Convective-gust nowcasting based on radar reflectivity and a deep learning algorithm

Abstract. Convective wind gusts (CGs) are usually related to thunderstorms, and they may cause great structural damage serious hazards, such as train derailment, service interruption, building collapse. Due the small-scale nonstationary nature of CGs, reliable CG nowcasting with high spatial temporal resolutions has remained unattainable. In this study, a novel model based on deep learning – namely, CGsNet is developed for 0–2 h lead times quantitative nowcasting, achieving minute–kilometer-level forecasts. physics-constrained established by training large corpora average surface speed (ASWS) radar observations; it can produce realistic spatiotemporally consistent ASWS predictions in events. By combining gust factor (1.77, ratio observed peak (PWGS) ASWS) predictions, PWGS forecasts estimated resolution 0.01∘ × 6 min resolution. shown be effective, an essential advantage spatiotemporal features CGs. addition, evaluation experiments indicate that exhibits higher generalization performance CGs than traditional method numerical weather prediction models. CG-nowcasting technology applied provide real-time