Marine heatwaves are increasingly recognized as major extreme events that affect marine ecosystems, fisheries, coastal environments, and regional climate interactions. Their formation and evolution are influenced by both local thermal anomalies and broader ocean–atmosphere dynamic processes. However, accurately predicting marine heatwaves remains challenging because their characteristics vary substantially among open-ocean regions, marginal seas, and coastal transition zones.

This study develops a data-driven prediction framework for short-term marine heatwave forecasting over the western Pacific and adjacent seas. The framework integrates long-term oceanic and atmospheric information to characterize the spatiotemporal evolution of surface thermal extremes. Considering the strong regional heterogeneity of marine heatwave occurrence, the study further emphasizes region-specific modeling and evaluation, aiming to improve the representation of marine heatwave development under different dynamic and geographic conditions.

The proposed framework is designed to learn both normal background variability and extreme-event-related anomaly patterns. By combining multi-source environmental information with regionalized deep learning models, the study seeks to enhance the prediction of marine heatwave intensity and evolution at multiple lead times. The model outputs are evaluated in physical temperature anomaly space to ensure interpretability and practical relevance for marine environmental monitoring.

Preliminary experiments indicate that incorporating broader environmental information can improve the model’s ability to capture regional marine heatwave evolution, particularly in dynamically complex marginal seas and coastal regions. This study provides a useful methodological basis for intelligent marine heatwave prediction and offers potential support for early warning of ocean extreme events.