Science Seminars

Study of plasmaspheric dynamics using data-driven empirical models: a neural network approach

Speaker: Xiangning Chu (UCLA)
Date: Thursday, Oct 19, 2017
Time: 4:00 PM
Location: SPSC W120

Seminar Abstract:

The plasmasphere is a region of cold dense plasma in the inner magnetosphere, extending from Earth’s upper atmosphere to the plasmapause. It is constantly in a dynamic state, with erosion and refilling occurring during geomagnetic storms. The plasmaspheric dynamics are important in understanding radiation belt physics because plasmaspheric density strongly influences energetic particle scattering, as well as plasma wave excitation and propagation. Previous empirical density models can provide statistically averaged density profiles that, however, do not resolve the dynamic evolution of the plasma density.
In this presentation, I will introduce the data-driven dynamic empirical models of the plasma density recently developed using a neural network approach. By taking time series of solar and geomagnetic indices as input, instead of using instantaneous values, these models are not only time-dependent but also history-dependent. They have good predictive abilities on both training database and out-of-sample data with a correlation coefficient of 0.95. Given their good predictive performance, they could provide unprecedented opportunities to gain insight into the plasmaspheric dynamics at any time and location. As an example, we will show that these models succeeded in reproducing various dynamic plasmaspheric features such as the erosion and recovery of the plasmasphere, as well as the plume formation. We will also show the dependence of the plasmasphere on geomagnetic activity such as magnetic storms, substorms, and enhanced convection and demonstrate the ability to reproduce plasmaspheric refilling with the model. Finally, we will show how these models could be used to discover new unexpected phenomena that are difficult to find otherwise.