Predicting Ionospheric Cusp Location From Solar Wind: An XGBoost Model Trained on 27 Years of DMSP Data

Abstract

The ionospheric cusp provides a pathway for magnetosheath plasma to enter the upper atmosphere. Predicting its latitude requires jointly considering solar wind driving, magnetic geometry, and their recent evolution. We train an XGBoost model on 39,668 Defense Meteorological Satellite Program cusp crossings from 1987 to 2014 using 74 solar wind, interplanetary magnetic field, geometric, and temporal features. The primary target is the absolute magnetic latitude of the equatorward precipitation boundary along the satellite track. Training on data before 2008 and testing on 2008–2014 yields a mean absolute error of 1.11°, more than 36% below the Newell coupling‐function baseline evaluated on the same test crossings (1.75°). Additional validation holds out complete days, years, contiguous time blocks, and storm events. Feature attribution and cumulative history‐feature tests show that solar wind history over the preceding 60 min provides most of the prediction improvement, consistent with the role of history on hourly scales emphasized in previous studies. These statistical associations do not establish causality. The model provides a foundation for forecasting along‐track cusp boundaries using upstream solar wind observations; the next step is to test its forecasting performance with real‐time L1 data streams.