Sentinel-2 Retrieval of Chlorophyll-a and Total Suspended Solids in Optically Complex Inland Waters: A Case Study of Lake Palic

This study focuses on the development of satellite-image-based models for estimating Chl-a and TSS concentrations in Lake Palic. Using Sentinel-2 images, we ensured high spatial resolution. Image processing was conducted using the SNAP (Sentinel Application Platform version 10.0.0, developed by the European Space Agency (ESA), Frascati, Italy) software with the C2RCC algorithm. A series of environmental indices was calculated and correlated with measured Chl-a and TSS values to support the development of a reliable predictive model. Two modeling approaches were applied: simple linear regression based on selected spectral indices and measured Chl-a and TSS concentrations, and machine learning methods to improve predictive performance and capture potentially nonlinear relationships. The models were evaluated using standard metrics and verified against field measurements. The overall process was implemented at two locations within the lake, the middle and the outflow. Spatial variability in optical conditions prevented the development of a single unified model. While Chl-a concentrations could be estimated with reasonable accuracy, TSS retrieval remained limited by optical complexity and spatial variability in water quality conditions. This suggests that the lake exhibits spatially variable optical regimes, in which different optically active constituents dominate the spectral response, thereby directly influencing model performance.