Online Catalog

GCRIO Home ->arrow Library ->arrow Consequences -> arrow Vol. 5, No. 2, 1999 -> arrow El Niño and the Science of Climate Prediction Search
U.S. Global Change Research Information Office logo and link to home
Updated 17 November 2004
Consequences (title)
Consequences Vol. 5, No. 2, 1999



See also: article overview





El Niño and the Science of
Climate Prediction


A review assessment published in

CONSEQUENCES vol 5 no 2, 1999, pp. 3-15

Not long ago, the term El Niño was seldom seen or heard outside the ivied walls of academia and research laboratories. In the last ten years it has been brought to the attention of nearly everyone, in every country, and is probably here to stay. During the winter of 1997-98, particularly, El Niño was on the news and in the headlines almost every day — and correctly or not, invoked to explain nearly every weather-related event, from mudslides, to ice storms, to common head colds. There were El Niño t-shirts and El Niño car sales, and the parochial term found its way into such bastions of popular culture as late-night TV monologues and comic strips.

This widespread notoriety of the El Niño may be new, but the concept is not, and efforts to understand it follow a winding path that can be traced back nearly a century. The name itself, as applied to natural events, dates back still further: to sea captains in the 1700s who sailed in wooden ships along the western coast of South America, and experienced wild swings, from time to time, in the temperature of the sea and the abundance of fish and other marine life.

We now have evidence that El Niño, and its counterpart La Niña, have been a fixture of the earth’s climate for at least the last 100,000 years. Fossilized corals that leave a readable record of year-to-year temperature variations during the last interglacial period (before the most recent Ice Age) show clear evidence of fluctuations in ocean temperature that are very similar to those observed today. What is behind this ubiquitous phenomenon, and why should it be of such interest and concern to all of us?

What It Is

El Niño is fundamentally a warming of the surface waters of the tropical eastern Pacific Ocean — from the South American coast to the International Date Line — that persists for three or more seasons. But it is far more than an oceanographic peculiarity. Tightly tied to these systematic ocean warmings are major swings in atmospheric conditions on a nearly global scale, which in turn invoke seasonal and longer-term changes in the Earth’s climate.

What is "normal" for ocean temperatures in the tropical Pacific, and how big a change does El Niño make? At the western (Asian) end of the tropical Pacific, the surface of the sea is consistently warm (about 29°C, or 84°F), the atmospheric pressure is low, and precipitation is frequent and intense (Figure 1). Conditions at the opposite end of the wide Pacific, some 15,000 kilometers to the east, are very different. There, nearer South America, the water is relatively cool (21 to 26°C, 70° to 79°F), the atmospheric pressure is high, and there is little rainfall. Colder surface waters persist in the equatorial eastern Pacific throughout the year, and are most pronounced in what is summer and autumn in the Northern Hemisphere, from July through November. The marked east-west difference in Pacific Ocean temperature is associated with what is known as a "thermally direct" circulation loop. Three of the four "sides" of the loop are rising air over the warmer waters of the tropical western (Asian) Pacific; descending air over the cooler waters at the eastern (American) end; and the horizontal, return flow of air along the ocean surface, from the cooler to the warmer ends of the Pacific, that is provided by the always-westward-blowing trade winds. A similar pattern of air circulation is set up when a fireplace is lighted at one end of a cool room.

Occasionally, however, the warm waters at the Asian end of the tropical Pacific begin to spread eastward. Accompanying these temperature changes, the regions of low pressure and heavy rainfall also migrate eastward, the central and eastern Pacific become warm and rainy, and the weather in the far western Pacific becomes somewhat drier and cooler. This is El Niño.

A weaker and far more benign warming of the ocean surface is an every year occurrence off the coasts of South America, in late December, and hence the origin of the name "El Niño", or "The Child", in reference to Christmas. Today, however, the term is reserved for the grander and more occasional events that we described above. Recently, the term "La Niña" has been coined to designate swings in tropical ocean temperature in the opposite direction, which also occur episodically. The characteristics of La Niña are cooler than normal conditions in the eastern Pacific, and wetter and warmer than normal conditions in the Far Western Pacific.

Closely associated with El Niño and La Niña is another meteorological phenomenon called the Southern Oscillation. It refers to a very large scale, seesaw exchange in sea-level air pressure between areas of the western and the southeastern Pacific. The general character of the Southern Oscillation has been known for more than seventy years. But not until much more recently was it recognized that the contraction or expansion of the warm waters in the west Pacific and the strong swings of the Southern Oscillation were part of a common phenomenon. The combination is now called El Niño/Southern Oscillation, or ENSO. El Niño defines the warm phase of ENSO, and La Niña the cold.

Increasingly, research has shown the importance of ENSO to seasonal climate on a worldwide scale, and to the many variables that are impacted by climate. ENSO has been linked to precipitation and temperature extremes in Africa, the Americas, Australia, and portions of Asia. It has been associated with epidemics, variations in crop yields, ecosystem disruptions, forest fires, famines, and even market fluctuations, to name but a few. The science of ENSO is fascinating, but these practical impacts are why we should care about it. The global impacts of ENSO are the topic of the next article in this issue of CONSEQUENCES. Here the focus is on what is known about ENSO, the questions that remain, the current status of our ability to simulate and predict its ups and downs, and the prospects for future predictions of these and other seasonal-to-interannual variations in climate.

Characteristics of ENSO in the 20th Century

Most of what we know of ENSO comes not surprisingly from data taken in the 20th Century. The instrumental record of atmospheric and ocean conditions in the century just ended is far from uniform in quality or coverage, but it is good enough to secure rough measures or indices of ENSO behavior. Perhaps the best known measure is the Southern Oscillation Index (SOI), defined as the difference in normalized sea level pressure between Tahiti and Darwin, Australia. Another that has become popular recently is a measure of sea surface temperature changes averaged over a fixed area of the equatorial eastern Pacific (90W to 150W, 5S to 5N), known as NIÑO3. A plot of these two indices for the period from 1882 to 1997 is shown in Figure 2. The close correspondence between the two indices — one meteorological, the other more oceanographic — demonstrates the strong connection between the atmosphere and ocean that lies at the heart of ENSO.

The strongest period (or cycle) that emerges when these records are analyzed statistically is about four years, but one can easily see that ENSO is anything but regular. It looks more like a jumble of many different periods, and is further characterized by epochs of relative quiescence (such as the mid-1930s and 1940s) and others of strong activity (the early 1900s and late 1900s). Even the fundamental "life cycle" of warm and cold events is seen to be quite variable, with some persisting for less than a year and others for several years. The causes of this rich variability are still debated, but an important consequence is that prediction is a challenging problem.

One feature that stands out is the rather protracted warm period of the early to mid 1990s. Indeed, the impact of this event and the very strong warm events in 1982-83 and 1997-98 were what made the past two decades the warmest of the century. Was this a random coincidence — the chance occurrence of unusually warm El Niños at the time of a pronounced rise in global surface temperature — or was one fueled by the other? We return to issues of ENSO and global greenhouse warming later in this review.

El Niños in Different Flavors

Some truly historic work was done in the early 1980s to organize existing data archives, and provide for the first time a consistent picture of the evolution in space and time of El Niño events. It was an important step, scientifically, for it allowed a generic, or "canonical" description of the life cycle of El Niño that was detailed enough to serve as a target for the theories and models that were subsequently developed. The method used to define the typical El Niño was to take the average of all El Niños that had occurred in the thirty-year period between the early 1950s and the late 1970s. The notion of averaging seemed entirely reasonable, for inspection of the limited information from each of the individual events suggested they all followed a similar pattern.

The resulting canonical description quickly became engrained into the thinking of the scientists who worked on ENSO problems, for they badly needed a clear and consistent example of what, precisely, they needed to watch for and explain. The significant characteristics — not unlike clues in a set of serial crimes — were:

• A relaxation of westward-blowing trade winds, and a weak warming of the sea surface in the central Pacific, at about the end of the year preceding the El Niño warming;

• A marked warming of the surface ocean off the coasts of Peru and Ecuador during the months of March to June of the El Niño year;

• Subsequent, westward expansion of the warming, to cover the entire eastern equatorial Pacific, and a concurrent, major collapse of the trades across most of the Pacific, culminating in a full scale event by the end (December) of the El Niño year;

• The return of stronger trades in the far Western Pacific starting in December, and continuing in ensuing months; and

• Rapid cooling of the waters off Peru and Ecuador, leading to colder than normal temperatures in the eastern Pacific by June of the year following El Niño.

Ironically, an unusually strong El Niño was taking form in 1982-83, at the very time that the canonical description was being published. And it followed a quite different pattern. At the time, there was still very little observational information on sea conditions that could be accessed by scientists quickly, and even a rough description of the event was not in hand until many months after its demise.

A Surprise

The 1982-83 event was the largest of the century at that time, dwarfing the composite picture by at least a factor of two. But more interestingly, the "life cycle" was different; it seemed to march to a different drummer — or no drummer at all. A major warming of the entire central and eastern Pacific, and collapse of the trade winds, developed rapidly late in 1982. These changes were not preceded by the expected warming event in the coastal area off Peru and Ecuador. Instead, in the subsequent year, 1983, an enormous coastal warming was noted there, even as the waters in the central Pacific were cooling.

The total duration of warm conditions was at least half a year longer than in the canonical description, and the amplitude of peak warming three times larger.

What we learned from the 1982-83 event was that El Niños come in more than one flavor. While tremendous progress was made from studies focused on the canonical event, the need to move beyond it was already undeniable.

Further Evidence of Change

What happened in ensuing years could only confirm this realization. The next two warm events began as did the 1982-83 El Niño, but each differed in how it subsequently played out. The first, beginning in 1986, persisted through all of 1987, to be followed by a strong but brief La Niña in mid 1988. The El Niño that began in late 1991 was not strong, but it persisted in the central Pacific for more than three years, while the eastern Pacific jumped back and forth three separate times between warmer and near-normal states, each lasting but a few months. This erratic behavior defied our simpler, text-book concept of El Niño to the point that there was (and still is) no consensus on whether the period following late 1992 should go down in the records as an extended El Niño, or repeated El Niños, or something entirely different.

The most recent El Niño, in 1997-98, was again unique. First, it set new records in terms of magnitude, eclipsing even the 1982-83 event. Its onset followed more the pattern of the canonical El Niño, beginning early in the year, with a very strong warming off the coasts of Peru and Ecuador. Its decline began early in 1998, and accelerated strongly during May, leading to colder ocean temperatures throughout much of the eastern Pacific in June. But in contrast to the canonical pattern, the extreme eastern Pacific did not follow suit until many months later.

Those who try to understand ENSO have gradually moved away from a focus on generic events, toward a picture of a highly variable cycle that is punctuated by particular extreme phases in both positive and negative sense. The terms "El Niño" and "La Niña" are surely here to stay, like the simple "electrons" and "protons" of early particle physics, but there is now a need for us, too, to expand our vocabulary.

Developing an Understanding of Enso

The pioneering work that led to our present understanding of ENSO traces back to Sir Gilbert Walker, who in the early 1900s discovered (and named) the Southern Oscillation.

Walker held the position of Director General of the Observatories in India, in the days of Rudyard Kipling. He assumed this post in 1904, following the devastating monsoon failure, and associated famine, in 1899 (an El Niño year). Walker set out to predict the years when the annual monsoon rains would fail to materialize, and in the process searched for correlates in sea level pressure and other variables on a global scale. He found that when the Southern Oscillation was notably weak there was heavy rainfall in the central Pacific, drought in India, warm winters in southwestern Canada, and cold winters in the southeastern U.S. No conceptual framework supported the patterns he found, however, and his work was largely dismissed for several decades.

Eventually, Walker’s correlations were reexamined in the light of decades of new and independent data, and found to hold. But it was the work of Jacob Bjerknes in the late 1960s that tied the Southern Oscillation clearly to the oceanographic phenomenon known as El Niño, which by this time was understood as a basin-scale warming of the Pacific Ocean, and not simply a phenomenon of the coastal region near South America. Bjerknes went on to propose a physical explanation for the linkage, which set the stage for the modern understanding of ENSO.

The Ocean and the Air

To understand Bjerknes’ hypothesis, one needs to appreciate some basic principles of ocean dynamics. The first has to do with the action of the westward-blowing trade winds on the near-surface ocean. If the Earth were not rotating, then the force of the westward winds would simply tend to generate a westward current in the near-surface waters. But the action of the so-called Coriolis force, due to the earth’s rotation, serves to deflect a westward current in the poleward direction in either hemisphere, producing a divergence of waters near the equator. The departing surface waters must be replaced by waters from below. Thus, westward winds result in a systematic upward flow, or upwelling, of water along the Pacific equator; the stronger the winds, the stronger the upwelling.

Temperature in the ocean decreases with depth. So whenever upwelling occurs, deeper and hence cooler waters are presented to the surface. Upwelling is therefore tied to cooling. This is why under normal conditions the equatorial oceans are cooler than neighboring areas to the north and south in the central and eastern Pacific (where the trades are strong).

The trade winds, in turn, are also linked to surface temperature patterns. In the tropics, the regions of heavy precipitation and large-scale ascending motion in the atmosphere tend to form over the warmest waters. Similarly, large scale sinking motion tends to occur over cooler regions. As the western (Asian) Pacific is very warm, and the eastern Pacific is normally cool, a circulation pattern with rising motion in the west, sinking in the east, and surface winds from cold toward warm (i.e., the westward trades), prevails. Bjerknes named this circulation pattern the Walker circulation.

Bjerknes’ hypothesis is essentially this: the normal patterns of trade winds, sea level pressure, and strong thermal contrast from west to east are mutually reinforcing. The strong temperature difference induces strong trades; the strong trades induce strong upwelling and cooling in the east Pacific, enhancing the thermal contrast. A La Niña state is simply an extreme version of the normal state, with especially strong trades and thermal contrast from west to east. El Niño, Bjerknes hypothesized, is also the result of the same positive feedback cycle, working in reverse. At these times, the eastern Pacific warms, reducing the thermal contrast with the west. The reduced thermal contrast weakens the force of the trade winds, which acts to further warm the east Pacific, and so on. This chain reaction leads inexorably to the extreme state that we call El Niño: a collapse of the trades, and elimination of the east-west thermal contrast.

There was one crucial element that Bjerknes could not explain. Why were there transitions? What drives and sustains the endless flip-flops that characterize the ENSO cycle? Bjerknes never resolved this puzzle. It awaited advances in ocean dynamics that were still years away.

Insights in the Modern Era (1970-1985)

Important to the progress in understanding the dynamics of the tropical oceans were a series of ocean observation programs conducted in the 1970s. In addition, the first ocean observing program to monitor the temperature of the upper ocean was established during this period, with the volunteer help of merchant ships that collected these readings as they crossed the ocean on regular shipping lanes. Data from these observing efforts confirmed earlier, theoretical predictions of disturbances in temperature and currents, known as Kelvin waves and Rossby waves, that were peculiar to the equatorial ocean. This was a critical advance in oceanography, for there was now a solid basis for "predicting" the way that upper ocean currents and thermal structure would adjust to changing wind patterns.

The interplay of winds and the surface layers of the ocean were the essence of ENSO. Any hope for predicting ENSO events demanded a solid understanding of how winds that blow across the surface of the ocean affect the deeper ocean, including the varying depth of the sharp transition zone, called the thermocline, that separates the Sun-warmed waters near the top from the abruptly colder waters of the depths below. Understanding Kelvin and Rossby waves and the response of the thermocline to these motions in the oceans were the key to the solution.

The existence of Kelvin and Rossby waves, and their reflections at the ocean boundaries, explained the changes in thermocline depth that had been found to characterize El Niño and La Niña states. Strong westward trade winds, that characterize normal and especially La Niña states, raise the depth of the thermocline in the east while deepening it in the west, through the actions of Kelvin and Rossby waves (Figure 1a). During El Niño, the trades slacken, and in due course the thermocline deepens in the east, and shallows in the west, compared to the normal state (Figure 1b).

As these important ideas were being worked out, they were quickly put to the test with rather simple numerical models of the ocean. Several studies in the late 1970s and early 1980s were able to demonstrate that the ocean changes that were being observed at that time were consistent with equations and models based on the action of Kelvin and Rossby waves. At the same time, a clearer notion of thermally-forced tropical wind circulation was being developed, offering similar capabilities to describe and simulate equatorial wind patterns with dynamical models. The tools for understanding the full ENSO cycle were now at hand.

Emergence of a Theory

The first attempts at dynamical explanations of ENSO involved ocean models and very simple prescriptions of wind patterns based on the ocean state. They showed the possibility of interactions between ocean and atmosphere that might lead to instabilities which would in turn generate El Niño.

A significant step forward came with the first coupled dynamical models of the atmosphere and ocean. These models were simplified in several respects, but were nonetheless complete enough to allow comparison with the real world, in terms of winds, currents, sea surface temperature, and variations in thermocline depth.

The first successful simulation of an ENSO cycle, with successive but irregular warm and cold events identifiably like nature, was achieved in the mid-1980s. This opened the door to a flurry of activity that led to the first theory for the ENSO cycle. The essential ideas invoke the peculiarities of oceanic Kelvin and Rossby waves and the different speeds at which they move back and forth through the waters of the equatorial Pacific.

The irregular nature of ENSO is a topic of continuing research. Why is the behavior so erratic? Why should there be so wide a wide range of event amplitudes and duration? Although these questions are yet to be resolved, a number of possibilities have been proposed. The first is that the complex, nonlinear nature of the ocean-atmosphere coupling can by itself lead to variations that are not periodic. In mathematics, this property is called deterministic chaos. Some rather simple physical systems behave in this way, as do the original ENSO coupled models.

A second possibility is that the irregularity arises from the action of the random variations that produce a kind of static, or "noise" in the climate system. There are numerous candidates. The atmosphere exhibits a lot of variability that is due to essentially random internal processes. Ordinary weather events — such as day to day fluctuations in winds and temperature and precipitation — amount to noise, when viewed in the context of longer-term processes such as ENSO. Very rapid changes of precipitation and winds, associated with thunderstorms, are ubiquitous in the tropics. Still other sources could perturb the ENSO process, including oceanic disturbances that enter the equatorial ocean from distant regions, or even external events such as volcanic eruptions. Any or all of these could play a role.

Progress in Observing ENSO

Perhaps the greatest accomplishment in climatology in the past decade and a half — and a primary legacy of the international Tropical Oceans/Global Atmosphere (TOGA) program — was the development and initiation of an ENSO observing system.

As noted earlier, the huge 1982-83 El Niño was well underway before even the scientific community was even aware of its existence, due to a total lack of rapid access information. As TOGA aimed its sights on the possibilities for ENSO prediction, what was once desirable became imperative: to monitor the tropical Pacific Ocean and atmosphere continuously and continually, from day to day and month to month and year to year, with universal, real time access to the data obtained.

A continuing set of measurements of upper ocean thermal structure already existed, dating from the 1970s, but these were transmitted to the scientific community only in delayed mode. Beginning in the early 1980s, greatly improved estimates of sea surface temperature became available, thanks to innovations that combined satellite data with that obtained from ships and floating instruments. The in situ information was essential in eliminating large-scale satellite biases; the satellite information was essential in providing the spatial detail missing from the sparse in situ network.

During the ten-year life of the TOGA program (1985-1995), tremendous progress was made. First, due to advances in theoretical understanding, the scientists involved had a clear picture of the importance and hence priority of subsurface observations of the ocean, particularly near the equator where the action of Kelvin and Rossby waves is vital. Second, the forward-looking focus on prediction made rapid delivery of data a high priority. As a result, the many technical problems of real-time monitoring were tackled and solved.

A New Observing System

Figure 3 shows the dramatic growth of ENSO observations during the TOGA years. By the end, an extensive network of moored buoys, drifting buoys, tide gauge stations, and volunteer observing ships was reporting information in real time via satellite relays.

The most recent El Niño was observed at an unprecedented level of detail. Week-by-week, even day-to-day changes in the ocean and atmosphere were monitored, and made available to anyone with access to the Internet. The steady increase in the quality and quantity of observations has played a vital role in the refinements in our understanding, in the discovery of new phenomena, and in the testing and validation of numerical models.

Even further progress is possible in the near future. The planned availability of satellite estimates of surface winds, using a spaceborne instrument called scatterometers, will soon do for winds what the combined satellite and in situ measurements did some years ago for sea surface temperature. The possibility of "smart" drifters, that can reside at programmable depths in the ocean, occasionally surfacing — like a submarine on patrol — to report the temperature at various depths via satellite, may soon allow an affordable means for measuring subsurface temperature (and other relevant quantities such as salinity) on a basin or even global scale. These and other advancements will undoubtedly add to what we know of ENSO, and aid the prediction of seasonal-to-interannual climate.

Predicting ENSO

It is common experience that weather forecasts become rather rapidly less reliable when projected more than a few days in advance. This loss of "skill" is the result of the atmosphere’s chaotic nature.

Forecasts of the weather rely heavily on what are called initial conditions — the array of temperatures and pressures which the forecaster’s model uses as a starting point for his or her projection. The more closely these describe what really pertains at that moment, the more accurate the forecast. Minor differences in initial conditions can significantly alter what is forecast, and the more distant the projection, the greater the disparity. Yet we can never know the actual initial state of the atmosphere precisely, because of incomplete and imperfect measurements. Thus, there are limits to how far ahead we can predict with useful skill, even if our predictive models were absolutely perfect. In the face of this reality, can we ever hope to predict climate, or future ENSO events?

Why Climate Can Be Predicted

First, climate prediction does not aim to foretell the details of day-to-day weather, one or more seasons ahead. Rather, it aims to predict some aspect of the statistics of weather, the simplest and most common being a seasonal mean: such as, precipitation in the western states will be lower than normal this winter. But even these more generalized projections would be hopeless were it not for the fact that additional (and potentially simpler) factors become more dominant in the climate system at longer time scales.

Most important among these, for ENSO, is the ocean. Theory dictates that ENSO is fundamentally a coupled process, arising from a tight linkage between atmosphere and ocean. Due to its great mass and thermal inertia, the ocean changes rather slowly, and by virtue of the strong coupling, it imparts to the atmosphere a degree of order, or determinism, that it might not otherwise possess. Interactions between the atmosphere and the land surface could potentially do the same, since the land is also slow to change.

As we noted earlier, predicting the behavior of ENSO is muddled somewhat by the presence of a seemingly chaotic element, and by the influence of shorter-term weather variations that we have called noise. Thus, we know ENSO prediction skill must always be limited. But the limits will be determined by the properties of the ocean-atmosphere coupling, and not the atmosphere alone. The distinction can change the time scales of useful predictions from days to seasons.

Initial ENSO Prediction Efforts

The first El Niño predictions were made in the 1970s, following the advancements of ocean dynamics that identified the importance of oceanic Kelvin waves in initiating these events.

The first and simplest scheme was diagnostic: observing the changes in trade winds, and inferring the ocean’s response. Later, a more quantitative approach was taken by employing a dynamic ocean model, forced by observed winds that projected the state of the ocean forward in time based on assumptions about future wind patterns. This showed a degree of skill at lead times of a few months, based on a retrospective analysis of how well the model could have forecast several past El Niño events based on trade-wind observations that existed at the time. But the need to assume future wind patterns limited the forecasts, particularly at lead times beyond three months.

In the mid-1980s other significant advancements were made, in the development of statistical methods of prediction and in forecast methodologies employing the first dynamical model that linked the oceans and the atmosphere. For validation, the forecasts were tested, in retrospect, on the basis of what was known of ocean conditions during the period from 1970 to 1985. Very little was expected from such experiments, given the many simplifications that were involved. Yet, surprisingly, when the observed and forecast values of NIÑO3 index were compared, the models demonstrated appreciable skill out to more than one-year lead-time.

The first actual forecast of equatorial Pacific Ocean temperatures was generated with the same model in early 1986. It called unambiguously for a warm event to develop later that year. The forecast was made public, and it generated a good deal of controversy. Many scientists were still highly skeptical of the notion that El Niño was predictable in theory, let alone in practice.

Later that year an El Niño event did indeed develop, albeit several months later than forecasted. This success set the stage for a highly intensified research effort in ENSO prediction under the international TOGA program, which indeed marked the beginning of the era of climate prediction.

An Intensified Prediction Effort

In the decade following the first successful forecasts, ENSO prediction has been transformed from a curiosity to an enterprise of global scale. As new and independent prediction efforts, both statistical and dynamical, have confirmed the earlier results, a growing consensus concerning the reality of forecast skill has emerged. Increasingly, research has attempted to clarify the actual limits of predictability, and to advance our capabilities toward that limit.

During the decade of the 1990s, the idea of an international scale organization dedicated to both the continual advancement of prediction, as well as the application of this new information to practical problems of direct benefit of societies, was born and developed. Today, the fledgling International Research Institute for Climate Prediction exists, and is working toward these goals.

Better Models

A significant effort has been directed, in recent years, toward more sophisticated models. While simplified models were effective in demonstrating predictive potential, they needed to be weighed and validated in the light of more complete and more realistic models. Most atmospheric scientists believe that, ultimately, the most accurate predictions will come from more complex models, and specifically the general circulation models, or GCM’s, that endeavor to include, as far as possible, all the processes that are known to affect the real climate system. But pressing GCM’s into the service of ENSO prediction is neither a short nor an easy process.

Because of their complexity, GCM's are substantially harder to optimize for specific problems such as ENSO. The first forecasts from these sophisticated models began in the early 1990s. Initially, they did not match the results of the simpler models in terms of predictive potential. Slowly but steadily, progress has been made. Today, the more elaborate forecast systems are competitive, and arguably the superior tools for ENSO prediction.

At least as important as improving models is improving the methods of initializing forecast models with observed data, all of which is subject to errors and uncertainty. The procedures currently being used are still relatively simple. More elaborate (and presumably more effective) machinery is available, but has yet to be tested.

The Track Record: Where do we Stand?

Today, more than twenty separate forecast systems are being run routinely to predict the evolution of ENSO. How good are they? Are the forecasts more reliable at some times than others?

In order to provide reliable answers to such questions, one needs a large number of forecasts, spanning varied states of the real climate system. A continual stumbling block is the lack of extensive data to make such evaluations. Prior to the 1980s, the quantity and quality of observational data decreases substantially, seriously hindering the ability to adequately initialize forecast models. We do not know enough about the state of the ocean at the time of onset of these earlier events. In addition, many forecast models have not been run extensively for past years because of resource limitations, and especially computer resources.

What we can do is evaluate the performance of available forecasts over the limited period of the past ten to fifteen years. This gives a sense, if not a robust measure, of current capabilities. As of 1994, one such study was done. It compared two dynamical coupled models, two purely statistical models, and one hybrid model. The finding was that each type of forecast model was, at that time, giving a comparable level of skill; namely, at the level of 65 to 70 percent accuracy for 6-8 month lead times. Though a respectable score, it left room for improvement. An interesting finding was that for all forecast systems, both seasonal and year-to-year variations in skill were apparent.

The explanation for these variations in skill is still unresolved, but the prevailing view is that the actual climate system undergoes fluctuations in predictability, depending on the season and the state of ENSO. Variable predictability is not uncommon in chaotic systems of many types. In the case of ENSO prediction, it carries a warning that the reliability of forecasts is not always the same. Future work will address ways to incorporate such information into the forecast products.

Improving the Models

The period from the mid-1990s to present has posed particular challenges to forecasters. No forecast system proved effective at predicting the protracted nature of the 1991-94 warm event. As mentioned earlier, the extended El Niño of those years, with quite rapid flip-flops from warm to normal in the east Pacific, and continuous warm temperatures in the central Pacific, did not fit the canonical El Niño pattern. It remains a target for improved models and forecast systems.

The most recent, powerful warm event of 1997-98 was a source of further discovery. For one, it gave us our first outright failure to predict the onset of El Niño, using the original coupled forecast system developed in 1985. Diagnosing this problem has been enlightening. The so-called Lamont forecast system, until this time, relied on available observations of ocean winds to establish initial ocean conditions: no direct ocean data were involved. It is obvious that in terms of predictive skill, so simple a system cannot be optimal, since it ignores so much that we now know of the climate system. Nonetheless, it had worked rather well for many years.

We found, after the fact, that the 1997 forecast could be salvaged in one of two ways. The first was to replace wind observations from ships with satellite-derived winds, which had become available in 1996. The second involved the assimilation of sea level information into the ocean model. The result was the most convincing demonstration to date of the impact that ocean data assimilation can have on ENSO forecasts.

A sobering aspect of the 1997 forecasts was the inability to forecast the onset with significant lead times. The improved forecasts could foretell the magnitude of the warming beginning only in about March, even though direct observations, taken before that time, left no doubt about what was developing in the tropical Pacific. This, too, has yet to be explained.

A possible explanation is the influence of the so-called Madden-Julian Oscillation, which was especially active during this period, and not considered in the forecast model.

The phenomenon involves convective disturbances that form in the eastern Indian Ocean and western Pacific Ocean, and tend to migrate eastward into the central Pacific. The disturbances have a preferred life cycle of one to two months, and often feature strong eastward winds and rain events. It is a matter of debate whether and how these events can interact with, or even trigger El Niño onsets. It is quite clear that they are not the fundamental cause of ENSO, since they are present to varying degrees every year, regardless of the phase of ENSO: El Niño, La Niña, or neither. Yet the possibility exists that the chance occurrence of a strong Madden-Julian event, at a time when ENSO predictability is otherwise low, could alter the subsequent evolution. If this is the case, then only with the ability to forecast the Madden-Julian phase could the ENSO evolution be predicted accurately at all times, several seasons in advance. At present, no forecast system captures intraseasonal signals such as the Madden-Julian oscillation with any skill. Once again, a target exists for improved models.

ENSO and Global Greenhouse Warming

For the past two or three decades, ENSO and global warming have been the two most studied climate problems. Yet, until very recently, they have been pursued independently, by nearly independent groups of researchers. The early global warming (or "global change") studies focused primarily on the role of atmospheric processes, with little if any emphasis on ocean dynamics, while ENSO research was focused first and foremost on this very topic. But as global change research has progressed, the importance of active coupling among atmosphere, ocean, land surface, and ice have been amply demonstrated, emphasizing, among other things, the probable connections between global change and ENSO.

How might ENSO and global change be related? It has been known for some time that ENSO has a significant effect on both local and globally averaged surface temperature. El Niño years stand out in any casual inspection of the record of global mean surface temperature. This is hardly a surprise, since ENSO alters the temperature of the surface ocean over an area that covers nearly one quarter of the Earth’s surface. Indeed, the prevalence of warm ENSO events in recent years has contributed significantly to the recorded increase in the average surface temperature of the Earth. But is there a causal connection between the two?

Some studies have suggested that the characteristics of ENSO might be very much modified by other aspects of the global environment. For example, some ENSO models have indicated significant increases in ENSO variability, and in the magnitude of ENSO extremes, as the Earth warms, overall. Unfortunately, the results depend on the details of temperature changes beneath the ocean surface, which is an issue of considerable uncertainty.

Reasons to Work Together

We now know that the tools that are being developed to assess the impacts of global warming must incorporate the effects of ENSO and other climatic phenomena as well. If not, what these prescribe may omit an important aspect of future climate change. For the many ENSO-sensitive regions of the world, significant changes in ENSO behavior could well be the most important expression of global greenhouse warming. On the research side, global change models, for example, will need to employ finer spatial and temporal resolution to capture the ocean processes that are critical to ENSO and to tropical air-sea interactions in general. For the most part, this has been lacking until now. The models developed for ENSO and climate variability research on the one hand, and global change studies on the other, need somehow to be joined.

It would prove fruitful to merge not only the models, but also more of the research activities of global change, and of climate variability and prediction in general. The prediction of ENSO and other seasonal-to-interannual variations provide an opportunity that is vitally important to global change research: namely, the validation of prediction and assessment tools. We know that the same climate processes that bring on year-to-year climate variability can also apply to changes of longer, decadal scale. What better way to demonstrate the adequacy or deficiencies of decadal and longer-term prediction tools than to test and hone them on shorter-term predictions?

What today’s global warming models need for validation are several extensively recorded cases of global climate changes, and these are in short supply. Climate has changed significantly in the distant past, but available paleo data are limited in what they can tell us, in more than general terms, about these prior global changes. A legacy of the last twenty years of ENSO research are data sets of individual, well-documented events, each lasting a year or two or more, that might be applied to this purpose.

The Challenges of Interannual Climate Prediction

The progress in ENSO research, and early successes in ENSO prediction, are truly exciting and significant. In a decade and a half we have not only identified a potential for limited seasonal-to-interannual climate prediction, but have also begun to exploit that potential for very practical, down-to-earth purposes. Institutions such as the IRI now exist that are dedicated to the application of improved climate forecasts to matters of public health, food security, natural resources, and human productivity. Indeed, a new age is upon us. But as our discoveries have opened new doors, they have also presented new challenges, and made us more aware of the complexity of the processes that we seek to predict.

Beyond ENSO

Although virtually all of the present skill in climate forecasting comes from ENSO and its direct consequences, there is reason for optimism that other elements of climate variability may also yield to practical prediction. An example is the known patterns of variability found outside the tropical Pacific, involving ocean temperature changes in the other tropical oceans. Our knowledge of tropical climate dynamics suggests that these also should have some determinism, and therefore some predictability. But they are not replicas of ENSO, and understanding is still lacking.

Considerably more effort is needed to understand better the mechanisms behind year-to-year monsoon variability, and year-to-year climate variations in the Atlantic basin, to name but two. Research into these questions offers the hope of predicting the future behavior of these phenomena, for practical purposes. But it could also improve predictions of ENSO as well, since evidence suggests that significant interactions may well exist between different sectors of the ocean.

A better understanding of these other phenomena will require an investment in more and better observations. But the payoff that is already apparent from the international investments in the ENSO observing system should make this an easy choice.

What we know today of ENSO itself is far from complete. The uncommon El Niños of the 1990’s have provided a rather sobering reminder of the limits of our understanding and prediction capability. Several unsolved issues have come to light. One is the role of intraseasonal variability, such as the Madden-Julian oscillation. At present, coupled models do not capture this variability very realistically, and we need to do better. Yet other signals exist in the observational record, including fluctuations at biennial (two year) time scales, and fluctuations at decadal or even multidecadal time scales. The latter two almost certainly involve higher latitude regions of the Pacific Ocean, and therefore somewhat different processes. It has been speculated that longer time scale processes might act as a modulator of ENSO, effectively altering its mechanics, and perhaps even its predictability, from decade to decade. Whether this is the case or not, improved understanding of these other sources of climate variability that coexist with ENSO can only serve to improve predictions, and to sharpen our estimates of the limits of forecast certainty.

Improved understanding demands more comprehensive models and improved methods of assimilating data. State-of-the-art climate models still suffer from systematic errors that surely limit their forecast performance. Among the most problematic are representations of the atmospheric boundary layer, atmospheric convection, clouds of all types, and ocean turbulent mixing. Some of these problems will demand focused process studies, including field observations.

Getting More From Models

As important, perhaps, is a more organized scheme for developing and evaluating models. Even at present, the more sophisticated climate system models challenge the intellectual and computing resources of even the largest climate research facilities in the US As knowledge increases and models improve, the demands on computing capability scale ever upward. How can we keep step?

At least part of the answer is to develop means for more effective collaboration among research groups and centers around the world. This means developing rigorous standards of evaluations, common benchmarks, and common programming and data format standards that will allow an easier transfer of technology from site to site. By exercising the discipline to develop such standards, the research community will greatly elevate the efficiency with which new models and forecast systems can be developed, tested, and improved. An effort aimed at precisely this end has recently developed within the US, spanning all major climate research groups and agencies supporting climate research and prediction. If it proves successful, the research community and the research agenda will benefit greatly.

In addition to better forecast models, alternative forecast methods should and will be explored in future research. One promising approach is that of ensemble systems. The basic premise is that, due to chaos and sources of unpredictable noise in the climate system, there is inherent uncertainty in any forecast. Our goal, then, should be to estimate not the exact state of the climate, but the probability of a range of possible states. This can be done by conducting many forecasts, from initial conditions varying very slightly, within the range of observational uncertainty. Then, information from the entire ensemble of forecasts allows for an estimate of uncertainty, and for probabilistic assessments. The ensemble need not be restricted to just one model; indeed, including several models of similar individual skill levels may produce the best result. Future climate forecasts and assessments will continually evolve in this direction, and indeed they must.

Another hurdle is computing. Even as today’s high-end climate models overburden the existing resources, the need to provide still more expensive, higher-resolution forecasts is being demonstrated. Under present circumstances, all too often forecasts are being evaluated on the basis of small, possibly unrepresentative samples of actual events. This leads to poor estimates of uncertainty, and unnecessarily slow or even erratic progress. But the problem is remediable. Increased investments in computing could be put to good use immediately; no new knowledge or technology is needed. The guaranteed return on these investments would be far more reliable assessments of forecast skill, and the means for researchers to innovate more efficiently and more effectively.

Climate prediction has come a long way in a short time. We are still faced with daunting challenges. But we also know that any advance in our understanding has the potential of immediate application in the everyday lives and needs of people, and society. Students today seem thoroughly energized by this problem, despite its complexities and challenges. If this is an indicator — as I indeed believe it to be — we can expect the new science of interannual climate prediction to continue to mature.

For Further Reading

"A current catastrophe: El Niño," by Pamela N. Knox. Earth, September, 1992.

"El Niño dynamics," by J. D. Neelin and M. Latif. Physics Today, December, 1998.

El Niño: Global Weather Disaster, by Thomas Y. Canby. National Geographic Magazine, vol 165, no 2, February 1984.

El Niño, La Niña, and the Southern Oscillation, by S. G. H. Philander. Academic Press, San Diego, Calif., 293 pp, 1990.

Learning to Predict Climate Variations Associated with El Niño and the Southern Oscillation. National Research Council, Washington, D.C., 1996.


Figure Captions

Figure 1. A schematic representation of conditions at and beneath the surface of the tropical Pacific Ocean, extending from the longitude of Australia on the left to the Americas on the right. Contours are isotherms of sea-surface temperature, with warmer surface temperature in darkest blue. The layer beneath the surface, labeled Thermocline, separates the Sun-warmed water near the ocean surface from the much colder waters of the deeper ocean below. Wide arrows along the equator indicate the direction of prevailing surface winds. Dashed lines depict the closed loop of vertical and horizontal circulation of air above the ocean, driven by convection and marked by towering tropical cumulus clouds. The upper diagram (a) is for "normal" conditions. When an El Niño is in progress (diagram b), the westward-blowing winds are reversed, warmer surface waters migrate eastward, and the thermocline becomes shallower in the west and deeper in the east. Regions of heavy tropical rain are also shifted eastward. (Based on a diagram by M. McPhaden, NOAA/PMEL.)

Figure 2. Two indices of oceanic and atmospheric conditions in the tropical Pacific that reveal the incidence of El Niño and La Niña conditions throughout the last 120 years. The Southern Oscillation Index (SOI), in blue, is a measure of the difference in barometric pressure between Tahiti and Australia. The NIÑO3 index, in black, indicates changes in sea-surface temperature in the eastern Pacific Ocean. As plotted here, peaks in either index indicate El Niño (i.e., warm) episodes; deep valleys are times of (cold) La Niñas. The near-perfect correspondence between the two indices is a reflection of how closely the tropical atmosphere and ocean are tied together in these ENSO events. (From Learning to Predict Climate Variations Associated with El Niño and the Southern Oscillation. National Research Council, Washington, DC, 1996.)

Figure 3. Enhancement of the tropical ocean observing system in the course of the international TOGA Program (1985-1995). Open circles are tide gage stations fixed on continental or island coastlines; black squares and diamonds are instrumented buoys that are held in place by deep anchors; arrows are instrumented, drifting buoys; continuous lines are the heavily-traveled lanes of commercial ships that supply observations of sea water temperature and other conditions on a voluntary basis. The denser observing network shown in the lower diagram, which remains in place, led to the breakthroughs that have made it possible to predict El Niño and La Niña events. (From McPhaden et al., Journal of Geophysical Research, vol 103, PP 14169-14240, 1998.)

Editor's Summary

Return to: Consequences Home Page
Return to: GCRIO Home Page
Next Article  
Consequences Volume 5 Number 2 1999

U.S. Global Change Research Information Office, Suite 250, 1717 Pennsylvania Ave, NW, Washington, DC 20006. Tel: +1 202 223 6262. Fax: +1 202 223 3065. Email: . Web: Webmaster: .
U.S. Climate Change Technology Program Intranet Logo and link to Home