The definition of robust climate analogues implies that the appropriate clustering method should emphasize the cluster properties of compactness and distance over connectedness (Janssen et al. 2012). Therefore methods which tend to generate spherical clusters seem adequate. As the analysis of the topological structure and potential dimension reduction (by SOM-based methods, e.g. Kohonen 1998) are not of major importance in our case the partitioning cluster method k-means (MacQueen 1967; Hartigan and Wong 1979) has been applied. This algorithm minimizes the total within-cluster sum-of-squares (TSS) criterion (Steinley 2006). If the data set consists of V variables and the number of groups is chosen to be K, the criterion is defined by:

中文翻译：

---

一种识别稳健气候类似物的新方法

稳健气候类似物的定义意味着适当的聚类方法应该强调紧密性和距离而非连通性的聚类特性（Janssen 等人，2012 年）。因此，倾向于生成球形簇的方法似乎是足够的。由于拓扑结构的分析和潜在的降维（通过基于 SOM 的方法，例如 Kohonen 1998）在我们的案例中不是很重要，因此划分聚类方法 k 均值（MacQueen 1967；Hartigan 和 Wong 1979）已被应用。该算法最小化了总的集群内平方和 (TSS) 标准 (Steinley 2006)。如果数据集由 V 个变量组成，并且组数选择为 K，则标准定义为：