全球变化知识资源中心

Possible changes in Atlantic meridional overturning circulation (AMOC) provide a key source of uncertainty regarding future climate change. Maps of temperature trends over the twentieth century show a conspicuous region of cooling in the northern Atlantic. Here we present multiple lines of evidence suggesting that this cooling may be due to a reduction in the AMOC over the twentieth century and particularly after 1970. Since 1990 the AMOC seems to have partly recovered. This time evolution is consistently suggested by an AMOC index based on sea surface temperatures, by the hemispheric temperature difference, by coral-based proxies and by oceanic measurements. We discuss a possible contribution of the melting of the Greenland Ice Sheet to the slowdown. Using a multi-proxy temperature reconstruction for the AMOC index suggests that the AMOC weakness after 1975 is an unprecedented event in the past millennium (p > 0.99). Further melting of Greenland in the coming decades could contribute to further weakening of the AMOC.

A persistent subpolar North Atlantic cooling anomaly is a conspicuous feature of the overall global warming pattern (Fig. 1). Model simulations indicate the largest cooling response to a weakening of the AMOC in this same region¹, suggesting this area has so far defied global warming owing to a weakening of the AMOC over the past century. The time history of the AMOC over this period is poorly known, however, owing to the scarcity of direct measurements. Because of the large heat transport associated with the AMOC, changes in sea surface temperatures (SSTs) can be used as an indirect indicator of the AMOC evolution².

Dima and Lohmann³ identified two distinct modes in global SST evolution, one associated with a gradual decline of the global thermohaline circulation and one due to multidecadal and shorter AMOC variability, and concluded ‘that the global conveyor has been weakening since the late 1930s and that the North Atlantic overturning cell suffered an abrupt shift around 1970’. Thompson et al.⁴ found that the SST difference between the Northern and Southern Hemisphere underwent a sudden decline by ~0.5 °C around 1970, with the largest cooling observed over the northern Atlantic. We interpret this as indicative of a large-scale AMOC reduction, as the most plausible explanation for such a rapid change in the interhemispheric temperature difference is the cross-equatorial heat transport of the AMOC (ref. 5). Drijfhout et al.⁶ regressed the AMOC strength and global-mean temperature on surface temperature fields in models and concluded that the conspicuous ‘warming hole’ south of Greenland is related to a weakening of the AMOC. They further found that a possible contribution of aerosol forcing to the cool patch as proposed in ref. 7 cannot be excluded.

Zhang et al.⁸, however, argue that the model simulation in ref. 7 overestimates the effect of aerosol forcing, by not accounting for any increase in ocean heat content in the North Atlantic over the second half of the twentieth century, in contrast to what is suggested by the observations. The observational data show a clear dipole response in the Atlantic, with the North Atlantic cooling and the South Atlantic warming when comparing 1961–1980 with 1941–1960. The maximum of South Atlantic warming is within the Benguela Current off southern Africa and the maximum of North Atlantic cooling is found within the Gulf Stream. These patterns are highly characteristic of AMOC changes and are found in many model simulations wherein the AMOC is weakened by freshwater ‘hosing experiments’. The Atlantic see-saw pattern is also evident in Fig. 1, where out of all Southern Hemisphere ocean regions the South Atlantic has warmed the most.

Terray⁹ analysed the CMIP5 model ensemble together with observed SST data to quantify the relative contributions of radiatively forced changes to the total decadal SST variability. Although in most models forced changes explain more than half of the variance in low latitudes, they explain less than 10% in the subpolar North Atlantic, where in most cases their contribution is not significantly different from zero (the notable exception is the model used by Booth et al. as mentioned above).

To put the twentieth-century AMOC evolution into a longer-term context, in the following we develop an AMOC index based on surface temperatures from instrumental and proxy data.

We take the results of a climate model intercomparison¹ to identify the geographic region that is most sensitive to a reduction in the AMOC (Fig. 1), which for simplicity we henceforth refer to as ‘subpolar gyre’, although we use the term here merely to describe a geographic region and not an ocean circulation feature. To isolate the effect of AMOC changes from other climate change, we define an AMOC index by subtracting the Northern Hemisphere mean surface temperature from that of the subpolar gyre (see Supplementary Information for an alternative index obtained by subtracting Northern Hemisphere SST). We thus assume that differences in surface temperature evolution between the subpolar gyre and the whole Northern Hemisphere are largely due to changes in the AMOC. This seems to be a reasonable approximation in view of the evidence on North Atlantic SST variability discussed in the introduction. We decided against using an index based on a dipole between North and South Atlantic temperatures^{2, 10}, as this might be affected by the large gradient in aerosol forcing between both hemispheres.

We test the performance of the index in a global warming scenario experiment for 1850–2100 with a state-of-the-art global climate model, the MPI-ESM-MR. This model has a realistic representation of the AMOC (refs 10, 11) based on criteria that include the magnitude and shape of the AMOC stream function and the realism of sites of deep-water formation. Without satisfying those criteria, we cannot expect realistic spatial patterns of SST response to AMOC variations and hence a good correlation of our temperature-based AMOC index with the actual AMOC. An analysis of ten global climate models found that a surface temperature response in the North Atlantic subpolar gyre is a robust feature of AMOC variability, although the details of this response depend on the quality of representation of the AMOC (ref. 10).

Figure 2 illustrates the high correlation of the AMOC index with the actual AMOC in the model, particularly on timescales of a decade and longer (smoothed curves). The correlation coefficient of the two smoothed curves after linear detrending is R = 0.90 and our temperature-based AMOC index predicts the actual AMOC changes in the model with an RMS error of 0.6 sverdrups (Sv; 1.1 Sv for the annual data), where the conversion factor of 2.3 Sv K⁻¹ has been fitted. Note that both individual components of the index—the subpolar gyre and the Northern Hemisphere surface temperature—increase during the twenty-first century in the simulation; it is the difference between the two which tracks the AMOC decline, as expected by our physical understanding of the effect of AMOC heat transport.

Simulated with the MPI-ESM-MR global climate model of the Max Planck Institute in Hamburg¹¹. a, Time series of the maximum overturning stream function (red) and the AMOC index (blue). Thin lines show annual values, thick lines smoothed curves over 11 years. b, Correlation coefficient r of the overturning stream function in the model with the AMOC index (shading), shown together with the mean stream function (grey contours in 5 Sv intervals).

To obtain a long-term reconstruction of the AMOC index requires long-term reconstructions of both the Northern Hemisphere mean temperature and SST of the subpolar gyre. For the Northern Hemisphere mean, Mann et al.¹² produced reconstructions using two different methods, composite-plus-scale (CPS) and errors in variables (EIV). Here we use the land-and-ocean reconstruction with the EIV method using all the available proxies, which is the reconstruction for which the best validation results were achieved (see Supplementary Methods of Mann et al.¹²). Based on standard validation scores (Reduction of Error and Coefficient of Efficiency), this series provides a skilful reconstruction back to AD 900 and beyond (95% significance compared to a red-noise null).

The subpolar-gyre series is derived from a spatial temperature reconstruction¹³, which reconstructs land-and-ocean surface temperatures in every 5° latitude by 5° longitude grid box with sufficient instrumental data to perform calibration and validation. The subpolar gyre falls within the region where the individual grid-box reconstructions are assessed to be skilful compared to a red-noise null¹³. In addition, we performed validation testing of the subpolar-gyre mean series, which indicates a skilful reconstruction back to AD 900 (95% significance compared to a red-noise null; see Supplementary Information for details).

Both time series as well as the resulting AMOC index are shown in Fig. 3. Remarkably, the subpolar gyre reaches nearly its lowest temperatures of the past millennium in the late twentieth century (orange curve), despite global warming. Mann et al.¹³ already noted that this region near Greenland is anomalous in being colder during the modern reference period (1961–1990) than even in the Little Ice Age (LIA). The AMOC index (blue curve) indicates a steady AMOC, with modest changes until the beginning of the twentieth century. There is indication of a maximum in the fifteenth century and a minimum around AD 1600. There is no sign in our index that a weak AMOC caused the LIA in the Northern Hemisphere¹⁴; rather the data are consistent with previous findings that the LIA reflects a response to natural volcanic and solar forcing^{15, 16, 17}, and if anything this surface cooling strengthened the AMOC at least during the first part of the LIA. The fact that LIA coldness seems to have been even more pronounced in South America than in Europe¹⁸ further argues against a weak AMOC, as the latter would have warmed the Southern Hemisphere. The twentieth century shows a gradual decline in the AMOC index, followed by a sharp drop starting around 1970 with a partial recovery after 1990 (discussed further below). This recovery is consistent with the finding of an AMOC increase since 1993 based on floats and satellite altimeter data¹⁹.

Data from the proxy reconstructions of Mann et al.^{12, 13}, including estimated 2-σ uncertainty bands, and from the HadCRUT4 instrumental data⁴⁹. The latter are shown in darker colours and from 1922 onwards, as from this time on data from more than half of all subpolar-gyre grid cells exist in every month (except for a few months during World War II). The orange/red curves are averaged over the subpolar gyre, as indicated on Fig. 1. The grey/black curves are averaged over the Northern Hemisphere, offset by 3 K to avoid overlap. The blue curves in the bottom panel show our AMOC index, namely the difference between subpolar gyre and Northern Hemisphere temperature anomalies (that is, orange/red curves minus grey/black curves). Proxy and instrumental data are decadally smoothed.

The most striking feature of the AMOC index is the extremely low index values reached from 1975 to 1995. It is primarily this negative anomaly that yields the cooling patch in the trend maps illustrated in Fig. 1. In the following we discuss this downward spike in more detail.

The significance of the 1975–1995 AMOC index reduction was estimated using a Monte Carlo method (see Supplementary Information). The annually resolved AMOC reconstruction from 900 to 1850 formed the basis for an ARMA(1,1) model which closely resembles the statistical properties of the data. 10,000 simulated time series of the same length as the AMOC index were constructed. The probability of reaching a similarly weak AMOC index as during 1975–1995 just by natural variability was found to be <0.005, based on the uncertainty of the proxy data and ignoring that this weakening is independently supported by instrumental data.

Figure 5 illustrates corroborating evidence in support of a twentieth-century AMOC weakening. The blue curve depicts the AMOC index from Fig. 3. The dark red curve illustrates the corresponding index based on the instrumental NASA GISS global temperature analysis. The green curve denotes oceanic nitrogen-15 proxy data from corals off the US north-east coast from ref. 25. These annually resolved δ¹⁵N data represent a tracer for water mass changes in the region, where high values are characteristic of the presence of Labrador Slope Water. The time evolution of the δ¹⁵N tracer agrees well with that of our AMOC index (Fig. 5). Ref. 25 reports four more data points from ancient corals preceding the twentieth century, the oldest one from AD ~500. These lie all above 10.5‰, providing (albeit limited) evidence that the downward excursion to values below 10‰ between 1975 and 1995 and the corresponding water mass change may be unprecedented in several centuries.

The blue curve shows our temperature-based AMOC index also shown in Fig. 3b. The dark red curve shows the same index based on NASA GISS temperature data⁴⁸ (scale on left). The green curve with uncertainty range shows coral proxy data²⁵ (scale on right). The data are decadally smoothed. Orange dots show the analyses of data from hydrographic sections across the Atlantic at 25° N, where a 1 K change in the AMOC index corresponds to a 2.3 Sv change in AMOC transport, as in Fig. 2 based on the model simulation. Other estimates from oceanographic data similarly suggest relatively strong AMOC in the 1950s and 1960s, weak AMOC in the 1970s and 1980s and stronger again in the 1990s (refs 41, 51).

Because the AMOC is driven by density gradients related to deep-water formation in the high-latitude North Atlantic, a weakening of the AMOC could be caused by a regional reduction in surface ocean density. Ref. 29 describes an ongoing freshening trend in the northern Atlantic in which the net freshwater storage increased by 19,000 km³ between 1961 and 1995, and the rapid AMOC drop in 1970 was preceded by a large-scale freshening known as the Great Salinity Anomaly^{30, 31}. This freshwater anomaly was described in 1988 as ‘one of the most persistent and extreme variations in global ocean climate yet observed in this century’³¹, the source of which has been linked to anomalous sea-ice export from the Arctic Ocean^{30, 31}. The freshwater volume anomaly of the Great Salinity Anomaly has been estimated as 2,000 km³ along the Labrador coast^{30, 31}.

Additional sources of freshwater addition are increasing river discharge into the Arctic Ocean³² and meltwater and iceberg discharge from the Greenland Ice Sheet (GIS). Because surface flow is directed northward and freshwater tends to remain near the surface owing to its low density, it is difficult to remove freshwater from the northern Atlantic, so an accumulation over longer timescales is plausible. According to a recent reconstruction of the total GIS mass balance from AD 1840 (ref. 33), the GIS was close to balance in the nineteenth century, but a persistent mass loss from Greenland began in AD 1900. The cumulative runoff and ice discharge anomaly (relative to the mean over 1840–1900) during AD 1900–1970 is estimated as 8,000 km³, of which 1,800 km³ was released after 1955 (Fig. 6). It is thus plausible that the accumulated freshwater input from Greenland may have made a significant contribution to the observed freshening trend. A comparable Southern Ocean freshening has likewise been linked to Antarctic ice sheet mass loss³⁴.

Data from Box and Colgan³³. Cumulative anomaly relative to the mean over 1840–1900, a pre-industrial period during which the Greenland Ice Sheet was approximately in balance.

URL: