Stable Attention Response for Reliable Precipitation Nowcasting

2026-05-13

AI Summary

This work addresses the significant challenge of precipitation nowcasting, which stems from the highly localized and rapidly evolving nature of atmospheric processes. Existing attention mechanisms often overlook cross-sample stability in their responses, leading to unreliable predictions. To tackle this issue, the authors propose HARECast, a novel framework that identifies, for the first time, the cross-sample instability of attention head response energy as a key source of forecasting error. HARECast introduces a general head-wise energy regularization technique, combined with grouped regularization, a reconstruction branch, and a diffusion-based predictor, to explicitly regulate attention behavior. The method is applicable to both single- and multi-modal architectures and demonstrates substantial improvements in forecast accuracy and stability on the SEVIR and MeteoNet benchmarks, confirming its effectiveness and robustness.

Abstract

Precipitation nowcasting remains challenging due to the highly localized, rapidly evolving, and heterogeneous nature of atmospheric dynamics. Although recent methods increasingly adopt attention-based architectures in both unimodal and multimodal settings, they mainly emphasize stronger representation learning and prediction capacity, while paying less attention to the stability of attention responses across samples. In this work, we show that cross-sample instability of attention-response energy is an important and previously underexplored source of forecasting unreliability. Empirically, inaccurate forecasts are associated with larger attention-response energy variance across heads and layers. Theoretically, we show that cross-sample variability can propagate through self-attention, and enlarge a lower bound on prediction error. Based on this insight, we propose HARECast, a Head-wise Attention Response Energy-regulated framework for precipitation nowcasting. HARECast explicitly models head-wise attention-response energy and stabilizes it through a group-wise regularization objective that reduces cross-sample fluctuations. The proposed formulation is generic and applicable to both unimodal and multimodal nowcasting architectures. We instantiate HARECast in a standard forecasting pipeline with reconstruction branches and a diffusion-based predictor, and evaluate it on commonly used benchmarks--SEVIR and MeteoNet. Experimental results demonstrate that HARECast achieves state-of-the-art performance.

Problem

• precipitation nowcasting
• attention stability
• attention-response energy
• forecast reliability
• cross-sample variability

Innovation

• attention stability
• precipitation nowcasting
• attention-response energy
• self-attention regularization
• HARECast