The information used in water quality remote sensing is primarily concentrated in the low-value regions of imagery， which are highly sensitive to atmospheric absorption and scattering processes， making atmospheric correction a critical component. Although current mainstream atmospheric correction methods exhibit a certain level of general applicability， their inherent atmospheric models fail to efficiently reflect the actual atmospheric conditions and water vapor effects at the time of imaging， thereby limiting their accuracy. To achieve high-precision atmospheric correction for water， this study utilizes the radiative transfer mechanism and Sentinel-2 data， extracting clean water pixels from the imagery as atmospheric control points to retrieve imaging-time atmospheric parameters that account for water vapor effects. Comparisons with FLAASH and Sen2Cor demonstrate the effectiveness of the proposed approach. Specifically： ① The corrected water spectra obtained through this method show high consistency with in situ measurements， achieving correlation coefficients above 0.856 and root mean square errors below 0.017， with reflectance values close to the actual measurements. ② This method not only enables effective extraction of complex natural boundary water， with an extraction rate of 96.78% and a Kappa coefficient of 0.958， but also extracts small area water， with an extraction rate of 89.68% and a Kappa coefficient of 0.871. These results demonstrate that atmospheric correction of Sentinel-2 data using this method is better suited for water quality remote sensing.