"The important part is merging good observations with good modeling. Just putting an instrument in the air or space doesnt get it done. Producing a better computer model wont get it done. You have to have good data assimilation tomake predictive modeling work."

"A storm is a thermodynamic engine," said Christopher Velden, an atmospheric science researcher at the University of Wisconsins Cooperative Institute for Meteorological Satellite Studies. "It runs because of a warm ocean and the transfer of that heat to the atmosphere through organized thunderstorms. To simulate this numerically, you have to deal with things like small-scale rain cells and cloud formation. There also are a number of secondary circulation and feedback mechanisms in play. All those behaviors are very difficult to capture mathematically."

Key to effective hurricane modeling is data assimilation, a term used by forecasters to describe the process by which satellite-borne sensor readings are transmitted, received, validated and then integrated into predictive software. It is a painstaking, laborious process that can take up to three years for every new instrument on each satellite launched.

"The important part is merging good observations with good modeling," says Stephen Lord, director of the NOAA Environmental Modeling Center. "Just putting an instrument in the air or space doesnt get it done. Producing a better computer model wont get it done. You have to have good data assimilation to make predictive modeling work."

One of the forecasters newest tools is an IBM supercomputer that replaces a six-year-old Cray C 90 at National Weather Service offices in Bowie, Maryland. When upgraded this coming September with additional microprocessors, the IBM will be able to process data at a speed of 2.5 trillion instructions per second, some 28 times faster than the Cray.

The added capability should enable forecasters to boost the accuracy of longer-range forecasts and to provide a more rapid means of testing new computerized weather models.