s r; a r; m n
Abstract
Soil hydraulic properties such as soil water characteristic curve are necessary prerequisite for modeling water movement and solute transport. Direct methods of estimating these hydraulic properties are time consuming and costly. Indirect methods, such as pedotransfer functions, estimate the hydraulic ...
Read More
Soil hydraulic properties such as soil water characteristic curve are necessary prerequisite for modeling water movement and solute transport. Direct methods of estimating these hydraulic properties are time consuming and costly. Indirect methods, such as pedotransfer functions, estimate the hydraulic parameters using easy-to-measure soil properties like particle size distributions, bulk density, or organic matter content. In this study, to estimate soil water characteristic curve, Rosetta pedotransfer function with artificial neural network approach, Soilpar-2, and different regression-based pedotransfer functions were compared and evaluated. For the purpose of comparison and evaluation of pedotransfer functions, statistical criteria of Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Modified Efficiency Coefficient (E'), and Modified Index of Agreement (d') were used. The results show that Rosetta, with mean values of the statistical criteria RMSE, MAE, E' and d' equal to 0.0310, 0.0247, 0.7956, and 0.9037, respectively, enjoyed high accuracy compared to the rest of pedotransfer functions. The results of this study showed that, to estimate soil water characteristic curve, the artificial neural network was more preferable than the regression pedotransfer functions with higher number of input parameters for the study area. The results also indicated that the adjusted Campbell pedotransfer function with RMSE, MAE, E' and d' equal to 0.0685, 0.0530, 0.5561 and 0.8075, respectively, presented the next best estimate of soil water characteristic curve for soils of the study area, after Rosetta.