Date of Award

Fall 2025

Project Type

Dissertation

Program or Major

Physics

Degree Name

Doctor of Philosophy

First Advisor

Amy M Keesee

Second Advisor

Hyunju Connor

Third Advisor

Jeremiah Johnson

Abstract

The dynamic relationship between the solar wind and the Earth's magnetic field can produce rapid changes in the geomagnetic field measured on the ground ($dB_H/dt$). The fluctuations can be highly localized, with differences of up to 100s of nT/min in a small geographic area. The field of space weather is entering a more data-driven era, with enormous amounts of publicly available datasets from satellites and ground measurements. In this work modern machine learning techniques were employed to make forecasting risk-assessment predictions of large and localized $dB_H/dt$, and model explainability methods were employed to understand the non-linear connections being made. These results were analyzed as a function of spatial and temporal components.

This dissertation details an effort to understand the drivers of large and localized spikes in the ground magnetic field. Twenty-four viable regions were identified for study. Additionally, modern machine learning techniques were used to forecast extreme events, and model explainability methods were utilized to enhance our understanding of the magnetospheric system and its connection to the solar wind. Using these regions and methods, the correlation between a measure of localization (RSD) and $dB_H/dt$, $B_Z^{IMF}$, and $V_X$ were confirmed. Through the use of model explainability techniques it was shown that the inertia of the internal geomagnetic system plays a crucial role in large and localized spikes during the recovery phase of geomagnetic storms. This method uncovered spatial and temporal asymmetries in how solar wind parameters not as often associated with geomagnetic phenomena are treated by the models, indicating a larger role as drivers than previously thought.

Download

Share

COinS