Global dynamic evolution of the cold plasma inferred with neural networks

The electron number density is a fundamental parameter of plasmas and a critical parameter in the wave-particle interactions. However, the distribution of cold plasma and its dynamic dependence on solar wind conditions remains poorly quantified. Existing empirical models provide us with statistical averages based on static geomagnetic parameters, but cannot reflect the dynamics of the highly structured and quickly varying plasmasphere environment, especially during times of high geomagnetic activity. Global imaging provides insights on the dynamics but does not provide quantitative estimates of number density. Accurately calculating the evolving distribution from first principles has also proven elusive due to the sheer number of physical processes involved.In this study, we propose an empirical model for reconstruction of global dynamics of the cold plasma density distribution based only on solar wind data and geomagnetic indices. We develop a neural network that is capable of globally reconstructing the dynamics of the cold plasma density distribution for L shells from 2 to 6 and all local times. First, we derive a plasma density database by using the NURD algorithm to identify the upper hybrid resonance band in plasma wave observations from Van Allen Probes [Zhelavskaya et al., 2016]. Then, we utilize the density database in conjunction with solar wind data and geomagnetic indices to train the neural network. To validate and test the model, we choose validation and test sets independently from the density database. We validate and test the neural network by measuring its performance on these sets and also by comparing the model predicted global evolution with global images of the He+ distribution in the Earth's plasmasphere from the IMAGE extreme ultraviolet (EUV) instrument.The present study demonstrates how we can reconstruct the global dynamics from local in-situ observations by using machine learning tools. We describe aspects of the validation process in detail and discuss the selected inputs to the model and their physical implication.