Estimating Vegetation Fraction Using Hyperspectral Pixel Unmixing Method: A Case Study of a Karst Area in China

Abstract

The rocky desertification is one of three major ecological problems in the karst areas in southwestern China. Vegetation fraction, bare soil, and bare rock are main typical surface characteristics obtained from remote sensing data when evaluating rocky desertification in these areas. How to estimate vegetation coverage more precisely is a challenging topic because the issues of complex surface coverage, highly spatial heterogeneity, and mixed-pixels should be addressed. Hyperspectral pixel unmixing is a better approach to solve these issues. In this paper, the Hyperion hyperspectral remotely sensed image is used as the source data, vegetation, soil, and rock are selected as three typical land cover features, and the pixel purity index (PPI) is utilized to distill the endmember spectral. Then, the pixel unmixing methods, including matched filtering (MF) and mixture tuned matched filtering (MTMF) are adopted to estimate vegetation coverage of the studied karst area, respectively. The results show that: 1) the maximum deviation between the ground-surveyed vegetation fraction and the MTMF-inversed one is acceptable, and so are the deterministic coefficient and the root mean square error (RMSE); 2) the MTMF-inversed results are more accurate than the ones inversed from the MF method and the MTMF-inversed vegetation coverage can be used to estimate the actual vegetation fraction. The results also demonstrate the applicability of the MTMF method in evaluating vegetation fraction in the karst regions.