This paper describes recent variations of the North Atlantic eddy-driven jet stream and analyzes the mean response of the jet to anthropogenic forcing in climate models. Jet stream changes are analyzed both using a direct measure of the near-surface westerly wind maximum and using an EOF-based approach. This allows jet stream changes to be related to the widely used leading patterns of variability: the North Atlantic Oscillation (NAO) and East Atlantic (EA) pattern. Viewed in NAO–EA state space, isolines of jet latitude and speed resemble a distorted polar coordinate system, highlighting the dependence of the jet stream quantities on both spatial patterns. Some differences in the results of the two methods are discussed, but both approaches agree on the general characteristics of the climate models. While there is some agreement between models on a poleward shift of the jet stream in response to anthropogenic forcing, there is still considerable spread between different model projections, especially in winter. Furthermore, the model responses to forcing are often weaker than their biases when compared to a reanalysis. Diagnoses of jet stream changes can be sensitive to the methodologies used, and several aspects of this are also discussed.
There is an emerging consensus among climate models that the midlatitude jet streams will shift poleward in response to greenhouse gas forcing (Meehl et al. 2007). The subtropical jets are expected to shift in line with an expansion of the tropics (Lu et al. 2007), and there is evidence that the eddy-driven component of the zonal wind will also shift poleward. The eddy-driven jets owe their existence to the westerly mean flow forcing associated with transient baroclinic eddies. These eddies accelerate the westerly flow in particular in the lower troposphere so that the eddy-driven jet can be distinguished from the subtropical jet by consideration of the low-level wind field. As shown by Lorenz and DeWeaver (2007), the multimodel mean response of the 850-hPa midlatitude westerly wind to greenhouse gas forcing shows a poleward shift in all regions and seasons. The transient eddy forcing that drives the jet streams is concentrated in the midlatitude storm tracks, and models are beginning to show some consistency in a poleward shift of these (Yin 2005). Jet stream changes are often described using patterns of variability such as the North Atlantic Oscillation (NAO) and the Northern Annular Mode (NAM), and by these measures there is also some level of qualitative agreement between models (Miller et al. 2006; Stephenson et al. 2006).
Encouraging though these consistent signals are, there are still considerable differences between the predictions of different climate models (Räisänen 2003; Miller et al. 2006; Sigmond et al. 2007). The principal aim of this paper is to demonstrate that this is particularly true for the North Atlantic jet stream in winter. The mean wind response is weak in this case (Lorenz and DeWeaver 2007), reflecting considerable spread in the regional atmospheric circulation responses predicted by the models (Woollings 2010). Yin (2005) described the storm track response in the zonal mean and, looking beyond this to regional details, reveals a different signature in the North Atlantic, comprising a downstream extension of the storm track in the multimodel mean, but again with considerable differences between individual models (Ulbrich et al. 2008; Laîné et al. 2009).
While Atlantic jet stream changes are often described using the NAO or annular mode frameworks, recent work has shown that in general more than one spatial pattern is required for this (Fyfe and Lorenz 2005; Seierstad et al. 2007; Sparrow et al. 2009). The leading circulation patterns represent a mixture of changes in the latitude, strength, and speed of the jet streams (Monahan and Fyfe 2006). Specifically in the North Atlantic, much of the jet stream variability can be described using a combination of the two leading patterns, namely the NAO and the East Atlantic (EA) pattern (Woollings et al. 2010, hereafter W10). In this paper we use both a direct jet latitude identification method and the combined NAO–EA framework to examine the response of the North Atlantic jet stream to greenhouse gas forcing in climate models. We also use these methods to describe jet stream changes in the recent historical record and we investigate various sensitivities to choices of methodology. The focus is on the winter season, December–February (DJF), but some results for summer, June–August (JJA), are shown for comparison.
2. Climate model simulations
We use data from 22 coupled climate models in the Coupled Model Intercomparison Project phase 3 (CMIP3) archive, which contributed to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC AR4). For present-day and future periods we use years 1960–99 of the twentieth-century control scenario (20C3M) and 2060–99 of the Special Report on Emissions Scenarios (SRES) A1B scenario, which is a commonly used midrange emissions scenario. The models are described by Randall et al. (2007), and many characteristics of the forced response are described by Meehl et al. (2007) and Christensen et al. (2007). Where available, up to three SRES A1B ensemble members from each model have been used to illustrate the uncertainty due to natural variability. A list of the models used is given in Table 1.
The multimodel mean response of 850-hPa zonal wind to forcing is shown in Fig. 1a for the winter season (DJF). The mean response does indeed constitute a poleward shift of the jet stream, in line with Lorenz and DeWeaver (2007) and the other studies described above. The response is largest over Europe downstream of the jet core. This is consistent with the mean flow forcing expected from the downstream extension of the storm track (Ulbrich et al. 2008). However, while the mean response seems clear, there is considerable spread between the different models, as evidenced by the standard deviation across the set of models (Fig. 1b). This is almost everywhere larger than the mean response, particularly in the core jet stream region over the Atlantic Ocean. Only farther downstream, over Europe and North Africa, is the mean response stronger than the standard deviation, suggesting an extension of the jet stream, which is more robust than the shift over the ocean basin.
Figure 2 shows examples of the response in four different climate models, showing that the spread in model responses is not simply quantitative. There are qualitative differences between these models in how the jet stream responds to forcing, that is, whether it shifts north or south and whether it strengthens or weakens. Note that these four models have been chosen as examples because of their differing responses. As suggested by the multimodel mean, several models do predict a poleward shift and a strengthening of the jet stream, as shown by the GFDL CM2.1 model. In addition, several models predict very weak responses, similar to HadGEM1. Figure 2 also serves to illustrate the considerable differences between the models in their representation of the twentieth-century period. The model climatologies and their responses in both winter and summer will be discussed further in section 4. First, we examine the behavior of the jet stream over the recent past and develop diagnostic methods to be applied to the model data.
3. The recent past
To characterize recent jet stream variations, we use data from the 40-yr European Centre for Medium-Range Weather Forecasts Re-Analysis (ERA-40) covering the period November 1957–August 2002 (Uppala et al. 2005). We use the diagnostics of jet stream latitude and speed derived by W10. These use the low-level zonal wind to specifically identify changes in the eddy-driven component of the zonal flow. The methodology is as described in W10, with the exception that the seasonal cycle is not removed here. Briefly, the daily mean zonal wind field is averaged over the levels 700–925 hPa in the sector 0°–60°W. A 10-day low-pass filter is then applied, and the jet speed and latitude on each day are simply identified by the maximum value of the resulting zonal wind section. In this paper we focus on the winter-mean jet latitude and speed, obtained by simply averaging the daily jet latitude and speed indices. By averaging the daily indices in this way the influence of random errors in the identification of the jet in individual wind profiles is reduced.
Figures 3a,b show time series of the jet latitude and speed in wintertime derived using this method. There are interesting contrasts between the two time series. The jet latitude exhibits a consistent poleward trend over this period,1 as seen by Strong and Davis (2007). Franzke and Woollings (2011) also find a significant trend using the same underlying jet stream diagnostic, although their approach does not distinguish between individual seasons. Despite large fluctuations from year to year the jet speed has remained relatively constant on longer time scales, apart from in the 1960s when it was unusually weak. The NAO describes changes in both the latitude and speed of the jet, so it is tempting to infer that these two quantities vary together to some extent. In fact, the correlation between the two time series in Fig. 3 is only 0.13. The jet latitude and speed are therefore largely independent, which suggests that they have had different dynamical influences over the ERA-40 period. It is clear that more than one index is needed to characterize the jet variability.
It is of interest to determine the spatial flow patterns that are linearly associated with changes in the jet latitude and speed. These are shown in Figs. 3c,d by regressing the monthly mean 500-hPa geopotential height (Z500) onto monthly means of the jet latitude and speed. The pattern associated with variations in latitude resembles the NAO, but the action centers are closer together so that the gradient between them occurs in a more concentrated region. The pattern associated with jet speed variations resembles a southward-shifted NAO pattern. As shown by W10, this means that it has an important contribution from the East Atlantic pattern.
It is clear that the jet latitude and speed patterns are not orthogonal, so they could never arise as separate EOFs in a principal component analysis. This is consistent with the work of Monahan and Fyfe (2006), who showed that EOF patterns in general cannot differentiate between different kinematic quantities such as the jet latitude and speed. The mixture of different physical effects is a well-known limitation of EOF analysis (e.g., Monahan et al. 2009). Despite this, there are some advantages to an EOF-based approach. First, the approach is objective, while there are several choices to be made in a direct analysis of the jet latitude and speed. Second, it provides a scaling with which to compare variations in different kinematic quantities. For example, using a jet identification approach, it is not clear how to compare the importance of a jet shift compared to a jet pulse. However, an EOF-based approach enables a comparison of these features according to the fraction of the flow variance explained. In this paper we apply both the direct jet stream and the EOF-based NAO–EA analyses to the climate model data and compare the results of the two methods. In this way we identify jet stream changes that are robust to the method of identification and also show how the jet changes relate to changes in these widely used circulation patterns.
To define the NAO and EA patterns, we use the first two EOFs over a confined Atlantic sector (20°–90°N, 75°W–15°E) in order to focus exclusively on variations related to the Atlantic jet stream. We use the monthly mean geopotential height integrated vertically from 1000 to 250 hPa (denoted zint). This is because the climate model responses to anthropogenic forcing are generally baroclinic so that the diagnosed jet stream response can be qualitatively different at different levels (Woollings 2008). An integrated measure over the depth of the troposphere seems most objective, but this is discussed further in section 5. Figure 4 shows the resulting EOF patterns in DJF, which we take to represent the NAO and EA patterns due to their close similarity with patterns in previous studies (e.g., Barnston and Livezey 1987).
The time series of the NAO and EA over the ERA-40 period are plotted in Figs. 5a,b. There is some similarity between the time series of the NAO and the jet latitude and between the EA and the jet speed, as might be expected given the patterns in Figs. 3c,d. However, these relationships are far from exact. Figure 5c introduces the NAO–EA state space, which will be used to diagnose the climate model behavior. The contours in Fig. 5c show how the NAO and EA patterns together describe changes in the eddy-driven jet stream. To derive these a linear regression approach has been used, in which the 850-hPa zonal wind has been regressed onto the time series in Figs. 5a,b. Each point in the state space is defined by its NAO and EA values and these are used with the regression patterns to derive the wind pattern at each point in the state space. The jet is identified in each of these patterns simply as the maximum westerly wind value as above. This provides fields of jet latitude and speed as a function of the NAO and EA indices: it is these fields that are contoured in Fig. 5c.
As seen in W10, the transformation from the NAO–EA to jet latitude and speed resembles a distorted polar coordinate transformation. An increase in jet latitude at constant speed would trace a clockwise arc in NAO–EA space, while an increase in jet speed at constant latitude would trace a relatively linear path away from the lower-left-hand quadrant. In several places the jet latitude contours become tightly bunched. This illustrates the limited ability of the two leading EOFs to describe all jet stream changes, although variations in latitude between 44° and 50° are well represented. The trace in this figure illustrates the low-pass filtered evolution of the system over the ERA-40 period. Several features are in good agreement with the direct jet stream diagnoses in Figs. 3a,b, for example the gradual weakening of the jet over the first 10 yr, followed by a more rapid strengthening and several clear meridional shifts each lasting several years. It is clear that both the NAO and the EA contribute to describing these variations.
4. Model results
a. Direct analysis
To directly analyze the jet latitude and speed in the models we have applied the simple jet identification method to the 850-hPa zonal wind field. This is performed for each month using the monthly mean wind data from the CMIP3 database and then the results are averaged to provide winter and summer climatological mean jet latitudes and speeds. The results for jet latitude in winter are shown in Fig. 6a with each model represented by an arrow pointing from its twentieth-century value to its twenty-first-century value. Most of the arrows lie below the horizontal line, which marks the ERA-40 jet latitude, showing that an equatorward bias in jet latitude is common to many of the models. Many of the models do predict a poleward shift of the jet in response to the forcing, but there are also several that predict a southward shift. Most of the models predict a weak change in jet latitude, with the response to forcing often being of smaller magnitude than the model bias. In the ensemble mean the jet is biased equatorward and exhibits a weak poleward shift of only ½° of latitude. Consistent with Fig. 1, the ensemble mean response is smaller than the spread between different models.
The results for jet speed are shown in Fig. 6b. Most of the models have an Atlantic jet that is too strong compared to ERA-40. There is no agreement between models on the response of jet speed to forcing, with the jet strengthening in some models and weakening in others. Shown in Fig. 6c is a simple measure of the tilt of the jet stream. The pronounced southwest–northeast tilt is a key characteristic of the North Atlantic jet, which is much less apparent in the Pacific and in the Southern Hemisphere. To measure the tilt we have simply identified the jet latitude in the eastern (0°–30°W) and western (30°–60°W) sectors of the Atlantic separately and taken the difference between them. This shows that most of the models underestimate the tilt of the jet stream, and there is no consistent response of the tilt to the forcing.
This analysis shows that in wintertime many CMIP3 climate models still exhibit the bias that has plagued generations of climate models. This can be summarized as a tendency to be too zonal (e.g., Woollings 2010) with a North Atlantic jet located too far equatorward, with insufficient tilt and excessively large wind speeds. The same analysis for the summer season (Fig. 6) shows less systematic bias. There is still a tendency for the jet latitude to be biased to the equatorward side, but the jet speed and tilt show little bias in the ensemble mean. There is, however, a wide range of behavior between the models, with some models over- or underestimating the jet speed or tilt by almost a factor of 2.
For both summertime jet speed and tilt the response to forcing is weak and inconsistent between models. However, there is general agreement between the models on a poleward shift of the summer jet in the twenty-first century, and this is reflected in the ensemble mean. Only a few models do not predict this, and the response is very weak for those models. The poleward jet shift is in general of the same magnitude as the bias in the models.
Finally, we investigate whether the presence of tilt in the jet stream influences the diagnoses of jet latitude and speed, in which the wind is simply zonally averaged across the Atlantic sector. To do this the jet latitude and speed were derived for each of the eastern and western sectors independently and the results were averaged to provide simple estimates of the jet latitude and speed in the absence of tilt. These estimates were found to be very highly correlated with the original method used above, with correlations of the jet latitude values of 0.99 in DJF and 0.94 in JJA, and a jet speed correlation of 0.99 in both seasons. There is, however, a consistent offset in jet speed, which is about 1 m s−1 higher when the tilt is taken into account. This reflects the smoothing of the jet profile when the tilt is not taken into account.
b. NAO–EA analysis
We now proceed to analyze the time-mean jet streams in the CMIP3 models using the EOF-based NAO–EA approach introduced above. To do this we calculate the climatological mean of the 1000–250-hPa vertically integrated geopotential height (zint) from each model run and subtract the ERA-40 climatology. The resulting anomaly field is then projected onto the NAO and EA patterns derived from ERA-40 (Fig. 4). To remove any influence of the global lifting of pressure surfaces in a warming atmosphere, the global mean difference of zint from the ERA-40 value is subtracted from each model run prior to projection. Some of the models have missing data in regions of high orography, such as over Greenland. In these cases the geopotential height is extrapolated down to 1000 hPa from the lowest level with temperature data, using hydrostatic balance and assuming a lapse rate of 6.5 K km−1. In this approach each model state is defined by a location in the NAO–EA state space as determined by ERA-40. This provides a sound framework with which to compare the model simulations to each other and to the observations. An alternative approach would be to use the NAO and EA patterns identified in each individual model, but then we do not have a consistent framework to compare the model biases. The alternative approach has been investigated elsewhere (e.g., Miller et al. 2006).
The results are shown for DJF in Fig. 8. Each arrow in this figure represents the results of one climate model. The base of the arrow corresponds to the location of the model’s twentieth-century control run with respect to the observed NAO and EA patterns. The origin corresponds to the full ERA-40 climatology, so the departure of each arrow’s base point from the origin reflects the bias in the model’s control state. Similarly, the head of the arrow marks the location of the A1B scenario model run so that the direction and length of the arrow describes the response of the model to anthropogenic forcing. The coordinates of each arrow are given in Table 1 to enable identification of the models.
Several conclusions can be drawn from this figure. First, many of the models have strong biases in their control runs. These often, but not exclusively, lie in the NAO−/EA+ quadrant, corresponding to equatorward biases in the jet location and positive biases in the jet speed. These biases are in agreement with the results of the direct analysis. One unit of the axes corresponds to one standard deviation of the monthly variability in ERA-40, which gives an indication of the scale of the biases. Several models have biases of around one standard deviation of the observed variability.
Second, there is considerable spread in the model responses to forcing, whether viewed with respect to the NAO–EA axes or the jet latitude and speed contours. As expected, there are several models that predict responses pointing in the NAO+/EA+ direction, corresponding to a strengthening and poleward shift of the jet. This is also reflected in the multimodel mean, shown by the red arrow. However, there are other models that predict equatorward shifts, and/or weakenings. There are also several models that show very weak responses, and here again the response to forcing is generally weaker than the model bias. The responses are also smaller than the observed change over the ERA-40 period, shown by the blue arrow, although this of course represents a shorter time scale change. Individual SRES A1B ensemble members are marked where available. These show significant spread in a few models, underlining that internal variability is not well sampled in the CMIP3 ensemble (Deser et al. 2012).
Figure 9 shows the equivalent picture for JJA. First, we note that the model biases appear even larger than those in winter, in some cases being close to two standard deviations of the observed variability. This may partly reflect the weaker observed variability in summer compared to winter, since the summertime biases are no larger than the wintertime biases in the direct analysis of Figs. 6 and 7. The model biases suggest a link between jet latitude and speed, with the more equatorward jets tending to be weaker. As in the direct analysis, there is greater agreement between the different model responses in this season. Many of the models show an increase in the NAO and EA values in response to forcing, and the contours suggest that this essentially corresponds to a poleward shift of the jet stream with little change in its strength. There are, however, a few models that, instead, show a weakening and/or equatorward shift of the jet. There is little observed change over the ERA-40 period in JJA.
As described in Woollings (2008), the atmospheric circulation response to greenhouse gas forcing has a baroclinic structure. A practical implication of this is that analyses performed at different vertical levels can give very different results. As an alternative to Fig. 8, we show in Fig. 10 results of a similar analysis but using the zonal wind at 850 hPa both to define the coordinate space and characterize the models. In this case, missing data values over orography are ignored, as the EOF patterns of the wind feature low values in these regions in any case. As described above, the low-level wind is specifically linked to changes in the eddy-driven, rather than subtropical, jet stream. Using this approach, the disagreement between models appears smaller but it is still evident. The low-level winds are generally too strong in the twentieth-century controls, but their latitudinal biases are smaller than those for integrated height. As before, many of the model responses are small, when compared to both the model biases and the level of observed variability.
Finally, in this section we note that in all of Figs. 8–10, the models occupy a region of state space where the relations between the jet variables and the patterns of variability are monotonic. This means that, in practice, an increase in the NAO in a model corresponds to a poleward shift of the jet and an increase in the EA corresponds to a strengthening of the jet. This is of course not the case for the state space in general. The relative utility of the two patterns is analyzed further in the next section.
5. Sensitivity to methodology
a. Comparison of direct and NAO–EA approaches
In this section we compare the results of the two different approaches. Figures 11a,b show scatterplots of the wintertime and summertime CMIP3 jet latitudes derived by the zint NAO–EA method versus those from the direct analysis. In general, the methods correlate reasonably well although there is some disagreement, especially in JJA. In DJF the disagreement between methods is greatest when the jet is far to the south, where the contours of jet latitude are tightly bunched in Fig. 8. The agreement between the methods is improved slightly when the EOFs are defined using the 850-hPa zonal wind, as might be expected (Fig. 11c).
The same analysis for the jet speed is shown in Figs. 11d–f. As seen for jet latitude, there is a relatively high correlation between the two methods despite some quantitative differences. The U850 EOF-based approach again shows a higher correlation with the direct analysis than the zint method does, although this time the zint method performs better in summer than winter. In all of Figs. 11d–f the majority of points lie below the diagonal, showing that the EOF-based approach systematically underestimates the jet speed when compared to the direct approach. We hypothesize that this arises due to the smoothing effect of using large-scale patterns to define the wind field.
The average correlation from Figs. 11a,b,d,e is 0.79, suggesting that the combined NAO–EA approach captures 62% of the spread between different jets in these models, when compared to the direct analysis. This comparison can be manipulated to infer the relative utility of the NAO and EA patterns in describing jet stream changes. For example, if the analysis is repeated using only the NAO zint pattern rather than both the NAO and EA, the correlation with the direct analysis for wintertime jet latitude decreases only slightly from 0.86 to 0.75. In contrast, if only the EA pattern is used, the correlation drops to 0.41. This shows that the NAO is clearly the dominant EOF in explaining jet stream shifts. The results for jet speed are different: if only the NAO is used, the correlation drops from 0.71 to 0.55, while if only the EA is used, the correlation is 0.58. This shows that the NAO and the EA have roughly the same power in describing changes in jet speed, although using the two patterns in combination gives significantly higher power.
Finally, we note that, despite some quantitative differences between the results of the two methods, the general conclusions with regard to the climate models are the same from each. Both methods show that most of the models have jets that are located too far equatorward in both seasons and are too strong in winter. Both methods show a consistent poleward shift of the jet in response to forcing, but only in summer. In winter there is considerable spread between the models that is much larger than the ensemble mean response. There is no signal of a consistent response in jet speed in either season.
b. Other sensitivities
There are other methodological choices to which the diagnosis of jet streams in climate models can be sensitive. The aim of this section is to highlight some that are particularly relevant to the work presented here. We begin by comparing two methods to determine the jet latitude and speed from the zonal wind field. Figure 12 shows, as examples, the DJF-mean zonal wind averaged over 925–700 hPa, 0°–60°W for the first 16 winters of ERA-40. First, the maximum wind speed in each section is marked as a dot: this provides a simple measure of the jet latitude and speed, as in W10 and as used here in Fig. 3. There are instances where this method is too crude, most notably for the 1965 section, but in general the point identified is reasonably representative of the jet stream as a whole.
The second method is based on area-weighted integrals, following Archer and Caldeira (2008). In this approach the jet latitude φ0 and speed U0 are defined by
The crosshairs in Fig. 12 show the values derived using this method for the example years. Again, this method often gives reasonable results for jet latitude but there are cases where the values are not especially representative of the jet stream as a whole (e.g., in 1958). Jet speeds are smaller using this method, and are less representative of the jet core. These examples highlight the difficulty of robustly identifying the jet latitude and speed. Other methods may provide more robust results, such as the Gaussian fitting of Monahan and Fyfe (2006), but this is not pursued here. Another means of reducing the impact of such nonrepresentative fits is to apply the procedure to daily or monthly data and then average the jet diagnostics up to longer time scales, as is done here and in Barnes and Hartmann (2010).
The sensitivity shown in Fig. 12 is still relevant to the results presented using the EOF-based approach because the jet speed and latitude contours in Figs. 8–10 have been derived using the first of the two methods discussed above. For comparison, Fig. 13 shows how the wintertime jet diagnostics as defined by the area-weighted approach vary as functions of the NAO and the EA. The nonlinearity is reduced when compared to the previous versions, but the overall pattern is similar. This shows that, while the EOF-based approach is more objective, some subjectivity remains in how the jet diagnostics are inferred from the EOFs. The differences between the contours in Figs. 8–10 and 13 are greatest for large magnitudes of the NAO and EA, suggesting that this analysis is not so robust in these areas of state space. This is in agreement with the results of Fig. 11a. However, most of the observational and model states plotted in Figs. 5c and 8–10 lie in an area of state space that is less sensitive to the method of jet identification.
Finally, we note that there are of course also sensitivities inherent in the EOFs themselves. As an example, we use the time series of the NAO and EA obtained from the NOAA Climate Prediction Center (CPC), which are based on rotated EOF analyses of the 500-hPa height,2 and show these in Fig. 14 for comparison with Fig. 5. One of the clearest differences from Fig. 5 is that the CPC EA pattern features a positive trend over the ERA-40 period, while the version derived from regional EOF analysis does not. Differences between the patterns themselves are minimal, as shown in Figs. 15a,b. Comparison with the map of linear trends in Z500 (shown in Fig. 15c) suggests that the subtle differences in the two EA patterns over Europe likely lead to the difference in trend shown by the two time series.
Figures 5c and 14c show that, despite this strong sensitivity of the trend to subtle differences in the spatial pattern, both methods reflect a strengthening and poleward shift of the jet stream over the ERA-40 period. One difference is that the CPC version shows the jet speed increasing more steadily throughout the ERA-40 period, while in the regional EOF version much of the increase in speed occurs during the first couple of decades. This is in better agreement with the direct analysis of Fig. 3, suggesting that the changes over the later decades in the CPC version reflect the trend over Europe. The advantage of the regional EOF approach used here is that the analysis is not contaminated with information from outside the Atlantic sector.
6. Investigating mechanisms
This paper has focused on diagnosing changes in North Atlantic jet characteristics in the CMIP3 models. While there is some support for the expected consistent poleward shift of the jet in the twenty-first century, especially in summer, one of the key results is the large spread between different models. It is very important to understand the mechanisms responsible for any predicted changes, so in this section we briefly examine two possible reasons for the spread in jet latitude response.
We first examine the results for evidence of a relation between a model’s response to forcing and its bias in twentieth-century jet latitude. Such a relation has been suggested in the Southern Hemisphere by Kidston and Gerber (2010) and in the North Atlantic by Barnes and Hartmann (2010). Figure 16a shows the relation between the response and the control value of wintertime jet latitude from the direct analysis of Fig. 6a. This shows only a very weak negative correlation, indicating that the differences in the jet latitude in the twentieth-century control simulations explain little or none of the model spread in the responses in this region.
The second potential cause examined here is climate sensitivity, which still varies widely between climate models. This is assessed by comparing the jet latitude response to the change in global mean surface air temperature over the same periods. The result (Fig. 16b) is a very weak positive correlation, suggesting that the differences in climate sensitivity between models also cannot explain the wide spread in predicted jet shifts.
The main conclusions of this paper are as follows.
The CMIP3 climate models show some agreement on a poleward shift of the jet stream in response to anthropogenic forcing. The spread between model projections is still large, however, both quantitatively and qualitatively. This is especially true in winter when the ensemble mean shift is very weak compared to natural variability. There is no consistent response of the jet speed or tilt to anthropogenic forcing.
Many climate models have jet stream biases (most often equatorward biases) that are larger than the models’ response to anthropogenic forcing. Many models are still too zonal in winter, underestimating the tilt of the jet and overestimating its speed.
The spread between models in the projected jet shift cannot be explained by either the differences in the control-state bias or climate sensitivity of the models.
The jet speed and latitude have varied independently over the ERA-40 period, suggesting that they may have different dynamical influences.
The NAO and EA patterns can be combined to describe climatological changes and interannual variability in the North Atlantic eddy-driven jet stream. This description is not perfect, however, capturing 62% of the spread between different jets in these models when compared to the direct analysis. Within this framework it is largely the NAO that describes jet shifts, whereas the NAO and the EA are of roughly equal importance in describing changes in jet speed.
We have also shown that diagnosis of the jet stream responses can be sensitive to various choices in the methodology used. In particular, the choice of vertical level is important. When only the low-level zonal wind is used, there is better agreement between the models. As in Woollings (2008), this suggests that the eddy-driven component of the flow may be responding more consistently across the models.
The layout of continents, mountain chains, and sea surface temperatures means that the North Atlantic jet stream is distinct from those in the North Pacific and much of the Southern Hemisphere in that the eddy-driven and subtropical jets are separated even in the time mean (Seager et al. 2002; Wilson et al. 2009; Brayshaw et al. 2009, 2011). This means that the response of the eddy-driven jet to forcing may be different in the North Atlantic to that seen elsewhere (Son and Lee 2005). In addition, the North Atlantic jet is particularly influenced by many different factors from the ocean to the stratosphere (e.g., Woollings 2010). Uncertainty in the jet stream response has a significant contribution to the uncertainty in the climate response to anthropogenic forcing (Karpechko 2010). Identifying which factors make the largest contributions to the jet stream uncertainty is an important topic for future research.
We are indebted to ECMWF for providing the ERA-40 reanalysis data and to the reviewers for their constructive comments. We acknowledge the modeling groups, the Program for Climate Model Diagnosis and Intercomparison (PCMDI), and the WCRP’s Working Group on Coupled Modelling (WGCM) for their roles in making available the WCRP CMIP3 multimodel dataset. Support of this dataset is provided by the Office of Science, U.S. Department of Energy.
List of Institutional Models
The institutional model names, and the acronyms used in Table 1, are listed below.
BCM2.0 Bjerknes Centre for Climate Research (BCCR) Climate Model, version 2.0
CCCma 3.1 T47 and T63 Canadian Centre for Climate Modelling and Analysis model
CNRM-CM3 Centre National de Recherches Météorologiques Coupled Global Climate Model, version 3
CSIRO 3.0 and 3.5 Commonwealth Scientific and Industrial Research Organisation Mark version 3.0
GFDL CM2.0 and CM2.1 Geophysical Fluid Dynamics Laboratory Climate Model version 2.0
GISS AOM Goddard Institute for Space Studies Atmosphere–Ocean Model
FGOALS G1.0 Institute of Atmospheric Physics (IAP) Flexible Global Ocean–Atmosphere–Land System Model, gridpoint version 1.0
INGV ECHAM4 Istituto Nazionale di Geofisica e Vulcanologia
INM-CM3 Institute of Numerical Mathematics Coupled Model, version 3.0
IPSL CM4 L’Institut Pierre-Simon Laplace Coupled Model, version 4
MIROC3.2(hires) and (medres) Model for Interdisciplinary Research on Climate 3.2, high-resolution version
MPI ECHAM5 Max Planck Institute
MRI CGCM2.3.2 Meteorological Research Institute Coupled General Circulation Model, version 2.3.2a
CCSM3 National Center for Atmospheric Research (NCAR) Community Climate System Model, version 3
PCM NCAR Parallel Climate Model
HadCM3 Third climate configuration of the Met Office Unified Model
HadGEM1 Hadley Centre Global Environmental Model version 1
See Cohen and Barlow (2005) for a discussion of the sensitivity to the choice of period.
Other differences in addition to the methodology, such as in the underlying dataset and the period used for calculation, may contribute to the differences shown here.