The geomagnetic storm and Kp prediction using Wasserstein transformer
Beibei Li
TL;DR
This work tackles the challenge of predicting planetary Kp index both 3- and 5-day horizons by fusing heterogeneous data sources (sun imagery, satellite measurements, and KP time series) within a Wasserstein-aligned Transformer framework. The model processes each modality through dedicated encoders, projects outputs to probability distributions, and enforces cross-modal distribution alignment via a Wasserstein-distance loss, while a multi-branch output head delivers coarse-to-fine KP classifications and a high-KP flag. A sliding-window data preparation with class-balanced expansion, along with online fine-tuning, enables robust forecasting and prevents data leakage, achieving competitive performance against NOAA baselines and capturing both quiet and storm periods. The approach demonstrates the value of integrating ML with traditional space-weather models for real-time, multi-day geomagnetic activity prediction, with potential to improve resilience of energy, communications, and navigation systems.
Abstract
The accurate forecasting of geomagnetic activity is important. In this work, we present a novel multimodal Transformer based framework for predicting the 3 days and 5 days planetary Kp index by integrating heterogeneous data sources, including satellite measurements, solar images, and KP time series. A key innovation is the incorporation of the Wasserstein distance into the transformer and the loss function to align the probability distributions across modalities. Comparative experiments with the NOAA model demonstrate performance, accurately capturing both the quiet and storm phases of geomagnetic activity. This study underscores the potential of integrating machine learning techniques with traditional models for improved real time forecasting.