Accurate and robust estimation of TDS, TOC, Chl-a and surface water temperature using Landsat-8, Sentinel-2, MODIS,...

This study evaluates the accuracy and robustness of a statistical approach utilizing Box-Cox transformations combined with linear stepwise regression to estimate four water quality parameters, chlorophyll-a (Chl-a), total organic carbon (TOC), total dissolved solids (TDS), and surface water temperature (SWT), using data from four satellite sensors: Landsat-8, Sentinel-2, MODIS, and ASTER. The accuracy of this methodology was assessed using the coefficient of determination (R²) and Root Mean Square Error (RMSE). Results indicate that all four sensors produced accurate water quality models, each exhibiting high R² values (≥ 0.84). MODIS recorded the highest R² for Chl-a (0.99) and demonstrated good performance in estimating TOC and TDS. ASTER provided the most accurate estimates for TOC (R² = 0.9752, RMSE = 2.33) and SWT (R² = 0.9435). Landsat-8 also shows good performance for all water quality parameters, with maximum R² values reaching 0.9621. Although Sentinel-2 exhibited greater variability, a high R² for SWT (0.9271) was achieved. The methodology demonstrated robustness since it effectively worked across water quality parameters and sensors despite differing spatial and temporal resolutions. Consequently, this approach enhances its suitability for routine water quality monitoring in developing countries, where accurate water quality estimation across different sensors is especially valuable given the high costs of sampling and monitoring.