Likely accelerated weakening of Atlantic overturning circulation emerges in optimal salinity fingerprint

Abstract

The long-term response of the Atlantic meridional overturning circulation (AMOC) to anthropogenic forcing has been difficult to detect from the short direct measurements available due to strong interdecadal variability. Here, we present observational and modeling evidence for a likely accelerated weakening of the AMOC since the 1980s under the combined forcing of anthropogenic greenhouse gases and aerosols. This likely accelerated AMOC weakening signal can be detected in the AMOC fingerprint of salinity pileup remotely in the South Atlantic, but not in the classic warming hole fingerprint locally in the North Atlantic, because the latter is contaminated by the "noise" of interdecadal variability. Our optimal salinity fingerprint retains much of the signal of the long-term AMOC trend response to anthropogenic forcing, while dynamically filtering out shorter climate variability. Given the ongoing anthropogenic forcing, our study indicates a potential further acceleration of AMOC weakening with associated climate impacts in the coming decades.

Fig. 1. Data-model comparison.

Time series of observed (thick lines) and multi-model ensemble mean (MMEM, thin lines; shading shows one standard deviation) (a) anomaly of Atlantic meridional overturning circulation (AMOC) intensity, (b) the salinity pileup fingerprint S_S, (c) the salinity pileup fingerprint S_SA, (d) monthly mean observation weight of salinity analysis over the subtropical South Atlantic (analysis value is more influenced by observations with weight closer to one), (e) the classical warming hole fingerprint T_NA and (f) key external forcing of CO₂ (from NASA GISS, brown), anthropogenic (ASR_NH, blue, Methods) and volcanic (shown as global stratospheric aerosol optical depths at 550 nm; from NASA GISS, grey) aerosols. Observed fingerprints (EN4 analysis data for S_S and S_SA; HadISST dataset for T_NA) are shown as 5-year running means. The AMOC anomaly from ECCOv4r4 reanalysis and RAPID measurements are shown in thick purple and blue lines with 1 and 2 offsets, respectively. The linear change of ECCOv4r4 AMOC is plotted in dashed purple line. Also shown in (d) is the observed AMO index (Methods). Model results are relative to the means of 1900–1950.

Fig. 2. Trend correlation between Atlantic meridional overturning circulation (AMOC) indices for period 1850–1985.

a Scatter for trends of AMOC intensity and its salinity-based fingerprint S_S across model ensemble means. Trends are calculated as the linear change from 1850 to 1985; (b) same as (a) but across model members. (c, d) same as (a, b) but for AMOC and the salinity-based fingerprint S_SA; (e, f) same as (a, b) but for AMOC and the warming hole fingerprint T_NA. Solid line in each panel marks the corresponding value in observations (1900–1985 for salinity indices and 1870–1985 for T_NA). Red and blue numbers in each panel are the correlation coefficients in CMIP6 (dots) and CMIP5 (triangles), respectively.

Fig. 3. Lead-lag correlation.

a–c Lead-lag correlation between Atlantic meridional overturning circulation (AMOC) intensity and southern AMOC intensity (gray), AMOC and the warming hole fingerprint T_NA (green), AMOC and the salinity-based fingerprint S_S (magenta), AMOC and the salinity-based fingerprint S_SA (orchid) for piControl (a), Hist-GHG (b) and Hist-aer (c) simulations. Time series is 11 years locally weighted scatterplot smoothing (LOWESS) filtered. d–f Scatter diagram of maximum correlations between AMOC fingerprints and AMOC when AMOC leads. Small magenta (S_S versus T_NA) and orchid (S_SA versus T_NA) dots are for each CMIP6 model member while big ones are for the ensemble mean. Vertical and horizontal dashed lines in (d–f) indicate the 95% significance level (determined by Monte Carlo method) for SSS indices (with a lag of 20 years) and for T_NA (with a lag of 5 years), respectively. For piControl simulation, the significance level is the averaged value over the seven simulations with different lengths (Table S1).

Fig. 4. Response of AMOC indices in CMIP6-DAMIP experiments.

Time series of anomalous multi-model ensemble mean (MMEM) Atlantic meridional overturning circulation (AMOC) intensity (a), salinity-based fingerprint S_S (b), salinity-based fingerprint S_SA (c) and warming hole fingerprint T_NA (d) in CMIP6 historical (purple), Hist-aer (blue), Hist-nat (green), Hist-GHG (orange) and Hist-stratO3(gray) simulations. Anomalies are relative to the means of 1850–1900.

Fig. 5. Trend-signal/variability-noise ratio of Atlantic meridional overturning circulation (AMOC) indices.

a–h Demonstrate the results from an ocean general circulation model (OGCM) experiment: a Surface heat flux forcing applied in the North Atlantic (black) which consists of two signals: 30-years variability signal (blue) and centennial trend signal (red). b Normalized anomaly of the AMOC at 30°N (black), AMOC at 30°S (gray), warming hole fingerprint T_NA (green) and salinity-based fingerprint S_S (magenta). Anomalies are relative to the corresponding control experiment. c Anomaly of the northern and southern AMOCs. Also shown are the cubic best-fit regression lines (dashed) for the two. d Lead-lag correlation between AMOC (northern AMOC) and southern AMOC (gray), AMOC and T_NA (green) and AMOC and S_S (magenta). e–g The decomposition of (e) northern AMOC, (f) T_NA and (g) S_S into long-term trend (red) and short-term variability (residual; blue). h Ratio of standard deviation between centennial and interdecadal AMOC variability as a function of latitude. Magenta and green stars mark the ratio of S_S and T_NA, respectively. i–l Same as (e–h) but for a coupled general circulation model (CGCM) simulation (see text).