Development of regional-scale pedotransfer functions based on Bayesian Neural Networks in the Hetao Irrigation District of China

In order to study determination the soil hydraulic parameters in the distributed hydrological models on farmland environmental effects resulted from water-saving practices of large scale irrigation district, the Bayesian Neural Networks and BP ANN model were applied to establish regional pedotransfer functions models based on the relationship of measured soil characteristic contents (saturated water content θs, residual water content θr and field water content θr), soil particle percentage, organic matter and bulk density and fitted VG model parameters of different soil texture classes from 22 soil water and salt monitoring points 110 soil samples in the Hetao Irrigation District. Then, the adaptability of two kinds of ANN models were evaluated by simulated and predicted results through the statistical results and SWRC figures. The several conclusions were reached: the ANN and BNN are both feasible PTFs methods. But, the training simulated accuracy of traditional BP model is better than that of BNN; however, the predicted accuracy of BNN model generally is better than the BP model. Furthermore, the number of input factors groups has significantly influenced the predictive accuracy of BP model. But there are little influences on the different inputs factors of BNN model. So, the BNN showed good robustness for the simple inputs. Second, the predicted SWRC has better fitness with measured and VG fitted curve than that of ANN. So, the BNN model is better than the traditional artificial neural network model has better adaptability in the peodotransfer function establishment when it uses only soil particle distribution. The BNN method is a practical method for regional pedotransfer function establishment.