基于支持向量机的土壤水力学参数预测

为了分析支持向量机在土壤水力学参数预测方面的效果，应用支持向量机构建用于预测土壤水力学参数的土壤传递函数，以土壤粒径分布、容重、有机质含量等土壤理化性质为输入项，分别预测土壤饱和导水率、饱和含水率、残余含水率，以及van Genuchten公式参数的对数形式。结果表明预测值和实测值不存在显著性差异，用支持向量机预测土壤水力学参数是可行的。不同输入项处理的预测分析表明，输入项为粒径分布、粒径分布和容重、粒径分布和有机质含量3种情况的预测效果差异不明显，而输入项为粒径分布、容重和有机质含量时预测效果优于前3种情况。支持向量机在预测土壤水力学参数方面的效果要优于多元线性逐步回归模型，而与BP神经网络模型相比不具有明显好的预测效果。

Prediction of soil hydraulic characteristic parameters based on support vector machine

Support Vector Machine(SVM) was applied to establish Pedo-transfer functions(PTFs) to predict the Soil Hydraulic Characteristic Parameters(SHCPs), which included the saturated soil hydraulic conductivity, saturated soil water content, residual soil water content, and the logistic form of van Genuchten model parameters. The input variables were set as different treatments selected from soil texture, soil bulk density and soil organic matter content. Results indicate that there is no obvious difference between predicted and observed values, and that SVM is suitable for SHCPs prediction. The predicted results of different input treatments indicate that the best prediction effect is from the treatment with all the input variables included soil texture, soil bulk density and soil organic matter content. But there is no obvious difference of the prediction effect among other treatments while the input variables is soil texture, or soil texture and bulk density, or soil texture and organic matter. Compared with other methodologies of PTFs, SVM perform better than multivariate stepwise regression model, but with similar prediction efficacy comparing with BP neural network model.