LAUSR

Abstract Several variables have been used to estimate the soil moisture retention curve (SMRC) so far, but one of the main points in using pedotransfer functions (PTFs) is the selection of readily available soil properties as inputs. In this study, 147 disturbed and undisturbed soil samples were taken from five provinces of Iran. Some soil properties such as particle size distribution (PSD), bulk density (BD), free swelling index (FSI) and SMRC were measured. A double exponential model was fitted to the SMRC data, and PTFs were developed to predict the model parameters, by multivariate linear regression and artificial neural network (ANN) methods. Using FSI with the soil basic properties (silt to sand ratio, clay and BD) as estimators improved the estimation of SMRC significantly for the full data set by ANN method (by 10.9 and 9.0% in the training and testing steps, respectively), and in the textural groups of “clay loam, silty clay loam”, “clay”, “loam”, “sandy loam” and “sand”, by the linear regression method in the range of 1.0 to 14.9%. Grouping the soil samples significantly improved the accuracy of the estimates, especially in the sandy soil textural group. In estimating SMRC using total data, ANN method had a greater accuracy compared with the linear regression method. In general, FSI significantly improved the accuracy of parametric estimation of the SMRC and grouping of data had a positive impact on improving estimates in some textural groups.