Automated Supervised Identification of Thunderstorm Ground Enhancements (TGEs)

TL;DR

This paper presents an automated supervised classification method that accurately detects Thunderstorm Ground Enhancements (TGEs) using a new dataset, combining machine learning with interpretability to improve detection and understanding of atmospheric phenomena.

Contribution

The study introduces a novel automated classification framework for TGEs that combines a TabPFN model with SHAP interpretability, achieving high accuracy and revealing data-driven detection thresholds.

Findings

Achieved 94.79% classification accuracy and 96% precision in TGE detection.

Identified data-driven thresholds that align with long-standing empirical criteria.

Demonstrated the method's potential for real-time radiation hazard monitoring.

Abstract

Thunderstorm Ground Enhancements (TGEs) are bursts of high-energy particle fluxes detected at Earth's surface, linked to the Relativistic Runaway Electron Avalanche (RREA) mechanism within thunderclouds. Accurate detection of TGEs is vital for advancing atmospheric physics and radiation safety, but event selection methods heavily rely on expert-defined thresholds. In this study, we use an automated supervised classification approach on a newly curated dataset of 2024 events from the Aragats Space Environment Center (ASEC). By combining a Tabular Prior-data Fitted Network (TabPFN) with SHAP-based interpretability, we attain 94.79% classification accuracy with 96% precision for TGEs. The analysis reveals data-driven thresholds for particle flux increases and environmental parameters that closely match the empirically established criteria used over the last 15 years. Our results demonstrate that modest but concurrent increases across multiple particle detectors, along with strong near-surface electric fields, are reliable indicators of TGEs. The framework we propose offers a scalable method for automated, interpretable TGE detection, with potential uses in real-time radiation hazard monitoring and multi-site atmospheric research.