SPARC logo (45 Ko)

Stratospheric Processes And their Role in Climate
A project of the World Climate Research Programme

Home Initiatives Organisation Publications Meetings Acronyms and Abbreviations Useful Links


The Arctic Oscillation and its role in stratosphere-troposphere coupling

Mark P. Baldwin, Northwest Research Associates, Bellevue, WA, USA (


In a white paper on the North Atlantic Oscillation (NAO) and tropical Atlantic variability, J. Marshall and Y. Kushir described climate changes that have taken place during the last few decades:

"Since the mid-1960s there has been a steady increase in wintertime storminess in the northeastern Atlantic and the North Sea. At the same time northern European countries bordering the Atlantic have experienced an upward trend in winter rainfall, while from the Iberian Peninsula to Turkey there has been a steady decline in rainfall. Winter air temperatures over northern Europe and Asia, from Scandinavia to Siberia rose steadily from the mid-1960s. During the same period the Middle East and North Africa cooled. Over Greenland and the Canadian arctic strong cooling occurred while the eastern half of the United States warmed slightly. In sub-Saharan Africa, rainfall steadily decreased after the wet decade of the 1950s, turning the Sahel in to a drought-stricken region. At the same latitude, mid-summer to late autumn tropical storms which were frequent during the 1940s and 50s became less frequent from the 1960s onwards. Some of these climatic trends may have been felt as far south as the semi-arid region of northeast Brazil, where austral summer rainfall, regulated by the annual migration of the Inter-tropical Convergence Zone (ITCZ), has displayed an upward trend."

Marshall and Kushnir then asked the following questions: "Is there a causal relation between these climatic events? Do they have a single orchestrating factor, and if so what is it? Can we discern the mechanisms that have governed these fluctuations? If so will we be able to foretell their future evolution?" Many of the answers may come from an understanding of a "mode of climate variability" called the Arctic Oscillation (AO).

The AO was first defined by Thompson and Wallace [1998] as the leading mode of variability of the extratropical Northern Hemisphere. They used an empirical orthogonal function (EOF) analysis of wintertime monthly-mean 1000hPa geopotential anomalies to obtain this leading mode, but it is important to realize that the AO is not just the leading mode of the surface, and not just of the winter months. It is also possible to obtain the AO without EOF analysis. Its strong zonal symmetry led Thompson and Wallace [2000] to call it the northern "annular mode." The terms "Arctic Oscillation," northern "annular mode," and "leading mode of the Northern Hemisphere" are all used in the literature to describe the same phenomenon.

The Arctic Oscillation "rivals ENSO in terms of its significance for understanding global climate variability and trends" [Wallace, 2000]. The well-known teleconnection pattern known as the North Atlantic Oscillation (NAO) may be regarded as the manifestation of the AO in the Atlantic sector, and maps of the AO and NAO over the Atlantic half of the Northern Hemisphere are nearly identical. Its similarity to the NAO has caused some confusion, and led Wallace to explain how the AO and NAO may be viewed as a single phenomenon viewed through two different paradigms. The AO provides a useful way through which to evaluate and understand climate variability and trends, the occurrence of significant weather events, and the role of stratosphere-troposphere coupling. The AO paradigm provides a framework in which remote linkages, such as between tropical Atlantic climate variability and North Atlantic variability can be better understood.

Since the 1960s the AO has experienced a "trend" toward a positive index, broadly consistent with the climatological changes described by Marshall and Kushnir, and possibly related to climate change. The changes in the AO at the Earth’s surface have been paralleled by a tendency for the high-latitude stratospheric polar vortex to be stronger and colder. The AO in the troposphere is strongly coupled to the strength of the stratospheric polar vortex, and stratospheric circulation anomalies are seen to propagate downward to the Earth’s surface, where they are reflected as changes in the magnitude and sign of the AO [Baldwin and Dunkerton, 1999]. Since the AO modulates the circulation of the northern troposphere, and even influences low latitudes, such as the strength of the trade winds across the subtropical and tropical Atlantic, intraseasonal changes in the stratospheric circulation are observed to precede both high-latitude and low latitude Atlantic climate anomalies and significant weather events. On longer timescales, modeling work by Shindell et al. [1999] has shown that increasing greenhouse gasses, which may be expected to cause a cooling and strengthening of the stratospheric polar vortex, may be linked to a trend toward a positive AO index, with hemispheric implications for surface climate. Also, the phase of the NAO (or AO) may influence the formation and intensity of hurricanes, and NAO-related statistics are used in seasonal hurricane formation.

Wallace [2000] noted that the AO paradigm transcends geographic and programmatic classifications, and therefore serves as a cross-cutting theme of interest to the climate prediction, Arctic, stratospheric, anthropogenic climate change, and atmospheric dynamics communities. The AO does not fall entirely within CLIVAR or SPARC, but it provides a framework in which possible significant advances in our understanding of climate variability may be made.

Spatial structure of the AO

Baldwin and Dunkerton [1999] expanded on the AO paradigm of Thompson and Wallace [1998], but calculated the leading mode as the first EOF from a single field consisting of geopotential at five levels (1000, 300, 100, 30, 10hPa), in order to study coupling of the stratosphere and troposphere. Their reasoning was that the leading mode of the lowest 30km of the atmosphere would better capture the relationship between the stratosphere and troposphere. Remarkably, their results were very similar to Thompson and Wallace, who used only the 1000hPa level. Figure 1 illustrates the signatures of the leading mode at nine levels, including the five levels used to define the AO. Each panel is calculated by a regression between the December-February AO index (the principal component time series from the EOF calculation) and geopotential. The panels are contoured in meters per standard deviation of AO index. The signature of the AO at 1000hPa, with the exception of the Pacific region, is nearly identical to SLP projected onto the NAO index (compare with Figure 2.1 of the Atlantic Climate Variability Prospectus).

Figure 1

Figure 1. Spatial signatures of the Arctic Oscillation at 10, 30, 50, 100, 200, 300, 500, 700, and 1000hPa. Each panel is produced by regressing the AO index (the time series of the first EOF of geopotential in the 1000?10hPa layer) with geopotential. Contour values are meters, corresponding to a one standard deviation anomaly in the AO Index. The contour interval of the 1000hPa panel is ±5, ±15,… for ease of comparison with Thompson and Wallace [1998]. The 1000hPa panel is very similar to the NAO pattern produced by J. Hurrell (e.g.,

The 500hPa panel again captures the same mode as Thompson and Wallace [1998], with one center of action over southern Greenland opposing a broad mid-latitude band. Figure 3a from Baldwin et al. [1994] illustrates that the 500hPa pattern resulting from singular value decomposition (SVD) between 50 and 500hPa is essentially similar to that shown in Figure 1 or that shown by Thompson and Wallace. In the stratosphere, the 50hPa pattern is much simpler and is dominated by a center of action displaced slightly off the pole toward Greenland, with a nearly zonally symmetric ring in mid-latitudes. As with the 500hPa pattern, the 50hPa pattern is nearly the same as either the SVD mode of Baldwin et al. [1994] or the AO signature of Thompson and Wallace [1998]. The 10hPa panel shows a pattern centered nearly over the pole, but with less zonal symmetry in mid-latitudes, featuring an opposing center of action over the eastern Pacific.

The above comparisons demonstrate that the AO pattern is extraordinarily robust, in that it may be recovered using a variety of data levels and techniques, and it is not sensitive to the number of months included in the winter season. Other than the seasonal cycle, the AO is the largest mode of variability in the extratropical northern troposphere-stratosphere.

Advantages of the AO Paradigm

Climate variability in the extratropical Atlantic region has been viewed largely through the paradigm of the NAO. The NAO may be described as "a large-scale see-saw in atmospheric mass between the subtropical high located near the Azores and sub-polar low near Iceland," and its influences are typically illustrated to extend from the east coast of the United States through Europe, and from the Tropical Atlantic through the North Atlantic to the Arctic. Fluctuations in the NAO index are associated with dramatic shifts in weather patterns over the Atlantic basin, and especially over Europe. Winds, storm tracks, rainfall, temperature, ocean circulation, fisheries, and related human impacts are all modulated by the phase of the NAO. Any predictability of the NAO would have immediate application and benefit over the entire Atlantic region.

The temporal variation of the NAO is often defined by an index of normalized, time-averaged pressure differences between stations representing its two centers of action, such as the Azores and Iceland. There is a key advantage to such a definition–the existing weather records allow the NAO index to be extended back in time to at least 1864. Further, by extending such an index through the use of paleo climate indicators such as tree rings and ocean sediments, proxies for the NAO and the tropical North Atlantic can be extended back several hundred years [e.g., Black et al., 1999]. These indices show that the NAO is apparently robust and has existed for hundreds of years. When the NAO index is correlated or regressed with gridded surface pressure data, the resulting north-south dipole pattern defines the spatial pattern of the NAO.

Reliance on only two stations necessitates time-averaging of the data, at least by month and more typically by season. Short-period fluctuations cannot be captured by such an index. Such station-based indices are not very well suited to representing the time-dependent behavior of the NAO [Wallace, 2000]. The spatial pattern of the NAO is very robust and is somewhat insensitive to the details of the index used, so that different time series can produce regression maps in surface pressure that are nearly identical [e.g. Hall and Visbeck, 2000]. This is an important consideration, as it allows a variety of indices to be used to define the spatial structure of the NAO, but it results in uncertainty in interpreting the physical meaning of the indices. A second consideration is that the spatial scale of the NAO is defined, and in effect limited, by the choice of index–the patterns will tend to be localized by the choice of stations.

Wallace [2000] carefully and eloquently explained that the AO and NAO represent a single phenomenon viewed through two paradigms. Wallace showed that the original definition of the NAO [Walker and Bliss, 1932], which is based on a linear combination of data from seven stations, defines a pattern nearly identical to the AO*. Wallace also demonstrated that the time series of the AO and that of Walker and Bliss’s original NAO index display an astounding correlation of 0.99. The AO/annular mode may be thought of as the leading wintertime low-frequency mode of variability of SLP, while the NAO is the leading mode in the Atlantic basin. Remarkably, maps of the two modes in the Atlantic region are nearly indistinguishable (but their respective indices are different).

* An occasional criticism of the AO paradigm is that the mode depends on a EOF calculation. The comparison with Walker's original NAO definition demonstrates that EOFs are not necessary to calculate the NAO or AO.Wallace [2000] speculates that Sir Gilbert Walker would have discovered the AO if hemispheric date had been available to him.

None of the names NAO, AO, or northern annular mode, are really adequate to describe the leading mode of variability of the Northern Hemisphere. It is sometimes pointed out that this mode is not confined to the North Atlantic or the Arctic, and that it does not oscillate, at least in the sense of a linear oscillator. Also, the mode is not zonally symmetric, especially in the troposphere, so it is not really annular. Indeed, some authors prefer to avoid all these names, employing terms such as "leading mode" or "leading coupled variability mode" [e.g., Perlwitz et al., 2000].

One key advantage of the AO/annular mode paradigm is conceptual. The AO may be thought of as a "free mode" of the northern troposphere [D. Hartmann, personal communication, 1999] that can be excited by a variety of sources. It may be thought of as a deep, hemispheric mode, with maximum variance over the North Atlantic, corresponding to the NAO centers of action. It is not surprising that such a free mode could act to connect widely separated areas of the troposphere (as opposed to "teleconnecting" specific regions). It is also natural to explore the existence of such a mode in the Southern Hemisphere [Thompson and Wallace, 2000]. Indeed, such "annular modes" are seen in zonal-mean zonal wind in numerical models; their existence does not depend on the Atlantic Ocean, or any ocean at all. In the case of the Earth’s Northern Hemisphere, land-sea contrasts (especially warm Atlantic Ocean extending to high northern latitudes [Wallace, 2000]) and mountain ranges produce the mode that we observe.

The concept of the AO superseding or coexisting with the NAO may cause some friction with programs and research focused around the NAO, but this need not be the case. The AO/annular mode paradigm, rather than diminishing the importance of the Atlantic region, inevitably leads to the conclusion that the Atlantic effects are more far-reaching and significant than previously thought. It is the Atlantic region that is profoundly linked to the leading mode of variability of the entire hemisphere.

Thompson and Wallace [personal communication, 1999] projected the surface AO signature map onto daily surface data to obtain a daily AO index. They confined their study to January?March, when the tropospheric and stratospheric circulations are coupled. They showed that the AO affects the frequency of occurrence of weather events such as blocking, cold air outbreaks, and high wind events throughout the Northern Hemisphere, including North America, Europe, and eastern Asia. They then selected high and low index days and examined a variety of weather events. Even in the Pacific Northwest, well outside the Atlantic basin (and not in a region highly correlated with the AO), they found that the AO index had a significant (at the 99% confidence limit) influence on weather events such as snow in Seattle (favored during low AO index days) and snow in the Cascade Mountains east of Seattle (favored during high AO index days).

Connection to the stratosphere

A strong link between the leading mode in the troposphere and the strength of the stratospheric circulation was shown in Kodera et al. [1991], Baldwin et al. [1994], and Perlwitz and Graf [1995]. In their white paper on Atlantic Climate Variability, Marshall and Kushnir noted that these studies demonstrated a strong statistical connection between the strength of the stratospheric cyclonic winter vortex and the tropospheric circulation over the North Atlantic. During winters of an anomalously strong stratospheric polar vortex, the NAO (or AO) tends to be in a positive phase with enhanced westerlies across the Atlantic, perhaps associated with changes in vertically propagating planetary waves. The strong vertical coupling may be largely due to the influence of the zonal-mean flow on vertically-propagating planetary waves. Another idea, suggested by Thompson and Wallace [1998] is the vertical coupling is accomplished, not by any particular planetary wave pattern, but by the zonally symmetric component of the flow itself, by interacting with whatever planetary waves happen to be present at the time. By thinking in terms of robust, dominant, hemispheric modes of variability, other fundamental connections to climate variability and the upper atmosphere have been made. Thompson and Wallace [1998] showed that the AO is, to date, the best measure of coupling of the tropospheric circulation to that of the stratosphere.

The AO index (the principal component time series) is defined only during the time period used in the EOF calculation, and for Figure 1 is based on December?February 75-day lowpass filtered data. A daily AO index for all seasons can be constructed by letting the spatial patterns in Figure 1 define the AO, then projecting these patterns onto daily data to obtain a unique AO index for each of the 17 data levels. By doing this, it can be seen whether the AO pattern in the stratosphere leads or lags the surface signature. Figure 2 illustrates the resulting time-height cross section of the AO, for 40 years of data from 1000 to 10hPa. The strong vertical coherence in Figure 2 shows that during winter, the stratospheric and tropospheric circulations tend to be coupled. It is also clear that the AO, especially outside the winter season, exists independently of the stratosphere. Figure 2 also demonstrates that large anomalies in the stratospheric circulation tend to propagate downward to the Earth’s surface, with an average propagation time of approximately three weeks. The downward propagation is variable, and is not apparent during some winters. The mechanism for downward propagation appears to be a result of wave, mean-flow interaction, but the details are not well understood. The downward propagation of the AO has been documented in a control run of the UKMO Unified GCM [P. McCloghrie, Oxford University, personal communication, 1999].

Figure 2

Figure 2. Time-height cross-section of the Arctic Oscillation, 1958-1997. The Arctic Oscillation is defined as the leading EOF of Northern Hemisphere 75-day low pass filtered geopotential anomalies in the 1000-10hPa layer. The EOF defines a spatial pattern of geopotential anomalies at each level from 1000 to 10hPa. The diagram shows a nondimensional measure of the magnitude of the patterns in Figure 1 as a function of height and time, in 75-day lowpass filtered data. Blue corresponds to a strong, cold polar vortex, while red indicates a warm, weak polar vortex. Red ticks indicate all major stratospheric warmings, while each red "C" indicates a less severe "Canadian" warming. The diagram illustrates that the characteristic Arctic Oscillation pattern in the middle stratosphere tends to occur before the corresponding pattern in the troposphere, with a variable downward propagation time averaging about three weeks. The downward propagation occurs for either positive or negative anomalies, but does not always occur.

The tight relationship between sudden stratospheric warmings and the AO is easily seen in Figure 2, wherein all major warmings are marked with a broad red tick, and Canadian warmings have a red "C." Within the stratosphere, the weak vortex during warmings is seen as a negative AO index. The multi-level AO index provides a statistic that allows the warming to be traced all the way to the Earth’s surface, in most cases. This view is consistent with synoptic studies of sudden warmings [e.g. Sherhag et al., 1970], which showed that some major warmings began as high as 60km, and progressed downward, in some cases, to the Earth’s surface.

In Figure 2 the shift over 40 years from predominantly red to predominantly blue reflects a trend toward positive AO index which has been discussed by Thompson and Wallace [1998] and Thompson et al. [2000]. This cooling trend in the stratosphere may be related to the observed warming trend in the troposphere, and could be enhanced by increasing greenhouse gas concentrations. Shindell et al., [1999] used several climate models to demonstrate that a trend in the Arctic Oscillation with increasing greenhouse gasses is captured only in models that have a well-resolved stratosphere. They concluded that the proper representation of stratospheric dynamics appears to be important to the attribution of climate change. Similarly, the phase of the AO can be affected by explosive tropical volcanic eruptions [Robock and Mao 1992].

The possibility that the quasi-biennial oscillation (QBO) [Baldwin et al., 2000] in the equatorial stratosphere acts to force the AO was discussed by Baldwin and Dunkerton [1999]. It appears that the phase and strength of the AO, especially in the stratosphere, are influenced by the QBO in a manner consistent with the QBO’s effect on the strength of the polar vortex. The AO tends to be in its negative phase (weak vortex) when the QBO is easterly.

In a series of papers Kodera and colleagues [e.g., Kodera et al., 2000] have shown that within the stratosphere, anomalies in zonal-mean zonal wind tend to propagate northward and downward during the winter season. The results have been confirmed by numerical experiments in which circulation anomalies begin as high as the lower mesosphere, and move slowly northward and downward, over a period of 2?3 months. The propagation is described well in the phase space of the first two EOFs of zonal-mean zonal wind, which tend to trace out a circular path, especially during sudden warmings. A synoptic interpretation of this process is the development of a sudden warming with a reduction in strength of the polar vortex and downward progression of the warming, sometimes to the Earth’s surface. The resulting surface signature of this process, which Kodera calls the Polar Jet Oscillation (PJO), is that of the Arctic Oscillation. The relationship between the AO and the PJO is not clear; in the lower stratosphere one could call the leading mode of variability the AO or the PJO.

Questions and Challenges

The introduction of the Arctic Oscillation has resulted in a flurry of new research and a refocusing of other investigations. With climate impacts potentially as important as ENSO, the AO is not yet well understood. Several fundamental questions remain, and there is active debate as to which paradigm is most useful, and even what name to use.

The SPARC Scientific Steering Group has organized an informal study group on stratosphere-troposphere coupling. The group will at first be based on an e-mail list of interested persons. If you would like to be added to the mailing list, please contact


Baldwin, M.P., X. Cheng, and T.J. Dunkerton, Observed correlations between winter-mean tropospheric and stratospheric circulation anomalies, Geophys. Res. Lett., 21, 1141-1144, 1994.

Baldwin, M.P., and T.J. Dunkerton, Propagation of the Arctic Oscillation from the stratosphere to the troposphere, J. Geopys. Res., 104, 30,937-30,946, 1999.

Baldwin, M.P., L.J. Gray, T.J. Dunkerton, K. Hamilton, P.H. Haynes, W.J. Randel, J.R. Holton, M.J. Alexander, I. Hirota, T. Horinouchi, D.B.A. Jones, J.S. Kinnersley, C. Marquardt, K. Sato, and M. Takahashi, The Quasi-Biennial Oscillation, Reviews of Geophysics, 38, submitted, 2000.

Black, D.E., L.C. Peterson, J.T. Overpeck, A. Kaplan, M.N. Evans, and M. Kashgarian, Eight centuries of North Atlantic ocean variability, Science, 286, 1709-1713, 1999.

Hall, A., and M. Visbeck, Annular modes and their climate impacts in the GFDL coupled ocean-atmosphere model, J. Clim., 11, submitted, 2000.

Kodera, K., K. Yamazaki, M. Chiba, and K. Shibata, A possible influence of the polar night jet on the subtropical tropospheric jet, J. Meteorol. Soc. Japan, 69, 715-721, 1991.

Kodera, K., Y. Kuroda, and S. Pawson, Stratospheric sudden warmings and slowly propagating zonal-mean zonal wind anomalies, J. Geophys. Res., 105, submitted, 2000.

Perlwitz, J., and H-F. Graf, 1995: The statistical connection between tropospheric and stratospheric circulation of the Northern Hemisphere in winter. J. Climate, 8, 2281-2295.

Perlwitz, J., H.-F. Graf, and R. Voss, The leading variability mode of the coupled troposphere-stratosphere winter circulaton in different climate regimes, J. Geophys. Res, 105, to appear, 2000.

Robock, A., and J. Mao, 1992: Winter warming from large volcanic eruptions. Geophys. Res. Lett., 19, 2405-2408.

Sherhag, R., K. Labitzke, and F.G. Finger, Developments in stratospheric and mesospheric analyses which dictate the need for additional upper air data, Met. Monographs, 11, 85-90, 1970.

Shindell, D.T., R.L. Miller, G.A. Schidt, and L. Pandalfo, Simulation of recent northern winter climate trends by greenhouse-gas forcing, Nature, 399, 452-455, 1999.

Thompson, D.W.J., and J.M. Wallace, The Arctic Oscillation signature in the wintertime geopotential height and temperature fields, Geophys. Res. Lett., 25, 1297-1300, 1998.

Thompson, D.W.J., and J.M. Wallace, Annular modes in the extratropical circulation. Part I: Month-to-month variability. J. Clim., to appear, 2000.

Thompson, D.W.J., J.M. Wallace, and G. Hegerl, Annular modes in the extratropical circulation. Part II: Trends, J. Clim., to appear, 2000.

Walker, G.T, and E.W. Bliss, World Weather V, Memoirs of the R. M. S., 4, 53-83, 1932.

Wallace, J.M., North Atlantic Oscillation / Annular Mode: Two paradigms ? One Phenomenon, Quart. J. Royal Met. Soc., to appear, 2000.


Back to SPARC Newsletter 14 Homepage