Lightning nowcasting based on high-density area and extrapolation utilizing long-range lightning location data

Abstract

Lightning poses varying degrees of threat to human production, life and personal safety because of its huge destructive power. Lightning prediction with intensity estimation is an important reference for disaster warning, determining lightning protection level and real-time and dynamic lightning risk assessment. Real-time observation of total lightning can compensate for the problem of limited range or shielding of radar observation and is of great significance for thunderstorm tracking and lightning nowcasting. In this paper, the VLF Lightning Location Network (VLF-LLN) is used to obtain real-time long-range lightning location data. Through density estimation based on the Gaussian mixture model and weighted according to time proximity, the spatio-temporal range and frequency information of lightning high-density area are obtained. Then, a spatio-temporal data extrapolation model ME-RNN is proposed to predict the intensity of lightning activity and its evolution pattern in the thunderstorm region, so as to realize the nowcasting of lightning activity. Based on the PredRNN architecture, the model integrates the MotionGRU operator optimized for the characteristics of lightning data, and integrates the generative adversative network (GAN) to capture the transient changes of lightning phenomena and predict its movement trend. At 6 min, 18 min and 36 min before the thunderstorm reached the area of concentration (AOC), the prediction accuracy of the model was 94.1 %, 79.4 % and 65.8 %, respectively. Furthermore, the ME-RNN model can track the active thunderstorm cells associated with lightning activity, and realize the prediction of the separation and merging of lightning high-density area, as well as the prediction of moving trend.