Interpretable Deep Learning Predicts 2026 Summer Drought Anomaly in Central China

Forum topic · 小凯 · 2026-08-21

Summary

A new arXiv paper (2608.19163) by Anran Wang, Wen Shi, Yong Luo, Jianbin Huang and colleagues presents an interpretable deep learning framework for seasonal climate prediction. The model translates dynamical atmospheric circulation forecasts—more reliable than precipitation forecasts—into precipitation estimates. Predictions initialized between March and May consistently indicate a dry anomaly over central China in summer 2026. Retrospective evaluation shows higher predictive skill in analogue years, which tend to feature central equatorial Pacific warming persisting from the preceding winter into summer. This warming induces an anomalous cyclonic circulation over the western North Pacific–South China Sea region, producing northerly winds and moisture divergence that jointly suppress rainfall. Layer-wise relevance propagation (LRP) independently identifies these northerly winds as the dominant predictive driver among all model inputs, and perturbation tests confirm the attribution: removing LRP-identified features eliminates the drought anomaly. The framework thus provides physically interpretable, evidence-based evaluation of AI-derived regional climate predictions before observations become available.

Overview

Field: Machine Learning / Climate Prediction Authors: Anran Wang, Wen Shi, Yong Luo, Jianbin Huang, Lijuan Chen, Junhu Zhao, Weixin Jin, Huihui Yuan Published: 2026-08-19 arXiv: 2608.19163

Abstract (translated from Chinese summary)

Seasonal precipitation anomalies are largely regulated by atmospheric circulation, which dynamical models predict with greater reliability than precipitation itself. The authors employ a deep learning model that translates dynamical circulation predictions into precipitation estimates. Predictions initialized from March to May consistently indicate a dry anomaly over central China in summer 2026.

Retrospective evaluations revealed higher predictive skill in analogue years, which also tended to feature 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, which induces northerly winds and moisture divergence that jointly suppress rainfall over central China.

Supporting this mechanism, layer-wise relevance propagation (LRP) independently identified these northerly winds as the dominant predictive driver among all model inputs. Perturbation tests support this attribution: removing the LRP-identified features effectively eliminates the drought anomaly.

Key contributions

A deep learning pipeline converting dynamical circulation forecasts into precipitation estimates
Consistent prediction of a summer 2026 dry anomaly over central China
Physical mechanism: central equatorial Pacific warming → anomalous cyclonic circulation → northerly winds + moisture divergence → suppressed rainfall
Interpretability via LRP, validated by perturbation tests
Enables evidence-based evaluation of AI seasonal forecasts before observations are available

--- *Auto-collected on 2026-08-21*