Probabilistic Approach to the Identification of Input Variables to Estimate Hydraulic Conductivity

Soil hydrologic data are required for catchment‐scale modeling but these data are often difficult and costly to obtain. Although pedotransfer functions (PTFs) have been used to generate these data, they are not easily transferable to other bioclimatic zones. As climate influences the development of soil structure, the incorporation of soil structure assessments may improve the effectiveness of pedotransfer functions. The objective of this study was to examine which types of categorical texture and structure data would be most useful in either improving current PTFs to estimate saturated hydraulic conductivity ( K s ) or allowing PTFs to be developed in areas where measured particle‐size distribution, organic matter (OM) content, and bulk density ( D b ) are lacking. As soil structure is categorical data, regression trees were used to determine which input data derived from the HYPRES database would be most useful in deriving new PTFs. Jackknife cross‐validation was used to generate randomized subsets of the data and the optimal size of the developmental ( n = 411) and test ( n = 91) data sets was derived experimentally. The relative importance of input variables was evaluated by considering the probability that the data were partitioned by each variable. The best model utilized field‐based information on soil horizon, soil structure (ped size), and soil textural class and, although the accuracy was no better than existing continuous PTFs, it has the added benefit of utility in data‐poor environments.