| Based on the soil water characteristic curve test,unsaturated soil hydraulic conductivity test,and conventional soil physical and chemical parameter series tests of 241 undisturbed loess soil samples in the field at low suction(<101KPa),the paper systematically studies the main influencing factors of unsaturated hydraulic conductivity and soil water characteristic curve.The soil Pedo-transfer function of the water movement parameters empricial model parameters using the soil physical and chemical parameters as the independent variables was established,including unsaturated soil hydraulic conductivity two-parameter power function,three-parameter power function,two-parameter exponential function model parameters and soil water characteristics Curve Brooks-Cory,vanGenuchten,Frelund-Xing model parameters.Based on the soil water characteristic curve and the unsaturated hydraulic conductivity,the soil Pedotransfer function of the two-parameter exponential function model of soil water diffusivity of the undisturbed loess was established.The effects of temperature on unsaturated hydraulic conductivity and soil water characteristic curve of undisturbed loess were discussed.The results show that:(1)Soil texture,structure and organic matter content are the main factors affecting the soil water characteristic curve and unsaturated hydraulic conductivity of undisturbed loess.Compared with the whole stage soil water characteristic curve and unsaturated water conductivity,the low suction stage shows greater variability.By single factor analysis,the single factor function relationship between 14 model parameters and each influencing factor was finally determined.(2)It is feasible to predict the parameters of the unsaturated hydraulic conductivity model of the undisturbed loess using the conventional soil physical and chemical parameters using BP neural network model and support vector machine model.The average absolute error of the two soil transfer functions for predicting the unsaturated soil hydraulic conductivity model parameters of the training sample are 0.0203 and 0.0151,and the average relative error are 0.0170,0.00937,the average root mean square error values are 0.476 and 0.161,respectively.The average absolute error of the two soil transfer functions for predicting the unsaturated soil hydraulic conductivity model parameters of the verification sample loess are 0.0182 and 0.0139,and the average relative error are0.0210,0.0161,the mean root mean square error values are 0.517 and 0.394,respectively.Based on model comparison,it is recommended to use a twoparameter exponential function and a support vector machine soil transfer function optimized based on particle swarm optimization as the optimal forecast model of unsaturated hydraulic conductivity of undisturbed loess soil.The method is used to predict the unsaturated hydraulic conductivity of the loess soil in the training sample,the average absolute error,the average relative error,and the average root mean square error are 0.0413,0.0381,and 0.394,respectively,and they have strong training ability on training samples.For the verification sample,the average absolute error,the average relative error,and the average root mean square error are 0.0425,0.0400,and 0.429,respectively,which have strong generalization ability to the verification sample.(3)It is feasible to predict the parameters of the undisturbed loess soil water characteristic curve model using non-linear model,BP neural network model and support vector machine model.Three soil transfer functions were used to predict the parameters of the loess soil water characteristic curve model of the training sample,the average absolute error between the predicted value and the measured values are 0.118,0.0111,0.000525,and the average relative error values are0.0965,0.0401,0.0249,and average root mean square error values are 1.436,0.0558,and 0.0619 respectively.For the verification sample,the average absolute error between the predicted value and the measured values are 0.0902,0.0147,and 0.00691,and the average relative error values are0.0809,0.0343,and 0.00325,and average root mean square error values are 0.781,0.0417,0.0146.(4)It is feasible to predict the parameters of the undisturbed loess soil diffusivity index function using conventional physical and chemical parameters of soil using BP neural network model based on genetic algorithm and support vector machine model based on particle swarm optimization algorithm.The average absolute error of the two-parameter exponential function of the undisturbed loess soil diffusivity predicted by the two soil transfer functions on the training samples is 0.0751,0.0415,and the average relative error is 0.0614,0.0409,mean root mean square error values are 3.094 and 2.016,respectively;mean absolute error of predicted loess soil diffusivity model parameters for validation samples values are 0.0233,0.0317,mean relative error values are0.0642 and 0.0573,and the mean root mean square error values are 1.442 and0.0511,respectively.It is recommended to use the support vector machine soil transfer function as the optimal prediction model of undisturbed loess soil diffusivity.(5)Temperature has a significant effect on the unsaturated hydraulic conductivity and soil water characteristic curve of the undisturbed loess.With the increase of temperature,the unsaturated hydraulic conductivity(with suction)and the characteristic curve of soil water show a trend of gradually shifting to the left.That is,under the same suction condition,the higher the temperature,the greater the unsaturated hydraulic conductivity of the soil,and the smaller the water content of the soil.Through single-factor analysis,the single-factor function relationship between the two-parameter power function,three-parameter power function,and two-parameter exponential function model parameters of the soil unsaturated hydraulic conductivity of the loess was finally determined.The single factor function relationship between temperature and Brooks-Cory,vanGenuchten,and Frelund-Xing model parameters of loess soil water characteristic curves was determined.Based on model comparison,it is recommended to use a two-parameter exponential function as the best fitting model for unsaturated hydraulic conductivity of loess soil under different temperature conditions.And the Frelund-Xing model is recommended to be used as the best fitting model for soil water characteristic curve model under different temperature conditions. |
PDF Full Text Request