January 26, 2000 http://www.phillynews.com/daily_news/2000/Jan/26/local/COMP26.htm

It's a snow job

New computer deceived government forecasters

by Chris Brennan Daily News Staff Writer

The National Weather Service last week proudly unveiled its new $35 million computer and praised its ability to improve stormy weather predictions.

Seven days later, Mother Nature sucker-punched the government's shiny new IBM with a surprise snowstorm.

Government forecasters were following the storm for five days, convinced, with the new computer's advice, the snow would fall over the Atlantic.

Instead, it landed on us with a school-closing, traffic-snarling thud, giving snow-removal workers little time to prepare for their biggest job of the new millennium. The storm made a left turn Monday night and charged ashore, with the weather service expecting 10 to 15 inches of snow in the region through this morning.

Mike Eckert, a senior forecaster at the weather service office in Bowie, Md., said the new computer works like handicapping horses at the track, compiling the latest weather information into several "computer models."

"Each model is a different horse in the race," Eckert said. "The job as a meteorologist is to understand the different strengths of the models."

It appears WCAU-TV (Channel 10) won that race this time.

The station warned viewers Monday night of a strong storm blowing into town yesterday.

"It's not often that the government computers are this far off," said Glenn "Hurricane" Schwartz, a meteorologist at the NBC affiliate. "The way the science has advanced in the last 20 years, the computers are getting better and better and surprises are getting less likely."

But Eckert said yesterday's storm could have been an even bigger surprise. The new computer is five times faster than the one it replaced.

"Speed is very important in meteorology," Eckert said. "An hour would make a big difference for the 11 p.m. news."

The surprise storm was a problem for PennDOT as morning commuters dodged snow plows.

"It was a tough morning because many people went to bed thinking we'd get an inch or two," said Andrew L. Warren, PennDOT's district administrator. "If they knew the storm was coming, they probably wouldn't have gone to work."

Francis J. Graff, PennDOT's maintenance manager in Philadelphia, saw Channel 10's forecast and scrambled his crews into work early yesterday.

"It snuck up on us by a day," Graff said. "We all thought it was going to be Wednesday, but we did get a jump on it anyway."

Arthur Wick is the computer chief at the National Weather Service in Maryland, which is spending $35 million to lease the IBM-SP supercomputer until 2002. The IBM can make 690 billion calculations per second and will be upgraded in September to 2.5 trillion calculations per second. Wick said the computer measures weather factors such as temperature, wind and humidity, and lays that information over a map of the country. Forecasters then try to figure out what the information means.

"I wouldn't characterize it as guessing," Wick said as his colleagues drifted out of his office for home. "People are bailing out of here because of the weather."

Eckert, the senior forecaster, looked at a snowy white-out from his office while predicting the weather for Philadelphia for the next few days. The snow will stop, just in time for a deep freeze, he said.

"It's going to be very cold, very windy," Eckert said. He described yesterday's storm as "a classic nor'easter," churning over the Atlantic, building power with warmer sea air before pushing ashore over the much colder land with a punishing punch.

Watching the storm Monday, he expected a "non-event."

"All the models kept it off the coast," he said. Then Monday morning, "the model started taking the system closer to land. This one took a little bit longer before we got a handle on what was going on."

The weather service was changing its forecast - calling for more snow - between 8 and 10 p.m. Monday, but that limited the amount of time to spread the warning, he said.

Warren M. Washington, a senior scientist at the National Center for Atmospheric Research in Boulder, Colo. said the accuracy of weather prediction by computer will always be limited by what information the machine is given to calculate and how it is interpreted by people.

Asked about the weather service predictions, Washington said, "It wasn't as accurate as it should have been." But he said forecasters can't always account for every "chaos aspect," little changes in weather that can cause major shifts.

"This was probably a very rapid, developing storm," Washington said. "The public should not expect a perfect forecast."

-- Homer Beanfang (Bats@inbellfry.com), January 26, 2000

Answers

Washington said "The public should not expect a perfect forecast" Oh OK, WHAT THE HELL should we expect when they paid 35 million dollars for the system??????????? ======consumer is....backing up

-- consumer (shh@aol.com), January 26, 2000.

...But Eckert said yesterday's storm could have been an even bigger surprise. The new computer is five times faster than the one it replaced.

"Speed is very important in meteorology," Eckert said. "An hour would make a big difference for the 11 p.m. news."

I love technology. Now we can get the wrong answer five times faster!

-- I'm Here, I'm There (I'm Everywhere@so.beware), January 26, 2000.

this is funny. automation not only enables us to do the same thing faster but we can also do things that could never be done without automation too. like make bad weather forecasts at lightening speed.

i used to do requirements for clinical information systems. our standard joke was that we could make bad doctors do bad medicine faster!! hopefully, however, once we added artificial intelligence to the systems, the bad doctors would get smarter?!!

-- tt (cuddluppy@aol.com), January 26, 2000.

I am a Computer Scientist and I resemble that remark! Seriously though folks, they continue to buy into the hype and crap from companies like MS -- and everyone else in this game. They seem to believe these things can predict a complex future but they can't... they just can't. Eventually the programmers will figure out how to get these predictions right. By that time they (we) will have soaked us (you) for a cool few billion dollars.

Neat game, eh? You pay for our development. We give you toys that are right .... oh, maybe 90% of the time...

Go figure.

-- Michael Erskine (Osiris@urbanna.net), January 27, 2000.