A Bayesian Version of the NIWA Two-Stock Hoki Model.
Ray Hilborn, Paul J. Starr, Vivian Haist · ResearchWorks at the University of Washington (University of Washington) · 2001
A Bayesian implementation of the National Institute for Water and Atmospheric Research (NIWA) hoki (Macruronus novaezelandiae) model (Cordue 1999) is described.This implementation, termed the UW/Seafic implementation, is based on the documentation and data provided in Cordue (1999) with minor differences.Overall model fit for the UW/Seafic model implementation is similar to that reported by Cordue.Differences in model fit occur primarily in data series that use age composition data from juvenile hoki.The model estimator was changed from a least-squares formulation to a maximum likelihood formulation to implement Bayesian methods to describe the posterior distributions of key parameters.Posterior distributions of the biomass trajectories and their associated confidence bounds show little effect of the inclusion of data in the model; that is, there is little attenuation in the width of the confidence bounds over the historical trajectory.This is interpreted as evidence that the model structure and the assumed bounds impose constraints on model output or that the data are not providing much information to the model.The high penalty weights imposed by the assumption that the survey proportionality constants should be similar among areas for the same survey type have a large effect on model biomass estimates.Biomass estimates for the western region hoki are nearly doubled when these penalty weights are relaxed while the biomass estimates for the eastern region hoki are about 20% smaller.The sensitivity of important model estimates on an untestable assumption is a poor attribute for this model.The authors conclude that the NIWA two-stock multi-area hoki model is unnecessarily complex and over-parameterized.Insufficient data exist to estimate the nearly 200 model parameters, and the estimates for most of these parameters are not well determined.The elaborate model structure allows inclusion of previously omitted data in a fashion consistent with current hypotheses of hoki population dynamics and stock structure.However, whether these additional data and the increased complexity of the model have improved the quality and precision of the stock abundance estimates is unclear.