Matt Briggs, left, a third-year graduate student in soil, crop and atmospheric sciences, and Richard Levine, a fifth-year grad student in statistics and biometry, look at a national radar summary on computer to verify the forecast.
By Blaine P. Friedlander Jr.
The chance of more accurate weather forecasts may soon improve as a result of a new method of statistical forecast analysis developed by two Cornell gradu ate students.
Based on a relatively new mathematical concept known as wavelets, the stu dents' methods could help forecasters develop a new prediction paradigm.
"This is not a forecast itself," said William M. Briggs, Cornell doctoral student in meteorological statistics. "It allows the meteorologist to gauge how good the forecast was after the fact. We use a mathematical concept known as wavelets to develop new error diagnostic procedures."
For example, the new method gauges field forecasts of temperature over a wide -area grid, then determines how close the forecast was to what actually happened. Using this information, meteorologists soon will be able to develop a better database and clearly see where the weaknesses are in their own forecasts.
Briggs and Richard A. Levine, Cornell doctoral student in statistics, presented their paper, "Wavelets and Image Comparison: New Approaches to Field Forecast Verifi cation," at the American Meteorological Society's 13th Conference on Probability and Statistics in the Atmospheric Sciences in San Francisco on Feb. 22.
Their method, which could later help hone weather forecasts, was developed in the environmental statistics program directed by George Casella, Cornell Lib erty Hyde Bailey Professor of Biometrics, and David Ruppert, Cornell profes sor of operations research. Briggs studies under Daniel Wilks, Cornell professor of atmospheric sciences.
"Briggs and Levine are applying new statistical methodology to an ex tremely difficult prediction problem," Casella said. "They face a lot of com plications. While trying to model physical processes and manage data sets, they're mixing a lot of complex characteristics. That makes it difficult."
But what are wavelets?
"Wavelets are mathematical tools that help remove the noise from the data," Levine said. "Wavelets remove the errors that may exist. In data, there is real versus muddled data; information is
muddled by extraneous stuff. But somewhere in that information is an obvious trend that can't be seen because there is so much junk. By running data through wavelets, it gives a clearer picture of what's going on."
As an analogy, suppose an old, scratchy sound recording were put onto a digital format -- scratches and all. The digitized music then could be electroni cally modified through a wavelet-oriented computer program and the extra neous scratches could be easily removed, leaving a more crisp sound intact.
With all the variables of weather to consider, forecasters have a tough time sifting through the noise, Levine said. This program arms the meteorologist with that much more information. For example, this method separates the insig nificant data from the truly useful information, so that meteorologists can better understand how wide-area forecasts of rain or temperature performed.
"This technique removes some of the subjectivity that now exists in the analy sis of the forecast. Wavelets act like a variable microscope, allowing us to study forecast performance at different scales," Briggs said. "This method can show the meteorologist that we're doing well forecasting here, but perhaps not so well there. We think incorporating wavelets can give important insights on the fore cast process."