A machine learning approach to modelling the spatiotemporal evolution of emissions and electron fluxes in the inner magnetosphere

Abstract:

The machine learning methodology based on Nonlinear Autoregressive Moving Average exogenous (NARMAX) models is employed. The NARMAX algorithms automatically deduces a model from input-output data, which can then be used to model the emissions and electron fluxes. These models use solar wind and geomagnetic indices as inputs. It is difficult to apply machine learning techniques to model parameters of the inner magnetosphere as these parameters vary spatially as well as temporally. Moreover, machine learning techniques require large amounts of data to develop a model and the data from the inner magnetosphere is sparse, even with a plethora of missions to draw upon, such as GOES, Cluster, Themis and the Van Allen Probes. Two methods are trialed to model the spatial variability of the waves and electron fluxes. The first was to bin the data into spatial regions and develop individual models for each bin and the second method was to use spatial coordinates at the time of the measurements as the inputs to the model. The first method was used to develop electron flux models at geostationary orbit. The second method was applied to model the Hiss, chorus, and magnetosonic waves and also the electron fluxes throughout the radiation belts from 1.5 RE to 6 RE.