A Novel Model for Forecasting Geomagnetic Indices Using Machine Learning

Widely used geomagnetic activity indices like Kp or Dst, derived from the combined data from several observatories distributed worldwide, are crucial to forecasting since solar-driven geomagnetic activity can significantly affect technology and human activities on Earth and in near-Earth space. We developed a new model to forecast geomagnetic indices by incorporating predicted data from individual observatories. Unlike previous models that rely solely on an index and overlook local physical effects, our approach accounts for each observatory separately in the forecasting process, allowing for index predictions that integrate the same physical principles as in the original calculations of the index. We demonstrate the model's performance for Kp and the newer Hpo indices (Hp60 and Hp30), which measure planetary disturbances with higher resolution than Kp and without its upper limit of 9. The model demonstrates good agreement, accurately capturing trends and overall behavior, even with sparse solar wind data.