Constructing a pixel unmixing model and estimating fractional vegetation cover for grassland using a separation index derived from visible spectra

The primary land cover types in desert grasslands are photosynthetic vegetation (PV), non-photosynthetic vegetation (NPV), and bare soil (BS). Although there have been some studies on the Fractional Cover of Photosynthetic Vegetation (fPV), Non-photosynthetic Vegetation (fNPV) and Bare soil (fBS) both domestically and internationally, with some achievements have been made, most of them focus on fPV,with relatively limited research on fNPV. NPV is the primary component of grassland cover during the non-growing season, significantly influencing grassland degradation and sandstorm disasters. Most NPV indices are established based on the spectral differences of NPV and BS in the short-wave infrared (SWIR), with very few applications in the visible spectrum (VIS). Multispectral images often lack a complete SWIR band, thus making it impossible to separate PV, NPV, and BS. NPV and BS exhibit highly similar spectral characteristics in the VIS band, with reflectance increasing almost linearly and displaying no obvious peaks or valleys. After repeated derivative calculations of the measured PV/NPV/BS endmember spectra using ASD TerraSpec Halo, it was found that near 550 nm, the BS derivative value peaked, the PV derivative value was minimal, and the NPV derivative value fell between them. Based on this spectral difference, five derivative indices and two original band reflectance indices were established. These seven indices were used to construct a triangular scatter plot with NDVI. In the DGSI-NDVI scatter plot, NPV was clearly clustered in the lower left corner; BS was almost clustered in the upper left corner; PV was distributed in the lower right corner. The distinct separation of these three feature types enabled the effective extraction of fPV, fNPV and fBS.