Practical uses of non-parametric methods in fisheries assessment modelling

Abstract

The vast majority of fisheries stock assessment modelling is parametric, where specific models are assumed and fitted to data, the results of which are used to assess stock status and provide scientific advice. Often, the assumed models may not acceptably explain the data, or the data are not informative enough to estimate the parameters of even the most simple models. Using a fully inferential statistical framework, artificial neural networks were fitted to example data sets (stock-recruit, catch and relative abundance) and key assessment quantities such as maximum sustainable yield and relative biomass depletion were estimated. The combination of flexibility and statistical rigor suggests there is an as yet under-utilised role for such approaches in stock assessment, and not just in data-poor scenarios.