Partitioning spatial dynamics in abundance of marine fisheries stocks between fine- and broad-scale variation: A Bayesian approach

Abstract

Fisheries stocks are often characterized by pronounced spatial patterning. This occurs at different scales, which may be driven by separate mechanisms. Regression kriging is often used to generate population estimates which may parse variation among these scales. A previous application of this and other frequentist methods to estimate the abundance of Atlantic sea scallops (Placopecten magellanicus) found that semi-parametric regression kriging performs best among model-based methods. We expanded upon these results by adapting the model structure for a Bayesian framework. This approach demonstrated the trade-off between abundance as a function of depth and broad spatial habitat features, and of fine-scale spatial aggregation. We applied our model to the 2015 belt transect survey of Atlantic sea scallops in Georges Bank and the Mid-Atlantic Bight off the eastern coast of the United States. The overall Bayesian predictions generally agreed with those of the frequentist method. However, the former favored broad spatial trends coupled with very fine-scale aggregation, while the latter favored finer trends and broad aggregation as the dominant driver of scallop abundance. Therefore, one can draw similar conclusions about abundance, yet disparate conclusions about the drivers. Careful consideration of the benefits and limitations of each approach is warranted.