Extraction Of Quantitative Relationships Between Soil Properties And Landscape Factors Based On Fuzzy Set Theory And Its Applications In Digital Soil Mapping

Detailed soil spatial and attribute information is required by many environmental modeling and land management applications. The accuracy and efficiency of conventional soil surveys, based on polygon mapping practice and the manual map production process, are low. Soil is a continuum both in spatial (geographic) domain and in attribute (property) domain, but traditional methods in soil classification and soil mapping take a Boolean approach and treat soil as distinct and discrete entities, rather than as a continuum. There are two limitations associated with soil mapping using the Boolean approach: the class assignment generalization and spatial generalization. So an appropriate way should be found for soil survey and soil mapping.With the recent development of Geographical Information System (GIS), digital terrain modeling, Remote Sensing (RS) technology, detailed predictive soil mapping becomes available. Recalling Jenny's (1941) famous equation, which he intended as a mechanistic model for soil development, S=f(Cl,O,R,P,T,...), implicitly, S stand for soil, cl represents climate, o organisms including humans, r relief,p parent material and t time. The function implies that unique environmental configurations reflect the unique status or properties of soil, and if we got the relation of soil and landscape we can infer the soil information by the corresponding landscape units.A case study in southern Anhui Province, subtropical China, was conducted to extract the relationships of terrain factors and soil properties by clustering the terrain factors based on fuzzy set theory. The similarity of each pixel to the typical ones was obtained from the classification results, the degrees of similarity are referred to as fuzzy membership. Then, the map of continuous soil horizon depth was drawn by means of ARC/info software after sampling at the area with high value of fuzzy membership and populating the similarity model using the linear and additive weighting function.Compared the predicted value set from the map to the independently collected field sample set, the derived soil map achieved an accuracy of 82%. Further investigation showed that the model works well in the area of low altitude, well-developed soil, thick soil horizon, stable surrounding for soil formation.The new approach can improve the efficiency and accuracy of soil mapping, and can show variance more clearly as well as hold more soil information than traditional soil map. It thus should be a feasible new method of soil survey and mapping.