Soil Mapping Using GIS, Expert Knowledge, and Fuzzy Logic

bility arises mainly from the limitations of the discrete data model and from the polygon-based mapping pracA geographical information system (GIS) or expert knowledgetice employed in conventional soil surveys. based fuzzy soil inference scheme (soil-land inference model, SoLIM) is described. The scheme consists of three major components: (i) a Zhu (1997a,b), Zhu and Band (1994), Zhu et al. model employing a similarity representation of soils, (ii) a set of (1996), and Zhu et al. (1997) have proposed a SoLIM inference techniques for deriving the similarity representation, and to overcome the limitations in conventional soil surveys. (iii) use of the similarity representation. The similarity representation This approach combines the knowledge of local soil allows the soil landscape to be considered as a continuum, and thereby scientists with GIS techniques under fuzzy logic to map overcomes the generalization of soils in conventional soil mapping. soils. Although based on new technology, the model The set of inference techniques is based on the soil factor equation remains based on the soil factor equation of Dokuchaeiv and the soil–landscape model. The soil–landscape concept contends (Glinka, 1927) and Hilgard (Jenny, 1961) and the soil– that if one knows the relationships between each soil and its environlandscape model described by Hudson (1992). This soil– ment for an area, then one is able to infer what soil might be at each landscape concept contends that if one knows the relalocation on the landscape by assessing the environmental conditions at that point. Under the SoLIM, soil environmental conditions over tionships between each soil and its environment for an an area are characterized using GIS or remote sensing techniques. area, then one is able to infer what soil might be at each The relationships between soils and their formative environmental location on the landscape by assessing the environmenconditions are extracted from local soil experts or from field observatal conditions at that point. The SoLIM employs GIS tions using a set of artificial intelligence techniques. The characterized and remote sensing techniques to characterize the soil environmental conditions are then combined with the extracted relaenvironmental conditions and uses a set of knowledge tionships to derive a similarity representation of soils over an area. acquisition techniques to extract soil–environmental reIt is demonstrated through two case studies that the SoLIM for soil lationships from local soil experts or from field observasurvey has many advantages over the conventional soil survey aptions. A set of inference techniques constructed under proach. Soil information products derived through the SoLIM are of fuzzy logic links the characterized environmental condihigh quality in terms of both level of spatial detail and degree of attribute accuracy. In addition, the scheme shows promise for improvtions with the extracted relationships to infer the spatial ing the efficiency of soil survey and subsequent updates through reducdistribution of soils. This paper discusses how the SoLIM ing time and costs of conducting a survey. However, the degree of addresses the key challenges faced in conventional soil success of the SoLIM highly depends on the availability and quality survey, and assesses the potential of the SoLIM to imof environmental data, and the quality of knowledge on soil–environprove soil surveys. The limitations of conventional soil mental relationships over the study area. survey approach are first discussed to provide a context for the SoLIM, which is followed by an overview of the SoLIM. The assessment of the SoLIM for soil survey D soil spatial and attribute information is through two case studies is described in the third part required for many environmental modeling and of this paper. land management applications (Beven and Kirkby, 1979; Burrough, 1996; Corwin et al., 1997; and Jury, Model and Process Limitations 1985). Currently, conventional soil surveys are the major of Conventional Soil Surveys source of soil spatial information for these applications. Conventional soil survey is also based on the soil– However, standard soil surveys were not designed to landscape equation or concept (Hudson, 1992). To map provide the detailed (high-resolution) soil information the soils over an area, field soil mappers will first estabrequired by some environmental modeling (Band and lish the soil–landscape model over the area through field Moore, 1995; Zhu, 1999a) and crop management appliinvestigation. The soil–landscape model captures the cations (Peterson, 1991). The format and detail of relationships between the soils in the area and the differconventional soil maps are not compatible with other ent landscape units. The soil mappers then manually landscape data derived from detailed digital terrain map the spatial extents of different soils or combinations analyses and remote sensing techniques (Band and of soils through photo interpretation (Fig. 1). Moore, 1995; Zhu, 1997a; Zhu, 1999a). This incompatiThe ability of soil scientists to conduct soil surveys accurately and efficiently is largely limited by two major A.X. Zhu and James Burt, Department of Geography, University factors, the polygon-based mapping practice and the manof Wisconsin-Madison, 550 North Park Street, Madison, WI 53706; ual map production process. The polygon-based mapping Berman Hudson, Soil Survey Interpretations, Natural Resources Conpractice is based on the discrete conceptual model (Zhu, servation Service, 100 Centennial Mall North, Lincoln, NE 68508; 1997a), which limits soil scientists’ ability to produce Kenneth Lubich, NRCS–USDA, 6515 Watts Road, Suite 200, Madison, WI 53719; Duane Simonson, NRCS–USDA, 1850 Bohmann Drive, Suite C, Richland Center, WI 53581. Received 3 Jan. 2001. Abbreviations: ANN, artificial neural network; CBR, case-based rea*Corresponding author ([email protected]). soning; GIS, geographical information system; SoLIM, soil-land inference model. Published in Soil Sci. Soc. Am. J. 65:1463–1472 (2001).