On the role of the ocean in projected atmospheric stability changes in the Atlantic polar low region



[1] The occurrence of destructive mesoscale ‘polar low’ cyclones in the subpolar North Atlantic is projected to decline under anthropogenic change, due to an increase in atmospheric static stability. This letter reports on the role of changes in ocean circulation in shaping the atmospheric stability. In particular, the Atlantic Meridional Overturning Circulation (AMOC) is projected to weaken in response to anthropogenic forcing, leading to a local minimum in warming in this region. The reduced warming is restricted to the lower troposphere, hence contributing to the increase in static stability. Linear correlation analysis of the CMIP3 climate model ensemble suggests that around half of the model uncertainty in the projected stability response arises from the varied response of the AMOC between models.

1. Introduction

[2] Polar lows are intense mesoscale cyclones which constitute a hazard to shipping and coastal areas in the northern North Atlantic [Rasmussen and Turner, 2003]. They are particularly prevalent in the Norwegian Sea and in the region between Greenland and Iceland [Harold et al., 1999; Kolstad, 2006; Zahn and von Storch, 2008]. It would be advantageous to determine how the occurrence and characteristics of these systems will respond to the changes arising from anthropogenic forcing. Zahn and von Storch [2010, hereinafter Z10] recently provided the first study of these changes, finding a significant decrease in the occurrence of polar lows using dynamically downscaled global CMIP3 climate model projections. Basin-averaged storm frequencies roughly halved by the end of the 21st century, with particularly large changes in the Greenland-Iceland region. This decrease was ascribed to an increase in atmospheric static stability which was shared by all of the global models considered. Marine cold-air outbreaks are often related to polar lows, andKolstad and Bracegirdle [2008]reached similar conclusions concerning the response of cold-air outbreaks to anthropogenic forcing. Given the key role of the stability changes it is important to understand the driving mechanisms underlying them and also the reasons for spread between different models. This will improve our understanding of the trustworthiness of the projected decline in polar lows and also indicate how uncertainty in the response may be reduced.

[3] The aim of this paper is to investigate whether changes in Atlantic Ocean circulation are important in determining the changes in the atmospheric static stability in the northern North Atlantic. The Atlantic Meridional Overturning Circulation (AMOC) transports heat into this region but is projected to weaken in the future under anthropogenic forcing [Meehl et al., 2007]. This weakening leads to a local minimum in the anthropogenic warming pattern in the North Atlantic with implications for the atmospheric storm track and large scale circulation [Woollings et al., 2012]. In this paper we show that it also has a strong influence on the atmospheric static stability. The stability may play a role in influencing the storm track but is of particular importance for the occurrence of polar lows which is more strongly controlled by stability [Bracegirdle and Gray, 2008]. Other factors such as large-scale atmospheric flow changes could influence the occurrence of polar lows [Zahn and von Storch, 2008; Blechschmidt et al., 2009; Noer et al., 2011], but these changes are less robust between models than the stability changes [Woollings, 2010].

2. A Hosing Experiment

[4] We begin by presenting some results from an idealised experiment in which the AMOC was artificially shut down in a climate model. This is the HadCM3 hosing experiment of Vellinga and Wu [2008], which was also used to investigate the atmospheric response to an AMOC shutdown by Brayshaw et al. [2009] and Woollings et al. [2012]. In the hosing simulation a strong and continuous freshwater perturbation was added to the North Atlantic to obtain an equilibrium collapsed state of the AMOC, and this can be compared to a control simulation.

[5] Figure 1 summarises the wintertime temperature response to hosing over the North Atlantic. The response in surface temperature is shown in Figure 1a and is similar to that shown by Vellinga and Wood [2002]. The cooling is particularly strong over the ocean surface as discussed by Laurian et al. [2010]. In contrast, the cooling in the mid-troposphere is much weaker (500 hPa;Figure 1b). The stronger cooling at lower levels indicates an increase in static stability and this is confirmed by a vertical section of the potential temperature in Figure 1c. This shows that the potential temperature reduction decays rapidly away from the surface. The static stability, as measured by the Brunt-Vaisala or buoyancy frequency (N2) is increased considerably. This increase is particularly strong in the boundary layer, where the change is greater than 50% for most of the region, but considerable changes extend through the depth of the troposphere. Very large changes occur in the boundary layer at high latitudes where the sea-ice extent increases.

Figure 1.

DJF mean differences between 20 years of the control and hosing HadCM3 simulations of (a) surface temperature (which equals SST over open ocean), (b) temperature at 500 hPa and (c) potential temperature along the section from (55N, 50W) to (75N, 20E), which is marked by a dashed line in a,b). Also shown in Figure 1c is the percentage increase in the Brunt-Vaisala frequencyN2 in red (with heavy contours every 100%).

[6] A large decrease in the strength of the AMOC can therefore be expected to have an influence on static stability and so on polar low behaviour. This vertical structure may also have implications for the impact on surface temperatures. Deser et al. [2004] described a nonlinearity in the atmospheric response to surface forcing in this region, whereby a cooling of the surface leads to a shallower penetration of anomalous heating than seen when the surface is warmed, due to the increase in stability. In this way the anomalous cooling due to an AMOC change is inhibited from mixing through the depth of the troposphere leading to large temperature changes at the surface.

3. CMIP3 Analysis

[7] We now present evidence that changes in the AMOC do indeed have an influence on the stability in climate models under anthropogenic forcing. For this purpose we analyse the ensemble of climate model simulations performed for the third Coupled Model Intercomparison Project (CMIP3). Model output from 13 coupled atmosphere-ocean general circulation models (AOGCMs) has been used, for which diagnostics of AMOC and atmospheric temperature changes were available. These are the same models as used inWoollings et al. [2012], and they are described in the Fourth Assessment Report of the Intergovernmental Panel on Climate Change [Randall et al., 2007]. The forcing scenarios and periods 20C3M (1961–2000) and SRESA1B (2061–2100) are used to characterise the end of the 20th and 21st centuries respectively. We focus on the extended winter season of October–March which coincides with the polar low season. The AMOC is diagnosed from the models as the maximum value of the annual mean meridional streamfunction at 45N in the Atlantic Ocean.

[8] To show the horizontal structure of the change in stability we follow Z10 and use the difference in temperature between the sea surface and the 500 hPa level as a simple measure of stability. This measure is routinely used by Norwegian forecasters to identify unstable conditions favourable for polar low development [Noer and Ovhed, 2003]. We define S = T500SST, so that an increase in S corresponds to an increase in stability (note the sign, so that S is the negative of vdT as used in Z10). The ensemble mean response in S is given in Figure 2a, which shows that stability increases across the basin but particularly strongly in the polar low genesis region to the south of Greenland, where ocean warming is moderated by the weakening of the AMOC. The structure of the stability response is very strongly influenced by the structure of the SST response, which is shown in black contours.

Figure 2.

(a) The CMIP3 multi-model mean response (2061–2100 minus 1961–2000) in stabilityS = T500SST. The SST response is shown in black contours. (b) Vertical section along the line from (45N, 50W) to (75N, 20E) of the potential temperature and static stability as in Figure 1cbut for the multi-model mean response. Green contours show the percentage change in the Brunt-Vaisala frequencyN2. (c and d) The same variables regressed on the AMOC response across CMIP3 model space (shown per Sverdrup weakening of the AMOC). Only gridpoints where 50% of the models are ice-free in the 20th century period are used in Figures 2a and 2c. The season used is October–March and the models used are listed inFigure 3. The colourbars to the right apply to both Figures 2a and 2b and Figures 2c and 2d, respectively.

[9] Figure 2bshows a vertical section of the multi-model ensemble mean response to forcing along a similar section to that inFigure 1c (chosen to pass through the region of strongest change). This confirms that the reduced warming south of Greenland is restricted to the lower troposphere, giving an increase in N2. In contrast, N2 decreases in the Norwegian Sea in the ensemble mean, due to the polar amplification of warming at low levels in the Arctic. This differs from the response in S, which is defined using SST rather than air temperature at low levels, and highlights the complexity of this region, where sea ice changes are very important. While a weaker AMOC can lead to increased ice extent, as in the hosing experiment, other processes such as the magnitude of polar amplification and the control sea ice extent are also important [Hodson et al., 2012]. In addition, the change in surface characteristics as sea ice retreats may be very important for the dynamics of polar lows themselves. To avoid these issues we focus in Figures 2a, 2c, and 3on gridpoints which are ice-free in the 20th century period, although this means that we may underestimate the importance of the AMOC in the Norwegian Sea. It is worth noting, however, that the regional model experiments ofZ10 also did not show a qualitatively different anthropogenic change in polar low occurrence in the region of ice retreat compared to the change over the open ocean.

Figure 3.

Scatterplots summarising the relationships between the temperature/stability responses and the changes in (a–c) AMOC and (d–e) global mean surface temperature. All ice-free gridpoints in the whole North Atlantic poleward of 45N were used. Each letter represents one of the CMIP3 models.r2 values give the fraction of variance explained by the best fit lines shown, with the red line in Figure 3c neglecting the outlier model g (which has an unusually large control AMOC of 26 Sv).

[10] Woollings et al. [2012] used a regression approach to show that the different surface temperature responses to forcing in this region can be explained by the spread in AMOC responses in the different models, and we now take a similar approach here. Figure 2c shows the regression across model space of the stability responses ΔS onto the AMOC response. This shows that a stronger AMOC weakening results in an increased stability, which is particularly strong to the south of Greenland but is also apparent in the Norwegian Sea. As before the spatial structure of the stability response is very similar to that of the SST response shown in black contours. The potential temperature and N2 regressions are shown in Figure 2d and these show a similar pattern to that of the more dramatic change in the hosing experiment (Figure 1c). In contrast to the ensemble mean response, both S and N2 are in agreement in this context, that a stronger AMOC decline is associated with increased stability in both the Greenland and Norwegian Sea polar low regions.

[11] A key question is to what extent the model spread in stability changes is influenced by the AMOC changes. Figure 3 shows scatterplots of the models' change in SST, T500 and S against two possible predictors: the change in the AMOC (Figures 3a–3c) and the change in global mean surface temperature (Figures 3d–3f). Firstly, we see that the changes in temperature at both levels, but particularly at 500 hPa, are strongly associated with the global mean warming (Figures 3d and 3e). However, the difference between the two levels (given by S) is independent of the mean warming (Figure 3f). In contrast, the AMOC change is more closely related to the surface temperature change (Figure 3a) than the change at 500 hPa (Figure 3b), and the stability S is highly correlated with the AMOC change (Figure 3c). This shows that while the global mean warming has a strong correlation with temperature at both levels, the vertical structure, and hence the stability, is strongly influenced by the change in the AMOC. In fact, the r2 values indicate that the spread in AMOC response explains 46 or 64% of the spread in stability change depending on whether the outlier model g is included. The slope of the red regression line gives a sensitivity of S of 0.2 K Sv−1. In the regional model experiments of Z10 this equates to a reduction of approximately 2 polar lows per season for every Sv of the AMOC weakening.

[12] This analysis has been repeated for both the North Atlantic and Iceland/Greenland sub-regions used byZ10(not shown). The result is that most of the signal in the stability-AMOC relation comes from the region between Iceland and Greenland (55–70N, 15–50W), withr2 = 0.64 for this region (excluding the outlier). The relation is much weaker in the North Atlantic region (65–80N, 20W–20E), with r2= 0.1, although as described above this may be underestimated by focusing only on ice-free regions.

[13] The ensemble mean AMOC reduction clearly plays a role in the ensemble mean increase in stability. However, the regression lines in Figure 3c suggest a significant increase in stability if the AMOC does not change, so other effects must also contribute to this.

4. Discussion and Conclusions

[14] The main conclusion of this letter is that the weakening of the AMOC in response to anthropogenic forcing has a strong influence on the increase in lower tropospheric stability in the northern North Atlantic in the CMIP3 simulations. While the warming at both the surface and mid-tropospheric levels is strongly related to the global mean warming, the temperature difference between the two levels is instead more closely related to the AMOC change. In addition, the structure of the change in stability strongly resembles the structure in the SST response. The weakening of the AMOC plays a large role in the ensemble mean stability increase, and around half of the model spread in the stability response can be explained by the varied AMOC response in the models.

[15] Comparison with the forced hosing experiment provides support for the interpretation that the ocean has a causal influence on the atmosphere. Further evidence of causality is given by Woollings et al. [2012], who show that the North Atlantic minimum in warming is not present in the CMIP3 slab model simulations which do not include changes in ocean circulation. An implication of this result is that, while the projected decline in polar low occurrence is robust across the ensemble, confidence in this response depends on our confidence in the AMOC response in these relatively coarse resolution ocean models.

[16] One caveat to this study is that we have not considered changes in the vicinity of the ice edge. The change in surface characteristics is likely very important in the region which is ice covered in the 20th century but ice free in the 21st [Deser et al., 2010; Medeiros et al., 2011]. Finally, we note that the link between AMOC and polar low changes is particularly simple in the context studied here, in which polar lows themselves are not resolved in the global models. In reality, changes in polar low occurrence could in turn have an influence on deep water formation and hence on the AMOC itself [Condron et al., 2008].


[17] 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 multi-model dataset. Support of this dataset is provided by the Office of Science, U.S. Department of Energy. BH was funded by the NERC TEMPEST project (NE/I00520X/1) and MZ by the NERC PREPARE project (NE/G015708/1). We would also like to thank Michael Vellinga for providing data from the HadCM3 hosing simulations, Jonathan Gregory for providing the CMIP3 AMOC data and two anonymous reviewers for their helpful comments.

[18] The Editor thanks two anonymous reviewers for their assistance in evaluating this paper.