REAL-TIME GEOMAGNETIC STORMALERTS USING AI-BASED FORECASTING MODELS

Abstract

Geomagnetic storms can severely impact critical technologies such as satellite communications, GNSS navigation, and power grids. Traditional forecast models based on empirical relationships and physics-based simulations often suffer from latency and limited spatial resolution. This paper explores the implementation of artificial intelligence (AI)-based forecasting models for issuing real-time geomagnetic storm alerts, leveraging historical solar wind data, IMF parameters, and geomagnetic indices. Multiple AI techniques including Random Forests, LSTM networks, and hybrid CNN-RNN models were trained on NASA OMNI and Kyoto WDC datasets. Results demonstrate a substantial improvement in prediction accuracy for Dst and Kp indices with lead times exceeding 1 hour. The developed AI framework has the potential to provide timely warnings for space weather disturbances in South Asia and globally.