Prediction of Proton Pressure in the Outer Part of the Inner Magnetosphere Using Machine Learning

The information on plasma pressure in the outer part of the inner magnetosphere is important for simulations of the inner magnetosphere and a better understanding of its dynamics. Based on 17-year observations from both Cluster Ion Spectrometry and Research with Adaptive Particle Imaging Detector instruments onboard the Cluster mission, we used machine-learning-based models to predict proton plasma pressure at energies from similar to 40 eV to 4 MeV in the outer part of the inner magnetosphere (L* = 5-9). Proton pressure distributions are assumed to be isotropic. The location in the magnetosphere, the property of stably trapped particles, and parameters of solar, solar wind, and geomagnetic activity from the OMNI database are used as predictors. We trained several different machine-learning-based models and compared their performances with observations. The results demonstrate that the Extra-Trees Regressor has the best predicting performance. The Spearman correlation between the observations and predictions by the model is about 70%. The most important parameter for predicting proton pressure in our model is the L* value, which relates to the property of stably trapped particles. The most important predictor of solar and geomagnetic activity is F-10.7 index. Based on the observations and predictions by our model, we find that no matter under quiet or disturbed geomagnetic conditions, both the dusk-dawn asymmetry at the dayside with higher pressure at the duskside and the day-night asymmetry with higher pressure at the nightside occur. Our results have direct practical applications, for instance, inputs for simulations of the inner magnetosphere or the reconstruction of the 3-D magnetospheric electric current system based on the magnetostatic equilibrium.