Artificial neural network prediction model for geosynchronous electron fluxes: Dependence on satellite position and particle energy

Abstract

Geosynchronous satellites are often exposed to energetic electrons, the flux of which varies often to a large extent. Since the electrons can cause irreparable damage to the satellites, efforts to develop electron flux prediction models have long been made until recently. In this study, we adopt a neural network scheme to construct a prediction model for the geosynchronous electron flux in a wide energy range (40 keV to >2 MeV) and at a high time resolution (as based on 5 min resolution data). As the model inputs, we take the solar wind variables, geomagnetic indices, and geosynchronous electron fluxes themselves. We also take into account the magnetic local time (MLT) dependence of the geosynchronous electron fluxes. We use the electron data from two geosynchronous satellites, GOES 13 and 15, and apply the same neural network scheme separately to each of the GOES satellite data. We focus on the dependence of prediction capability on satellite's magnetic latitude and MLT as well as particle energy. Our model prediction works less efficiently for magnetic latitudes more away from the equator (thus for GOES 13 than for GOES 15) and for MLTs nearer to midnight than noon. The magnetic latitude dependence is most significant for an intermediate energy range (a few hundreds of keV), and the MLT dependence is largest for the lowest energy (40 keV). We interpret this based on degree of variance in the electron fluxes, which depends on magnetic latitude and MLT at geosynchronous orbit as well as particle energy. We demonstrate how substorms affect the flux variance.