wrf_gpu
User's Guide › Reference › Credits, License & Citation

Credits, License & Citation

wrf_gpu exists because of the decades of open science behind the Weather Research and Forecasting (WRF) model. This page credits that work, explains the AI authorship of this rewrite, and covers licensing and how to cite the project.

WRF is developed and maintained by NCAR/UCAR and the WRF community. wrf_gpu is an independent, WRF-compatible re-implementation validated against WRF as an oracle — it is not WRF, not a Fortran-source port, and not affiliated with or endorsed by UCAR/NCAR. Please credit and cite the WRF team; see the links and the citation below.

Credit to WRF / NCAR / UCAR

The scientific model wrf_gpu reproduces — the Advanced Research WRF (ARW) dynamical core and its physics — was created by the WRF developers at the National Center for Atmospheric Research (NCAR), operated by the University Corporation for Atmospheric Research (UCAR), together with NOAA, NCEP and the broader WRF community. Their documentation is the model this guide is patterned after, and their users' guide is the reference every operational WRF user knows.

Naming

WRF® is a registered trademark of the University Corporation for Atmospheric Research (UCAR). Describe this project as WRF-compatible / WRF-derived, never as "WRF" itself. See LICENSE_NOTES.md in the repository.

AI authorship — written by GPT × Claude

wrf_gpu is, unusually, an AI-written scientific model. The code, the validation harnesses, the proof objects, and this documentation were authored by large language models under a manager-agent process:

  • an Anthropic Claude model (the manager and a primary implementer/reviewer), and
  • an OpenAI GPT model (a primary implementer and adversarial critic),

with a human director setting goals and gates. The mascots on the logo — a clay-orange robot (Claude) and a green robot (GPT) propping up a happy, GPU-powered storm cloud — are a nod to exactly that. Every substantive claim passes an independent AI critic before release, which is why the repository carries a proof object for essentially every statement. See How wrf_gpu Works → Agent-driven development.

What AI authorship means for trust

It means the project leans hard on falsifiable evidence rather than authority: sealed fixtures prove their dynamics/thermo fields cell-for-cell against WRF, and the open gaps (control/configuration parity, 24–72 h skill, terrain ceiling, multi-GPU throughput) are stated plainly rather than smoothed over. Treat it as a transparent research artifact you can audit, not a drop-in operational replacement for WRF.

License

  • wrf_gpu is released under the MIT License.
  • Files that translate, contain or derive from third-party code and data keep their upstream terms: AER's RRTMG/RRTM terms (these files may not be sold), NCAR's BSD 3-Clause notice for the MMM physics translations, and the UCAR public-domain notice for WRF-derived material. See THIRD_PARTY_NOTICES.md.
  • Third-party data (observation datasets such as station data, and any met_em / CPU-WRF reference outputs) is not redistributed here; you supply your own.

How to cite

If you use wrf_gpu, please cite both the underlying WRF model and this project.

Cite WRF

Skamarock, W. C., et al. (2019). A Description of the Advanced Research WRF
Model Version 4. NCAR Technical Note NCAR/TN-556+STR. doi:10.5065/1dfh-6p97

Cite wrf_gpu

Cite the repository and the release you used (tag + commit), for example:

wrf_gpu: a GPU-native, WRF-compatible regional weather model (v0.3.0).
https://github.com/wrf-gpu/wrf_gpu

A methods write-up describing the GPU rewrite and its validation accompanies the project; check the repository for the current citation and any DOI.

Contributing & source

wrf_gpu is a forkable, auditable artifact. The source, proof objects, ADRs, and contribution guide (CONTRIBUTING.md) are on GitHub: github.com/wrf-gpu/wrf_gpu.