Estimating ecological indicators of karst rocky desertification by linear spectral unmixing method

•Spectral indices KRDSIs had significant linear correlation with coverage rates of soil and rock.•Reliable abundances were retrieved by partially constrained linear spectral unmixing (LSU) algorithm.•The abundances retrieved by partially constrained LSU algorithm could characterize the coverage rates of land covers.•Karst rocky desertification was mapped by the abundances of rock and vegetation.

Coverage rates of vegetation and exposed bedrock are two key indicators of karst rocky desertification. In this study, the abundances of vegetation and exposed rock were retrieved from a hyperspectral Hyperion image using linear spectral unmixing method. The results were verified using the spectral indices of karst rocky desertification (KRDSI) and an integrated LAI spectral index: modified chlorophyll absorption ratio index (MCARI2). The abundances showed significant linear correlations with KRDSI and MCARI2. The coefficients of determination (R2) were 0.93, 0.66, and 0.84 for vegetation, soil, and rock, respectively, indicating that the abundances of vegetation and bedrock can characterize their coverage rates to a certain extent. Finally, the abundances of vegetation and bedrock were graded and integrated to evaluate rocky desertification in a typical karst region. This study suggests that spectral unmixing algorithm and hyperspectral remote sensing imagery can be used to monitor and evaluate karst rocky desertification.