Flood monitoring and reservoir management in the transboundary Chenab River Basin using machine learning and remote sensing techniques

Flood mapping and upstream reservoir monitoring are crucial for the downstream countries. In August 2020, significant flooding occurred along the Chenab River in Pakistan. According to reports the flood resulted from an unwarned dam discharge by the upstream country India. This study utilizes Machine Learning (ML) and Deep Learning (DL) models, including the Random Forest Classifier (RFC), K-means clustering, change detection, and threshold to map flooding along the Chenab River. It also performs hypsometric analysis of the Salal Reservoir for flood forecasting using Synthetic Aperture Radar (SAR), Sentinel-1 (S1) and Digital Elevation Model (DEM). Hydrographic analyses investigate the pre- and post-flood scenarios of the upstream region for the flooding event. The threshold method produced the best results, with an overall accuracy (Kappa coefficient) of 98.50% (0.97). The K-means clustering method also performed well, with an overall accuracy (Kappa coefficient) of 96.50% (0.93). The pre-flood extent of the downstream river was 327.21 km², which increased to 833.49 km² during the flood, resulting in a total inundated area of 506.28 km². The intra-annual variation of the Salal reservoir’s analysis of the year 2020 revealed that the reservoir water levels peaked between 491 and 492 m, with a maximum area of 5.19 km² just before the spillways were opened and triggered subsequent flooding. These levels eventually decreased to 489–490 m, covering an area of 4.28 km² after the flood event. This study presents a unique approach for monitoring inaccessible reservoirs and concludes that SAR-based techniques are reliable and cost-effective for mapping flood extents.