Prediction of Tsunami Height on the Sunda Strait Coastal Area Based on Seafloor Morphology at the Earthquake Epicenter Using Spatial Linear Regression and Multi-target Regression
Abstract
The Sunda Arc has the potential for a megatsunami with devastating impacts along the west coast of Sumatra and southern Java. Machine learning models still face challenges in predicting tsunami wave heights along the Sunda Arc with high accuracy. Various factors, such as the diverse characteristics of earthquake locations and coastal conditions, make predicting tsunami heights difficult. Tsunami prediction is currently performed using various methods, such as numerical modeling, which incurs high computational costs, and data-driven machine learning, which is highly dependent on data availability. Machine learning research that has been conducted uses features from earthquake data without paying attention to the spatial characteristics of the earthquake location, so it does not achieve optimal accuracy. To obtain the characteristics of a tsunami area, spatial regression is one of the effective methods to map the earthquake potential based on spatial information. This research proposes a framework to predict the impact of tsunamis along the coastline. By combining spatial regression and Multi-target regression to predict tsunamis in the Sunda Arc area, this research proposes multi-target regression to predict tsunami height in 73 locations at once. Spatial linear regression serves as an intermediate spatial feature-extraction mechanism to represent seafloor morphological unevenness and spatial heterogeneity around earthquake epicenters before the multitarget tsunami-height prediction stage. The research is evaluated using R-squared and RMSE to determine the ability of the model to predict and produce the best R-squared value of 0.93196 and RMSE of 0.18188 m.