土壤质量评价中少量样本最小数据集的构建——以内蒙古杭锦旗黄河南岸灌区典型地块为例

为研究土壤质量评价过程中样本容量较少情况下最小数据集的构建方法。采集并测定内蒙古杭锦旗黄河南岸灌区典型地块的12个土壤理化指标，利用聚类分析、相关分析和主成分分析等探讨了少量样本最小数据集的构建方法。结果表明:1)通过聚类分析、相关分析和主成分分析所构建的最小数据集中的指标包括土壤容重、饱和含水量、土壤电导率、阳离子交换量、硝态氮、速效钾、速效磷；2)全体数据集与最小数据集的土壤质量指数呈显著正相关，R2达到了0.735，Nash有效系数为0.917，偏差系数为0.057；3)基于全体数据集和最小数据集计算得出的土壤质量指数变化范围分别是0.47~0.73和0.37~0.75，平均值为0.56和0.53，表明研究地块土壤呈现中等质量水平。研究发现，在黄河南岸灌区典型地块基于少量样本进行土壤质量评价的过程中，使用聚类分析、相关分析、主成分分析3种分析方法可以构建最小数据集，且检验精度较高。

Construction of minimum data set with small number of samples for soil quality assessment: A case study of a typical land in the south bank of Yellow River irrigation area of Hangjin Banner, Inner Mongolia

In order to study the construction of minimum data set by using the small number of samples for soil quality evaluation, this research collected and measured twelve soil physical and chemical indicators of a typical field in the south bank irrigation area of the Yellow River in Hangjin Banner, Inner Mongolia. The methods for constructing the minimum data set with small number of samples by cluster analysis, correlation analysis and principal component analysis were investigated. The results showed that: 1)The indexes in the minimum data set established by cluster analysis, correlation analysis and principal component analysis are soil bulk density, saturated water content, soil conductivity, cation exchange capacity, nitrate nitrogen, available potassium and available phosphorus; 2)The soil quality index of both the total data set and minimum data set is significantly positively correlated, where R2 is 0. 735, the Nash effective coefficient is 0. 917, and the deviation coefficient is 0. 057; 3)Based on the total data set and minimum data set, the variation range of soil quality index are respectively 0. 47-0. 73 and 0. 37-0. 75 and with an average of 0. 56 and 0. 53, indicating the soil quality level of the study field is moderate. In conclusion, for soil quality evaluation in the typical irrigated area on the south bank of the Yellow River, the minimum data set can be constructed based on a small number of samples by using three methods including cluster analysis, correlation analysis and principal component analysis, and the test accuracy is high.