Estimating Saturated Hydraulic Conductivity Using Decision Tree Analysis.

Pedotransfer functions (PTFs) are useful tools that can help predict saturated hydraulic conductivity (Ksat). However, at present, most PTFs only utilize soil texture and bulk density. There are numerous other properties, including soil structure, which are available in soil survey databases that may be useful for more accurately predicting Ksat_.  In our study we want to determine which of these, if any, will give a more accurate prediction of Ksat.  We have used soil profile descriptions from the S-124 regional project dataset.  This is a Southern Cooperative Bulletin Series which contains 21 soil series descriptions from all over the Southeastern United States. These bulletins contain qualitative soil structure descriptions that we will quantify to determine the most important properties in the prediction of Ksat.  We have broken down the descriptors into four qualitative and two quantitative groups: horizon notation, texture class, ped size, crack orientation, bulk density, and particle size distribution.  Qualitative type variables are represented by zeros and ones, determined by the samples’ membership in the respective groups. Preliminary runs show reasonable estimations for Ksat using the decision-tree model developed for a European soil data-base. However, the sand texture class prediction was lower than expected with a value of 13.21 cm/day.  The overall root mean squared residual (RMSR) was .3911, thus indicating a good prediction of Ksat when compared to values measured in the field.