Chlorophyll-a (Chl-a) is the main component of phytoplankton and an important index of water quality. Pearson correlation analysis is conducted on measured Chl-a concentration and band reflectance to determine the sensitive bands or multiband combinations of the Chl-a to input to a support vector machine (SVM) model. An indicator β is defined to evaluate the model performance of fitting and prediction. The model performs well with the lowest β (decision coefficient, (R2) = 0.774; root mean square error (RMSE) = 22.636 μg/L) of the validation set. The model test results prove that the model performs well. We analyze the impact factors of the model. The seasonal factor affects the model performance significantly; thus, samples from different seasons should be combined to train the model and inverse the water quality. Noise points reduce the model accuracy significantly; therefore, obvious outliers must be excluded at first. Additionally, the sampling method affects model accuracy, and systematic sampling in the descending order of Chl-a concentration is recommended. The combination of SVM algorithm and remote sensing technology provides a convenient, scientific, and real-time method to monitor and control water quality.
Keywords: Chlorophyll-a concentration; Donghu Lake; Lake eutrophication; Remote sensing; SVM model; Water resources monitoring.