An Unsupervised Fuzzy Clustering Approach for the Digital Mapping of Soil Organic Carbon in a Montaneous Region of China

Spatial distribution of soil attributes is the basic information required for land surface process simulating and ecological modeling. Purposive sampling method based on typical points which employed environmental factors has been widely used in digital soil mapping (DSM) to acquire soil spatial properties at different scales. Clustering analysis of soil environmental covariates was performed to explore for sampling points representative of different grades of soil spatial distribution and to formulate a sampling designing method based on representativeness grade. This method was used to predict soil organic matter (SOM) content in the surface layer of grassland soil within a 4 km2 area of the Bayanbulak District, Xinjiang Uyghur Autonomous Region. Six terrain factors, including elevation, slope, aspect, planform curvature, profile curvature, and topographic wetness index, were clustered by fuzzy c-means method. Fuzzy membership distribution of 9 groups of environmental factors was derived to position 18 soil samples in the area with membership larger than 0.9. Then, SOM map was predicted with fuzzy membership model. Finally, 35 individual soil samples (16 regular sampling points, 9 cross-sectional sampling points, and 10 sampling points according to altitude) were collected as the verify point. The results showed that purposive sampling combined with FCM is a low cost and efficiency mapping method with satisfactory prediction precision and model stability and could be possibly applied to small-scale region with the similar landscape conditions.