Automated soil inference under fuzzy logic

Soil information is essential to any terrestrial ecological modelling and management activity. Polygon soil maps produced from soil surveys are currently the major source of information on the spatial distribution of soil properties for a variety of land analysis and management activity. However, there are some major problems regarding the use of current soil maps in geographic analysis and especially in geographic information systems (GIS). These problems include limited coverage at a fixed scale, locational errors, attribute errors, and insufficient information in the mapping units. Much of these problems are due to the crisp logic and cartographic techniques with which soil maps are produced. Under crisp logic standardly used in soil classification and mapping, an area belongs to one and only one soil mapping unit, and is separated from other mapping units by sharp boundary lines. However, soil in a landscape is a continuum and the discretization of such a continuum into distinct spatial and categorical groups results in a significant loss of information. This paper presents a methodology to infer and represent information on the spatial distribution of soil. The methodology combines fuzzy logic with GIS and expert system development techniques to infer soil series from environmental conditions. The methodology for every point in an area produces a soil similarity vector (SSV) showing the similarity of the soil at the point to a prescribed set of soil series. The SSV produced from this methodology can be used to infer local soil properties at values intermediate to the typical or central values assigned to each possible series. Preliminary results from the methodology using a limited set of environmental variables for an experimental watershed in Montana are presented.