Historical analogues of the recent extreme minima observed in the Atlantic meridional overturning circulation at 26°N
- First Online:
- Cite this article as:
- Blaker, A.T., Hirschi, J.J., McCarthy, G. et al. Clim Dyn (2015) 44: 457. doi:10.1007/s00382-014-2274-6
- 426 Views
Observations of the Atlantic meridional overturning circulation (AMOC) by the RAPID 26°N array show a pronounced minimum in the northward transport over the winter of 2009/10, substantially lower than any observed since the initial deployment in April 2004. It was followed by a second minimum in the winter of 2010/2011. We demonstrate that ocean models forced with observed surface fluxes reproduce the observed minima. Examining output from five ocean model simulations we identify several historical events which exhibit similar characteristics to those observed in the winter of 2009/10, including instances of individual events, and two clear examples of pairs of events which happened in consecutive years, one in 1969/70 and another in 1978/79. In all cases the absolute minimum, associated with a short, sharp reduction in the Ekman component, occurs in winter. AMOC anomalies are coherent between the Equator and 50°N and in some cases propagation attributable to the poleward movement of the anomaly in the wind field is observed. We also observe a low frequency (decadal) mode of variability in the anomalies, associated with the North Atlantic Oscillation (NAO). Where pairs of events have occurred in consecutive years we find that atmospheric conditions during the first winter correspond to a strongly negative Arctic Oscillation (AO) index. Atmospheric conditions during the second winter are indicative of a more regional negative NAO phase, and we suggest that this persistence is linked to re-emergence of sea surface temperature anomalies in the North Atlantic for the events of 1969/70 and 2009/10. The events of 1978/79 do not exhibit re-emergence, indicating that the atmospheric memory for this pair of events originates elsewhere. Observation of AO patterns associated with cold winters over northwest Europe may be indicative for the occurrence of a second extreme winter over northwest Europe.
The Atlantic meridional overturning circulation (AMOC) is part of the global ocean conveyor which transports warm and saline surface waters to the North Atlantic (Broeker 1987; Dickson and Brown 1994; Kuhlbrodt et al. 2007). On their journey towards the Nordic Seas these surface waters gradually become denser as they release heat to the atmosphere. Eventually, the increasing density leads to the sinking of the water masses and they are returned southward as cold and dense North Atlantic deep water. In the subtropical North Atlantic the surface and deep branches of the AMOC result in a maximum net northward heat transport of around 1.3 PW (Ganachaud and Wunsch 2000; Lumpkin and Speer 2007; Johns et al. 2011). The AMOC has been identified as a key ocean mechanism which contributes to the comparatively mild European climate. A large fraction of the heat released to the atmosphere by the AMOC is carried eastward towards Europe by the predominant westerly winds, leading to warmer temperatures in northwestern Europe than at similar latitudes in western Canada (Rhines and Häkkinen 2003; Broeker 1987; Sinha et al. 2012). The potential importance of the AMOC to the climate of the North Atlantic and Europe was a key motivation for the deployment of a transatlantic mooring array at 26.5°N, known as the RAPID-WATCH/MOCHA/WBTS array (hereafter referred to as the 26°N array) (Rayner et al. 2011).
Observation- and model based studies have shown the AMOC transport at 26.5°N to exhibit substantial variability on short timescales (Cunningham et al. 2007; Kanzow et al. 2009, 2010; Hirschi and Marotzke 2007; Baehr et al. 2007, 2009; Balan Sarojini et al. 2011; McCarthy et al. 2012). The observed intra-annual peak-to-peak range of the AMOC, computed using daily means, can be as large as 30 Sv (Cunningham et al. 2007). Between April 2004 and April 2009, the AMOC transport at 26.5°N has a mean of 18.5 Sv, with a standard deviation of 4.7 Sv, when computed using 5 day means. The origin of the observed AMOC variability is only partly understood. Some variability, such as the observed seasonality of the AMOC and interannual variability seen in numerical models, can be linked to the seasonal variability in the wind stress curl along the African coast (Kanzow et al. 2010; Chidichimo et al. 2010; Sinha et al. 2013).
Observations from the 26°N array revealed a recent strong short-term reduction in the strength of the AMOC (McCarthy et al. 2012). The event lasted for 3–4 months over the late winter and early spring of 2009/2010, and resulted in the April–March annual average for that year reducing to 12.8 Sv, 30 % lower than the mean for the previous 5 years. This is the first such event to occur in the 7 year long record from the 26°N array, and it is not known whether events of this kind have occurred in the past. It was followed the year after by a similarly anomalous minimum, raising the question of whether the two events were linked, or whether they occurred consecutively by coincidence. To try and address this question we employ ocean models forced with historical observations of surface fluxes. We examine the characteristics of anomalously low AMOC events in five 1/4° model integrations which all cover the period 1958-2001 and two further model integrations, one at 1/4° and one at 1/12°, which cover the period between 2001 and 2011.
Outline The remainder of this article is organized as follows. Section 2 describes the observation based datasets and the ensemble of model simulations used. Our analysis is presented in Sect. 3. Finally, we summarise our findings in Sect. 4.
2.1 Observation based datasets
For the analysis presented here we use the following observation based datasets.
The AMOC measured by the 26°N array is the sum of the transport measured through the Florida Straits, the geostrophic component derived from measurements of the density difference between the eastern and western boundaries, Ekman transport derived from CCMP/Quikscat winds, and a compensation term which ensures that there is no net transport through the section (Rayner et al. 2011). Ocean models suggest that the sum of these is a close approximation to the true AMOC which one could compute if the northward velocity at all points through the section were known (Hirschi and Marotzke 2007; Hirschi et al. 2003). Data from the 26°N array (http://www.rapid.ac.uk) are provided at 12 hourly resolution with a 10 day low pass filter applied. Observations of all component time series are available from April 2004 to present. For our analysis we compute 5 day averages to facilitate comparison with the model simulations, for which output is stored as 5 day averages.
To explore the geographical nature of the events we use monthly mean sea level pressure (SLP) and surface air temperature (SAT) data from the NCEP/NCAR reanalysis dataset (Kalnay et al. 1996). We relate the events identified to the monthly indices of the North Atlantic and Arctic Oscillations following Hurrell (1995) (NCAR 2012a, b). To examine the mechanism of re-emergence we make use of SST data from the NOAA optimal interpolation dataset (Reynolds et al. 2002) http://www.esrl.noaa.gov/psd/ and also of the ENACT (EN3) reanalysis from the UK Met Office (Ingleby and Huddleston 2007) http://www.metoffice.gov.uk/hadobs/en3/.
2.2 Model description and runs
This table shows the model configurations used
Five simulations start from 1958, and form the ensemble which we will analyse in this paper. Aspects of the AMOC in some of these runs have been published previously (Grist et al. 2010, 2012; Blaker et al. 2012; Hirschi et al. 2013; Duchez et al. 2014b). All of these simulations are on the ORCA025 grid, for which horizontal resolution is nominally 1/4°. South of 20°N the model grid is isotropic Mercator, and north of 20°N the grid becomes quasi-isotropic bipolar, with poles located in Canada and Siberia to avoid numerical instability associated with convergence of the meridians at the geographic North Pole. At the Equator the resolution is approximately 27.75 km, becoming finer at higher latitudes such that at 60°N/S it becomes 13.8 km. The models in the ensemble include both 64 and 75 vertical levels with a grid spacing increasing from 6 m near the surface (1 m for the 75 level runs) to 250 m at 5,500 m. Bottom topography is represented as partial steps and bathymetry is derived from ETOPO2 (U.S. Department of Commerce 2006). To prevent excessive drifts in global salinity due to deficiencies in the fresh water forcing, sea surface salinity is relaxed towards climatology with a piston velocity of 33.33 mm/day/psu. Sea ice is represented by the Louvain-la-Neuve Ice Model version 2 (LIM2) sea-ice model (Timmerman et al. 2005), and version 3 of the same for one of the runs.
Climatological initial conditions for temperature and salinity were taken in January from PHC2.1 (Steele et al. 2001) at high latitudes, MEDATLAS (Jourdan et al. 1998) in the Mediterranean, and Levitus et al. (1998) elsewhere. These initial conditions were applied to all runs with the exception of N112, which started from the final state of N102, and VN206, which started from the final state of N206. Starting from rest the models simulate the period indicated in the right hand column of Table 1, with surface forcing comprising of 6-hourly mean momentum fields, daily mean radiation fields and monthly mean precipitation fields supplied by the DFS3, DFS4.1 and CORE2 datasets (Brodeau et al. 2010; Large and Yeager 2004, 2008) and linearly interpolated from the time mean fields by the model. Model output is stored as 5-day averages, although for disk storage considerations, we have only retained monthly mean values for some of the older datasets.
One 1/4° run, VN206, is forced by the ERA Interim dataset (Dee et al. 2011) and extends to March 2011. We also examine the output of a recently completed ORCA12 (1/12°) run, which again uses the NEMO ORCA code, but is now eddy resolving for much of the global ocean.
The ensemble of model integrations we have compiled here are all closely related. They are runs of the same model with minor/moderate variations in the code version, applied surface forcing, ice model used and intial conditions, and all reproduce plausible ocean states. For each of the model runs we compute the AMOC and component parts equivalent to those which are measured and used to construct the time series at 26.5°N. Florida Straits transport in the model integrations is computed as the integral of the meridional velocity through the Florida Straits. The upper mid ocean (UMO) component is computed by integrating the east-west density difference derived geostrophic transport across the basin. Ekman transport is computed using the surface wind stress. A uniform compensation velocity which ensures no net volume transport through the section is applied. Details of the component decomposition method are provided in the Appendix.
3 AMOC analysis
3.1 Comparison with observations
The extreme events which occur during the winters of 09/10 and 10/11 both reach their peak in late December. Taking a 30 day average of the observations and subtracting from the time mean observational values, the anomalous Ekman component contributes 52 % of the first AMOC minimum and around 80 % of the second. Both simulations (ORCA12 and VN206) agree well with the observations for the first event (55 and 51 % respectively). For the second event the Ekman component contributes 50 % (ORCA12) and 46 % (VN206) for the two simulations. An observed negative anomaly in the UMO transport, which began early in 2009 and was associated with a partial shift of the circulation from the deep overturning to the upper gyre circulation, was found to contribute significantly to the minimum event of 2009/10 (McCarthy et al. 2012). To establish how well the UMO anomaly is represented by the simulations we compute the accumulated UMO transport anomaly in the manner of Bryden et al. (2014) (Fig. 1c). The accumulated transport anomaly is the cumulative summation of UMO anomalies from April 2004 (the start of the observational period) to the end of the simulations/observations. Presenting the anomalies in this way emphasises anomalies of a consistent sign, whilst reducing the visual effect of differences in the timing of short term variability between the simulations and observations. It indicates that the models also capture around two thirds of the anomalous transport associated with the UMO.
This table shows the mean and standard deviation (brackets) for each time series shown in Fig. 1
3.2 Historical analogues
We construct a composite of the AMOC anomalies (\(\varPsi '\)) from the five eddy permitting hindcasts that span the time period 1958-2001 by removing the linear trend from each ensemble member and then removing the seasonal cycle. The AMOC anomalies are then averaged to produce the ensemble mean (Fig. 2). This composite reveals several strongly negative events, some substantially in excess of 2 standard deviations from the mean. There is a strong negative event in 1962/63, one in 1980/81 and another in 1983/84. There is also a minimum in 1986/87 with a duration of 3–4 months. We can also identify two pairs of events, one pair in 1968/69 and 1969/70, and another pair in 1977/78 and 1978/79. One further example in the time series which may also be a weaker analogue of the 2009/10 event occurs during 1996/97 and 1997/98. The event in the winter of 1996/97 occurs slightly later (around March) than other events in the time series (typically January-February). It coincides with an anomalously strong northward Ekman transport anomaly in February, which we suggest is likely to have reduced the impact and altered the timing of the negative event.
From the observed events we know that a large fraction of the minimum arises from the Ekman component [Fig. 1, and also McCarthy et al. (2012)], and the high correlation between the Ekman component and the AMOC (0.87) can be seen in Fig. 2a, d. It is worth noting that the ensemble spread for the Ekman component is very small, which is a reflection of the similarities among the forcing datasets and how strongly the Ekman component is controlled by the surface forcing. Whilst variability of the Ekman component is the largest single contributor, it does not explain all of the variability in the AMOC. The value of two standard deviations in the AMOC is 4 Sv. Two standard deviations of the Ekman component alone is 2.7 Sv, whilst the variability associated with the other components combined is slightly higher than this at 2.95 Sv. However, we note that for months which exhibit a negative AMOC anomaly of greater than 2 standard deviations, we find that on average the Ekman component contributes 62 % of the anomaly. Some interaction between the FS transport and the geostrophic transport can also be seen in Fig. 2b, c, which have a correlation of −0.55. This is partly attributable to the inclusion of the FS compensation in the UMO transport, and partly to variation in the path of the Gulf Stream, with some fraction of the transport occasionally passing to the east of the Bahamas instead of through the FS. This may be an artefact of the model resolution. The ensemble spread is large for the FS transport and the geostrophic transport, and this reflects the differences in timings of short timescale chaotic events (e.g. mesoscale ocean eddies, Gulf Stream meanders), which account for about 30 % of the total AMOC variability (Hirschi et al. 2013).
The distribution of extreme AMOC events is asymmetric. There are only two positive anomalies stronger than 6 Sv, one in 1984 and one in 1994, both of which coincide with strong positive anomalies in the Ekman transport. In contrast there are 6 negative events which exceed 6 Sv. However, the time series is too short and there are insufficient events to say whether this is statistically significant.
3.3 Latitudinal characteristics
3.4 Atmospheric and SST conditions
Examining first the most recent events, the winters of 2009/10 and 2010/11 both exhibited a strongly negative NAO index, with December 2010 recording the second lowest NAO index (−4.62) since records began in 1825 (Osborn 2011). Spatial plots of the winter mean SLP for the winters of 2009/10 and 2010/11 (Fig. 7e, f) show that these winters, that of 2009/10 in particular, bear the characteristics of a negative Arctic oscillation (AO) pattern with a pronounced anomaly in the N. Atlantic sector and over Russia. The AO is another pattern of atmospheric variability closely related to the NAO (Fig. 6b), and is derived using the first principle component of winter (DJFM) SLP anomalies poleward of 20°N (NCAR 2012b). However, whilst the two indices are closely related, strongly negative AO years do not necessarily coincide with a strongly negative NAO. The NAO is considered by some to be a regional expression of the AO (see e.g. Thompson and Wallace 2000), and Wallace (2000) suggested that they can be considered manifestations of the same basic phenomenon. February 2010 recorded the lowest AO index (−4.266) since reliable records began in 1950 (L’Heureux et al. 2010; NOAA 2013). Similar spatial plots for the winters of the other pairs of events in 1968/69−1969/70 and 1977/78−1978/79 reveal that the winters of the first year of the pair all exhibit similar strong negative AO conditions. The anomaly patterns, particularly over the North Atlantic sector, are seen in each case to persist in the following winter.
Examining SAT anomalies for 1968/69–1969/70, 1977/78–1978/79 and 2009/10–2010/11 reveals a widespread cool anomaly across much of Siberia for the winters of the pairs of events (Fig. 9). For the years associated with single events cool anomalies are weaker and located over Europe (Fig. 10). The temperature anomalies associated with the pairs of events also show the tendency to persist for the following winter. This is most pronounced for the winters of 2009/10 and 2010/11 where the cold anomalies over Eurasia and North America occur mainly over the same regions. For the other two pairs of events the largest anomalies in the 2 m air temperature vary in location from one year to the next. For the 1968/69–1969/70 pair the first event coincided with exceptionally low temperatures over Siberia (e.g. Hirschi and Sinha 2007), whilst anomalies were small over Europe. During the second event in contrast the coldest anomalies occurred over Northwestern Europe. A large variability in the temperature distribution in different NAO negative winters is consistent with earlier studies (e.g. Heape et al. 2013).
The atmospheric conditions during the winters of 1968/69 and 1977/78 were similar to those experienced in the winter of 2009/10, with the NAO and AO indices in an extreme negative state (Fig. 6). Taws (2013) examined SST fields from EN3 for evidence of previous re-emergence signatures. Figure 12a shows lagged pattern correlations referenced to March (March through to April of the following year) of SST anomalies for the North Atlantic computed from EN3 (Ingleby and Huddleston 2007). It shows that both March 2010 and the following winter and March 1969 and the following winter exhibited higher levels of correlation (between 1 and 2 standard deviations of the 1960–2011 period) between subsequent winters, and the similarity of the timing and strength of the correlation suggests that both periods experienced re-emergence events. March of 1978 and the following winter show no significant increased correlation, suggesting that re-emergence did not occur during this period. None of the other years show a strong indication of re-emergence. Figure 12b is equivalent to 12a, but instead shows a composite of the lagged pattern correlations computed from SST anomalies in the ensemble of ORCA025 models. The models also show significantly higher correlations between the winter months of 1968 and 1969, in this case between 2 and 3 standard deviations, indicating that there was a re-emergence event. The model ensemble also found no significant correlation between March 1978 and the following winter, indicating that this period did not experience a re-emergence event.
To further examine the strong indicators of reemergence for 1968/69 we plot surface and subsurface temperature anomalies for the region 5–65°N, 80–10°W. An SST anomaly pattern similar in amplitude and distribution to the one presented in Fig. 11 is present in March 1969 (Fig. 13a). After persistent AO negative atmospheric conditions over the winter months this anomaly is coherent down to the base of the winter mixed layer (Fig. 13d). The pattern continues to persist into September at the winter mean mixed layer depth, below the shallow summer mixed layer (Fig. 13e), and through to January of 1970 (Fig. 13f), where the seasonal mixing returns the anomaly to the surface (Fig. 13c). The SST anomaly in September of 1969 (Fig. 13 b) is uncorrelated with the SST anomaly present in March 1969 (see Fig. 12b).
The mechanism of re-emergence, whereby temperature anomalies formed in the deep winter mixed layer are trapped subsurface by the shallow summer mixed layer and then re-entrained into the mixed layer during the onset of the following winter occurs annually. However, in order to provide a memory to the atmosphere a strong and coherent SST anomaly pattern must form during the first winter. A strong negative AO or NAO state which is persistent throughout the winter would facilitate this. There are a number of individual AMOC minima events which occur in the composite time series (Fig. 2), some of which are associated with negative NAO states, some of which are not. A possible explanation for why re-emergence events only occurred during strong negative AO states is that these events are consistently stable over the North Atlantic sector throughout winter (December to March). To examine this we construct a time series of the maximum monthly NAO index occurring during the winter months (DJFM). The mean and standard deviation of this time series are 1.7 and 1.22 respectively. For the 1968/69 and 2009/10 winters the maximum monthly mean NAO index reached during winter was −1.4 and −1.5 respectively, around 2.5 standard deviations lower than the mean winter maximum. The maximum NAO index during winter of 1977/78 was also anomalously low at −0.5, 1.75 standard deviations below the mean. The winter maximum NAO index for the three individual events examined (1962/63, 1980/81 and 1987/88) were all within 0.5 standard deviations of the mean.
Interestingly the responses in the AMOC are stronger for 1978/79 than for 1969/70, indicating that processes other than re-emergence might be important for providing memory to the atmosphere from one winter to the next, resulting in the wind stress anomalies which provide the Ekman anomaly in two consecutive winters. Of course it is also possible that two consecutive events occur by chance, and that no processes for providing memory from one winter to the next are involved.
4 Summary and discussion
In light of the anomalous minima events recorded in the AMOC for the winters of 2009/10 and 2010/11 by the 26°N array (McCarthy et al. 2012) we employ an ensemble comprising five 1/4° ocean model realisations to investigate how frequently these events have occurred in the past, and in particular whether there are mechanisms which might give rise to multiple events occurring in consecutive winters.
We first make comparisons between two model integrations (one 1/4° and one 1/12°) which cover the period for which we have observations and determine that the NEMO ORCA ocean model is able to adequately reproduce the interannual variability of the AMOC captured by the mooring array, as well as the timing and amplitude of the anomalously strong minima which occur during the winters of 2009/10 and 2010/11. The model is not able to capture all of the variability seen in the observations, and we do not expect it to given that around 30 % of the variability is chaotic (Hirschi et al. 2013). The modelled amplitudes of the 2009/10 and 2010/11 events are slightly lower than observed, but so are both the mean and standard deviation of the AMOC represented in the model.
Examining the ensemble of NEMO runs which extend back to 1958 we have identified a number of events for which a strong reduction of the AMOC transport occurred. In some cases (e.g. 1962) these are individual events, but we also identify two pairs of events which occur in consecutive years during 1969/70 and 1978/79, which are historical analogues of the recently observed events. A possible third weaker analogue occurred in 1997/98.
We compare boreal winter atmospheric conditions for the years during which extreme negative anomalies in the AMOC were observed. The AMOC minimum event which occurred in the winter of 1962 coincides with an extended period of negative NAO, but no significant AO index. Other examples of individual events such as in 1981 and 1990 do not coincide with negative NAO states. In fact 1990 corresponds to one of the strongest positive AO and NAO indices for the period we examine. The pairs of events identified in 1969/70, 1978/79 and 2010/11 are all associated with strong negative AO indices, with particularly strong SLP patterns over the North Atlantic sector, and corresponding with strongly negative NAO indices. The event in the winter of 2009/10 coincides with one of the strongest negative values of the NAO index since records began in 1860, and there is evidence that the re-emergence mechanism contributed to persistence of the SST anomalies (Taws et al. 2011), and to the renewed development of negative NAO conditions (Buchan et al. 2014). We show here that the events of 1968/69 are also connected to an occurrence of re-emergence, but the equally strong, if not stronger, AMOC minima seen in 1978/79 are not linked to a re-emergence of SST anomalies. SAT anomalies for all 3 events in 1969/70, 1978/79 and 2010/11 show a strong negative anomaly which covers much of Siberia, and other studies suggest that mechanisms related to sea ice cover over the Arctic and North Atlantic (e.g. Deser et al. (2007)) or to snow cover and thickness over Eurasia (Peings et al. 2012; Fletcher et al. 2009; Gong et al. 2004, 2003) may provide memory to the atmosphere which allows the negative NAO conditions over the North Atlantic sector to persist.
The minimum of the events frequently coincides with anomalously strong southward Ekman transport, which constitutes around half of the anomaly. We find that the simulations also capture a substantial amount of the low frequency variability of the UMO transport, indicating that this signal is also surface forced. It may be Sverdrup transport (Duchez et al. 2014b) or related to other mechanisms such as a lagged ocean response to the surface forcing such as the one described by Sinha et al. (2013), whereby changes in windstress influence ocean transport through adjustments of the vertical velocity and vortex stretching. Some of the model runs respond strongly for a given event whilst others show weaker responses, indicating that the events are not purely a response to the surface forcing but that there is some dependence on the ocean state or that ocean physics may affect the strength or timing of the non-Ekman component of the anomalies. For example, a study which compared two of the simulations used here (N102 and N112) found that up to 30 % of the total variability of the AMOC is attributable to chaotic processes such as mesoscale eddies and internal and planetary waves (Hirschi et al. 2013). Bryden et al. (2014) examine the 2009/10 event and conclude that some of the observed reduction is due to modes of ocean variability which are not associated with recent atmospheric forcing. Understanding these modes of variability could lead to improvements in the model representation of such events.
An interesting implication of the association of the extreme negative events with negative AO patterns and the re-emergence of SST anomalies is that it may be possible to predict the onset of negative NAO conditions and a second consecutive AMOC minimum (Maidens et al. 2013). There may also be scope for improved prediction of the AMOC for a third consecutive winter, since the anomaly composite reveals that there are no cases since 1958 where we find three or more consecutive extreme negative events.
This work was supported by the NERC funded RAPID-WATCH project VALOR (NE/G007772/1) and was also part of the DRAKKAR project. Data from the RAPID-WATCH MOC monitoring project are funded by the Natural Environment Research Council and are freely available from www.noc.soton.ac.uk/rapidmoc. Sarah Taws was funded by a NERC Quota Studentship, with added support from the UK Met Office. NCEP Reanalysis Derived data were provided by the NOAA/OAR/ESRL PSD, Boulder, Colorado, USA, from their Web site at http://www.esrl.noaa.gov/psd/. NAO Index Data was provided by the Climate Analysis Section, NCAR, Boulder, USA. We thank two anonymous reviewers for their useful and constructive comments.