A new algorithm to find earthquake clusters using neighboring cell connection and rate analysis.

Abstract:

To study earthquake interaction, it is important to objectively find a group of earthquakes occurred closely in space and time. Earthquake clusters are chosen with previous techniques that characterize them as mainshock-aftershock sequences or swarm sequences by empirical laws (e.g., Omori-Utsu; ETAS) or direct assumptions about physical processes such as stress transfer, transient stress loading, and fluid migration. Recent papers instead proposed non-parameterized techniques such as a kernel-based smoothing method. The cumulative rate clustering method (CURATE, Jacobs et al., 2013) is one of the approaches without any direct assumptions. The CURATE method was applied in New Zealand and provided a good result for selecting the swarm sequences comparing with the ETAS model. However, it is still difficult to choose a proper confined area and a time interval for extracting sequences from the catalog. To avoid arbitrariness in space and time parameters, here we propose a new method modifying the CURATE approach. We first identify the spatial clusters by looking into the spatial distribution in a 2-D cell-gridded map. The spatial clusters defined as multiple neighboring cells, each of which contains at least one earthquake in a time period T. From the selected spatial clusters, we then evaluate temporal clustering which is defined as a transient increase of seismicity rate comparing to the rate before the target event. We tested this method focusing on shallow crustal seismicity, northern Honshu, Japan. We chose the parameter range from T = 1 to 100 days and cell size = 0.01°to 0.1°. As a result, the number of the clusters increase with longer T and larger cell size. By choosing the T = 30 days and cell size = 0.05°, we successfully selected the long-lasting aftershock sequences associated with the 2004 M6.8 Chuetsu and 2007 M6.8 Chuetsu-oki earthquakes, while other empirical models and CURATE method failed to decluster.