Probabilistic Representation of Genetic Soil Horizons

Published soil survey reports typically describe soil series concepts in the form of aggregated information: ranges in soil properties, interpretations, and limitations that are derived from a collection of field-described soil profiles. While aggregated soil properties are readily estimated via standard statistical functions (mean, median, etc.), an aggregated representation of horizonation (e.g., genetic or functional horizon designation and depth) is typically difficult to construct. Variation in horizon designation use among soil scientists and different soil description systems, changes in horizon designation standards over time, variable depths at which horizons occur, and the uncertainties associated with these are all factors that complicate the process of delivering an aggregated representation of horizonation. In this chapter, we propose alternatives to the typical “representative profile,” e.g., the selection of a single soil profile to represent a collection. Two possible methods for aggregating a collection of soil profiles into synthetic profiles are presented, describing depth-wise probability functions for each horizon. Both methods rely on an expert-guided description of generalized horizon designation (e.g., a subset of horizon designation labels that convey a reasonable “morphologic story”) along with associated rules (regular expression patterns) used to correlate field-described to generalized horizon designation. The first method is based on (1-cm interval) slice-wise evaluation of generalized horizon designation; the second is based on a proportional-odds logistic regression model fit to depth-slices. These methods are demonstrated using USDA-NRCS soil survey data (USA).