Table of Contents | Director's Message | Executive Summary | MMM Achievements
Education and Outreach | Community Service | Awards | Publications | People | ASR 2004 Home

 

MMM Achievements

WRF: Weather Research and Forecast Model Development

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 will accelerate the transfer of research advances into operations. The model is being developed as a collaborative effort among the NCAR MMM Division, NOAA's National Centers for Environmental Prediction (NCEP) and Forecast System Laboratory (FSL), the Department of Defense's Air Force Weather Agency (AFWA) and Naval Research Laboratory (NRL), 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. 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 WRF model is state-of-the-art, transportable, and efficient in a massively parallel computing environment. It is designed to be modular, and a single source code will be maintained that can be configured for both research and operations. It offers numerous physics options, thus tapping into the experience of the broad modeling community. Advanced data assimilation systems are being developed and tested in tandem with the model. WRF is 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.

Significant development and testing for WRF continued during the past year, leading up to the release of the research-quality version (V2.0) in May. The basic 3DVAR package has been undergoing testing, and has been integrated with the model into the V2.0 release to insure compatibility in code supported for community use. The first operational implementation of WRF began at NCEP in September, running an ensemble configuration in their six High Resolution Domains. The following sections highlight the recent progress in the various areas of WRF development.

WRF Model numerics and dynamic cores

Click image for larger version.

Figure 1.
Comparison of simulated and observed temperature and precipitation for 1990/10 – 1991/3.

The WRF model framework presently hosts two dynamical cores: a split-explicit Eulerian model, based on mass vertical coordinate, called the Advanced Research WRF (ARW) core that was developed at NCAR, and the NMM (Nonhydrostatic Mesoscale Model) model, developed by Zavisa Janjic (NOAA/NCEP) from the operational hydrostatic ETA model. Additionally, a semi-implicit semi-Lagrangian prototype is being developed in an effort led by Jim Purser (NOAA/National Centers for Environmental Prediction). The WRF framework is being used to comparatively evaluate the relative accuracy and efficiency of their numerical techniques used in the cores, in a controlled computational environment, in retrospective forecasts by the DTC, in convective resolving forecasts in the 2004 warm season, and in ongoing idealized testing.

Click image for larger version.

Figure 2.
Text

Click image for larger version.

Figure 3. Text

William Skamarock has characterized the resolution capabilities of the ARW core using kinetic energy spectra computed from the 2003 BAMEX real-time forecasts run at both 10 and 4 km horizontal resolution. Figure 1 shows spectra computed from observations and from ARW model forecasts for the full forecast period during the BAMEX field program (5 May 2003 through 14 July 2003).

The ARW model reproduces the observed spectra very well, and spectra also show that the dissipation (damping) mechanisms in the model start becoming noticeable at around 6-8 dx as indicated in Figure (Skamarock_figure_2.ai). Examination of Spectra from other models, including the NCEP NMM, MM5, and COAMPS, indicates that the ARW model has significantly greater resolution capabilities than the other models based on their current configurations. Additionally, the differences in the spectra from the models can be traced to differences in the dissipation mechanisms used in the models.

In a complementary analysis, W. Skamarock has also computed structure functions from the WRF forecasts. The flatness factor for the WRF BAMEX forecasts is shown in Figure 2. It indicates that the strong intermittency observed in the mesoscale and cloudscale is reproduced in the WRF model forecasts, lending further evidence to the correctness of the WRF spectra at convective-resolving scales. In an effort related to WRF model dynamics and core development, W. Skamarock, Jim Doyle (ONR/NRL) and Peter Clark (UK Met Office), have begun developing a standard test set for nonhydrostatic dynamical cores of NWP models. Historically, a number of tests cases have been used to examine the efficiency and accuracy of discrete solvers for the nonhydrostatic Navier Stokes equations used in high-resolution NWP models. These test cases, including 2D and 3D mountain waves, gravity currents, thermals, and moist convection, are generally not well documented in the literature. W. Skamarock and his collaborators have put together a set of test cases that probe the robustness of the solvers with respect to nonhydrostatic motions. The need for a nonhydrostatic model test suite, and an initial set of test cases, was endorsed by over 80 scientists, representing both operations and research, at the Short Range Numerical Weather Prediction (SRNWP) workshop in Bad Orb, Germany in October 2003. The initial test suite is briefly described on the web at http://www.mmm.ucar.edu/individual/skamarock/test_cases/test_cases.html

Guenther Zaengl (University of Munich, Germany) visited in Spring 2004, and collaborated with Jimy Dudhia on WRF diffusion, damping, and a new vertical coordinate. G. Zaengl's diffusion changes are potential improvements for WRF applications in complex terrain, while the change to the upper damping layer will soon be added to WRF to allow its use with real-data simulations. G. Zaengl also provided a new vertical coordinate code for WRF that improves structures over mountains by having coordinate surfaces that have terrain slope effects confined to lower levels of the model atmosphere. In collaboration with J. Dudhia and Joseph Klemp, he has continued to further adapt this to a generalized time-dependent coordinate.

WRF computational framework

WRF in High-performance Computing

 

Click image for larger version.

Figure 4.
Text

A major element of the WRF software design is to provide good performance over a wide range of computing platforms. Through efforts led by John Michalakes, WRF is now ported to and running routinely on many of the fastest high performance computers in the world, as listed by the Top 500 Super Computing Sites organization (as of June 2004, see www.top500.org), including systems at NCAR, NOAA/NCEP and FSL, the Pittsburgh Supercomputing Center, Pacific Northwest National Laboratory, NCSA (University of Illinois), and the Navy Oceanographic Office. Through the efforts of Tom Henderson, working with the vendor, WRF has also been ported to the Cray X-1, the first scalable-parallel vector computing system. Recent WRF performance benchmarks in terms of sustained performance demonstrate good scaling across a large number of processors, as shown in Figure 4, below. In addition to parallelization of the WRF model, many of the WRF pre-processing and off-line nesting utilities have been parallelized by David Gill, avoiding bottlenecks for running large to very-large domains with the WRF system.


Grid Nesting Schemes

Click image to view a larger animation.

Figure 5. Text

The capabilities of the horizontal mesh refinement infrastructure have been expanded this year by J. Michalakes and D. Gill, with extensive testing conducted by J. Dudhia, W. Wang, and James Bresch during the 2004 hurricane season. The WRF model now performs both two-way (lateral forcing of the fine grid by the coarse grid, and feedback of the fine grid to the coarse domain) and one-way (no feedback from the fine grid to the coarse grid) horizontal grid nesting. Grid nesting was part of the research release of the WRF model (V2.0) made available to the user community last spring. The one-way nesting forecasts may be handled inside WRF as a single forecast where feedback is disabled, or through multiple sequential WRF forecasts (the technique that is preferable for real-time runs and used by operational centers). The two-way nesting options allow the model to generate an initial condition for the fine grid, or to alternatively ingest a fine grid domain's higher resolution orographic information. Both the one-way and two-way nesting options permit any integer ratio for the mesh refinement, as illustrated in the animation [nesting_compare.avi] showing fine-grid nesting ratios of 2, 3, 4, and 5. All portions of the WRF nesting code were developed from the ground up to be parallel, scalable, and low overhead. Preliminary tests show that the computational overhead for two-way nesting is around 8%, well below the original target of no more than 15% additional overhead for nesting.

J. Michalakes has implemented moving two-way interacting nests in WRF and demonstrated the capability with a Hurricane Ivan case from 11-13 September, 2004. The animation in Figure 5 shows simulated radar reflectivity for Hurricane Ivan in output from a moving 4 km nest on 80 processors of the NCAR IBM bluesky computer.

Without moving nests a researcher must create a nested domain large enough to contain the hurricane as it moves for the duration of the simulation. A smaller nest that can move with the storm requires only a fraction of the grid points of a non-moving domain, providing proportional savings in computational cost. The additional overhead for moving the nest within a running simulation is under 2 percent of total run time.

WRF as a component in coupled modeling systems

Click image for a larger animation.

Figure 6. Text

As part of a project funded by the U.S. Department of Defense Program Environment and Training (PET) program, J. Michalakes used WRF to demonstrate a flexible, efficient software infrastructure for operational forecasting of regional environments. The WRF I/O API was extended with the Model Coupling Environment Library (MCEL) developed by Matthew Bettencourt of Air Force Research Laboratory to handle coupled interactions between four environmental models into a coupled system for operational theater scale forecasting. The WRF atmospheric model, the ADCIRC ocean circulation model, the SWAN wave model, and the LSOM sediment optics model have been coupled together using the MCEL implementation of the WRF I/O and Model Coupling API to simulate a 24 hour forecast over the Yellow Sea. Figure 6 shows the coupling of the models, with output from the 24-hour forecast for November 25, 1999, the date of a high-wind induced ferryboat accident in that domain (note this image is clickable and viewable as an animation on-line).

WRF Model physics

The release of Version 2 of WRF in May 2004 included many new physics packages that were the culmination of several collaborative projects in recent years:

  • Fei Chen and Mukul Tewari (both RAP) worked with J. Dudhia on putting the unified Noah land-surface model into WRF V2. They also added some improvements to the urban representation and the emissivity treatment of the LSM this year.
  • Tanya Smirnova (NOAA/FSL) collaborated with W. Wang on adding the RUC LSM to WRF Version 2. This land-surface model is the one used operationally by FSL in the Rapid Update Cycle (RUC) model.
  • Georg Grell and Deszo Devenyi (both NOAA/FSL) provided their new Grell-Devenyi ensemble cumulus parameterization scheme, which was implemented by W. Wang and J. Dudhia into WRF. It is also part of the new RUC physics.
  • Song-You Hong and Jeong-Ock Lim (both Yonsei University, Seoul, Korea) worked on updated versions of the old NCEP 3-class and 5-class microphysics, and also provided a new 6-class scheme, including graupel. These new schemes known as WSM3, WSM5 and WSM6 are now part of WRF Version 2, and represent a collaborative effort with J. Dudhia and Shuhua Chen (U California, Davis).
  • S.-Y. Hong and J. Dudhia also finalized the Yonsei University planetary boundary layer scheme (YSU PBL), and added it to Version 2. This is an updated form of the MRF PBL that takes a new approach in representing the PBL top entrainment layer explicitly, and has been used as the preferred choice in WRF forecasts this year including hurricane and the 4 km central US forecasts.
  • Barry Lynn and Alex Khain (both Hebrew University of Jerusalem, Israel) visited for a week in summer 2004, and initiated work with Jimy Dudhia on adding a spectral bin microphysics parameterization to WRF. This is suitable for the fine-scale modeling of cloud processes with grid sizes well below 1 km.

Tom Black (NOAA/NCEP/EMC) and J. Dudhia added NCEP's updated physics packages to Version 2. This brings into WRF the version of the physics used operationally by NCEP, where notable changes have been made since the previous set of schemes brought into WRF Version 1 a few years ago. The updated packages included the Mellor-Yamada-Janjic PBL, the Betts-Miller-Janjic cumulus parameterization, the Ferrier Eta microphysics, and the GFDL radiation scheme. These physics packages can also be run by NCEP with their NMM core, making possible direct comparisons and swapped physics tests between dynamical cores as was done for the Developmental Testbed Center's retrospective forecast project.

WRF idealized and case-study testing

George Bryan (MMM/ASP) and Jason Knievel (RAP) evaluated squall line structures produced by the WRF Model in an idealized framework. They extended the research of an earlier study to include the newly released ARW core, and they explored a wider range of shear environments. They found that a previously documented spurious updraft pattern does not occur in strong shear conditions, even when a low-diffusion setting is used in the model. Additionally, they demonstrated that squall lines in weak shear are much more sensitive to implicit model diffusion than are squall lines in strong shear. Based on this work, they summarized their recommendations to WRF Model users in a paper at the WRF Users Workshop.

Click image for larger version.

Figure 7. Text

Click image for larger version.

Figure 8. Text

Axel Seifert (ASP), in collaboration with Morris Weisman, investigated the effects of cloud microphysics on WRF forecasts with explicit convection. Four cases from the BAMEX field phase were simulated with a four-km WRF using different microphysical schemes, including a sophisticated two-moment five-class scheme. All simulations revealed a very weak effect of the microphysical schemes on the mesoscale organization of the convection, for example in the June 10th case the formation and propagation speed of the squall line itself (Figure 7), although the schemes produce significantly different microscale cloud structures. The two-moment scheme is able to distinguish between the high reflectivity at the leading edge of the convection and a trailing stratiform region with lower reflectivity. The Lin scheme shows a lack of stratiform regions, while the Reisner scheme, which was mainly developed for stratiform clouds, overestimates the reflectivity in general. The surface precipitation (Figure 8) shows that the Lin scheme tends to overestimate the amount of precipitation in convective regions, while the other schemes give slightly better results. The two-moment scheme is available for the WRF community and, although it increases the WRF runtime by about 30%, may be attractive for research and special applications.

WRF experimental real-time forecasting

WRF real-time forecasts were continued this year in support of a variety of scientific goals. These interests include testing the robustness of the newly developed nesting capability, the convection-resolving forecasts over sub-continental US domain, and hurricane track and intensity forecasting.

In support of nesting development and WRF Version 2 release, a two-domain nested configuration at 36 and 12 km was set up in April to test and evaluate the new capability's correctness and robustness. Real-time runs and others since April have shown that the nesting is stable and robust.

Click image to view a larger version.

Figure 9. Text

Convective weather remains a significant challenge for numerical weather prediction systems, and is recognized as a major contributor to poor warm season quantitative precipitation forecasting (QPF). Based on the preliminary findings of BAMEX real-time, 4 km simulations from 2003, a twice-as-large 4 km domain was configured by Wang and Weisman to continue the explicit convection forecasts for the spring and summer of 2004. The extended domain included more mountainous regions to the west, and it was designed to test model's ability to simulate diurnally varying convection. These 4 km forecasts were then compared to equivalent 10 km WRF forecasts as well as to other operational models, which all employed convective parameterization.

Click image to view a larger animation.

Figure 10.
Text

The animation illustrates the model forecast for convection beginning on 24 May 2004 in comparison to the observed radar reflectivity. These simulations show much promise for improving convective forecasting out to 36 h, especially with respect to forecasting convective system mode and propagation, but also highlight existing challenges, related especially to the systematic over prediction of rainfall. The early findings from these enlarged domain forecasts showed that the model has indeed improved in capturing the diurnal variability of the convection. The overall success of these forecasts was also largely due to the use of explicit convective resolution over a very large domain, thus avoiding issues of mismatched convective representations across domain boundaries (e.g., explicit versus parameterized). Further sensitivity testing of these cases to changes in microphysics, boundary layer physics, resolution, etc., is ongoing in an attempt to improve known problem areas. The apparent improvements in convective forecast guidance at 4 km resolutions were found to be extremely useful for operations planning during the recent BAMEX field program, and were also highly praised by NWS and SPC forecasters, who used the WRF output for their daily severe weather outlooks. During the spring and summer of 2003, 2004, approximately 150 36 h real-time forecasts were conducted with the NCAR version 1.3 and 2.0 of WRF-ARW.

Click image to view a larger animation.

Figure 11. Text

Click image for larger version.

Figure 12. Text

To examine the ability of the updated WRF Version 2.0, which includes the nesting capability (released in May 2004), to predict the tracking and intensity of tropical cyclones, W. Wang, J. Bresch, and D. Gill conducted a series of real-time forecasts of hurricanes during the 2004 season, with the support of SCD. A number of powerful storms made landfall in the U.S. in September, and W. Wang, J. Bresch, and D. Gill ran WRF for hurricanes Charley, Frances, Ivan, and Jeanne. Nested WRF runs were performed with inner nests as fine as 4 km, within 12-km larger-area domains for 48-hour forecasts. A separate single domain, 12 km configuration was also run for 5-day track forecasts. While the high-resolution runs were designed to test the model's ability to simulate the realism of the storm intensity and structure, the low-resolution configuration was performed to examine the model's ability to predict tracks for up to 5 days. These experiments demonstrated that WRFV2 with its new two-way nesting capability has the potential to predict the intensity of the storms. During the period when Ivan was a category 4 and 5 storm, the nested WRF simulated surface winds in exceeding 60 m/s . The five-day, 12 km track forecasts showed that the model can predict the tracking of these systems with the accuracy similar or better than the current operational models in real-time mode. The figure below shows the comparison of the five-day track forecast from 0000 UTC, September 11 initialized WRF and that issued from National Hurricane Center at 0300 UTC on the same day. WRF's track was very accurate and predicted the landfall near Mobile Bay, Alabama, with a timing error of about 9 hours.

WRF model data assimilation

An upgraded version of the WRF 3D-Var (V2.0) data assimilation system was released to the research community in May 2004. The code, documentation and online tutorial are available from the WRF web-site http://www.wrf-model.org/wg4. New capabilities available to the community in V2.0 include a vertical velocity analysis, radar radial velocity assimilation, preconditioned conjugate gradient minimization (to improve convergence), and an improved assimilation of surface observations via the use of boundary layer parameterizations in the interpolation from model to observation locations. The development and testing work for the WRF 3DVAR is discussed in the section on Mesoscale Data Assimilation.

WRF Atmospheric Chemistry Modeling (WRF-Chem)

Click image for larger version.

Figure 13. Text

W. Skamarock has developed a new advection scheme for WRF-ARW chemistry (WRF-Chem) applications, applicable more generally to scalar transport in WRF-ARW applications. The advection scheme is based on a timesplit implementation of Lin and Rood’s unlimited-timestep advection algorithm, and it is implemented using a Piecewise-Parabolic Method (PPM) flux evaluation. The full scheme has both positive definite and monotonic options, is conservative and consistent with the discrete continuity equation in the WRF-ARW Runge-Kutta time integration. The large timestep capability of the new advection scheme will allow the use of timesteps for scalar transport that are much larger than those used in the WRF-ARW dynamics. In WRF-Chem applications, tens to hundreds of scalars are transported, and this transport dominates the processor time in a WRF-Chem application. The new transport scheme will reduce this the transport processor time by the ratio of the transport timestep to the dynamics timestep, resulting in major cost savings in addition to providing positive definite or monotonic behavior. Preliminary tests of the new scheme in the WRF model is shown in Figure 13, where transport within a highly nonlinear gravity current is reproduced by the new advection scheme at timesteps a factor of five larger than the stable timestep for the dynamical solver, with no discernable loss of accuracy.

WRF Regional Climate Modeling

Click image for larger version.


(a) Simulated T (b) Observed T (c) Simulated P (d) Observed P

Figure 14. Comparison of simulated and observed temperature and precipitation for 1990/10 – 1991/3.

A regional climate model (RCM) provides high-resolution climate scenarios important for impact assessment and resource management. The Weather Research and Forecasting (WRF) model was designed specifically for high-resolution applications and has provided a base for the development of a new RCM for simulations using 1-30 km grid spacing. The development of a regional climate capability for WRF is being carried out through an NCAR Opportunity Fund project led by B. Kuo and J. Dudhia. Ruby Leung (visitor from Pacific Northwest National Lab), and J. Dudhia are incorporating physics packages from the Community Climate System Model (CCSM) into WRF. First, the CAM3 radiation package was implemented and tested, and the work is now continuing with coupling the Community Land Model (collaborating with Gordon Bonan and Mariana Vertenstein, CGD). With James Done, several changes have been implemented to the model, including modification to the treatment of lateral boundary conditions, updating of sea surface temperature and vegetation parameters, and a consistent treatment of fractional cloud cover and cloud liquid/ice water path in the cloud and radiation parameterizations for cloud-radiation feedback.

Click image for larger version.


Figure 15. Scatter plots showing simulated versus observed elevation, surface temperature, precipitation, and snow water equivalent at the snow telemetry stations in the western U.S.

J. Done, R. Leung, C. Davis, and B. Kuo have evaluated WRF for four-month simulations and compared results with observations and with MM5. They conducted a cold-season simulation of the western U.S. at 30 km spatial resolution for the period October 1990 to March 1991 and compared the model results with observed temperature, precipitation, and snow data. Figure 15 compares the observed and simulated cold-season mean temperature and precipitation and shows a model warm bias of up to 3oC and an overprediction of precipitation, and compares the observed and simulated temperature, precipitation, and snow water equivalent at the snow telemetry stations that are typically located at high elevation. Snowpack is grossly underpredicted by the model. Although our results are similar to results from regional climate simulations using MM5 at 40-km grid spacing, it is thought the benefits of WRF will become apparent for higher resolution simulations.

WRF Forecast Verification

J. Knievel, D. Ahijevych, and K. Manning demonstrated that the realism of numerical weather prediction models can be evaluated by examining the temporal modes of the simulated rainfall. Because of its growing prominence in the operational and scientific communities, the Weather Research and Forecasting (WRF) Model was chosen for the study. Its simulated rainfall frequency was compared to rainfall frequency diagnosed from the WSR-88D network. Simulations and observations were similar overall in the normalized amplitudes of their diurnal and semidiurnal modes. The modes' phases were different, though, and in the WRF Model the simulated rain fell too early, and light rain was too frequent. The model also did not produce the distinct, nocturnal maximum in rainfall frequency so integral to the hydrologic cycle of the Great Plains.

C. Davis has collaborated with Barbara Brown, Randy Bullock and Daran Rife, all from RAP, to develop and apply new verification approaches to assess the skill of high-resolution forecasts of rainfall and surface winds. Traditional scores are problematic for verifying flows approaching the limits of predictability because they reward forecasts having unrealistically low spatial and temporal variability, and are mute regarding the causes of forecast error. With B. Brown and R. Bullock, C. Davis has developed and applied an algorithm to define coherent areas of precipitation, emphasizing mesoscale convection, and compared properties of these areas with observations obtained from NCEP Stage IV precipitation analyses (gauge and radar combined). Both simulations on a 22-km grid and 4-km grid, the latter with no cumulus parameterization, have been evaluated (for different time periods). In general, the WRF model produced too many large rain areas in both cases, large being defined as 20 grid lengths or more. The intensity distributions differed substantially, however. The WRF with parameterized convection had rainfall too narrowly distributed about the mean, especially in the later afternoon when the cumulus parameterization was most active. The intensity distribution among rain systems in the 4-km forecasts was generally too broad, especially in the late afternoon. Both sets of forecasts exhibited considerable regional variation in statistics. The 22-km WRF failed to produce the nocturnal MCS maximum in the Great Plains and underestimated the diurnal cycle of convection near the Gulf of Mexico, in agreement with Davis et al. (2003). The 4-km WRF exhibited the largest positive size and intensity bias associated with systems over the Midwest and Mississippi Valley regions, but little size bias over the High Plains, Ohio Valley and Southeast. Overall, the skill scores produced by the algorithm agreed with those obtained from manual inspection by Done et al.(2004).

C. Davis and D. Rife compared forecast skill of MM5 for 30-km and 3.3-km-grid forecasts over the White Sands Missile Range in Central and Southern New Mexico. They compared forecast time series at surface stations with observations by devising and applying non-traditional verification scores. In particular, they defined objects as two-hour temporal changes greater than one standard deviation in magnitude and verified whether forecasts produced observed changes, allowing for temporal error of the forecast changes. Whereas root-mean-square error statistics showed little benefit of the higher-resolution, the object-based technique revealed notably greater skill of finer-scale forecasts immediately over and within roughly 10-20 km of complex terrain features. However, there was only modest benefit over flat areas greater than 10-20 km from terrain features. In addition, higher skill was evident for the cross-barrier wind component than the along-barrier wind component. Other verification metrics, such as temporal anomaly correlation and time series variance were examined and pointed to similar benefits of higher-resolution.

WRF Workshops, Tutorials and Community Support

International WRF Workshops and scientific visits

In an effort to introduce the WRF modeling system to the international atmospheric science community, B. Kuo and J. Klemp organized two major international WRF workshops in FY04; one at the Seoul National University in December 2003, and the other at Taiwan’s Central Weather Bureau in September 2004. J. Klemp, B. Kuo, B. Skamarock, J. Dudhia, D. Barker, J. Knievel, and Georg Grell (NOAA/FSL) gave lectures that cover the various aspects of the WRF modeling system, from dynamics, numerics, data assimilation, chemistry modeling, to real-time forecast applications and Development Testbed Center (DTC). These workshops were all very well attended, with more than 100 participants in the audience, showing strong interest from the international community in the WRF modeling system. J. Klemp and B. Kuo also visited the New Zealand Weather Service and the Chinese National Meteorological Center to discuss the use of WRF for forecast applications over the Southern hemisphere and China. These international workshops and scientific visits are very useful in helping the international atmospheric science community to transition from MM5 to WRF for research and operational applications.

WRF community support

The number of WRF users continued to increase in the last year. Another 800 people (in addition to the previous 1400) have visited WRF site and downloaded the code. Over 960 users have subscribed to the WRF news list, compared to 600 people reported last year. In May 2004, the WRF ARW, SI, and 3DVAR Version 2.0 was released in which the nesting capability was included. Two minor version releases were followed in late May and early June. The WRF User page has been redesigned and made its debut with the WRF Version 2 release in May. The number of user emails continues to increase. During FY04, W. Wang and other MMM staff answered 1130 user emails, which is a 66% increase over the previous year.

W. Wang, J. Klemp, J. Dudhia and J. Powers organized the Fifth WRF and Fourteenth MM5 Users' Workshop, which took place 22-25 June. One hundred papers were received for the workshops, and among them, sixty-five were WRF-related. About 170 people from 90 institutions and 20 countries participated.

Following the Users' Workshop, a four-day WRF tutorial was offered on 28 June - 1 July. The WRF model, WRF Standard Initialization and WRF 3DVAR were taught at the tutorial. This was the first WRF tutorial where practice sessions were offered. It was no small task to accommodate 115 people (from 75 institutions and 15 countries) at the tutorial. A new User's Guide was developed and distributed at the tutorial. Cindy Bruyere, Shu-hua Chen (University of California, Davis), J. Dudhia, D. Gill, Yong-Run Guo, W. Huang, J. Michalakes, and Wang, along with John Smart, and Paula McCaslin (NOAA/FSL), lectured at this WRF tutorial.

Developmental Testbed Center and testing for WRF operational implementation

This year was the initial spin-up for the Center. Louisa Nance was hired during the final quarter of FY03, and Bob Gall became director about the same time. One additional employee was hired during FY04 (Meral Demirtas) who will be starting early in FY05. Strong links were forged between the NCAR and FSL components of the Central DTC located in Boulder. A Terms of Reference has been drafted describing a distributed DTC with components at several locations, including the central DTC in Boulder, consisting of components at FSL and NCAR, and initially one distributed component at NRL in Monterey, CA.

Completion of WRF System Test

Click image for larger version
.
Figure 16. One such event is the strong storm that impacted a significant portion of the East Coast between 15-17 February 2003 shown in the attached figures.

During FY03 and FY04, an extensive series of tests were performed on the two WRF Cores (NMM and ARW) and their attendant physics packages to determine readiness of the WRF system for it’s initial operational implementation at NCEP, October 1, 2004. This pre-implementation testing of WRF was a collaborative effort by scientists from NCAR, FSL, NCEP, AFWA, and Northup Grumman. The testing involved eight configurations, four for each core (baseline physics, swapped physics and perturbed initial/boundary conditions) applied to four regions around the US for four different seasons. This testing included over 2000 separate runs of the WRF system. Standard verification statistics were computed for each deterministic forecast, as well as ensemble statistics for various combinations of the eight configurations, using NCEP software. A select subset of individual events are also being analyzed for the purpose of gaining a physical understanding of the differences between the two WRF Cores and their attendant physics packages. One such event is the strong storm that impacted a significant portion of the East Coast between 15-17 February 2003 shown in Figure 16. Initial indications are the results for the cores themselves were similar with considerable differences related to the physics packages. The tests have led to operational implementation.

Initial WRF Visitor Program

A major goal of the DTC will be to tap the extensive expertise in the NWP research community for much of the testing that will be done in the DTC. This year we began a modest visitor program with four visitors; Bill Gallus and Isidora Jankov (Iowa State University), Dave Dempsey (San Francisco State University), and Ying Lin (NCEP ). The visitors worked on grid structure for the ARW core (Dempsey), physics options within the cores (Gallus and Jankov), minimum requirements for a cloud resolving resolution—grid spacing, physics options, number of levels (Gallus and Jankov), and verification techniques (Lin).

High Resolution Summer Experiment

Click image for larger version

Figure 17. Text

The DTC participated, with several other groups (NCEP, CAPS, NSSL, NWS Central Region); in a follow on experiment to the previous year’s of a high resolution (4km) forecast experiment over the Midwest. The purpose of these experiments was to examine whether forecasters found useful information in the high- resolution forecasts beyond that available in the lower resolution operational models in the 36-hour timeframe. The answer was “yes”. Figure 17 shows 1-h forecasts from the NMM and ARW cores using a cold start and a third forecast using the University of Oklahoma (CAPS) data assimilation system for a hot start with the ARW. At this time range, the hot start (which it was initialized with radar information) produced the best forecast. Figure 17 shows the same forecast cycle after 12 hours. At this longer time range the initial precipitation fields resulting from the assimilation system used in the hot-start forecast have died out and now the cold start forecasts look much better. At longer time ranges (not shown), all three models produce similar forecasts. This result was found consistently throughout the experiment.

MM5 Development and community support

Final MM5 Enhancements: Preparation for MM5's final release in FY05 are continuing. Many changes are being incorporated for the release (MM5 Version 3.7). Two of the major new components come from Guenther Zaengl (University of Munich, Germany) who has provided an improved sub-grid parameterization scheme for mountainous terrain, and a solar radiation modification to allow for terrain slope and shadow effects. Zaengl worked with C. Bruyere and J. Dudhia during his visit in Spring 2004 to finalize this code for Version 3.7. He has also provided separate code with a new vertical coordinate for MM5 that represents flow over complex terrain better, and an improved upper radiative boundary condition for nested domains. These codes will be made available to users separately from Version 3.7.

C. Bruyere and J. Dudhia are also adding a new version of the microphysics scheme by Paul Schultz (NOAA/FSL), modifying the Grell cumulus parameterization as suggested by Georg Grell (NOAA/FSL), making improvements to the MRF PBL scheme based on work by Yubao Liu and Fei Chen (RAP), and have changed the cloud ice representation in the CCM2 radiation in response to tests carried out by Cliff Mass and co-workers at the University of Washington.

MM5 Experimental Real-Time Mesoscale Modeling: To support flight forecasting and research activities over Antarctica, J. Powers, K. Manning, Michael Duda, and Syed Rizvi have continued to develop the Antarctic Mesoscale Prediction System (AMPS). NCAR has been collaborating in this project with David Bromwich of The Ohio State University's Byrd Polar Research Center. AMPS is an experimental real-time mesoscale modeling system generating twice-daily MM5 forecasts at resolutions as high as 3.3 km in support of the United States Antarctic Program. AMPS also serves a broad range of international activities, such as the Global Ocean Ecosystem Dynamics Program and the Antarctic operations of Italy, the UK, Australia, Chile, Germany, and South Africa.

Over the past year AMPS has also provided support for emergency medical evacuations. In September 2003 a medevac from Amundsen-Scott South Pole Station was necessitated by a worker suffering from a failing gall bladder before flight operations to the Pole had started for the summer season. AMPS was used in the fight forecasting for the Twin Otter rescue aircraft. Similarly, in April 2004, another critical medical situation required the airlift of three workers from McMurdo Station, this time after flight operations had ceased for the year, and AMPS was there to help. Over the past year AMPS began using 3-dimensional variational data assimilation (3DVAR). Analyses found this to be superior to the previous objective reanalysis package. This is significant not only for the immediate gain in forecast accuracy, but more importantly for the long-term positioning of AMPS to take advantage of indirect satellite atmospheric measurements, notably the COSMIC (Constellation Observing System for Meteorology, Ionosphere, and Climate) observations. The latter should greatly enhance the initialization of models targeting the polar regions. Future work in AMPS will revolve around implementation and testing of the WRF model over Antarctica. This application is expected to offer a unique testbed for this next-generation modeling system.

Next Topic: Mesoscale Data Assimilation

 

Table of Contents | Director's Message | Executive Summary | MMM Achievements
Education and Outreach | Community Service | Awards | Publications | People | ASR 2004 Home

National Center for Atmospheric Research University Corporation for Atmospheric Research National Science Foundation Annual Scientific Report - Home Atmospheric Chemistry Division Advanced Studies Program Atmospheric Chemistry Division Climate and Global Dynamics Division Environmental and Societal Impacts Group High Altitude Observatory Mesoscale & Microscale Meteorological Division Research Applications Program National Center for Atmospheric Research Scientific Computing Division