高分四号卫星是我国首颗高空间分辨率地球静止卫星, 在浑浊二类水体的遥感定量监测方面应用潜力很大。 为评价高分四号多光谱数据经大气校正后水体反射率的精度, 以太湖为研究区, 使用同步MODIS数据辅助的Gordon单次散射改进算法, 对2016年7月21日和2016年8月17日两景高分四号多光谱数据进行大气校正, 并通过与地面同步实测光谱数据、 以及地球静止水色卫星GOCI数据大气校正结果的协同比对, 验证高分四号多光谱数据的大气校正效果, 为该卫星产品的水色遥感应用提供借鉴和参考。 结果表明, 红光B4波段校正精度最高, 平均绝对误差(MAPE)为10.71%； 绿光B3波段校正精度次高, MAPE为13.21%； 近红外B5波段校正精度次低, MAPE为33.06%； 蓝光B2波段校正精度最低, MAPE为53.55%。 其中B3, B4和B5波段校正精度高于GOCI, 主要原因在于高分四号的空间分辨率远高于GOCI, 混合像元导致的精度误差相对较小, 充分显示了高分四号作为一颗高空间分辨率地球静止卫星在水色遥感方面的优势； 而B2波段低于GOCI, 表明高分四号的蓝光波段尚有改进空间, 今后有必要对该波段进行重新定标等处理； 在未得到有效处理的情况下, 水色遥感应用应尽量避开该波段。 总体而言, 高分四号多光谱数据校正精度较高, 可以较好的应用于内陆二类浑浊水体的定量遥感监测。

GF-4 satellite is the first geostationary satellite with high spatial resolution in China, which has great potential in the aspects of inland water monitoring. This paper takes the Taihu Lake as study area to evaluate the accuracy of GF-4 derived spectral reflectance after atmospheric correction, aiming to provide valuable information for the application of GF-4 data to water color remote sensing. The MODIS-aided improved Gordon atmospheric correction algorithm was used to conduct atmospheric correction for GF-4 images obtained on 2016-07-21 and 2016-08-17. In situ measurements of quasi-synchronous samples and atmospheric corrected GOCI (Geostationary Ocean Color Imager) data were used to validate the GF-4 derived spectral reflectance. The results show that the red-light B4 band had the highest accuracy with RMSE (Root Mean Square Error) of 1.25×10-3 and MAPE (Mean Absolute Percentage Error) of 10.71%, the green-light B3 band had RMSE of 2.43×10-3 and MAPE of 13.21%, and the near-infrared B5 band had RMSE of 1.95×10-3 and MAPE of 33.06%. The blue-light B2 band had the worst accuracy with RMSE of 6.81×10-3 and MAPE of 53.55%. The accuracy of B3, B4 and B5 bands of GF-4 were higher than those of GOCI. This is because that the spatial resolution of GF-4 is much higher than that of GOCI, so the errors caused by mixed pixel were relatively smaller, showing the advantage of GF-4 as a high-resolution geostationary satellite when used in water color remote sensing. The accuracy of B2 band of GF-4 was lower than that of GOCI, indicating the B2 band of GF-4 still has room for improvement. In the future, the B2 band should be corrected separately, and the usage of this band in water color remote sensing should be avoided if it is not corrected properly. Overall, the multispectral data of GF-4 has relatively high accuracy and can used to monitor the inland case II turbidwater. In order to improve the accuracy of GF-4 derived spectral reflectance, several researches should be conducted in the future. First, aerosol observation stations should be established to obtained long-term data needed by the multiple scattering algorithm forlocalaerosol. Second, the study area should be divided into several sub-regions in order to reduce the errors caused by the assumption that the air condition was the same for the whole study area. The third, the spatial resolution of GF-4 was 50 m, while the effective area of thespectrograph probe was only 1 m2. Spectrum data should be collected at multiple points within one GF-4 pixel, so that the errors caused by scale mismatch can be reduced.