• WRF Model prototypes for integrating the dynamical equations
• WRF computational framework
• WRF Model physics
• WRF experimental real-time forecasting
• WRF case-study evaluation and testing
• WRF model data assimilation
• WRF community release and related activities
• WRF project management

Related website: http://wrf-model.org

The overall goal of the WRF Model project is to develop a next-generation mesoscale forecast model and data assimilation system that will advance both the understanding and prediction of mesoscale weather, and promote closer ties between the research and operational forecasting communities. The model is being developed as a collaborative effort among the NCAR MMM Division, NCEP's Environmental Modeling Center (EMC), the NOAA/Forecast System Laboratory's Forecast Research Division (FRD), the Department of Defense's Air Force Weather Agency (AFWA), the Center for the Analysis and Prediction of Storms (CAPS) at the University of Oklahoma, and the Federal Aviation Administration (FAA), along with the participation of a number of university scientists. Primary funding for MMM participation in WRF is provided by the NSF/USWRP, AFWA, FAA and the DoD High Performance Modernization Office. With this model, researchers seek to improve the forecast accuracy of significant weather features across scales ranging from cloud to synoptic, with priority emphasis on horizontal grids of 1-10 kilometers. The model will incorporate advanced numerics and data assimilation techniques, a capability for multiple, relocatable nests, and improved physics, particularly for the treatment of convection and mesoscale precipitation systems. It will be well-suited for a range of applications, from idealized research to operational forecasting, and it will have the flexibility to accommodate future enhancements.

The WRF model has a host of desirable characteristics. It is designed to be modular, and a single source code will be maintained that can be configured for both research and operations. It will be state-of-the-art, transportable, and efficient in a massively parallel computing environment (accommodating vector architectures as well). Data assimilation systems and adjoint and tangent linear forms (for 3DVAR analysis and 4DVAR assimilation) will be developed in tandem with the model itself. Numerous physics options are being offered, thus tapping into the experience of the broad modeling community. It will be maintained and supported as a community mesoscale model to facilitate wide use in research, particularly in the university community. Research advances will have a direct path to operations. With these hallmarks, the WRF model is unique in the history of numerical weather prediction in the U.S.

The WRF system has advanced substantially in the past year. This advancement has been facilitated by real-time forecasting, continued system case testing and analysis, the development of its data assimilation system, and a new release of WRF for community evaluation. When WRF becomes sufficiently mature to be used operationally, it is expected to (1) replace the Meso-Eta model for the operational Threats forecasts at NCEP, (2) replace the MM5 model for operational use by AFWA, and (3) take on the function of rapid updating, now served by the RUC model. The following sections illustrate the progress and highlights in the various areas of WRF development.

Joseph Klemp and Skamarock continued their research to improve the accuracy of numerical simulations over complex terrain. In collaboration with Oliver Fuhrer (Swiss Federal Institute of Technology, Zurich), they documented the presence of artificial disturbances that can arise in flow over small-scale terrain, demonstrated that these errors are contained in the linear system of equations, and explained their behavior through analytic solutions to the steady-state, finite-difference equations. Their analysis documents that these errors arise if the order of accuracy in computing the metric terms associated with the terrain following coordinates is not the same as the accuracy used for the horizontal advection. They extended their analysis to explain how these distortions also arise in semi-Lagrangian models and how they can be eliminated. This work led to improvement of the numerics in the WRF Eulerian prototypes, and also the Canadian MC2 Model.

Prototype dynamical cores continue to be developed within the WRF framework, to facilitate the comparative evaluation of the relative accuracy and efficiency of Klemp and Skamarock's numerical techniques, in a controlled computational environment. Two completed prototypes are the split-explicit Eulerian models based on mass and height vertical coordinates; both are available as run-time selectable cores within the WRF model framework and both have been developed at NCAR. A semi-implicit semi-Lagrangian prototype is being developed in an effort led by Jim Purser (NOAA/NCEP).

A hybrid coordinate model formulation was examined by Zuwen He (University of Miami) with assistance from Skamarock, Stan Benjamin, (NOAA/FSL) and Rainer Bleck (Los Alamos National Laboratory). Also, the nonhydrostatic model, developed by Zavisa Janjic (NOAA/NCEP) from the operational hydrostatic Eta model, is being placed into the WRF framework to allow comparison of the NCEP model with the existing WRF cores.

During the past year, the Eulerian, terrain-following mass coordinate, split-explicit, flux-form prototype has been chosen as the official WRF core to be supported for community use. This decision followed extensive testing of both the mass and height vertical-coordinate WRF cores. Several months of daily NWP forecasts, and idealized test simulations covering a broad range of scales, including simulations of synoptic-scale baroclinic waves in a periodic channel with a 100-km horizontal grid and a supercell thunderstorm evolution on a 1-km grid, were used to evaluate the cores. Compared to the height coordinate prototype core, the mass coordinate core has a more realistic upper boundary condition (specified pressure as opposed to a rigid lid) and reverts to the standard terrain following sigma-p hydrostatic formulation with a simple switch in the model.

The Eulerian mass and height coordinate WRF cores use the 3rd order timesplit Runge-Kutta scheme (RK3) developed by Lou Wicker (NOAA/National Severe Storms Laboratory) and Skamarock. This scheme allows the use of both upwind (odd ordered) and centered (even ordered) high-order flux-divergence operators, exhibits low dispersion errors, and allows a larger time step than the traditional leapfrog schemes when used with either odd or even higher-order advection operators.

The Runge-Kutta scheme has proven to be both robust and efficient. For example, using daily operational forecasts produced at NCAR and NOAA/FSL, Mike Baldwin (NOAA/NSSL) and Matt Wandishin (University of Arizona) produced power spectra for precipitation, and compared it with observationally-derived spectra and spectra derived from the operational Eta model. They found that the WRF model reproduced the observed spectra much better than even the much higher resolution operational Eta model (see Fig. 16) lending further evidence that high order numerical methods should be employed in high resolution NWP models.

Figure 16. Three-hour accumulated precipitation (top) from a 4 km observational analysis (radar and rain gage) and the WRF and operational Eta models from 12 Z 4 June 2002 runs, valid at 18 Z. The spectra for these precipitation fields (bottom) show that the 10 km WRF model forecast maintains the variance in the precipitation field down to at least 4 times the grid spacing (approximately 40 km). By comparison, the variance for the operational Eta model drops off much sooner, at greater than 10 times the grid spacing. The observational analysis, Eta output and spectral analysis were provided by Mike Baldwin and Matt Wandishin of NOAA/NSSL.

In another study using the height coordinate, Tetsuya Takemi (Osaka University, Japan) and Rotunno used idealized squall-line simulations to calibrate the turbulent kinetic energy (TKE) mixing parameterization in the WRF model. They found that the high order methods used in the WRF core resulted in significantly more resolution of the energy spectra, with the retention of more power in the shorter scales, compared to existing leapfrog-based models (see Fig. 17). Numerical filters that excessively damp the small-scale eddies and mask the TKE mixing generally have been needed in cloud model formulations. The high-order methods used in the WRF models removes the need for these non-physically-based filters and results in energy removal occurring predominantly within the physically-based TKE parameterization, in addition to better resolution of the tail of the energy spectra.

Figure 17. Spectra from squall-line simulations using the WRF model and also the WRF model configured with filtering parameterizations necessary in other existing cloud models. The extension of the spectra to higher wave numbers demonstrates the improved numerics and filtering characteristics of the WRF model formulation.

WRF is designed to serve a diverse community of users, from researchers to operational NWP centers, and in many cases includes the introduction of local modifications to the code. WRF's software is designed to be flexible, maintainable, extensible, efficient, and portable to a range of high-performance computing platforms. WRF software supports efficient and reasonably rapid software development, and promotes and facilitates code reuse, thereby leveraging the investment in scientific and infrastructure development by other groups and contributing to the pool of such software. The software infrastructure is called the WRF Advanced Software Framework (ASF), the foundation for the WRF model itself (now in release 2.1.1 (April 2002)) and for the WRF 3DVAR system.

Development this past year was focused on computational and I/O performance enhancement, new dynamical core development, nesting, and model coupling. Figure 18 shows WRF model performance exceeding .1 Teraflop/s on 512 processors of the Terascale Computing System at the Pittsburgh Supercomputing Center. It also shows WRF performance on 256 processors, using the new IBM Blue Sky system at NCAR. In addition to performance, the plot shows efficiency as the percentage of peak performance the model is able to utilize.

Figure 18. WRF model performance (mflop/sec) V. no. of processors on the Terascale Computing System (TCS) at the Pittsburgh Supercomputing Center (pink triangles) and on the IBM Blue Sky (IBM) system at NCAR (blue diamonds). Efficiency as the percentage of peak performance that the model can utilize is shown by the pink squares for the TCS and by the blue squares for the IBM.

Work continued on two new dynamical cores being developed by NCEP as run-time options to WRF: the Nonhydrostatic Mesoscale Model (NMM; Javisa Janjic) core and the semi-implicit semi-Lagrangian (Jim Purser) core. Nesting in WRF is being implemented as two-way interactive domains that are mesh aligned and coincident (that is, non-rotated and with each point in the parent domain corresponding to a point in the nest). Nests may move, overlap, and telescope to arbitrary depth. Nests may be instantiated or destroyed on-the-fly. A prototype nesting capability (currently one-way) has been developed and is undergoing testing and additional enhancement to input lower boundary data such as terrain and land use for the high-resolution domain.

Inter-model coupling involves grid-to-grid translation, interpolation, and other computational transformations, and efficient communication between component models running in single-program multiple-data (SPMD) and multi-program multi-data (MPMD) modes. Initial implementation of coupling WRF is based on the Model Coupling Toolkit (MCT; Jacob and Larson, Argonne National Laboratory) and the Model Coupling Executable Library (MCEL; Bettencourt, University of Southern Mississippi). Ultimately, software being developed under the NASA Earth System Modeling Framework (ESMF) project will provide this functionality.

WRF model version 1.2 was released in April 2002, and is supported by NCAR for the user community. The main accomplishments in version 1.2 were the release of the new "mass coordinate" dynamical core developed by Klemp and Skamarock, and the addition of the land-surface model by Fei Chen (NCAR/RAP). The Standard Initialization pre-processor for WRF for real-data ingest was upgraded by Brent Shaw, John Smart, and Paul Schultz (all NOAA/FSL) in collaboration with David Gill, to enable these new capabilities to work in WRF real-data simulations. Additional physics in the 1.2 release also included the new Kain-Fritsch cumulus parameterization from Jack Kain (NOAA/NSSL), an efficient new microphysical scheme designed for the Eta operational model by Brad Ferrier (NOAA/NCEP), and the Eta model's Mellor-Yamada-Janjic PBL scheme added by Tom Black (NOAA/NCEP) and Jimy Dudhia.

For WRF to fulfill its objective as a scientific research model, it is essential that it incorporate state-of-the-science physical packages as they become available. The integration of new physics modules for WRF continues with the help of outside collaborators. Song-You Hong (Yonsei University, South Korea) who visited in the summer of 2002, worked with Dudhia to complete the incorporation of a revised planetary boundary layer scheme that improves the representation of PBL-top entrainment, and the PBL mixing of heat, moisture and momentum, compared to the current scheme used in MM5 and WRF that was developed by Hong and Hua-Lu Pan (NOAA/NCEP). This new scheme has been developed by Hong and Y. Noh (Yonsei University, South Korea) using knowledge gained from large-eddy simulations. Hong has also collaborated with Dudhia and visitor Shuhua Chen (University of California, Davis) on a modified simple ice microphysical parameterization that contains some new ideas in representing ice number concentration in single-moment ice schemes. Georg Grell (NOAA/FSL) collaborated with Wei Wang and Dudhia to incorporate a new ensemble cloud parameterization scheme in WRF that is being tested in preparation for release in the community version. Work is also under way to add FSL's RUC-model land-surface model, developed by Tatania Smirnova (NOAA/FSL), into WRF. Hall is developing a new, two-moment, five-species microphysical parameterization for WRF, MM5, and RUC.

Related website: http://www-ad.fsl.noaa.gov/fvb/rtvs/ihop/station/index.html

WRF real-time forecasts continued over the past year. The goals have been to evaluate the model under a variety of weather conditions, and to test the model's robustness. The focus of the real-time runs this year was the mass coordinate WRF model. Wang has been running three configurations of the mass model in real-time: 1) a 22-km CONUS domain, run at NCAR since September 2001, and with precipitation forecasts verified at NSSL by Baldwin (NOAA/NSSL/Storm Prediction Center); 2) a central-US 10-km domain run at NCAR, designed to capture the convective events in the Rocky Mountain and central plains regions; and, 3) a 10-km CONUS domain, run at FSL/NOAA since May 2002. Bresch built web pages to display these forecasts, which can be viewed at http://rain.mmm.ucar.edu/mm5/pages/wrf.html.

In addition to precipitation verification carried out at NSSL, the precipitation forecasts at stations from the central-US 10-km domain were verified by Andrew Loughe (NOAA/FSL) during the field experiment IHOP (http://www-ad.fsl.noaa.gov/fvb/rtvs/ihop/station/index.html). The WRF forecasts during IHOP have also been examined closely by a number of scientists at NCAR (Weisman) and at NSSL (Jack Kain).

The central U.S. 10-km forecasts have also served to verify the land-surface physics in WRF, using field data taken during IHOP. Time series output was produced from WRF forecasts and is being compared to field data by Fei Chen (NCAR/RAP).

An exciting endeavor this past year was the running of a CONUS domain at 10-km resolution. This was facilitated by FSL/NOAA, where the model was run on its computer "jet" twice daily. Hourly surface forecasts have been provided on the web since September 2002. An example of the forecast for tropical storm Isidore, initialized at 1200 UTC Sept 25 2002, is shown in Movie 3. This loop illustrates realistic detail in the surface hourly rainfall as the storm makes landfall in Louisiana.

Related website: http://www.mmm.ucar.edu/wrf/WG7

Trier and Davis completed and analyzed WRF simulations comprising diverse categories of observed meteorological events, including a wintertime midlatitude oceanic cyclone, a tropical cyclone, a long-lived warm-season mesoscale convective system, and a shallow arctic cold front in the lee of steep terrain. These simulations were run with the mass coordinate version of the model, with 10- to 30-km horizontal grid spacings, over regional to continental scale domains. Sensitivity of solutions to cumulus and explicit microphysical parameterizations were found, with the tropical cyclone simulation being particularly sensitive to which cumulus parameterization was used, and the long-lived continental mesoscale convective system being particularly sensitive to which explicit microphysical parameterization was used. These and additional case studies will be included in an upcoming semi-automated test suite, designed to allow WRF users to run the model with different configurations (e.g., grid spacing and domain sizes) and model physics options, and to compare results with available observations and with completed control simulations.

Trier and Davis also began examining WRF physics dependencies in tests at 1- to 12-km horizontal grid spacing. For two-dimensional simulations of deep convection, available WRF microphysics schemes and cumulus parameterizations have been varied for prototype thermodynamic stability and wind shear regimes. Figure 19 illustrates the strong sensitivity of precipitation to horizontal resolution, and, at the coarser grid spacings (e.g., 12 km), whether cumulus parameterization is used. In Fig. 19-a, with 1-km horizontal grid spacing the precipitation moves eastward at approximately 7 m/s and the mesoscale flow structure after several hours of simulation (not shown) resembles that of observed squall lines with trailing stratiform precipitation. When a cumulus parameterization is employed with 12-km horizontal grid spacing, the precipitation is quasi-stationary (Fig. 19-b) and does not exhibit the pronounced mesoscale flow features evident in the control case (not shown). Incorrect evolution and movement of mesoscale regions of precipitation is a shortcoming that has been noted in previous studies of parameterized convection.

Figure 19 (a and b). Hovmoller (time versus distance) diagrams of surface rainwater mixing ratio (g/kg) for (a) a fully explicit two-dimensional simulation of deep convection that utilized 1-km horizontal grid spacing, and (b) a two-dimensional simulation that employed 12-km horizontal grid spacing and utilized the Kain-Fritsch cumulus parameterization.

Jason Knievel addressed the verification and improvement of WRF's prediction of precipitation, especially precipitation resulting from MCSs. The focus was the model's treatment of temperature and humidity at the ground and in the lower troposphere. Using verification data from the International H2O Project (IHOP) conducted in May and June of 2002, preliminary indications are that the model often does not properly capture in its initial fields, nor accurately predict, pronounced stable layers in the lowest few kilometers.

Figures 20 and 21 present the observed soundings for Dodge City, Kansas valid for 1200 UTC 15 June 2002. As can be seen, the observed stable layer to 780 hPa (Fig. 20) is absent from the model sounding (Fig. 21).

Figure 20. Observed sounding for Dodge City, KS for 1200 UTC 15 June 2002. Dewpoint is in green and temperature is in red. Wind barbs: full barb= 10kt, half barb=5 ky, pennant= 50 kt.

Figure 21. WRF sounding for Dodge City, KS for 1200 UTC 15 June 2002. Dewpoint is in blue and temperature is in black. Wind barbs: full barb= 10kt, half barb=5 kt, pennant= 50 kt.

Perhaps as a result, the model inaccurately generates rain over large areas where stability is actually too high to foster precipitating clouds. It has been found that increases in the vertical resolution of the model do not solve the problem. Knievel's investigation of this issue will continue into 2003.

Bresch investigated WRF model behavior in a series of real-data cases, using a 10-km grid over the south-central United States. Forecasts were initialized twice daily for a one-week period at the beginning of July 2002 - a period with heavy rains in Texas. While it was found that forecasts were realistic, producing heavy rains in the San Antonio area, WRF tended to have too large an areal coverage of light rain. Figure 22 shows the temperature bias as a function of height for all radiosonde stations within the WRF domain, at several forecast times. WRF's temperature biases have been rather small (less than 0.5 deg K) at all forecast hours, with a slight cold bias at 850 hPa in the upper troposphere. Other variables also showed acceptably small biases.

Figure22. Temperature bias as a function of height for all radiosonde stations withing the WRF doman, at several forecast times.

One of the primary objectives of the WRF developmental effort is to improve the ability to represent and forecast convective systems in the 6-24 hour time frame. In order to test the current capabilities and limitations of the WRF model for such scenarios, Weisman ran 4-km simulations for convective cases centered on the western high plains, congruent with the International H2O project (IHOP) research domain. This region is rich in mesoscale features that could be important to convective triggering and evolution, including the dry line, nocturnal low-level jet, and cold and warm frontal passages. The enhanced observational database accompanying IHOP, including extra soundings, surface observations, and boundary layer moisture measurements, will also boost our ability to verify specific forecasts. While 4 km is nominally considered the minimum grid resolution necessary to represent MCSs explicitly, the use of 4-km rather than 12-km grids (which still require convective parameterizations) for the IHOP cases demonstrated significant improvement in representing system-scale structure for the larger convective systems (see Fig. 23). However, more isolated convective outbreaks were not as well represented.

Figure 23. Column maximum reflectivity at 00 UTC 16 June 2002 from a) the observed national radar composite, b) a 4 km resolution WRF simulation using the LIN microphysics scheme, c) a 10 km resolution WRF simulation using the LIN microphysics along with the Kain-Fritsch convective parameterization scheme, and d) a 10 km resolution WRF simulation using the NCEP3 microhysics scheme along with the Kain-Fritsch convective parameterization scheme, all initialized at 12 UTC 15 June using Eta model initial conditions.

As described earlier in section A.3 Mesoscale Data Assimilation, one of the strengths of WRF will be its data assimilation system, which has been a focus for the Mesoscale Prediction Group's Data Assimilation subgroup, where Barker, Guo, Huang, and Xiao continued to extend the capabilities of WRF 3DVAR.

As described earlier in the Wrf Model Physics section, in conjunction with the continuous WRF model development, WRF Version 1.2 was released in April 2002. The Standard Initialization (SI) pre-processor for WRF real-data ingest was upgraded in this release by Brent Shaw, John Smart, and Paul Schultz (NOAA/FSL) in collaboration with Gill, to enable the new capabilities in WRF 1.2 to work in real-data simulations. The post-processing software for NCL, Vis5D, and GrADS were updated by Skamarock, Wang, and Cindy Bruyere to work with the new mass dynamical core in the WRF model.

In June 2002, the Third WRF Users Workshop was held at NCAR. This brought together over 120 WRF developers, and the scientists testing and using WRF, to discuss progress, research, and issues. Following this, also in June 2002, was the WRF Tutorial. Lectures and discussions were conducted for the community on the operation and status of the WRF modeling system. There were 80 participants in the tutorial.

There are 418 users currently registered on WRF users' email list and over 900 people have visited the WRF Users' page to download the WRF software.

Related website: http://wrf-model.org

During the past year, a Memorandum of Agreement (MOA) was negotiated among the WRF principal partners in which the commitments of the respective organizations to the WRF effort are clarified. This MOA is in the process of being signed by NCAR, NOAA, FAA, AFWA, and the Navy. A number of WRF planning meetings were held during the year: the WRF Oversight Board met (February, Washington DC) to review the MOA and to address commitments to WRF; a workshop was held (February, Washington DC) for WRF Working Groups to review the status of development efforts, and to refine the plans and schedules for future work; and the WRF Science Board met (June, Boulder) to discuss priorities for WRF development activities.

In August, Powers assumed duties as the NCAR WRF project manager. Powers assists Klemp, the WRF project coordinator, in facilitating the NCAR contributions to WRF development and in interacting with the other WRF partner organizations.