A High-Resolution Data Fusion Framework for Water Body Assessment Using Sentinel-2 and Gradient Boosting

Abstract

Water pollution is becoming a very significant problem nowadays, and therefore, there is a great demand for efficient solutions. Water sampling is costly and timeconsuming, and it does not occur often enough to provide continuous control over water pollution. To address this issue, the proposed system uses Sentinel-2 imagery alongside ERA5-Land weather parameters for monitoring. In particular, the system calculates Normalised Difference Vegetation Index (NDVI) and Normalised Difference Water Index (NDWI) to assess vegetation and water presence. Next, the system uses the obtained data as input to the Machine Learning (ML) framework, specifically to build a CatBoost regressor that uses gradient boosting. It models complex dependencies among spectral bands and water-quality parameters, accurately predicting them with high $\mathbf{R}^{\mathbf{2}}$ values, including pH, dissolved oxygen, ammonia concentration, and turbidity. Moreover, in the proposed system, an additional component classifies water bodies based on their quality for fish breeding. Thus, less reliance on traditional water sampling and a more environmentally friendly approach were achieved.