Item

Abstract

A conceptual framework is proposed, to identify flood-affected locations that should be considered in order to lessen the consequences of naturally occurring disasters. Sentinel-1 data are used to evaluate the performance of automatic Otsu’s method and machine learning (ML) algorithms (Random Forest (RF), Support Vector Machine (SVM), CART, Minimum Distance (MD), K-nearest neighbour (KNN) and KD Tree KNN (KD-KNN)) to characterize the flooded region. The study provided a holistic spatial assessment of flood inundation in the region due to the impact of the extreme precipitation. The most adequate performance based on compound value is achieved by KNN, followed by SVM model and Otsu’s thresholding method. The validation site results reveal that Vertical transmit and Vertical received (VV) polarisation performs significantly better than Vertical transmit and Horizontal received (VH) polarisation. The flood extent with the highest accuracy was obtained using both Otsu’s thresholding method and MD, resulting in overall accuracies of 94.98% and 88.98%, respectively. These accurate flood extent maps were then utilized to estimate the potential number of individuals and buildings at risk in the study area, through the integration of population data from Gridded Population of the World Version 4 (GPWv4), Global ML Building Footprints by Microsoft, and building data from OpenStreetMap.