In this paper, we propose a community detection method to find regions in spatial data with higher correlations. We construct a ‘correlation deviation graph’ that considers the influence of global correlation caused by the node's geographical location. We compare the performance of the correlation deviation graph to conventional correlation graph construction approaches for community detection, using synthetic data, and show that the proposed method has the best performance in the presence of a global correlation structure. We also compare the performance of signed spectral clustering and the signed Louvain algorithm for community detection on the correlation graphs, and find that the signed spectral clustering algorithm performs best. We apply our algorithms to simulated earthquake ground motion data in the Los Angeles region. The results suggest that communities of high correlation in ground shaking tend to be associated with common geological conditions and relative location along the rupture strike direction.

在本文中，我们提出了一种社区检测方法来在空间数据中找到具有更高相关性的区域。我们构建了一个“相关偏差图”，其中考虑了由节点地理位置引起的全局相关性的影响。我们使用合成数据将相关偏差图的性能与用于社区检测的常规相关图构建方法进行了比较，结果表明，该方法在存在全局相关结构的情况下具有最佳性能。我们还比较了有符号谱聚类和有符号Louvain算法在相关图上进行社区检测的性能，发现有符号谱聚类算法性能最好。我们将算法应用于洛杉矶地区的模拟地震地面运动数据。