Answers are grounded in this paper and cite their sources.
Understanding habitat-environment dynamics is crucial for sustainable pelagic fisheries. However, existing forecasting methods often ignore vertical environmental constraints and struggle with zero-inflated, spatiotemporally complex fishery data. To address this, we propose a Hybrid Gaussian Spatiotemporal Transformer (HGP-ST) model that incorporates 3D vertical environmental features to predict yellowfin tuna catch per unit effort (CPUE). Using 2010–2023 monthly longline fishery data and environmental data, we constructed multiscale environmental features encompassing surface dynamics, physiological thresholds, thermocline characteristics, and vertical stratification differences. The HGP-ST model introduces a Gaussian process prior before Transformer fusion to mitigate modeling instability under sparse sample conditions. Concurrently, its dual-branch architecture mitigates the zero-value inflation problem by estimating the probability and intensity of capture events. The results demonstrate that the HGP-ST model achieves improved predictive performance compared to traditional baselines. Vertical stratification and physiological thresholds are identified as key environmental drivers contributing to the model's prediction output. Furthermore, the model captures the tropical-to-subtropical catch per unit effort (CPUE) spatial gradient and tracks the intra-annual evolution of central fishing grounds. This approach improves the representation of complex fisheries data, providing a reliable data foundation for resource assessment and operational planning in the Western and Central Pacific Ocean (WCPO).
No takes yet. Share an insight, caveat, or question.
Ji et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: