Enhancing Water Body Detection in Satellite Imagery: A Novel Hybrid Approach Exploiting Spectral and Reflectance Characteristics

Abstract

Water body mapping using satellite data is necessary for environmental assessment and proper allocation and utilization of water resources. Existing methods for extracting water from Sentinel-2 data using water indices often produce mixed, misleading results because of mixed pixels, spectral ambiguity, atmospheric changes, and optical distortions such as shadows and glare. In contrast, while the colour image is rich in hue detail, it lacks the spectral data needed to classify water. This work aims to develop a combined water fusion approach that uses RGB reflectance and Sentinel-2 bands, along with water indices such as NDWI and MNDWI, to produce multidimensional features for both spectral and spatial aspects. This approach leverages both spectral and RGB data, improving discrimination of water features and enhancing the reliability of water extraction under different conditions. To demonstrate the method's effectiveness, researchers tested it on Indian water bodies using Mean Squared Error (MSE), Peak Signal-to-Noise Ratio (PSNR), and Mean Absolute Error (MAE). These indicators assess result reliability across regions. This hybrid framework shows how remote sensing can support sustainable water resource management at scale.