What is the possibility that weather computers are experiencing Y2K problems? The east coast storm this week was a fiasco for the NWS and as of yet they are at loss to explain their erroneous predictions. I saw a guy from NWS on the national news last night, and he said "We are looking into the possibility that the computer models were fed bad data".

Hmmm...

-- Old Man Winter (@ .), January 27, 2000

Answers

---I've seen several stories and comments by local weather people talking about the same thing. They currently have 3 different models for this next 2 wave stom coming. and they are really different! And this week is the first time I ever heard weathermen complain about the computers, so yes, I do think your hypothesis might have merit, from time coincidence alone. Lot of number crunchin going on in those mainframes, who knows what software or combination they are running, and also, what other computers and data they are linked to.

-- zog (zzoggy@yahoo.com), January 27, 2000.

you're trying to blame bad weather forecasting on y2k? What's next, the weather itself? A little old lady had her bag snatched today, i wonder if y2k was the cause?

-- Mr. Sane (hhh@home.com), January 27, 2000.

Come come Mr. Sane. what background do you have to make this comment. I've worked in the IT field for over 28 years. I have written much code in modeling software. Yes its very possible that the software is getting data loss due to y2k errors.

justtthink@y2k.com

-- justthinkin com (justthinkin@y2k.com), January 27, 2000.

Our weather forecasters here have been wrong 50% of the time. It goes with the job.

-- (I'm@pol.ly), January 27, 2000.

I find it interesting that some forecasters did disagree with the models -- but wouldn't (or couldn't?) openly disagree with the computer-generated prediction. I can hear it now: "Look, we paid $35 million for this setup -- who do you think you are?"

-- I'm Here, I'm There (I'm Everywhere@so.beware), January 27, 2000.

Our local forcaster in the Northeast did say that the NEW computers did cause some problems with being able to predict this past storm accurately. They are looking into the matter as to not have egg on their faces again. They were off by a whole day as to when the storm would start. Luckily I had my stores and did not have to go out!

-- Diane (cptlauthor@aol.com), January 27, 2000.

Right 50% of the time???? How do you get such good ones??

-- walt (walt@lcs.k12.ne.us), January 27, 2000.

Gee, up here in the great Northwest, they predict either rain. showers or partial clearing with showers and I'd bet they are right 98% of the time!

Seriously, I read somewhere in the last couple of days (can't remember where - Shakey's senility kicking in maybe) that the software wasn't performing to expectation and more training was needed in interpreting the data that was presented. That was the reason given as to why they only got the storm warnings 2 or 3 hours ahead of the storm.

-- Valkyrie (anon@please.xnet), January 27, 2000.

href=http://www.msnbc.com/news/362109.asp>MSNBC News

When good forecast models go bad

How high-tech tools fooled forecasters

By David Montgomery and D'Vera Cohn WASHINGTON POST

WASHINGTON, Jan. 26  The worlds best computer weather models, located in Bowie, had digested reams of data from weather balloons, ocean buoys, observation posts, satellites and ships. Supported by a new IBM supercomputer firing up enhanced software for the first time, the models calculated the highs and lows, the ridges and troughs, the moisture and motion of the atmosphere.

ON MONDAY at 3:30 p.m., based on the results, the National Weather Service issued this prediction for the Washington area yesterday:

A 40 percent chance of light snow. ... Total accumulation less than one inch.

In the cold and snowy light of yesterday morning, with the aid of 20-20 hindsight, Dewey Walston, a meteorologist with the Weather Service, had something to get off his chest about those computer models. He typed up a post-mortem addressed to other forecasters.

Wow, Walston wrote. Eating a lot of humble pie this morning. ... Too bad the [models] cant answer all the phone calls.

Walston went on to observe that the Monday morning computer weather assessment, predicting conditions for yesterdays rush hour, were the most horrible I have seen in my 10 [years] in the NWS.

The models, he added, did a pitiful job.

The models are run several times a day, and data from a new computer run became available in the early evening, predicting the storm much more accurately. After analyzing the results, federal forecasters issued the first weather alert for Washington before 10 p.m. More alerts quickly followed for cities up the Eastern Seabord. Far from less than 1 inch, this was the worst snowstorm since 1996, a bomb  to use a weather term. It was the sort of winter northeaster that has dumped snow over such a large area only 28 times in the last half-century.

In a day of meteorological hand-wringing and second-guessing, federal forecasters admitted they and their computers had been late in calling it right. But they denied that they had been anything less than thorough in tracking the storm.

I wouldnt characterize it as catching us off guard, said Louis Uccellini, director of the National Centers for Environmental Prediction, a branch of the Weather Service. He noted that the medium-range forecast pointed to the possibility of a significant East Coast storm by Wednesday. We were watching this thing like a hawk.

TRUSTING COMPUTERS

But Patrick Michaels, the state climatologist for Virginia, suggested that a subtle bit of human psychology had foiled the forecasters. He said that one of the computer models, called Eta (sounds like ate-a) after the Greek letter, is so good (it nailed the storm of January 1996 by more than a day), that forecasters are loath to contradict it. So they didnt, even though they could see radar evidence Monday of heavy late-afternoon snow in Raleigh, N.C., and other contrary data.

Here we have the best model we know of insisting that the main precipitation shield is to the south and east of Washington, Michaels said.

And here we have our eyeballs looking at the precipitation shield advancing north and west.

The forecasters  including Michaels  didnt believe their eyeballs. The moral of this snowstorm, Michaels suggested, should be:

The computer Etaed my forecast.

Uccellini, whom Michaels praised as knowing more about East Coast snowstorms than any living human being, pointed out that the Weather Service issued the more accurate forecast hours before people had to get up and face the day. The only alterations through the night were to add a few inches to the total predicted accumulation.

But the Weather Services revision of the analysis still came too late to make the early evening broadcast news, the last time many people tune in to the weather. The bad news came later.

Sue Palka, weathercaster at WTTG-TV (Channel 5), went on the air for Mondays 10 p.m. news predicting eight inches and then spent yesterday doing live updates every half-hour. Its been insane today. My back is killing me, she said late yesterday. I just realized an hour ago that I hadnt eaten all day.

Sometimes the computer models disagree with one another. Thats when a meteorologist proves his or her mettle, sorting out the conflicting data, said Andy Woodcock, a meteorologist with the Weather Service. But on Monday, forecasters had three key models in essential agreement, all predicting the storm would pass far to the east of the major East Coast cities. To reject a unanimous chorus of computer models is the kind of stuff that gets you awards if youre right, Woodcock said. Youre a gutsy dude if you do it. Advertisement

No gutsy dudes were on the case Monday.

So why didnt the models catch this earlier?

Id love to have two or three years to do a research project on that, Uccellini told a group of reporters during a telephone news conference.

The models had predicted that a weather system over the Great Lakes would help push the East Coast storm out to sea, but that did not happen. Chris Davis, a meteorologist with the National Center for Atmospheric Research, a consortium of universities and other institutions, said he suspects that the raw data going into the computer models from balloons, aircraft and satellites may have been insufficient. Surprisingly strong snow in Raleigh and other evidence independent of the computers raised some concern among the Weather Services forecasters in Sterling as well as meteorologists at AccuWeather, the private service based in State College, Pa., that supplies many Washington clients, including the D.C. government.

Uccellini said the new supercomputer, recently unveiled with fanfare, helped matters. Its first run of enhanced software happened to be Monday morning. The supercomputer detected some possible problems with the original forecast, he said, and by evening, its superior number-crunching ability helped pull together the revised forecast more quickly.

Another problem in the prediction was that the standard ratio of water to snowflakes did not apply. Usually, when forecasters know how much water is coming down, they use a ratio of 1 inch of water for 10 inches of snow to predict how much snow will fall. But this storm was more intense, 1 to 20.

Its what makes snow forecasting so much fun, Uccellini said.

This storm was a northeaster, with winds that were strong but not blizzard-strength, except for brief gusts. It is also sometimes called a bomb, Uccellini said. That signifies the explosive development of a storm off the coast accompanied by a quick drop in central pressure.

Such storms that dump snow on all the major cities of the East Coast tend to be episodic, Uccellini said.

They come in groups, he said. You can go without them for four or five years, and then you can have two or three happen in one or two years.

) 2000 The Washington Post Company

-- rocky (rknolls@no.spam), January 27, 2000.

Weather Forecasters and Economists...the only two fields where you can be considered an expert even though you have never been correct in a prediction.

-- JB (noway@jose.com), January 27, 2000.