Analysis of spectral mixing mechanism of photosynthetic vegetation (PV)/non-photosynthetic vegetation (NPV) and bared soil (BS) mixture would be essential to establish the optimal spectral mixture model and further improve the estimation accuracy of sparse vegetation coverage in Hunshandake (Otindag) sandy land, Inner Mongolia of China. Over the past several decades, remote sensing has been widely utilized for estimating the fractional cover of vegetation. However, most efforts have been devoted to the estimation of fractional cover of photosynthetic vegetation (fpv) rather than fractional cover of non-photosynthetic vegetation (fnpv), although the latter is equally important, especially in desertified regions. Among of which, linear spectral mixture analysis was the most popular approach since its simplicity and operability, while the effects of the multiple scattering and the resulting nonlinear mixing problem is seldom concerned. Therefore, taking the Otindag sandy land as the study area, the mixed spectra were acquired over 47 sample plots through utilizing two Hyperion images, and the endmember(PV/NPV and BS) spectral and fraction were acquired through field investigation. Then, linear spectral mixture model(LSMM)and non-linear spectral mixture model (NSMM) including different multiple scattering combinations were adopted to decompose the mixed spectra to determine the optimal spectral mixture model based on the root mean square error (RMSE) of the unmixing and estimation accuracy of fpv and fnpv. The results show: (1) The LSMM consists of PV/NPV and BS endmembers performs fairly well in Hunshandake (Otindag) sandy land, with a RMSE of 0.12 for fpv(R2=0.84) and a RMSE of 0.13 for fnpv (R2=0.66); (2) The performance of NSMMs, which consider different multiple photon scattering effects scenarios, do not improve significantly whether in unmixing RMSE or estimation accuracy of fpv and fnpv. Moreover, non-linear mixing effects among different endmembers has little effect on the estimation accuracy of fpv, but will result in a significant reduction of estimation accuracy on fnpv.