Achieving Computationally Trackable Modelling of Erosion in the Alps using the Instructed Glacier Model (IGM)

Initially, landscape evolution models (LEMs) were derived by observing the landscape and its change through cartography metrics. Since then, the techniques have shifted to more process-based numerical methods and have been used to reconstruct a wide range of physical landforms through computer simulations. One such application is reconstructing the Alps. However, as we seek to model the landscape evolution within a topographically driven region of the Alps, we require high-order ice dynamics to accurately capture the underlying physics. As such, existing landscape evolution models, ones that incorporate both glacial and fluvial erosion, are incapable of modelling millions of years of erosion at a high spatial resolution due to the computational cost. However, by extending the efficient physics-driven AI iceflow model, the Instructed Glacier Model (IGM), to also act as a LEM, we are able to reduce the computational load by 1-2 orders of magnitude. As IGM only replaces the iceflow solver, this then means we can incorporate existing state-of-the-art process-based erosion models within the geomorphology literature, including but not limited to abrasion, quarrying, fluvial, and hillslope processes. We then seek to show how powerful this new model is by validating it on existing benchmark papers across the aforementioned processes while also simultaneously reducing the computational load. As such, this allows one to do landscape evolution modelling over millions of years at a high spatial resolution, presenting itself as a potential option to model the entire Quaternary period, and possibly beyond. Finally, this exposes new research applications that rely on ensemble approaches as well as inverse techniques as the computational demand for doing such simulations is now feasible.