Studying the effects of microtopography on surface flow across spatial scales with the SERGHEI model

Microtopography has been recognised to have complex impacts in runoff generation and catchment connectivity. When studying small scale flows, microtopography evidently dominates the water flow paths and the inundation of the surface. However, as the spatial scale of observations increases, the effect of microtopography is mixed and diffused, and is difficult to isolate in catchment hydrological signatures. Numerical simulations that study microtopography must explicitly resolve the scales of microtopographic features, which are orders of magnitude smaller than the typical domains of interest, such as hillslopes and catchments. The high computational cost that usually constrains such simulations can be approached by means of High-Performance Computing. In this work, we present the model framework SERGHEI (Simulator of Environment, Rainfall, Geomorphology, Hydraulics and Ecology), a high-performance parallelised model based on the Kokkos programming framework. The Kokkos framework enables SERGHEI to run efficiently on heterogeneous systems and multiple graphics processor units (GPU). Hence, SERGHEI is suitable to address very large computational studies. Using SERGHEI, we carry out several numerical experiments of rainfall-runoff in idealized catchments with different shapes of microtopography. Here, we discretise the catchment at a very high spatial resolution which explicitly and completely resolves microtopography. In this contribution, we firstly present the implementation strategy of SERGHEI which enables us to run these numerical experiments in feasible time (minutes per simulation). Results of these experiments are compared both domain-wide and in spatially-distributed and multiscale manner, by assessing hydrological signatures and hydrodynamic distributions in subdomains of increasing size. Our findings suggest the existence of some threshold spatial scale at which the effects of microtopography may become averaged, and therefore robustly parametrisable into subgrid modelling approaches or simple ponding models. We present some possible strategies that would make use of such a parametrisation.