E3SM plan for version 4
E3SM version 4, which is expected to be released in the beginning of 2028, will be the first coupled model release designed to run on exascale computers. This is made possible by rewriting the atmosphere and ocean components in C++ using the Kokkos library to provide performance portability across all DOE exascale computers as well as conventional CPU computers like E3SM’s own Chrysalis machine. The MALI land ice component is also exascale-enabled by C++/Kokkos. Other E3SM components (land and sea ice) won’t be able to take advantage of the GPUs on exascale computers until version 5 (but are smaller contributors to total run time).
Resolution
Access to exascale resources and shifting project focus towards shorter timescales allow us to increase E3SM’s lowest supported grid spacing to ~12 km in the atmosphere, ~10 km for the ocean and sea ice, and ~3 km over land regions. Access to bigger and faster GPU-enabled computers means we can make these changes while maintaining similar simulation speed and simulated-year capacity as we’re currently getting for the E3SMv3 high-resolution configuration. The E3SMv4 high-resolution configuration will have 3 km resolution in the atmosphere and 18 km to 2km resolution for the ocean, sea ice, and land components. This refinement will greatly improve E3SM’s representation of extreme weather events like tropical cyclones, atmospheric rivers, and mesoscale convective systems. Because weather impacts are often quite local in nature, increased resolution also makes E3SM more relevant and useful for resource management and energy infrastructure planning. The resolution in the ocean allows for mesoscale eddies to be resolved, removing the need for parameterizations that have been a major source of bias in previous versions of E3SM. Because version 4 won’t be able to handle all of the use cases of version 3 – biogeochemistry and human/Earth-system interactions in particular – the E3SM project will continue to use and support the version 3 model until the version 5 release.
E3SM AI Emulator
We envision routinely performing 50 year coupled simulations at this resolution. Large ensembles of much longer simulations will be possible by developing emulators to be released alongside all new versions of E3SM.
Grids
Another exciting feature of E3SMv4 is the use of a unified Model for Prediction Across Scales (MPAS) grid to cover all model components on the Earth’s surface. Avoiding the need for complex handling of cells that are partially land and partially ocean, this unified grid framework will allow seamless connectivity and consistent flux exchanges between the land, river, ocean, and sea ice model components. This approach eliminates interpolation errors inherent in multi-grid coupling systems, significantly improving the accuracy of freshwater routing and laying the groundwork for robust two-way coupling essential for simulating coastal processes. Beyond seamless coupling, the MPAS grid offers specific advantages for land and river modeling compared to traditional latitude-longitude grids. Its inherent support for variable resolution allows for computationally efficient grid refinement over specific areas of interest, such as key watersheds or coastal regions. Grid generation techniques allow observed river networks and key hydrological feature points (like dams or gauges) to be embedded directly into the model grid, ensuring a more geometrically accurate representation of the actual river system and facilitating model benchmarking. Furthermore, the MPAS grid avoids the severe cell distortion found in latitude-longitude grids at high latitudes, providing a more accurate representation of geographical features and potentially improving numerical stability in polar regions. This combination of consistent coupling and enhanced representational fidelity is crucial for advancing E3SM‘s ability to simulate the water cycle and its interactions across the Earth system.
E3SM Atmosphere Model (EAMxx)
While version 4 is still under active development, its atmosphere component (called EAMxx) is expected to include P3 microphysics, SHOC turbulence/cloud physics, similar gravity wave drag to the high-resolution version of E3SMv3. While the Zhang-McFarlane deep convection scheme is being ported to EAMxx, the model is being tested without a deep convection scheme as a candidate release configuration. The Modal Aerosol Model (MAM) used in previous E3SM releases has been ported to C++/Kokkos. This code (called MAMxx) is being tested and evaluated for possible inclusion in the E3SMv4 release, which will also include the Simple Prescribed Aerosol (SPA) capability already available in EAMxx.
E3SM Ocean Model (OMEGA)
In addition to switching languages and greatly accelerating the model on GPUs, the E3SMv4 ocean component (called OMEGA) has a completely redesigned model framework relative to the old MPAS model. This improves readability and interpretability for users and streamlines model configuration. In addition to computational and infrastructure improvements, OMEGA will also update the equation of state (TEOS-10) and move to a non-Boussinesq formulation. The latter development is critical for accurate simulation of sea level change. Use of TEOS-10 could make OMEGA truly energy conserving, a first for global ocean modeling. In addition to OMEGA, great strides have been made in better incorporation and acceleration of the ocean surface wave model WAVEWATCH III (WW3). Further sophistication in coupling E3SMv4 will include full coupled waves via a sea state and wave-sea ice interactions. In addition, WW3 in E3SMv4 will be the first hybrid ML wave model, via a new AI nonlinear source term parameterization. This parameterization has enabled dramatically increased physical fidelity without an increase in computational cost.
E3SM Sea Ice Model
The E3SMv4 sea ice model will add physical and biogeochemical improvements to coupling. The first significant change planned is a switch to variable thermohaline coupling between ocean and sea ice consistent with the OMEGA TEOS-10 equation of state. Instead of sea ice being assigned a fixed salinity from the perspective of the ocean model, mushy-layer physics in the MPAS-SeaIce column physics (Icepack) will continually drain salt after forming, offering a more realistic catalyst for ocean overturning. This change also addresses a coupling instability in E3SMv3 HR without artificially large marine runoff flux spreading. Two changes to sea ice morphology are planned: first, a time-varying floe size distribution will evolve during the attenuation of surface ocean waves by sea ice fracture whenever waves are switched on. Second, we also plan to incorporate variable density sea ice from macroporous ridging pending trials of the computational expense of this new scheme in Icepack. Improvements to the radiative properties of sea ice are in the development and testing stages including dust deposition, application of hexagonal scattering in snow on sea ice following success in the E3SMv3 Southern Ocean Regionally Refined Mesh (SORRM) configuration over land snow, and potential expansion to more spectral bands commensurate with RRTMG beyond just the visible and near-infrared split.
E3SM Land Ice Model
The largest polar change to occur in E3SMv4, however, will be introduction of a dynamical Greenland ice sheet. The MALI Greenland ice sheet model will be coupled through the E3SM Land Model (ELM), which is also being enhanced to include a deep firn model that represents the evolution of snow to firn over Greenland and Antarctica. Direct iceberg calving at the edge of Greenland has been finalized, and long control simulations are about to commence to confirm stability and computational economy of MALI-inclusive B-case configurations. MALI is GPU enabled and is expected to take a smaller fraction of total run time than the sea ice model.
E3SM Land Model (ELM)
Following the successful use of prognostic vegetation structure to replace static remote sensing-based datasets in the v3 physics campaigns, this dynamic vegetation configuration will be the default for all v4 simulations. Based on recent success in developing an AI method to dramatically accelerate the land model spinup for prognostic vegetation (and associated soil components), this AI method will become the default land spinup approach for all v4 simulations. A new capability allowing subgrid transport of surface water across topographic subdomains has been tested at the global scale and is expected to be turned on for the v4 campaign. A new representation for lakes and wetlands will be implemented in v4, eliminating the adverse impact of excessive land evaporation on land-ocean fluxes. The FATES model which brings prognostic vegetation biogeography (shifting area distribution of different vegetation types) is being evaluated at the global scale and is expected to be introduced in v5.
E3SMv4 Science Targets
The high-resolution coupled model envisioned for E3SMv4, along with its AI emulator, will enable a wide range of scientific endeavors to support DOE’s energy mission as well as hypothesis-driven research to deepen our understanding of the Earth system. For example, the much higher resolution of E3SMv4 compared to other Earth system models – including previous versions of E3SM – will greatly improve our ability to capture different types of storms and their precipitation characteristics. This will have great benefit because floods and storms contribute to a large proportion to natural hazards in the U.S. and worldwide, leading to challenges for managing energy supply, demand, and delivery and planning for energy infrastructure development. E3SMv4 will also be better equipped to simulate different flood generation mechanisms, including coastal flooding caused by storm surge and pluvial and fluvial processes over land due to its higher resolution and unified land/ocean/river mesh. Numerical experiments and initialized hindcasts and predictions will be performed to understand the large-scale and regional drivers of floods and storms, their subseasonal-to-decadal variability, and their local-to-regional impacts on the U.S. energy and water sectors. Modeling land and land-atmosphere coupling at high resolution will also open an opportunity to better understand the roles of vegetation in the surface environments. E3SMv4 will be used to perform experiments to test hypotheses of the coupled vegetation-environment and the impacts of vegetation on hazards such as heat waves, aridity, drought, and wildfire, with significant implications for society and the economy. Lastly, polar processes play key roles in mid-latitude circulations and environments and dominate the uncertainty in future sea levels, affecting coastal populations and energy infrastructure. Leveraging the high resolution that contributes to improved modeling of sea ice, land ice, and their coupling with the ocean, E3SMv4 will be used to evaluate the pathways and underlying processes of rapid sea level changes and examine the spatial pattern and seasonality of Arctic sea ice loss and the implications for the U.S. energy sector.
The E3SM team is proud to be at the forefront of Earth system modeling innovation, driving advancements that will not only enhance scientific understanding but also provide critical insights for energy infrastructure planning and resource management. As development progresses, E3SM version 4 promises to set a new benchmark for precision, scalability, and relevance in addressing the complex challenges of a changing Earth system.
This article is a part of the E3SM “Floating Points” Newsletter, to read the full Newsletter check:
