Imperfect emergency brake: Can delayed Solar Radiation Modification revert AMOC and SPG weakening?

Solar Radiation Modification (SRM) is a collection of hitherto hypothetical methods that would reflect a small fraction of incoming solar radiation, thereby cooling the Earth and reducing the impact of greenhouse gas forcing, albeit imperfectly. The best-researched method so far is Stratospheric Aerosol Injection (SAI), which would work by injecting a reflective aerosol (e.g. sulphate) or a precursor gas (e.g. SO2) into the stratosphere. Previous studies (e.g., Tilmes et al, 2018, 2020, Xie et al., 2022) have shown that SAI and other SRM methods can reduce or even prevent Atlantic Meridional Overturning Circulation (AMOC) weakening. No dedicated study has however been done on the effect of SRM on the Subpolar Gyre (SPG). Also, most SRM modelling studies focus on present-day (2020) or at least speedy initialization of SRM. In reality, SRM might only begin many decades from now, if at all. In our study, we investigate whether delaying SRM will cause irreversible changes to the AMOC and the SGP. To this end we compare three scenarios in the CESM2 model: Control: An extreme warming scenario (RCP8.5) without SAI SAI2020: As Control, but keeping global mean surface temperature constant by means of SAI from 2020 onwards SAI2080: As Control, but starting SAI from 2080 such as to bring global mean surface temperature to 2020 levels and keeping it constant thereafter. These are extreme scenarios, not intended to represent plausible policy choices but meant to investigate whether irreversibility can occur in principle. We find that in Control AMOC weakens from 16 Sv in 2020 to 7 Sv in 2100, while in SAI2020, it only weakens to 12 Sv. In SAI2080, AMOC stops weakening after 2080, but does not recover (at least till 2100) to the strength it has in SAI2020. Thus, delayed SAI cannot quickly revert AMOC weakening, if at all. This has effects on the local climate, in particular overcooling around the North Atlantic, and even the interhemispheric temperature gradient. In addition, we find for Control, that deep convection (i.e. deep mixed layers in winter) ceases in the Labrador sea around 2050 and south of Iceland around 2070. Under SAI2020, deep convection remains active south of Iceland. Under SAI2080, deep convection does not recover by 2100. We conclude that SAI is not a perfect emergency brake for global warming: If action is delayed, changes in ocean circulation persist at least for several decades. However, we stress that other, including political, factors must be taken into account when considering (near-term) SAI, and that phasing out greenhouse gas emissions must remain the primary tool of climate policy.