Comparison of seismicity declustering methods using a probabilistic measure of clustering

We present a new measure of earthquake clustering and explore its use for comparing the performance of three different declustering methods. The advantage of this new clustering measure over existing techniques is that it can be used for non-Poissonian background seismicity and, in particular, to compare the results of declustering algorithms where different background models are used. We use our approach to study inter-event times between successive earthquakes using earthquake catalog data from Japan and southern California. A measure of the extent of clustering is introduced by comparing the inter-event time distributions of the background seismicity to that of the whole observed seismicity. Theoretical aspects of the clustering measure are then discussed with respect to the Poissonian and Weibull models for the background inter-event time distribution. In the case of a Poissonian background, the obtained clustering measure shows a decrease followed by an increase, defining a V-shaped trend, which can be explained by the presence of short- and long-range correlation in the inter-event time series. Three previously proposed declustering methods (i.e., the methods of Gardner and Knopoff, Reasenberg, and Zhuang et al.) are used to obtain an approximation of the residual “background” inter-event time distribution in order to apply our clustering measure to real seismicity. The clustering measure is then estimated for different values of magnitude cutoffs and time periods, taking into account the completeness of each catalog. Plots of the clustering measure are presented as clustering attenuation curves (CACs), showing how the correlation decreases when inter-event times increase. The CACs demonstrate strong clustering at short inter-event time ranges and weak clustering at long time ranges. When the algorithm of Gardner and Knopoff is used, the CACs show strong correlation with a weak background at the short inter-event time ranges. The fit of the CACs using the Poissonian background model is successful at short and intermediate inter-event time ranges, but deviates at long ranges. The observed deviation shows that the residual catalog obtained after declustering remains non-Poissonian at long time ranges. The apparent background fraction can be estimated directly from the CAC fit. The CACs using the algorithms of Reasenberg and Zhuang et al. show a relatively similar behavior, with a time correlation decreasing more rapidly than the CACs of Gardner and Knopoff for shorter time ranges. This study offers a novel approach for the study of different types of clustering produced as a result of various hypotheses used to account for different backgrounds.