Optimal TEC Forecast Models Based on Machine Learning and Time Series Analysis Techniques: A Preliminary Study on the Ring of Fire

Abstract

Geomagnetic storms are one of the major factors causing Total Electron Content (TEC) anomalies. Analyses of TEC fluctuations also provide a valuable understanding of the mechanisms of earthquakes and tsunamis. However, there is no clear consistency in investigations of TEC disturbances that should be considered simultaneously in both solar and seismic activities. Therefore, based on Machine Learning (ML) and time series analysis techniques, we build TEC forecast models to study relationships among ionospheric anomalies, geomagnetic storms, and earthquakes. Robust statistical tests are used to select the optimal models and estimate forecast performance. Depending on the quality of input data and sampling rates, the forecast performance can get from ~2.0 to ~2.5 TECU for 3-day predictions using daily time series and reach up to ~1.3 TECU using one-minute time series. These models present significant relationships between the ionosphere, solar activity, and seismic events, which can be applied to hazard warning systems.