Interpretable AI predicts a 2026 summer dry anomaly in central China
A deep learning model translates dynamical circulation predictions into precipitation estimates. Predictions initialized from March to May 2026 consistently indicate a dry anomaly over central China in summer 2026. Retrospective evaluations show higher predictive skill in analogue years featuring central equatorial Pacific warming persisting from winter into summer. Layer-wise relevance propagation (LRP) identifies northerly winds as the dominant driver. Perturbation tests confirm that removing LRP-identified features eliminates the dry anomaly.
Researchers employed a deep learning model to translate atmospheric circulation predictions into precipitation estimates. Initialized from March to May, the model consistently predicts a dry anomaly over central China for summer 2026. Retrospective evaluations revealed higher predictive skill in analogue years with central equatorial Pacific warming persisting from the preceding winter into summer. This warming favors an anomalous cyclonic circulation over the western North Pacific-South China Sea-South China region, inducing northerly winds and moisture divergence that suppress rainfall. Layer-wise relevance propagation independently identified these northerly winds as the dominant driver, and perturbation tests confirmed that removing LRP-identified features eliminates the dry anomaly.
The model uses deep learning to map circulation predictions to precipitation, with layer-wise relevance propagation providing interpretability. Perturbation tests validate the attribution by removing LRP-identified features, which eliminates the predicted dry anomaly. The approach demonstrates physically interpretable explanations for AI-derived seasonal forecasts.
Interpretable AI for seasonal climate prediction could improve trust and adoption in meteorological and agricultural sectors. The ability to attribute predictions to specific atmospheric drivers may support decision-making for water resource management and disaster preparedness.
The framework offers a method for producing explainable seasonal forecasts, which could be valuable for climate-sensitive industries such as agriculture, insurance, and energy. Enhanced interpretability may facilitate regulatory acceptance and operational deployment.
Observable next signals include validation of the 2026 summer precipitation anomaly against actual observations, further retrospective testing on additional years, and potential application of the interpretability framework to other regions or seasons.