基于主成分分析土壤水分扩散率单一参数模型的BP神经网络模型

应用水平土柱法测定了杨凌地区典型粘壤土的水分扩散率，利用土壤水分扩散率的单对数模型和双对数模型对其进行了拟合，建立了土壤水分扩散率单一参数模型，基于主成分分析建立了单一参数模型中参数B的BP神经网络模型。结果表明：利用主成分分析可将研究区域土壤容重、有机质含量、粘粒含量、粗粉粒含量和砂粒含量综合成3个主成分；基于主成分分析建立的BP神经网络模型拟合的单一参数模型参数B]的均方根误差RMSE为0.308 2；将拟合得到的参数B代入单一参数模型中对土壤水分扩散率进行预测，除去其中较大值的预测结果偏低外，其余土壤水分扩散率预测结果都比较接近实测值，预测结果的均方根误差RMSE为0.257 8，可利用基于主成分分析建立的BP神经网络模型预测单一参数模型中的参数B。

BP artificial neural network model of one-parameter soil moisture diffusivity model based on principal components analysis

Soil hydraulic property parameters are necessary for the solution of the basic soil hydraulic properties equation. Unsaturated soil hydraulic properties are important physical parameters for modeling soil water and salt movement，in which soil water diffusivity is one of the most important parameters. However，because these parameters have a strong spatial variability，making a direct determination of them in a larger region is often not feasible，and the determination of soil variability by the spatial structure often has a large number of errors. To overcome these shortcomings，soil structure and hydraulic parameters of the function must be established in order to find a determination of soil moisture characteristics of the more simple and effective method. In the paper，soil water diffusivity of typical clay loam in Yangling City，Shaanxi Province，China was measured with horizontal soil column method，then single and double logarithm models of soil water diffusivity were applied to fit above measured values，based on above fitted results，a single parameter model was established，and on this basis a BP artificial neural network of single parameter model was established based on the principal components analysis. The results showed that bulk density，organic matter content，clay content，silt content and sand content could be converted into three principal components；RMSE of parameter B which were fitted with established BP artificial neural network model was 0.308 2；Except large values of soil water diffusivity that its forecasted value was low，soil water diffusivity forecasted based on fitted parameter B was close to its measured value，and RMSE of forecasted soil water diffusivity was 0.257 8，which indicated that established BP artificial neural network model could be used to forecast parameter B in single parameter model.