Improvement of NDVI mixture model for fractional vegetation cover estimation with consideration of shaded vegetation and soil components
Xihan Mu, Yang Yang, Hui Xu, Yuhan Guo, Yongkang Lai, Tim R. McVicar, Donghui Xie, Guangjian Yan
Abstract
The fraction of green vegetation is a widely-used indicator of vegetation abundance at regional and/or global scales. The pixel mixture model, especially the dimidiate pixel model (DPM, also referred to as two-endmember model) based on the normalized difference vegetation index (NDVI), plays an important role in the accurate estimation of fractional vegetation cover (FVC) via remote sensing. The two components in the traditional DPM are vegetation and soil (both sunlit and shaded). However, to date, the influence of shaded vegetation and shaded soil has not been fully considered in the NDVI-based DPM. Herein we analyze the necessity and feasibility of processing shaded components separately. The shaded soil was found to largely affect the canopy NDVI and can be combined with the vegetation (both sunlit and shaded) as one of the two components in DPM due to the high NDVI of shaded soil under a small percentage of diffuse sky radiation (< 10 % of the total hemispherical radiation in red band in this study). This finding partially explains why the canopy NDVI is oversensitive to background. The DPM was then improved with the solar and view angles to account for the fraction of shaded soil. We performed simulation and field measurements to validate the proposed models to varying factors including the vegetation structure, soil background, solar and view geometry, and slope gradient. The improved DPMs outperformed the traditional DPM ( i.e. , where no effect of shaded soil is considered) when estimating the NDVI and FVC of the mixed pixel. The FVC estimated with traditional DPM results in the RMSE from 0.14 to 0.31, and that with the improved DPMs range from 0.04 to 0.13. The decrease of uncertainty by using the improved DPMs was generally over 50 % when compared to the output from a traditional DPM. The proposed DPM maintains the advantage of an easy-of-use two-component mixture model yet is more accurate than traditional ones and thus expected to improve the FVC estimation from satellite data. • Shaded components are added in the NDVI-based dimidiate pixel model (DPM). • Shaded soil largely affects the canopy NDVI yet shaded vegetation doesn't. • Newly proposed DPMs are derived with solar and view angles to consider shadows. • New DPMs reduce 50 % uncertainty while maintaining ease of use as traditional DPM. • Partially overcome NDVI's oversensitivity to background and improve accuracy of FVC.