Testing stress shadowing effects at the South American subduction zone

The seismic gap hypothesis assumes that a characteristic earthquake is followed by a long period with a reduced occurrence probability for the next large event on the same fault segment, as a consequence of the induced stress shadow. The gap model is commonly accepted by geologists and is often used for time-dependent seismic hazard estimations. However, systematic and rigorous tests to verify the seismic gap model have often failed so far, which might be partially related to limited data and too tight model assumptions. In this study, we relax the assumption of a characteristic size and location of repeating earthquakes and analyse one of the best available data sets, namely the historical record of major earthquakes along a 3000 km long linear segment of the South American subduction zone. To test whether a stress shadow effect is observable, we compiled a comprehensive catalogue of mega-thrust earthquakes along this plate boundary from 1520 to 2015 containing 174 earthquakes with M-w > 6.5. In our new testing approach, we analyse the time span between an earthquake and the last event that ruptured the epicentre location, where we consider the impact of the uncertainties of epicentres and rupture extensions. Assuming uniform boundary conditions along the trench, we compare the distribution of these recurrence times with simple recurrence models. We find that the distribution is in all cases almost exponentially distributed corresponding to a random (Poissonian) process; despite some tendencies for clustering of the M-w >= 7 events and a weak quasi-periodicity of the M-w >= 8 earthquakes, respectively. To verify whether the absence of a clear stress shadow signal is related to physical assumptions or data uncertainties, we perform simulations of a physics-based stochastic earthquake model considering rate and state-dependent earthquake nucleation, which are adapted to the observations with regard to the number of events, spatial extend, size distribution and involved uncertainties. Our simulations show that the catalogue uncertainties lead to a significant blurring of the theoretically peaked distribution, but the distribution would be still distinguishable from the observed one forM(w) >= 7 events. However, considering the stress transfer to adjacent fault segments and heterogeneous instead of constant stress drop within the rupture zone can explain the observed recurrence time distribution. We conclude that simplified recurrence models, ignoring the complexity of the underlying physical process, cannot be applied for forecasting the M-w >= 7 earthquake occurrence at this plate boundary.