Tsunami detection using early-stage traveling ionospheric disturbances and machine learning based on GNSS measurements

Abstract

Natural hazards, such as earthquakes and tsunamis produce shock waves that travel into the ionosphere and produce traveling ionospheric disturbances (TIDs). TIDs should be detected in the case of earthquakes and tsunamis, as they can assist in monitoring natural hazards and provide early warning for tsunamis. A key challenge in detection is that TIDs also occur in the ionosphere as ordinary phenomena that are not caused by tsunamis. To address this issue, it must be noted that after a tsunami, TIDs are generated by two sources: acoustic waves from the earthquake shock, and gravity waves from tsunami propagation. Usually, tsunami TID detection is performed using gravity waves; however, gravity waves are weaker than acoustic waves and are more similar to ordinary ionospheric TIDs. Thus, to reduce false detections in near-field regions where tsunamis are strong and acoustic waves are present, this study focuses on early-stage TIDs—believed to be generated by acoustic waves—and incorporates their characteristics into a Levenberg–Marquardt backpropagation neural network for detection. GNSS time series of two giant tsunami events, the 2011 Tohoku and the 2004 Sumatra earthquakes and tsunamis, were considered. Because of the dense GNSS receiver coverage, the 2011 data were considered as the training and testing datasets, and the 2004 Sumatra dataset served as an independent validation case. The observations were classified into two categories: tsunami TIDs and non-tsunami TIDs. The test showed an accuracy of 99.8% in tsunami detection, and validation with observations from another region confirmed the model’s capacity to detect tsunami TIDs. This accuracy is higher than that of most previous detection systems and demonstrates its promising potential. Incorporating early-stage TID properties into a neural network enhances the practical reliability of tsunami detection.