Detailed soil spatial information is essential for agricultural, environmental, and landuse planning applications and analyses. Existing soil attribute data is no longer sufficient to meet the needs of many of these needs since it does not explicitly describe the spatial variability of soil properties. Current statistical methods for predicting spatial variability in soil properties rely on the assumptions of linearity and stationarity; however, evidence suggests that soil properties may be related to terrain attributes in a nonlinear fashion and the assumption of stationarity is difficult to meet. Fuzzy membership values (FMVs) generated by the Soil Landscape Inference Model may be useful for predicting soil property values in areas where the relationship between soil property values and terrain attributes is perceived to be non-linear.

Predictive models based on existing soil survey data, terrain, and FMVs were developed for soil properties on gently rolling and steep transects in southwestern Dane County, Wisconsin. Models based on regression with topographic variables had R2 values 0.1-0.3 higher and mean absolute error (MAE) values 1.1-9.0 times lower than other models on the gently rolling transect. Models based on regression with fuzzy membership values had R2 values 0.1-0.8 higher and MAE values 1.5-17 times lower than other models on the steep transect. These results imply that when relationships between soil property values and terrain attributes are non-linear, inclusion of FMVs in soil property prediction equations may improve model results.