The primary source of anthropogenic mortality for the endangered leatherback sea turtle (Dermochelys coriacea) and threatened loggerhead sea turtle (Caretta caretta) is thought to be bycatch in commercial fisheries, including the U.S. pelagic longline fishery. The National Marine Fisheries Service (NMFS) is charged with annually estimating the number of takes by this fishery. These bycatch estimates have serious implications for species conservation and fishery practices. However, there is no consensus on the most appropriate estimation methodology, and the resulting estimates are highly uncertain. This is partially because only 5-8% of vessels have fishery-independent observers who record bycatch data and because takes occur very infrequently. Also, since NMFS distributes observers in proportion to the amount of fishing in the previous year-fishing area-calendar quarter, there are some strata with low historical coverage. Due to variability in the system, it can be difficult to extrapolate bycatch estimates to these strata, and some bycatch hotspots may be missed through this method of distributing observers. In this study we use a simulation model to examine the bycatch estimation performance of the delta-lognormal approach and a generalized linear model under two spatial sea turtle distributions (non-clumped, clumped).

Results/Conclusions

The results of this simulation modeling analysis suggest that while a leading estimation method has the potential to be a powerful tool for estimating sea turtle bycatch, the suitability of an estimation method largely depends on the circumstances. The sea turtle spatial distribution affects the structure of the observer data upon which bycatch estimates are made, and the data structure influences which method is most appropriate. This study is being expanded to include more estimation methods, several spatial fishing vessel distributions, and various spatial observer distributions. Further, this project not only evaluates methods for estimating sea turtle bycatch by the pelagic longline fishery but also provides results potentially applicable to any zero-heavy dataset.