Estimating saturated hydraulic conductivity for southeastern soils using decision tree analysis

Pedotransfer functions (PTFs) are used to predict saturated hydraulic conductivity (Ks) from more easily measured soil properties. Our objective was to determine if soil morphology was an important factor in predicting Ks using PTFs. We used soil profile descriptions for nine soils from the S-124 regional project dataset describing soils of the southeastern United States. Our best decision-tree model predicted log10 Ks (cm day-1) with an average log10 root mean square residual (RMSR) of 0.8017. The best models used bulk density and texture but not soil morphological descriptors. Sand textural class predicted the highest Ks. For the finer textured soils, the splits were based on bulk density. The NRCS method predicted Ks with a RMSR of 0.9562. Morphological descriptors of soil structure may not have been important because bulk density acted as a surrogate for structure.