Extended Data Fig. 10: Homogeneous vs heterogeneous segmentation RSF model results.

Comparison of unstable slip segmentation for models with a. homogeneous shear modulus and effective normal stress; b. heterogeneous shear modulus derived from FWI; c. heterogeneous effective normal stress from FWI. Upper panels show the mechanical parameters used in each model. For all models, d_c is constant for the entire fault. The b-a value listed refers to the peak velocity-weakening value of a curve otherwise identical to that of Extended Data Fig. 9. The middle panels show the logarithm of normalized slip velocity (slip velocity / plate velocity) with time and distance from the trench for multiple cycles of each model. The lower panels show the percentage of total cumulative slip accumulated at velocities faster than, slower than, and equal to plate velocity over all model cycles. These plots help distinguish persistent slip segmentation from temporary spontaneous slip segmentation. In a, spontaneous segmentation can be seen in the middle panel at portions of the fault (indicated by arrows) where unstable slip fronts tend to arrest over multiple cycles. This type of slip segmentation, which has been reported in previous rate-state friction models of slow-slip^39,48, is considered ‘spontaneous’ because it arises spontaneously from stress and slip interactions on a homogeneous fault, rather than at transitions in geometric, mechanical or loading conditions. The lower panel of a shows that this segmentation has little effect on the cumulative distribution of fast slip on the fault over hundreds of cycles, implying that segment boundaries are temporary and migrate over many cycles. In contrast, the segment boundaries in b-c are long-lived, ‘persistent’ barriers to unstable slip that influence the spatial distribution of the long-term accumulation of fast slip. These segment boundaries occur at sharp gradients in shear modulus b and effective normal stress c. Note that the FWI-derived shear moduli b and effective normal stresses c promote longer-recurrence transient slip events of ~25 years b and ~10-15 years c relative to the shorter-term slip cycles of ~2 years in the homogeneous case a; however, these differences in recurrence interval are most likely caused by the different values of b-a between the three models, which were selected to highlight segmentation patterns of slow-slip events between the three cases.