[2603.21768] Extending Precipitation Nowcasting Horizons via Spectral Fusion of Radar Observations and Foundation Model Priors

Computer Science > Machine Learning arXiv:2603.21768 (cs) [Submitted on 23 Mar 2026] Title:Extending Precipitation Nowcasting Horizons via Spectral Fusion of Radar Observations and Foundation Model Priors Authors:Yuze Qin, Qingyong Li, Zhiqing Guo, Wen Wang, Yan Liu, Yangli-ao Geng View a PDF of the paper titled Extending Precipitation Nowcasting Horizons via Spectral Fusion of Radar Observations and Foundation Model Priors, by Yuze Qin and 5 other authors View PDF HTML (experimental) Abstract:Precipitation nowcasting is critical for disaster mitigation and aviation safety. However, radar-only models frequently suffer from a lack of large-scale atmospheric context, leading to performance degradation at longer lead times. While integrating meteorological variables predicted by weather foundation models offers a potential remedy, existing architectures fail to reconcile the profound representational heterogeneities between radar imagery and meteorological data. To bridge this gap, we propose PW-FouCast, a novel frequency-domain fusion framework that leverages Pangu-Weather forecasts as spectral priors within a Fourier-based backbone. Our architecture introduces three key innovations: (i) Pangu-Weather-guided Frequency Modulation to align spectral magnitudes and phases with meteorological priors; (ii) Frequency Memory to correct phase discrepancies and preserve temporal evolution; and (iii) Inverted Frequency Attention to reconstruct high-frequency details typically lost in s...