Evaluation of a Bayesian State-Space Stock Assessment Model for Kuskokwim River Chinook Salmon

Reliable estimates of run abundance and productivity are necessary in scientifically defensible salmon management, yet such estimates are often impeded by natural and sampling variability.  Bayesian state-space models are becoming increasingly popular in salmon stock assessments to cope with data-limitations and different sources of uncertainty, including both measurement error and process variation.  We applied this approach to 38 years of harvest and incomplete escapement sampling of the Kuskokwim River Chinook Salmon stock to reconstruct historical drainage-wide run abundance while simultaneously estimating stock-recruitment parameters.  We evaluated sensitivity of biological reference point estimates to various structural assumptions including (1) assumed variances, (2) whether parameters should be allowed to vary over time, and (3) whether accounting for the effects of size-selective harvest on spawning stock biomass should be incorporated (i.e. escapement quality).  Sensitivity analyses suggested that the model is robust to assumed observation variances within realistic ranges.  Inspection of residuals from escapement indices indicated that these counts showed time-series patterns which justified incorporation of a first order autoregressive process on scaling parameters (i.e. catchability).  Structuring the model to include annually-varying escapement quality allows for investigation of the hypothesis that size-selective harvest plays an important role in salmon population dynamics.