Increased risk of near term global warming due to a recent AMOC weakening

Some of the new generation CMIP6 models are characterised by a strong temperature increase in response to increasing greenhouse gases concentration1. At first glance, these models seem less consistent with the temperature warming observed over the last decades. Here, we investigate this issue through the prism of low-frequency internal variability by comparing with observations an ensemble of 32 historical simulations performed with the IPSL-CM6A-LR model, characterized by a rather large climate sensitivity. We show that members with the smallest rates of global warming over the past 6-7 decades are also those with a large internally-driven weakening of the Atlantic Meridional Overturning Circulation (AMOC). This subset of members also matches several AMOC observational fingerprints, which are in line with such a weakening. This suggests that internal variability from the Atlantic Ocean may have dampened the magnitude of global warming over the historical era. Taking into account this AMOC weakening over the past decades means that it will be harder to avoid crossing the 2 °C warming threshold.

This Supplementary Information provides additional analysis on the attribution of the AMOC weakening to internal variability (Supplementary Note 1) and the relationship between the GSAT and the AMOC over the historical period using three additional ensembles of historical simulations (Supplementary Note 2). Supplementary Figures 1 to 9 and Supplementary  Table 1 are also displayed.

Supplementary Note 1. Estimation of the relative influence of internal variability and forced response in the AMOC weakening
In order to investigate and try to attribute the relative influence of internal variability and forced response in the AMOC weakening, we regress the AMOC onto its forced and unforced components (see Methods). This indicates to which extent one or the other component of the AMOC covaries as a function of the total AMOC variations ( Supplementary Fig. 1). For all the members, the maximum regression for the forced response is 0.45 Sv Sv -1 , with a lot of members having small regression coefficients (mean = 0.15 Sv Sv -1 ). Therefore, internal climate variability appears to be the main driver of the AMOC variability in the IPSL-EHS over the 1940-2016 period although there appears to be a small forced component in the ensemble mean towards the end of the historical period.

Supplementary Figure 1 | Forced and unforced components to the AMOC.
Linear regression coefficient (Sv Sv -1 ) between the time evolution of the low-pass filtered AMOC strength anomaly in each member of the IPSL_EHS and the forced response of the low-pass filtered AMOC strength anomaly (red), as well as with the internal variability of the low-pass filtered AMOC strength anomaly (blue). The forced response is defined as the ensemble mean and the internal variability as the difference between the full AMOC time series in each member and the forced response. All time series are considered over the 1940-2016 period. The error bars represent the 95% confidence intervals of the regression coefficient.

Supplementary Note 2. Test of the relationship between the GSAT and the AMOC trends over the recent decades in three other large ensembles
In order to test the relationship found in the IPSL ensemble, we evaluated three other ensembles performed with models with different climate sensitivities (Supplementary Figure  2). The number of models analysed is limited as we can only consider models with publication on the ESGF of at least 30 historical members and the relevant variables needed to estimate the AMOC. The MPI Grand Ensemble 1 (MPI-GE) was performed using the MPI-ESM1.1 model, characterized by a lower low-frequency internal climate variability and a lower effective ECS (2.8 K) than IPSL-CM6A-LR. The two other ensembles available were performed with the CNRM-CM6-1 2 and the CanESM5 3 models, both characterized by relatively high effective ECS values of 4.9 K and 5,6 K, respectively, estimated from an abrupt4xCO 2 experiment.
A positive relationship between the AMOC and GSAT trends is found in the MPI-GE, with the closest members to the observed 1940-2014 GSAT trend exhibiting a negative AMOC trend ( Supplementary Fig. 2a). In that respect the MPI-GE is consistent with the IPSL-CM6A-LR model ensemble. The fact that a large fraction of the MPI-GE members have larger GSAT trends than observed despite a low climate sensitivity (effective ECS of 2.8 K) is surprising at first sight. However it should be noted that MPI-ESM1.1 has a fairly weak aerosol forcing of about -0.5 Wm -2 according to Mauritsen et al. (2019) 4 . This is in contrast with best estimate of -0.9 Wm -2 in chapter 7 of the IPCC Fifth Assessment Report 5 and ranges of -1.6 to -0.6 Wm -2 (68% likelihood) and -2.0 to -0.4 Wm -2 (90% likelihood) in the authoritative review by Bellouin et al. (2019) 6 . Since then there have been several studies providing observational evidence for a more negative aerosol effective radiative forcing 7-10 . The MPI-ESM1.1 aerosol forcing is also outside the range of aerosol ERF for the subset of CMIP6 models reported in Zanis et al. (2020) 11 . We can thus conclude that matching both the GSAT and AMOC trends in the MPI-GE is only possible because the aerosol forcing is on the low side. In contrast we estimated the aerosol IRF and ERF of IPSL-CM6A-LR to be -0.7 and -0.6 Wm -2 , which is not very strong for a large ECS model. Had the MP-ESM1.1 a stronger (more negative) aerosol ERF, it would not have been possible for the MPI-GE to match both the observed GSAT and AMOC trends. The analysis of the MPI-GE therefore does not invalidate our hypothesis.
The CNRM ensemble is characterized by a very strong positive relationship between the AMOC and the GSAT trends since 1940 and a larger low-frequency natural variability than the other ensembles ( Supplementary Fig. 2b). Some historical members match both the GSAT and the AMOC 1940-2014 trends, and the members with the lowest RMSE over the 1900-2014 are among these members. The CNRM-CM6-1 model could have been compatible with an even lower observed GSAT trend despite its rather large ECS of 4.9 K. This model thus supports our hypothesis with IPSL-CM6A-LR.
Finally, for the CanESM5 ensemble, the GSAT and AMOC trends over the 1940-2014 period show a weak relationship (Supplementary Fig. 2c). None of the historical simulations is close to the observations in terms of the GSAT trend. Given that this model has little natural variability, it does not invalidate our hypothesis though not much can be concluded.