Allocating catches to species in mixed species data – A model-based approach

Abstract

Stock assessments should ideally be conducted at the species level, but this can be difficult when commercial catches are reported by species group rather than by species. This is a common problem if species are hard to identify or it is too time consuming to do so. Species split algorithms are applied in such cases. We outline a GAM-based approach for splitting species-aggregated catch and effort data to species and apply them to two species groups (tiger and endeavour prawns) in Australia’s Northern Prawn Fishery. The best models incorporate spatial, temporal and biophysical covariates, demonstrating strong explanatory power and robust performance in cross-validation tests. The results suggest that the annual split to species by location and day of the year is constant over time for the endeavour prawns. However, evidence exists for long-term trends (i.e., non-stationary) when splitting the tiger prawns. This highlights the importance of regularly updating this information.