土壤属性空间预测的广义回归神经网络方法研究  

Study on Method of Generalized Regression Neural Network for Spatial Prediction of Soil Properties

作　　者：刘付程[1] 郭衍游[1] 闫晓波 

机构地区：[1]淮海工学院测绘工程学院,江苏连云港222001 [2]胶州市环境保护局,山东胶州266300

出　　处：《淮海工学院学报（自然科学版）》2008年第1期68-71,共4页Journal of Huaihai Institute of Technology:Natural Sciences Edition

基　　金：江苏省教育厅2006年度"青蓝工程"资助项目;淮海工学院自然科学基金资助项目(Z2004020)

摘　　要：土壤属性的空间变异性是农业和环保决策的重要依据,研究如何利用有限的样本数据来获得更为详尽的土壤属性空间分布信息对于科学决策具有重要意义。提出了一种基于广义回归神经网络模型(GRNN)的土壤属性空间预测方法,并将其与Kriging方法进行了比较。结果表明,GRNN方法能较好地刻画土壤属性的空间分布特征,其计算方法简单,预测精度较高;在给定的样本数据条件下,GRNN方法的预测精度总体上要优于Kriging方法,表明GRNN方法在土壤属性空间预测中的应用是有效可行的。The spatial variability of soil properties is the basis of decision-making in agricultural management and environmental protection, and it is also of great significance in scientific decision-making to utilize limited sample data to obtain detailed spatial information of soil properties. The Generalized Regression Neural Network （GRNN） method is put forward in this paper to predict spatial distribution of soil properties, and it is compared with the Kriging Method. Results show that GRNN could better characterize the spatial distribution of soil available Zn, with a simple algorithm and high precise estimates. And the prediction errors of GRNN are lower than those of Kriging Method given the interpolation data set of soil samples described in this paper. Therefore, GRNN is superior to Kriging and can be used to predict soil properties with better results.

关 键 词：广义回归神经网络 土壤属性 空间预测 

分 类 号：S152[农业科学—土壤学] TP183[农业科学—农业基础科学]