The following statement(s) have expired and are here for historical purposes and do not represent statements of the AMS that are “in force” at this time.
(Adopted by AMS Council 14 January 2001)
Bull. Amer. Met.Soc., 82, 701
The skill of seasonal climate prediction has improved substantially over the past two decades, largely in response to increased understanding of the El Niño/Southern Oscillation (ENSO) phenomenon. Routine, scientifically based, skillful, seasonal climate forecasts are now possible for some parts of the world, for some seasons. These seasonal climate predictions are able to project the mean conditions and some of the statistical characteristics of the climate a season or two in advance. These seasonal predictions are primarily of use to organizations that have a decision-making process that can intelligently use probabilistic input and that are engaged in activities that are sensitive to seasonal climate variations and involve significant economic stakes.
Parallel with these important advances in seasonal climate prediction is the increased understanding of the limit of predictability of day-to-day weather changes, which theoretical and experimental studies set in the range of 1–2 weeks. Thus, while it may be possible to predict certain statistical properties of the climate for the next season and for the season after that, there is no scientific basis for the deterministic prediction of day-to-day weather beyond a week or two. Claims of skillful predictions of day-to-day weather changes beyond this limit have no scientific basis and are either misinformed or calculated misrepresentations of true capabilities. The AMS Statement on Seasonal to Interannual Climate Prediction, which follows, provides an outline of the current capabilities in seasonal climate and the prospects for future improvements.
The "climate" for a season is commonly defined by the average and other statistics related to temperature or precipitation over a region for a prescribed three-month period. Scientifically based seasonal climate prediction (e.g., prediction of the coming season's average temperature and precipitation) has come of age over the past two decades. Until 20 years ago, seasonal climate predictions were based exclusively on empirical/statistical techniques that provided little understanding of the physical mechanisms responsible for relationships between current conditions and the climate anomalies (departures from normal) in subsequent seasons. Empirical (statistical) tools are still utilized for seasonal prediction, but now within the context of a greatly improved understanding of the underlying physical processes. Mathematical models analogous to those used in numerical weather prediction, but including representation of atmosphere–ocean interactions, are being used to an increasing extent in conjunction with, or as an alternative to, empirical methods. The purpose of this statement is to provide an overview on the current state of understanding of seasonal to interannual climate variability and of the current state of the art of climate prediction in this time frame to the nonspecialist and the general public. Climate change and climate variations over several years or tens of years (often termed decadal climate variability) are not explicitly discussed in this statement.
2. Seasonal climate predictions
a. Limits of extended weather forecasts
Weather refers to the day-to-day state of the atmosphere (e.g., the position of fronts and individual storms, daily precipitation and temperature, as well as humidity, winds, and sea level pressure). Meteorologists have long had the goal of extending accurate predictions of day-to-day weather out to several days and perhaps even to weeks, months, and seasons in advance. While great progress has been made on forecasts for timescales up to almost a week, all current applied and theoretical studies lead to the conclusion that the day?to?day fluctuations in weather are not predictable beyond one to two weeks. Beyond that time, errors in the initial conditions (i.e., the data that define the state of the atmosphere at the start of the forecast period) no matter how small they are assumed to be, grow to the point that they overwhelm whatever valid information that the forecast might contain. In the late 1960s MIT atmospheric scientist Edward N. Lorenz discovered this extreme sensitivity to initial conditions in his efforts to produce extended forecasts of the behavior of a highly simplified model of the atmosphere run on a computer. This so called "chaotic" behavior has subsequently been shown to be an inherent property of nonlinear systems like the atmosphere.
b. El Niño/Southern Oscillation (ENSO) and the basis for seasonal climate prediction
While nature has conspired to limit our ability to forecast day-to-day weather, there exists a firm scientific basis for the prediction of seasonal mean climate anomalies (i.e., departures from normal of averages and other statistics of weather over a season or longer). Seasonal climate anomalies result from complex interactions between the atmosphere and the underlying surfaces: that is, the world oceans and land surfaces. The atmosphere, which fluctuates very rapidly on a day-to-day basis, is tied to the more slowly evolving components of the earth system, which are capable of exerting a sustained influence on climate anomalies extending over a season or longer, far beyond the 1–2 week limit of deterministic atmospheric predictability. The atmosphere is particularly sensitive to tropical sea surface temperature anomalies such as those that occur in association with ENSO.
The interaction, or coupling, between the oceans and the atmosphere at seasonal timescales was first appreciated and understood in the context of the ENSO phenomenon: the episodic warming and cooling of the sea surface temperature in the equatorial central and eastern Pacific (popularly known as "El Niño" and "La Niña" events). However, ENSO is not solely an oceanic phenomenon. In addition to the surface and subsurface of the equatorial Pacific Ocean, ENSO also encompasses the ocean's interactions with the global atmosphere on timescales of several seasons. The first successful computer models of the ocean–atmosphere interactions associated with ENSO were performed in the 1980s. Since that time increasingly sophisticated and realistic computer models have been developed in support of seasonal climate prediction.
Slowly evolving land surface conditions (i.e., soil moisture, vegetation, and snow cover) are also believed to feed back upon the atmospheric circulation, but these effects have proven more difficult to simulate in computer models and incorporate into climate prediction schemes.
3. Current Practice in seasonal prediction
Most of the current seasonal predictive skill over many regions of the world comes from the strong influence of ENSO. State?of?the?art statistical models and computer models that link the tropical oceans with the global atmosphere have demonstrated measurable predictive skill on seasonal timescales. These models are routinely used to predict sea surface temperature anomalies in the tropical Pacific. The predicted sea surface temperature anomalies can, in turn, be linked to seasonal temperature and precipitation anomalies over the globe. These empirically derived "teleconnections" can be simulated in global atmospheric models forced with prescribed sea surface temperature anomalies. By exploiting them, it is possible to provide useful seasonal forecasts for some regions of the world out to one season, two seasons and, in some cases, even out to three seasons in advance with a level of skill significantly better than statistical forecasts based on persistence. These forecasts of seasonal mean climate anomalies can, in turn, be linked to changes in the likelihood of extreme weather events that tend to occur more or less frequently during El Niño years than during La Niña years. Examples include intense winter cyclones in California and hurricanes.
Short-term weather variability (which, as mentioned above, is unpredictable beyond the first week or two) introduces uncertainty or "noise" into the seasonal statistics. For example, a strong Arctic cold air outbreak or a heavy precipitation event associated with a particular storm can significantly affect the seasonal mean temperature (or precipitation total) regardless of the sea surface temperature anomalies. Because of this weather noise and the lack of understanding of all of the components of the climate system, seasonal climate predictions are inherently probabilistic. In contemporary state-of-the-art seasonal climate predictions, computer models of the atmosphere are generally run in groups, or ensembles, of 10 or more for a given sea surface temperature forecast. For each member of the ensemble, the model is run with slightly different "initial" conditions. As discussed above, the influence of these arbitrary initial conditions are lost after a week or two. The differences among the ensemble members give forecasters some measure of the likelihood that a particular seasonal climate state will be above, near, or below normal. Typically these categories are defined such that there are equal long-term probabilities that the temperature and precipitation will be above, near, or below average—that is, 33% probability for each category, expressed as (33, 33, 33). Seasonal temperature and precipitation forecasts are couched in terms of shifts in these equal probabilities to favor one or two of the categories. For example, a typical seasonal forecast might read (45, 30, 25) which would be interpreted as a 45% chance that the seasonal mean temperature (or precipitation) would be in the upper third of the historical temperature (precipitation) distribution, a 30% chance that that the temperature would be in the middle third, and a 25% chance that the temperature would fall in the lowest third.
The advances in the skill of seasonal climate prediction over the past 20 years are attributable to 1) major advances in empirical and theoretical understanding of the ocean and the atmosphere and how they interact; 2) the development of computer models to simulate and predict these processes, and particularly those associated with ENSO; and 3) improved in situ and satellite observations, especially over the equatorial Pacific.
The demonstrated climate prediction skill associated with ENSO leads many climate researchers and forecasters to believe that advances in climate prediction will come from a better understanding and simulation of teleconnections involving the other ocean basins and from the inclusion of land surface conditions in climate prediction models. The importance of the other ocean basins in determining the mean seasonal climate has been demonstrated, for instance, in prediction studies for Brazil and western Africa, where sea surface temperature anomalies over the tropical Atlantic Ocean have been shown to play an important role. Sea surface temperature anomalies over the Indian Ocean have some influence on seasonal climate in eastern Africa, Southern Asia, and Australia. Routine forecasting of global sea surface temperatures is now being attempted at a small number of forecast centers, and research is under way to determine whether land conditions can be predicted as well. Other modes of atmospheric circulation variability, in particular the mode variously known as the North Atlantic Oscillation (NAO) or the Arctic Oscillation (AO), also exhibit variability on the seasonal-to-interannual time frame that might represent a potential source of predictability.
4. Climate predictability on other timescales
The feasibility of seasonal forecasting depends on the fact that over a season, the effects of shorter term (weather) events tend to average out, revealing the smaller but more consistent influence of the ocean and land surface on the atmosphere. Prediction in the intermediate range of about two weeks to two months is rendered more challenging by the presence of a higher "noise level" imposed by the inherently unpredictable day-to-day atmospheric variability, which is less completely averaged out. Hence the forecasts in the range from several weeks to less than a season tend to be less skillful than seasonal climate forecasts. Nevertheless, evidence is mounting that under some conditions—strong regional boundary forcing, for example, by the non-ENSO-related ocean or the land surface—useful skill may be realizable on shorter climate timescales. This aspect of the climate predictability problem is the subject of active research. One hope for improving predictions at these timescales lies with further understanding and an improved capability to model a mode of tropical climate variability with periods of 20 to 60 days, termed the Madden–Julian oscillation. Another potential source of prediction lies with the influence of land surface conditions, most notably soil moisture but also including snow cover. Volcanic eruptions may represent an episodic opportunity for prediction of global temperatures over several seasons, and "global warming" may provide, in ways not yet understood, some additional basis for improving seasonal predictions.
The skill of probabilistic forecasts of seasonal climate one to two seasons in advance has improved dramatically during the 1980s and 1990s. Advances in understanding the wide array of processes that contribute to seasonal to interannual climate variability offer the hope of continuing improvements in the decades to come. Owing to the chaotic nature of day-to-day weather fluctuations, such forecasts will always remain probabilistic and be subject to considerable uncertainty, but they can nevertheless be of substantial value in mitigating the adverse impacts of climate fluctuations upon human activities such as agriculture and fisheries, and the allocation of fuel, energy, and water. To realize these potential gains it will be necessary to maintain the integrity of the global weather observing system and to provide enhancements, as needed, for monitoring processes in the oceans and on the land surface that contribute to variability on these timescales.
References and suggested further reading
Barnston, A.G., A. Leetmaa, V.E. Kousky, R. E. Livezey, E. A. O'Lenic, H. van den Dool, A. J. Wagner and D. A. Unger, 1999: NCEP forecasts for the El Niño of 1997–98 and its U.S. impacts. Bull. Amer. Meteor. Soc., 80, 1829–1852.
Mason, S. J., L. Goddard, N. E. Graham, E. Yulaeva, L. Sun, and P. A. Arkin, 1999: The IRI seasonal climate prediction system and the 1997/98 El Niño event. Bull. Amer. Meteor. Soc., 80, 1853–1873.