Long Short Term Memory (LSTM) Network Models for Ionospheric Anomalies Detection: A Case Study for Mw=7.7 Awaran, Pakistan Earthquake

Abstract

Since ionospheric variability changes dramatically before the major earthquakes (EQs), the detection of ionospheric anomalies to EQ forecasts has become a new trend in the current era. Therefore, there is a call to identify highly accurate, advance, and intelligent models to identify these anomalies. In this study, we have proposed a deep learning-based method, long short-term memory (LSTM) network, to detect ionospheric anomalies using the Total Electron Content (TEC) time series of Awaran, Pakistan (Mw=7.7) EQ on September 24, 2013. We have taken 45 days of TEC data with a 2-h temporal resolution and train the models with an accuracy of 0.07 TECU. After fitted models with optimal hyperparameters, we have applied both to forecast TEC values for one week before the EQ. The anomalies, high differences (crossing the threshold value) between forecasted and observed TEC, are an indication of abnormal activities, e.g. earthquake, space weather, etc. In this study, we detected anomalies for the Awaran EQ. We conclude our results with the identification of ionospheric anomalies that occurred before the EQ results showed that strong positive anomalies are recorded 3 days before (on Sep 21) the EQ. These anomalies are thought to be related to Awaran EQ due to the quiet space weather conditions on the anomalies days. This study brings new insights into the AI techniques in seismoionospheric EQ forecasting.