Time-dependent forecast of large earthquakes from physics-informed probabilistic approach

The elastic energy that fuels large earthquakes accumulates heterogeneously along faults, resulting in complex earthquake occurrence patterns. Although earthquake cycle simulations help capture such complexity in seismic hazard models, their high computational cost prevents widespread use and uncertainty quantification. Here, we propose a physics-based probabilistic method to forecast the timing and magnitude of large earthquakes. The method combines fracture mechanics theory and observational constraints on seismic coupling, historical seismicity and fracture energy to forecast the arrest locations of long ruptures that have saturated the seismogenic depth. We apply this approach to estimate the time-dependent probability of earthquakes exceeding M8.5 across the entire Chilean subduction zone. Sensitivity analysis shows the most critical model parameter relates to the scaling between fracture energy and slip. We constrain the model parameters by matching the time between mega-earthquakes on the Valdivia segment, validate the constrained model based on the seismicity of other segments, then apply it prospectively over the whole Chile megathrust. Our results highlight how earthquake potential on a given megathrust segment can be altered by earthquakes on neighboring segments, which reduce the energy available for multi-segment ruptures. The proposed physics-informed framework constitutes a new approach for time-dependent seismic hazard analysis on large faults.