A Deep Learning-Based Precipitation Nowcasting Model Fusing GNSS-PWV and Radar Echo Observations

Nowcasting plays a critical role in disaster warning systems, and recent advancements in deep learning have shown great potential in improving the accuracy and timeliness of such predictions. This study proposes a novel deep learning-based model for precipitation nowcasting, which integrates global navigation satellite system (GNSS)-derived precipitable water vapor (PWV) data with radar observations. The model introduces two key innovations: multi-source data fusion and time-dimension attention mechanism. These advancements enhance the model’s capability to accurately forecast precipitation events, particularly under challenging conditions with high rainfall intensity. In comparative experiments conducted using radar and GNSS data from Hong Kong, the model, incorporating both data fusion and the attention mechanism, demonstrated the best overall performance, with critical success index (CSI) scores increasing by 26% and Heidke skill score (HSS) scores by 23% at the 30 mm/h threshold. Moreover, it effectively simulates rainfall regions and their changing trends, demonstrating the complementary value of GNSS PWV data to radar observations.