Wrapping up a Decade of Development

February 16, 2025

Over the course of 2024, E3SM celebrated its decadal anniversary through “Decade of Development” stories in its newsletter, Floating Points.

This story celebrates E3SM by remembering and presenting its timeline. The project began in 2014 with a bold 10-year vision to model the earth system on exascale machines, achieving this milestone in 2023 with the Gordon Bell Prize. In 2018 the project released E3SMv1, addressing key questions on the water cycle, biogeochemistry, and cryosphere. Subsequent phases released E3SMv2 in 2021 and SCREAMv1 in 2023, a GPU-enabled model that broke performance records – modeling more than 1 simulated year per day (SYPD). E3SMv3 (released in 2024), reduces biases and introduces human-Earth system feedbacks. Looking ahead, the team is developing a new decadal vision, targeting 2027 for E3SMv4, a GPU-enabled, fully coupled model foundational for Earth system digital twins from local, regional to global scales. Future plans continues to support DOE’s energy mission via integrating advanced technologies like AI and machine learning to accelerate ground breaking scientific discoveries, exert leadership in performance-portable software for DOE next-generation architectures, and delivering E3SMv5 and E3SMv6 by 2033.

This article celebrates E3SM’s decade of progress through key achievements. Nearly 1.5 petabytes of native simulation data and over 70,000 CMIP6 datasets have been published. Over 400 journal papers have been cited 13,000+ times, with an h-index of 63. The open-source project on GitHub boasts 67,000+ commits, 3,500+ merged pull requests, and 300+ stars, with millions of lines of code contributed by 50+ monthly contributors. Led by Lawrence Livermore National Laboratory, E3SM involves 8 DOE labs and academic partners, with 117 staff members in December 2023. The website has drawn over 13,000 users globally since April 2023, reflecting E3SM’s broad impact.

This story reflects on the key lessons that emerged after a decade of E3SM development. Science simulation groups emphasized incorporating “sanity checks” to catch errors early, conservative planning with the willingness to drop/delay features to avoid overall science goal delays, and operationalizing workflows to minimize mistakes. Infrastructure teams highlighted the importance of version control tools like Git/GitHub for collaborative development, pull requests for code review, and comprehensive overnight testing. The coupled model group stressed best practices for reproducibility, extended coupled simulations, cross-group collaboration, and automated diagnostics. Project management focused on open, transparent practices, including detailed documentation, open meetings, agile planning with long- and short-term roadmaps, and bi-weekly progress tracking. These lessons have streamlined workflows, improved productivity, and ensured robust development processes.

This article explores the E3SM evolution over the past decade in computational resources, code review processes, and organizational structure. Launched in 2014, the project is nearing its 10-year goal of creating exascale-capable models, including GPU-enabled EAMxx and OMEGA, with v4 set for 2027. Early challenges with limited DOE supercomputer access were mitigated by dedicated clusters like Anvil, Compy, and Chrysalis. Code review processes matured from informal checks to a formalized system in 2019, with 2022 updates requiring comprehensive feature testing. Organizationally, E3SM transitioned through three phases: Phase 1 emphasized shared leadership, Phase 2 focused on long-term goals with streamlined leadership, and Phase 3 re-aligned with national priorities, renaming groups to emphasize science goals.

This article is a part of the E3SM “Floating Points” Newsletter, to read the full Newsletter check: