中文摘要：

在天然文岩渠流域大量实测土壤剖面数据的基础上,评价了12种根据基本土壤性质预测不同层次土壤饱和水力传导率的土壤转换函数方法的效果,同时还探讨了多元回归和BP人工神经网络两种构建方法的适用性。结果表明：基于BP神经网络方法的土壤转换函数预测精度均显著优于根据多元回归建立的土壤转换函数,其中基于BP-ANN的Wosten1999函数对于表层和底层土壤预测精度最高,而Li2007方法对第二层土壤预测效果最好;不考虑分层因素时,基于BP-ANN的Wosten1999函数预测效果最好。此外还利用GIS空间插值,对天然文岩渠流域不同深度的土壤饱和导水率进行可视化表达,为模拟该地区的土壤水分运动提供参数支持。

英文摘要：

In this study, based on soil profile data in Tianranwenyanqu Basin, we evaluated the effect of 12 familiar pedo-transfer functions according to the fundamental soil properties to estimate the saturated soil hydraulic conductivity, and then explore the applicability of the multiple regression and BP Artificial Neural Network. The results showed that the prediction accuracy of pedo-transfer functions based BP-ANN is much better than from multiple regression, Wosten1999 based BP-ANN has the highest prediction accuracy for the surface and the bottom layers, Li2007 for the second layer, while Wosten1999 based BP-ANN is the best model without layering. Besides, we use GIS spatial interpolation to express visually the saturated soil hydraulic conductivity at different depths, which could provide basic parameters for modeling soil water movements in this region.