Abstract: The classification of 72 replicated fertilizer dispersed experiments in rice growing areas of southern China during two years was carried out by using hierachical clustering methods by amalgamation rule of single linkage. The ranges of Euclidean distances between replications for yields, constants and N, P, K partial coefficients of polynomial regression equations were determined. The variables and their thresholds for classification were calculated and tested, according to the precision of classification, i.e. probability that replications of same experimental site were classified into same group. The constant of polynomial equation was a good variable for classification with its precision average of 73.45%. If necessary, the N, P, K partial coefficients could also be selected as classification variables, and the precision was estimated to be more than 65%.

Key words: cluster analysis, threshold, fertilizer experiments, rice

摘要： 采用多点试验中同一试验点的两个重复分于同一类的可能性大小，也即重复识别率作指标，以系统聚类法中常用的最短距离法对按统一方案分布于南方稻区72个氮、磷、钾有重复肥料试验进行分类，确定了肥料试验处理产量、肥效方程常数项及偏回归系数项重复间欧氏距离的变幅、分类指标及其阈值。水稻肥料试验分类效果以肥效方程常数项分类最好，重复识别率两年平均值达73.45%；若按肥效方程氮、磷、钾一次项进行分类，重复识别率一般在65%以上，能基本满足肥料试验结果的分类要求。

关键词: 聚类分析, 阈值, 肥料试验, 水稻