A Unified Model for Remotely Estimating Chlorophyll a in Lake Taihu, China, Based on SVM and In Situ Hyperspectral Data

Accurate estimation of chlorophyll a (Chla) in inland turbid lake waters by means of remote sensing is a challenging task due to their optical complexity. In order to explore the best solution, we observed water quality parameters and measured water reflectance spectra in Lake Taihu for 14 days in November 2007. After initial wavelength analysis and iterative optimization, the best three-wavelength factor (TWF) was determined as [R _rs ^-1(661) - R _rs ^-1(691)] R _rs(727). Linear models and a support vector machine (SVM) model with TWFs as the inputs were established for retrieving Chla concentration level. It is found that linear models with a single TWF performed worse than the SVM model. The SVM model is highly accurate, whose R ² and root-mean-square error are 0.8961 and 2.67 mg/m³, respectively. Validation of the SVM model using data sets obtained at another sampling time reveals small errors. Thus, this model can be used to extract Chla concentration levels in Lake Taihu waters but is not restricted by the sampling time. These findings underline the rationale of the TWF model and demonstrate the robustness of the SVM algorithm for remotely estimating Chla in Lake Taihu waters.