The Atlantic meridional overturning circulation (AMOC)—a system of ocean currents in the North Atlantic—has a major impact on climate, yet its evolution during the industrial era is poorly known owing to a lack of direct current measurements. Here we provide evidence for a weakening of the AMOC by about 3 ± 1 sverdrups (around 15 per cent) since the mid-twentieth century. This weakening is revealed by a characteristic spatial and seasonal sea-surface temperature ‘fingerprint’—consisting of a pattern of cooling in the subpolar Atlantic Ocean and warming in the Gulf Stream region—and is calibrated through an ensemble of model simulations from the CMIP5 project. We find this fingerprint both in a high-resolution climate model in response to increasing atmospheric carbon dioxide concentrations, and in the temperature trends observed since the late nineteenth century. The pattern can be explained by a slowdown in the AMOC and reduced northward heat transport, as well as an associated northward shift of the Gulf Stream. Comparisons with recent direct measurements from the RAPID project and several other studies provide a consistent depiction of record-low AMOC values in recent years.
Access optionsAccess options
Rent or Buy article
Get time limited or full article access on ReadCube.
All prices are NET prices.
Subscribe to Journal
Get full journal access for 1 year
only $3.90 per issue
All prices are NET prices.
VAT will be added later in the checkout.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
We acknowledge the World Climate Research Programme’s Working Group on Coupled Modelling, which is responsible for CMIP, and we thank the climate modelling groups listed in Extended Data Table 1 for producing and making available their model output. For CMIP, the US Department of Energy’s Program for Climate Model Diagnosis and Intercomparison provides coordinating support and led the development of software infrastructure in partnership with the Global Organization for Earth System Science Portals. Data from the RAPID-WATCH meridional overturning circulation monitoring project were generated with funding from the Natural Environment Research Council and are freely available from www.rapid.ac.uk/rapidmoc. We thank L. Jackson for the GloSea5 reanalysis data, and E. Frajka-Williams for the AMOC reconstruction from satellite altimetry and cable measurements. We also thank the personel of National Oceanic and Atmospheric Administration's GFDL for investeing time and resources into the development of CM2.6, which was evaluated in this research. A.R. was funded by the Marie Curie Horizon2020 project CONCLIMA (grant number 703251). PIK is a Member of the Leibniz Association.Reviewer information
Nature thanks S. Gulev, A. Schmittner and the other anonymous reviewer(s) for their contribution to the peer review of this work.
Extended data figures and tables
Observed linear SST trends (using annual HadISST data), calculated for different timespans to test the robustness of the linear SST trend pattern to the starting and ending years of the timespan. The pattern is normalized with the respective global mean SST trend. Regions that show below-average warming or cooling are in blue; regions that show above-average warming are in red.
Linear SST trends during a CO2-doubling experiment using the GFDL CM2.6 climate model (top), and observed trends during 1870–2016 (HadISST data, bottom), both normalized with the respective global mean SST trends and using data from the November–May season. Regions that show cooling or below-average warming are in blue; regions that show above-average warming are in red. Note again that owing to the much greater climate change in the CO2-doubling experiment, the signal-to-noise ratio for the modelled SST trends is better than that for the observations, and thus the noise level is suppressed by the normalization.
a, The distribution (grey bars) of all local trends, normalized to the global trends, from the HadISST data for 1870–2016, for latitudes between 60° S and 75° N. The distribution is located around µ = 1 with a standard deviation of σ = 0.66 (grey bars). The 5th and 95th percentiles are marked in darker grey. The distribution of the 1870–2016 trends for grid cells assigned to the subpolar gyre regions is shifted to lower or even negative values, with a median of = −0.17 (blue). The distribution of trends for grid cells in the Gulf Stream region are shifted to higher values, with a median of = 2.4 (red). The distributions are normalized to account for the different sample sizes of global, subpolar gyre and Gulf Stream regions. b, As for panel a, but for the CO2-doubling run of the CM2.6 model, with µ = 1.1, σ = 0.48, = −0.02 and = 2.4. The standard deviations of the model data are expected to be smaller than those of the observations because of the larger climate-change signal by which the model data are normalized; this reduces the ‘noise’ of short-term variability relative to the climate signal.
a, The evolution of the Gulf Stream (GS) separation point compared with the AMOC strength in the CM2.6 control and CO2-doubling runs, as indicated by the Gulf Stream index44. The graph shows a link between a weaker AMOC and a northward shift of the separation point. b, Time series of the southward transport of the deep ocean current (summed between depths of 1,000 m and 4,000 m) at 40° N in the region between the US coast and 65° W (see Methods), showing a weakening DWBC during the CO2-doubling experiment. The thin lines show annual values, the thick lines show the 20-year LOWESS-smoothed values.
Extended Data Fig. 5 Linear SST trends from a CO2-doubling experiment using the GFDL CM2.6 climate model, and observed long-term trends from different SST data products, normalized with the respective global mean SST trends.
The trend from 1870 to 2016 was calculated using those datasets that provide data until the present (HadISST16, ERSSTv554, ERSSTv455, ERSSTv3b56 and Kaplan57). Otherwise, it was calculated from 1870 to the end of the available time period (SODA58 and COBE59; see Extended Data Table 1). The SODA data are given for a depth of 5 m instead of the surface; thus, the long-term trend differs for regions with ice cover. For the SODA data, the normalization was adjusted with surface SST data instead of the data at a 5-m depth, to make this dataset comparable to the others. All datasets show a prominent cooling in the subpolar gyre region; the high-resolution data (HadISST, COBE and SODA) also show pronounced warming in the Gulf Stream region.
We calculated the AMOC anomaly from two AMOC indices and two model-based conversion factors. In red is the AMOC anomaly as defined by Rahmstorf et al.7 (HadCRUT4 data), updated with the latest data to 2016. In blue is the AMOC anomaly as defined herein (HadISST data). Thick lines are smoothed by a 10-year LOWESS filter. This smoothing filter is lower than that used in Fig. 6, in order to compare and show the two indices with a higher time resolution.
Extended Data Fig. 7 Sensitivity to the extension of the subpolar gyre region regarding sea-ice cover.
a, Left panel, our original subpolar gyre region (blue outline) and the average November–May sea-ice cover from 1870 to 2016 (blue shading, from HadISST data). Right panel, a reduced subpolar gyre region (green outline) that is always ice-free, compared with the maximum sea-ice cover for the November–May season from 1870 to 2016. b, Comparison of the AMOC indices based on these two regions. The thin lines show annual values, the thick lines show the 20-year LOWESS-smoothed values.
a, We calculated the AMO index from the HadISST dataset after Trenberth and Shea60. This index is defined as the weighted mean SST over the North Atlantic (0° N to 80° N), relative to the mean SST from the period 1901–1970, but with the global mean SST (averaged over the global oceans from 60° S to 60° N) removed. The thin lines show annual values, the thick lines indicate the 20-year LOWESS-smoothed values. We show our AMOC index for comparison. b, As for panel a, but here the AMO index is compared with the interdecadal variability of our AMOC index—that is, the detrended 20-year LOWESS-smoothed index. The comparison shows that the AMO index has similar interdecadal variability to the AMOC index but is lacking the climatic trend found in the latter.
a, Comparison of our AMOC index with the interdecadal variability in the NAO index (after Hurrell61), calculated as the sea-level pressure at the Lisbon station minus the sea-level pressure at the Stykkisholmur/Reykjavik station for the months December to March (DJFM). The thin lines show annual values, the thick lines show the 20-year LOWESS-smoothed values. The linear trend over the whole time period is shown with dashed lines. b, Lagged cross-correlation between the AMOC index and the NAO index shows that peak negative correlation occurs when the AMOC leads the NAO by three years, with R = −0.54. The red lines mark the 95% significance level.