Optimized stratified random surveys best estimate multispecies abundance in a rapidly changing ecosystem

Abstract Data from fishery-independent surveys are critical inputs to stock assessments, ecosystem-based fishery management, and applied ecological research. However, environmental change may affect species distributions and their availability to surveys, with consequences for the consistency and precision of abundance estimates over time. We investigated whether defining survey stratum boundaries by environmental conditions improves the precision and accuracy of abundance estimates in a multispecies survey. We fitted univariate spatiotemporal species distribution models to 16 stocks (14 species) using historical observations of fishery-independent bottom trawl survey catch-per-unit-effort and sea bottom temperature in the eastern and northern Bering Sea from 1982 to 2022. These spatiotemporal models were used to simulate species distributions and survey observations under a variety of environmental conditions and survey designs. The predicted density of each species at each location and time was passed to a multivariate optimization routine to determine whether this could increase the accuracy of estimates of abundance per unit of survey effort across species relative to traditional survey designs. Historical and projected future abundances for 10 of the 16 stocks were estimated more precisely under optimized designs–up to 4× as precise as the existing design. The accuracy of the estimate of abundance precision was always lowest for systematic sample allocation and highest for random or balanced random sampling within strata, suggesting that designs optimized with historical biological and environmental data lead to a better ability to quantify survey precision. The approach developed here can be applied in other ecosystems experiencing change to support the design of flexible survey designs that could increase the efficiency of sampling marine resources under current and future climates.