采样数量与空间插值方法对土壤属性预测精度的影响

发布时间：2018-08-03 11:33

【摘要】：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。
[Abstract]:Different sampling amount and spatial interpolation method will affect the precision of soil attribute prediction. In this paper, the typical arid farming area of Huaibei Plain in Anhui Province is taken as an example. The random sampling method is used to select 50 samples from 550 soil samples as validation samples, and then 500400300200 and 1005 subsets of samples are selected. The spatial variation of soil organic matter in the study area was analyzed by four spatial interpolation methods: ordinary Kriging method, inverse distance weight interpolation method, local polynomial interpolation method and radial basis function interpolation method. The difference of prediction accuracy between different sample size and interpolation method is compared by independent verification and cross validation. It is found that the precision of soil attribute prediction is greatly affected by different sampling quantities. In the county scale of Huaibei Plain, 200 samples are more reasonable in soil sampling for the purpose of digital soil attribute mapping. The results of cross-validation and independent verification show that the common Kriging method is the best, the local polynomial interpolation method and the radial basis function interpolation method are the second, and the inverse distance weight interpolation method is the worst.
【作者单位】：武汉大学遥感信息工程学院;安徽理工大学测绘学院;武汉市规划研究院;中国矿业大学环境与测绘学院;
【基金】：国家级大学生创业创新训练计划(201510361114) 安徽省高校自然科学研究项目(KJ2015A034)资助
【分类号】：S151.9

【相似文献】