EarthWorks: Creating a Storm-Resolving Community Model

October 10, 2024

Dave Randall

Hosted by Eric Maloney

Abstract

A large team of CSU and NCAR scientists has been working since the summer of 2020 to create a global storm-resolving model (GSRM) based on the Community Earth System Model (CESM). The model, now called CESM-EarthWorks or CESM-E for short, uses the Model for Prediction Across Scales (MPAS) atmospheric dynamical core developed by the Mesoscale and Microscope Meteorology Laboratory of the National Center for Atmospheric Research, the MPAS-Ocean and sea ice models developed at the Los Alamos National Laboratory, the Community Land Model (CLM), and the physical parameterizations of the Community Atmosphere Model (CAM). The dynamical cores of the atmosphere and ocean models are non-hydrostatic. Most of the computations can be performed on either CPUs or GPUs.

The components of the model are coupled together using the CMEPS mediator, which makes use of the Earth System Modeling Framework (ESMF). All components use the same geodesic grid. The target horizontal grid spacing, which has now been reached, is 3.75 km. The model can also be run with coarser and, in principle, finer horizontal grids. The MPAS dynamical cores also allow local grid refinement by stretching. The target vertical grid, which can now be used, includes 93 levels, with a top near 80 km above the surface of the Earth.

We are currently planning a DYAMOND (data ocean) simulation with the 3.75-km horizontal grid and 93 levels. We are also conducting a short coupled simulation with the same horizontal grid and a lower model top.

All GSRMs need massive computing resources, with many processors and a lot of memory. Large volumes of input and output data are required. Analysis of the large output datasets is also a challenge. For EarthWorks, an additional complication is that the model had to be implemented using the CESM infrastructure with minimal changes. Because CESM was not designed for use with kilometer-scale grids, many software issues were encountered along the way. These had to be diagnosed,and fixed in order to reach our goals.

In this talk I will outline the model development process, with some discussion of the problems that we have overcome, and present some recent results.