This paper presents two-stage clustering approach for accurate analysis of earthquake catalogs where aim is to categorize events in terms of aftershock (AF) clusters or independent backgrounds (BGs). In stage I, the Gaussian kernel-based temporal density estimation is used for grouping of events based on their occurrence time. From the graph, local peak (maxima), local minima and their timing information are utilized to group the events into significant time zones. In stage II, on events of each time zone, coordinate and magnitude information is combined together (weighted mechanism) to determine effective local weighted spatial density (\(\rho ^{\mathrm{w}}\)). Based on \(\rho ^{\mathrm{w}}\) and event distance (\(\delta\)), a decision graph is drawn to find out the spatial cluster centroids for each time zone. Event’s assignment to the centroid is carried out based on its nearest neighbor of higher density. Outliers (non-clustered) are also detected in stage II which is considered as independent backgrounds. The experimental analysis is carried out on historical seismicity of California, Himalaya, Japan and Sumatra–Andaman region. The results indicate that obtained AFs and total number of events follow a similar cumulative and \(\lambda\) rate, whereas BGs have linear cumulative and consistent \(\lambda\) rate. It is also observed that AFs and total events have similar ergodic behavior, quantified from the inverse TM metric plot. The competitive performance of the proposed approach is obtained over state-of-the-art declustering methods.

本文提出了一种两阶段的聚类方法，用于对地震目录进行精确分析，其目的是根据余震（AF）簇或独立背景（BG）对事件进行分类。在阶段I中，基于高斯核的时间密度估计用于基于事件的发生时间对事件进行分组。从图中可以看出，局部峰值（最大值），局部最小值及其时间信息可用于将事件分为重要的时区。在阶段II中，在每个时区的事件中，将坐标和量级信息组合在一起（加权机制），以确定有效的局部加权空间密度（\（\ rho ^ {\ mathrm {w}} \））。基于\（\ rho ^ {\ mathrm {w}} \）和事件距离（\（\ delta \）），绘制决策图以找出每个时区的空间聚类质心。将事件分配给质心是基于其最近的密度较高的邻居。在阶段II中也检测到异常值（非聚集），该阶段被视为独立背景。对加利福尼亚，喜马拉雅山，日本和苏门答腊－安达曼地区的历史地震活动进行了实验分析。结果表明，获得的AF和事件总数遵循相似的累积和\（\ lambda \）速率，而BG具有线性累积且一致的\（\ lambda \）率。还可以观察到，AF和总事件具有类似的遍历行为，从逆TM度量图中可以量化。相比于最新的去簇方法，该方法具有竞争优势。