[Remote sensing retrieval of vegetation coverage in arid areas based on multiple endmember spectral unmixing]

Vegetation coverage is an important indicator in the assessment of terrestrial ecosystem and land desertification. By using the data acquired from the novel hyperspectral sensor HIS in Chinese HJ-1 small satellite, the suitable endmember spectrum was extracted by the combination of pixel purity index and endmember average root mean square error. Then, the vegetation coverage (FVC) in Shihezi area of Xinjiang, Northwest China was retrieved by the model of multiple endmember spectral mixture analysis (MESMA). With the comparison of the FVCs retrieved from the linear spectral analysis (LSMA) model and the measurement results, the FVCs retrieved from the MESMA model were evaluated. The results showed that the MESMA model enabled the use of different endmember combinations for different image pixels, and thus, could perform better than the LSMA model in the estimation of regional FVCs. As compared with the LSMA model, the correlation coefficient between the FVCs retrieved from the MESMA model and the measured FVCs increased from 0.766 to 0.838, while the root mean square error decreased from 0.375 to 0.196.