ICES Journal of Marine Science: Journal du Conseil

Tate,, Alissa;Lo,, Johnny;Mueller,, Ute;Hyndes, Glenn, A;Ryan, Karina, L;Taylor, Stephen, M

doi: 10.1093/icesjms/fsz228pmid: N/A

Abstract Evaluation of fisheries management and sustainability indicators can be supported by a reliable index of harvest rate. However, the most appropriate model that accounts for recreational fisheries is largely unknown. In order to adjust for these factors, generalized linear models were applied to data from shore-based recreational fishing surveys conducted in Western Australia between 2010 and 2016. Five candidate error distributions (lognormal, Gamma, Zero-Altered Gamma, Tweedie, and delta-lognormal) and seven independent variables (year, month, target species, fishing platform, fishers’ avidity, time of day, and day type) were examined for commonly caught nearshore teleost species. Zero-Altered Gamma and Tweedie models performed best overall, although model performance and explanatory variables varied between species. Standardized harvest rates for Australian herring (Arripis georgianus) declined from 1.88 ± 0.17 (mean ± s.e.) fish per fishing party per day) in 2010 to 0.86 ± 0.07 in 2016, while harvest rates for School whiting (Sillago spp.) increased from 0.44 ± 0.21 in 2010 to 0.94 ± 0.34 in 2016. The standardized harvest rates for both species generally showed less fluctuation among years and consistently had smaller errors than the raw harvest rates. Overall, the results suggest that the choice of error distribution, as well as explanatory variables, is species dependent when assessing shore-based fisheries. The approach used could easily be adapted to other recreational fisheries to establish reliable species-specific harvest rates that can detect variability against thresholds set in harvest strategies. Introduction Global population growth has resulted in increased pressure from commercial, artisanal, and recreational fishing on fish stocks (Post et al., 2002; Cooke and Cowx, 2004; Lewin et al., 2006), with recreational harvest exceeding commercial catches in some areas (Cooke and Cowx, 2004; Lewin et al., 2006) and an estimated 47.1 billion fish captured annually by recreational fishers worldwide (Cooke and Cowx, 2004). Understanding and accounting for recreational fishing is a critical component of many fisheries management strategies. In many jurisdictions, the lack of mandatory requirements to report shore-based effort and harvest, and the widespread access aquatic resources, make it challenging to quantify the impacts of shore-based recreational fishing on fish stocks (Pollock et al., 1994; Smallwood et al., 2012). As a result, shore-based activity is often overlooked or disregarded in stock assessments and harvest strategies (Hartill et al., 2012). Harvest rate (number of fish kept per unit of effort) and catch rate (number of fish kept and released per unit of effort) can provide indices that contribute to stock assessments of fisheries (Maunder and Punt, 2004; The National Academy of Sciences, 2017). These indices can be used for assessing the status of individual species or a fishery, and in some cases, as indicators of stock abundance (Maunder and Punt, 2004). However, raw harvest rates may not be comparable over time (Campbell, 2004; Maunder and Punt, 2004), as they do not account for influencing variables such as changes in management, environmental conditions, and biological variation (Reid and Montgomery, 2005; Rangel and Erzini, 2007; Beckley et al., 2008; Strehlow et al., 2012; Pope et al., 2016). Standardization is based on statistical modelling to adjust harvest rates according to explanatory variables (Maunder and Punt, 2004). Generalized linear models (GLMs) have been applied to catch per unit effort data from commercial (Ortiz and Arocha, 2004; Carruthers et al., 2011; Marriott et al., 2014) and boat-based recreational fisheries (Ghosn et al., 2012; Su and He, 2013; Pope et al., 2016); however, these techniques have rarely been applied to shore-based recreational fisheries (O’Neil and Faddy, 2002). Data from shore-based recreational fishing surveys are costly to collect with fishers often difficult to reach, limiting the survey design and resulting in catch and effort estimates that may not be reliable indicators of stock abundance. Catch and catch rate estimates are often zero-inflated (Ye et al., 2001; O’Neill and Faddy, 2003; Shono, 2008; Taylor et al., 2011; Webley et al., 2011; Yu et al., 2011; Heermann et al., 2013) and as such alternative error distribution models may be more appropriate than those based on Gaussian assumptions when establishing an index of shore-based recreational fishing quality and harvest rate strategies (Maunder and Punt, 2004). Catch and release motivations of recreational fishers can vary depending on target species, psychological, social, and environmental factors (Henry and Lyle, 2003). Harvest rates provide information on only the retained catch, whereas catch rates also include the released catch as reported by fishers. Previous studies have applied GLMs to recreational fishing data where harvest (Webley et al., 2011; Hartill et al., 2013) and catch rate (Ortiz and Brown, 2002; Forbes et al., 2015) are the dependent variable when standardizing recreational fishing data. An estimated 612 000 Western Australian residents participate in recreational fishing (Department of Primary Industries and Regional Development, 2018), and approximately 50% of the recreational harvest is taken from the shore (Henry and Lyle, 2003). Shore-based recreational fishing in the Perth metropolitan area of Western Australia is a popular lifestyle activity (Smallwood et al., 2011). This fishery extends along the coastline adjacent to the majority of the state population, and routine assessment of harvest is difficult to obtain due to the lack of a licence frame for shore-based fishing, large geographical scope and often remote fishing locations (Fletcher and Santoro, 2014). However, based on previous research (Henry and Lyle, 2003), the recreational harvest is known to exceed that of the commercial harvest for some species. Estimates of shore-based recreational harvest rates for these species can be compared over time and included in management strategies and monitoring plans, to facilitate management of the stocks. This study evaluates potential indices of harvest rate for nearshore teleost species by applying GLMs to data from four shore-based recreational fishing surveys conducted between 2010 and 2016. The study has focused on key target species, particularly the Australian herring (Arripis georgianus) and School whiting (Sillago spp.). The main objectives were to: (i) identify the variables that contribute to temporal changes in harvest rates; (ii) identify the most appropriate error distribution model(s) for estimating and standardizing harvest rates; and (iii) compare raw and standardized harvest rates. Methods Survey design The survey covered a 60 km stretch of coastline adjacent to the city of Perth, Western Australia, encompassing 56 known shore-fishing locations including groynes, jetties, and beaches. The pilot survey in 2010 occurred over 3 months from April to June. For the following three survey years, the duration of each survey occurred over a 5-month period from February to June (Smallwood et al., 2012). The primary sampling unit was survey day, stratified by month (5 months per year from February to June) and day type (weekdays and weekend days). A total of six randomly selected weekend days (including public holidays) and six randomly selected weekdays were sampled per month with three morning (6 am–1 pm) and three afternoon (1 pm–8 pm) shifts for each day type. On each survey day, a random selection of locations, platforms, starting location, and travel direction (north or south) was made. The survey ran for a total of 60 days each year (12 days per month) with each shift lasting 7 h. Instantaneous counts of people actively engaged in line fishing were taken upon arrival at each location for all beach types. Interview data were collected for each fishing party, whereby a randomly selected fisher was selected for each party, included the number of fishers, gear type and number (i.e. rod or hand line), time spent fishing, and avidity of a randomly selected fisher in each party (recalled number of days fished in past 12 months). The standard fish name and the number of each species kept (observed by the interviewer) and released (reported by fisher) were recorded for each fishing party. Unlike the species kept, the released catch was not sighted by survey officers and often recalled over a number of hours. For this reason, only the kept species (harvest) was used in analysis. The analyses focused on Australian herring and School whiting. Southern school whiting (Sillago bassensis), Western school whiting (Sillago vittata), and Yellowfin whiting (Sillago schomburgkii) were aggregated to a single species group, School whiting (Sillago spp.). These species were aggregated because reliable identification at a species level is difficult to do during field interviews and attempts to separate the data into the different species at a later time would have introduced bias. These species have similar life history characteristics (Hyndes and Potter, 1997), occur in nearshore waters (Hyndes et al. 1996), and are treated as one stock for management purposes (Smith-Vaniz and Jelks, 2006; Brown et al., 2013). Results for other less common nearshore species, Southern garfish (Hyporhamphus melanochir), Western Australian salmon (Arripis truttaceaus) and Silver trevally (Pseudocaranx spp. Complex) are presented in the Supplementary data. Statistical modelling Harvest rates determined from recreational fishing surveys have been considered to be continuous (Su and He, 2013), as well as a discrete variable (Ochwada-Doyle et al., 2014; Ryan and Conron, 2019). In the current study, effort and catch were collected on a party basis to meet survey objectives (Pollock et al., 1994; Jones and Pollock, 2012), with estimation of harvest rate accounting for the number of fishers in each party and averaged across the primary sample unit (survey days). Therefore, harvest rate was considered to be a continuous variable. The harvest rate (⁠ HR̂d ⁠) for day d ⁠, defined as fish per fishing party per day, was calculated as follows: HR̂d= ∑i=1FdcdihdiFd, where cdi is the kept catch (number) of a species for the ith fishing party, hdi the corresponding incomplete trip length (in decimal hours) multiplied by the number of fishers and Fd the total number of fishing parties interviewed on day d ⁠. Therefore, the overall raw harvest rates (⁠ HR̂ ⁠) were calculated as follows: HR̂=∑i=1nsHR̂dns, where ns is the total number of survey days. The mean-of-ratios estimator was used to calculate harvest rates for individual species because incomplete trip data were collected (Hoenig et al., 1997). Fishing trips of 15 min or less in duration were excluded from the analysis similar to previous studies using this survey technique, as fishers often take this amount of time to set up their gear and have not yet put a line in the water (Hoenig et al., 1997; Pollock et al., 1997). In addition, fine scale effort data (e.g. hook size and bait type) were not collected as it requires longer interview time and increases respondent burden. The raw harvest rates were standardized using five different GLMs. A GLM is based on the assumption that the dependent variable Y follows a distribution from the exponential family, which can include the normal, Poisson, and Gamma distributions. The expected value of the dependent variable is given by the following equation: EY=μ=g-1η, where g is the link function which relates the expected value μ=EY of the response Y to the linear predictor η of m explanatory variables. η=β0+∑j=1mβjxj=β0+β1x1+β2x2+⋯+βmxm=Xβ. The vector β'=[β0, β1, …, βm] is a vector of unknown parameter coefficients for the explanatory variables X=x1, x2, …, xm ⁠, typically estimated with maximum likelihood or a Bayesian technique. The variance VarY is a function of the mean and obeys VarY=ϕVμ=Vg-1η, where the dispersion parameter ϕ is a constant and Vμ is the variance function of the underlying distribution. The type of data collected can vary greatly in fisheries science, as does the model selection including choice of link function, transformations, and distribution (McCullagh and Nelder, 1983). The data structure used in this analysis is continuous data collected from incomplete fishing trips, similar to data used by Ye et al. (2001) and Webley et al. (2011). Based on initial assumptions and examination of the data, five distributions were selected: Lognormal, Gamma, Zero-Altered Gamma (ZAG), Tweedie, and delta-lognormal. The lognormal model assumes that the log-transformed harvest rates are normally distributed. As zero-catch data were present, an offset value was added to all harvest rate values prior to applying a lognormal model. The offset value is not determined uniquely, and often the reasons for choosing the offset value are unclear. Following Shono (2008), we chose 0.1 times the maximum value of the response variable as the offset value. The Gamma distribution has been widely used in fisheries applications to describe the right-skewed continuous data that are often present (Ye et al., 2001; Webley et al., 2011). Like the lognormal distribution, the Gamma distribution has a positive probability mass for positive values only. To allow the inclusion of zero-catch data, an offset value was added here also. Unlike the case of the lognormal distribution, there are no recommendations from the literature that could be implemented. We, therefore, tested nine different offset values from 0.02 to 0.09 for their impact on overall estimates and an offset value of 0.02 was found to provide the best fit-statistics to minimize the model residuals, and thus applied to subsequent analyses. The ZAG, Tweedie, and delta-lognormal models were selected for this study, as they explicitly address zero-inflated data. The models differ in the treatment of zeros. For the ZAG model, it is assumed that zero and non-zero data arise from separate processes; the probability of obtaining a zero harvest is modelled using a binomial distribution, and for the non-zero catches, a Gamma distribution model is assumed. Unlike the ZAG model, the Tweedie model does not treat zero and non-zero data separately. Tweedie models are a special case of an exponential dispersion model where the variance is proportional to p (power of the mean). The Tweedie model with power variance value between 1 and 2 represents the distribution of a continuous positive variable with positive mass at 0. Three different power variance values (1.1, 1.5, and 1.9) for the Tweedie model were tested for each species. These three models are, henceforth, referred to as Tweedie #1, Tweedie #2, and Tweedie #3, respectively. The delta-lognormal model estimates the ratio of zero catch with a binomial error distribution and applies a lognormal error for the positive catch rates. Seven variables were selected for analysis: year, avidity, day type, month, platform, target species, and time of day (Table 1). The process for standardizing harvest rates was to first establish the order in which to introduce the seven selected variables into each model, with year as a fixed variable. For the remaining six variables, their importance was determined based on scaled deviance. The magnitude of the scaled deviance was used to rank each variable from 1 (lowest) to 6 (highest), and the average rank for each of the six variables determined the order in which they were added to each candidate model. Table 1. Type of Independent variables used in the analysis along with the harvest rate (dependent variable). Variables Levels Dependent variable  Harvest rate (number of fish kept per day) Independent variables  Year 2010, 2014, 2015, 2016  Avidity High avidity (≥30 days fished per year, by recall), Low avidity (<30 days fished per year)  Time of day AM (06:00–13:00), PM (13:00–20:00)  Month February, March, April, May, June  Day type Weekday, Weekend/Public holiday  Target species Targeted, Not targeted  Platform Beach, Groyne, Jetty Variables Levels Dependent variable  Harvest rate (number of fish kept per day) Independent variables  Year 2010, 2014, 2015, 2016  Avidity High avidity (≥30 days fished per year, by recall), Low avidity (<30 days fished per year)  Time of day AM (06:00–13:00), PM (13:00–20:00)  Month February, March, April, May, June  Day type Weekday, Weekend/Public holiday  Target species Targeted, Not targeted  Platform Beach, Groyne, Jetty Also listed are the levels of each of the independent variables. Open in new tab Table 1. Type of Independent variables used in the analysis along with the harvest rate (dependent variable). Variables Levels Dependent variable  Harvest rate (number of fish kept per day) Independent variables  Year 2010, 2014, 2015, 2016  Avidity High avidity (≥30 days fished per year, by recall), Low avidity (<30 days fished per year)  Time of day AM (06:00–13:00), PM (13:00–20:00)  Month February, March, April, May, June  Day type Weekday, Weekend/Public holiday  Target species Targeted, Not targeted  Platform Beach, Groyne, Jetty Variables Levels Dependent variable  Harvest rate (number of fish kept per day) Independent variables  Year 2010, 2014, 2015, 2016  Avidity High avidity (≥30 days fished per year, by recall), Low avidity (<30 days fished per year)  Time of day AM (06:00–13:00), PM (13:00–20:00)  Month February, March, April, May, June  Day type Weekday, Weekend/Public holiday  Target species Targeted, Not targeted  Platform Beach, Groyne, Jetty Also listed are the levels of each of the independent variables. Open in new tab Variables were introduced in a forward stepwise approach for each candidate model and fivefold cross-validation was then used to identify the most appropriate model for each species. Cross-validation was run seven times for each candidate model, introducing a new variable with each run, as per the pre-determined order for each species. The selection of explanatory variables for each species was made by comparing the AIC values, with lower values indicating a more parsimonious model. Model selection for each species was performed by comparing the predicted harvest rates to observed harvest rates. The criteria for this decision were a high Pearson’s correlation coefficient (r) and low root mean-squared error (RMSE) estimate between the predicted and observed value. The ability to produce non-negative estimates was also considered. Two-way interactions between significant variables for each species were tested; however, results showed they had no significant impact on the accuracy and precision of standardized harvest rates, and therefore, are not presented. Once the best model was selected for each species, further comparisons were made between the standardized yearly (estimated as marginal means) and raw harvest rates. The differences between means are determined according to the lack of overlap of the confidence intervals. All analyses were completed using R version 3.4.4 (R Core Team, 2017). Results A total of 6266 interviews were conducted throughout the 4 years [2010 (n = 1511); 2014 (1620); 2015 (1439); 2016 (1696)]. Australian herring was the most commonly targeted species (28.4% of interviewed fishing parties) of the 103 unique species caught. Fishers caught (kept and released) more Australian herring than any other species (5918) compared with School whiting (1490), Southern garfish (772), Silver trevally (474), and Western Australian salmon (279; Table 2). Table 2. Summary of the number of interviews per year, number of each species caught by fishers (kept and released) per year, and the number of species harvested with proportion of zero catches for each species. Year Number of interviews Species Fish caught (n) Kept Fish caught (n) Released Interviews where fishers harvested fish (% zero catch) 2010 1511 Australian herring 2485 304 371 (75.45) School whiting 183 43 44 (97.09) Southern garfish 661 40 127 (91.59) Western Australian salmon 16 0 7 (99.54) Silver trevally 85 129 39 (97.42) 2014 1620 Australian herring 1364 135 239 (85.25) School whiting 509 86 77 (95.25) Southern garfish 33 0 18 (98.89) Western Australian salmon 4 3 4 (99.75) Silver trevally 13 99 8 (99.51) 2015 1439 Australian herring 937 125 168 (88.33) School whiting 328 39 46 (96.80) Southern garfish 28 0 6 (99.58) Western Australian salmon 33 44 19 (99.75) Silver trevally 20 40 14 (99.03) 2016 1696 Australian herring 519 49 130 (92.33) School whiting 253 49 55 (96.76) Southern garfish 9 1 4 (99.76) Western Australian salmon 138 41 80 (95.28) Silver trevally 11 77 8 (99.53) Year Number of interviews Species Fish caught (n) Kept Fish caught (n) Released Interviews where fishers harvested fish (% zero catch) 2010 1511 Australian herring 2485 304 371 (75.45) School whiting 183 43 44 (97.09) Southern garfish 661 40 127 (91.59) Western Australian salmon 16 0 7 (99.54) Silver trevally 85 129 39 (97.42) 2014 1620 Australian herring 1364 135 239 (85.25) School whiting 509 86 77 (95.25) Southern garfish 33 0 18 (98.89) Western Australian salmon 4 3 4 (99.75) Silver trevally 13 99 8 (99.51) 2015 1439 Australian herring 937 125 168 (88.33) School whiting 328 39 46 (96.80) Southern garfish 28 0 6 (99.58) Western Australian salmon 33 44 19 (99.75) Silver trevally 20 40 14 (99.03) 2016 1696 Australian herring 519 49 130 (92.33) School whiting 253 49 55 (96.76) Southern garfish 9 1 4 (99.76) Western Australian salmon 138 41 80 (95.28) Silver trevally 11 77 8 (99.53) Open in new tab Table 2. Summary of the number of interviews per year, number of each species caught by fishers (kept and released) per year, and the number of species harvested with proportion of zero catches for each species. Year Number of interviews Species Fish caught (n) Kept Fish caught (n) Released Interviews where fishers harvested fish (% zero catch) 2010 1511 Australian herring 2485 304 371 (75.45) School whiting 183 43 44 (97.09) Southern garfish 661 40 127 (91.59) Western Australian salmon 16 0 7 (99.54) Silver trevally 85 129 39 (97.42) 2014 1620 Australian herring 1364 135 239 (85.25) School whiting 509 86 77 (95.25) Southern garfish 33 0 18 (98.89) Western Australian salmon 4 3 4 (99.75) Silver trevally 13 99 8 (99.51) 2015 1439 Australian herring 937 125 168 (88.33) School whiting 328 39 46 (96.80) Southern garfish 28 0 6 (99.58) Western Australian salmon 33 44 19 (99.75) Silver trevally 20 40 14 (99.03) 2016 1696 Australian herring 519 49 130 (92.33) School whiting 253 49 55 (96.76) Southern garfish 9 1 4 (99.76) Western Australian salmon 138 41 80 (95.28) Silver trevally 11 77 8 (99.53) Year Number of interviews Species Fish caught (n) Kept Fish caught (n) Released Interviews where fishers harvested fish (% zero catch) 2010 1511 Australian herring 2485 304 371 (75.45) School whiting 183 43 44 (97.09) Southern garfish 661 40 127 (91.59) Western Australian salmon 16 0 7 (99.54) Silver trevally 85 129 39 (97.42) 2014 1620 Australian herring 1364 135 239 (85.25) School whiting 509 86 77 (95.25) Southern garfish 33 0 18 (98.89) Western Australian salmon 4 3 4 (99.75) Silver trevally 13 99 8 (99.51) 2015 1439 Australian herring 937 125 168 (88.33) School whiting 328 39 46 (96.80) Southern garfish 28 0 6 (99.58) Western Australian salmon 33 44 19 (99.75) Silver trevally 20 40 14 (99.03) 2016 1696 Australian herring 519 49 130 (92.33) School whiting 253 49 55 (96.76) Southern garfish 9 1 4 (99.76) Western Australian salmon 138 41 80 (95.28) Silver trevally 11 77 8 (99.53) Open in new tab Order of variables Target species and time of day were the most significant variables for Australian herring. However, for School whiting, the harvest rates were influenced more by platform than time of day. Avidity and day type had little significance for either species (Table 3). Table 3. Summary of the most parsimonious model and optimal combination of variables (highlighted in bold) for each of the five focus species, as determined by fivefold cross-validation. Species Order of variables Preferred model Australian herring Year + Target species + Time of day + Platform + Month + Day type +Avidity Tweedie School whiting Year + Target species + Platform + Month + Time of Day + Avidity +Day type Tweedie Southern garfish Year + Target species + Platform + Month + Day type + Time of day + Avidity ZAG Western Australian salmon Year + Target species + Platform + Month+ Time of day + Avidity + Day type Lognormal Silver trevally Year + Target species + Time of day + Platform + Month + Day type + Avidity Lognormal Species Order of variables Preferred model Australian herring Year + Target species + Time of day + Platform + Month + Day type +Avidity Tweedie School whiting Year + Target species + Platform + Month + Time of Day + Avidity +Day type Tweedie Southern garfish Year + Target species + Platform + Month + Day type + Time of day + Avidity ZAG Western Australian salmon Year + Target species + Platform + Month+ Time of day + Avidity + Day type Lognormal Silver trevally Year + Target species + Time of day + Platform + Month + Day type + Avidity Lognormal Open in new tab Table 3. Summary of the most parsimonious model and optimal combination of variables (highlighted in bold) for each of the five focus species, as determined by fivefold cross-validation. Species Order of variables Preferred model Australian herring Year + Target species + Time of day + Platform + Month + Day type +Avidity Tweedie School whiting Year + Target species + Platform + Month + Time of Day + Avidity +Day type Tweedie Southern garfish Year + Target species + Platform + Month + Day type + Time of day + Avidity ZAG Western Australian salmon Year + Target species + Platform + Month+ Time of day + Avidity + Day type Lognormal Silver trevally Year + Target species + Time of day + Platform + Month + Day type + Avidity Lognormal Species Order of variables Preferred model Australian herring Year + Target species + Time of day + Platform + Month + Day type +Avidity Tweedie School whiting Year + Target species + Platform + Month + Time of Day + Avidity +Day type Tweedie Southern garfish Year + Target species + Platform + Month + Day type + Time of day + Avidity ZAG Western Australian salmon Year + Target species + Platform + Month+ Time of day + Avidity + Day type Lognormal Silver trevally Year + Target species + Time of day + Platform + Month + Day type + Avidity Lognormal Open in new tab Model performance For Australian herring, minimal model improvement (in accordance to AIC) occurred after the addition of platform as the fourth variable. For School whiting there was no significant improvement in the model after the second variable, target species, was entered into the model (Figure 1). As it is difficult to statistically compare models that use different distributions, cross-validation was used to compare the candidate models. Results for Australian herring showed that the estimated harvest rates for the lognormal and Tweedie #1 models correlated best with the raw harvest rates. The lognormal and Tweedie #1 models (r = 0.59, RMSE = 1.82) appeared to produce the best fit for this species overall; however, Tweedie model #1 was selected as the best model for Australian herring because instances of negative harvest rates (albeit small) arose when using the lognormal model (Table 4). For School whiting, there was very little difference in the correlation r between the estimated and raw harvest rates across all models with values ranging from 0.51 to 0.53. Negative estimates were an issue for the non-zero-inflated models (lognormal and Gamma). By a very small margin across all the performance measures, Tweedie model #3 (r = 0.53, RMSE = 2.81) was determined to be the best model for standardizing harvest rates of School whiting (Table 4). The delta-lognormal model performed worst for both species and has a tendency to underestimate harvest rates. For Australian herring and School whiting, the RMSE was lower for distribution models compared to the raw mean estimates. This suggests that the application of a distribution model typically results in more accurate estimates than using raw mean estimates. The coefficients of the final GLMs with significant variables and error distribution to describe harvest rates are provided in the Supplementary data. Figure 1. Open in new tabDownload slide AIC results from fivefold cross-validation (using selected Tweedie model and variables for each individual species) figure shows results from each of the fivefold after the addition of each variable into the model [1= year (fixed variable), 2 = year + variable 2]. Figure 1. Open in new tabDownload slide AIC results from fivefold cross-validation (using selected Tweedie model and variables for each individual species) figure shows results from each of the fivefold after the addition of each variable into the model [1= year (fixed variable), 2 = year + variable 2]. Table 4. Cross-validation results of each candidate model for all five focus species. Species Candidate model r RMSE Bias Negative estimates (−) Australian herring Year + Target species + Time of day Raw 0.14 2.61 0.00 Lognormal 0.59 1.82 0.08 − Gamma 0.58 1.84 0.00 ZAG 0.55 2.00 −0.14 Tweedie (1.1) 0.59 1.82 0.00 Delta-lognormal 0.49 2.36 −0.23 School whiting Year + Target species Raw −0.03 2.93 0.00 Lognormal 0.52 2.87 0.15 − Gamma 0.51 2.85 0.02 ZAG 0.53 2.82 0.02 Tweedie (1.9) 0.53 2.81 0.02 Delta-lognormal 0.46 2.68 −0.25 Species Candidate model r RMSE Bias Negative estimates (−) Australian herring Year + Target species + Time of day Raw 0.14 2.61 0.00 Lognormal 0.59 1.82 0.08 − Gamma 0.58 1.84 0.00 ZAG 0.55 2.00 −0.14 Tweedie (1.1) 0.59 1.82 0.00 Delta-lognormal 0.49 2.36 −0.23 School whiting Year + Target species Raw −0.03 2.93 0.00 Lognormal 0.52 2.87 0.15 − Gamma 0.51 2.85 0.02 ZAG 0.53 2.82 0.02 Tweedie (1.9) 0.53 2.81 0.02 Delta-lognormal 0.46 2.68 −0.25 The criteria to establish the most parsimonious model for each species (highlighted in bold) were a high correlation coefficient (r) and a low bias estimate (Bias) between the expected value and true value, a low RMSE indicating a better fit in comparison to other models, and non-negative estimates (−ve estimates). Open in new tab Table 4. Cross-validation results of each candidate model for all five focus species. Species Candidate model r RMSE Bias Negative estimates (−) Australian herring Year + Target species + Time of day Raw 0.14 2.61 0.00 Lognormal 0.59 1.82 0.08 − Gamma 0.58 1.84 0.00 ZAG 0.55 2.00 −0.14 Tweedie (1.1) 0.59 1.82 0.00 Delta-lognormal 0.49 2.36 −0.23 School whiting Year + Target species Raw −0.03 2.93 0.00 Lognormal 0.52 2.87 0.15 − Gamma 0.51 2.85 0.02 ZAG 0.53 2.82 0.02 Tweedie (1.9) 0.53 2.81 0.02 Delta-lognormal 0.46 2.68 −0.25 Species Candidate model r RMSE Bias Negative estimates (−) Australian herring Year + Target species + Time of day Raw 0.14 2.61 0.00 Lognormal 0.59 1.82 0.08 − Gamma 0.58 1.84 0.00 ZAG 0.55 2.00 −0.14 Tweedie (1.1) 0.59 1.82 0.00 Delta-lognormal 0.49 2.36 −0.23 School whiting Year + Target species Raw −0.03 2.93 0.00 Lognormal 0.52 2.87 0.15 − Gamma 0.51 2.85 0.02 ZAG 0.53 2.82 0.02 Tweedie (1.9) 0.53 2.81 0.02 Delta-lognormal 0.46 2.68 −0.25 The criteria to establish the most parsimonious model for each species (highlighted in bold) were a high correlation coefficient (r) and a low bias estimate (Bias) between the expected value and true value, a low RMSE indicating a better fit in comparison to other models, and non-negative estimates (−ve estimates). Open in new tab Standardized harvest rates The raw and standardized annual harvest rates (mean ± s.e.) for Australian herring showed a downward trend between 2010 (1.69 ± 0.19 and 1.88 ± 0.17 fish per fishing party per day) and 2016 (0.48 ± 0.08 and 0.86 ± 0.07). Overall, harvest rates for Australian herring from the raw approach appear to have been underestimated those from the standardized approach (Figure 2). Figure 2. Open in new tabDownload slide Raw harvest rates (fishing party per year) compared with standardized harvest rates (using selected Tweedie model and variables) for each species across each of the 4 years (2010, 2014–2016). Figure 2. Open in new tabDownload slide Raw harvest rates (fishing party per year) compared with standardized harvest rates (using selected Tweedie model and variables) for each species across each of the 4 years (2010, 2014–2016). Similarly, the standardized annual harvest rates for School whiting provide a clearer trend than the raw harvest rates. The standardized harvest rate for this species shows an increasing trend between 2010 (0.44 ± 0.21 fish per fishing party per day) and 2016 (0.94 ± 0.34). There was greater annual variability in the raw harvest rates (2010, 0.51 ± 0.15; 2014 0.90 ± 0.26; 2015, 0.54 ± 0.14; 2016, 0.73 ± 0.19; Figure 2). Discussion Harvest rates for shore-based recreational fisheries Based on a roving–roving survey (Pollock et al., 1997) of shore-based recreational fishers in Western Australia, we have shown that it is important to account for the effects of variables that influence harvest rates, such as target species, fishing platform and time of day, in determining harvest rates. By standardizing harvest rates using appropriate GLM models, the errors around the estimates were reduced indicating improvement in the precision of standardized harvest rate estimates compared to raw harvest rates. Raw harvest rates over time are used as indices to assess the status of individual species or a fishery (Gulland, 1983). However, raw harvest rates are problematic as they do not account for environmental conditions, biological variation, or management changes. When using estimates of raw harvest rates as indices to represent a fishery, assumptions are often made regarding the influence of these variables (Maunder and Punt, 2004). Estimates of harvest and catch rates from recreational fishing surveys are a valuable tool in fisheries management as they provide a method of monitoring catches and facilitate sustainable management of valuable resources (Ghosn et al., 2012; Leigh et al., 2017; Ryan and Conron, 2019). Thus, it is important for fisheries management to provide indices that can be used to evaluate changes in recreational fishing quality or success (Ghosn et al., 2012), or alternatively, form the basis of subsequent estimation of total harvest and catch with complementary estimates of total effort. Without appropriate standardization, the use of raw catch rates can lead to errors in stock assessments, and subsequent management decisions (Maunder and Punt, 2004). Standardizing shore-based recreational catches The GLM approaches used in our study are commonly applied to catch and effort data for commercial and boat-based recreational fisheries (Yu et al., 2011; Ghosn et al., 2012; Su and He, 2013). Each model used in our study addressed the presence of the zero-inflated catch rate data differently. For Australian herring and School whiting, the models that actively addressed the presence of zeros performed better than those that did not. The Tweedie models (power variance 1.1 and 1.9) performed better than ZAG, delta-lognormal, lognormal, and Gamma distribution models. This result is consistent with those of Shono (2008), who tested Tweedie and delta-lognormal distributions on by-catch of silky shark (Carcharhimus falciformis). Our study also showed that when the percentages of zeros were high, models that handled zero-inflation provided better estimates than commonly used statistical models. The negative harvest rates that occurred when using the lognormal and Gamma models for these species ruled these models out; however, this result may also be due to the offset value chosen. Other species we tested (Southern garfish, Western Australian salmon, and Silver trevally) were harvested infrequently. For these species with low sample sizes, the performance of Tweedie and delta-lognormal models was not as good as the ZAG and lognormal model, which ultimately provided the best fit for these species (see Supplementary data). There are various factors that influence harvest and catch rates of shore-based recreational fisheries (Rahikainen and Kuikka, 2002; Reid and Montgomery, 2005; Campana et al., 2006; Deroba and Bence, 2009; Carruthers et al., 2011). In our study, target species was the most significant variable for both species (as well as the other three species), but particularly for Australian herring. In other studies, the majority of the recreational harvest was taken by fishers targeting species. Fishers who target a species are more likely to have a broader understanding of the factors that increase their chance of landing their target species (Su and He, 2013; Pope et al., 2016). It is very likely that fishers who target Australian herring have gained knowledge of the preferred environmental conditions, schooling behaviour, time of day and time of the year, increasing the probability of a successful fishing trip for the target species (Su and He, 2013; Pope et al., 2016). Gear type used by fishers also varies depending on the target species. Specific gear type information was not collected in this survey but links to fisher targeting success are known from other sources (Cooke et al., 2005; Cerda et al., 2010). The impact of year, target species and platform may be attributed to the ecology of the species. For example, Australian herring are migratory species that move between nearshore and inshore waters (Fairclough et al., 2000; Ayvazian et al., 2004) and rely on ocean currents for migration, whereas School whiting remain in sandy nearshore embayments or estuarine habitats, increasing the likelihood of their capture by shore-based recreational fishers accessing these nearshore platforms (Hyndes et al., 1996; Fowler et al., 2008). Regardless of the factors that influence harvest and catch rates, the ability to standardize harvest and catch rates by incorporating these factors into appropriate models allows harvest and catch rates to be more accurately assessed for stock assessments of fisheries. Standardized annual harvest rates The validity of raw harvest and catch rates and their ability to account for changes occurring over time is increasingly questioned in fisheries literature (Maunder and Punt, 2004). By implementing an appropriate regression model that accounts for the biological, attitudinal, and environmental factors significant to a fishery, standardized harvest rates reduce the influence of these variables, improving the precision of the estimates (O’Neil and Faddy, 2002; Taylor et al., 2011; Su and He, 2013). For all species of focus in this study, the standardized harvest rates had smaller error margins than the raw harvest rates, indicating that standardizing harvest rates for each individual species can ultimately provide more precise estimates than raw harvest rates. Australian herring, whose harvest rates were more precise once standardized, experienced a decline in standardized harvest rates from 2010–2016. This decline is consistent with recent concerns over stock status and sustainability risk assessment (Smith and Brown, 2014). Declining stock numbers have prompted reviews of assessments, with management changes through a reduced daily bag limit of 30–12 in 2015 (Ryan et al., 2017). The change in bag limit could provide an explanation for the decline in harvest rate between 2015 and 2016. As this was the only such instance and as the change was only relevant for the last survey year, it was not included in this analysis. However, with a longer time series, the influence of the change in bag limit on harvest rates could be examined further. Standardized estimates specific to each species provide reliable estimates of rates that can be applied to current management to detect any variability against thresholds set in harvest strategies. This study compares harvest rate, which is the retained catch observed directly by trained interviewers. Catch rates, on the other hand, that include both the retained and released catch, can introduce bias as released catch are self-reported by fishers and cannot be validated. Many nearshore species, including Australian herring and School whiting, are popular target species for both boat and shore-based fishers in Western Australia. Low release rates in this study were consistent with reported release rates from boat-based recreational fishers in Western Australia where the released catch for these species are minor, and the most common reason for release is “too small” (Ryan et al., 2017). The post-release mortality for these species as a result of handling stress is also minimal for Whiting and similar nearshore species (Butcher et al., 2006). Therefore, the inclusion of the released catch into the analysis would impact very little on the overall trend of harvest rates in this study. Overall, harvest rates proved a more reliable and accurate index to contribute to stock assessments of these popular nearshore species. In 2015/16, Australian herring and School whiting were the most commonly caught finfish by boat-based recreational fishers in Western Australia (Ryan et al., 2017). However, the commercial catch for these species was relatively low in 2016, with 61.7 t of Australian herring (targeted) and 31.8 t of School whiting (targeted and by-catch) captured by commercial fisheries in the West Coast Bioregion (Fletcher et al., 2017). The majority of recreational fishing effort for these species in Western Australia occurs in the West Coast Bioregion (Ryan et al., 2017). However, due to the lack of a sampling frame for large-scale off-site surveys, the shore-based recreational harvest of both species is currently unknown. Harvest rates for these species provide a current index of shore-based recreational fishing quality for these species. This methodology provides an approach that can be applied to similar fisheries where commercial and recreational fisheries do not overlap spatially, or fisheries that only have a recreational component. Conclusions This study demonstrated a method for improving the precision of harvest rates for shore-based recreational fisheries using appropriate models to standardize harvest rates for influential variables. Understanding how these variables influence harvest rates of each species have implications for future shore-based recreational fishing surveys in Western Australia and elsewhere. For example, the standardized harvest rates outlined in this study will assist in monitoring the stock status of numerous finfish species in the region (Smith et al., 2013) as part of an Ecosystem-based approach to fisheries management (Wise and Fletcher, 2013). It is recommended that model selection should be specific to each individual species as influencing factors can vary between species. By applying this approach to future surveys, previously defined threshold levels can be improved and comparisons made with past estimates or pre-set performance indicators and reference points. This will contribute to assessments of stock status and support sustainable management of individual species in multispecies fisheries. Acknowledgements The authors would like to thank the Department of Primary Industries and Regional Development (DPIRD) for funding, planning, and implementation of this survey as well as the field interviewers and data entry staff. We thank Dr Claire Smallwood and Dr Kenneth Pollock for the survey design and DPIRD staff Dr Ainslie Denham, Dr Kim Smith, Tim Nicholas, and Shane Walters for their valuable comments. We are extremely grateful to all the recreational fishers who participated in the surveys and to the ICESJMS reviewers and the editor for their insightful comments. References Ayvazian S. G. , Bastow T. P. , Edmonds J. S. , How J. , Nowara G. B. 2004 . Stock structure of Australian herring (Arripis georgiana) in southwestern Australia . Fisheries Research , 67 : 39 – 53 . Google Scholar Crossref Search ADS WorldCat Beckley L. E. , Fennessy S. T. , Everett B. I. 2008 . Few fish but many fishers: a case study of shore-based recreational angling in a major South African estuarine port . African Journal of Marine Science , 30 : 11 – 24 . Google Scholar Crossref Search ADS WorldCat Brown J. , Dowling C. , Hesp A. , Smith K. , Molony B. 2013 . Status of nearshore finfish stocks in south-western Western Australia Part 3: Whiting (Sillaginidae). Fisheries Research Report 248. Perth, Western Australia. 128 pp. Butcher P. , Broadhurst M. K. , Brand C. B. 2006 . Mortality of sand whiting (Sillago ciliata) released by recreational anglers in an Australian estuary . ICES Journal of Marine Science , 63 : 567 – 571 . Google Scholar Crossref Search ADS WorldCat Campana S. E. , Marks L. , Joyce W. , Kohler N. E. 2006 . Effects of recreational and commercial fishing on blue sharks (Prionace glauca) in Atlantic Canada, with inferences on the North Atlantic population . Canadian Journal of Fisheries and Aquatic Sciences , 63 : 670 – 682 . Google Scholar Crossref Search ADS WorldCat Campbell R. A. 2004 . CPUE standardisation and the construction of indices of stock abundance in a spatially varying fishery using general linear models . Fisheries Research , 70 : 209 – 227 . Google Scholar Crossref Search ADS WorldCat Carruthers T. R. , Ahrens R. N. M. , McAllister M. K. , Walters C. J. 2011 . Integrating imputation and standardization of catch rate data in the calculation of relative abundance indices . Fisheries Research , 109 : 157 – 167 . Google Scholar Crossref Search ADS WorldCat Cerda M. , Alos J. , Palmer M. , Grau A. M. , Riera F. 2010 . Managing recreational fisheries through gear restrictions: the case of limiting hook size in the recreational fishery from the Balearic Islands (NW Mediterranean) . Fisheries Research , 101 : 146 – 155 . Google Scholar Crossref Search ADS WorldCat Cooke S. J. , Barthel B. L. , Suski C. D. , Siepker M. J. , Philipp D. P. 2005 . Influence of circle hook size on hooking efficiency, injury, and size selectivity of bluegill with comments on circle hook conservation benefits in recreational fisheries . North American Journal of Fisheries Management , 25 : 211 – 219 . Google Scholar Crossref Search ADS WorldCat Cooke S. J. , Cowx I. G. 2004 . The role of recreational fishing in global fish crises . BioScience , 54 : 857 – 859 . Google Scholar Crossref Search ADS WorldCat Department of Primary Industries and Regional Development. 2018 . Annual Report 2018. Department of Primary Industries and Regional Development, Western Australia. 248 pp. Deroba J. J. , Bence J. R. 2009 . Developing model-based indices of Lake Whitefish abundance using commercial fishery catch and effort data in Lakes Huron, Michigan, and Superior . North American Journal of Fisheries Management , 29 : 50 – 63 . Google Scholar Crossref Search ADS WorldCat Fairclough D. V. , Dimmlich W. F. , Potter I. C. 2000 . Length and age composition and growth rates of Australian herring Arripis georgiana in different regions . Marine and Freshwater Research , 51 : 631 – 640 . Google Scholar Crossref Search ADS WorldCat Fletcher W. , Mumme M. , Webster F. 2017 . Status Reports of the Fisheries and Aquatic Resources of Western Australia 2015/16: The State of the Fisheries. Perth, Western Australia. 255 pp. Fletcher W. J. , Santoro K. 2014 . Status Report of the Fisheries and Aquatic Resources of Western Australia 2013/14 State of the Fisheries. Perth, Western Australia. 359 pp. Forbes J. P. , Watts R. J. , Robinson W. a. , Baumgartner L. J. , Steffe A. S. , Murphy J. J. 2015 . Recreational fishing effort, catch, and harvest for Murray Cod and Golden Perch in the Murrumbidgee River, Australia . North American Journal of Fisheries Management , 35 : 649 – 658 . Google Scholar Crossref Search ADS WorldCat Fowler A. J. , Steer M. A. , Jackson W. B. , Lloyd M. T. 2008 . Population characteristics of southern sea garfish (Hyporhamphus melanochir, Hemiramphidae) in South Australia . Marine and Freshwater Research , 59 : 429 – 443 . Google Scholar Crossref Search ADS WorldCat Ghosn D. , Collins D. , Baiada C. , Steffe A. 2012 . Catch per unit effort and size composition of striped marlin caught by recreational fisheries in southeast Australian waters. Fisheries Research Report Series, 30. Cronulla. 41. pp. Gulland J. A. 1983 . FAO/Wiley series on food and agriculture, Volume 1. Fish Stock Assesment A Manual of Basic Methods. Food and Agriculture Organisation of the United Nations, Rome, Italy, Italy. 384 pp. Hartill B. , Bian R. , Rush N. , Armiger H. 2013 . Aerial-access recreational harvest estimates for snapper, kahawai, red gurnard, tarakihi and trevally in FMA 1 in 2011–12. 48 pp, doi: 10.13140/RG.2.1.4517.2080. Hartill B. W. , Cryer M. , Lyle J. M. , Rees E. B. , Ryan K. L. , Steffe A. S. , Taylor S. M. , et al. 2012 . Scale- and context-dependent selection of recreational harvest estimation methods: the Australasian experience . North American Journal of Fisheries Management , 32 : 109 – 123 . Google Scholar Crossref Search ADS WorldCat Heermann L. , Emmrich M. , Heynen M. , Dorow M. , König U. , Borcherding J. , Arlinghaus R. 2013 . Explaining recreational angling catch rates of Eurasian perch, Perca fluviatilis: the role of natural and fishing-related environmental factors . Fisheries Management and Ecology , 20 : 187 – 200 . Google Scholar Crossref Search ADS WorldCat Henry G. W. , Lyle J. M. 2003 . The National Recreational and Indigenous Fishing Survey FRDC Project No. 99/158. Canberra, Australia. 161 pp. Hoenig J. M. , Jones C. M. , Pollock K. H. , Robson D. S. , Wade D. L. 1997 . Calculation of catch rate and total catch in roving surveys of anglers . Biometrics , 53 : 306 – 317 . Google Scholar Crossref Search ADS WorldCat Hyndes G. A. , Potter I. C. 1997 . Age, growth and reproduction of Sillago schomburgkii in south-western Australian, nearshore waters and comparisons of life history styles of a suite of Sillago species . Environmental Biology of Fishes , 49 : 435 – 447 . Google Scholar Crossref Search ADS WorldCat Hyndes G. A. , Potter I. C. , Lenanton R. C. J. 1996 . Habitat partitioning by whiting species (Sillaginidae) in coastal waters . Environmental Biology , 45 : 21 – 40 . Google Scholar Crossref Search ADS WorldCat Jones C. M. , Pollock K. H. 2012 . Chapter 19 Recreational angler survey methods: estimation of effort, harvest, and released catch. In Fisheries Techniques. 3rd edn, pp. 883 – 916 . Ed. by A. V. Zale, D. L. Parrish, and T. M. Sutton. American Fisheries Society, Bethesda, Maryland. Leigh G. M. , O’Neill M. F. , Stewart J. 2017 . Stock assessment of the Australian east coast tailor (Pomatomus saltatrix) fishery. Department of Agriculture and Fisheries, Queensland. 167 pp. Lewin W.-C. , Arlinghaus R. , Mehner T. 2006 . Documented and potential biological impacts of recreational fishing: insights for management and conservation . Reviews in Fisheries Science , 14 : 305 – 367 . Google Scholar Crossref Search ADS WorldCat Marriott R. J. , O’Neill M. F. , Newman S. J. , Skepper C. L. , Neill M. F. O. , Newman S. J. , Skepper C. L. 2014 . Abundance indices for long-lived tropical snappers: estimating standardized catch rates from spatially and temporally coarse logbook data . ICES Journal of Marine Science , 71 : 618 – 627 . Google Scholar Crossref Search ADS WorldCat Maunder M. N. , Punt A. E. 2004 . Standardizing catch and effort data: a review of recent approaches . Fisheries Research , 70 : 141 – 159 . Google Scholar Crossref Search ADS WorldCat McCullagh P. , Nelder J. A. 1983 . Generalized Linear Models . Chapman & Hall , London . 506 pp. Google Scholar Crossref Search ADS Google Preview WorldCat COPAC Ochwada-Doyle F. , Mcleod J. , Barrett G. , Clarke G. , Gray C. 2014 . Spatial patterns of recreational exploitation in eastern Australian ROFAs: implications for zonal management . Fisheries Management and Ecology , 21 : 383 – 397 . Google Scholar Crossref Search ADS WorldCat O’Neil M. F. , Faddy M. J. 2002 . Analysis of recreational fish catches—dealing with highly skewed distributions with many zeros, 3rd World Recreational Fishing Conference, 2124 May 2002, Northern Territory, Australia, pp. 67–69. O’Neill M. F. , Faddy M. J. 2003 . Use of binary and truncated negative binomial modelling in the analysis of recreational catch data . Fisheries Research , 60 : 471 – 477 . Google Scholar Crossref Search ADS WorldCat Ortiz M. , Arocha F. 2004 . Alternative error distribution models for standardization of catch rates of non-target species from a pelagic longline fishery: billfish species in the Venezuelan tuna longline fishery . Fisheries Research , 70 : 275 – 297 . Google Scholar Crossref Search ADS WorldCat Ortiz M. , Brown C. 2002 . Standardized catch rates for sailfish (Istiophorus platypterus) from United States recreational fishing surveys in the Northwest Atlantic and Gulf of Mexico . International Commission for the Conservation of Atlantic Tunas , 54 : 772 – 790 . WorldCat Pollock K. H. , Hoenig J. M. , Jones C. M. , Robson D. S. , Greene C. J. 1997 . Catch rate estimates for roving and access point surveys . North American Journal of Fisheries Management , 17 : 11 – 19 . Google Scholar Crossref Search ADS WorldCat Pollock K. H. , Jones C. M. , Brown T. L. 1994 . Angler Survey Methods and Their Applications in Fisheries Management . American Fisheries Society , Bethesda, MD . 363 pp. Google Preview WorldCat COPAC Pope K. L. , Chizinski C. J. , Wiley C. L. , Martin D. R. 2016 . Influence of anglers’ specializations on catch, harvest, and bycatch of targeted taxa . Fisheries Research , 183 : 128 – 137 . Google Scholar Crossref Search ADS WorldCat Post J. R. , Sullivan M. , Cox S. , Lester N. P. , Walters C. J. , Eric A. , Paul A. J. , et al. 2002 . Canada’s recreational fisheries: the invisible collapse? Fisheries , 27 : 6 – 17 . Google Scholar Crossref Search ADS WorldCat R Core Team. 2017 . R: A Language and Environment for Statistical Computing . R Foundation for Statistical Computing , Vienna, Austria . http://www.r-project.org/ (last accessed 1 July 2017). COPAC Rahikainen M. , Kuikka S. 2002 . Fleet dynamics of herring trawlers—change in gear size and implications for interpretation of catch per unit effort . Canadian Journal of Fisheries and Aquatic Sciences , 59 : 531 – 541 . Google Scholar Crossref Search ADS WorldCat Rangel M. O. , Erzini K. 2007 . An assessment of catches and harvest of recreational shore angling in the north of Portugal . Fisheries Management and Ecology , 14 : 343 – 352 . Google Scholar Crossref Search ADS WorldCat Reid D. D. , Montgomery S. S. 2005 . Creel survey based estimation of recreational harvest of penaeid prawns in four southeastern Australian estuaries and comparison with commercial catches . Fisheries Research , 74 : 169 – 185 . Google Scholar Crossref Search ADS WorldCat Ryan K. L. , Conron S. C. 2019 . Comparing indicators of recreational fishing in Port Phillip Bay, Australia, from 2008 to 2011 with variability from a background period (2003–07). Marine and Freshwater Research, 70: 1345–1357. Ryan K. L. , Hall N. G. , Lai E. K. , Smallwood C. B. , Taylor S. M. , Wise B. 2017 . Statewide survey of boat-based recreational fishing in Western Australia 2015/16. Fisheries Research Report 287. 205 pp. Shono H. 2008 . Application of the Tweedie distribution to zero-catch data in CPUE analysis . Fisheries Research , 93 : 154 – 162 . Google Scholar Crossref Search ADS WorldCat Smallwood C. B. , Pollock K. H. , Wise B. S. , Hall N. G. , Gaughan D. J. 2011 . Quantifying recreational fishing catch and effort: a pilot study of shore-based fishers in the Perth Metropolitan area. Fisheries Research Report 216. Perth, Western Australia. 56 pp. Smallwood C. B. , Pollock K. H. , Wise B. S. , Hall N. G. , Gaughan D. J. 2012 . Expanding aerial–roving surveys to include counts of shore-based recreational fishers from remotely operated cameras: benefits, limitations, and cost effectiveness . North American Journal of Fisheries Management , 32 : 1265 – 1276 . Google Scholar Crossref Search ADS WorldCat Smith K. , Brown J. 2014 . Biological synopsis of Australian herring (Arripis georgianus). Fisheries Research Report 251. Perth, Western Australia. 36 pp. Smith K. , Brown J. , Lewis P. , Dowling C. , Howard A. , Lenanton R. , Molony B. , et al. 2013 . Status of nearshore finfish stocks in south-western Western Australia Part 1 : Australian herring. Fisheries Research Report 246. Perth, Western Australia. 194 pp. Smith-Vaniz W. F. , Jelks H. L. 2006 . Australian trevallies of the genus Pseudocaranx (Teleostei: Carangidae), with description of a new species from Western Australia . Memoirs of Museum of Victoria , 63 : 97 – 106 . Google Scholar Crossref Search ADS WorldCat Strehlow H. , Schultz N. , Zimmermann C. , Hammer C. 2012 . Cod catches taken by the German recreational fishery in the western Baltic Sea, 2005-2010: implications for stock assessment and management . ICES Journal of Marine Science , 69 : 1769 – 1780 . Google Scholar Crossref Search ADS WorldCat Su Z. , He J. X. 2013 . Analysis of Lake Huron recreational fisheries data using models dealing with excessive zeros . Fisheries Research , 148 : 81 – 89 . Google Scholar Crossref Search ADS WorldCat Taylor S. M. , Webley J. A. C. , Mayer D. G. 2011 . Improving the precision of recreational fishing harvest estimates using two-part conditional general linear models . Fisheries Research , 110 : 408 – 414 . Google Scholar Crossref Search ADS WorldCat The National Academy of Sciences. 2017 . Review of the Marine Recreational Information Program (MRIP) . National Academy of Sciences . WorldCat COPAC Webley J. A. C. , Mayer D. G. , Taylor S. M. 2011 . Investigating confidence intervals generated by zero-inflated count models: implications for fisheries management . Fisheries Research , 110 : 177 – 182 . Google Scholar Crossref Search ADS WorldCat Wise B. S. , Fletcher W. J. 2013 . Determination and development of cost effective techniques to monitor recreational catch and effort in Western Australian demersal finfish fisheries. Final Report for FRDC Project 2005/034 and WAMSI Subproject 4.4.3. Fisheries Research Report No. 245. Department of Fisheries, Western Australia. 168 pp. Ye Y. , Al-Husaini M. , Al-Baz A. 2001 . Use of generalized linear models to analyze catch rates having zero values: the Kuwait driftnet fishery . Fisheries Research , 53 : 151 – 168 . Google Scholar Crossref Search ADS WorldCat Yu H. , Jiao Y. , Winter A. 2011 . Catch-rate standardization for Yellow Perch in Lake Erie: a comparison of the spatial generalized linear model and the generalized additive model . Transactions of the American Fisheries Society , 140 : 905 – 918 . http://www.tandfonline.com/doi/abs/10.1080/00028487.2011.599258. Google Scholar Crossref Search ADS WorldCat © International Council for the Exploration of the Sea 2019. All rights reserved. For permissions, please email: [email protected] This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)