Land use land cover mapping and snow cover detection in Himalayan region using machine learning and multispectral Sentinel-2 satellite imagery

The Chamoli district of Uttarakhand, India, was recently devastated by a natural disaster i.e. flash flood (07 February 2021), causing significant damage to life, property, infrastructure, and the landscape. For effective policy making decisions, a detailed information about the various types of land cover presented in the region is required. This paper aims to (i) perform Land Use Land Cover (LULC) mapping, (ii) detection of snow cover in Himalayan region district Chamoli, Uttarakhand, India, using single-date Sentinel-2 imagery. Here, three Machine Learning approaches (Random Forest (RF), Support Vector Machine (SVM), and k-nearest Neighbor (KNN)) were employed to classify Sentinel-2 satellite imagery, which includes Near-infrared (NIR) and visible light bands (Blue, Green, and Red). Results demonstrated the successfully mapping of LULC classes, individual snow cover mapping and snow cover area estimation. The study achieved an overall classification accuracy of 91.01%, 89.67%, and 87.88% for SVM, RF and KNN, respectively. SVM reported the highest accuracy and an increase of + 1.31% and + 3.11% compared to RF and KNN, respectively. Snow, Forest and Sand class achieved higher accuracy, whereas Built-up and Water bodies obtained lower class-specific accuracy by all three classifiers. The analysis revealed that all classifier mapped snow cover effectively with F1-score 97.97%, 97.72% and 96.80% by SVM, RF and KNN respectively. Results indicated that estimated snow cover area is 2758 km², 2742 km², 2744 km² by SVM, RF and KNN respectively. Results of this study demonstrated the use of Sentinel-2 multispectral satellite data is a viable option for mapping snow cover.