Sentinel 2A影像去云下的丘陵地区植被覆盖度反演

Retrieval of Vegetation Coverage Fraction in Hilly Areas with Sentinel 2A Image of Cloud Removal

光学遥感影像作为一种重要的数据源,被广泛应用于植被覆盖度反演,但南方丘陵地区常年多云雾,受云雾污染,像元质量下降。受丘陵地形的限制,地块破碎化严重,加之南方丘陵植被类型多,植被光谱变化较大,这些因素造成了植被覆盖度提取的准确度降低。为了提高丘陵地区植被覆盖度反演的精度,提出一种整合光学与SAR影像的反演框架。以Sentinel1A和Sentinel 2A影像作为遥感数据源,首先利用Sentinel 1A的光谱特征去除Sentinel 2A影像上的云像元,再使用光谱归一化减弱植被的光谱变化,在此基础上将扩展线性光谱混合模型用于反演植被覆盖度。结果表明:该框架能有效恢复云污染的像元,端元和影像归一化后,降低光谱异质性,反演丘陵地区植被覆盖度的精度较高,均方根误差为0.14,相关系数为0.95,接近地表植被覆盖的真实情况。

Optical remote sensing image was widely used as an important data source for the retrieval of vegetation coverage fraction.How ever,with heavy cloudy and fog in the southern hilly area,the quality of the pixels was degraded by cloud pollution.Due to the limitation of hilly terrain,the fragmentation was more serious.In addition,due to the large number of vegetation types in the southern hilly area and the large spectral variability in vegetation spectrum,these factors had reduced the accuracy of vegetation coverage fraction in hilly areas.In order to im prove the accuracy of vegetation coverage inversion in hilly areas,this paper proposed a retrieval framework for integrating optics and SAR imag es.Sentinel 1 A and Sentinel 2 A images used as remote sensing data sources,firstly,Sen2 cor atmospheric correction model was used to obtain sentinel 2 A images,with ground surface reflectivity and cloud mask products.Then cloud mask and sentinel 1 A images used to mark the non-cloud pixel locations on the sentinel 2 A image at a minimum distance from the cloud pixels,the non-cloud pixel was used to replace the spec tral curves of the cloud pixels with the smallest distance on the sentinel 2 A image to remove the effects of clouds and shadows.finally,in order to solve the spectral variability in the vegetation,the spectral normalization technique was adopted.On this basis,the extended linear spectral mixture model(ELMM)was used to obtain the unmixing coefficients.The experimental results showed that the framework could effectively recover cloud-contaminated pixels in sentinel 2 A image,when the endmembers and images were normalized,which reduced the spectral heter ogeneity in each class.The retrieval accuracy was highly accurate,and the root mean square error(RMSE)was 0.14 and the correlation coeffi cient was 0.95,which was close to the real situation of surface vegetation cover.