In order to improve the precision of phytoplankton chlorophyll-<i>a</i> (chl<i>a</i>) concentration retrieval, this study classified the
data into two groups (the high and the low) by chla concentration with the threshold of 50μg·L<sup>-1</sup>. And then build the
statistical models for each group. Particularly, a modifying factor OSS/TSS was used to unmixing the spectra in the low
model to improve the low relationship between spectral reflectance and chl<i>a</i> concentrations. As a result, the
concentration classification model allowed estimation of chl<i>a</i> with a root mean square error (RMSE) of 21.12μg·L<sup>-1</sup> and
the determination coefficient (R<sup>2</sup>) was 0.92, comparing with RMSE of chla estimation was 35.72μg·L<sup>-1</sup> and R<sup>2</sup>=0.72 in
the traditional model. It shows that concentration classification is a helpful method for accurate remote chla retrieval in
eutrophic inland waters.