Community Earth System Model 3 (CESM3): Plans, progress, timelines

Sep 25, 2024

Information on the science and infrastructure updates and timelines for the forthcoming CESM3 release. CESM3, which includes scientific advances in all components, is intended to be the model that is used for the CMIP7 DECK and Fast Track simulations.

Latest update: September 25, 2024

The CESM development team is working towards a scientific release of version 3 of the Community Earth System Model (CESM3). CESM3 will contain numerous science and infrastructure improvements and is the model configuration intended for the simulations requested in the CMIP7 DECK and Fast Track.

The timeline for a public release, which will include code and scientifically supported configurations and model output for most DECK simulation (PI Control, Historical, and several standard idealized simulations), is targeted for summer 2025, presuming relatively smooth tuning / problem solving progress over the latter part of 2024. As the release date firms up, we will update the timeline here.

The component models that will be included in CESM3 are shown in Fig. 1. And, a detailed list of the changes that are slated for inclusion in each of the component models of CESM3 are listed in Table 1 below. Note that the list may not be fully complete and that some planned developments may not make it into the final default model configuration.

The aim of the CESM Project is for the released version to be the same model version that is used for CMIP7, which means that the timing of the release will be dictated to some degree by the availability of updated CMIP7 forcings. \ The intention is to use the Spectral Element dynamical core with a nominal 1deg resolution for atmosphere and land and an ~80km model top (mid-top, resolving the stratosphere and lower mesophere) and the MOM6 ocean model with ~2/3 deg resolution. CO2 emissions-driven mode is intended to be the default configuration for historical and scenario simulations.

Those who would like to track progress and issues that the CESM development team are working on are welcome to review the CESM3 Development Tracking github page. You can also follow along on development activities at the CESM code github page. For those interested in approximate computational costs of different CESM3 configurations for the purpose of proposal development, you can find timing estimates on Derecho here.

Table 1: Changes across CESM component models between CESM2 and CESM3

Component	Change	Reason
Ocean	Transition to MOM6 dycore Eulerian z-coordinate to Arbitrary Lagrangian‐Eulerian (ALE) algorithm B-grid to C-grid Linear free surface to nonlinear free surface More accurate advection schemes Non-Boussinesq dynamics (option)	The development of the POP model was discontinued. MOM6 ALE algorithm has many advantages, such as elimination of vertical CFL condition, flexible choices of vertical coordinates, reduction of numerical (spurious) mixing, wetting/drying (vanishing layers). C -grid provides better wave propagation at higher resolutions than POP B-grid. It also allows for single-point channels. Nonlinear free surface allows much higher near-surface vertical resolution (~ 1m) than could be achieved with POP (~ 10 m). More accurate and truly monotonic advection schemes, for both tracers and momentum, to better represent transport processes. The non-Boussinesq option properly represents thermosteric sea level.
Ocean	Ocean bathymetry representation. POP had step-like bathymetry while MOM6 has a continuous bathymetry MOM6 allows for sub-grid-scale channels	Better representation of topography-flow interactions and transport through narrow channels, such as Gibraltar.
Ocean	Surface and bottom boundary conditions Virtual salt flux in POP (constant volume) changed to the natural boundary condition in MOM6 (variable mass, enthalpy conserving) Inclusion of geothermal heating in MOM6 Single surface layer frazil ice formation in POP, while in MOM6 frazil formation is vertically distributed (whole water column)	Natural BC on freshwater is the correct physically based BC. It provides the correct closure for the hydrological cycle, including sea level rise from glacial melt. The non-Boussinesq option properly represents thermosteric sea level change. There is also an enthalpy flux associated with mass exchanges that were neglected in POP improving energy conservation across the system. Geothermal heating is a significant forcing for the circulation in the abyssal ocean. Depth-dependent frazil formation provides a more physically-based representation of ice-ocean interactions.
Ocean	Lateral sub-grid-scale closures Numerical reformulation of GM/Redi and lateral eddy mixing to accommodate ALE coordinates Change from stratification dependent mixing coefficients to mesoscale eddy kinetic based schemes Changed from harmonic anisotropic viscosity in POP to biharmonic enstrophy cascade (Leith) closure with option to energy backscatter (option) in MOM6	Leveraging developments in the multi-institutional Ocean Energy and Transport Climate Process Team to better represent mesoscale eddy transport processes. These new closures are scale-adaptive across resolutions.
Ocean	Vertical sub-grid-scale closures Major reformulation of the KPP boundary layer scheme, now called FPMix Communication of spectral wave information between ocean and wave model	More accurate representation of influences of wind waves and swell on near-surface momentum and tracer mixing.
Ocean	Grid resolution Dipole grid in POP to tripole grid in MOM6 1.125 degree in POP to ⅔ degree in MOM6 (workhorse) 60 layers (10 m thick surface layer) in POP to 65-75 layers (2.5 m thick near the surface) in MOM6 A more complete hierarchy of horizontal resolutions in MOM6 (from 2 degree to 1/36 degree)	Better representation of the mean flow and wave propagation in the workhorse resolution. Better aspect ratio in the horizontal grids (more numerically accurate). Better representation of surface boundary layer, air-sea interaction, and biogeochemical processes.
Ocean	Regional one-way nesting capability	POP did not have open boundary conditions, so could not be used for regional simulations. Regional ocean simulations opens up the possibility of additional actionable science within CESM.
Ocean	Parameterizations in POP currently lacking in MOM6 Estuary Box Model (EBM) Overflows	Currently seeking funding to adapt the EBM to MOM6. Unclear if overflow parameterization is needed in MOM6 when using an isopycnal coordinate.
Ocean	Output and post-processing More flexible output streams and variable specification in MOM6 than in POP	More user-friendly selection of output variables, their frequency and file streams.
Sea Ice	Transition to CICE6	CICE6 is the development sea ice model being updated by the CICE Consortium. The CICE6 code has been significantly reconfigured and refactored such that the dynamics and thermodynamics have been separated. CICE6 now calls Icepack, which is responsible for calculating column physics.
Sea Ice	Updates to sea ice physics (e.g. snow on sea ice, melt ponds, landfast ice)	Needed improved physics for better simulation of sea ice evolution and for applications work.
Sea Ice	Implementation of a Floe Size Distribution (FSD)	Implementation of a joint distribution of ice thickness and floe size. Floe sizes are important in physical processes including new ice formation, welding of floes, lateral growth and melt, and fracture of floes by ocean surface waves. Representing the FSD has been shown to strongly impact the sea ice concentration and seasonal processes driving sea ice evolution.
Land	New surface datasets & dataset generation with ESMF (cime) infrastructure	Improved / updated source datasets available; Enables higher resolutions and new features (e.g. transient urban capabilities and shifting cultivation)
Land	Land-atmosphere exchange: Biomass heat storage, Surface roughness, and Dust emissions.	Improved calculations of biophysical fluxes and dust emissions sources.
Land	Improved cryosphere representation: Snow physics Snow thermal properties Excess ice	More realistic and physical representations of snow-aerosol-radiation interactions and thermal properties improves the CLM simulated snowpack evolution and soil temperatures in many regions. Excess ice reduces the rate of arctic soil warming and allows for projections of ground subsidence that may occur with permafrost thaw.
Land	Hydrologic developments including: Representative hillslope hydrology (option) Irrigation improvements	The hillslope hydrology model captures aspect and elevation effects on energy and water balance and will be an optional feature for land-only simulations. The irrigation improvements consider irrigation sources (unconfined vs. confined aquifers) and application methods (sprinkler vs drip).
Land / River	Development of mizuRoute river model (option)	This vector-based river model allows for more flexibility in routing schemes and catchment definition. Moreover it’s being developed in-house, which will allow us to replace RTM and Mosart river models in CESM with mizuRoute. mizuRoute will be an optional feature for land-only simulations in CESM3.
Land	Ecosystems and biogeochemistry: Arctic & boreal phenology and allocation Implementation of matrix solver for semi-analytical spin-up (SASU) FATES ecosystem demography model, including land use and land cover change (option) Microbial-explicit soil biogeochemical model (MIMICS) (option)	Work here was aimed at improving regional biases in plant productivity and survival. With SASU we aim to accelerate the spin up of ecosystem carbon and nitrogen pools to facilitate more rapid model development and parameter calibration. Ongoing developments with FATES and MIMICS aim to provide more ecologically realistic simulations of biotic response to climate change, both above- and below-ground. FATES developments include satellite and no-competition configurations that we will be able to run in transient simulations over historical and SSP scenarios. FATES and MIMICS will be optional features for land-only simulations.
Land	Crop Model: Bioenergy crop development Tillage and residue removal Crop phenology improvements Updated planting and harvest dates	CLM was the only land model with an explicit crop model in CMIP6 archive. Developments here provide expanded functionality and are aimed at addressing known biases and limitations in our representation of crop growth and management.
Land	Urban Model: Transient urban datasets Explicit air-conditioning adoption	CLM was also the only land model with an explicit urban model in the CMIP archive. Developments here provide transient urban capabilities and consideration of air-conditioning adoption.
Land	Model infrastructure: Parameter perturbation experiment (PPE) Streamlined workflow for single point simulations with flux tower data (NEON and PLUMBER) New compsets, including noAnthro and anomaly forcings Development of a simple land model (SLIM) a new CESM component. Reduced complexity or “fast modes” for NWP	Development of a sparse grid and application of SASU spinup provides opportunities for machine learning approaches to calibrate land model parameters in ways that are new for CESM3. Additional infrastructure and tutorials were built for running single point simulations with data from flux tower sites on HPC, containers, and in the cloud. New compsets allow for paleo simulations and idealized land simulations while SLIM allows for simplified representation of land features in CESM. Finally, reduced complexity modes with reduced land unit and PFT numbers allow for faster solutions for application in high resolution and NWP applications.
Atmosphere	Changed dynamical core from FV (used for CAM4,5,6) to spectral-elements (SE)	Scalability (strong scaling) More accurate transport Variable and high resolution capability More advanced thermodynamics Horizontal molecular diffusion/thermal conductivity operators for WACCM-x FV not being actively developed (over 20 years old)
Atmosphere	Switched from MG to PUMAS microphysics code base (incl. several science changes)	Fix bug likely causing large climate sensitivity in CESM2 Add missing processes, adding hydrometeor types
Atmosphere	Updated L-scale CLUBB code with prognostic momentum transport	More consistent with the rest of CLUBB physics (other moments are prognosed so more consistent to prognose momentum fluxes as well)
Atmosphere	Convective gustiness parameterization	Missing process for source of latent heat flux (particularly over ocean in the tropics at low wind speeds) Improve tropical variability (e.g. MJO)
Atmosphere	New radiation code base (RRTMG-P)	Modernize radiations code (with GPU support) and improve radiation algorithms
Atmosphere	New gravity wave drag parameterization [not in code base yet]	Missing drag sources to improve high-top stratospheric wind biases
Atmosphere	Topography software improvements (e.g. command line interface, not smoothing PHIS over ocean, support for ESMF file format,...)	Ease of use for community Improve ocean-atmosphere coupling
Atmosphere	Re-order physics	To alleviate spurious surface wind oscillations CLUBB call was moved to after coupler call.
Atmosphere	CAM physics support for z-based dynamical core MPAS	Provide non-hydrostatic capability in CAM and support SIMA/EarthWorks
Atmosphere	Deep convection scheme (ZM) modification: alternative convective parcel calculation	Adapt ZM for higher boundary layer resolution
Atmosphere	Richarson number based free-atmosphere mixing	CLUBB does not appear to provide enough free -atmosphere mixing hence a Richardson-number based vertical mixing scheme from CAM4 has been enabled where CLUBB is not active Improves stability of model
Atmos chemistry	GHG chemistry updates	Improved SOA and aerosol description, coupling with biogenic SOA precursors
Atmos chemistry	OASISS	Online DMS emissions based on Online Air-Sea Interface for Soluble Species
Atmos chemistry	Improved chemical mechanisms TS2,3, SLH chemistry	Different science questions required chemical mechanisms of different complexity. Air quality studies require more tropospheric chemistry, short-lived halogens are important for improved climate and ozone
Atmos chemistry	Sectional Aerosol Model CARMA in CESM2	Option to use the more comprehensive sectional aerosol model CARMA as an alternative to the standard modal aerosol model.
Atmos chemistry	GEOS-Chem	An alternative chemistry module; increase collaborations between GEOS-Chem and CESM communities.
Atmos chemistry	HEMCO emissions	Allows online regridding and incorporating regional inventories into global inventories.
Atmos chemistry	CAMchem MPAS	Non-hydrostatic dynamical core to allow simulations at <5km horizontal resolution
Atmos chemistry	TUV-x	Improve the representation of photochemical decomposition of reactive species generating different radicals (O¹D, OH, Cl/Br/I atoms) as well as the penetration of downwelling radiation to the lower troposphere. Also improvements for the treatment of aerosols.
Whole Atmosphere	Inline TUV-x	Improved representation of photolysis and heating rates.
Whole Atmosphere	SE dynamical core for WACCM and WACCM-X	Maintain consistency with CAM. Enables high resolution simulations.
Land Ice	Multiple ice sheet evolution capability.	Allow simultaneous evolution of more than 1 land ice grid (e.g. Greenland and Antarctica).
Land Ice	Ice-ocean interaction	Two-way field exchanges between the ocean and land-ice components.
Land Ice	Data GLC option	Streamlined coupling with CTSM and NUOPC.
Land Ice	CISM physics improvements: Basal sliding scheme Iceberg calving scheme	Improving the realism of ice sheet simulations.
Land Ice	Support for dynamic simulations of mountain glaciers in CISM.	Study water availability in ESM.
Paleoclimate	Use paleo to inform model development	Perform simulations of ice-age and hothouse climates that could be used for model assessment and tuning
Software Engineering	Software engineering support for most of the above items	Most science changes also require software engineering contributions; we are not calling those out separately here.
Software Engineering	Migration of driver/coupler infrastructure to use the Earth System Modeling Framework (ESMF). Key pieces of this include: New driver/mediator: The Community Mediator for Earth Prediction Systems (CMEPS) New data models: The Community Data Models for Earth Prediction Systems (CDEPS) Simplified process and tool for creating land surface datasets leveraging ESMF remapping	There were many reasons for making this change; key among these are: Interoperability and collaboration with other modeling systems based on ESMF, such as NOAA’s Unified Forecast System (UFS), which also uses CMEPS and CDEPS The use of ESMF allows runtime remapping between component grids, making it much easier to introduce new grids into CESM ESMF regridding allows new capabilities like the exchange grid and dynamic masking (which supports ice sheet coupling) The driver run sequence can now be viewed and modified as a simple text file ESMF gives all of the functionality of the previously-used Model Coupling Toolkit (MCT) and adds new features useful to CESM like ESMF-managed threading (which allows different components to use different numbers of threads)
Software engineering	Extensive work to support ultra-high resolution (3.75km) simulations: required changes in many components as well as ESMF for the sake of memory scalability, performance, etc.	There is interest in ultra-high resolution for various projects, but pushing the system to this limit exposed memory and performance bottlenecks that we have needed to address.
Software engineering	Better support for new grids, including: Out-of-the-box regionally-refined grids Easier mechanisms for incorporating user-defined grids (leveraging ESMF’s runtime remapping)	It is becoming much more common for users to do simulations on custom grids, such as regionally-refined grids, so it was important for this use case to be easier.
Software engineering	Support for cloud / container platforms	Providing CESM via a Docker container that can be run on the cloud or other systems allows users to get CESM running quickly without needing their own computing resources and porting effort. This has been especially useful for tutorials, automated testing and supporting local development on laptops / desktops.
Software engineering	Various improvements to I/O performance via the ParallelIO (PIO) library	I/O performance can be a bottleneck, particularly at high resolutions. We have made improvements such as optimizing the ordering of variables being written and introducing initial capabilities for asynchronous I/O to improve performance.
Software engineering	Split CIME into its functional pieces, keeping CIME itself as mainly the Python-based Case Control System.	In CESM2, CIME included not only the Python-based Case Control System but also various Fortran-based infrastructure, such as the driver, coupler and share code. We have split these pieces into separate repositories to facilitate continued collaboration on the pieces shared with E3SM (the Case Control System) while providing separate control over the unshared pieces.
Software engineering	New tool for managing the versions of components and externals in a given CESM version: git-fleximod	The tool we had developed for managing versions of components and externals in CESM2, manage_externals, was becoming difficult to maintain and had unneeded complexity because all of our repositories are now in git (rather than subversion). We developed a replacement tool, git-fleximod, which has similar functionality but is much simpler due to leveraging recent git features, and so will be easier to maintain moving forward.