Detection precursor of sumatra earthquake based on ionospheric total electron content anomalies using N-Model Articial Neural Network

Abstract

Indonesia is a country located between the Indo-Australian, Euresian and the Pacific plate. Based on these facts, earthquakes are frequent in Indonesia, especially in Sumatra. Therefore, an early detection of an earthquake, also known as an earthquake precursor, is required. At the moment, some research is exploring the earthquake relation with Total Electron Content located in ionospere. Machine learning methods and artificial intelligence are used to detect earthquake precursors. This study focuses on the N-ANN (N-Model Neural Network Model) method for detecting earthquake precursors. In addition, this study uses the Dst (Disturbance Storm Time) Index to subtract the effects of geomagnetic storms from TEC. TEC data uses TEC GIM (Global Ionospheric Maps) at 00:00. The observed earthquakes were the December 2004 to March 2005 earthquakes. The experiments show that N-ANN is more stable with the 5 model ANN, 3 hidden layer and 2 neurons. Earthquake precursors found 3 to 0 days before the earthquake occurred. The experimental results on 16 earthquake events reach 76% accuracy, 81% recall, and 93% precision. It can be concluded that N-ANN can be considered to detect earthquake precursors for early detection of earthquakes as a warning system.