Failure to track a stable AMOC state under rapid climate change

Abstract: The Atlantic Meridional Overturning Circulation (AMOC) is a tipping element of the climate system. The current estimate of the global warming threshold for the onset of an AMOC collapse is +4C. However, such a threshold may not be meaningful because AMOC stability depends on the rate of radiative forcing and background climate state. Here, we identify an AMOC stabilising mechanism that operates on timescales longer than present-day radiative forcing increase. Slow forcing permits coherent adjustment of surface and interior ocean properties, supported by enhanced evaporation and reduced sea-ice extent, counteracting destabilising feedbacks. This mechanism is explicitly demonstrated in a slow CO2 increase experiment (+0.5 ppm/yr), in which the AMOC remains stable up to +5.5C of global warming. By contrast, under intermediate- and high-emission scenarios, the AMOC collapses at substantially lower warming levels (+2.2C and +2.8C, respectively). Our findings demonstrate the strong radiative forcing path dependence of AMOC tipping and imply that limiting the rate of radiative forcing is critical for reducing the near-term risk of an AMOC collapse.

• The paper demonstrates that AMOC tipping depends on the rate and pathway of radiative forcing, with stability at +5.5°C under slow CO₂ increases versus collapse at +2.2–2.8°C under faster scenarios.
• The paper attributes slow-forcing stability to enhanced evaporation, sea-ice retreat, and coordinated surface–interior density adjustment that preserve deep-water formation pathways.
• The paper shows that a coupled box model reproduces rate-induced tipping and suggests monitoring buoyancy-flux and NADW transformation indicators rather than relying on a universal warming threshold.

The Atlantic Meridional Overturning Circulation (AMOC) is widely treated as a tipping element whose collapse is associated with a global warming threshold, commonly quoted at +4.0°C with a range of +1.4°C to +8.0°C. In "Failure to track a stable AMOC state under rapid climate change" (2602.09964), van Westen, Börner, and Dijkstra argue that such a threshold is not a well-defined quantity: AMOC tipping depends strongly on the radiative forcing pathway, not merely on the warming level reached. Using the Community Earth System Model (CESM v1.0.5) and CMIP6 output, they demonstrate that a slow CO₂ increase keeps the AMOC stable up to +5.5°C of global warming, whereas intermediate- and high-emission scenarios trigger collapse at +2.2°C and +2.8°C, respectively.

Experimental design

All simulations branch from a quasi-equilibrium pre-industrial freshwater hosing simulation in which a slowly varying freshwater flux $F_H$ over 20°N–50°N was compensated elsewhere, producing statistical equilibria at fixed $\overline{F_H}$. The authors focus on the equilibrium at $\overline{F_H} = 0.45$ Sv, which lies in the destabilising salt-advection feedback regime, closer to the saddle-node bifurcation located near $F_H \approx 0.5$ Sv in this model version. From this state they run historical forcing followed by RCP2.6, RCP4.5, and RCP8.5 extended to year 2500, plus a dedicated CO₂ ramp simulation in which CO₂ rises linearly from 284.7 ppm at +0.5 ppm yr⁻¹ for 1750 years, reaching 1120 ppm (4×CO₂) by model year 1708. A linear rather than exponential ramp was chosen deliberately: an exponential continuation of the initial growth rate would have produced roughly 6140 ppm, while the linear ramp still yields approximately +0.3°C per century of warming.

Under RCP2.6 the AMOC recovers; under RCP4.5 and RCP8.5 it collapses around model years 2110 and 2060, at GMST anomalies of +2.2°C and +2.8°C. In stark contrast, the CO₂ ramp simulation maintains a stable AMOC throughout, with a GMST anomaly of +5.5°C over its final 50 years — exceeding even the RCP8.5 maximum. This single comparison constitutes the paper's central evidence against a universal warming threshold.

Mechanisms stabilising the AMOC under slow forcing

The analysis rests on water mass transformation (WMT) diagnostics. Under thermal wind balance assumptions, the adiabatic AMOC can be reconstructed from surface buoyancy fluxes between 40°N–65°N, yielding $\Psi_{\mathrm{NADW}}$, with near-zero values indicating that an adiabatic AMOC cannot be sustained. In the CO₂ ramp simulation, both the interior density level of maximum overturning ($\sigma_2^{\mathrm{max}}$) and the surface sinking region lighten coherently, so adiabatic pathways remain open. In the RCP4.5 and RCP8.5 simulations, by contrast, the surface waters lighten faster than the interior, closing these pathways — direct evidence that rate-dependent effects govern not only AMOC weakening but tipping itself.

Two dominant stabilising processes operate under slow warming. First, enhanced evaporation amplifies the Atlantic's net evaporative character, effectively imposing a negative freshwater flux forcing; the surface freshwater flux declines by 0.069 Sv per degree of warming ($R^2 = 0.98$ over 25-year windows). Second, retreating North Atlantic sea ice limits sea-ice insulation effects on deep convection and removes a meltwater source south of Greenland. After water mass transformation, the resulting salinity anomalies are exported in the NADW between 1000–3500 m depth, increasing $F_{\mathrm{ovS}}$ — the AMOC-induced freshwater transport at 34°S — whose larger values and reduced variance indicate greater stability under quasi-equilibrium conditions. Notably, Labrador Sea deep convection reactivates from model year ~770, producing a nonlinear ~3 Sv strengthening around model year 800 tied to sea-ice retreat.

A caveat applies here: the $F_{\mathrm{ovS}}$ stability indicator is only valid under quasi-equilibrium conditions, which do not hold for the RCP scenarios. The CO₂ ramp's mean warming rate of 0.03°C per decade is a factor of ten slower than RCP8.5 over the 21st century and than currently observed rates.

Corroboration across models

The salinifying NADW response appears in CESM's Hist/RCP4.5 run at $\overline{F_H} = 0.18$ Sv once warming exceeds ~+2.5°C, but not in Hist/RCP2.6 at $\overline{F_H}$0 Sv, suggesting the response scales with warming level rather than hosing strength. Among CMIP6 models under extended SSP1-2.6 and SSP2-4.5, most show increased NADW salinity at 1000–2000 m depth, though signals are small and IPSL-CM6A-LR freshens instead. The stabilising response cannot be verified under SSP5-8.5 because all available CMIP6 models collapse there. Inter-model differences in near-surface salinity at 34°S point to unexplored roles of atmospheric bridges and Agulhas Leakage transport.

Dynamical interpretation

An idealised 5-box ocean model captures the essential dynamics. Without temperature–freshwater coupling ($\overline{F_H}$1), raising the subpolar atmospheric temperature anomaly to +5°C shifts the saddle-node bifurcation from $\overline{F_H}$2 Sv to 0.342 Sv, so any $\overline{F_H}$3 above this value guarantees bifurcation-induced collapse regardless of rate. With coupling ($\overline{F_H}$4), however, the effective freshwater forcing declines as the ocean warms, and whether the system tracks the 'AMOC on' state depends on the ratio of atmospheric forcing timescale to ocean adjustment timescale. For $\overline{F_H}$5 Sv °C⁻¹, slow warming (+0.01°C yr⁻¹) permits tracking while fast warming (+0.05°C yr⁻¹) causes failure at the moving saddle-node. Critical warming rates are mapped as functions of $\overline{F_H}$6 and $\overline{F_H}$7, and remain qualitatively robust up to +10°C anomalies — demonstrating that forcing pathways exist under which the AMOC does not collapse even under extreme warming. Because the forcing is state-dependent, this constitutes a more complex rate-induced tipping scenario than the classical single-parameter case.

Limitations and open questions

Several limitations are acknowledged explicitly. The critical radiative forcing rate for AMOC collapse in the CESM could in principle be determined but requires computational resources beyond the scope of the study. The stabilising mechanism operates on timescales longer than present-day forcing increases, so warming thresholds may retain some utility on decadal timescales, though their usefulness diminishes as warming rates slow. The proposed conversion of a warming threshold into a warming-rate threshold is complicated by dependence of the critical rate on background climate state and on the strength of the stabilising mechanism. The CMIP6 corroboration suffers from low signal-to-noise ratios, limited scenario availability (only GISS-E2-1-G extends to SSP2-4.5 through 2500), and the absence of collapsing-model data under high emissions. Contributions from atmospheric moisture transport and Agulhas Leakage to inter-model spread are identified but not resolved.

Conclusion

This paper establishes, in a comprehensive coupled climate model, that AMOC tipping is governed by the radiative forcing path rather than by a unique global warming threshold: the same model sustains a stable AMOC at +5.5°C under slow forcing yet collapses at +2.2–2.8°C under faster scenarios. The underlying physics — coherent surface–interior density adjustment enabled by enhanced evaporation and reduced sea-ice cover — is corroborated qualitatively across CMIP6 models and rationalised dynamically in a box model exhibiting genuine rate-dependent tipping. The practical implication is that limiting the rate of radiative forcing, not only its cumulative magnitude, is critical for reducing near-term AMOC collapse risk; the authors estimate a critical warming rate near +0.29°C per decade, comparable to projected rates of +0.27 to +0.36°C per decade. Physics-based indicators such as $\overline{F_H}$8 and the surface buoyancy flux, whose sign change marks collapse onset, are proposed as more robust monitoring tools than warming-level thresholds.