Triggering of large earthquakes is driven by their twins

We present a rigorous pseudo-prospective testing framework to evaluate four generalizations of the ETAS (epidemic-type aftershock sequence) model that all feature space-varying background events rates. We demonstrate that the model, with (i) a magnitude-dependent Omori law exponent and (ii) a modified Gutenberg-Richter magnitude distribution that links triggered and triggering earthquake magnitudes, overperforms strongly. These two properties (i) and (ii) have been previously documented empirically and derived from general physical and scaling reasonings. One of the novel consequences of this new superior model is that, above magnitude 6.3 (in California), earthquakes are triggered mainly by other earthquakes of similar magnitudes. In contrast, small non-observable events mostly trigger smaller events. This spectacular feature is further validated by the rationalization of earthquake doublets, which are shown to emerge from the magnitude correlations inherent in the best model, and are not the result of some arbitrary selection bias. Our results fundamentally question the existing paradigm that the smallest earthquakes dominate triggering whatever the magnitude of the target events.