Improving estimation of length–weight relationships using spatiotemporal models

Abstract

Length–weight relationships (LWRs) are an essential component of fishery stock assessments. They are used to develop indices of condition and to convert length data into estimates of biomass. Attempts to capture variability in underlying ecological processes within statistical modeling frameworks for LWRs have typically relied on the inclusion of environmental variables. Here, using a case study of sea scallop ( Placopecten magellanicus), we demonstrate that introducing spatiotemporal random effects into generalized linear mixed models can improve LWRs. We compare models with and without potentially informative environmental variables. We find that the explicit incorporation of spatiotemporal dependence structures reduces bias and increases precision in the estimation of weight. The combination of both spatiotemporal effects and environmental variables provided the best predictions in most years. Spatiotemporal random effects can provide a comprehensive means of improving LWRs for various species, even when influential environmental variables are unavailable.