Research On One-Parameter Models Of Soil Water Movements And The Estimates Of Parameters In These Models

Soil moisture movement is a complex process, the parameters of which is the foundation how to build the agricultural water irrigation system, the irrigation time and quantity of water for irrigating. Also, researching the movement of nutrients in the soil is based on the studying of soil water transportation, as well as agricultural pollutants migration. Therefore, studying on the soil water movement parameters’spatial variability is of great significance for these aspects. However, the measurement of getting soil water movement parameters takes a lot of manpower and resources. Considering how to get these data quickly and accurately through some indirect methods will make strong practical significance.In this paper, infiltration field work and laboratory experiments were combined together, and the researching technical route is according the experimental data analysis and computer simulation. According to the soil water dynamics theory, the data of field infiltration experiments was analyzed in order to build geostatistical models of stable infiltration rate and 120min cumulative infiltration respectively. Then, the analysis of soil infiltration characteristics spatial variability on single scale was following. At the same time, joint multi-fractal analysis method would be used to analyze the infiltration characteristics correlation among soil properties on multi-scale. Data from laboratory experiments were analyzed and processed, and the one-parameter models based on the parameters of soil water movement (diffusion rate of soil moisture, soil moisture characteristic curve, unsaturated soil and hydraulic conductivity) were built. Then was the following this one parameter’s BP neural network models, which revealed the relationships between single parameter and soil physics and chemistry properties. The findings as follows:(1) Appropriate infiltration equations were chose to fit the infiltrating procedure, which were Kostiakov formula and Philip formula. Comparing the fitting accuracy of these two formulas, the Kostiakov formula was selected to calculate the fitting parameters such as stable infiltration rate and cumulative infiltration.(2)Statistical method was used to analyze single-scale spatial variability of the infiltration characteristics. According to semivariogram function fitting results, there had significant difference between the steady infiltration rate and 120min cumulative infiltration. Space variability of stable infiltration rate was stronger than that of 120min cumulative infiltration, but they all belong to strong variation; At 500m intervals of sampling scale, stable infiltration rate and 120min cumulative infiltration showed strong spatial variability, and both variables had self-similarity.(3) Under the use of joint multi-fractal method, the infiltration parameters’spatial variability at multi-scale had be analyzed, and found out which was the greater impact factor among soil physical and chemical characteristics. According to the results of multifractal analysis, initial soil water content and soil bulk density had more influence on the spatial variability of those two variables (120min cumulative infiltration and stable infiltration rate). While soil bulk density had the greatest impact on 120min cumulative infiltration capacity with the correlation coefficient reaching 0.700, the impact of initial water content impacted the on soil infiltration rate with the maximum correlation coefficient reaching about 0.500.(4) Through the way of Principal Component Analysis, soil physical and chemical characteristics (including initial moisture content, soil bulk density, organic matter content, clay content, coarse silt and sand content), comprehensive four main components could be got. And these four main components, containing all the information of soil physical and chemical properties, achieved the purpose of dimension reduction. Based on this, the four main components were used as input variables of BP neural network models. At the same time, soil water movement parameters’(including soil water diffusivity, soil moisture characteristic curve, unsaturated soil hydraulic conductivity) one-parameter models’ parameter (A or B) were chosen as output variables. Then, the appropriate number of hidden layer of the neural network should be set before training. In that training process,46 groups of data were randomly selected as training samples, and the later 6 groups of data would be selected as predicting samples to analyze the accuracy of the model.BP neural network models based on one-parameter models of soil water movement built in this thesis, can directly estimate soil water movement parameters from the soil physical and chemical parameters. Under this situation, these models can provide method, which solving the difficulties of getting the data of soil moisture movement parameters from experiments in the large area. But the prediction effect is relatively poor, which required further studying.