Mapping of Waterlogged Areas and Silt-Affected Areas After the Flood Using the Random Forest Classifier on the Sentinel-2 Dataset

EasyChair Preprint 8527

Abstract

Flood Mapping is an important activity that helps in understanding the spatial extent of the flood over the impacted region thereby helping emergency responders in chalking out plans for future emergencies. The main of this study is mapping waterlogged areas and silt-affected areas after the submergence of floods. In this study, Random Forest (RF) classifier is used for map-ping waterlogged areas and silt affected areas using a pixel-based supervised classification approach. For the classification process, six land use/cover classes covering a total area of 1491.84 km2 of the Khagaria district of Bihar, India have been used. A four-band Sentinel-2 dataset at 10 m spatial resolution has been used for both pre-flood and post-flood datasets. The overall accuracy (OA) and Kappa score (K) for pre-flood classified data acquired using RF are (OA=84.95%, k=0.817). Whereas overall accuracy and kappa score for post-flood classified data using RF are (OA=83.325%, k=0.798) respectively. The results of post-flood classified data have shown that waterlogged areas and silt-affected areas have increased significantly from 22.40 km2, 7.22 km2 to 245.60 km2, 81.53 km2 respectively. Also, the classifier has shown fair Producer’s and User’s accuracy for the affected class that consists of Water-logged areas and Silt-affected areas. Further-more, quantitative analysis of post-flood classified data shows there is a significant increase in waterlogged areas and silt-affected areas.

Keyphrases: Flood mapping, Land use land cover, Random Forest, Sen2Cor, Sentinel-2, Waterlogged, remote sensing

@booklet{EasyChair:8527,
  author    = {Shivam Rawat and Rashmi Saini and Annapurna Singh},
  title     = {Mapping of Waterlogged Areas and Silt-Affected Areas After the Flood Using the Random Forest Classifier on the Sentinel-2 Dataset},
  howpublished = {EasyChair Preprint 8527},
  year      = {EasyChair, 2022}}