In the poker game of long-range weather forecasting, Dr. Donald L. Gilman is the National Weather Service's version of the old-time riverboat gambler.

Last year, for instance, the chief of the Weather Service's prediction group went out on a limb and bet his shirt that the East and West coasts would experience milder-than-average winter temperatures.

"That wasn't one of our better efforts," Gilman will acknowledge, his self-confidence seemingly unshaken even after he watched most of the country battered last winter by unexpected cold and frost. Like any serious gambler, Gilman sees occasional losses as just another part of the tricky game of predicting what the weather will be.

His predictions -- particularly his annual long-range winter forecast -- can help determine the routes ships will take, the funds local governments will set aside for snow removal, the schedules for new building construction. But to hear Gilman speak of the uncertainties of his own craft, unswerving faith in the weatherman's words seems woefully misplaced.

"It is in some ways like detective work," Gilman explained in an interview. "It's a bit like old-fashioned medicine. It is scientific, but it is not a strict science. It's somewhere between physics and economics. Among the physical sciences, it is somewhat soft."

His detective's clues are in the atmosphere: dust clouds that form after volcanic eruptions, sea temperatures, upper-level winds and pressure patterns. There is also a dash of seat-of-the pants intuition, and a belief that Mother Nature, like the murderer in a Mickey Spillane thriller, will behave in certain predictable ways, based on past experience.

"There's some tendency to consistency," Gilman said. "If North Dakota was warm in the fall, there's a pretty good chance it will be warm in the winter." But he cautioned, "There aren't any strong, dependable relationships -- only weak ones."

Reaching for yet another analogy from his seemingly endless repertoire, Gilman added, "It's like trying to play the horses when you don't know everything about the horse. You may know his speed and his track record, but you may not know how the horse is feeling that particular day."

After 26 years with the National Weather Service, the boyish-looking 53-year-old forecaster has become something of a national celebrity. It's an unexpected role for a meteorologist who wrote his 1957 doctoral dissertation on "Empirical Orthogonal Functions Applied to 30-Day Forecasting."

Now Gilman is busy fielding telephone calls from NBC News, posing for photographs for U.S. News and World Report and planning for his news conference this morning at which he will unveil the Weather Service's annual winter forecast.

His division issues detailed 30-day and 90-day forecast bulletins that subscribers use to help determine everything from the crop outlook for the Nile Valley to the price of soybeans on the commodities index. But nothing attracts the nation's attention quite like the extended winter weather forecast, and his annual press conferences, he says with a slight smile, are akin to a circus.

The prediction is based on a consensus among the branch's five forecasters and often turns out to be the polar opposite of forecasts from the Old Farmer's Almanac, the Scripps Institution of Oceanography and other climatologists.

In November 1982, Gilman said the eastern United States had a 60 to 65 percent chance of experiencing a warmer-than-average winter, while most of the country would be wetter than usual. The bold prediction flew in the face of the Farmer's Almanac's dire warnings of "incredible" cold. It also ignored that fall's observations of unusually long hair on the woolly bear caterpillar, which historically has presaged a bad winter.

When it was over, Gilman was close to the mark, and the almanac and the caterpillar were wrong.

Last year, however, it was Gilman who was forced to eat the proverbial crow after he predicted another winter with milder than usual temperatures on the east and west coasts. The last two weeks of 1983 were brutally and unexpectedly cold -- with new lows in Florida damaging citrus crops. "The last two weeks of December were so extreme they colored the rest of the winter," Gilman said.

The veteran forecaster makes no apologies, however. Instead he points to last winter as an example of how his is not a science of certainties, but an art of probabilities. The long-range forecasts are never flat predictions of more cold or precipitation, but are always couched in terms of percentages -- a 60 percent chance of more cold, or a 70 percent chance of precipitation. Reporting in probabilities, he said, "brings people right up against the limitations of what they're getting."

"I'd rather not call that guessing," he added quickly. "It's more than that. What we're working with are weak indications -- hints."

Many times, the long-range predictions from Gilman's branch differ from others because he has no use for sun spots, the telltale weather sign for many forecasters. "Sun spots we don't touch at all," he said. "I think the evidence is much too shaky there. No simple connection has been shown between sun spot variation and the weather."

But that doesn't make Gilman a traditionalist. In fact, he said he looks forward with relish to the day when mathematical models fed into sophisticated computers can produce a more accurate accounting of future weather patterns.

Already the so-called "numerical weather predictions" can be used with some accuracy for short-term forecasting. He predicts it will be at least another decade before mathematical equations can be developed that would be of substantial help in long-range forecasting.