Multi-objective unstructured meshes for improved mountain snow hydrology with the Canadian Hydrological Model (CHM)

Abstract:

The melt of seasonal snowcovers in cold regions provides downstream regions with a critical supply of freshwater, impacting ecosystems and human society such as agricultural, industrial, and municipal users. Late-lying snowpacks can persist into summer and can maintain stream flows through periods of low precipitation. The heterogeneity in these late-lying snowpacks stems from energetic differences (slope and aspect), mass transport (blowing snow and avalanching), and precipitation variability. Snowdrift-permitting scales of 1 m – 250 m are required to accurately simulate this snow cover and process heterogeneity. However, these spatial scales can be computationally intractable for large extents when using fixed resolution structured grids. Application of unstructured surface discretizations allow for large reductions in computational elements (70%+) while preserving critical land-surface heterogeneity. Here, an overview of the unstructured mesh generation approach in the Canadian Hydrological Model (CHM) is detailed. A case study using horizontal advection of snow (blowing snow) is shown that takes advantage of the multi-scale resolution. Lastly, recent developments in the distributed domain decomposition of an unstructured mesh across multiple MPI processes is detailed. This ensures limited domain communication and enables efficient global linear algebra solutions.