Seismic and geodetic observations show that slow slip events (SSEs) in subduction zones can happen at all temporal and spatial scales and propagate at various velocities. Observation of rapid tremor reversals (RTRs) indicates back-propagating fronts traveling much faster than the main rupture front. Heterogeneity of fault properties, such as fault roughness, is a ubiquitous feature often invoked to explain this complex behavior, but how roughness affects SSEs is poorly understood. Here we use quasi-dynamic seismic cycle simulations to model SSEs on a rough fault, using normal stress perturbations as a proxy for roughness and assuming rate-and-state friction, with strengthening behavior at high slip rate. SSEs exhibit temporal clustering, large variations in rupture length and propagation speed, and back-propagating fronts at different scales. We identify a mechanism for back propagation: as ruptures propagate through low-normal stress regions, a rapid increase in slip velocity combined with rate-strengthening friction induces stress oscillations at the rupture tip, and the subsequent ”delayed stress drop’ induces secondary back-propagating fronts. Moreover, on rough faults with fractal elevation profiles, the transition from pulse to crack can also lead to the re-rupture of SSEs due to local variations in the level of heterogeneity. Our study provides a possible mechanism for the complex evolution of SSEs inferred from geophysical observations and its link to fault roughness.

The speed-up of glaciers following ice shelf collapse can accelerate ice mass loss dramatically. Investigating the deformation of landfast sea ice enables studying its resistive (buttressing) stresses and mechanisms driving ice collapse. Here, we apply offset tracking to Sentinel-1 A/B synthetic aperture radar (SAR) data to obtain a 2014-2022 time-series of horizontal velocity and strain rate fields of landfast ice filling the embayment formerly covered by the Larsen B Ice Shelf, Antarctic Peninsula until 2002. The landfast ice disintegrated in 2022, and we find that it was precipitated by a few large opening rifts. Upstream glaciers did not accelerate after the collapse, which implies little buttressing effect from landfast ice, a conclusion supported by the near-zero correlation between glacier velocity and landfast ice area. Our observations suggest that buttressing stresses are unlikely to be recovered by landfast sea ice over sub-decadal timescales following the collapse of an ice shelf.