Prediction of saturated hydraulic conductivity Ks of agricultural soil using pedotransfer functions

Abstract

The determination of the saturated hydraulic conductivity Ks on a field scale presents a challenge in which several variables have to be considered. As there is no benchmark or reference method for the Ks determination, the suitability of each available method has to be evaluated. This study is aimed at the functional evaluation of three publicly available types of pedotransfer functions (PTFs) with different levels of utilised predictors. In total, ten PTF models were applied to the 56 data sets including the measured Ks value and the required predictors (% sand, silt and clay particles, dry bulk density, and organic matter/organic carbon content). A single agricultural field with a relatively homogenous particle size distribution was selected for the study to evaluate the ability of the PTF to reflect the variability of Ks. The correlation coefficient, coefficient of determination, mean error, and root mean square error were determined to evaluate the Ks prediction quality. The results showed a high variability in Ks within the field; the measured Ks values ranged between 10 and 1261 cm/day. Although the tested PTF models are based on a robust background of soil databases, they could not provide estimates with satisfactory accuracy unless local soil data were incorporated into the PTF development.

Keywords

• functional evaluation
• machine learning
• neural network, non-linear regression
• soil hydraulic properties