The estimation of chlorophyll-a concentrations (Chla) in lakes using remote sensing is convenient, but its use remains challenging in large eutrophic water bodies due to the great spatial and temporal variations of its optical properties. Combining the sampling location and date information with Chla data, this study divided the lake water into three types, I, II and III, and then built an optimal Chla estimation model for each type based on 11 datasets collected from 2004 to 2012 in Taihu Lake, China. The resultant model expression is Chla = exp (ax2 + bx + c), where x is R701/R677, (1/R686-1/R695) × R710 and (R690/R550-R675/R700) / (R690/R550 + R675/R700). For the Chla ranging from 2 to 192 mg/m3, the root-mean-square error (RMSE) of the new model decreased up to 5.1 mg/m3 compared to that of previous band combination models, such as band ratio, three-band and four-band models when directly validated. The RMSE of the re-parametrization model (the lowest RMSE < 12 mg/m3) is also lower than for those models (the lowest RMSE > 16 mg/m3), indicating that the Chla estimation model that considers the spatial and temporal variations has a better performance and validation accuracy and therefore is more effective for remote sensing monitoring of water quality.
Keywords: Case II water; Data partition; Estimation model; Hyperspectral remote sensing; Water quality.