Thunderstorm Detection and Prewarning via Multisource Fusion of Precipitation and Atmospheric Electric Field Features

Traditional detection and prewarning methods based on a single atmospheric electric field (AEF) data source often fail to accurately assess thunderstorm weather conditions. This paper proposes a thunderstorm detection method that employs weighted multi-source fusion of precipitation and AEF features, achieving a high-precision evaluation through multi-modal feature fusion and machine learning model optimization. Based on the detection range of AEF apparatus, spatially matched radar chart precipitation data are extracted to construct physically meaningful precipitation statistic features and time-varying AEF features. Using feature importance analysis, the physical contribution of each feature is determined, and these features are weighted and fused accordingly to generate enhanced features. Furthermore, we establish an improved gradient boosting decision tree model that enhances classification performance in the weighted feature space by adaptively adjusting the learning rate. Experimental results demonstrate that this method achieves a highly competitive F1-score of 0.92 in thunderstorm recognition tasks, representing a 35.3% improvement over the traditional AEF threshold method. The proposed weighted fusion framework provides a novel solution for thunderstorm detection and prewarning through the synergistic use of ground-based AEF and radar-based precipitation data.