Monitoring the environment and managing water bodies are crucial for preserving ecosystems and ensuring sustainable resource utilization. This study aims to propose a robust approach for segmenting water bodies by combining various complementary attributes extracted from deep learning techniques. By leveraging deep contextual features learned from attention regions and enhanced edge detection, our method significantly improves the accuracy of detecting water bodies in low-resolution satellite imagery. Utilizing S1 and S2 Sentinel imagery data and exploring multi-band features, we enhance key attributes through advanced UNet architectures—specifically, Ki-UNet and Attention UNet—and integrate complementary features for downstream tasks. Additionally, we assembled a local dataset from the Khyber Pakhtunkhwa (KPK) region of Pakistan, facilitating precise water body segmentation from satellite or aerial views. Our ensemble model achieved remarkable accuracy and Intersection over Union (IoU) scores, reaching up to 99.01% and 96.2%, respectively, surpassing state-of-the-art models. This research provides automated, accurate segmentation techniques essential for environmental management and resource assessment, offering a promising solution for delineating water bodies. github link : (https://github.com/SaadBaloch96/Dataset/blob/main/README.md)