Weather Research and Forecasting Model
by Carol Rasmussen
For the past couple of years, a group of scientists from the research and forecasting communities has been taking the first steps on the long road to a new kind of model--one designed for all of them. Joseph Klemp of NCAR's Mesoscale and Microscale Meteorology Division (MMM) reported on the proposed weather research and forecasting (WRF, pronounced "warf") model at the U.S. Weather Research Program's conference this spring.
Why develop a new model now? Klemp says, "Certainly numerical weather prediction models have been evolving over the past 30 years and have been improving dramatically in their capabilities. However, we still have a long way to go." Existing models cannot fully use some of the new, small-scale observational data that are now available. Model treatments of some physical processes have proved to be inadequate. Perhaps most important, though, is the chance to combine expertise from both types of modelers. Klemp explains, "The model's purpose is both to improve our understanding and prediction of mesoscale weather and to promote closer ties between the research and operational forecasting communities. It isn't going to be your model or my model, it's going to be our model."
Participants come from four organizations: MMM, NOAA's National Centers for Environmental Prediction (NCEP), NOAA's Forecast Systems Laboratory (FSL), and the University of Oklahoma's (OU) Center for Analysis and Prediction of Storms (CAPS). "Each of the organizations involved supports a state-of-the-art model, but they all recognize the need for new and better ways of doing things," says Klemp. "If we pool our resources we'll end up with a better outcome." Scientists from other institutions are also contributing, including Lou Wicker at FSL, Bob Wilhelmson at the University of Illinois, and Dave Dempsey at San Francisco State University.
The participants anticipate that:
- MMM will maintain the model, freely distribute it to and support its use by the community, and migrate its own research to the new model. The new model will take the place of the NCAR/Pennsylvania State MM5 model.
- FSL will gradually transfer its development efforts from its RUC model to the WRF model, which will replace the RUC model when it achieves operational status.
- NCEP will implement the WRF model as a high-resolution nest within its operational regional forecast model (Eta) and consider it as a replacement for its regional model, based on the merits of its performance.
- CAPS will conduct fundamental studies in small-scale data assimilation and parameter retrieval appropriate for use in the new model.
What will it look like?
The model developers have outlined their priorities for the WRF model. It should be accurate and efficient over a broad range of scales, from cloud-sized to synoptic scale. However, design emphasis is on a high-resolution model with a horizontal grid of one to ten kilometers. "If we're forced to make compromises between coarse resolution and fine resolution, we'll put the emphasis on fine," says Klemp. The model will have improved physics, especially cloud microphysics and representations of convection and turbulence. Data assimilation will be improved, with an emphasis on making better use of small-scale observations. The model will adhere to FORTRAN 90 standards and use standard interfaces for its physics so that scientists can "plug and play" their own physics packages within the model. The scientists envision developing components in steps so that individual parts can be evaluated without slowing the overall process. Finally, Klemp emphasizes, "We are not starting from scratch. We will adapt the best features of existing models and develop new techniques where deficiencies are identified."
The developers are currently wrestling with "numerical model issues," Klemp reports, such as the prognostic equation set, grid staggering, vertical coordinates, terrain representation, and time integration. They are testing alternative approaches to some of these problems on idealized simulations "where we know what the correct answer is."
The groups have also made progress in developing the overall software framework for the model so that it can run efficiently on various types of parallel computers, such as distributed memory and shared memory. "There's a driver layer on top that figures out how to work with the machine, so that the model code that scientists work with remains as simple as possible," Klemp explains.
Nine WRF working groups have been created to formulate detailed plans and begin the development work in their respective areas:
Although the collaborators are poised to continue the testing and decision-making needed to develop the model, ongoing funding is still in question. Klemp hopes that a full-physics, research-quality model will be available within two or three years and that the model will be operational in about five years. Achieving this time line, however, will be contingent on establishing a nucleus of support for each of the four organizations participating in the project.
- Dynamical model framework, led by Bill Skamarock (MMM)
- Software architecture, led by John Michalakes (MMM)
- Standard model initialization procedures, led by Stan Benjamin (FSL)
- Data assimilation system, led by Jim Purser (NCEP)
- Model physics, led by John Brown (FSL)
- Testing, verification, real-time implementation, led by Kelvin Droegemeier (OU)
- Postprocessing software, led by Henry Neeman (OU)
- Community model support, led by Jimy Dudhia (MMM)
- Operational implementation, led by Geoff DiMego (NCEP)
For further information about the model, contact Klemp at 303-497-8902 or
In this issue...
Other issues of UCAR Quarterly
Edited by Carol Rasmussen,
Prepared for the Web by Jacque Marshall
Last revised: Tue Apr 4 15:10:57 MDT 2000