Summary The Sichuan Basin in China has experienced a number of devasting earthquakes in the past 20 years, particularly on the Longmen Shan fault (LMSFZ) with the 2008 Wenchan and 2013 Luschan events. This study employs a hierarchical, four-dimensional (latitude, longitude, depth, time) clustering framework to characterize seismic activity in the Sichuan Basin. After the identification of spatial features of the region (e.g., faults), we then apply two cluster algorithms on the Longmen Shan fault data and compare the identification of the 2008 and 2013 events. In particular, we apply and compare Density-Based Spatial Clustering of Applications with Noise (DBSCAN) and Bayesian Gaussian Mixture Model (BGMM) on the identification of mainshock-aftershock sequence (as well as any foreshock events). By applying temporal clustering to the dataset and comparing DBSCAN and BGMM methods, we find distinct differences between the results. Specifically, we find that DBSCAN identifies a simple mainshock-aftershock sequence, while BGMM produces a more complex foreshock-mainshock-aftershock sequence. However, both scenarios have been identified within previous work on these events, highlighting that additional analysis is required and that single cluster algorithms should be applied with caution. The work here in comparing machine learning techniques within an integrated clustering framework is timely and will serve as a guide for more in-depth analysis on earthquake patterns and fault dynamics using these methods.
Wan et al. (Wed,) studied this question.