Prediction Of Saturated Hydraulic Conductivity Of Surface Soil In Sand-dune-and-meadow Interlaced Region Of Horqin With Pedo-transfer Functions Method

This paper takes the tuo-survey area, typical sand type in Khorchin as the study area. Two hundred and forty sampling points were set in the wild area, and what’s more, the physical and chemical properties were determined, including the moisture content, dry density, organic matter, saturated hydraulic conductivity and so on. Analysis of the physicochemical properties of the data, draw the following conclusion:1. The surface soil of nine types of landforms were sampled, which included the mobile sand dunes, semi-fixed sand dunes, fixed sand dunes, poplar in sand dune area, cultivated land in sand dune area, low coverage meadow, high coverage meadow, cultivated land in meadow area, and abandoned land. And analyzed the physical and chemical parameters of surface soil under different geomorphic types.2. These four kinds of soil---Campbell, Cosby, Wosten, Saxton, were chosen to transfer the function and to predict the saturated hydraulic conductivity of surface soil in this region. And the results showed that the predicted values of these soil transfer functions differed from the measured values a lot, and the correlation coefficient was less than0.3, so the precision was difficult to meet this region.3. This paper combined the methods of principal component analysis and nonlinear regression analysis. In this way, the soil transfer function of the saturated hydraulic conductivity of surface soil in this region was set up. Furthermore, in the soil transfer function, we only used organic matter and soil particle size parameters. Because of the correlation coefficient of predicted values and measured values was0.648, this transfer function could be applied to predict the saturated and hydraulic conductivity of surface soil in the region of Khorchin sand dune.4. Application of BP netural network technology is used to predict the surface soil saturated hydraulic conductivity. On this basis, the paper selected these five kinds of soil characteristic paramrters---soil bulk density, organic matter content, saturated moisture content, the average partecle size, and particle size of the standard deviation as the input variables, ans to validate the established PTFS funtion. The result shoes that correlation coefficient of forecast values and measure values, which is used to build and test data of the model, were grater than0.7, and have a good correlation. Therefore, it points out the parameters of BP netural network model is chosen reasonably.5. After analysing and evaluating comprehensively the effectiveness, applicability and accuracy of these two function models, we can find that the soil transfer function (PTFs) built by BP netural network technology is superior to the traditional regression method. Hoewver, the predictions effect of the soil transfer function (PTFs) built by the two models are both ideal. So they can be both applied in Khorchin test area.