Surface water body extraction and Change Detection Analysis using Machine Learning Algorithms: A Case study of Vaigai Dam, India

Abstract

Surface water mapping is crucial to conserve and to plan water resources. The water body extraction and surface water extent estimation from the satellite images are challenging because the different land types have similar spectral responses. In this paper, the Machine Learning (ML) classifiers are trained to segment water bodies from satellite images. The features extracted through Convolutional Neural Network (CNN) and spectral indices methods are used for training. Gaussian Naive Bayes (GNB), Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), Adaptive Boosting (AdaBoost), eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), and Categorical Boosting (CatBoost) are the ML classifiers considered. Linear Imaging Self Scanning Sensor-III (LISS-III) images provided by the Resourcesat-2 satellite have been used for experimentation. The experimental results show that the RF and GNB are the best and least-performing ML classifiers for water body extraction. Additionally, the water extent of Vaigai dam is determined using the segmented maps. The surface water extent has good agreement with the rainfall and water capacity of the reservoir.