C-NEMS: an open-source continuation of EIA’s Project BlueSky
We went public with this announcement today and wanted to post here, since this community can play an active role in helping to shape the effort. We really appreciated everyone’s feedback during Project BlueSky!
Background
The U.S. Energy Information Administration has run the National Energy Modeling System (NEMS) for over 30 years. It is used to produce the Annual Energy Outlook, and most US-focused energy models calibrate to it in some way. Each sector in NEMS is represented with the most relevant mathematical structure for its market dynamics, and the sectors are coupled and solved as a system that reaches equilibrium rather than posed as a single optimization problem. While continuously updated, NEMS has become a patchwork of accumulated legacy code, which has become difficult to maintain and extend.
In 2022 EIA launched Project BlueSky to design a successor and released a proof-of-concept prototype in early 2025 under Apache 2.0 (EIAgov/BlueSky). When things got chaotic in the U.S. federal government last year, EIA was faced with tough decisions and decided to pause further development. Some of us who worked on it there, along with new partners, have restarted the effort outside government as Community NEMS (C-NEMS).
Overall Goal
A long-term aspirational goal is that C-NEMS could eventually be picked up by EIA as a successor to NEMS. There is no commitment on their side, but building toward that possibility shapes our approach:
-
A model able to produce detailed sectoral representations similar to NEMS along with outputs that track the Annual Energy Outlook.
-
Permissive licensing (Apache 2.0), so adoption carries no copyleft complications for a federal agency.
-
Clear documentation and code standards aimed at handoff to a team that did not write the code. Python was chosen as the modeling language of choice since it is most widely used.
-
Data provenance traceable to publicly available sources.
Ultimately, C-NEMS is being built with particular institutional requirements in mind. That said, we want to leverage the collective expertise of the openmod community.
Design approach
Here are some of the key features, which will be familiar to this community:
-
Heterogeneous governing dynamics across sectors, coupled into a multi-sector equilibrium. This is the part that differs most from open energy modeling frameworks we are familiar with, where the entire system is defined as a single optimization problem.
-
A solution approach for that coupled system that is substantially faster than NEMS in the prototype. We are currently testing formulations that could reduce run times further.
-
Nonlinear formulations to capture endogenous tradeoffs and human behavior: endogenous oil and gas supply, consumer choice in demand modules, endogenous technological learning.
-
Sensitivity analysis and uncertainty quantification supported natively rather than through a handful of deterministic scenarios.
-
Documented, automated pipelines from raw public data to model inputs, in a separate repository.
Where we expect overlap
We are aware this is still a crowded landscape and that several of the ideas above are not new here. Please include pointers to current or prior work we should be building on rather than developing ourselves.
Status
The electricity module is furthest along, a natural gas market module is next, and the data pipeline work is underway. It is not ready for production analysis, or computational testing. The motivation and design thinking are in DeCarolis et al. (2025).
What would be most useful
Feedback of the coupling and solution approach, particularly from anyone who has attempted similar multi-sector formulations. Inputs on the data pipeline and its connection to the model are also welcome
C-NEMS is under active development. Check out our C-NEMS GitHub page here. The cnems-models repository houses the source code associated with each module. The cnems-inputs repository defines the data pipelines developed by Catalyst Cooperative that map raw data to model input files.
Sign up for email updates, open an issue, contribute a pull request, or reach out to info@c-nems.org if you’d like to help shape where it goes next.
Development is currently supported by Arnold Ventures and the William and Flora Hewlett Foundation. Contributors are at Carnegie Mellon, American University, Ohio State, Catalyst Cooperative and Western Spark, with an advisory board of current and former energy modellers and analysts.