Species-distribution shifts are becoming commonplace due to climate-driven change. Difficult decisions to modify survey extent and frequency are often made due to this change and constraining survey budgets. This often leads to spatially and temporally unbalanced survey coverage. Spatio-temporal models are increasingly used to account for spatially unbalanced sampling data when estimating abundance indices used for stock assessment, but their performance in these contexts has received little research attention. We therefore seek to answer two questions: (1) how well can a spatio-temporal model estimate the proportion of abundance in a new “climate-adaptive” spatial stratum? and (2) when sampling must be reduced, does annual sampling at reduced density or biennial sampling result in better model-based abundance indices? We develop a spatially varying coefficient model in the R package VAST using the eastern Bering Sea (EBS) bottom trawl survey and its northern Bering Sea (NBS) extension to address these questions. We first reduce the spatial extent of survey data for 30 out of 38 years of a real survey in the EBS and fit a spatio-temporal model to four commercially important species using these “data-reduction” scenarios. This shows that a spatio-temporal model generally produces similar trends and density estimates over time when large portions of the sampling domain are not sampled. However, when the central distribution of a population is not sampled the estimates are inaccurate and have higher uncertainty. We also conducted a simulation experiment conditioned upon estimates for walleye pollock (Gadus chalcogrammus) in the EBS and NBS. Many species in this region are experiencing distributional shifts attributable to climate change with species historically centered in the southeastern portion of the survey being increasingly encountered in the NBS. The NBS was occasionally surveyed in the past, but has been surveyed more regularly in recent years to document distributional shifts. Expanding the survey to the NBS is costly and given limited resources the utility of reducing survey frequency versus reducing sampling density to increase survey spatial extent is under debate. To address this question, we simulate survey data from alternative sampling designs that involve (1) annual full sampling, (2) reduced sampling in the NBS every year, or (3) biennial and full sampling in the NBS. Our results show that annual sampling, even with reduced sampling density, provides less biased abundance information than biennial sampling. We therefore conclude that ideally fishery-independent surveys should be conducted annually and spatio-temporal models can help to provide reliable estimates.
Read full abstract