旱作区阳离子交换量土壤转换函数的研究

Study on Pedo-Transfer Functions of Cation Exchange Capacity in Dry Farming Region

摘要: 为减少大范围测定阳离子交换量(cation exchange capacity, CEC)工作量，探寻一种利用较容易测定的土壤理化性质预测CEC简单、高效的方法。本文基于旱作区152个表层0~10 cm土壤的pH、质地、有机质等数据，建立BP神经网络、支持向量机、多元线性回归3种CEC土壤转换函数，并比较分析了各方法的预测精度和敏感性。结果表明，CEC主要受有机质、黏粒的影响且呈明显的正相关关系；支持向量机预测精度最高，其决定系数(R² = 0.58)、效率系数(E = 0.57)均高于BP神经网络和多元线性回归，均方根误差(RMSE = 5.41)均低于其他模型，多元线性回归模型预测精度最低。支持向量机方法在旱作区内能够较好地预测阳离子交换量。

Abstract: In order to reduce the workload of large-scale determination of cation exchange capacity (CEC), a simple and efficient method for predicting CEC by using easily measured soil physical and chemical properties was explored. Researcher have used different input soil properties to derive pedo-transfer functions (PTFs) to predict soil CEC. Based on the pH, texture and organic matter (OM) of 152 topsoil (0~10 cm) samples in dry farming region, the CEC soil transfer function was established by back propagation neural network (BP-NN), support vector machine (SVM) and multiple linear regression (MLR), and the prediction accuracy and sensitivity of each method were compared and analyzed. The results showed that CEC was mainly positively correlated with OM and clay. The prediction accuracy of SVM was the highest, with its decision coefficient (R² = 0.58) and efficiency coefficient (E = 0.57) higher than BP-NN and MLR. The root mean square error of SVM (RMSE = 5.41) is lower than other models, and MLR model has the lowest prediction accuracy. SVM method can better predict cation exchange capacity in dry farming.