Comparisons of autoregressive integrated moving average (ARIMA) and long short term memory (LSTM) network models for ionospheric anomalies detection: a study on Haiti (Mw = 7.0) earthquake

Abstract

Since ionospheric variability changes dramatically before the major earthquakes (EQ), the detection of ionospheric anomalies for EQ forecasting has been a hot topic for modern-day researchers for the last couple of decades. Therefore, there is a need to identify highly accurate, advance, and intelligent models to identify these anomalies. In the present study, we have discussed artificial intelligence techniques e.g. autoregressive integrated moving average (ARIMA), and long short-term memory (LSTM) network, to detect ionospheric anomalies using the total electron content (TEC) time series over the epicenter of Mw 7.0 Haiti EQ on January 12, 2010. We have considered 20 days of TEC data with a daily 2-h interval and trained the models with an accuracy of 1.28 and 0.07 TECU for ARIMA and LSTM, respectively. Both ARIMA and LSTM results showed that the negative anomalies are recorded 5 days before the EQ (January 7), while strong positive anomalies are recorded 1–2 days before the EQ (January 11–12) that are consistent with the findings of previous studies. Moreover, the quiet space weather conditions during the analyzed period indicate that the observed variations could be considered precursors to the impending Haiti EQ. Our analysis suggests that the performance of the LSTM model is more robust as compared to the ARIMA model in terms of detection of seismoionospheric anomalies.