New AI Model Forecasts Extreme Temperature Events With Unprecedented

Extreme temperature events are becoming more frequent as warming alters weather patterns, but forecasting them has remained difficult. Researchers at the Hangzhou Institute for Advanced Study of the University of Chinese Academy of Sciences developed a deep-learning model, Hankelformer, aimed at predicting non-stationary and extreme conditions. Published in National Science Review, the approach pairs structured time-series augmentation with contrastive learning. Across benchmark datasets, Hankelformer outperformed leading forecasting methods, cutting mean squared error by up to 34%. The model focuses on rare, short-lived, highly nonlinear transitions that standard time-series tools can miss.