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

2026-03-23

AI Summary

This work addresses the limited forecast skill of radar-only nowcasting beyond short lead times, which stems from the absence of large-scale atmospheric context. To overcome this limitation, we propose PW-FouCast, a novel framework that fuses radar observations with meteorological priors from the Pangu-Weather foundation model in the frequency domain via a Fourier-based backbone network, enabling synergistic modeling of multi-source heterogeneous data. The method innovatively introduces Pangu-Weather-guided frequency-domain modulation and a frequency-domain memory mechanism to correct phase errors, alongside a reverse frequency-domain attention module to reconstruct high-frequency details. Evaluated on the SEVIR and MeteoNet benchmarks, PW-FouCast significantly extends the reliable forecast horizon while preserving high-fidelity precipitation structures, outperforming current state-of-the-art approaches.

Technology Category

• Computer Vision: Multi-modal Vision
• Machine Learning: Multimodal Learning
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

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 spectral filtering. Extensive experiments on the SEVIR and MeteoNet benchmarks demonstrate that PW-FouCast achieves state-of-the-art performance, effectively extending the reliable forecast horizon while maintaining structural fidelity. Our code is available at https://github.com/Onemissed/PW-FouCast.

Problem

• precipitation nowcasting
• radar observations
• foundation models
• representational heterogeneity
• forecast horizon

Innovation

• spectral fusion
• foundation model priors
• frequency-domain modeling
• precipitation nowcasting
• Fourier-based backbone