Plasmaspheric dynamics studied using a three-dimensional machine learning based plasma density model in the inner magnetosphere

Abstract:

Plasmaspheric density and composition strongly influence wave growth and propagation, as well as energetic particle scattering. Previous statistical, empirical plasma density models of the inner magnetosphere have limited capability to make accurate predictions. Consequently, these models cannot be used to adequately quantify complex global processes and nonlinear responses to driving conditions, factors of critical importance during storms. Recent advancements in machine learning techniques have enabled a more dynamic study of the space environment. Here we present a three-dimensional dynamic electron density model based on an artificial neural network. This model uses a feedforward neural network which was generated using electron densities from satellite missions of CRRES, ISEE, IMAGE, POLAR, and Van Allen Probe. The three-dimensional electron density model takes spacecraft location and time series of solar wind and geomagnetic indices (flow speed, SYM-H, AL, and AE) obtained from NASA’s OMNI database as inputs. When compared with the out-of-sample data, the three-dimensional model predicts equatorial and field-aligned density profiles from satellite measurements with an error of less than 0.16. When the three-dimensional model is applied to a number of magnetic storms, successful reconstruction of the expected plasmaspheric dynamics, such as the plasmaspheric erosion, plume formation in three dimensions was achieved.