Number 31, December 2001

Student Editorial

The Need for Probablistic Forecasting

Now that autumn has given way to winter, the meteorological community is once again presented with the opportunity to both promote and practice probabilistic forecasting. I'm sure many meteorologists reading this article will think, "Probabilistic forecasting? Every other conference I go to has a talk on probabilistic forecasting! Why is the WeatherZine's Student Editorial Editor wasting a column talking about something that is already common knowledge in the meteorological community?"

Before I answer this question, let me first explain probabilistic forecasting for those not familiar with the term. As any forecaster knows, there are days when forecasters are rather sure of what will happen, and there are days when forecasters are rather uncertain. A probabilistic forecast includes not only information regarding the expected weather for the near future, but also the degree of certainty that the forecaster has in the forecast. Some broadcast meteorologists will note uncertainty in the timing of a frontal passage, or how far a swath of heavy precipitation will stretch. Some, in an attempt to assure viewers that they are the best meteorologist in the market, will not make such statements for fear that uncertainty will be mistaken as incompetence. The forecasters that truly provide the greatest service to their viewers or clients frequently share information regarding uncertainty. Unfortunately, this information is shared too infrequently.

The need for probabilistic forecasting, and the consequences of not meeting this need, were quite evident in the aftermath of the Nor'easter of March 4-6, 2001. On March 3, based on forecasts of a record storm, the entire East Coast was gearing up for the event. Municipalities had workers stay overtime to handle problems that would arise from the storm. Factories were shut down for the first time in years. In a race to be the first with the story, broadcast meteorologists warned of a near-record blizzard and told viewers to stock up on supplies and stay tuned to their local news. What these broadcast meteorologists failed to tell their viewers was that the track of the storm, as well as the location of the rain-snow line, was quite uncertain, and that in an area as populous as the North Atlantic, a fifty-mile margin of error could make the difference between getting a foot of snow or an inch of rain. One could also argue that communication from the National Weather Service contributed to this misplaced sense of certainty in the forecast process.

While the model forecasts correctly predicted the intensity of the storm, they incorrectly predicted its location. The majority of snow in fact ended up in New England and avoided major populations. The New York metropolitan area lost millions of dollars, leading Byram Township, New Jersey, mayor Richard Bowe to threaten to take local broadcast meteorologists to court. Said Bowe, "People who give the wrong information should be held accountable for losses sustained by those who follow that information." National Weather Service meteorologist Bill Goodman responded, "We use our knowledge of the weather and computer models to make our best judgment. This is not a perfect world." That is certainly true, but while the meteorological community is aware of its limitations, it does not always communicate those limitations to the general public. This is where probabilistic forecasting can improve the meteorological community's ability to serve the public.

So back to the question posed earlier in this article. Why discuss a need for probabilistic forecasting when readers of this column likely already recognize this need? The answer is simple. The greater meteorological community recognized a need for probabilistic forecasting long before March 2001, but that hasn't led to a significant change in the forecasting process. To the degree that change has occurred, it has been relatively minor with respect to decision makers' needs for the best information that the forecasting community can provide. One must conclude that this is either the result of the greater forecasting community’s failure to recognize the need for probabilistic forecasting, or forecasters simply ignoring this need for some reason. To provide such information to viewers and clients of professional forecasters, the meteorological community must facilitate a transition to probabilistic forecasting, whether through research, product development, or simply dialogue, as we await the first Nor'easter of 2002.

Those of us concerned with the societal impact of weather must do what we can to develop the knowledge and means necessary to issue forecasts that indicate the appropriate level of uncertainty. In the larger picture, we must continue to bridge the gap between our respective research areas and the situations in which our research can make a difference in people's lives.

Russ Chibe
Department of Atmospheric Science,
Colorado State University
rchibe@atmos.colostate.edu