Bayesian hierarchical stock-recruitment models for setting conservation limits for Atlantic salmon stocks in Scotland

Atlantic salmon (Salmo salar L.) populations in Scotland are subject to active management and conservation practices which require biological reference points (BRPs), specifically conservation limits, defined at the level of the stock. Acquiring the data necessary to independently derive these BRPs for all managed populations in Scotland is prohibitive, motivating the use of Bayesian hierarchical stock-recruitment models. These models provide a framework for the joint analysis of multiple monitored stocks, and the transportation of BRPs to non-monitored stocks. This framework was adapted to introduce nationally relevant and available covariates that might explain variation in recruitment dynamics among stocks and reduce uncertainty in posterior predictions of BRPs. Model selection was designed to maximise the prediction of BRPs for new stocks via leave-one-group-out cross-validation. Out-of-sample predictive performance was maximised by including information on latitude, land usage within the catchment and historic catch per area of salmon habitat in the model. The extensions to Bayesian hierarchical stock-recruitment methods presented here, when applied at a national scale, result in more locally discriminative posterior predictions compared to existing methods and are readily applicable to other stocks and species.