|Scenes From a Storm
Slicing and dicing a blizzard of data, IBM's Deep Thunder computer model generates short-term 3-D forecasts of local weather. Working eight hours in advance, Deep Thunder predicted a thunderstorm in the New York metro area on the evening of May 31. A forecast for 8 p.m. shows the previous hour of rainfall, vertical winds at high altitudes, and rain yet to come (brown columns). Such focused forecasts can warn of flash floods, freezes, and other local weather that wider-angle models might miss.
|Restless Weather, Clearing Skies|
A time sequence traces how the May 31 New York storm unfolded. Early forecasts called for a warm, humid day. But as a cold front moved in at 7 p.m., winds shifted (arrows) and moisture rose, forming clouds. At this point the model correctly predicted that turbulent zones of rising air (blue) would form at the storm's core.
As the thunderstorm pushed eastward around 8 p.m., its anvil-shaped clouds were predicted to dump rain (green trail)—as much as two inches (5 centimeters) in some areas. After the actual storm, radar data confirmed that this part of the forecast was right on target.
By June 1, the storm had dissipated. The forecast was at most 30 minutes off, says IBM's Lloyd Treinish. "To learn what went wrong, we sometimes plug data from the actual storm back into the model and run it again."