A hybrid physical–deep learning approach for high-resolution detection of marine heatwaves

Marine heatwaves (MHWs) have intensified globally in recent years, driving widespread coral bleaching, ecosystem degradation, and escalating economic losses. In some coastal regions, coral bleaching rates have exceeded 80%, while fisheries-related damages have been estimated at over USD 3.1 billion annually. Despite these impacts, the characteristics and underlying mechanisms of nearshore MHWs remain poorly constrained, largely due to the lack of high-resolution sea surface temperature (SST) observations. Thermal infrared satellite products are subject to persistent data gaps caused by cloud cover, particularly in coastal environments where strong land–sea interactions, multiscale physical processes, and pronounced spatial heterogeneity limit conventional MHW detection.

A physics-informed deep-learning framework is developed to reconstruct all-weather, high-resolution SST fields for nearshore regions. By integrating physical constraints with multi-source geophysical predictors, the approach generates a 2 km-resolution SST dataset with high accuracy, achieving a root-mean-square error of 0.30 °C, a mean bias of 0.01 °C, and a coefficient of determination (R²) of 0.99 against independent reference observations. The reconstructed SST fields enable robust identification of nearshore MHWs and resolve fine-scale thermal structures that are not captured by existing coarse-resolution datasets.

Based on the reconstructed SST product, the spatiotemporal evolution of nearshore MHWs is systematically characterized, and associated physical and ecological implications are examined. Case studies in the South China Sea and the Mediterranean Sea reveal multiple unprecedented extreme events in recent years, including record-breaking MHWs in the Mediterranean during the past three years in terms of both intensity and spatial extent. High-resolution analyses further reveal an enhanced spatial correspondence between MHWs and coral reef distributions, indicating intensified thermal stress in ecologically vulnerable coastal zones. Accurate, all-weather, high-resolution SST reconstruction therefore provides a critical basis for advancing nearshore MHW detection and improving assessments of emerging coastal climate risks.