摘要
用随机方法从262个采样点中抽取200个点作为已知有机质含量的数据集,将所有采样点的碱解氮作为辅助数据预测有机质的空间分布。利用有机质信息的普通克立格法的方差解释量和预测精度最低,而回归克立格法因在预测过程中加入了回归残差而使方差解释量最大、预测精度最高。为了分析采样数对不同方法预测精度的影响,从上述已知有机质含量的200个点中分别随机抽取40、80、120、160个点构成4个数据集,分别利用它们的有机质信息和不同方法预测了有机质的空间分布,结果表明:对于每个数据集,4种方法的预测精度顺序均为RGK>COK>RG>OK,线性回归法的预测精度随采样点的增加基本不变,而其它三种方法的预测精度却逐渐提高。
This paper presents four methods for estimating spatial distribution of soil organic matter and examines these methods＇ sensitivity to the sampling number. The four methods are ordinary kriging （OK）, simple linear regression （RG）, cokriging （COK）, and regression-kriging （RGK）. All sampling sites are randomly divided into two groups： interpolation dataset （200 points） and validation dataset （62 points）. The organic matter of interpolation subset methods, ordinary and alkalizable nitrogen of all observations are used to mapping soil organic matter. kriging Among four , only using the information of organic matter, yields lowest accurate predictions and smallest proportion of the total variation, while regression-kriging using secondary data （alkalizable nitrogen） yields highest accuracy and largest variation explainable. To examine the effect of sampling number on the performance of four mapping methods, four subsets of 40,80,120,160 sampling sites are randomly selected from the interpolation dataset. For each subset, organic matter is estimated over the study area by four methods, respectively. The results show that the accuracy performances of four methods are RGK 〉 COK 〉 RG 〉 OK. Moreover, the results indicate that the performance of simple linear regression remain stable, and that others perform better when the sample size of organic matter increased.
出处
《地理科学》
CSCD
北大核心
2007年第5期689-694,共6页
Scientia Geographica Sinica
基金
北京市自然科学基金(4061002)
农业部948项目(2006-G63)资助
关键词
土壤有机质
线性回归
克立格
协克立格
回归克立格
采样数
organic matter
linear regression
ordinary kriging
cokriging
regression-kriging
sample size
