Un modèle espace–état pour l’évaluation des stocks de morue de la sous-division 3Ps

Abstract

I describe an implementation of a state-space stock assessment model (SSM) for 3Ps cod. An assessment challenge for this stock is that there are uncertainties and possible biases in fishery landings information that are difficult to quantify. I use the censored likelihood approach based on the best available information on potential inaccuracies in landings to address this uncertainty. Hence, the SSM only uses information on lower and upper bounds of what the real fishery landings were. Information on the age-composition of the catches is included in the model fitting using a likelihood based on the continuation ratio logits of catch proportions-at-age. Otherwise, the SSM I describe is formulated like the typical SAM used in ICES assessments, with some differences in the stochastic model for variation in fishing mortalities (F’s) and the likelihood function for survey indices. I assume that survey indices have a normal distribution with a constant coefficient of variation. An advantage of this approach is that indices with zero values can be used for estimation, whereas they cannot be used with the more commonly used lognormal distribution. This is an important issue when there are many zeros that are not distributed at random throughout the survey ages and years, which is the case for 3Ps cod. Many model formulations were presented and compared in terms of AIC/BIC and other model results. Retrospective patterns were presented for some model formulations. The retrospective patterns were large enough to be a concern. The model formulations had trouble finding a good fit to the DFO RV and Sentinel indices. I did not find an acceptable assessment model formulation, and I concluded that further research on assessment inputs is needed before I could recommend a reliable assessment model formulation.