A Study On The Methods Of Vegetation Coverage Remote Sensing Estimation In Desert Oasis Based On LSMM

When studying vegetation cover in arid oasis region, it is easily to meet the problem of miscarriage in image pixel classification because that there are large number of mixed image pixel in the medium resolution remote sensing image. Spectral mixture model simulate the spectral composition in image pixel, it can penetrate inside each pixel to unmix all of the composition elements. By this way, it can get the composition ratio of the various coverings corresponding earth surface in the pixel, and get the earth surface vegetation coverage in sub-pixel precision. Spectral mixture model is the ideal model in solving the image pixel mixing problem in regional scale. In this paper, we choose Landsat-8OLI image as the data source, obtain the true surface reflectance values of the earth cover materials in the study area by through the tasks of radiometric calibration, atmospheric correction, geometric correction and water mask, and combine with the field measured data which is plesiochronous with the image data. We constructed the four-endmember model, five-endmember model and six-endmember model in different methods, and carried out Linear Spectral Mixtrue Model to unmix the image data in study area using the three kinds of endmember model, obtained the vegetation coverage in the study area at different endmember model,combining with the measured data for validation. The main conclusions obtained are as follows.1. Linear spectral mixture model can get different expectation decomposition results when constructed under different constraints. In all kinds of LSMM, Fully-constrained LSMM have good physical meaning. It is suitable for vegetation coverage estimation study by remote sensing.2. The number of the endmember theoretically should not exceed the number of bands of image data values in linear spectral mixture model, but in actually using, the number of the endmember mainly depend on the number of the bands of the image data which is neither noise nor related bands after endmember minimum noise fraction transform (or principal component transform) because of the correlation between bands. The number of multi-spectral data band itself is limited, we should expand the spectral dimension of the original image to increase the spectral information, and extract a greater number of end-member.3. Four-endmember model was built based on the traditional theory of V-Ⅰ-S, it is a high inductive depict of the earth surface coverage. When building five-endmember model, we evaluated the vegetation type characteristics of the study area, choose a new endmember which can represent desert vegetation based on the original image spectral dimension expanding, and added the new endmember to participating model decomposition. When building six-endmember model, we focused on the measured vegetation information in study area, by spectral characteristics comparison and field demonstration, we added the five-endmember a new endmember which is mainly come from the measured spectral characteristics of vegetation.4. Under qualitative analysis and plot measured data validation, we believe that the final result of remote sensing estimation of the three kind of endmember model is in line with the distribution of vegetation cover characteristics of the study area as a whole. The result of estimation of five and six endmember model are more detailed to show the status of the distribution of a wide range of vegetation desertification, they are in line with the actual situation and be considered to have higher accuracy. Considering the result of regression analysis between measured values and model estimation, the five-endmmer model estimation has the most correlation with the field measured values, with lowest degree of dispersion scatterplot.5. In remote sensing estimation of vegetation coverage in arid zones based on linear spectral mixture model, the endmember selection should mainly based on image endmember, and take the measured spectral characteristics of vegetation as a reference. If the reference endmember was directly chosen to participate model decomposition, more error would be take in and thus the decomposition estimation accuracy would be reduced.6. By mapping analysis, we thought that the core oasis in Shule river piedmont region were concentrated in three county administrative center and the surrounding location, and the artificial oasis was the mainly oasis. Vegetation coverage in descending order of natural health vegetation was around the core area, sparse desert vegetation in the region was widely distributed, reflecting the characteristics of regional desertification. Gobi desert took the largest share. Overall, the piedmont region vegetation coverage spatial distribution was influenced by the local spatial distribution of water resources and use pattern, vegetation cover of most positions in this area was below30%, and the desertification degree was high.