Chlorophyll-a(Chl.a) concentration is an important indicator of water quality evaluation, which is affected by the seasonal changes of suspended matter concentration. How to weaken the spectral interferences of suspended matter is an important issue in the retrieval of Chl.a concentrations with remote sensing in inland water. Based on six field surveys in 2011-2013 in Guishui River, the inversion ability of Chl.a concentrations with three-band model and support vector regression model were evaluated, in which baseline correction and first order differential method were used to weaken the spectral noises. Reflectance of 750 nm was defined as a baseline, a connecting line of reflectance from 500 nm to 750 nm was defined as another baseline. Baseline correction was defined as spectral reflectance minus baseline values. The validation was performed using data of samples in July 2013, the results demonstrated that SVR model was better than the three-band model in reversion of the Chl.a concentration in seasonal turbid water. Band reflectance combinations after baseline correction screening as the input variables can improve the inversion accuracy of the SVR estimation model. The determination coefficient was 0.68 and the root mean square error was 3.38 μg/L. Linear baseline correction can slightly improve the inversion accuracy of the three-band model.