SM033-0004
Modeling radiation belt electrons with information theory and neural networks

Monday, 14 December 2020
Poster
Simon Wing, Johns Hopkins University, Baltimore, MD, United States, Aleksandr Y Ukhorskiy, JHU/APL, Laurel, United States, Thomas Sotirelis, Johns Hopkins Univ, Laurel, MD, United States, Jay Johnson, Andrews University, Department of Engineering, Berrien Springs, MI, United States, Drew L. Turner, Johns Hopkins University Applied Physics Laboratory, Laurel, MD, United States, Romina Nikoukar, Johns Hopkins Applied Physics Laboratory, Laurel, MD, United States and Giuseppe Romeo, Johns Hopkins University Applied Physics Lab, Abingdon, MD, United States
Abstract:
An empirical model of radiation belt electrons is developed using RBSP data 2013-2018. The model inputs the solar wind and magnetospheric parameters and outputs radiation belt electron phase space density (psd). The process of selecting input parameters is complex. Many solar wind and magnetospheric parameters are linearly and nonlinearly correlated or anticorrelated with one another, making it difficult to determine which parameters would carry relevant and which would carry redundant information. Information theory is used to determine the relevant input parameters and their response lag times. It is also used to determine the effect of solar wind parameters as a function of L*. The input parameters are ranked based on their information transfer to the radiation belt electrons. Using this ranking as a guide for selecting input parameters, the radiation belt electron model based on neural networks is developed. The preliminary result shows that the model predictive efficiency (PE) is ~0.66, which is comparable to those obtained by some previous models.
<< Previous Abstract | Next Abstract >>