LAUSR

This article presents a multipath imaging system for thunderstorm developments, wherein data are three-dimensional atmospheric electric-field signals (3DAEFSs) collected with a self-made 3DAEF apparatus (3DAEFA). In this way, thunderstorms are presented in a staged and visual form. To start with, entropy-based intervals are constructed from historical AEF data, to classify denoised AEFS components, according to the entropy value of each component. Furthermore, AEFS time sequences are reconstructed with a reference to whether components within the same entropy-based interval are sequential or not, providing time information for the subsequent clustering-based spatial denoising to thunderstorm point charge coordinates. Finally, predicted value (PV) intervals, which are used to divide and then reconstruct AEFS time periods, are acquired to realize the point charge multipath imaging corresponding to periods, based on the established stacked autoencoder and the extreme gradient boosting (SAE-XGBoost) model. Empirical results demonstrate that the multipath better visualizes the whole process of thunderstorm activities. Comparisons with radar charts further confirm that the proposed system effectively images charge multipaths and provides a valid reference for visual thunderstorm monitoring.