Enhancing Weather Information with Probability Forecasts
(Adopted by AMS Council on 13 January 2002)
Bull. Amer. Met.Soc., 83
Much of the informational content of meteorological data, models, techniques, and forecaster thought processes is not being conveyed to the users of weather forecasts. Making and disseminating forecasts in probabilistic terms would correct a major portion of this shortcoming. It would allow the user to make decisions based on quantified uncertainties with resulting economic and social benefits. Widespread implementation of probability forecasts would require forecasters to become more familiar with user needs, and users to be educated on probability forecasts and how to make optimum use of this new information. The American Meteorological Society endorses probability forecasts and recommends their use be substantially increased.
Current situation Weather forecasts have improved dramatically over the past few decades and particularly in the last 20 years (Bull. Amer. Meteor. Soc., 79, 2161-2163). Forecasts produced by operational forecasters using the new observational data and results of improved numerical models have become more accurate at practically all time- and space scales for all weather elements. These forecasts are valuable in daily operations to users, including the general public, the military, aircraft operators, businesses, and emergency managers. These forecasts contribute very useful and often critical information for decision making.
However, there is much more information available than is being provided to users. Present-day forecasts are predominantly "categorical" in that the uncertainty inherent in the forecast is not made explicit. To make this information available would require that the uncertainty be quantified and put into understandable terms. This quantification would almost certainly involve numerical probabilities.
A probability forecast of a weather event can be a forecaster's judgement of the likelihood that the event will occur. Probability forecasts can also be produced directly from numerical models and postprocessing of model results, and climatological forecasts can be expressed in those terms. All these forecasts have the element of uncertainty, but such forecasts should be considerably more useful to a user than categorical forecasts, and perhaps critically so.
Some progress has been made, especially in recent years, in providing forecasts in probabilistic terms. Forecasts of probability of precipitation (PoP) have been made for over 30 years and are well accepted by a large clientele. More recently, probability forecasts are issued routinely by the National Centers for Environmental Prediction for a variety of weather phenomena, such as tropical cyclone "strike probabilities" and intensity; convective outlooks; heavy snow/icing outlooks; and 6-10 day, 8-14 day, monthly, and seasonal outlooks for temperature and precipitation. National Weather Service forecast offices and river forecast centers are beginning to produce probability forecasts of river stage, volume, and flow. However, these probability forecasts are still only a small fraction of all forecasts issued.
Effective use of probability forecasts requires that users understand them. The probability of an event is a familiar concept. For instance, what is the probability a die roll will be a 6; what is the probability Dasher will win the race; what is the probability I will win the lottery tomorrow? Though the exact value of the probability may not be known, the concept is familiar and understandable. A misunderstanding that is often encountered regards the definition of the event. For example, for a forecast of 30% PoP for Boston tomorrow, a person may be unsure as to whether that means it will rain over 30% of the Boston area tomorrow, it will rain for 30% of the time tomorrow somewhere in Boston, there is a 30% probability it will rain somewhere in Boston tomorrow, or some other interpretation. Widespread use of probability forecasts will require significant efforts to educate the user in the definition of the event being forecast.
Opportunities An operational forecaster is provided considerable guidance for making probability forecasts. The statistical postprocessing of the output of dynamic weather prediction models can and does provide well-calibrated (reliable) probability estimates. The maturing technology of ensemble forecasts can also provide, with minimal postprocessing, probability estimates of specific weather events, such as precipitation amount for a given interval of time at a specific place being over, say, 0.25 inches; some forecasts of this nature are now being made available. Such "objective" forecasts are statistical estimates of the conditional relative frequency of the event. The relative frequency of a die roll producing a 6 is known to be one-sixth, under the assumptions of a fair die and a fair roll, but if we did not know this or suspected a loaded die, we could determine the probability of a 6 for that specific die by repeated rolls. This calculated relative frequency would be an estimate of the probability of the event, and in this case, a very good estimate, provided the number of rolls was large. Numerical weather models and their postprocessing do not yet produce as good an estimate for weather events, but they do provide useful results up to a week in advance, and are continually improving. With appropriate statistical processing, the objective probability forecasts will blend into climatological relative frequencies at the long range, which are also useful to some users.
Probability forecasts offer several benefits over categorical forecasts. They contain more information, because the uncertainty in the forecast is specifically expressed; the user is made aware of that uncertainty and can use that information in decision making. Probability forecasts can be used with thresholds to make decisions, where the thresholds can vary from user to user and purpose to purpose. Availability of probability forecasts would allow users to make the go/no?go decision based on quantitative uncertainties and his/her own threshold for making the decision. For instance, a school superintendent in a hilly area might cancel school with a lower probability of 2 inches or more of snow than one in a flat area where the journey to school in snowy conditions would be less dangerous.
Too often, the roles of the forecaster and the decision maker are confused, or blended into the forecast itself. Specific probability forecasts allow the roles to be separate, as they should be. A probability of 10 percent that flood waters will overtop a levee may influence one merchant to move stock to higher ground, but another, possibly because of the high cost of moving, may not move stock until the probability is 20 percent. Given only a categorical forecast (e.g., the crest will be 6 inches below the levee top), the user may choose to ignore it, to form his/her own probability of the overtopping to occur, or to base the operational decision on other information, but does not have the opportunity to use the full extent of information available. Entire municipalities may be lulled into inaction when there is in reality a significant chance that the town is in danger.
Probability forecasts would have significant economic benefits for the nation. Because a significant portion of the economy is weather sensitive, a new economic sector of weather risk management has come into being. This management industry provides a "hedging tool," allowing companies to even out their weather sensitive costs. Better management by these companies benefits the general public in the form of lower cost for commodities, such as power. Since this is a growing industry, the increase of probability forecasts at this time is especially appropriate.
Challenges A number of challenges arise with the use of probability forecasts. The examples cited above deal with single specific events; however, many forecast decisions are complex. For instance, for quantitative precipitation, one would like a probability distribution such that a probability of any desired amount, say 0.5 to 1.0 inches, could be obtained for any desired time interval. Forecasters will need to be educated to handle these complexities. New ways for displaying and communicating probabilistic information are needed. Users must be educated on how to make optimum use of this new information. Although the concept of probability should be known, the actual use of the information may be challenging. Forecasters need to be aware of the specific user's needs (e.g., emergency managers) and help to devise methods and models for the user to employ in formulating plans of action. Guidance forecasts of weather variables will become even more important and must be communicated in probabilistic terms to operational forecasters so that they can make the best possible forecasts. Care must be taken that these forecasts are well calibrated. For instance, ensemble forecasts generally do not span the full range of meteorological possibilities, and probabilities estimated as relative frequencies directly from them may be too sharp.
Summary In general, present day weather forecasts do not contain a quantification of the uncertainty that is inherent in them. Probability forecasts based on the forecaster's thought processes and/or available models and techniques would substantially benefit users of weather forecasts. Successful implementation and use of probability forecasts will require forecasters to understand user needs for this information and to be trained in how to best use the guidance produced by models to make probability forecasts. Similarly, users of weather information must be trained in how to best interpret and use this valuable resource - probability forecasts.
About the AMS | Policy Program | Conferences,
Meetings, and Symposia ]
[ Education Programs and Resources ]
[ History of Earth Sciences | Journals and Publications | Local Chapter Information | Member Services ]
[ News and Information | Frequently Asked Questions (FAQs) ]
[ Disclaimer | Contacts at AMS | Email AMS Web Administrator ]
on Logo to Return to AMS Home Page
© 2002 American Meteorological Society
Headquarters: 45 Beacon Street, Boston, MA 02108-3693
Phone: 617-227-2425; Fax: 617-742-8718