The National Weather Service, which suffers from the unique problem of having more satellite data than it knows what to do with, got a huge burst in number-crunching power Tuesday as a new parallel-processing supercomputer became operational.
The $35 million IBM SP, which has been undergoing testing since November, is five times faster than the Cray C-90 computer currently in use. The Cray C-90 has been processing data from the service's four dedicated satellites and ground stations since 1994. Once the IBM gets an additional upgrade in September, it will be 28 times faster than the Cray, pumping out 2.5 trillion calculations per second.
"We have satellite systems facing us that are going to have 200 times the amount of data coming at us. And that's all going to be put into our models."
The new computer will handle the data assimilation and numerical modeling for the NWS meteorologists, just like the old one did, but the increased power allows for more complicated models that will yield more accurate and far-reaching weather forecasts.
"Right now, we predict weather systems the size of the state of New Jersey, and three, four or five days in advance," said Louis Uccellini, director of NOAA's National Centers for Environmental Prediction. "Now we'll be able to predict systems the size of individual counties within a state and extend those forecasts out five, six or seven days in advance."
The computer will be used to predict short-term terrestrial weather only. It will not be used for space weather predictions, or any terrestrial climate predictions more than 14 days in advance.
But even with the new IBM, meteorologists aren't even close to exhausting the data that the satellites produce.
"[The] good news in the meteorological community is that there's a lot of data, the bad news is that there's a lot of data," said NWS director John J. Kelly Jr. "I'm not sure our computational power will ever catch up with our data."
"We have satellite systems facing us that are going to have 200 times the amount of data coming at us," Uccellini said. "And that's all going to be put into our models."
The IBM runs four weather models simultaneously -- one for global predictions, and the other three for local predictions -- all of which interact. At the same time, the computer handles data assimilation, which means taking in all the satellite and ground-based data, sifting them for their most reliable components and giving each piece of data its proper weight and relevance so that they do not conflict when fed into the model.
"The data assimilation folks could probably use one of these computers just for data assimilation," Uccellini said. "And they're going to be sharing with the models people."
Along with a computer reserved for that task, Kelly said, the National Weather Service would like a faster computer to do "climate and seasonal forecasting," that could better predict the effects of long-term atmospheric effects like El Nino or La Nina.
Kelly speculated on the possibility that funding for a new climate-dedicated supercomputer may be part of President Clinton's 2001 budget, due out by the end of January.
"Let's just say we'll talk in a couple of weeks about it," Kelly said, laughing.
advertisement
