A Machine Learning‐Based Model of Three‐Dimensional Ion Fluxes in the High‐Altitude Earth's Northern Cusp

Abstract Magnetic reconnection at the dayside magnetopause is the primary pathway for transferring mass, momentum, and energy from the solar wind into the terrestrial magnetosphere. Previous studies have shown that the spatiotemporal dynamics of dayside magnetic reconnection can be inferred remotely from properties of Earth's cusps, specifically their ion dispersion signatures, as well as structure and location within the magnetosphere. Despite the abundance of in situ cusp observations from multiple space‐based instruments, intermittent sampling spacecraft moving quickly along their trajectories precludes characterization of the cusp's global response to dynamic upstream drivers. To overcome this limitation, we leveraged existing data sets to develop a machine learning‐based model of three‐dimensional ion fluxes in the high‐altitude northern cusp that are modulated by solar‐wind forcing. This model is based on a residual neural network trained on solar wind parameters from NASA's OMNIWeb database, and ion flux measurements from the CIS/HIA instrument aboard ESA's Cluster mission. The model reproduces longitudinal and latitudinal displacements of the cusp under ‐ and ‐dominated interplanetary magnetic field regimes, in close agreement with reconnection theory. To quantify uncertainty and evaluate performance, we use an ensemble of 100 identical networks with randomized training and validation splits, reserving a fixed test set shared by all models. The statistics indicate consistent predictions within the central cusp region. Also, we used Shapley Additive exPlanations to visualize input feature contributions to the model's outputs. The analysis suggests that the magnitude of ion fluxes within the cusp is most strongly associated with , and solar wind parameters.