International audienceMass wasting caused by large-magnitude earthquakes chokes mountain rivers with several cubic kilometres of sediment. The timescale and mechanisms by which rivers evacuate small to gigantic landslide deposits are poorly known, but are critical for predicting post-seismic geomorphic hazards, interpreting the signature of earthquakes in sedimentary archives and deciphering the coupling between erosion and tectonics. Here, we use a new 2D hydro-sedimentary evolution model to demonstrate that river self-organization into a narrower alluvial channel overlying the bedrock valley dramatically increases sediment transport capacity and reduces export time of gigantic landslides by orders of magnitude compared with existing theory. Predicted export times obey a universal non-linear relationship of landslide volume and pre-landslide valley transport capacity. Upscaling these results to realistic populations of landslides shows that removing half of the total coarse sediment volume introduced by large earthquakes in the fluvial network would typically take 5 to 25 years in various tectonically active mountain belts, with little impact of earthquake magnitude and climate. Dynamic alluvial channel narrowing is therefore a key, previously unrecognized mechanism by which mountain rivers rapidly digest extreme events and maintain their capacity to incise uplifted rocks

The “precipiton” method is a particle‐based approach that consists of routing elementary water volumes on top of topography with erosive and depositional actions. Here we present an original way to calculate both river depth and velocity from a method that remains embedded in the precipiton framework. The method solves the governing equations for water depth, where the water depth is increased by a constant quantity at each precipiton passage and decreased by a value based on a flow resistance equation. The precipitons are then routed downstream on top of the resulting water surface. The method is applicable even if the precipitons are routed one by one (i.e., independent of each other), which makes it simple to implement and computationally fast. Compared to grid‐based methods, this particle method is not subject to the classic drying‐wetting issue, and allows for a straightforward implementation of sediment transfer functions between the river bed and running water. We have applied the method to different cases (channel flow, flow over topographic barriers, and flood prediction on high‐resolution lidar topography). In all cases, the method is capable of solving the shallow water equations, neglecting inertia. When coupled with erosion and sediment transport equations, the model is able to reproduce both straight and braided patterns with geometries independent of grid size. Application of the model in the context of multithread rivers gives new insight into the development of braiding instability.