Different sampling sizes and spatial interpolation methods will affect the accuracy of prediction of soil properties.This study selected Mengcheng County in Northern Plain of Anhui Province as an example.50 samples were selected from total 550 soil samples as validation subset using random sampling method.The five samples subsets,500,400,300,200,100 samples,were extraction from remaining 500 samples.Ordinary Kriging,inverse distance weighting method,local polynomial,radial basis function interpolation were used to research spatial variability of soil organic matter,and independent validation and cross-validation were used to compare the accuracy of predictions of different sampling sizes and spatial interpolations.The results showed that:different number of samples will have a greater effect on the regional soil properties prediction accuracy in northern Anhui plains county scale,digital soil mapping property for the purpose of soil sampling,approximately 200 samples is reasonable sampling number.Different interpolation methods on prediction accuracy was not significant from a cross-validation and independent verification of the root mean square error,the ordinary Kriging optimal,local polynomial interpolation and radial basis function method followed by inverse distance weighting interpolation results most difference.%不同的采样数量以及空间插值方法会对土壤属性预测精度产生影响.本文以安徽省淮北平原典型旱作农业区蒙城县为例,采用随机抽样的方法,从550个土壤样点中随机抽取50个样点作为验证样本,之后抽取500、400、300、200和100五个样本子集,采用普通克里格法、反距离权重插值法、局部多项式插值法和径向基函数插值4种空间插值方法,解析研究区土壤有机质空间变异,并利用独立验证和交叉验证方法比较不同样本数量和插值方法的预测精度的差异.发现不同采样数量对区域内土壤属性预测精度会产生较大影响,在淮北平原县域尺度上,以数字土壤属性制图为目的的土壤采样中,200个样点是比较合理的采样数量.不同插值方法对预测精度影响不显著,从交叉验证与独立验证的结果均表明,普通克里格法最优,局部多项式插值法和径向基函数插值法次之,反距离权重插值法的结果最差.
展开▼
