Written and maintained by CASRAI Editorial Board
Last updated
WRF (the Weather Research and Forecasting model) is an open mesoscale numerical weather prediction system developed for atmospheric research and operational forecasting. This guide treats it as research software: where it comes from, what goes in and out of a run, how it is obtained, and what to cite so a published result can be reproduced. It is written for researchers who use WRF output or plan to run it; it does not rank forecast models.
What is WRF modelling?
WRF modelling means running the Weather Research and Forecasting model to simulate the atmosphere over a limited region, either from observed or analyzed conditions (a real-data case) or from idealized conditions (an idealized case). NCAR’s Mesoscale and Microscale Meteorology Laboratory describes WRF as a state-of-the-art mesoscale numerical weather prediction system serving both atmospheric research and operational forecasting (UCAR MMM, WRF model page, read 2026-11-07).
The dynamical solver most researchers mean by “WRF” is the Advanced Research WRF (ARW). According to the WRF user guide (version 4.4, chapter 5), ARW is a fully compressible, nonhydrostatic model with a run-time hydrostatic option, uses an Arakawa C-grid, and integrates with second- or third-order Runge-Kutta time schemes and second- to sixth-order advection options. Since version 4.0 the default vertical coordinate is hybrid; setting hybrid_opt=0 reverts to the terrain-following coordinate.
Who develops WRF
UCAR states that development began in the late 1990s as a collaboration among NCAR, NOAA (NCEP and ESRL), the US Air Force, the Naval Research Laboratory, the University of Oklahoma and the FAA. Support runs through the WRF user forum at forum.mmm.ucar.edu. For wider context on the field, see What is meteorology and What is atmospheric science.
Components and the real-data workflow
The user guide describes a fixed sequence. WRF Preprocessing System (WPS) output feeds the initialization program, which feeds the model.
| Step | Program or file | What the user guide says |
|---|---|---|
| 1 | WPS output met_em* |
Real-data input produced by WPS |
| 2 | namelist.input |
Run settings are edited here, including the physics suite ('CONUS' or 'tropical') |
| 3 | ./real.exe |
Creates wrfinput_d0* and wrfbdy_d01; success is indicated by “real_em: SUCCESS EM_REAL INIT” |
| 4 | ./wrf.exe |
Runs the model, writes wrfout_d0* (and restart files wrfrst*); success is indicated by “wrf: SUCCESS COMPLETE WRF” |
Other executables are ideal.exe (idealized cases), ndown.exe (one-way nesting) and tc.exe. Two-way nesting uses feedback=1. The code is built with ./configure then ./compile; netCDF is the only mandatory library, and build options are serial, shared-memory (smpar), distributed-memory (dmpar) and the combined dm+sm. The guide’s troubleshooting advice for CFL errors is to reduce the time step or adjust w_damping.
Inputs, in plain terms
A real-data run needs gridded analysis or forecast fields processed by WPS, which is why WRF is a regional tool that depends on boundary data from a larger-scale model. Global systems such as ECMWF and GFS are different kinds of systems, not rival regional configurations of the same thing.
Download, registration and licence
UCAR’s source-download page says download requires registration, which also subscribes the registrant to the WRF news list, and asks users to read the WRF Public Domain Notice before downloading. Since version 4.0, releases are published on the public WRF-Model GitHub release page. The UCAR WRF page shows version 4.8.0 as the latest at the time of reading; check the GitHub release page for the current release before you start, and record the exact version you use.
How to cite WRF
NCAR’s citing page gives the following (read 2026-11-07):
- Version 4: Skamarock, W. C., Klemp, J. B., Dudhia, J., Gill, D. O., Liu, Z., Berner, J., Wang, W., Powers, J. G., Duda, M. G., Barker, D. M., and Huang, X.-Y. (2019). A Description of the Advanced Research WRF Version 4. NCAR Technical Note NCAR/TN-556+STR, 145 pp. doi:10.5065/1dfh-6p97.
- Version 3: Skamarock et al. (2008), NCAR/TN-475+STR, 113 pp., doi:10.5065/D68S4MVH.
- Version 2: Skamarock et al. (2005), NCAR/TN-468+STR, 88 pp., doi:10.5065/D6DZ069T.
- Software itself: the WRF-ARW model DOI 10.5065/D6MK6B4K is a persistent link to the WRF information page, citable alongside the technical note.
- Physics schemes: NCAR directs users to a separate WRF physics references page for the citation of each scheme.
For reproducibility, a methods section should also state the exact WRF version, the physics options, the domain and nesting set-up, the driving data, and the namelist.input (deposited as supplementary material). See computational reproducibility, FAIR4RS and reproducibility infrastructure; a DOI makes the software and the technical note findable.
WRF-Chem
WRF-Chem, the chemistry-enabled configuration, is listed separately on the UCAR download page (via the NOAA ruc.noaa.gov WRF-Chem pages), and some WRF releases have no matching WRF-Chem release. Check version compatibility before combining them. Its own user guide is the reference for it.
Strengths and limits
- Strength: run on real or idealized conditions, with nesting and selectable physics, in a documented, openly distributed code base with a user forum.
- Strength: a stable, citable reference for each major version.
- Limit: a real-data run needs externally supplied input fields from WPS, and the build and namelist set-up take expertise.
- Limit: no accuracy comparison with other models is made on this page, because none of the primary sources cited here makes one.
Alternatives and neighbouring systems
| System | Role relative to WRF |
|---|---|
| ECMWF | A global forecasting centre; its products are a different kind of system from a regional WRF configuration |
| GFS | A global model from NOAA, commonly the type of larger-scale data a regional run is driven by |
FAQ
What is WRF modelling?
Running the Weather Research and Forecasting model to simulate the atmosphere over a region from real or idealized conditions, for research or forecasting (UCAR MMM).
What is the most accurate weather model?
No single answer is given here: accuracy depends on region, variable, lead time and configuration, and the NCAR sources describe WRF’s design, not a ranking. Look for a published verification study for your region and variable.
Which is better, ECMWF or GFS?
They are global systems from different organisations, and this guide does not rank them. Read verification results for your use case; for WRF, either can serve as source data for boundary conditions.
Will AI replace meteorology?
This page makes no prediction. WRF is a physics-based numerical model, and what its documentation covers is that approach.
Is WRF free to download?
UCAR’s page requires registration and refers users to the WRF Public Domain Notice; releases since 4.0 are on GitHub.
Sources
- UCAR NCAR MMM, WRF model page (read 2026-11-07)
- WRF source download page (read 2026-11-07)
- WRF user guide v4.4, chapter 5 (read 2026-11-07)
- Citing WRF (read 2026-11-07)
