Study On The Estimation Method Of Photosynthetic/Non-Photosynthetic Vegetation Coverage In The Hunshandake (Otindag) Sandy Land

Photosynthetic vegetation(PV) and Non-photosynthetic vegetation(NPV) as the two main kinds of vegetation cover forms in Hunshandake sandy land, play a essential role in intercepting rainfall, preventing desertification and consolidating soil, land surface material and energy cycles, soil nutrient retention and other aspects. Its have a very important significance for understanding desert vegetation degradation and recovery trend, carrying out ecological engineering management effectiveness evaluation estimating the fractional coverage of photosynthetic vegetation (CPV) and fractional coverage of non-photosynthetic vegetation (CNPV) timely and accurately. However, the spectral mixing mechanism analysis of PV/NPV and BS mixture is the basic prerequisite to achieve CPV and CNPV accurate estimation of Otindag desert.Therefore, we choose Hunshandake sandy land as the study area. Firstly, we constructed a typical PV/NPV and BS type endmember library covering destination via field survey. Then, on the basis of Hyperion hyperspectral data combined with ground investigation data, we tried using linear spectral mixture model (LSMM) and non-linear spectral mixture models (NSMMs) respectively to explore the spectral mixing mechanism of PV/NPV and BS mixture so that we could discover the optimal spectral mixture model for estimating CPV and CNPV of Hunshandake desert. Finally, we used two different endmember selection apporachs to calculate CPV and CNPV information of Zhenglan Banner which is a typical experiment district of Hunshandake sandy land draw support from GF1 and Landsat8 images data; meanwhile, we evaluated the estimation precision by means of ground truth data. The main conclusions are as follows:(1) The LSMM consists of PV/NPV and BS endmembers performs fairly well in Hunshandake sandy land, with a RMSE of 0.12 for CPV(R2=0.84) and a RMSE of 0.13 for CNPV (R2=0.66). The performance of NSMMs, which consider different multiple photon scattering effects scenarios, do not improve significantly whether in unmixing RMSE or estimation accuracy of CPV and CNPV. Moreover, non-linear mixing effects among different endmembers has little effect on the estimation accuracy of Cpv, but will result in a significant reduction of estimation accuracy on CNPV.(2) Different endmember selection method will have some impact for estimation results, variable endmember selection apporachs show better advantage relative to fixed endmember selection apporach in estimation accuracy. From the estimation results of Zhenglan Banner surface vegetation coverage based on GF1 and Landsat8 data we can see that the multiple endmember apectral mixture analysis (MESMA) and automaticed Monte Carlo unmixing used variable endmember selection method have a accuracy promotion compared with a full limited LSMM model which used a fixed endmember selection method.Among them, for GFl image data, the full constraint LSMM have a RMSE of 0.1163 for CPV(R2=0.72) and a RMSE of 0.1352 for CNPV (R2=0.71), the MESMA have a RMSE of 0.1002 for CPV (R2=0.83) and a RMSE of 0.0778 for CNPV (R2=0.74), the Auto MCU have a RMSE of 0.1066 for CPV(R2=0.80) and a RMSE of 0.0607 for CNPVv(R2=0.77); for Landsat8 image data, the full constraint LSMM have a RMSE of 0.1415 for CPV(R2=0.74) and a RMSE of 0.1378 for CNPV(R2=0.73), the MESMA have a RMSE of 0.1208 for CPV(R2=0.80) and a RMSE of 0.1173 for CNPV (R2=0.79), the Auto MCU have a RMSE of 0.1001 for GPV(R2=0.82) and a RMSE of 0.0629 for CNPV (R2=0.76).(3) Compared with Landsat8 data shows, GF1 data can get higher CPV and CNPV estimation accuracy and could be further used for larger area sparse vegetation parameters remote sensing quantitative inversion.