OTOMATISASI PEMILIHAN WAKTU TIBA GELOMBANG-P GEMPA MENGGUNAKAN MACHINE LEARNING

Abstract

The accuracy of earthquake wave first-break picking is important in seismic data processing. The first-break process is usually picked manually by visual inspection with or without STA/LTA algorithm. The principle of STA/LTA algorithm is the division between short moving average and longer time moving average capable of detecting sudden spikier signal with higher signal/noise ratio. However, both of these methods have some disadvantages such as longer time processing, lower picking accuracy and uncertainty to picker subjectivity. This study is trying to estimate P-wave arrival automatically using one of machine learning algorithm, Artificial Neural Network (ANN). Machine learning is one of the topics that has developed rapidly in recent years, but its application in seismology, especially in Indonesia, is still low. The purpose of this study is to create a Python based P-wave auto first-break picking program. The program is tested on synthetic wave and real earthquake data that obtained from Badan Meteorologi, Klimatologi, dan Geofisika (BMKG). The testing results show that the prediction of first break on synthetic and real data are pretty accurate. One of the factors that affect the accuracy of first break picking prediction is data filtering, because the predictions on filtered data has lower error percentage.