FOR IMMEDIATE RELEASE CONTACT: July 21, 1993 Michael Schneider Pittsburgh Supercomputing Center 412-268-4960 Judith Wall University of Oklahoma 405-325-1701
"The model did a very good job of predicting storm type, motion and general characteristics," says Kelvin Droegemeier, associate professor of meteorology and co-founder of the University of Oklahoma's Center for Analysis and Prediction of Storms (CAPS). A research team led by Droegemeier developed the computer model, the Advanced Regional Prediction System (ARPS).
By providing more timely predictions, ARPS has the potential to reduce property damage and save lives, especially in areas like Oklahoma, which is prone to heavy flooding and tornadoes. Forecasts in Oklahoma now typically give about 30 minutes warning for a severe storm. With ARPS advance warning could be issued four or five hours earlier.
During the month-long experiment, forecasters at the National Weather Service's Experimental Forecast Facility in Norman, Okla. each morning determined the basic structure of the atmosphere. If there was any potential for storms that afternoon, these data became the starting conditions for a run of the ARPS model. Using connections on the Internet, an international computer network, a CAPS researcher in Oklahoma fed the data to the C90 in Pittsburgh by noon (central daylight time). By 1 p.m., output was available to forecasters in Oklahoma, in time for the daily afternoon forecast.
ARPS accurately predicted the general characteristics and motion of severe storms that ravaged Oklahoma on Mother's Day weekend, May 8 and 9. Forecasters gained enough confidence in the model that on May 21 they referred to it in an official National Weather Service release indicating the potential for intense thunderstorms over central Oklahoma that evening.
Because the ARPS model is very demanding computationally, the experiment depended on a collaborative effort between Droegemeier's research team and staff at the Pittsburgh Supercomputing Center, who scheduled the model to run on four of the C90's 16 processors each day during May. The CRAY C90 supercomputer used for the experiment is the newest, most powerful system from Cray Research, and the C90 at Pittsburgh, installed in late 1992, is the first one in the United States available for public research.
ARPS usually took about 45 minutes to complete its daily run, which gave a prediction that simulated four hours of storm evolution. Along with other output, the model provided forecasters with a sequence of images they could display on computer screens as a moving picture of how the storm would evolve during late afternoon and early evening.
Until now, storm-scale weather models have been used mainly for research, and an important goal of the experiment was to evaluate how ARPS would work in an operational setting. "We wanted to gain experience working with forecasters," says Droegemeier, "to understand what they need to use a model like this as part of daily forecasting. It's like in football," he adds, "you can practice and you can scrimmage, but you don't really know how you're doing until you're in a game."
Because the experiment was directed primarily at testing operational procedures, the model runs left out information -- surface features of the land and Doppler radar data -- that will be included in a more full-scale ARPS evaluation next year. Nevertheless, experienced forecasters found that even this simplified version of the model was useful. "It gave us an integrated estimate of rain water at various levels," says Paul Janish, a forecaster who participated in the experiment, "and we were able to evaluate other parameters to give us an idea what type of storm might develop."
Although computer models are a standard part of weather forecasting, existing models -- such as those that predict weather patterns across the United States -- can't predict movement and severity of individual storms. ARPS predicts over a smaller area and gives detailed readings on key storm parameters such as rainfall, wind direction and velocity, temperature and pressure. It also predicts in time increments related to the way storms actually develop -- every few minutes rather than every 12 hours.
In addition to providing more advance warning, ARPS should also make it possible to give more precise information about the type of storm likely to form, where it will go and how long it will last. "It's one thing to say there are going to be thunderstorms," says Janish, "it would be another to say thereีs a likelihood of strong thunderstorms with potential tornadoes, or that tornadoes seem most likely in this part of the state and will form three hours from now."
A fully operational ARPS model also has the potential to reduce costs for the airline industry, which loses millions of dollars annually due to weather-related delays at airports and re-routing of flights.
The Pittsburgh Supercomputing Center, a joint project of Carnegie Mellon University and the University of Pittsburgh together with Westinghouse Electric Corp., was established in 1986 by a grant from the National Science Foundation with support from the Commonwealth of Pennsylvania. The Center for Analysis and Prediction of Storms is one of the first 11 science and technology centers created in 1988 by the National Science Foundation. Its mission is to demonstrate the practicability of storm-scale weather prediction, with an emphasis on intense thunderstorms and related phenomena. CAPS envisions its research culminating in a prototype for regional prediction centers around the United States, with ARPS as the prototype numerical model.
Related article, with graphics, from Projects in Scientific Computing, PSC research report.
go back to contents page