Research On Remote Sensing Indicators Extraction Of Karst Rocky Desertification Using Multiple Endmember Spectral Unmixing

Karst environment has been a hot topic in contemporary geoscience research field, karst rocky desertification is one of the major ecological problems facing our country in southwest karst region. Karst regions are typically ecological fragile zones constrained by geological setting, with high complexity and heterogeity. In karst regions, rocky desertification is one of the most serious problems of land degradation. The bare rock ratio and vegetation coverage are essential assessing indicators of karst rocky desertification. With high subjectivity and low efficiency, the use of remote sensing was mostly focused on visual interpretation and computer-assisted digital processing of aerial photographs and satellite images. However, remote sensing methods cannot directly be exploited to extract the information of karst rocky desertification owing to the high complexity and heterogeneity of karst environment, mixed pixels generally exist. Remote sensing is difficult to directly provide the ecological indicators of assessing karst rocky desertification.Feature abundance generated by spectral unmixing can be characterized by the feature coverage of karst rocky desertification area. After the summarization of the remote sensing techniques of rocky desertification information extraction, this paper is aimed at the uncertainty of remote sensing information for rocky desertification and complex terrain in the karst area, topographic correction was processed for the images of this area. For the strong variability of endmember in karst rocky desertification area, MESMA was applied to the inversion feature typical karst area abundance. According to the mixed pixel imaging mechanism, based on the introduction of linear mixture model and the basic principle of MESMA, three kinds of endmember extraction algorithms was used to extract endmembers. The selected endmember extraction algorithms are respectively pure pixel index(PPI), sequential maximum angle convex cone(SMACC), vertex component analysis(VCA), the results showed the VCA is the best endmember extraction algorithm of the three. Due to the swath of Hyperion data was narrow, simulation of the wide hyperspectral data based on endmember extraction was adopted to obtain the wide hyperspectral data which covered the whole study area. The umixing results of the hyperion data showed significant linear relationship with spectral indices. The coefficients of determination R2 were 0.84 and 0.81 for vegetation and rock respectively, the results showed that the abundance can characterize the surface coverage. The unmixing results of multispectral data and simulation hyperspectral data which covered the entire study showed karst rocky desertification of the Qibainong town presented continuous distribution, the rural eastern region of rocky desertification is a lesser degree.