Spatiotemporal clustering of great earthquakes on a transform fault controlled by geometry

Minor changes in geometry along the length of mature strike-slip faults may act as conditional barriers to earthquake rupture, terminating some and allowing others to pass. This hypothesis remains largely untested because palaeoearthquake data that constrain spatial and temporal patterns of fault rupture are generally imprecise. Here we develop palaeoearthquake event data that encompass the last 20 major-to-great earthquakes along approximately 320 km of the Alpine Fault in New Zealand with sufficient temporal resolution and spatial coverage to reveal along-strike patterns of rupture extent. The palaeoearthquake record shows that earthquake terminations tend to cluster in time near minor along-strike changes in geometry. These terminations limit the length to which rupture can grow and produce two modes of earthquake behaviour characterized by phases of major (Mw 7–8) and great (Mw > 8) earthquakes. Physics-based simulations of seismic cycles closely resemble our observations when parameterized with realistic fault geometry. Switching between the rupture modes emerges due to heterogeneous stress states that evolve over multiple seismic cycles in response to along-strike differences in geometry. These geometric complexities exert a first-order control on rupture behaviour that is not currently accounted for in fault-source models for seismic hazard. The rupture mode between major and great earthquakes is controlled by transform fault geometry, according to simulations of a reconstructed record of 20 palaeoearthquakes along the Alpine Fault, New Zealand.