The Potential of Multispectral Vegetation Indices Feature Space for Quantitatively Estimating the Photosynthetic, Non-Photosynthetic Vegetation and Bare Soil Fractions in Northern China

Guoxiong Zheng,Liangliang Jiang,Zhihai Gao,Anming Bao,Cun Chang,Tao Chen,Xiaosong Li

doi:10.14358/pers.85.1.65

Abstract

Non-photosynthetic vegetation (NPV) is widely distributed in the arid and semi-arid area, especially in the sandy areas. The hyperspectral-based cellulose absorption index (CAI) is an accepted method for estimating the cover fractions of NPV. However, the spaceborne hyperspectral data currently available to us are very limited. In this study, we tried to identify one or more combinations based on the multispectral vegetation indices feature space model to quantitatively estimate the PV, NPV and bare soil fractions of the Otindag Sandy Land in northern China. Three frequently-used green vegetation indices, NDVI, EVI and OSAVI, and nine multispectral-based indices sensitive to NPV were used to examine the spatial patterns based on the field measured endmember spectra and non-growing and growing season Landsat-8 OLI image reflectance spectra. The capabilities of these different combinations were tested in this study area using mosaicked Landsat-8 OLI imagery. The results show that the feature space of different combinations based on the field measured spectra and image reflectance spectra has good consistency. The separability of feature space determines the availability of this model. The normalized difference senescent vegetation index (NDSVI) and brightness index (BI) were found to have greater potential to combine with the three selected green vegetation indices for simultaneous estimation of the fractional cover of PV, NPV, and bare soil in the Otindag Sandy Land because of their clear and separable feature space. We obtained the best and medium-precision estimates for NDVI-NDSVI (fPV: RMSE=0.26; fNPV: RMSE=0.17) and OSAVI-BI (fPV: RMSE=0.27; fNPV: RMSE=0.25) for 104 field observations.

Full Text