WRF tags Isabel
NCAR’s new model gets high marks for predicting the course and timing of the historic hurricane.
The most powerful weather model of its kind met the most powerful Atlantic storm of the young century last month—with memorable results.
As Hurricane Isabel bore down on the East Coast, NCAR researchers broke out the newly developed Weather Research and Forecasting Model (WRF, pronounced “worf”) to simulate the storm. As early as 15 September, the model generated high-resolution forecasts showing the storm would strike North Carolina’s Outer Banks in the middle of the day on 18 September. That forecast proved to be remarkably prescient: the storm tracked exactly as expected and made landfall within the predicted time frame.
“I think WRF did great,” says Jordan Powers, a Mesoscale and Microscale Meteorology (MMM) Division scientist who manages WRF operations.
WRF, which was conceived by researchers and forecasters in the late 1990s, is now in the testing stage. NCAR scientists used it during the late spring and summer this year to capture massive midwestern thunderstorms during the Bow Echo and MCV (Mesoscale Convective Vortex) Experiment, better known as BAMEX. The model performed remarkably well, enabling researchers to anticipate the development of thunderstorm complexes up to 36 hours in advance.
Last month, scientists began using WRF to predict tropical storms. But nothing could challenge the new model like Isabel, a storm of historic proportions that achieved category 5 strength—the strongest possible rating for a hurricane on the Saffir-Simpson scale—while still well out over the Atlantic Ocean.
“We thought it was a good opportunity to apply the model to a situation that was attracting a lot of interest as the storm approached the mainland,” Jordan says. “We had run WRF in real time for BAMEX, which was over land and in the center of the country. Now we wanted to see how WFR behaved at very high resolution with a storm like a hurricane over the ocean.”
Hurricane and Blue Sky
On 12 September, MMM’s Joe Klemp, Bob Gall, Wei Wang, and Jordan made the decision to run five-day forecasts of Isabel, instead of the 36-hour forecasts that had been run for tropical storms and for Isabel in its early stages. They also decided to create exceptionally high-resolution simulations as the storm approached land. WRF would operate on a model grid with data points as close as just 4 kilometers (2.5 miles), bringing into focus Isabel’s internal structure, including the eyewall and rain bands. In contrast, existing hurricane models generally operate on a much coarser grid of at least 18 kilometers (11.3 miles).
Running such long-range and high-resolution forecasts threatened to use up much of MMM’s reserved computing time at the Scientific Computing Division. But because Isabel was such a high priority, NCAR director Tim Killeen granted a special allocation that boosted MMM’s computer allotment. SCD staff expedited WRF’s Isabel forecasts.
Beginning 13 September, the IBM Blue Sky supercomputer in the Mesa Lab hummed with calculations as WRF’s five-day forecasts were updated twice daily. On 15 September, WRF began to zoom in on Isabel to create high-resolution, 48-hour forecasts. MMM’s Jim Bresch posted plots on the Web.
The results impressed even scientists who work with WRF on a daily basis. Wei recalls that the initial conditions entered into the model set up a fairly weak storm. But WRF respondedby strengthening Isabel into a major hurricane.
“The model was able to take the initial condition and spin up quite well,” Wei says. “That was very exciting to see. I hadn’t been sure what to expect.”
WRF’s early 10-kilometer runs indicated that Isabel would veer farther to the north and strike the New Jersey shore. In the following days, however, WRF corrected the track. Researchers also were able to view full-color forecasts that resembled radar images and showed areas of intense rain and wind gusts within the large storm. “They were very realistic images,” Wei says.
Wei hasn’t done a full-scale analysis comparing the forecasts to the nuances of the actual storm. She suspects, however, that WRF slightly overestimated the storm’s strength at landfall, in part because the model does not entirely capture interactions between the ocean and the atmosphere.
WRF was not the only model that won kudos for tracking Isabel. Thanks to the ever-increasing power of computers and more data about the atmosphere, forecasting models in general shone in the days leading up to the hurricane’s landfall. The National Hurricane Center, using models with coarser resolution than WRF, predicted the storm’s track with a high degree of accuracy—although the storm’s intensity proved more difficult to forecast.
WRF builds upon the capabilities of current models, such as the MM5 (a mesoscale model developed by NCAR and Pennsylvania State University)and Eta (developed by the National Weather Service’s National Centers for Environmental Prediction). It is designed to meet the needs of both researchers and forecasters. The model uses a state-of-art computer code, developed by an interdisciplinary team led by MMM’s John Michalakes, that can be run on a wide range of computer platforms ranging from a desktop workstation to the Earth Simulator supercomputer.
“Research advances will have a direct path to operations, thereby providing society with better forecasts,” explains MMM director Bob Gall. “This link between research and operations makes WRF unique in the history of numerical weather prediction in the United States.”
Using the experience of Isabel and other storms, researchers will continue to tweak WRF. The National Centers for Environmental Prediction plans to begin using it for high-resolution forecasts in September 2004, and other agencies, including some branches of the military, subsequently will use WRF for their specialized operational needs. NCAR will maintain the model to facilitate wide use in research, particularly in the university community.
In the next few months, MMM plans to release a more sophisticated research version that will enable scientists to nest a high-resolution grid within a larger grid of coarser resolution. That way, scientists can focus in on areas of concern while using computer time as efficiently as possible. In time, WRF may also be coupled with ocean and/or wave models for even better forecasts of hurricanes.