Comparing the Performance of Multiple Linear Regression and Regression Tree to Predict Saturated Hydraulic Conductivity and the Inverse of Macroscopic Capillary Length

The objective of this study was to compare the performance of multiple linear regression and regression tree for deriving pedotransfer functions (PTFs) to predict soil saturated hydraulic conductivity (Kfs) and inverse of macroscopic capillary length parameter (*). Therefore, Kfs and * of 60 points of Azadegan plain in Shahrekord region were measured by multiple constant head method using single ring apparatus. Using some of the readily available soil data of two first pedogenic layers of the soils as inputs, multiple linear regression and regression tree were then applied to derive the PTFs. The accuracy and reliability of the derived PTFs were evaluated using root mean square error (RMSE), mean error (ME), relative error (RE) and Pearson correlation coefficient (r). Results indicated that regression tree predicted the parameters better than multiple linear regressions. Values of relative error (RE) and the mean square error (RMSE) for * estimated by regression tree was 0.24 and 0.019 (cm/min), respectively, that was 0.03 and 0.023 (cm/min) lower than those of multiple linear regression. Furthermore, results showed that bulk density, geometric mean and weight mean of peds diameter had the most major effects on saturated hydraulic conductivity and macroscopic capillary length. Regression tree and multiple linear regressions overestimated and underestimated the soil saturated hydraulic conductivity, respectively.