Abstract
Optical Satellite imagery is very useful in mapping floods. However, optical sensors can operate during day time only and cannot capture images of impacted areas during cloud cover. As such, Synthetic Aperture Radar (SAR) data is very useful due to its unique ability to penetrate clouds and ability to capture images during day and night. The primary objective of this study is to obtain the flood inundation map by leveraging the potential of high-quality optical (Sentinel-2) and SAR (Sentinel-1) imagery using machine learning techniques. The dataset consists of four 10m spatial resolution Sentinel-2 optical bands integrated with dual-polarized (VH, VV) Sentinel-1 bands along with four Sentinel-2 spectral indices. Four Machine learning classifiers (Random Forest (RF), Rotation Forest (RoF), Support Vector Machine (SVM), and k Nearest Neighbors (KNN)) were used to obtain the flood inundation Maps. All the classifiers obtained good accuracy in the range of 88% to 91%. Rotation forest achieved the best performance among all classifiers with an overall accuracy of 91.2% and a kappa score of 0.881.

This publication has 29 references indexed in Scilit: