Detection of Suspended-Matter Concentrations in the Shallow Subtropical Lake Taihu, China, Using the SVR Model Based on DSFs

Accurate detection of suspended-matter concentrations in water columns is an important task in remotely sensing water color. This letter aims to identify an optimal model for estimating suspended-matter concentration in the optically complex Lake Taihu of China. Remote sensing reflectance R_rs(λ), inherent optical properties, and constituent concentrations of the Lake water were synchronously measured in November of 2007. After the effects of water constituents on R_rs(λ) were analyzed, the definitive spectral factors were determined, which are indicative primarily of total suspended matter (TSM). Several methods were compared in modeling the relationship between R_rs(λ) and TSM. Results show that the support vector regression (SVR) model performs best with a root-mean-square error of 4.7 mg · l^-1 (R² = 0.968). Its predictive errors in four seasons were also assessed with the mean absolute percentage errors varying in the range of 22.0%-60.0%. Thus, the SVR model can be used to reliably retrieve TSM concentrations in Lake Taihu. This finding offers new insights into the optical signals of in-water constituents in optically complex lakes.