Evaluation of the impacts of different treatments of spatio-temporal variation in catch-per-unit-effort standardization models

Abstract

Abstract Many stock assessments heavily rely on indices of relative abundance derived from fisheries-dependent catch-per-unit-effort (CPUE) data. Therefore, it is critical to evaluate different CPUE standardization methods under varying scenarios of data generating processes. Here, we evaluated nine CPUE standardization methods offering contrasting treatments of spatio-temporal variation, ranging from the basic generalized linear model (GLM) method not integrating a year-area interaction term to a sophisticated method using the spatio-temporal modeling platform VAST. We compared the performance of these methods against simulated data constructed to mimic the processes generating fisheries-dependent information for Atlantic blue marlin (Makaira nigricans), a common bycatch population in pelagic longline fisheries. Data were generated using a longline data simulator for different population trajectories (increasing, decreasing, and static). These data were further subsampled to mimic an observer program where trips rather than sets form the sampling frame, with or without a bias towards trips with low catch rates, which might occur if the presence of an observer alters fishing behavior to avoid bycatch. The spatio-temporal modeling platform VAST achieved the best performance in simulation, namely generally had one of the lowest biases, one of the lowest mean absolute errors (MAEs), and 50% confidence interval coverage closest to 50%. Generalized additive models accounting for spatial autocorrelation at a broad spatial scale (one of the lowest MAEs and one of the lowest biases) and, to a lesser extent, non-spatial delta-lognormal GLMs including a year-area interaction as a random effect (one of the lowest MAEs and one of the best confidence interval coverages) also performed adequately. The VAST method provided the most comprehensive and consistent treatment of spatio-temporal variation, in contrast with methods that simply weight predictions by large spatial areas, where it is critical, but difficult, to get the a priori spatial stratification correct before weighting. Next, we applied the CPUE standardization methods to real data collected by the National Marine Fisheries Service Pelagic Observer Program. The indices of relative abundance predicted from real observer data were relatively similar across CPUE standardization methods for the period 1998–2017 and suggested that the blue marlin population of the Atlantic declined over the period 1998–2004 and was relatively stable afterwards. As spatio-temporal variation related to environmental changes or depletion becomes increasingly necessary to consider, greater use of spatio-temporal models for standardizing fisheries-dependent CPUE data will likely be warranted.