Exploring AI Progress in GNSS Remote Sensing: A Deep Learning Based Framework for Real‐Time Detection of Earthquake and Tsunami Induced Ionospheric Perturbations

Abstract Global Navigation Satellite System Ionospheric Seismology investigates the ionospheric response to earthquakes and tsunamis. These events are known to generate Traveling Ionospheric Disturbances (TIDs) that can be detected through GNSS‐derived Total Electron Content (TEC) observations. Real‐time TID identification provides a method for tsunami detection, improving tsunami early warning systems (TEWS) by extending coverage to open‐ocean regions where buoy‐based warning systems are impractical. Scalable and automated TID detection is, hence, essential for TEWS augmentation. In this work, we present an innovative approach to perform automatic real‐time TID monitoring and detection, using deep learning insights. We utilize Gramian Angular Difference Fields (GADFs), a technique that transforms time‐series into images, in combination with Convolutional Neural Networks (CNNs), starting from VARION (Variometric Approach for Real‐time Ionosphere Observation) real‐time TEC estimates. We select four tsunamigenic earthquakes that occurred in the Pacific Ocean: the 2010 Maule earthquake, the 2011 Tohoku earthquake, the 2012 Haida‐Gwaii, the 2015 Illapel earthquake. The first three events are used for model training, whereas the out‐of‐sample validation is performed on the last one. The presented framework, being perfectly suitable for real‐time applications, achieves 91.7% of F1 score and 84.6% of recall, highlighting its potential. Our approach to improve false positive detection, based on the likelihood of a TID at each time step, ensures robust and high performance as the system scales up, integrating more data for model training. This research lays the foundation for incorporating deep learning into real‐time GNSS‐TEC analysis, offering a joint and substantial contribution to TEWS progression.