A Convolutional Neural Network‐Based Model for Precipitation Nowcasting Leveraging Data From Gauge Stations

Abstract Rainfall nowcasting, the short‐term prediction of precipitation, is a vital component of early warning systems aimed at mitigating the effects of extreme weather events. In this study, we develop a deep learning approach based on convolutional neural networks (CNNs) for rainfall nowcasting and train it exclusively on data from rain gauge stations. Specifically, we adapt the original U‐Net architecture and tailor it for regression tasks to forecast short‐term precipitation at the Forio rain gauge station on the Island of Ischia, Italy. Two model input configurations are examined to assess the potential value of incorporating data from multiple sources: (a) using data solely from the Forio station, and (b) integrating data from multiple rain gauge stations across the island. The CNN‐based nowcasting performance is evaluated across various lead times, ranging from 10 min to 3 hr. The results show that both models achieve high predictive accuracy ( across most lead times), with the single‐station model outperforming the multi‐station configuration, indicating that adding data from other stations does not necessarily improve forecasting performance. These findings contribute to the advancement of rainfall nowcasting to inform real‐time early warning systems.