Deep Learning Model for Identifying Ionospheric Electric Field Perturbations and Seismic Correlation

The quest for dependable earthquake-induced perturbations has driven investigations into the ionosphere. Recent advancements in satellite observations and machine learning offer novel opportunities to investigate these potential correlations via data-driven methodologies. This thesis examines ionospheric electric field disturbances through the application of deep learning methodologies to global space-based data. The research employs electric field measurements from the DEMETER satellite ICE mission, concentrating on very low frequency (VLF) electric field power spectral density data to examine ionospheric variability on a global scale. A deep learning framework using an LSTM autoencoder is developed to learn the underlying background behavior of the ionosphere and to detect anomalous perturbations in the data. This study utilised an adaptive rolling-window training strategy to address the highly non-stationary characteristics of geophysical datasets, continuously updating the model in response to fluctuating background conditions. This methodological framework integrates deep learning and statistical validation to systematically examine potential correlations between identified ionospheric anomalies and seismic events. The findings indicate that deep learning models can effectively extract significant ionospheric electric field anomalies from extensive satellite datasets and detect statistically significant associations with following earthquakes under the given framework. Additionally, the study attempts a feasibility study on a multi-instrument approach by integrating ground-based ionosonde observations to examine potential coupling between disturbances in the lower and upper ionosphere. This thesis develops an adaptable and extensible deep learning framework for the detection of ionospheric perturbations from global space-based observations and to study the association with earthquakes. The proposed methodology is intended to be flexible for integration with alternative datasets and signatures supporting ongoing research into lithosphere–ionosphere interactions and enhancing the advancement of data-driven approaches for investigating potential seismo-ionosphere coupling.