Soil-landscape modeling across a physiographic region: Topographic patterns and model transportability

Soil-landscape modeling techniques have been developed as a quantitative method to predict patterns of soil properties from observed patterns in soil-forming factors. However, transportability of these models to unsampled landscapes is unknown. Our objective was to develop quantitative soil-landscape models for multiple study sites and examine the similarity of these quantitative models, and therefore the similarity of soil-landscape relationships among areas with similar soils. We collected high-resolution digital elevation models (DEM) for six study sites across the Pennyroyal physiographic region of Kentucky, and for each study site used terrain attributes derived from the DEM to collect discrete soil samples using a stratified random sampling design for morphological, physical, and chemical characterization. For three of these sites we examined the inherent differences in terrain attributes among sites, and developed quantitative soil-landscape models that predict the spatial patterns in A-horizon depth, surface soil organic carbon content, and surface sand and silt content. The other three sites were used to test the transportability of these models. Terrain attribute distributions differ significantly among study sites, with regional terrain attributes (upslope contributing area, topographic wetness index) being more similar among fields than local terrain attributes (slope gradient, slope curvature). Predictive models explained from 28% to 67% of the variation in soil properties. The terrain attributes that best predicted soil variability were similar across all three fields used for model development, with slope gradient, elevation, slope curvature, and upslope contributing area appearing in most of the models. However, applying models from one field to other fields within the same physiographic region produced inconsistent results. In general, prediction quality decreased with distance from the site of model development. Further sampling, modeling, and validation at additional field sites are required to properly establish model transportability.