Fine-scale help
with foul weather:
A forecasting tool for all seasons

While standard weather-forecasting models keep an eye on the jet stream and other continent-scale factors, thunderstorms, snow squalls and other localized hazards can sneak through to cause havoc. A smaller-scale weather-forecast model nourished through an NCAR-university partnership is now used worldwide for commerce, defense, and basic research.

For over 100 scientists and crew members cloaked in Antartica's 24-hour midwinter darkness, the situation aboard the Magdalena Oldendorff looked grim. The supply ship, strengthened to plow through polar sea ice, was en route from the Antarctic coast to Cape Town. In mid-June 2002, the ship became trapped in pack ice. Buffeted by snow and bitterly cold winds approaching hurricane force, the ship had no means of escape in sight.

Only a few years ago, orchestrating a rescue in such conditions would have been nearly impossible. Weather reports were (and are) scarce across the Antarctic, and few computer forecast models had been tailored for the continent's unique geography and harsh climate. Fortunately for those aboard the Magdalena Oldendorff, things have changed since 2000. A mesoscale model cultivated at Pennsylvania State University and refined at NCAR and Ohio State University has been adapted to spot the brutal punches of wintry weather that pummel Antarctica.

Forecasts from this Antarctic Mesoscale Prediction System (AMPS) helped the South African Weather Service guide the SA Agulhas from Cape Town to pick up the stranded ship's 78 Russian scientists and most of its 28 German crew members. Two helicopters from the Agulhas finished bringing the last passengers aboard on July 1, when AMPS had targeted a brief window of favorable weather.

The dramatic rescue came only a year after another challenging mission in harsh Antarctic weather. In early April 2001, Ronald Shemenski—the only physician at the South Pole's Amundsen-Scott research outpost—suffered a gallstone attack. Another attack could have endangered Shemenski's life and put the entire crew at risk. Relying heavily on a forecast system built around AMPS, meteorologists with the U.S. Antarctic Program foresaw an end to a stretch of wind and blowing snow and identified a five-day window when aircraft might attempt a rescue. A Twin Otter hopscotched from Chile to the nearby Antarctic Peninsula and finally flew onto the Pole's runway—lit by a string of improvised torches—on April 24. Within days, Shemenski was recuperating in a Denver hospital.

Zooming in on weather's worst

Conditions like these were far from the minds of the scientists who created the Penn State/NCAR mesoscale model. It was launched in the 1970s by a team led by Penn State assistant professor Richard Anthes, now the president of UCAR. An ever-expanding group of collaborators have built variations and improvements to the model over the years. Then and now, the goal has been to simulate mesoscale weather: thunderstorms, snow bands, and other features that range roughly from a county to a state in size. The national-scale models guiding public forecasts in the 1970s could only slice and dice weather about every 190 kilometers (120 miles), a resolution far too coarse to track local weather features.

By 1980, creative graduate students at Penn State were adapting the model to suit their needs. Among these innovators was Ying-Hwa "Bill" Kuo, now an NCAR scientist. At the time, says Kuo, "there were no systematic support services" for users outside the university. That problem was addressed in the mesoscale model's fourth incarnation, MM4, created for a major study of acid deposition. With Anthes and Kuo in Boulder, the model became a dual effort between Penn State and NCAR. "We basically cleaned up the MM3, made it more efficient and more user-friendly, and started providing user services," recalls Kuo.

The model's most recent version, MM5, debuted in 1991. It quickly became a lodestone for university researchers. Once they or their students signed on, they could pull software off the Internet free of charge and take advantage of support and regular workshops at NCAR. "Our user count has increased really rapidly over the last ten years," says Kuo. The growth has been especially strong among government and industry scientists, who now make up about half of all MM5 users.

What makes MM5 so appealing to so many? For starters, the model is remarkably limber, thanks in large part to a flexible nested grid developed by NCAR's Georg Grell. Like a virtual magnifying glass, the nested grid allows users to focus on specific regions at a scale that would be too computationally intensive if applied to the model's entire domain. The nested grid can detail weather features as finely as every 1 km (0.6 mi). Jimy Dudhia (NCAR), David Stauffer (Penn State), and colleagues have devised other improvements, such as the ability to assimilate data in four dimensions (space and time) and to run the model free of hydrostatic assumptions that limit the range of vertical motions depicted.

Overall, MM5 now features a rare blend of breadth and detail that allows weather to be tracked from cloud scale to global scale.

  Ringed by
sea ice
and covered
with thick
glaciers,
Antarctica
is a
unique
forecasting
challenge.

A world of uses

National defense agencies have put MM5 to operational use in eclectic settings, with NCAR scientist Thomas Warner serving as liaison. Five Army test ranges, from Alaska to the East Coast to the desert Southwest, employ the model to produce ultra? high-resolution forecasts. A similar configuration provided outlooks for the 2002 Winter Olympic Games in Salt Lake City, with the output used to predict the transport of hazardous material if any were released by a terrorist. Forecasts for Afghanistan were generated in the fall of 2001 for the National Ground Intelligence Center. And post-event MM5 analysis of weather conditions during the 1991 war in the Persian Gulf contributed to the assessment of whether Gulf War syndrome might be related to accidental exposure to nerve gas. The battles against Colorado wildfires in the summer of 2002 also benefited from MM5 forecasts.

MM5 has made its mark on everyday aspects of the weather business as well. A number of private firms have adapted the model to produce finely detailed outlooks, in some cases parsed every hour. Some of the local forecasts most commonly seen on television and the World Wide Web now rely on MM5-derived forecast products.

Several public-private partnerships are doing innovative work with MM5. For instance, states from Texas to North Carolina now issue air-quality alerts to the public using an MM5-based monitoring system. It was developed by the Southeast Center for Mesoscale Environmental Prediction, a consortium that includes an air-quality modeling firm, a television station, and North Carolina State University.

Stretching the envelope to 90° south

Although MM5 is exceptionally pliable, adapting it isn't necessarily a snap. Kuo says that the polar forecasts were an interesting outcome of a May 2000 workshop on Antarctic weather prediction, sponsored by NSF and held at Ohio State University. Not long ago, the southernmost continent got little attention from forecasters and modelers. Surface stations are widely scattered, and each part of Antarctica is bypassed by weather satellites for several hours each day. Moreover, the high, cold, dry environment presents inherent problems for computational forecasting. According to Kuo, "All your model problems get amplified."

Still, he decided to give the workshop participants a five-minute presentation on experimental predictions over Antarctica from a global version of MM5 developed by Dudhia. Even at the model's coarsest resolution—120 km (75 mi)—winter storms pinwheeled around the continent's margins, evolving in ways familiar to veteran forecasters. This led to a joint effort by Ohio State and NCAR to develop AMPS. At its heart was a polar version of MM5 that zeroed in on Antarctica.

The Polar MM5 emerged from a research group at Ohio State led by meteorologist David Bromwich, a 25-year veteran of polar analysis and modeling. As Bromwich recalls, "The standard physics in MM5 did not work well over Antarctica and Greenland, because conditions there are so different from the central United States, for which MM5 was primarily developed." Existing code was substantially modified to simulate clouds, radiation, the boundary layer, and sea ice in the polar regime. Long-range, global-scale outlooks for Antarctica now extend out to five days, while forecasters can examine the Ross Sea region and other points of interest in hourly increments at a scale of 3.3 km (2 mi).

Only a few months after the Ohio State meeting, NCAR was putting daily runs of the PolarMM5 on the World Wide Web as part of AMPS. Producing day-to-day forecasts isn't part of NCAR's mission. However, these outlooks furthered both research and operations. "Whenever the system went down, people [using it] would call us up," says Jordan Powers, who oversees the daily operation of AMPS at NCAR. A subsequent meeting solidified relations between the NCAR? Ohio State collaboration and the NSF-coordinated U.S. Antarctic Program. The austral summer of 2001?02 made a strong case for AMPS' merit. Navy meteorologist Arthur Cayette manages forecasting for the Antarctic Program, where maximum flight time has to be wrung out of nearly every day during the brief summer. On more than three-quarters of the season's toughest forecast days, AMPS provided the determining guidance, according to Cayette. "There was an extended period of extreme weather during January, at the height of the summer season. Without the use of AMPS forecasts, we would have been faced with conservative thinking, and opportunities captured by forecasting periodic breaks in the weather would have been lost."

Next: a partnership from square one

Despite its continued growth, MM5's days are actually numbered. "Even though MM5 is a very flexible system, some elements are more than 30 years old," says Kuo. "To capture advances in modeling research, we need to implement new software architecture, high-accuracy computational schemes, and a new dynamical framework."

That framework will be provided in the context of a brand-new system, the Weather Research and Forecasting Model. Now in the midst of development, WRF is being built by NCAR, NOAA's Forecast Systems Laboratory and its National Centers for Environmental Prediction, the Air Force Weather Agency, and the University of Oklahoma's Center for the Analysis and Prediction of Storms. Since the late 1990s, scientists from each of these organizations have been crafting a forecast model designed from the start with both research and operational uses in mind.

In 2001 a beta-test version of WRF went online with daily forecasts. The results thus far are encouraging to NCAR's Joseph Klemp, one of the lead developers for WRF.

Weather
Research and
Forecasting

Kuo points to the "cultural exchange" now taking place in WRF among research modelers, who are always looking ahead to the next improvement, and operational modelers, who are keenly aware of the need for consistency and reliability in public forecasts. WRF could make both groups happy, Kuo believes. Operational versions of the model could advance with as much deliberation as needed. Other WRF variants could be tweaked more quickly to produce the kinds of regional innovations made possible by MM5. On the forecasting frontiers of tomorrow, some version of WRF could even help save another life—or a shipload of lives.