Influence of model misspecification, temporal changes, and data weighting in stock assessment models: Application to swordfish (Xiphias gladius) in the Indian Ocean

Abstract Results from stock assessment modelling are often highly sensitive to model assumptions. It is therefore important that models are correctly specified and sensitivity analyses are conducted to evaluate the impact of model uncertainty. Model misspecification, changes in parameters over time, and data weighting are common issues that arise when developing stock assessment models. We conducted a stock assessment for swordfish Xiphias gladius in the Indian Ocean, using an integrated age-structured model, and evaluated estimates of management quantities under alternative assumptions about (1) changes in catchability for CPUE-based indices of abundance, (2) changes in gear selectivity, (3) weighting of abundance indices, and (4) the impact of sex-specific growth. The results indicated that assuming time-blocks for both catchability and selectivity may be appropriate to reflect the changes in fishing operations of Japanese and Taiwanese longline fleets. This assumption also provided better model performance and more optimistic assessment results because it implied that the decline in indices of abundance resulted from changes in catchability rather than depletion of biomass. Inappropriate choices for selectivity curves can deleteriously affect model performance, and attempting to account for model misspecification through downweighting of data may not be appropriate. Care must be taken when modelling changes in selectivity because selectivity can be distorted to accommodate changes in catchability. More generally, substantial changes in catchability (e.g., due to changes in targeting) may not be fully addressed in CPUE standardization and may require modelling changes over time in catchability within the stock assessment model. Finally, we found that misspecification of growth is at least as influential, if not more so, than misspecification of catchability and selectivity, or data weighting.