气候分析应用聚类方法的评价
EVALUATION OF CLUSTERING PROCEDURES IN CLIMATIC CLASSIFICATION
-
摘要: 对聚类分析的华德法、重心法和平均联系法的理论分析指出:在组间距离相同的条件下,华德法优先合并小的组,最终产生大小较为均匀的组,这不能准确地反映气候的现实。在组内平方和增量相同的条件下,重心法优先合并大的组,最终产生一个很大的组和很多极小的组,背离气候的现实。平均联系法基本上避免了上述偏歧,产生凝聚、方差均匀的组,较好地反映气候现实。中国东部大陆夏季降水的分区证实了以上结论。Abstract: Theoretical analyses on three clustering procedures show that:being influenced by the group size of merger at any step of fusion,Wards method gives the priority to the merger with less observations,and finally produces clusters of close size.On the contrary,centroid method gives the priority to the merger with more observations,and finally produces a huge cluster and many small clusters some of which contain only one observation.Average linkage method is almost not influenced by the group size and produces cohesive clusters with close variance.It duplicates climatic reality with the greatest accuracy.A regionalization of year-to-year variation in summer precipitation in eastern China verifies above mentioned conclusions.