A Deep Learning Algorithm for Water Body Mapping from Sentinel 2 Data

Water is vital for the life on the Earth. Human health, food security, and economic growth are all water dependent. Monitoring of water bodies and their spatial changes is crucial for understanding the impact of human activities and climate changes on aquatic ecosystems. Remote sensing data provides large amounts of data that have been extensively used for monitoring water bodies, geometry, topology, associated attributes, and their changes. However, water body delineation is challenging due to sensor limitations, cloud presence, and atmospheric conditions. This paper presents a novel approach leveraging Convolutional Neural Networks (CNNs) for extracting water bodies from Sentinel-2 imagery. The efficacy of the proposed algorithm is rigorously evaluated across heterogeneous terrains encompassing diverse riverine and lacustrine features. Key metrics such as overall accuracy, F1 score, precision, and recall are employed to quantitatively assess algorithmic performance. Our findings underscore the promising potential of deep learning techniques in accurately delineating water bodies and monitoring their dynamic behaviour within intricate environmental contexts at local, regional, and global scales.