Deep learning and surplus production models for multivariate autoregressive modeling and simulation of the jack mackerel fishery associated with environmental conditions

The jack mackerel (Trachurus murphyi) fishery in the Southeast Pacific Ocean is strongly influenced by environmental conditions, particularly sea surface temperature (SST). This study aims to improve the predictive performance of fishery models by incorporating environmental variability into surplus production models and by applying advanced deep learning techniques. We developed and validated multivariate autoregressive models based on Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks to simulate jack mackerel catches under contrasting environmental scenarios. Sea surface temperature and fishing effort were used as the main explanatory variables, with lag structures included to capture delayed relationships between environmental conditions and fishery yields. The results show that both the CLIMPROD model and the deep learning architectures provide robust predictive performance. The selected CLIMPROD model achieved an R² of 0.87, while the best CNN and LSTM models attained R² values of 0.94 and 0.85, respectively. These approaches improve upon more classical formulations by accounting for the nonlinear effects of environmental variability on fishery dynamics. Simulations under different SST scenarios indicate that warmer periods tend to produce higher jack mackerel catches than colder periods, highlighting the influence of SST on fishery productivity. Overall, the findings support the value of incorporating environmental variability into fishery models to improve the accuracy and reliability of management-oriented analyses.