| 摘要: |
| 在美国土壤水分物理性质数据库(UNSODA 2.0)的基础上,考虑土壤质地不分类和分类2种情况,分别构建了基于支持向量回归机(SVR)的土壤传递函数模型,比较了在土壤质地不分类和分类情况下预测土壤水力学参数(水分特征曲线和饱和导水率)的效果,并与建立在相同数据库上的基于神经网络的Rosetta模型的预测效果进行了比较。结果表明:土壤质地不分类的情况下,输入参数越多,基于SVR模型的预测效果越好;土壤质地分类情况下,基于SVM分类建模的预测结果普遍好于不分类情况。无论土壤质地是否分类,样本和输入参数相同的条件下,基于SVR的模型预测的效果都优于Rosetta模型。 |
| 关键词: 土壤质地分类 传递函数 支持向量回归机 神经网络 |
| DOI：10.11841/j.issn.1007-4333.2010.06.017 |
| 投稿时间：2010-04-19 |
| 基金项目:国家科技支撑计划项目(2008BADA7B05); 公益性行业科技项目(200803036); 教育部新世纪优秀人才支持计划项目(NCET-07-0809)第一作者:聂春燕,硕士研究生,E-mail:niechunyanzhy@163.com通讯作者:陈薇,教授,主要从事统计分析、可拓学及其应用研究,E-mail:chenwei@cau.edu.cn |
|
| Comparison of predicting results of soil hydraulic parametersby SVR and rosetta models |
|
NIE Chun-yan1, HU Ke-lin2, SHAO Yuan-hai1, CHEN Wei1
|
| (1.College of Science,China Agricultural University,Beijing 100193,China;2.College of Resources and Environmental Sciences,China Agricultural University,Beijing 100193,China) |
| Abstract: |
| Based on the database of America soil water physical characteristics (UNSODA 2.0),the PTFs models for both classified and unclassified soil textures were constructed by using Support Vector Regression (SVR) approach,and the predictive results of soil texture under both classified and unclassified situations for soil hydraulic parameters (V-G model parameter and soil saturated hydraulic conductivity),as well as the results from the Rosetta model based on neural network and constructed by the same database,were compared.The results indicated that the more the number of parameters input,the better the results obtained by using SVR model in the situation of soil texture unclassified;however,in the situation of the classified,the results obtained from separately constructed SVM model are generally better than that of the unclassified.No matter whether the soil texture classified or not,on the occasion of the same sample and parameters input,the results obtained from the model based on SVR have an advantage over Rosetta model. |
| Key words: soil texture classification pedotransfer functions support vector regression neural network |
