Topological Comparison Between the Stochastic and the Nearest‐Neighbor Earthquake Declustering Methods Through Network Analysis

On short timescales, earthquakes cluster in both time and space, eventually complicating the analysis of seismicity. One basic goal is to partition the earthquake catalogue into two classes of events — background events, regarded as spontaneous or independent earthquakes, and clustered events, including events triggered by other earthquakes. It is generally supposed that the background events represent a region’s long-term spatiotemporal behaviour of seismicity, while the identification of earthquake clusters is important to understand and forecast the evolution of seismicity on short time scales.
However, as noted by LML External Fellow Jiancang Zhuang and colleagues in a new paper, there is no commonly agreed method for separating earthquake clusters from each other or from background seismicity. Existing clustering algorithms often identify earthquake clusters in different ways and, consequently, provide different declustered versions of an earthquake catalogue. To get new insights on this issue, Zhuang and colleagues use tree graph representations and tools from network analysis to compare the performance of two specific clustering methods – one based on nearest-neighbour distances and another on a stochastic point process. Both data-driven methods can be satisfactorily applied to decompose the seismic catalogue into background seismicity and sequences of clustered earthquakes, and also allow studying the internal structure of the identified sequences, since they provide the connections between the events within each cluster.
The authors found that the two declustering algorithms produce similar partitions of the earthquake catalogue, although they may differ in the internal structure outlined for individual clusters. In particular, the nearest-neighbour method tends to provide simpler structures than stochastic declustering method. The authors demonstrate that these features can be usefully quantified with centrality measures widely used in network analysis, and argue that these tools will likely be of increasing value in earthquakes clustering analysis.
The paper is available at

Leave a Reply

Your email address will not be published. Required fields are marked *