Weather forecasting has come a long way since Old Farmer's Almanacs first blew off the shelves in 1792. A network of ground-based weather stations and satellites collect data, which are pumped through sophisticated computer models and evaluated visually by the world's best meteorologists.
So why can't the local TV weather reporter get the forecast right? Part of the problem, say Indiana University Bloomington computer scientists Beth Plale and Dennis Gannon, is that weather models are run on fixed schedules, irrespective of current weather conditions.
Plale and Gannon are developing a new software infrastructure for running weather models as part of their National Science Foundation project Linked Environments for Atmospheric Discovery, or LEAD. With the new software, meteorologists will be able to start up weather models automatically in response to current weather conditions. The IU researchers believe a model that predicts weather in real-time will be more accurate and, ultimately, more reliable.
Weather is complicated stuff, and present-day models for predicting weather were designed with that fact in mind. Yesterday's computers and networks were comparatively slow. Pumping hundreds of variables through a complex model took a while.
Today's computers can handle complicated calculations with ease and can even be strung together to work on different parts of a model at the same time, collaborating and cooperating in what's called a "grid." With improvements in high-speed Internet access and performance, computers participating in a grid don't even need to be on the same continent. The time is right, Plale and Gannon believe, to rev up the quality of humanity's weather prediction tools.
"While fixed-schedule forecasts are useful for predicting large continent-wide weather patterns, they fall short in predicting the occurrence of smaller—or 'mesoscale’—phenomena such as tornados, severe storms, and flash floods," Plale says. "With the advent of small-scale radars, small enough to mount on cell towers, enough real-time data is now being collected in local regions such as Indianapolis or Chicago to improve forecasts significantly."
While the LEAD project is exploratory, early results look promising. As the technology matures, Plale and Gannon expect their model to be widely used by meteorologists, as well as by educators at the secondary, collegiate, and post-graduate levels.
"Working with meteorologists on a project that has such a visible, immediate, and well-defined benefit to society in terms of saving lives and taxpayer dollars is tremendously exciting," Plale says. "And the project's goals of bringing meteorology research into classrooms has the potential to increase the number of young people in this country who choose science as a career. These broader goals are what make the hard work worth undertaking."
Plale is an assistant professor of computer science at IU Bloomington. Gannon is chair of the IUB Department of Computer Science chair and is also CEO of Pervasive Tech Labs, a consortium of researchers dedicated to the creative use and application of information technology.
http://www.cs.indiana.edu/~plale/
