Abstract
In many marine fisheries assessments, population abundance indices from surveys collected by different states and agencies do not always agree with each other. This phenomenon is often due to the spatial synchrony/asynchrony. Those indices that are asynchronous may result in discrepancies in the assessment of temporal trends. In addition, commonly employed stock assessment models, such as the statistical catch-at-age (SCA) models, do not account for spatial synchrony/asynchrony associated with spatial autocorrelation, dispersal, and environmental noise. This limits the value of statistical inference on key parameters associated with population dynamics and management reference points. To address this problem, a set of geospatial analyses of relative abundance indices is proposed to model the indices from different surveys using spatial hierarchical Bayesian models. This approach allows better integration of different surveys with spatial synchrony and asynchrony. We used Atlantic weakfish (Cynoscion regalis) as an example for which there are state-wide surveys and expansive coastal surveys. We further compared the performance of the proposed spatially structured hierarchical Bayesian SCA models with a commonly used Bayesian SCA model that assumes relative abundance indices are spatially independent. Three spatial models developed to mimic different potential spatial patterns were compared. The random effect spatially structured hierarchical Bayesian model was found to be better than the commonly used SCA model and the other two spatial models. A simulation study was conducted to evaluate the uncertainty resulting from model selection and the robustness of the recommended model. The spatially structured hierarchical Bayesian model was shown to be able to integrate different survey indices with/without spatial synchrony. It is suggested as a useful tool when there are surveys with different spatial characteristics that need to be combined in a fisheries stock assessment.
A hierarchical Bayesian model of the influence of run length on sequential predictions
