Hyper-resolution global hydrological modelling using HPC: what's new and what's keeping us?

Abstract:

Global hydrological models (GHMs) have proven to be valuable tools to assess the impacts of climate and socioeconomic change on global water resources and to identify hotspots of water stress and groundwater depletion. Although many GHMs still operate at the relatively coarse resolution of 0.5 degrees (~50 km), a number of models (e.g., Lisflood, mHM, PCR-GLOBWB, WaterGAP) have global versions that operate 5 arc-minutes (~10km). However, the next step towards hyper-resolution global modelling, i.e., at 30 arc-second (~ 1km) resolution, is severely hampered by the computational burden associated with GHMs. This is partly due to the fact that, contrary to fully-implicit PDE-type model codes such as ParFlow and HydroGeoSphere, GHMs have been slow in adopting massive parallelization and high-performance computing (HPC) to speed up calculations. In this presentation I will provide an overview of where current and future global hydrological models could benefit from HPC techniques: surface hydrology, groundwater, human water use and surface water routing. I will provide examples of work done so far (e.g. PCR-GLOBWB, HydroBlocks, mHM) and identify the main hurdles for efficient scaling, especially in case of routing. Apart from the challenge of speeding up GHM-codes, I will also address the parameterization of hyper-resolution GHMs, which is highly problematic, in particular of human water demand and water use. I will end by pointing out that HPC could help in resolving the parameterization challenge.