International Journal on Advanced Science, Engineering and Information Technology, Vol. 12 (2022) No. 1, pages: 16-22, DOI:10.18517/ijaseit.12.1.14237

Tsunami Potential Prediction using Seismic Features and Artificial Neural Network for Tsunami Early Warning System

Astri Novianty, Carmadi Machbub, Sri Widiyantoro, Irwan Meilano, - Daryono


Tsunamis are categorized as geophysical disasters because tectonic earthquakes triggered most of their occurrences. The high number of deaths due to tsunami catastrophe has made many countries develop a tsunami early warning system (TEWS), especially countries prone to tectonic earthquakes. One of the crucial subsystems in a TEWS is the tsunami potential prediction subsystem. To provide an early warning of tsunami, the prediction must be carried out based on early observation of the seismic event before the tsunami. In this short time of computation, the calculation of seismic parameters can only produce some roughly estimated features. Hence, a proper inference method that can decide accurate predictions upon the features is urgently needed for the TEWS. Some existing TEWSs are using rule-based inference to decide the prediction and often overestimate the prediction of tsunami potential. This study tries to develop a tsunami-potential prediction system using the machine learning approach as its inference method. Seismic features extracted from P-wave seismic signals are used as input data for learning and classification using a backpropagation artificial neural network (ANN). The accuracy result is then validated by K-fold cross-validation. Our simulation results show that the utilization of backpropagation ANN has given better accuracy in tsunami prediction compared to one of the existing TEWS that does not use machine learning for its prediction. At least for some seismic events that occurred during 2010-2017, the proposed system results in fewer overestimated predictions than the existing TEWS referred.


Tsunami; prediction; backpropagation ANN; seismic; early warning system.

Viewed: 1396 times (since abstract online)

cite this paper     download