Predicting unseen chub mackerel densities through spatiotemporal machine learning: Indications of potential hyperdepletion in catch-per-unit-effort due to fishing ground contraction

Abstract

In fisheries management, accurate estimates of fish stock abundances are crucial for sustainable harvesting practices. Traditional methods often rely on catch-per-unit-effort (CPUE) data, assuming fishing effort is uniformly distributed across the stock range. However, this assumption is often violated, leading to potential biases in CPUE-based abundance indices (AI). In the present study, we focused on chub mackerel stock in the East Asian Marginal Seas (EAMS), where shifting fishing grounds and ocean warming have raised concerns regarding the reliability of the nominal CPUE trend. We developed a spatiotemporal machine learning approach to predict the CPUE values while taking into consideration environmental variables and changes in fish distribution. Our model accounts for unexploited areas, thereby addressing the sampling biases inherent to traditional CPUE analyses. The results suggest that recent declines in the nominal CPUE observed in Japan do not reflect the actual stock declines but instead reflect biases due to shrinking fishing areas. These findings highlight the need for more sophisticated methods in fisheries management to ensure sustainable practices and highlight the importance of considering environmental and distributional changes in fish stock assessments.

Keywords

• Abundance index
• Hyperstability
• Fisheries sampling bias
• Fishing ground contraction
• Model stacking
• Spatiotemporal cross-validation