Tsunami Threat Level Classification Based on Deep Learning: A Case Study in the Java Megathrust Region

The potential tsunami hazard due to the rupture of the megathrust plate in southern Java is real. The results of the study are in line with the call to strengthen Indonesia's existing tsunami early warning system, especially in the southern region of the island of Java. One important form of early warning that benefits the community is the tsunami threat level in the affected area. In this study, tsunami threat level information is generated based on data on the maximum height of tsunami waves. The most accepted method for providing tsunami hazard information is to perform a nonlinear tsunami simulation. However, this method requires a large amount of time and computational load, while the period for the tsunami to reach the coast is relatively short. Therefore, this study proposes a Convolutional Neural Network (CNN) approach to produce a short classification of tsunami hazard levels. Our case study uses the Java megathrust earthquake hypothesis (Mw 6.0 - 8.8) with Pacitan Bay, Perigi Bay, and Parangtritis Beach as observation points. The results show that our proposed method has good performance with a validation accuracy of 91.16 and a test accuracy of 87.10. © 2021 IEEE.