The resolution of Earth system models is increasingly being towards the kilometer scale (k-scale). However, multiple technical challenges remain in performing global simulations using the E3SM Land Model (ELM) at the k-scale that will contain ~221 million grid cells. First, the existing raw surface parameters for running k-scale ELM simulations will be derived from coarse resolution and outdated datasets. Second, the ELM surface dataset file at k-scale that naively covers the entire globe, when 70% of grid cells are ocean, will be unnecessarily large at ~6 TB. Third, the default configuration of computing the map between the atmosphere and land component on the fly when the two components use the same grid incurs a high initialization cost. Fourth, ELM will only have 1 plant functional type (PFT) in each grid cell at k-scale, and ELM’s default approach of allocating 17 PFTs per grid cell will require excess memory and result in restart files that are excessively large.
In this work, we summarize the progress made to address the challenges mentioned above in performing k-scale ELM simulations. First, a new set of global land surface parameters with a resolution of 1 km for multiple years from 2001 to 2020, utilizing the latest and most accurate available datasets, has been developed under the Integrated Coastal Modeling Project (ICoM). ELM k-scale simulations performed under ICoM over the contiguous United States demonstrate that land surface parameters contribute significantly to the spatial heterogeneity of simulated soil moisture, latent heat, emitted longwave radiation, and absorbed shortwave radiation. Next, a set of idealized ELM simulations were performed on the Perlmutter with domains containing 1 million, 10 million, and 100 million grid cells. A 91% reduction in file size is achieved if the ELM surface dataset contains only active land grid cells and assumes 1 PFT per grid cell. The use of a mapping file for exchanging data between the atmosphere and land model significantly reduces model initialization time. Additionally, the use of 1 PFT per grid cell achieves a 10% reduction in simulation memory usage and a 70% reduction in the size of the ELM restart file. The developments summarized in this study are necessary first steps towards performing global k-scale ELM simulations.
