Bycatch mitigation for commonly caught shark species in the Gulf of Mexico reef bottom longline fishery

Abstract

Objective

The Gulf of Mexico (GOM) reef bottom longline fishery typically not only targets groupers and snappers but also interacts with 27 species of sharks, which are primarily discarded as bycatch. Slow growth, late maturity, and low fecundity in a landscape of increasing fishing pressure make sharks comparatively more susceptible to overfishing and endangered status than other fishes. The purpose of this study was to determine which gear and/or environmental variables best predict the shark catch per set for commonly caught shark species in the GOM reef bottom longline fishery.

Methods

We considered 12 commonly caught shark species that vary from the abundant Atlantic Sharpnose Shark Rhizoprionodon terraenovae to the critically endangered Scalloped Hammerhead Sphyrna lewini. Catch per set, effort, gear, and environmental data were taken from the National Oceanic and Atmospheric Administration National Marine Fisheries Service observer dataset for the GOM reef bottom longline fishery (2009–2017) and were used to fit generalized additive models. The Bayesian information criterion and 10‐fold cross‐validation were used to select the best set of variables that predicted catch per set to determine gear configurations, fisher activities, and environmental conditions contributing to higher shark catch per unit effort. We modeled each species individually, all species combined, and species grouped by similar ecology.

Result

Gear and fishing method variables were consistently included in the best predictive models across species and were the only potential basis for a single strategy that could decrease bycatch across all 12 species. Patterns of environmental variables were only consistent across species with similar ecology and habitat.

Conclusion

Sharks as a group should not be lumped together, as the effects of mitigation measures become confounded and directly managing trade‐offs between species when minimizing bycatch becomes impossible. Focusing on gear rather than environmental variables is the best apparent option to potentially reduce shark catch per set across commonly caught species while minimizing trade‐offs.

INTRODUCTION

Incidental shark catch is an important source of mortality that must be addressed to sustainably manage shark species (Karp et al. 2011; McCully et al. 2013; Clarke et al. 2014). Shark life history is generally described by K-selection, with late maturity, long gestation periods, few energetically demanding offspring, and a long lifespan when compared to bony fishes (Cortés 1998). This results in a low rate of population growth and poor productivity, making sharks particularly vulnerable to overfishing even at low levels of harvest (Cortés 1998, 2002). There is a mismatch between productivity and exploitation, which is further complicated by the fact that mortality occurs even when sharks are not being targeted, as sharks are commonly caught as bycatch in fisheries targeting bony fishes (Wosnick et al. 2023).

The U.S. Gulf of Mexico (GOM) reef fish bottom longline (GOMRBLL) fishery typically targets groupers Epinephelus spp. and snappers Lutjanus spp. (Scott-Denton et al. 2011; Karp et al. 2011). In this fishery, shark species are considered bycatch and the reduction of shark bycatch is necessary to meet legislative mandates under the Magnuson–Stevens Fishery Conservation and Management Act (Karp et al. 2011; National Oceanic and Atmospheric Administration 2016), which state that conservation and management measures must be in place to minimize bycatch. The GOMRBLL fishery discards most of the sharks encountered regardless of the species or the condition of individuals caught (Scott-Denton et al. 2011). The high shark discards and the vulnerability of sharks make shark bycatch reduction desirable regardless of the individual species' current exploitation status.

In this study, we consider the 12 most commonly caught shark species, which differ in terms of their ecology, current exploitation status (Tables 1 and 2), management, and protections. The purpose of this study is to determine which gear and/or environmental variables best predict shark catch per set for commonly caught shark species in the U.S. GOMRBLL fishery. We consider individual species as well as species grouped by size (small vs. large) and habitat (coastal vs. deepwater [DW]), with the aim of exploring the modification of gear and/or fishing practices based on environmental conditions that will reduce encounters for sharks as a group. An optimal strategy to reduce shark bycatch would reduce the catch of all shark species without reducing the catch per unit effort (CPUE) of the target species. However, it is expected that bycatch per set of sharks in different species groups will have different important explanatory variables and trends. Based on the results, we propose mitigation strategies that aim to collectively reduce the interaction of commonly caught shark species with GOMRBLL gear.

TABLE 1 Summary of stock status and ecology group for shark species included in this study (Rosa et al. 2006; Santana et al. 2006; SEDAR 2006, 2007, 2012b, 2012a, 2013, 2015, 2016, 2017, 2018, 2020; Burgess and Branstetter 2009; Conrath 2009; Cortés 2009; Ebert et al. 2009; Morgan et al. 2009; Musick et al. 2009; National Oceanic and Atmospheric Administration 2016; Rigby et al. 2017, 2019; Ferreira and Simpfendorfer 2019; UN Environment Programme World Conservation Monitoring Centre 2020; National Marine Fisheries Service 2022). CITES, Convention on International Trade in Endangered Species of Wild Fauna and Flora; IUCN, International Union for Conservation of Nature; SEDAR, Southeast Data, Assessment, and Review.

Common name	Scientific name	Overfishing	Overfished	Ecology group	Management notes
Atlantic Sixgill Shark/Bigeye Sixgill Shark	Hexanchus vitulus/Hexanchus nakamurai	Unknown	Unknown	Deepwater	Prohibited; IUCN Red List: data deficient
Sharpnose Sevengill Shark	Heptranchias perlo	Unknown	Unknown	Deepwater	Prohibited; IUCN Red List: near threatened
Night Shark	Carcharhinus signatus	Unknown	Unknown	Deepwater	Prohibited; IUCN Red List: vulnerable
Blacktip Shark	Carcharhinus limbatus	No	No	Large coastal	IUCN Red List: near threatened
Scalloped Hammerhead	Sphyrna lewini	Yes	Yes	Large coastal	In year 10 of a 10-year rebuilding plan, under assessment (SEDAR 77); recreationally prohibited; CITES Appendix II (2014); IUCN Red List: critically endangered
Nurse Shark	Ginglymostoma cirratum	Unknown	Unknown	Large coastal	IUCN Red List: data deficient
Sandbar Shark	Carcharhinus plumbeus	No	Yes	Large coastal	In year 18 of a 66-year rebuilding plan; IUCN Red List: vulnerable
Silky Shark	Carcharhinus falciformis	Unknown	Unknown	Large coastal	Listed under CITES Appendix II (2017); IUCN Red List: vulnerable
Tiger Shark	Galeocerdo cuvier	Unknown	Unknown	Large coastal	IUCN Red List: near threatened
Atlantic Sharpnose Shark	Rhizoprionodon terraenovae	No	No	Small coastal	IUCN Red List: least concern
Blacknose Shark	Carcharhinus acronotus	Yes	Yes	Small coastal	In year 10 of a 30-year rebuilding program; IUCN Red List: near threatened
Smooth Dogfish	Mustelus canis	No	No	–	IUCN Red List: near threatened

TABLE 2 Catch per year (2009–2017) for shark species in the Gulf of Mexico reef bottom longline fishery according to the National Oceanic and Atmospheric Administration observer program (National Marine Fisheries Service 2018).

Species	Year	Total
2010	2011	2012	2013	2014	2015	2016	2017
Atlantic Sixgill Shark/Bigeye Sixgill Sharkc,d	15	10	15	94	25	37	46	1	243
Night Sharkc,d	48	40	19	25	14	52	40	4	242
Sharpnose Sevengill Sharkc,d	20	14	11	9	13	3	77	17	164
Blacktip Sharkb	23	29	5	26	272	5	6	1	367
Scalloped Hammerheadb	28	44	4	36	119	28	85	40	384
Nurse Sharkb	31	28	10	62	193	132	95	42	593
Sandbar Sharkb	172	139	21	351	179	210	133	127	1332
Silky Sharkb	117	43	59	73	58	22	105	42	519
Tiger Sharkb	216	271	23	347	154	107	140	115	1373
Atlantic Sharpnose Sharka	2031	2421	339	1866	1522	292	1060	281	9812
Blacknose Sharka	523	442	88	278	243	154	312	33	2073
Smooth Dogfish	482	421	133	998	488	583	1025	1746	5876

METHODS

Catch, effort, gear, and environmental data were taken from the National Oceanic and Atmospheric Administration (NOAA) National Marine Fisheries Service (NMFS) observer data set for the GOMRBLL fishery during the period 2009–2017 (National Marine Fisheries Service 2018). Since 2006, a mandatory observer program, which is jointly implemented by the GOM Fishery Management Council and the NMFS Southeast Fisheries Science Center, has monitored the commercial reef fishery in the GOM. National Marine Fisheries Service observers were allocated to vessel trips through several methods over the years, including stratified random sampling, proportional (to effort) sampling, and voluntary cooperation (Scott-Denton et al. 2011; National Marine Fisheries Service 2018). The observer program collects data at the set level on variables that are shown to be important in predicting shark bycatch. Only species that had at least one positive observation in every year of the time series were analyzed, resulting in a total of 12 shark species (Table 2). Catch was defined as the number of individual sharks caught in a set regardless of whether they were retained. We modeled catch per set rather than catch per hook or catch per hook-hour because we included soak time and the number of hooks as predictor variables to test for the effects of these effort variables on catch. Modeling catch per set is also consistent with other bycatch studies in this fishery (Smith et al. 2018, 2019). Catch by species (or group of species) was modeled as a function of environmental and gear variables using a generalized additive model (GAM; Equation 1; Guisan et al. 2002).

Delta-lognormal, delta-gamma, and negative binomial error distributions were explored, but the negative binomial distribution was used because it is most appropriate for data consisting of small counts that may be overdispersed. We used the R package DHARMa (Hartig 2017), which uses a simulation-based approach to transform residuals into a standardized scale, and we compared the data to the distribution that would be expected under a negative binomial distribution (Hartig 2017). Diagnostic plots of DHARMa residuals revealed that the negative binomial was more appropriate than other distributions and adequate for all shark species (Figure S1 [available in the Supplementary Material separately online]).

Candidate environmental variables included year, season, location (latitude and longitude), and fishing depth (m), while the candidate gear variables included soak time (h), hook shape, hook size, and the number of hooks set (Equation 1; factor levels in Table 3; continuous variable distributions in Figures S2–S5). Smoothers were placed on continuous variables (s is a spline smooth for individual variables; te is a two-dimensional tensor smooth for latitude and longitude), and categorical variables were treated as fixed effects. Variable coefficients for categorical variables were defined as difference relative to the reference level, which is the first category (Table 3):1$\begin{array}{c}\text{Catch}~\text{Year}+\text{Season}+\mathit{te}\left(\text{Latitude},\text{longitude}\right)+s\left(\text{Depth}\right)\hfill \\ +s\left(\text{Soak time}\right)+\text{Hook shape}+\text{Hook size}+s\left(\text{Hooks}\mathrm{set}\right).\hfill \end{array}$

TABLE 3 Factor levels used for season, hook shape, and hook size predictor variables considered in the generalized additive model approaches to predict bycatch per set for commonly caught shark species (N = 10,783 sets).

Variable	Number of levels	Levels (n)
Year	8	2010 (1370)a; 2011 (2332); 2012 (524); 2013 (2134); 2014 (860); 2015 (655); 2016 (1695); 2017 (458)
Season	3	Jan–Apr (4784)a; May–Aug (2798); Sep–Dec (3201)
Hook shape	2	Offset (5004)a; straight (5779)
Hook size	5	≤11 (432)a; 12 (653); 13 (5728); 14 (2216); ≥15 (999)

The dredge function from the R package MuMIn (Barton 2015) was used to test all possible variable combinations within the full model above. The Bayesian information criterion (BIC), maximum likelihood estimation, and 10-fold cross validation were used to identify which combination of variables produced the best model performance, balancing fit and parsimony to optimize predictive ability. The BIC was chosen specifically because it penalizes more complex models, thus preventing overfitting (Schwarz 1978). Models with a BIC weight greater than 0.01 were further considered as candidate models, where model weight $w_i$ for model i was calculated from the difference in BIC (${∆}_i$) between model i and the best model: $w_i={\mathrm{e}}^{-{∆}_i/2}/\sum {\mathrm{e}}^{-{∆}_i/2}$ (Burnham and Anderson 2004). Cross validation of each candidate model was then used to determine the best predictive model among all candidate GAMs. A 10-fold cross validation procedure (Then et al. 2015) randomly allocated each data point to one of the 10 folds. Nine-tenths of the data were used for training, while one-tenth was used as the test data set. The GAM cross validation results were assessed using the root mean square error (RMSE) and mean absolute error (MAE), which were calculated by comparing the catch predicted from the model fitted to the training data set for the sets in the test data set to the true values (Stow et al. 2009; Grüss et al. 2019). The best of several candidate models would have the smallest RMSE and MAE. These metrics can also be compared across models for different species to determine which species' catch can be predicted most accurately and precisely from the models. The RMSE and MAE were calculated across all 10 folds, and the mean was used to select the best model.

The fitting and cross validation procedures were performed for the catch of each species individually; the catch of all small coastal sharks (SCS), all large coastal sharks (LCS; as defined by NMFS [SEDAR 2006]; Table 1), and all DW sharks; and the catch of all species combined. Final models were selected based on the results of the BIC ranking and cross validation. Retained variables and their fitted coefficients were compared across species and species groups to examine for broad patterns that can be used to design mitigation strategies for influencing as many species as possible.

RESULTS

Shark bycatch per set in the GOMRBLL fishery was primarily characterized by the following species, in order of decreasing total numbers encountered (Table 2): Atlantic Sharpnose Shark, Smooth Dogfish, Blacknose Shark, Tiger Shark, Sandbar Shark, Nurse Shark, Silky Shark, Scalloped Hammerhead, Blacktip Shark, Night Shark, Atlantic Sixgill/Bigeye Sixgill Shark, and Sharpnose Sevengill Shark. Over two-thirds (~70%) of shark encounters were made up of Atlantic Sharpnose Sharks and Smooth Dogfish.

Model selection

Forty-one models (Table S1 [available in the Supplementary Material separately online]) had a BIC weight greater than 0.01 and underwent a 10-fold cross validation procedure. For all cases in which more than one candidate model was being considered, the MAE and RMSE values were similar across candidate models within a species or species group; therefore, the models had similar predictive ability (Figure S6), so the best models selected based on the BIC were used for all following analyses (Table 4). In the BIC best models for each species or species group, hook size was excluded the most while depth was only excluded from one model. The group of all species combined was the only case in which the full model was selected.

TABLE 4 Retained variables in best models selected based on the Bayesian information criterion (BIC). “NA” indicates that the variable was not selected, while the plus symbol (+) indicates that the variable was selected. LogL, log likelihood.

Species	Hook shape	Hook size	Depth (m)	Number of hooks set	Soak time (h)	Season	Latitude, longitude	Year	BIC	R 2	logL
All species	+	+	+	+	+	+	+	+	42,882	0.382	−21,170
Atlantic Sixgill Shark/Bigeye Sixgill Sharkc,d	NA	NA	+	NA	NA	NA	NA	+	1408	0.082	−652
Night Sharkc,d	NA	NA	+	NA	NA	+	NA	NA	1521	0.064	−711
Sharpnose Sevengill Sharkc,d	NA	NA	+	NA	NA	NA	+	NA	846	0.08	−380
Deepwater sharks	NA	NA	+	+	NA	+	NA	+	2908	0.17	−1367
Blacktip Sharkb	+	NA	+	+	NA	NA	+	NA	1354	0.099	−610
Scalloped Hammerheadb	NA	NA	+	NA	NA	+	NA	+	2545	0.073	−1191
Nurse Sharkb	NA	+	+	NA	NA	+	+	+	3286	0.125	−1509
Sandbar Sharkb	+	NA	+	NA	+	+	NA	+	6806	0.066	−3295
Silky Sharkb	NA	NA	NA	+	NA	+	+	+	3698	0.054	−1706
Tiger Sharkb	NA	NA	+	+	+	NA	NA	+	8000	0.042	−3919
Large coastal sharks	+	+	+	+	+	NA	+	+	16,144	0.175	−7834
Atlantic Sharpnose Sharka	+	+	+	NA	+	+	+	+	20,784	0.26	−10,174
Blacknose Sharka	NA	+	+	NA	+	NA	+	+	9052	0.136	−4392
Small coastal sharks	+	+	+	NA	+	+	+	+	23,374	0.296	−11,471
Smooth Dogfish	+	NA	+	+	NA	+	+	NA	9609	0.459	−4649

Predicted catch patterns were variable over time, with few long-term patterns (Figure 1). When all species were combined, the year effect was significantly dampened (Figure 1). Season 2 (May–August) was associated with higher catches of Sandbar Sharks, Smooth Dogfish, and Silky Sharks and lower catches of all other groups and other individual species (Figure 1). When all species were considered together, the three seasons had similar predicted catch.

[IMAGE OMITTED. SEE PDF]

Models for the Atlantic Sharpnose Shark, Sandbar Shark, Smooth Dogfish, Blacktip Shark, the LCS group, and all species combined retained hook shape as an explanatory variable in the BIC best model (Table 4; Figure 1). All models indicated that offset hooks had a higher shark catch than straight hooks (Figure 1). The catch on small hooks was higher for the SCS group than for the LCS group, as expected (Figure 1). However, none of the individual species or species groups showed a discernible pattern. Across all species and species groups, there was no single hook size that would minimize shark bycatch.

Depth was retained as a predictor of catch for all species and species groups except the Silky Shark, with the expected higher catch for coastal species in shallower water and DW species in deeper water (Table 4; Figure 2).

[IMAGE OMITTED. SEE PDF]

Models for the Blacknose Shark, Atlantic Sharpnose Shark, Sandbar Shark, Tiger Shark, SCS group, LCS group, and all species combined retained soak time as an explanatory variable (Table 4; Figure 2). All species or species groups except the Blacknose Shark had an overall increasing trend in catch with increasing soak time (Figure 2).

The Smooth Dogfish, Blacktip Shark, Silky Shark, Tiger Shark, LCS group, DW group, and all species combined retained the number of hooks set as an explanatory variable (Table 4; Figure 2). For every species or species group that retained the variable, there was a clear increase in catch with an increase in the number of hooks.

The Blacknose and Nurse sharks had the lowest predicted catch in the western portion of the study area. This contrasted with the Smooth Dogfish, Blacktip Shark, and Sharpnose Sevengill Shark, which had their highest predicted catch in that same location (Figure 3). The Smooth Dogfish and Silky Shark had their lowest predicted catch at and around the intersection of 28° N, −82.5° W, which corresponds with Tampa Bay. When the species groups were compared, they were remarkably similar, with an almost uniform prediction across the study latitudes and longitudes (Figure 3).

[IMAGE OMITTED. SEE PDF]

Model predictive ability

Combined species groups had higher median MAE and RMSE values than the individual species that comprised each corresponding species group (Figure S6). The DW species group had relatively low median MAE and RMSE values of 0.1 and 4.0, respectively, when compared to the other combined species groups. However, the DW group contained the highest single MAE and RMSE, with values over 20 and over 600, respectively. The LCS group had a lower median MAE than the SCS group, which was reflected by the corresponding individual species' MAE medians.

These MAE and RMSE values indicated that our ability to predict bycatch per set varied by species. Generally, LCS were better predicted than SCS, DW sharks were better predicted than coastal species, and models had the worst predictive ability when all species were combined.

DISCUSSION

Habitat is a key source of differentiation between species and species groups, while the use of straight hooks and lower soak times appear to uniformly reduce shark bycatch per set. Interestingly, the magnitude of each variable's effect was dampened when all species were combined even when the effects were consistent across species, particularly when it comes to the effect of environmental variables. Furthermore, the predictive ability of models was reduced when species were grouped and was the worst for all species combined.

Shark bycatch in the pelagic longline fishery can be influenced by bait type, soak time, hook shape, leader length and material, depth, and special shark-targeting gear (Clarke et al. 2014). Offset hook shape, number of hooks, and soak time all showed a positive relationship with catch. The use of offset hooks is also associated with increased release mortality due to a higher probability of gut-hooking (Kerstetter and Graves 2006). Hook size did not clearly delineate between small and large species, but this may be partly explained by the fact that the hook size variable recorded the manufacturer's size label, which may not be consistent across manufacturers. The actual measured size of the hook could potentially be a better variable for predicting the bycatch of sharks of varying sizes.

Effects of environmental variables on catch rates were consistent with the ecology of each species or species group, with coastal species caught more in coastal locations and shallow waters while DW species were caught more in deep waters. Although we expected this result, it is important to confirm, as other analyses of mitigation strategies based on gear configurations found that bycatch does not necessarily correspond to the species' typical habitat depth (Clarke et al. 2014).

No discernible yearly patterns were found, but for most species and species groups that retained the season variable, high catch was predicted to occur in May–August. Several of the target species have closed seasons, and there is a large area closure in the eastern GOM during summer months. It is unclear whether the changes in catch with time of the year are related to changes in temperature or changes in regulation.

Caveats and future research

Further research should directly examine trade-offs and the consequences of prioritizing one shark group over another or attempting to balance the needs of multiple groups at once. Because the species caught as bycatch in this fishery vary in their status, their productivity, and whether bycatch is an important source of their mortality, a management strategy evaluation could be used to weigh the trade-offs involved in mitigating bycatch for multiple species simultaneously. There is a documented relationship between Atlantic Sharpnose Shark CPUE and bait type (Driggers and Hannan 2019), with mackerel-baited hooks having a higher CPUE than squid-baited hooks. It is possible that combining this with the findings of the current study could be a way to mitigate the trade-offs in reducing the bycatch per set of multiple species simultaneously. For example, if DW species were prioritized, encouraging fishers to shift to shallower waters and bait their hooks with squid could allow for avoidance of DW sharks while reducing the catch per set of Atlantic Sharpnose Sharks.

One potential method for the balancing of trade-offs is implementing measures to avoid gear types or habitats with high bycatch in accordance with the findings discussed here and considering the mitigation hierarchy for sharks proposed by Booth et al. (2020), which was designed to assist in making science-based management decisions at the fishery level.

The impact of any proposed mitigation in this study does not consider the impact on the target species or any other bycatch species. Other studies on shark bycatch have found that reducing longline soak time can reduce shark bycatch per set without affecting Red Grouper Epinephelus morio CPUE (Foster et al. 2017) and that soak time for a pelagic longline could have minimal effects on the CPUE of other target species (Rice et al. 2012). Changing the hook shape to reduce shark bycatch may have unintended consequences for other species in this fishery—target and bycatch alike. Trade-offs and consequences for other species in the fishery should be incorporated in future studies.

We were unable to consider more rarely caught species due to data limitations. Other rarely caught sharks are prohibited species: the Dusky Shark Carcharhinus obscurus, Atlantic Angel Shark Squatina dumeril, and Smalltail Shark Carcharhinus porosus were caught 37, 7, and 4 times, respectively. Rare occurrence of high-CPUE events does not equate to insignificant catch (O'Farrell and Babcock 2021), and effective management requires an understanding of these rare events.

Recommendations

Considerations of species ecology coupled with fisher choice of gear and methodology have the potential to reduce shark bycatch per set in the GOMRBLL fishery. Species vary in the specific environmental and gear variables that can inform either voluntary changes or management actions to mitigate bycatch per set. However, there are indicators that are consistent across all species and groups, including hook shape, the number of hooks set, and soak time. Focusing on gear modifications is the only way that we found to reduce catch of all 12 species at once. Encouraging the use of straight hooks rather than offset hooks could reduce the catch of several shark species without increasing the bycatch of other shark species. To reduce shark bycatch, the number of hooks and the amount of time for which the hooks are in the water could be reduced. Future studies should explore and determine the optimal number of hooks to decrease shark bycatch while maximizing target catch.

Environmental and location-based variables show more variation across species and appear to be consistent with the ecology of each species. A strategy to minimize shark bycatch per set would need to manage trade-offs and prioritize some species over others. For example, depth was important for all but one species/species group, but the coastal species showed a pattern in direct opposition to the DW species. The environmental variable that appeared to be the most consistent across species was season. For seven species or species groups, the period January–April had the highest predicted catch per set. Therefore, a time closure that corresponds to January–April could move effort to times of the year when these sharks have lower bycatch rates. However, this would potentially be detrimental to the Sandbar Shark, Smooth Dogfish, and Silky Shark. Incentives that encourage more effort in May–August rather than January–April would move effort into the highest catch time for Sandbar Sharks and Smooth Dogfish, whereas encouraging more effort in September–December rather than January–August would move effort into the highest catch time for Silky Sharks but could be beneficial for Scalloped Hammerheads. For environmental conditions, the EcoCast integrated models of Hazen et al. (2018) suggested that avoidance of multiple bycatch species is possible by using fine time/area scales. A fine-scale, real-time approach like EcoCast, which predicts the occurrence of multiple species on a probability surface, could be the best implementation of measures with inconsistent results across species. Applied measures can be chosen based on which shark species have the highest probability of being in a given location.

CONCLUSION

This study supports the importance of the NOAA NMFS observer program. Although the observer program has covered only about 1% of fishing effort in recent years (National Marine Fisheries Service 2020), observer programs are generally the most accurate and reliable source of bycatch information (Suuronen and Gilman 2019). Observer programs collect data at the set level, whereas logbooks often report at the trip level and tend not to include fine-scale gear and environmental data. These set-level variables were shown to be important in predicting shark bycatch. Focusing on gear rather than environmental variables is the best apparent option to potentially reduce shark catch across commonly caught species. Other options based on environment and location force the acknowledgment of trade-offs. Sharks as a group should not be lumped together, as the signals become confounded and the direct management of trade-offs becomes impossible. At the very least, species should be analyzed in subgroups based on ecology; although predictive ability and the magnitude of variable signals are reduced, the data requirements are minimized while still maintaining the integrity of the patterns for the species that they represent.

ACKNOWLEDGMENTS

This publication was made possible by the Educational Partnership Program of the NOAA Office of Education (Award Number NA16SEC4810007). Its contents are solely the responsibility of the award recipient and do not necessarily represent the official views of the U.S. Department of Commerce, NOAA. Additional funding was provided by the NOAA through the Cooperative Institute for Marine and Atmospheric Science (Award Number NA20OAR4320472). Work was completed at the Florida Fish and Wildlife Conservation Commission Research Institute. We thank Enric Cortés, Paul Richards, Steve Smith, and the observers and staff of the GOMRBLL observer program.

CONFLICT OF INTEREST STATEMENT

The authors declare that they have no conflicts of interest.

DATA AVAILABILITY STATEMENT

Data sharing is not applicable to this article as no new data were created or analyzed in this study.

ETHICS STATEMENT

This article does not contain any studies involving animals or human participants performed by any of the authors.

References

Barton, K. (2015). MuMIn: Multi‐Model Inference. R package version 1.18. https://cran.r‐project.org/package=MuMIn

Booth, H., Squires, D., & Milner‐Gulland, E. J. (2020). The mitigation hierarchy for sharks: A risk‐based framework for reconciling trade‐offs between shark conservation and fisheries objectives. Fish and Fisheries, 21(2), 269–289. [DOI: https://dx.doi.org/10.1111/faf.12429]