Wide range of possible trajectories of North Atlantic climate in a warming world

The study addresses how anthropogenic external forcing modulates internal decadal variability of North Atlantic sea surface temperatures (SSTs). Decadal climate prediction is valuable for adaptation and risk management, but prediction reliability depends on understanding sources of decadal variability, including internal variability, which can cause large and time-evolving uncertainties under changing forcing. The North Atlantic exhibits strong interannual-to-multidecadal variability that affects heat waves, tropical cyclones, Arctic sea ice, monsoons, and global precipitation. While mechanisms of North Atlantic SST variability have been widely studied, it remains unclear how human-induced forcing alters internal variability at decadal scales. Projected weakening of the Atlantic meridional overturning circulation (AMOC) under warming, linked to increased heat and freshwater input and reduced deep convection, motivates examining how internal variability may evolve in response to these mean-state changes. The research questions are: whether internal decadal variability of northern North Atlantic SST increases under warming, when and how distinct climate trajectories emerge across realizations, and what mechanisms trigger and sustain them.

Prior work documents North Atlantic variability in observations and models and its climate impacts (e.g., on heat waves, cyclones, sea ice, monsoons, and precipitation). Studies debate the roles of ocean circulation versus surface fluxes in Atlantic multidecadal variability and emphasize links among NAO, deep convection, AMOC, and SST. Projections indicate AMOC weakening and emergence of the North Atlantic warming hole with associated atmospheric impacts, though the magnitude and patterns vary by model. Large ensembles have been used to study internal variability and forced responses across modes like ENSO and PDO. Evidence suggests human-induced changes in variability are widespread. Observational studies indicate an AMOC slowdown and a warming hole fingerprint, but uncertainties remain. This study builds on these findings by using a 100-member initial-condition large ensemble to separate forced and internal components and to diagnose time-evolving internal variability under external forcing.

- Model and forcing: Community Earth System Model Version 2 Large Ensemble (CESM2-LE), 100 members. Forcing: CMIP6 historical (1850–2014) and SSP3-7.0 (2015–2100). Components at nominal 1° resolution (CAM6, POP2, CICE5, CLM5). Ensemble spread arises from different oceanic/atmospheric initial states and micro-/macro-perturbations; two biomass burning forcings (original and low-pass filtered) used during 1997–2014. - Time focus and filtering: Winter season defined as DJFM. Decadal variability isolated via 11-year running mean of DJFM averages. - Indices and regions: Northern North Atlantic SST (NNASST) index defined as area average over 50–80°N, 90°W–40°E. Labrador Sea region defined with POP2 masks, northern boundary at 65°N. AMOC index at 50°N defined as the maximum Eulerian-mean overturning streamfunction below 500 m. Northward heat transport (NHT) computed at 50°N. NAO index defined as normalized DJFM sea level pressure difference between model grid points nearest Lisbon, Portugal and Stykkisholmur, Iceland. - Group classification: To identify divergent trajectories, ensemble members were classified based on their winter NNASST during the five years with maximum ensemble STD (2056–2060). Members with NNASST greater than (less than or equal to) the ensemble mean plus (minus) one standard deviation of the CESM2 piControl NNASST during any of these years were assigned to the warm (cold) group. Result: 40 warm and 31 cold members. Robustness tested against season/domain choices and alternative classification methods. - Mixed layer and density analyses: Assessed Labrador Sea mixed layer depth (MLD) evolution and differences between groups. Decomposed Labrador Sea density differences (upper 1500 m) into temperature and salinity contributions using a linear equation of state with thermal expansion (αθ) and haline contraction (βs) coefficients computed relative to piControl reference values (upper 295 m, DJFM means). - Heat and salinity budget analyses: For the upper 295 m in the Labrador Sea, decomposed total temperature tendency (°C/yr) and salinity tendency (g/kg/yr) into contributions from resolved advection, parameterized advection (mesoscale/submesoscale), diabatic vertical mixing (vertical diffusivity), surface fluxes (net air–sea heat flux including ice formation for heat; precipitation, evaporation, river/ice runoff, ice melt/growth for salinity), and a residual term (including lateral diffusion and KPP non-local mixing not available in outputs, plus small Robert filter tendencies). Budgets computed for DJFM, then low-pass filtered. - Feedback and trigger diagnostics: Compared NAO, wind stress magnitude over the Labrador Sea, surface heat fluxes, and sea-ice cover between groups to diagnose stochastic triggers and hypothesized positive feedbacks (vertical mixing–surface salinity feedback; sea ice–wind stress feedback). - Timing and external forcing context: Examined ensemble-mean vertical salinity gradient evolution in the Labrador Sea to assess when feedbacks become active under increasing surface freshening. - Cross-model assessment: Analyzed seven additional large ensembles (CanESM2, CanESM5, ACCESS-ESM1-5, MPI-ESM, CSIRO-Mk3.6, MIROC6, GFDL-SPEAR) under historical plus future scenarios (RCP8.5, SSP3-7.0, SSP5-8.5; ensemble sizes ≥25). Computed a warming hole index (ratio of regional-to-global SST change from early 2000s baseline to 2070–end) to characterize ensemble-mean North Atlantic response. Evaluated time-evolving ensemble spread of DJFM SST over a common subpolar domain (50–65°N, 40–20°W) after low-pass filtering and time-normalization for each model; compared to CESM2-LE results.

- Internal variability increase: The ensemble spread of northern North Atlantic SST (NNASST) more than doubles in the mid-21st century, peaking during 2056–2060. Distinct warm and cold trajectory groups emerge with ensemble means diverging from ~2035 and remaining separated through 2100. - Spatial evolution: Differences originate in the Labrador Sea in the 2030s and expand across the subpolar and mid-to-high latitude North Atlantic, indicative of earlier development of the North Atlantic warming hole in the cold group and transport of colder water from the Labrador Sea. - Deep convection and timing: All members show shoaling Labrador Sea MLD under warming, but the cold group exhibits an accelerated reduction, with earlier shutdown: median year reaching 80 m MLD is 2052 (cold) vs 2063 (warm), an 11-year difference. Differences in NHT and AMOC trajectories begin in 2029 and 2032, respectively; the cold group has reduced NHT and slightly weaker AMOC. - Density controls: Upper-ocean (0–100 m) Labrador Sea density is significantly lower in the cold group starting in the 2020s, peaking in the 2050s. Salinity differences dominate density differences (freshening in cold group), while temperature contributions are smaller and of opposite sign (damping). - Salinity budget: The cold group’s smaller (more negative) total salinity tendency during winters from the 2020s to ~2058 is primarily due to weaker diabatic vertical mixing, limiting upward mixing of saltier subsurface waters. Surface salinity fluxes and parameterized advection generally oppose this difference; resolved advection shifts from amplifying (2031–2040) to opposing (from ~2049). Residual terms modestly reinforce negative tendencies from ~2036. - Heat budget: The cold group has a smaller total temperature tendency (early 2030s–late 2050s), with major contributions from weaker vertical mixing and residual processes; both resolved and parameterized advection contribute moderately. Surface heat flux differences consistently damp the group differences due to reduced turbulent heat loss in the cold group. - Trigger and feedbacks: A stochastic NAO difference (cold group has lower NAO from 2027–2038) likely triggers initial weaker convection via reduced westerlies, reduced heat loss, and weaker mechanical stirring (lower wind stress). Under enhanced surface freshening, two positive feedbacks amplify and sustain divergence: (1) vertical mixing–surface salinity feedback (weaker convection limits upward salt flux, further freshening and reducing density); (2) sea ice–wind stress feedback (colder SST and greater sea ice reduce melt and mechanical stirring near the ice edge, further weakening mixing and deep convection). Evidence includes significant group differences in sea ice cover, wind stress magnitude, and vertical-mixing-driven salinity tendencies near the ice edge. - Role of external forcing: The ensemble-mean vertical salinity gradient in the Labrador Sea increases over time due to Arctic sea-ice melt and runoff, making the system more sensitive to vertical mixing and enabling feedback activation in the early 21st century. Eventually, pervasive surface freshening shuts down deep convection in all members, leading to possible re-convergence as feedbacks cease. - Cross-model robustness: In seven additional large ensembles, models that develop a pronounced North Atlantic warming hole (CESM2, MPI-ESM, CSIRO-Mk3.6, MIROC6, GFDL-SPEAR) show a distinct increase in subpolar SST ensemble spread in the 21st century (timing varies). Models without a warming hole (CanESM2, CanESM5, ACCESS-ESM1-5) show decreased spread relative to the 20th century, even when restricting to relative-cooling regions. This links increased internal variability to models with strong deep convection shutdown and AMOC slowdown signatures.

The findings demonstrate that anthropogenic forcing can enhance internal decadal variability of North Atlantic SST by shifting the mean state toward conditions that activate nonlinear positive feedbacks. A brief stochastic atmospheric trigger (NAO) around the early 2030s leads to persistent, multidecadal divergence in SST trajectories once the ocean’s surface freshening strengthens vertical salinity gradients, amplifying differences through mixing–salinity and sea ice–wind stress feedbacks. This mechanistic link explains the timing of increased variability and connects it to mean-state changes such as deep convection shutdown and AMOC weakening. Cross-model comparisons indicate that increased subpolar SST internal variability is a robust feature among models exhibiting a North Atlantic warming hole, highlighting the broader relevance of the mechanisms beyond a single model. These results imply that, despite larger internal variability, there may be enhanced multidecadal predictability if the trajectory (warm vs cold) can be identified near the tipping period (~late 2020s in CESM2), offering potential societal benefits for climate risk management.

The study reveals a dramatic mid-21st-century increase in internal variability of northern North Atlantic SST in CESM2-LE, with distinct warm and cold trajectories emerging from the early 2030s and persisting for decades. The divergence is stochastically triggered (NAO variability) and amplified by two positive feedbacks involving vertical mixing–surface salinity and sea ice–wind stress interactions, which become active under externally forced surface freshening. Multi-model evidence links increased subpolar SST internal variability to the formation of the North Atlantic warming hole, indicating robustness across models that simulate strong deep convection shutdown and AMOC slowdown. Future research directions include: (1) systematic monitoring of observed NNASST and related ocean metrics (MLD, NHT, AMOC) to identify which trajectory the real world is following, potentially enabling multidecadal outlooks; (2) extending dynamical decadal predictions beyond the usual ~10-year horizon once positive feedbacks activate (around 2027 in CESM2), to achieve skillful long-lead forecasts; and (3) exploring how impacts and predictability associated with natural external forcings (e.g., volcanic eruptions) may change under global warming.

- Model dependence: Mechanisms and regional expressions may depend on CESM2 specifics (e.g., dominant deep water formation in the Labrador Sea, parameterization schemes). While cross-model analysis supports robustness for “warming hole models,” inter-model differences in physics and forcing scenarios affect timing and magnitude. - Observational constraints: Many key variables (MLD, AMOC, NHT) lack long observational records, limiting direct validation of the proposed timing and mechanisms; SST is better observed but indirect. - Diagnostic completeness: Some budget terms (lateral diffusion, KPP non-local mixing) were not available and are included in residuals, introducing uncertainty in partitioning contributions. - Forcing scenario and ensemble design: Results reflect SSP3-7.0 forcing in CESM2-LE and specific ensemble initializations, which may influence timing of tipping and divergence; other scenarios may shift timing. - Classification choices: Although tested for robustness, group definitions based on thresholds and periods (2056–2060) may influence member counts and specific trajectories.