Good practices for including environmental data to inform spawner-recruit dynamics in integrated stock assessments: Small pelagic species case study

When assessing the status of a fish population for advising management, the spawner-recruit (S-R) relationship, which represents stock productivity, is an important biological process to consider, including the extent to which environmental drivers should be explicitly accounted for in recruitment estimation. Fishery managers are increasingly making requests for analysts to investigate the utility of including environmental data in stock assessment applications. Consequently, research is needed that addresses both specific environmental processes that may influence stock parameters and robust methods for including such data in the assessment. Two methods for incorporating environmental information in an integrated assessment model were evaluated in this study, based on: 1) including an environmental covariate as an additional component of the S-R function; or 2) using the covariate outside the S-R relationship as a survey index of recruitment. The methods were implemented in a broadly employed stock assessment software package (Stock Synthesis) using population/fishery processes generally applicable to an exploited small pelagic fish species (Pacific sardine). Simulation analysis was used to examine how environment-recruitment considerations in the assessment influenced model performance. The quality (bias and precision) of estimates of abundance and derived quantities generated from the models were examined statistically. Findings indicated that both methods for including an environmental index in the model resulted in relatively high quality estimates for a diverse group of output variables, depending on assumptions regarding bias correction and penalties associated with the recruitment deviation estimates. Several estimated quantities were found to be generally unbiased and relatively precise (e.g., stock biomass and average recruitment), regardless of the method implemented in the model; whereas, the choice of the method did impact the quality of results associated with more sensitive quantities (e.g., virgin recruitment and depletion). Recommendations for approaches based on both methods, as well as practical considerations associated with the methods are discussed.