Stable Attention Response for Reliable Precipitation Nowcasting

TL;DR

This paper introduces HARECast, a novel framework that stabilizes attention responses in precipitation nowcasting models, leading to improved reliability and state-of-the-art forecasting performance.

Contribution

It identifies cross-sample instability of attention responses as a key unreliability factor and proposes a regularization method to stabilize attention, applicable to various architectures.

Findings

HARECast reduces attention-response energy variance across samples.

Stabilized attention responses improve forecast accuracy.

HARECast achieves state-of-the-art results on benchmark datasets.

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.