Improving Soil Maps with a Semantically Calibrated, Digital Model for Delineating Hillslope Position.

Attempts to digitally automate soil classification have only been partially successful and causes of misclassifications have been unclear, perhaps because of the large number of parameters used in the models. Digital soil classification models often do not carefully consider how the individual digital attributes correlate to the components of the mental soil models used by soil mappers. To better understand the relationship between digital attributes and the factors driving soil mappers' delineation decision process, and hence, the end product (the soil classification map), further study is needed that focuses on simpler combinations of digital attributes and specific components of mental soil models. We use semantic modeling to correlate sections of a hillslope from discipline-specific, semantic meanings to the appropriate scale and geometric thresholds of digital terrain calculations. In order to perform semantic modeling for soil mapping, we employed a simple, decision-tree model relating hillslope position points observed by NRCS soil scientists with profile curvature, elevation, and slope. This model is calibrated to, and tested against, the scale which experienced soil mappers use in evaluating hillslope position. The calibration of the digital terrain calculation scale was performed by varying the grid resolution and analysis neighborhood size for profile curvature. This simplified approach also allowed for the evaluation of uncertainty’s effects on the results. Because all of the parameters of this model are based on a DEM, different error realizations for the DEMs could be assessed for impact on the model’s success rate. The results are useful as delineation guides for current soil survey procedures and as a building block for digital soil classification models.