In a paper in today's issue of Physical Review Letters, the researchers show that not all chaos on a weather map is equal, and outline a technique they've developed for identifying regions they call chaos hot spots.

These hot spots shift location on a regular basis and cover about 20 percent of the global map at any given time, write the Maryland research team, which is led by world leaders in chaos dynamics, numerical weather prediction, and massive databases.

"This work has tremendous potential for improving both the accuracy of existing forecasts and for increasing the length of time into the future that the weather can be predicted accurately," said math professor James Yorke, principle investigator for the research project.

"Because of expertise in the three areas essential to this project, the University of Maryland is uniquely capable of building on the forecasting improvements of the last three decades," said Yorke, who is director of the university's Institute for Physical Science and Technology and a member of the university's chaos theory group that U.S News and World Report currently ranks as the world's best.

Weather is what scientists call a complex chaotic system whose central property is that a tiny change in one part of the system can become magnified over time into a major change elsewhere. This means that a small localized weather change not accounted for in computer forecasting models can cause the actual weather pattern to gradually diverge from the models until what occurs in the sky over our heads is very different from what the weather person predicted a few days before.

Since 1992, the National Weather Service has provided "ensemble forecasts," in which a computer model generates a main forecast and several slightly adjusted forecasts that provide a range of possible outcomes for the weather. The forecast issued by local meteorologists represents a synthesis of these different models.

The ensemble approach and other improvements that brought about accurate 3 and 5 day forecasts were developed by a co-leader of the Maryland team, Eugenia Kalnay, during her tenure at the National Weather Service. Kalnay, who is chair of the university's department of meteorology, was director of the National Weather Services's Environmental Modeling Center from 1987 through 1997.

For their current findings, the Maryland researchers looked at global wind predictions from five of these ensemble forecasts at a particular level (the level at which atmospheric pressure is 500 millibars) in the atmosphere. Placing these five forecasts on the map, the researchers then looked at how each forecast deviates from the main forecast in wind strength and direction.

By analyzing squares that are 688 miles by 688 miles (1100 km by 1100 km) in a global map, they identified regions in which these deviations in wind vectors tend to line up with one another. The aligned wind vectors transform the regions in which they reside into chaos hot spots where good observations become most crucial for reducing forecasting errors. All other points on the map are less important for forecasting, the authors say.

According to team member and lead author D. J. Patil, the current work uses wind vectors to identify hot spots, because these measurements are readily available for many points on global weather maps. However, he noted that findings about chaos hot spots also apply to other variables that affect weather patterns such as temperature, humidity and barometric pressure.

The team's current findings are part of an ongoing project started last year that is supported by a $1 million grant from the W.M. Keck Foundation.

The project's next step is to look for global hot spots based on the running of a hundred possible forecasts rather than just the five used in this work. The team then plans to move from a global perspective down to the regional views of chaos hot spots that can translate into better regional and local forecasts.

These steps will require refining of the initial work and further development of methods for dealing with the huge data sets needed in weather prediction.

"Going from a global to a more precise and therefore more data-rich regional view means the chaos hot spots will become more numerous and harder to pinpoint, and the weather impact of small atmospheric changes in these hot spots increases," Patil said.

At the same time, the team will be determining the best way to use observations of wind, temperature or other atmospheric conditions to correct the weather modeling of the unstable regions or hot spots that are a key to improved forecasts.

According to Patil, the team will try to rank chaos hot spots based on which ones can best help keep "good forecasts from going bad."

"In some areas, your forecast doesn't get any better no matter how many readings you take, so we want to be able to target those hot spots where frequent readings can provide information that really improves forecasts," Patil said.

Maryland's chaos weather team is led by Yorke, Kalnay and Larry Davis, chair of the department of computer science. Davis, founder of the university's Institute for Advanced Computer Studies, is an acknowledged leader in high performance computing and computer vision. Team members who co-authored the Physical Review Letters paper are D. J. Patil, Brian R. Hunt, Eugenia Kalnay, James A. Yorke, and Edward Ott.

Graphics from the paper are available at this URL. - By Lee Tune

Related websites:

Yorke home page

Kalnay home page

Patil home page

[Contact: James Yorke, Eugenia Kalnay, Larry Davis, Dhanurjay (DJ) A.S. Patil, Lee Tune]

25-Jun-2001