R. 2013. Forming a partnership to avoid bycatch. Fisheries , 38: 434– 444. Google Scholar CrossRef Search ADS   O’Keefe C. E., Cadrin S. X., Stokesbury K. D. E. 2014. Evaluating effectiveness of time/area closures, quotas/caps, and fleet communications to reduce fisheries bycatch. ICES Journal of Marine Science , 71: 1286– 1297. Google Scholar CrossRef Search ADS   Pálsson O. K., Björnsson H., Guðmundsson S., Otteson Þ. 2015. Mælingar á brottkasti þorsks og ýsu 2013. Discard measurements of cod and haddock 2013. Report 183, Hafrannsóknastofnunin (Marine Research Institute), Reykjavík, Iceland. 11 pp. [in Icelandic] Pol M. V., Herrmann B., Rillahan C., He P. 2016. Selectivity and retention of Pollock Pollachius virens in a Gulf of Maine trawl fishery. Fisheries Research , 184: 47– 53. Google Scholar CrossRef Search ADS   R Core Team. 2015. R: A Language and Environment for Statistical Computing . R Foundation for Statistical Computing, Vienna. https://www.R-project.org/ (last accessed 13 February 2017). Schopka S. A. 2007. Friðunarsvæði við Ísland og skyndilokanir á Íslandsmiðum. Sögulegt yfirlit. Area closures in Icelandic waters and the real-time closure system. A historical review. Report 133, Hafrannsóknastofnunin (Marine Research Institute), Reykjavík, Iceland. 86 pp. [in Icelandic]. Scott W. B., Scott M. G. 1988. Atlantic fishes of Canada. Canadian Bulletin of Fisheries and Aquatic Sciences , 219: 286– 288. Wallace S., Turris B., Driscoll J., Bodtker K., Mose B., Munro G. 2015. Canada’s Pacific groundfish trawl habitat agreement: a global first in an ecosystem approach to bottom trawl impacts. Marine Policy , 60: 240– 248. Google Scholar CrossRef Search ADS   © International Council for the Exploration of the Sea 2017. All rights reserved. For Permissions, please email: journals.permissions@oup.com

Journal

ICES Journal of Marine ScienceOxford University Press

Published: Jan 1, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off