Present-day North Atlantic salinity constrains future warming of the Northern Hemisphere

The North Atlantic (NA) plays a pivotal role in regulating global and Northern Hemisphere (NH) climate through the Atlantic Meridional Overturning Circulation (AMOC), which transports heat and carbon from the tropics. Under global warming, the AMOC is projected to weaken, with associated changes in ocean heat transport and freshwater fluxes, contributing to phenomena such as the NA subpolar 'warming hole' and large uncertainties in future climate projections. The NA, while covering only ~15% of ocean area, stores about a quarter of the ocean’s anthropogenic carbon. However, Earth system models (ESMs) exhibit a wide spread in AMOC intensity and NA carbon uptake, leading to uncertainty in NH and global warming projections. The study aims to reduce this uncertainty by developing an emergent constraint based on present-day sea surface salinity (SSS) in the NA subpolar region, a well-observed metric and a proxy for AMOC strength, to constrain future NA carbon uptake and NH surface warming.

Prior research highlights the NA’s outsized role in anthropogenic carbon uptake and climate regulation, with AMOC strength linked to variability in heat and carbon transport. Studies have reported expected AMOC decline under warming and its impacts on hemispheric energy balances, the ITCZ, and regional climate phenomena. The NA accounts for a substantial fraction of anthropogenic carbon storage, with uptake closely tied to ocean mixing and deep convection processes. Earlier modeling phases (CMIP5 and CMIP6) show large intermodel spreads in AMOC and NA carbon uptake, driving uncertainty in warming projections. Emergent constraints have successfully reduced uncertainties in other climate metrics (e.g., climate sensitivity, precipitation extremes, Arctic sea-ice albedo). Salinity has been used to constrain carbon uptake in the Southern Ocean, and NA subpolar SSS is recognized as a proxy for AMOC strength, often correlating more strongly with AMOC than SST. Despite the Southern Ocean’s dominant share of global carbon uptake, model uncertainty there is high and not well correlated with NH warming, underscoring the value of NA-focused constraints.

• Models and scenarios: Analyzed 31 ESMs (18 CMIP6, 13 CMIP5). Historical simulations (1850–2014 for CMIP6; to 2005 for CMIP5) are followed by high-emission scenarios (SSP5–8.5 for CMIP6; RCP8.5 for CMIP5) through 2100. One ensemble member per model was used to weight models equally.
• Regions and periods: NA subpolar region defined as 55°W–15°W, 45°N–65°N; NH as 0–360°E, 10°N–90°N. Present-day climatology is 1981–2010; future period is 2071–2100; pre-industrial baseline is 1850–1900.
• Observations: Present-day SSS from World Ocean Atlas 2018 (WOA18) annual climatology (1981–2010), with standard errors to quantify uncertainty; an additional multi-observation SSS dataset (1993–2014) is referenced. Model outputs were interpolated to a 1°×1° grid.
• Variables and diagnostics: NA cumulative anthropogenic carbon uptake computed as the difference in air–sea CO2 flux (fgco2) between historical+scenario runs and concurrent pre-industrial control, encompassing both forced and natural changes. AMOC stream function calculated from meridional velocity or provided streamfunction variables, with AMOC index as the maximum streamfunction at 26°N below 50 m. Global heat–carbon coupling parameter α derived from regression of anthropogenic ocean carbon storage onto ocean heat storage, computed via vertical integrations with fixed density and heat capacity.
• Statistical analyses: Intermodel relationships between SSS, AMOC strength, NA carbon uptake, and NH/global warming assessed using linear regressions. For figure relationships, both total least squares (TLS) and ordinary least squares (OLS) are used as appropriate. Significance assessed by two-sided Student’s t-tests. Intermodel uncertainty is one standard deviation across models.
• Emergent constraint framework: Established linear emergent relationships between present-day SSS (predictor, x) and projected outcomes (y): NH surface temperature change and NA cumulative carbon uptake (2071–2100). Prior model distributions assumed Gaussian. Conditional PDFs P(y|x) derived from regression prediction errors; observational PDF P(x) from WOA18 SSS mean and standard error; constrained PDFs P(y) obtained by integrating the product P(y|x)P(x) over x. Out-of-sample testing performed using CMIP5 models. Sensitivity assessed for SSP1–2.6 as well.
• Radiative feedback assessment: Linked NA carbon uptake to NH downward longwave radiation changes and total surface heat flux components to explain the greenhouse effect modulation across models.

• Intermodel relationship: Future NH surface temperature change is significantly and negatively correlated with future cumulative NA carbon uptake in CMIP6 (r = −0.73). Models with larger NA carbon uptake project weaker NH warming.
• Mechanism: Greater NA carbon uptake lowers atmospheric GHG concentrations, reducing increases in downward longwave radiation over the NH and yielding less surface warming.
• Present-day SSS as proxy and predictor:
  • Present-day SSS in the NA subpolar region correlates strongly with present-day AMOC (SSS–AMOC r ≈ 0.90; SST–AMOC r ≈ 0.75), and with present-day NA cumulative carbon uptake (r = 0.66).
  • Present-day SSS correlates with future SSS (r = 0.60), indicating persistence of relative AMOC strength across models.
  • Future SSS correlates with future NA cumulative carbon uptake (r = 0.69).
• Emergent constraint relationships using present-day SSS (1981–2010):
  • Correlation with future NH surface temperature: r = −0.72.
  • Correlation with future NA cumulative carbon uptake: r = 0.88.
  • Observed NA subpolar SSS (WOA18): 34.75 ± 0.04 psu; CMIP6 ensemble mean: 34.41 ± 0.70 psu.
• Constrained projections under SSP5–8.5 (2071–2100 vs 1850–1900):
  • NH surface warming: from 5.9 ± 1.4 °C (prior) to 5.5 ± 1.0 °C (after constraint), a 30% reduction in uncertainty.
  • NA cumulative carbon uptake: from 50.1 ± 10.7 PgC (prior) to 54.7 ± 5.1 PgC (after constraint), a 53% reduction in uncertainty.
  • Global mean surface temperature: from 4.7 ± 1.1 °C to 4.4 ± 0.8 °C, a 23% uncertainty reduction (Extended Data Fig. 7).
• Model improvements and comparisons:
  • CMIP6 shows improved decadal subpolar NA SST variability vs CMIP5, but similar uncertainty ranges.
  • Out-of-sample test with CMIP5 supports the constraint: uncertainty reductions of 39% for NH warming and 26% for NA carbon uptake.
• Additional insights:
  • The ocean has not saturated in NA carbon uptake by 2100 under SSP5–8.5 across ESMs.
  • The global heat–carbon coupling parameter α is significantly related to subpolar NA SSS in the present day (r ≈ −0.60), suggesting a role of global circulation in NA uptake.
  • The emergent relationship is stronger in ESMs with interactive ocean biogeochemistry (r = −0.72) than without (r = −0.53).

The study demonstrates that present-day SSS in the NA subpolar region, a robust observational metric and strong proxy for AMOC strength, provides a powerful emergent constraint on future NA carbon uptake and NH warming. Models with higher present-day SSS (stronger AMOC) transport more carbon-enriched surface waters poleward and into the deep ocean, leading to larger carbon uptake that suppresses the greenhouse effect via reduced increases in downward longwave radiation, and hence smaller NH surface warming. This causally consistent chain—SSS → AMOC → NA carbon uptake → NH radiative response—explains the negative intermodel correlation between NA carbon uptake and NH warming. Applying the observed SSS tightens projections, lowering both the mean projected NH and global warming and substantially narrowing uncertainties. The results suggest current ESMs may underestimate present-day subpolar NA SSS, resulting in an overestimation of NH warming and underestimation of NA carbon uptake. The constraint is robust across high and low emissions scenarios and is stronger in models that include ocean biogeochemistry, underlining the importance of interactive carbon cycle processes in constraining climate projections. While the Southern Ocean dominates global ocean carbon uptake, its model spread is not well-linked to NH warming, reinforcing the unique leverage of the NA for constraining NH and global temperatures.

By identifying present-day subpolar NA SSS as an emergent constraint, the study substantially reduces uncertainty in projections of NH surface warming and NA cumulative carbon uptake under SSP5–8.5. Constrained estimates indicate lower NH and global warming and higher NA carbon uptake than the unconstrained CMIP6 ensemble, implying that current ESMs likely overestimate future warming. The work highlights the critical role of AMOC-mediated carbon uptake in shaping NH climate and underscores the value of sustained, accurate SSS observations in the NA subpolar region. Future research should: (1) maintain and expand long-term SSS and AMOC observations to refine constraints; (2) further investigate mechanisms linking SSS, AMOC, and carbon–heat coupling across basins; (3) apply model independence and performance weighting within emergent-constraint frameworks; (4) extend analyses to additional scenarios and to coupled biogeochemical feedbacks; and (5) evaluate regional climate impacts and extremes under constrained warming trajectories.

• AMOC observational constraint is limited by short (2005–2014) and highly variable records, yielding a weaker emergent relationship than SSS-based constraints and potential biases in constrained means.
• The relationship between SSS, SST, and AMOC is complex; using SSS as a single proxy may oversimplify underlying dynamics and feedbacks.
• Model dependence and structural similarities may violate the assumption of model independence (one ensemble member per model used). Performance weighting was not applied.
• ESM biases (e.g., underestimation of present-day subpolar NA SSS) can influence emergent relationships.
• Southern Ocean carbon uptake uncertainty remains large and its relationship to NH warming is weak, limiting broader applicability beyond the NA focus.
• The emergent relationships are established under specific periods and scenarios; robustness across different forcings and internal variability states, while partly tested (e.g., SSP1–2.6, CMIP5), may warrant further evaluation.