Assessing climate change impacts for small-scale fisheries in the Gulf of California using deep learning

The complexity of small-scale multi-species fisheries, which often lack detailed catch data, creates significant obstacles to measuring and addressing the potential impacts of climate change, hindering efforts to forecast future conditions and develop suitable management strategies. Deep learning techniques offer a promising tool for recognizing patterns and predicting trends despite these limitations, offering valuable insights for fisheries management in data-limited settings. We introduce the first application of a Mixture of Experts deep learning architecture to forecast catches in multi-species fisheries for the Gulf of California, Mexico, under a future high-emissions scenario. Unlike previous single-species studies, our approach simultaneously models catch forecasts for more than 80 species across multiple habitat groups under future climate change scenarios. Our results showed diverse responses in fisheries catch across habitat groups, with reef and benthic fish expected to decline significantly (-12.46 % and −9.37 %) during the 2050s-2060s, followed by recovery in the 2070s-2080s. We applied a Shapley Additive Explanations (SHAP) analysis to evaluate the importance of features for each predictive model. This analysis revealed the impact of temperature at different depths on each fishery, and a sensitivity analysis highlighted the magnitude of these effects. Our findings indicate that climate impacts will vary across the Gulf of California, emphasizing the need for region-specific management approaches and underscoring the importance of maintaining diverse fishing portfolios to build resilience against climate-driven changes.