Estimation of saturated hydraulic conductivity by using gene expression programming and ridge regression (A case study in East Azerbaijan province)

Soil hydraulic conductivity is an important physical characteristic which used in modeling of water, solutes and pollutants transport. Direct measurement of soil hydraulic conductivity is time-consuming and costly, and due to experimental error and soil heterogeneity, results sometimes are unrealistic. On other hand, it could be estimated by easily measurable soil properties. The purpose of this study is development of genetic programming and linear regression models to estimate the saturated hydraulic conductivity of soil using readily available soil properties. For this reason, 160 soil samples with different properties were taken from different area of East Azerbaijan province of Iran. Then some physical and chemical characteristics of soil such as sand, silt, clay and organic matter, bulk density, pH and EC were measured. Then the data was divided into the two different data sets, namely training (75% of data) and testing datasets (25% of data). GeneXproTools 4.0 and Statistica software’s was used to calibration of Genetic programming and regression models, respectively. Six pedotranfer functions (PTFs) with a combination of different mathematical operators was designed by genetic programming. Finally, one of the PTFs which was more accurate than the others was selected. As well as ridge regression was used to development of regression PTFs. The accuracy and reliability of PTFs were determined by R2, RMSE, and MAE criteria. Results showed that the genetic programming PTF (GP-PTF) is more accurate and reliable than the regression-PTF. So that R2, RMSE (Cm h-1) and MAE (Cm h-1) of GP-PTF for training dataset were 0.91, 1.82 and 1.23 respectively, and for test dataset were 0.92, 2.27 and 1.59, respectively. While values of the above mentioned criteria of regression-PTF for training dataset were 0.70, 3.48 and 2.07, respectively, and for test dataset were 0.76, 3.11 and 1.88, respectively.