A Method Of Prediction For Soil Hydraulic Properties Based On Pedotranfer Functions And Its Applications

Soil hydraulic properties are important parameters in models for soil water and solute transport.Unfortunately,direct measurement of the hydraulic properties is relatively time-cousuming,troublesome and therefore costly.Instead of measuring the hydraulic properties,estimation techniques have been developed such as pedotransfer functions(PTFs).On the basis of linear regression,extended nonlinear regression or artificial neural network,PTFs convert easily measurable basic soil properties(e.g.,soil texture,organic matter content and bulk density)to the hydraulic properties.Owing to the significant discrepancies among the methods for the development of PTFs,it is important to select the proper method or even propose a new one to better predict soil hydraulic properties.Additionally,relatively little research has been done to evaluate estimates from PTFs chiefly with respect to some practical problems that may limit the applications of PTFs to a certain extent.The comparison of various PTFs and their applications were the focus of the research.To achieve the final goal,five initial problems as listed below must be solved in the course of study:(1).How to develop a more effective pedotransfer function to better describe the correlation between basic soil properties and the hydraulic characteristics.In this paper,a new PTF,named as PCRRNN,was developed to predict van Genuchten parameters based on principal component regression and artificial neural network.In order to further detect the effectiveness of PCRRNN,quantitative comparison of the errors generated by PCRRNN and other PTFs(developed by linear regression,extended nonlinear regression and artificial neural network)was conducted.(2).How to evaluate the published PTFs of Wosten(Wosten,1997),Rawls(Rawls and Brakensiek,1985)and Campbell(Campbell,1985;Campbell and Shiozawa,1992)for unsaturated soil hydraulic conductivity.According to the hydraulic conductivity functions including the models of van Genuchten-Mualem,Brooks-Corey and Campbell-Norman,three published PTFs including Wosten,Rawls and Campbell were selected to predict hydraulic conductivity of sandy soils of Mellendorf and Wunstorf Cities,Germany.None of the above PTFs have yet been verified against these data.Quantitative comparison of the errors generated by Wosten,Rawls and Campbell was conducted.Additionally,different PTFs for the estimation of saturated hydraulic conductivity were also evaluated.(3).How to apply multimodel method to simulate one-dimensional soil water movement.The HYDRUS-1D model was used to simulate one-dimensional soil water flow in a wheat-maize field of Qingdao City.Results showed that the satisfactory agreement between the observed and calculated soil water content did indicate applicability of HYDRUS-1D model to the flow processes at this site.Subsequently,eight published PTFs including Bruan,Canarache,Gupta,Hall,Petersen,ROSETTA,Vereecken and Varallyay were used to predict soil water retention characteristics.The HYDRUS-1D model was used next to simulate flow using hydraulic parameters obtained with the PTFs.Significant differences in prediction accuracy were found among eight model outputs.It was clear that uncertainties in PTF selection have a significant effect on simulated soil moisture.To deal with these uncertainties,four multimodel methods(e.g.,simple averaging,superensemble forecasting,superensemble with principal component analysis and Akaike imformation criterion)were used in flow simulation based on soil moisture monitoring data and eight model outputs.(4).How to evaluate the effectiveness of point(Hall and Gupta-Larson)and parametric PTFs(Wosten and Vereecken)to detect the structure of soil water retention spatial variability.Point(Hall and Gupta-Larson)and parametric PTFs(Wosten and Vereecken)were used to predict soil water retention characteristics(e.g.,?33 and ?1500)in the region of Smeaton,Canada.Quantitative comparison of the errors generated by the above PTFs with three sampling intervals(e.g.,3m,6m and 12m)was conducted.Results showed that among the four PTFs,Gupta-Larson was found to be the best one,followed by Wosten and Vereecken.The accuracy of Hall was the worst of all the PTFs.The sampling interval had no significant effect on PTF predictions.Additionally,geostatistics was also used to analyze the spatial variability of predicted soil water retention characteristics.(5).How to quantify the uncertainty in PTFs-based spatial estimation of van Genuchten parameters.Spatial distribution of basic soil properties was predicted using Cokriging with multispectral remotely sensed data(e.g.,ETM).Linear regression was used to develop PTFs for prediction of van Genuchten parameters at the same time.Subsequently,spatial distribution of van Genuchten parameters was predicted based on PTFs and spatial distribution of basic soil properties.The predictions were evaluated in terms of their variability.Results showed that the variability of the PTFs estimates was significantly smaller than that of the observations,but the changing trend of the predictions was consistent with that of the observations.Additionally,bootstrap and Lation hypercube sampling(LHS)methods were used to analyze the uncertainty in PTFs predictions.Results showed that uncertainty of the estimated van Genuchten parameters was attributed to two sources:the PTF intrinsic uncertainty and uncertainty of the estimated basic soil properties.When the PTF intrinsic uncertainty was considered,for the ?s,ln(?)and n parameters,33%,31%and 47%of measurements were encompassed by the 95%predictions intervals.When the PTF intrinsic uncertainty and the uncertainty of the estimated basic soil properties had been considered,however,84%,66%and 90%of measurements were encompassed by the 95%predictions intervals.Through this study,the conclusions can be summarized as follows:? Compared with the traditional method(linear regression,extended nonlinear regression and artificial neural network),the new method PCRRNN was more accurate to predict van Genuchten parameters.? Among the three published PTFs including Wosten,Rawls and Campbell,Campbell and Rawls were respectively the best for predicting unsaturated hydraulic conductivity of sandy soils in the regions of Mellendorf and Wunstorf.The accuracy of Wosten was the worst of all the PTFs.Additionally,The predictions of unsaturated hydraulic conductivity could be significantly affected by the value of the saturated hydraulic conductivity.An input of measured saturated hydraulic conductivity may not improve the accuracy of unsaturated hydraulic conductivity.? Among the four multimodel methods,superensemble forecasting was found to be the worst method,followed by simple averaging.Akaike imformation criterion(AICc)was better than the above two methods,and the performance of AICc was no different from using the best in training model.Using principal component analysis(PCA)in the superensemble appeared to be the best method,and the performance of superensemble with PCA was comparable to that of flow simulation with the HYDRUS-1D model by using measured hydraulic properties as inputs.?Compared with the parametric PTFs(Wosten and Vereecken),point PTFs(Hall and Gupta-Larson)were better at describing the spatial variability of soil water retention characteristics.? The uncertainty of the estimated basic soil properties dominated over the PTF intrinsic uncertainty and determined the spatial distribution of the uncertainty of the estimated van Genuchten parameters.Only two types of uncertainty were considered,the spatial variability of the measured van Genuchten parameters was well captured.The development of PTFs and their applications are basis for the combination of PTFs and hydrological model.This is of great significance and ecological values to the regional water resources management and agricultural non-point pollution control.