All year round, Washingtonians are an exacting bunch. But as the mercury falls, the public's expectations of weather forecasts seem to rise.

Predictability is important, of course. Each evening, you want to know that the electricity will be on for the morning coffee, that the car will start and what tomorrow's weather will be. Occasionally, the power goes out, the car doesn't start and the forecast is wrong. More and more often, though, it is right. Although there are limits to what we forecasters can predict, today's 24-hour forecast is correct about 90 percent of the time, and the two- to three-day outlook is as accurate as a 24-hour forecast was 10 to 15 years ago. Yet I sometimes feel that the public's expectations have increased even more than our accuracy. That's a bit of a quandary for those of us who make a living predicting the future.

Forecasting, especially in Washington winters, is a mix of meteorology and psychology. As a meteorologist at Channel 4, I'm concerned not only with storms, but with people's reaction to my forecast, especially if it includes the "S" word. I know that any time snow is predicted, everyone will rush to the grocery store, expecting they'll be marooned at home whether there are two inches or 20. I want to get it right. But television is a competitive medium, and I have an eye on the other channels. I certainly don't want to underestimate a storm that the other guy is predicting. On the other hand, I don't want to be crying wolf.

In Washington, the real trick is to be right about water, which is probably the most important variable. I guarantee it's the most important in winter, because in our region it can exist as a liquid (rain or clouds), a solid (snow, sleet or ice) or a gas (humidity)--at practically the same temperature. That rain/snow/sleet mix is often right on the edge and hard to predict. That's why Fredericksburg's weather can be entirely different from Frederick's.

I've heard more than once that meteorology is an inexact science. It is no less exact than any other science, but it is the science that involves more predictions than any other. It's the nature of those predictions that's different.

I've also heard, "If we can send a man to the moon, why can't we predict tomorrow's weather better?" It might surprise you, but I would argue that sending a man to the moon is more straightforward than predicting tomorrow's weather--much of the time. There are two reasons. Unlike the movement of planets, the moon, the sun, a ball thrown into the air or a clock's pendulum, the atmosphere is not a linear system. A tiny weather event such as a small puffy cloud forming in Colorado can become a monster cluster of thunderstorms in Kansas that produces tornadoes and causes widespread damage. But it's impossible to measure and track every puffy cloud in the world at the same time.

The second reason has to do with the "butterfly effect"--the concept of chaos theory proposed by MIT meteorologist Edward Lorenz--that a slight difference in initial measurements can have a dramatic effect on results. Lorenz asked, "Does the flap of a butterfly's wings in Brazil set off a tornado in Texas?" The atmospheric magnification of small events and the sensitivity to initial conditions do more than make forecasting a challenge. Ultimately, they limit our ability to predict.

Meteorology is perhaps the only science that people put to the test every day. But in my experience, it is still little understood.

I am a meteorologist trained in physics who has been forecasting the weather on TV for 30 years, and I still need to begin my workday many hours before I go on the air. Preparing a forecast is like examining a patient--looking at blood counts and X-rays, measuring blood pressure, taking a history and eventually offering a diagnosis. I also look at pressure readings, temperatures, X-rays (okay, radar images) and other things to get a picture of the day's atmospheric health. Then I look at mathematical/physical models of the atmosphere, which give me a picture of large-scale meteorological features as they move and develop . . . but always with that darned butterfly flying around. I am also guided by local climatological statistics and geographic variables. I rely, too, on my decades of experience. During an August heat wave, putting together a forecast may take less than an hour. If a major snowstorm is approaching, I may be working at it until seconds before I'm on the air. Then it's a question of communicating the science in accessible and human terms: the psychology of meteorology.

Many folks don't really want to know tomorrow's forecast. They want someone to tell them what to do. Should they stay home? Is it safe to drive? Will school be canceled? Should they do their homework? (The last one is easy; always do your homework.) There's a balancing act involved in giving people information they want and can use as part of an objective forecast. For example, I sometimes talk about snow being "conversational" (a dusting) versus "shovelable" (three to five inches), to convey how the weather will affect viewers personally.

Once on a cold winter morning, a woman who was planning a trip to Pennsylvania called to ask me about the weather there. I told her what I thought it would be that day--flurries in the mountains, cold and windy. "How much snow?" she asked. No more than a dusting to an inch or two, I said. "Will it be slippery?" she inquired. "What will the roads be like?" I could tell I wasn't giving her the information she really needed so I said, "Why not wait until tomorrow? It will be partly sunny." "Oh good," she said. "That's what I wanted to hear."

"Partly sunny" is a perfect example of where we could really do a better job of forecasting and communicating. People often ask me about the difference between partly sunny and partly cloudy. There really isn't one. I could give much more useful information--the sun will shine tomorrow 60 percent of the time--but that would involve math and the laws of probability, which is something I'm not sure viewers or news directors want, despite the fact that most folks cope with these mathematical concepts in other realms: the odds in Las Vegas or the point spread of the Sunday game.

Some of us are more guilty than others of stretching the scientific limits of forecasting. Not long ago, I saw a TV meteorologist in Utah who was forecasting specific temperatures 10 days out, even though accuracy of that kind is impossible. There are also popular myths that confuse what the science can do, and day-by-day predictions months in advance like those in farmers almanacs that perpetuate those myths.

But no matter how much I invoke science and probability, sometimes I'm just plain wrong--and everyone seems to remember when I am wrong. How would you like to have a job where every once in a while you know you're going to make a mistake in front of thousands and thousands of people? Before the big snowstorm last March, I predicted we'd get a dusting to one or two inches of snow. We had eight inches. After the storm, boy was I kidded. I thought the viewers deserved an explanation, so I described on air what went wrong and why, and tried to use it as a learning tool.

The limits of predictability were demonstrated last summer as millions were evacuated from the East Coast because of Hurricane Floyd. The most probable track of the storm was correctly forecast: The storm would remain offshore south of North Carolina, but there was a mathematical chance that it would veer to the west and hit Florida, Georgia or South Carolina. Probability entered into the decision to evacuate, but we meteorologists did not communicate that well. The evacuations were in part meteorological, in part risk management and in part political decisions. The next time a mass evacuation is a possibility, meteorologists should make clear to the public the odds of a storm's landfall, and the limits of predictability, so that people can play a part in the deciding whether to evacuate.

The limits of predictability make me think I should include a confidence index in my forecasts. Sometimes I give a guarantee when I am almost certain about the next day--but never where snow is concerned. On those nights, who knows, I may start saying something like this: "Tomorrow, I'm 80 percent confident that it will snow. There's a 10-percent chance of a dusting or up to two inches, a 60-percent chance of two to four inches and a 30-percent chance of four to six." That sounds like I'm just giving myself more wiggle room, but isn't that kind of information more useful than merely saying it's going to snow? In any case, here's one prediction I'm sure about: Whatever my snow forecast is, there's a 100-percent chance of a rush on milk at the Giant.

Bob Ryan, a past president of the American Meteorological Society, has been with Washington's Channel 4 (NBC) for 20 years. If his forecasts this winter are more right than wrong, he's hoping to make that 21.

Highlights in Modern Forecasting

1800s

Weather charts are created at the Smithsonian Institution in 1849 from reports sent via the mail, and later by telegraph; kites with meteorgraphs are used to probe the atmosphere; In 1870, Congress passes a joint resolution requiring the war secretary to set up weather monitoring at military installations, thereby creating the U.S. Weather Bureau.

1919

Norwegian meteorologist Vilhelm Bjrknes discovers that interacting air masses generate cyclones. From his study of cold- and warm-air clashes, storm "fronts" are recognized; the Weather Bureau begins observations by aircraft.

1921

The University of Wisconsin begins broadcasting weather reports by radio to farmers; two years later, Weather Bureau reports, forecasts and warnings are transmitted by 140 radio stations in 39 states.

1928

U.S. commerical airlines jointly hire their first weatherman; MIT opens the country's first university meteorology program.

1934

Franklin D. Roosevelt asks the National Sciences Academy to investigate weather-induced airplane crashes; a year later, a hurricane warning service is established.

1941

The Extended Forecast Division of the Weather Bureau begins in Washington; during World War II, five-day forecasts are issued to the military services twice a week.

LATE 1940s

The first local TV weather reports are broadcast; radar is used to study and track thunderstorms; the first systematic experiments to investigate the physics of precipitation are conducted over the skies of Massachusetts by American chemist and meteorologist Vincent Joseph Schaefer; two USAF Air Weather Service meteorologists learn how to predict the likelihood of tornado development.

1952

ENIAC, one of the first electronic digital computers, takes five minutes to produce a 24-hour weather forecast. Within three years, the Joint Numerical Weather Prediction Unit is turning out routine computer-generated forecasts, which prove to be 35-percent accurate.

1959

The first meteorological satellite sensor is launched; a year later, the world's first weather spacecraft goes up (TIROS I) and is operational for 78 days, sending back images of shifting cloud cover, frontal systems and wind fields; the American Meteorological Society (1920) begins a seal-of-approval program to provide a measure of competence for weather forecasters.

1966

Geostationary satellite goes into orbit 22,000 miles over the equator, sending the first full views of Earth.

1970s

Satellite cloud pictures are used during TV weather forecasts, with weather forecasters drawing highs, lows and fronts by hand.

1980s

Computer graphics allow for use of color images and animation in weather forecasts; in 1982, a former ABC-TV "Good Morning America" weatherman launches the Weather Channel.

1990s

Doppler radar, developed during WWII and improved extensively through the 1970s, is installed at National Weather Service stations across the country.

Sources: National Weather Service, Weatherwise magazine, Encyclopedia Britannica