The decrease in ocean heat transport in response to global warming

Oceans redistribute heat by transporting excess tropical ocean heat uptake toward higher latitudes, where it is released to the atmosphere. In the Atlantic Ocean, this is primarily accomplished by the Atlantic meridional overturning circulation (AMOC), which drives northward heat transport throughout the basin and a unique equatorward heat transport in the South Atlantic. In the Indo-Pacific, subtropical cells (STCs) dominate a more zonally symmetric OHT about the equator, with global OHT being poleward at almost all latitudes and thereby helping regulate Earth’s climate by reducing equator-to-pole temperature gradients. Under global warming, atmospheric heat transport (AHT) is expected to increase, largely due to enhanced latent heat transport in a warmer, moister atmosphere. OHT changes tend to oppose AHT changes (Bjerknes compensation), but this compensation is incomplete in a warming world because the ocean absorbs over 90% of excess heat, altering the top-of-atmosphere energy balance. The mechanisms by which OHT decreases under warming, and whether changes are uniform across basins, remain unclear. While increased stratification and vertical temperature gradients generally enhance OHT, a weakened ocean circulation can reduce it. Models consistently project a weakening AMOC under greenhouse forcing due to reduced surface densification at high latitudes (warmer, fresher conditions), but the magnitude differs across CMIP5 and CMIP6. Indo-Pacific OHT changes are less well documented compared to the Atlantic, Arctic, and Southern Ocean. This study investigates how OHT changes by late 21st century across basins, decomposing contributions into overturning and gyre components, and further into velocity-, temperature-, and nonlinear-driven changes, comparing CMIP5 and CMIP6 scenarios.

Prior work establishes that oceans take up the majority of excess planetary heat under anthropogenic forcing and transport it poleward, with the Atlantic OHT dominated by the AMOC and the Indo-Pacific by STCs. Bjerknes compensation suggests opposing OHT and AHT responses, though incomplete under strong ocean heat uptake. Numerous studies indicate AMOC weakening under warming due to reduced surface density in high latitudes, with impacts on North Atlantic OHT and the appearance of a surface warming hole. Differences in AMOC changes between CMIP5 and CMIP6 have been highlighted, with suggestions that CMIP6 may be more sensitive, potentially due to aerosol forcing. OHT changes in the Indo-Pacific have been less extensively documented than in other basins. Observationally constrained estimates of OHT and reanalysis products provide benchmarks but differ in methods and magnitude (e.g., RAPID array estimate of 1.33 PW at 26.5°N vs lower modeled MMM and reanalysis values).

The study analyzes changes in meridional ocean heat transport (OHT) using 52 ensemble members from 22 CMIP5 models and 52 ensemble members from 24 CMIP6 models (Supplementary Tables 1–2). Historical simulations and future scenarios are examined: CMIP5 RCP2.6 and RCP8.5, and CMIP6 SSP1-2.6 and SSP5-8.5. Reference periods are 1970–1999 (historical) and 2070–2099 (future), chosen as common 30-year windows across CMIP phases. Ensemble means and standard deviations are weighted by the number of ensemble members per model. Statistical significance of changes/differences is assessed via a two-sample weighted Kolmogorov–Smirnov test at 5% significance. OHT is computed on native ocean model grids (B- or C-grid), acknowledging that sections do not always align with constant latitude; reported latitudes are section means. Global OHT is partitioned into Atlantic and Indo-Pacific basins north of 34° S and the Southern Ocean south of 34° S. OHT is computed from monthly 3D temperature T(x,y,z,t) and meridional velocity v(x,y,z,t): OHT(y,t) = ∬ T v dx dz × cp × ρ0 / 1e15, with cp = 4000 J kg−1 K−1 and ρ0 = 1026 kg m−3. For each basin and latitude, the section-average velocity is removed (v ← v − v̄) before computing OHT to ensure zero net transport. Model artifacts near the Indonesian Throughflow (due to unresolved narrow straits) that produce spikes in some models are masked at affected latitudes. The total OHT is decomposed into overturning (zonal-mean) and azonal (gyre) components using T = ⟨T⟩ + T′, v = ⟨v⟩ + v′: OHT_ov = ∬ ⟨v⟩⟨T⟩ dx dz (implemented via v′T′ formulation consistent with their definition), and OHT_az = ∬ v′T′ dx dz, following their stated decomposition. Changes in OHT for total, overturning, and gyre components are further decomposed into contributions driven by velocity changes, temperature changes, and a nonlinear residual. Velocity-driven changes fix temperature to the historical seasonal cycle while allowing velocity to vary; temperature-driven changes fix velocity to the historical seasonal cycle while allowing temperature to vary; the nonlinear term is the residual. Monthly-mean data limit the resolution of eddy-driven transports, particularly in the Southern Ocean and at gyre boundaries; many models do not include parameterized (bolus) eddy velocities in archived velocity fields. Where available, meridional ocean temperature transport (hfy) is used to estimate total ocean temperature transport (OTT), including parameterized eddies, for comparison. The meridional overturning circulation (MOC) is computed per basin and latitude from meridional velocity with section-mean velocity removed, yielding a streamfunction comparable to archived MOC (differences ≲1 Sv if computed correctly). To assess the role of higher climate sensitivity in CMIP6, CMIP6 OHT changes are scaled by the relationship between global mean temperature (GMT) change and OHT change derived via linear regression within CMIP6. The scaled CMIP6 OHT = OHT_CMIP6 − a (GMT_CMIP6 − GMT_CMIP5), effectively adjusting CMIP6 to the GMT of CMIP5 for scenario comparisons. Significance of differences between scaled CMIP6 and CMIP5 accounts for the applied scaling.

- Across scenarios, multimodel means (MMMs) project a decrease in poleward OHT north of the equator and south of ~10° S, with stronger reductions in CMIP6 than CMIP5, even after adjusting for higher CMIP6 climate sensitivity. - Atlantic basin (near 26–26.5° N): MMM reductions in northward OHT are approximately: - CMIP5 RCP2.6: −0.093 PW (−9.3%); CMIP5 RCP8.5: −0.200 PW (−20.0%). - CMIP6 SSP1-2.6: −0.232 PW (−23.3%); CMIP6 SSP5-8.5: −0.304 PW (−30.5%). - CMIP6 scaled: SSP1-2.6: −0.206 PW; SSP5-8.5: −0.295 PW. - The CMIP6 response in SSP1-2.6 is ~2.5× larger than CMIP5 RCP2.6; in SSP5-8.5 it is ~1.5× larger than CMIP5 RCP8.5. - Indo-Pacific basin (20° S): MMM reductions in poleward (southward) OHT are approximately: - CMIP5 RCP2.6: −0.097 PW; CMIP5 RCP8.5: −0.123 PW. - CMIP6 SSP1-2.6: −0.157 PW; CMIP6 SSP5-8.5: −0.194 PW. - CMIP6 scaled: SSP1-2.6: −0.142 PW; SSP5-8.5: −0.176 PW. - Southern Ocean (55° S): MMM southward OHT is reduced by 0.071–0.268 PW, large enough in CMIP6 SSP5-8.5 to reverse direction at 55° S from −0.154 PW (southward) to +0.114 PW (northward) (+0.063 PW in scaled CMIP6). - Decomposition indicates the reductions in Atlantic OHT are predominantly overturning-driven and dominated by velocity-driven changes that closely mirror AMOC declines; temperature-driven changes partly offset these south of the subpolar gyre. Gyre contributions are secondary and mainly relevant in the subpolar North Atlantic. - AMOC weakens substantially in future scenarios, with stronger declines in CMIP6 than CMIP5 across both weak and strong forcing scenarios, consistent with larger velocity-driven OHT reductions. - In the Indo-Pacific, OHT changes are dominated by overturning south of the equator. Velocity-driven reductions diminish OHT in both hemispheres; in the Northern Hemisphere these are largely counteracted by temperature-driven increases due to enhanced vertical temperature gradients. South of ~15° S, weakening of the deep overturning cell (reduced AABW inflow and formation) dominates, with limited compensation from temperature-driven changes because deep cells are less affected by upper-ocean stratification. - In the Southern Ocean near 55° S, overturning-driven OHT change is primarily temperature-driven (increased vertical temperature gradients from upper-ocean warming peaking near ~45° S). Gyre-driven OHT shows no significant change near 55° S. - Historical benchmarking: Modeled MMM Atlantic OHT at 26.5° N (1.001/0.998 PW in CMIP5/6) is lower than the RAPID observational estimate (1.33 PW), and falls between indirect observation-based (1.11 PW) and reanalysis-based (0.72 PW) estimates, indicating a tendency to underestimate Atlantic OHT. - Scaling CMIP6 to CMIP5 GMT reduces but does not eliminate CMIP5–CMIP6 differences in many regions; near 55° S, scaling makes CMIP5 and CMIP6 OHT changes very similar and often statistically indistinguishable.

The results directly address how and where poleward OHT declines under global warming and how these changes differ across basins and model generations. Decreases are widespread in the Northern Hemisphere and south of ~10° S, with the Atlantic showing robust, overturning-driven reductions tied to AMOC weakening. This reduction produces limited divergence in OHT with latitude until ~40° N and contributes to the North Atlantic “warming hole,” which also reduces local air–sea heat loss. In the Indo-Pacific, reduced OHT south of the equator is dominated by the weakening of the deep overturning cell and reduced AABW formation, whereas in the Northern Hemisphere Indo-Pacific, wind stress–related weakening of the STC reduces OHT but is largely offset by enhanced vertical temperature gradients that increase temperature-driven OHT. In the Southern Ocean, OHT reductions are largely temperature-driven (enhanced vertical gradients) and can even reverse net transport near 55° S in high-forcing CMIP6 scenarios. The stronger CMIP6 response relative to CMIP5—especially for Atlantic OHT and AMOC—suggests that future climate impacts mediated by ocean circulation may be larger than previously thought. The extent to which CMIP6 differences reflect improved realism versus higher aerosol forcing sensitivity remains unresolved; some evidence points to CMIP6 models being overly sensitive to aerosols, potentially exaggerating AMOC decline compared with CMIP5, while proxy reconstructions imply that even larger AMOC sensitivity may be plausible than simulated historically. Scenario design differences (SSP vs RCP) could also contribute to larger CMIP6 changes due to faster early forcing increases, although single-model comparisons (e.g., CanESM5) show smaller SSP–RCP differences than the MMM. Overall, the findings underscore the critical role of overturning circulation changes in shaping regional climate responses, polar amplification, and ocean–atmosphere heat exchange.

By systematically decomposing future OHT changes across basins and CMIP phases, the study shows a robust, widespread reduction in poleward OHT under global warming, driven predominantly by overturning circulation declines (AMOC in the Atlantic; deep cell/AABW in the Southern Hemisphere), partly offset by temperature-driven effects from increased vertical stratification. The reductions are stronger in CMIP6 than CMIP5, persisting even after accounting for higher climate sensitivity, with the Atlantic exhibiting the largest differences due to a stronger projected AMOC decline. These changes have important implications for regional climate patterns, including the North Atlantic warming hole and reduced polar amplification. Future work should better constrain AMOC sensitivity and aerosol forcing impacts, extend observational benchmarks of OHT (particularly in the Southern Ocean and South Atlantic), improve representation and availability of eddy/bolus transports in model archives, and refine Indo-Pacific process representations (e.g., STCs and Indonesian Throughflow) to reduce uncertainties in basin-scale OHT projections.

Key limitations include: (1) Use of monthly mean data misses higher-frequency variability and eddy contributions, especially important in the Southern Ocean and at gyre boundaries; parameterized bolus velocities are generally not included in archived velocity fields, though OTT (hfy) is available for some models and suggests notable Southern Ocean impacts. (2) Differences in OHT estimation methods among reanalysis- and observation-based products complicate benchmarking; models tend to underestimate Atlantic OHT at 26.5° N compared with RAPID. (3) Model grid curvature and native-grid calculations mean OHT sections do not follow constant latitudes and restrict analysis to ≤65° N; spikes associated with unresolved Indonesian Throughflow require data masking at affected latitudes in some models. (4) Not all required variables are available for all models/ensembles, reducing sample sizes for some diagnostics. (5) The MMM OHT does not include bolus transport, leading to discrepancies with observations in the Southern Ocean that are reduced when OTT with bolus is considered. (6) CMIP6’s higher climate sensitivity and potential aerosol forcing sensitivities complicate direct comparisons with CMIP5; scaling by GMT addresses part of this but residual differences remain. (7) Observational constraints are sparse south of 34° S, increasing uncertainty in Southern Ocean assessments.