A Neural Network Approach for Identifying Relativistic Electron Pitch Angle Distributions in Van Allen Probes Data

Abstract:

A myriad of physical phenomena occur in the inner magnetosphere, in particular at the Earth’s radiation belts, which can be a result of the combination of both internal and external processes. However, the connection between physical processes occurring deep within the magnetosphere and external interplanetary drivers it is not yet well understood. In this work we investigate whether a selected set of interplanetary structures affect the local time distribution of three different classes of high energy electron pitch angle distributions (PADs), namely normal, isotropic, and butterfly. We split this work into two parts: initially we focus on the methodology used which employs a Self-Organized Feature Map (SOFM) neural network for identifying different classes of electron PAD shapes in the Van Allen Probes’ Relativistic Electron Proton Telescope (REPT) data. The algorithm can categorize the input data into an arbitrary number of classes from which three of them appears the most: normal, isotropic and butterfly. Other classes which are related with these three also emerge and deserve to be addressed in detail in future works. We also discuss the uncertainties of the algorithm. Then, we move to the second part where we describe in details the criteria used for selecting the interplanetary events, and also try to investigate the relation between key parameters characterizing such interplanetary structures and the local time distributions of electron PAD shapes.