Forecasting Geomagnetic Disturbances with Interpretable Deep Learning

Abstract

Geomagnetic disturbances significantly impact Earth, affecting spacecraft operations, power grids, communication systems, among others. The Kp index, a widely used geomagnetic disturbance measure, requires accurate prediction to achieve effective space weather monitoring. In this paper, we present an interpretable deep learning approach to predicting the Kp index.
We leverage SHAP (SHapley Additive exPlanations) and PDP (partial dependence plots) to analyze feature importance and model decision-making. Our approach provides reliable forecasts while offering insight into the underlying factors that influence geomagnetic activity. Experimental results demonstrate the good performance of the proposed approach.