Tsunami Vulnerability and Risk Assessment Using Machine Learning and Landsat 8

Abstract

Tsunami is a disaster that often occurs in Indonesia, and there are no valid indicators to assess and monitor coastal areas based on functional land use and based on the land cover, which refers to the biophysical characteristics of the earth’s surface. One of the recommended methods is the vegetation index. The vegetation index is a method from LULC that can provide information on how severe the tsunami’s impact was on the area. In this study, an increase in the vegetation index was carried out using machine learning. This study aimed to develop a tsunami vulnerability assessment model using the Vegetation Index extracted from Landsat 8 satellite imagery optimized with KNN, Random Forest, and SVM. The stages of the study were 1) extraction of Landsat 8 images using algorithms NDVI, NDBI, NDWI, MSAVI, and MNDWI; 2) prediction of vegetation indices using KNN, Random Forest, and SVM algorithms. 3) accuracy testing using the MSE, RMSE, and MAE,4) spatial prediction using the Kriging function, and 5) tsunami modeling vulnerability indicators. The results of this study indicated that the NDVI interpolation value is 0 - 0.1, defined as vegetation density, biomass growth, and moderate to low vegetation health. The NDWI value was 0.02 - 0.08, and the MNDWI value was 0.02 - 0.09, interpreted as surface water along the coast. MSAVI is a value of 0.1 0, defined as vegetation’s absence. The NDBI interpolation value was -0.05 - (-0.08), interpreted as the existence of built-up land with social and economic activities. From the research results on the ten areas studied, there were three areas with conditions with a high level of tsunami vulnerability, two areas with medium vulnerability, and five areas with low vulnerability to tsunamis.