Retrieval Of Green And Non-green Fractional Vegetation Cover In Desert Steppe Based On NDVI-DFI Linear Spectral Mixture Analysis Model

Estimation of Fractional Vegetation Cover of Green Components,f _GVand Fractional Vegetation Cover of Non-green Components,f _NGV are important for assessing the growth conditions of vegetation in arid areas and monitoring environmental change and desertification.This study took the desert steppe in the middle of Siziwang Banner of Inner Mongolia Autonomous Region as the research area,to estimate the f _GV and f _NGV by using NDVI-DFI Linear Spectral Mixture Analysis Model based on Landsat-8 OLI and Sentinel-2 MSI images.The endmember matrixes of spectral indices of NDVI and DFI were calculated for green vegetation,non-green vegetation and bare soil at different regional scales as the parameters of NDVI-DFI Linear Mixture Analysis Model,in order to obtain its more accurate values.And the accuracy assessment for final results of estimated fractional vegetation covers had been completed by using field based ground truth data set.The main research results and relevant conclusions are as follows:(1)For the Sentinel-2 MSI based FVC estimation,the f _GV and f _NGVgot their best accuracies(R~2=0.78,RMSE=4.86%,MAE=4.23%;R~2=0.39,RMSE=8.20%,MAE=7.30%)when the endmember eigenvalues were determined based on the 4 data tiles(including the study area and surrounding area)mosaicking scale respectively.(2)For the Landsat-8 OLI based FVC estimation,the f _GV and f _NGV got their best accuracies(R~2=0.45,RMSE=9.57%,MAE=6.81%;R~2=0.39,RMSE=9.42%,MAE=7.90%)when the endmember eigenvalues were determined based on the cross-ecotype scale and scene scale respectively.(3)For the Landsat-8 OLI based FVC estimation,the descending order by different scales for estimated accuracies(by RSME)of the f _GV and f _NGV were as follows respectively:9.57%at cross-ecotype scale,18.51%at study area scale and 25.83%at image scene scale,and 9.42%at the image scene scale,10.92%at the study area scale and 16.79%at the cross-ecotype scale.For the Sentinel-2 MSI based FVC estimation,the descending order by different scales for estimated accuracies(by RMSE)of the f _GV and f _NGV were as follows respectively:4.86%at the 4 data tiles mosaicking scale,5.72%at the study area sclae and 14.52%at the cross-ecotype scale,and 8.20%at 4 data tiles mosaicking scale,17.45%at cross-ecotype scale,24.22%at study area scale.(4)On the whole,Sentinel-2 based FVC accuracies were better than Landsat-8 OLI based corresponding accuracies for both of green and non-green vegetation.