Abstract: Abstract: Topographical complexity of landscapes and diversity of terrain features have caused to interweave with the scattering mechanism and intensity of full polarization Synthetic Aperture Radar (SAR) data. This makes it difficult for the traditional Wishart-H/α (polarization entropy/scattering angle) classification to effectively classify land types in Karst areas. In this study, a complex Wishart distance measure was used for the class clustering of land type in research areas. A super-box clustering was carried out using the H/α plane. The semi-supervised classification was also carried out, according to the complex Wishart distance between the average coherence matrix of super-box and sample clustering. The obtained data of land classification were further processed using the Polarimetric-Total-Power (SPAN) and the Normalized Differential Vegetation Index (NDVI), where the SPAN was sensitive to buildings and bare rock land, whereas, the NDVI was sensitive to wood land, grass land and cultivated land. An effective classification of land types was finally realized in the Karst areas. Specifically, the complex Wishart distance between two types of sample was calculated to determine the similarity of scattering characteristics in the samples of different land types, and the clustering was also performed during this time. Eight kinds of super-box clustering were divided into using the H/α plane. Three kinds of preliminary clustering were obtained, including the first type of water body, the second type of construction and bare rock land, as well as the third type of wood land, grass land and cultivated land. The SPAN was then used to classify the building and bare rock land, using a threshold in the way of super-box segmentation. The wood land, the grass land, and the cultivated land were classified by introducing a combination of the NDVI and Digital Elevation Matrix (DEM). The improved method can be used to effectively classify the water body, woodland, grassland, farmland, construction land, and bare rock land, with the overall accuracy of 81.45%. To verify the improved method, the another full-polarization SAR data was selected from the typical Karst areas, where the terrain was relatively flat, while the topography was relatively single. This case study demonstrated that the land classification was successfully implemented, where the overall classification accuracy reached 85.66%. The finding can provide a novel way to accurately classify various land types, and thereby serve as an ideal supplementary means to monitor rocky desertification evolution in Karst areas.