Multimodal Measurement Framework for Thunderstorm Charge Motion: Spatiotemporal Sensor Fusion With Bayesian-Optimized Localization

Xu Yang, Hongyan Xing, Fa Zhu, Yong Chen, David Camacho, Xudong Dong, Witold Pedrycz

Abstract: The accurate measurement of thunderstorm cloud point charge motion is critical for analyzing thunderstorm dynamics, but remains challenging due to the limitations of single-modal atmospheric electric field (AEF) data. This article presents an enhanced multimodal sensor fusion framework that synergizes 3-D AEF measurements with radar echo intensity (REI) and precipitation data. The methodology introduces several key instrumentation advancements: 1) a spatiotemporal calibration method that resolves the sampling rate conflict (1 s AEF versus 6 min radar) and incorporates two AEF features other than statistics—amplitude of change (AC) and zero-crossing time—to capture nonstationary AEF signal dynamics better; 2) a physics-constrained feature fusion strategy that integrates the AEF features with REI and precipitation data, followed by a Cohen’s d-based selection to construct a highly discriminative feature vector; 3) a multiobjective Bayesian optimization (MOBO) scheme for the random forest (RF) model, simultaneously minimizing classification error and model complexity; and 4) a trajectory postprocessing module that applies smoothing and denoising to the 3-D point-charge localization results, yielding physically plausible motion paths. Validated against multiple weather events, the framework demonstrates superior performance in weather attribute (WA) classification and achieves dynamic, high-fidelity visualization of charge trajectories that show strong consistency with independent radar observations, establishing a new channel for extreme weather monitoring.

External IDs:doi:10.1109/tim.2025.3637944