Regression-Based Prediction of Surface Water Quality Using LISS-III Spectral Data

Accurate and cost-effective monitoring of river water quality is essential for sustainable water resource management, particularly in heavily polluted river stretches such as the Yamuna River at Mathura, Uttar Pradesh, India. Conventional field-based monitoring methods provide reliable measurements but are often limited in spatial coverage and require significant time and resources. This study explores the potential of satellite remote sensing combined with regression modeling techniques to predict key surface water quality parameters using IRS LISS-III spectral data. A total of 46 water samples were collected during the pre-monsoon season (March 2024) along a 12 km stretch of the Yamuna River between Devraha Baba Temple and Chintaharan Ghat. The collected samples were analyzed following APHA (2017) standards for parameters including Total Suspended Solids (TSS), Total Dissolved Solids (TDS), Turbidity, Chlorophyll-a, Biochemical Oxygen Demand (BOD), and Chemical Oxygen Demand (COD). These field observations were correlated with corresponding IRS LISS-III spectral bands (B2–B5) and derived spectral indices to develop predictive models. Three modeling approaches—Multiple Linear Regression (MLR), Stepwise Regression (SWR), and Support Vector Regression (SVR)—were applied to evaluate the relationship between spectral reflectance and water quality parameters. The MLR models demonstrated moderate predictive performance with coefficient of determination (R²) values ranging from 0.44 for TSS to 0.78 for Chlorophyll-a. The SWR models improved prediction efficiency by selecting significant spectral predictors and achieved R² values up to 0.79 for BOD and Chlorophyll-a. Among all methods, SVR showed the highest predictive capability, with R² values reaching 0.85 for COD and 0.82 for TDS, indicating its effectiveness in capturing nonlinear relationships between spectral signals and water chemistry. The results highlight that optically active parameters such as Chlorophyll-a and COD exhibit strong spectral responses in the visible and near-infrared regions. Overall, the study demonstrates that integrating remote sensing data with statistical and machine learning techniques provides a reliable and scalable framework for estimating surface water quality. This approach can significantly enhance spatial monitoring of river pollution and support environmental management and policy decision-making for sustainable river basin management.