Lightning Warning Algorithm Based on Multi-Station Atmospheric Electric Field and Data Augmentation

Abstract

An effective lightning warning system can ensure the safety of aircraft and promote the development of a low-altitude economy. Compared with weather radars, ground-based atmospheric electric field mills can monitor electric field variations in low-altitude regions in real-time without being affected by ground clutter. To address current challenges in lightning warning methods based on atmospheric electric field data—such as limited lightning location samples and a high false alarm rate (FAR)—this thesis proposes a lightning warning model that integrates multi-station atmospheric electric field data with meteorological variables such as temperature and humidity, combined with data augmentation techniques. First, temporal and lagging features of the electric field are extracted and fused with multidimensional meteorological data including temperature, humidity, wind speed, and total cloud cover. A spatial-temporal density-based spatial clustering of applications with noise (ST-DBSCAN) is employed to annotate samples across multiple stations. The mode-normalized Wasserstein generative adversarial network with gradient penalty (MN-WGAN-GP) is used to generate synthetic samples with distributions similar to real data. Finally, a lightning warning algorithm is constructed based on categorical boosting (CatBoost). Experimental results show that the model achieves a probability of detection (POD) of 86.49% and a FAR of 32.87% on the test set. The proposed algorithm contributes to the development of refined regional lightning warning technologies and ensures the safety of low-altitude operations.