Evaluating multispecies survey designs using a joint species distribution model

Abstract

Survey designs should be efficient as marine survey programs are usually expensive and time-consuming; however, surveys have rarely been evaluated for multiple species. In the present study, we evaluated multispecies fisheries surveys with respect to three influential factors, i.e., sampling methods, estimation methods and sample size. A joint species distribution model (JSDM) developed in north Yellow Sea, China was used as the operating model to simulate the spatial distribution of multiple species simultaneously. We examined the precision of multispecies abundance estimation using diverse sampling methods [random sampling (RDS), systematic sampling (SYS), stratified random sampling (SRS), generalized random-tessellation stratified sampling (GRT) and spatial coverage sampling (SPC)], estimation methods [arithmetic mean (Arm), universal kriging (Ukr), multivariate distribution model (Mvd), and boral model (Brm)], and a range of sample sizes (from 30 to 300). The results showed significant differences in estimation among sampling methods, where GRT and SYS yielded less relative absolute bias (RAB) over all and RDS showed the least precision. Regarding estimation methods, Mvd and Arm showed the best performances and Brm yielded the least precision. Significant interactions existed between sampling and estimation methods. Arm worked best with GRT, likewise Mvd with SYS and Ukr with SPC. SPC and Mvd showed the best performances for a small sample size (N = 30), and all sampling and estimation methods provided similar results for a large sample size (N = 300). Generally, doubling sample size resulted in a decrease of RAB by 0.097 on average, a rate depending on species, sampling and estimation methods. This study contributed to an integrative framework for evaluating designs of multispecies fisheries surveys.