Impact of the GeoMIP G1 sunshade geoengineering experiment on the Atlantic meridional overturning circulation

We analyze the multi-earth system model responses of ocean temperatures and the Atlantic Meridional Overturning Circulation (AMOC) under an idealized solar radiation management scenario (G1) from the Geoengineering Model Intercomparison Project. All models simulate warming of the northern North Atlantic relative to no geoengineering, despite geoengineering substantially offsetting the increases in mean global ocean temperatures. Increases in the temperature of the North Atlantic Ocean at the surface (∼0.25 K) and at a depth of 500 m (∼0.10 K) are mainly due to a 10 Wm−2 reduction of total heat flux from ocean to atmosphere. Although the AMOC is slightly reduced under the solar dimming scenario, G1, relative to piControl, it is about 37% stronger than under abrupt4 × CO2. The reduction of the AMOC under G1 is mainly a response to the heat flux change at the northern North Atlantic rather than to changes in the water flux and the wind stress. The AMOC transfers heat from tropics to high latitudes, helping to warm the high latitudes, and its strength is maintained under solar dimming rather than weakened by greenhouse gas forcing acting alone. Hence the relative reduction in high latitude ocean temperatures provided by solar radiation geoengineering, would tend to be counteracted by the correspondingly active AMOC circulation which furthermore transports warm surface waters towards the Greenland ice sheet, warming Arctic sea ice and permafrost.


Introduction
The Atlantic Meridional Overturning Circulation (AMOC) plays a critical role in the climate through its transport of heat and freshwater from the tropics to higher latitudes (Vellinga and Wood 2002), which is particularly effective at warming the North Atlantic and transporting heat which melts sea ice, reduces snow cover and melts the floating parts of terrestrial glacier systems. The process involves warm saline surface water flowing northward to high latitudes where it cools, sinks and returns southward at depth. Observational evidence regarding the strength of the AMOC is limited: the mean over 2004-2012 is 17.2 Sv with 10 day filtered root mean square variability of 4.6 Sv (McCarthy et al 2015). The AMOC is a key means by which heat is sequestered into the ocean's interior and thus modulates the trajectory of climate change (Buckley and Marshall 2016). However, simulated AMOC varies widely between climate models (Gregory et al 2005, Intergovernmental Panel on Climate Change IPCC 2013b) and the mechanism driving variability is not well known (Buckley and Marshall 2016), which results in a wide range of results for current and projected changes in the AMOC. Despite this, numerical model projections do robustly suggest that the AMOC will weaken over the 21st century Original content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence.
Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. (Cheng et al 2013, Intergovernmental Panel on Climate Change IPCC 2013a.
A reduction in the density of the surface waters of the northern North Atlantic can inhibit the sinking of surface waters and deep water formation, weakening the AMOC. Contributing factors are increased freshwater flux into the northern North Atlantic via reduced sea ice growth, increases in precipitation minus evaporation or increased run off from land; surface warming of the northern North Atlantic; and reduced surface wind stress mitigating oceanic mixing (Mikolajewicz and Voss 2000). Under greenhouse gas warming increased fresh water input into the polar and sub-polar seas is projected due to increased precipitation at high latitudes (Dai et al 2001, Lehner et al 2012, Intergovernmental Panel on Climate Change IPCC 2013c.
Geoengineering has been proposed as a way of mitigating anthropogenic climate change, especially increasing global mean temperatures (Royal Society 2009). Reducing incoming solar radiation almost immediately leads to a drop in surface temperatures (e.g. Robock et al 2009), though the cooling is not homogenous. Although there are regional differences in the efficacy of geoengineering, the temperature anomalies relative to the present day are much smaller in magnitude than under purely greenhouse gas forcing scenarios (Yu et al 2015, Kravitz et al 2014. To date, little research on the response of ocean te et al mperatures or the oceanic circulation to geoengineering has been published. Cao et al (2016) used the HadCM3L model to conduct a uniformly reduced solar irradiance geoengineering simulation on the millennial time scale, and found out the AMOC under geoengineering remains closer to that of the control preindustrial simulation than it would under greenhouse gas forcing alone. McCusker et al (2015) used a climate model to investigate the impact of stratospheric sulphate aerosol geoengineering on Antarctica, and concluded that geoengineering would not reduce upwelling of warm water near actively retreating glacial margins, such as Pine Island Glacier, and therefore is not successful at counteracting the trend of increased ice mass loss from Antarctica. A global climate model study by Tilmes et al (2014) showed that just reducing the solar radiation reaching the Arctic would still lead to a slowdown of the AMOC. Muthers et al (2016) showed the important role of chemistry-climate interaction in the prediction of the AMOC strength under solar radiation reduction scenarios. Here we make use of multi-model ensemble data from the Geoengineering Model Intercomparison Project (GeoMIP) experiments (Kravitz et al 2011) to investigate changes in the AMOC as a result of a solar radiation management (SRM) scenario.
In this study we investigate the response of the ocean to a reduction in solar irradiance (G1, see the description in the experiments and data section) to mitigate the warming that arises from increased atmospheric concentrations of greenhouse gases. We address the issues of how the G1 scenario alter the surface and subsurface ocean at global and regional scales and, in particular, we explore how it impacts the AMOC and how the experiment helps to elucidate the mechanisms of AMOC change under greenhouse gas forcing.

Experiments and data
The GeoMIP experiments are built on the CMIP5 framework (Taylor et al 2012), including the G1 scenario that are used in this study (figure S1 available at stacks.iop.org/ERL/12/034009/mmedia), (Kravitz et al 2011). G1 is a highly idealized experiment, facilitating analysis of dominant radiative effects and the climate system responses. G1 is based on the CMIP5 abrupt4 Â CO 2 experiment and starts from a stable preindustrial climate (the CMIP5 experiment piControl; Taylor et al 2012). It imposes two large counteracting step function forcings: a quadrupling of atmospheric CO 2 (as done in abrupt4 Â CO 2 ), and a reduction in incoming solar radiation.
In this study we analyze monthly output from 7 climate models ( (table S1), to determine changes in ocean temperatures with depth, the AMOC, ocean heat transport and atmospheric wind stresses. However, some models lack heat flux and water flux data (table S2). Since there is a sudden change in forcing between that specified in piControl and that for abrupt4 Â CO 2 (figure S1), there will be significant transient effects over the first 10 years of the simulation. The fast response to abrupt climate change occurs in only a few years while oceanic and some land responses will be considerably longer than the 50 year period of the G1 forcing (e.g. Kravitz et al 2013, Cao et al 2016. Previous studies have removed the fast response transient by considering only the last 40 years of simulation results, and we follow that approach here. Although oceans will not have reached equilibrium during the 50 years of simulations, Cao et al (2016) note that the climate response over a 1000 year period with HadCM3L, is much more variable on century time scales under abrupt4 Â CO 2 than G1. Therefore, all maps and zonal averages here are calculated using years 11-50 of the geoengineering simulations, thus excluding the first 10 years of both G1 and abrupt4 Â CO 2 . We compare G1 with abrupt4 ÂCO 2 and piControl.

Ocean temperature response
The change in global average ocean surface temperature under G1 relative to piControl is between À0.25 K and 0.23 K for the seven models, whereas warming of Environ. Res. Lett. 12 (2017) 034009 between 2.33 K and 3.84 K occurs in the abrupt4 ÂCO 2 simulations (figures S2(a), (b)). An ensemble model cooling of almost 3 K of the surface under G1 relative to abrupt4 Â CO 2 demonstrates the ability of solar radiation management to offset ocean surface warming. Spatial patterns of sea surface temperature (SST) anomalies under G1 display regional differences ranging from a cooling of around 0.3 K in the tropics between 30°S to 30°N to a warming of up to 0.2 K in the areas between 30°N and 70°N, and 40°S and 70°S (figure 1 (b)). A warming in the northern North Atlantic under G1 emerges in the convection zones in the Labrador Sea, which would make the surface waters lighter and hence increase their stability, inducing a weakening of the AMOC (Hu et al 2004). Although a warming in the midlatitude areas occurs under G1, it is much smaller than the 2.5 K warming under abrupt4 Â CO 2 ( figure 1(a)). G1 is therefore successful at moderating SST increases in these areas.
The largest subsurface (500 m) warming (up to 0.13 K) under G1 occurs south and west of Greenland (figure 1(d)), with all models in agreement on the warming trend. Less pronounced warming (0.10 K) is projected for the subsurface (500 m) ocean layers around Antarctica (figure 1(d)). Nevertheless, this means almost all waters that can access the cavities beneath Antarctic ice shelves and around ice sheet margins will be interacting with a warmer ocean than in the control simulation, though the temperature rises are much smaller than under abrupt4 Â CO 2 . Ice sheet modelling and palaeoclimate studies demonstrate that the stability of the Antarctic ice sheet is extremely sensitive to even small increases in ocean temperatures (e.g. Golledge

AMOC and heat transport response
The AMOC index here is defined as the annual-mean maximum volume transport streamfunction at 30°in the North Atlantic. The ensemble mean AMOC value which is about 18.8 Sv for the 7 piControl simulations (figure 2(a)) is consistent with the observed AMOC amplitude (17.2 ± 4.6) Sv measured by the RAPID-MOCHA array over 2004(McCarthy et al 2015. However, only 3 of the 7 models we use have mean piControl AMOC indices that are within the range of the observed AMOC. The models common to the Cheng et al (2013)   Under the abrupt4 Â CO 2 scenario (figures 2(b) and (c); figure S3(a)), all models predict a weakening of the AMOC, ranging from 5.0 Sv to 13.5 Sv by the 50th year of the simulation, with a mean weakening of 8.1 Sv ( figure 2(b)). Previous studies show that models with stronger AMOCs in their control run exhibit larger weakening in a warming world (Gregory and Tailleux 2011). We therefore also plot relative changes of the AMOC under abrupt4 ÂCO 2 and G1 ( figure 2(c)). According to this metric, BNU-ESM is no longer an outlier. Relative reductions in the AMOC in the 50th simulation year average 44%, and range from 26% to 60%, in reasonable agreement with the Gregory et al (2005) estimates of 10% to 50% declines over a 140 year simulation during which the CO 2 concentration quadruples. Compared with abrupt4 Â CO 2 , G1 successfully mitigates weakening of the AMOC (figures 2(b) and (c); figure S3(b)). Ensemble mean anomalies in the 50th year of the simulation are 1.3 Sv (À7%) in G1, compared with 8.1 Sv (À44%) in abrupt4 Â CO 2 . This difference can also be clearly seen from the northward heat transport changes in the North Atlantic ( figure S4). In experiment G1, heat transported northward is not reduced as in abrupt4 Â CO 2 , there is still strong heat transport into the North Atlantic which keeps the northern North Atlantic warm.
Reductions in the AMOC with increasing global temperatures reduce poleward heat transport south of 60°N (figure S4). The net reduction of heat northwards across 30°N is about 0.25 PW (30% relative to piControl) in abrupt4 Â CO 2 and about 0.05 PW (8% relative to piControl) in G1. The net reduction of the heat transport south of 60°N in both abrupt4 Â CO 2 and G1 is consistent with the warming of the subsurface between 30°N and 60°N in the Atlantic Basin (figure 1). Between 60°N to 70°N, there is a slight increase of northward heat transport to the high latitudes under abrupt4 Â CO 2 (figure S4(a)) because of the increased flow of North Atlantic water across this latitude, as seen in a previous study (Hu et al 2004). While under G1, the northward heat transport returns to piControl levels.

Mechanisms for AMOC changes
The AMOC is primarily sensitive to changes in three different air-sea interactions which act on different time scales: heat flux, freshwater flux, and wind stress. At short time scales (months to seasonal), the wind stress changes can be the dominant contribution to the To clarify the causes of AMOC decreases under abrupt4 Â CO 2 and G1, we next consider the change in the heat flux through the ocean surface at the three deep convection regions in the northern North Atlantic which are located at the Labrador Sea, the Irminger Sea basin and the Nordic Seas ( figure 4). A previous study showed that the models we use here has the same specific convection regions as observed (Heuzé et al 2015). Under abrupt4 Â CO 2 , the heat flux from the ocean to the atmosphere in all the deep convection regions decreases by up to 70 Wm −2 ( figure 4(a)), which is about 75% of the mean piControl heat flux, demonstrating a tremendous reduction in heat lost from ocean to atmosphere. Figure S5 shows a reduction of the temperature difference between air and sea surface in the northern North Atlantic, which would induce a reduction in sensible heat loss from ocean to atmosphere under abrupt4 Â CO 2 .
Meanwhile, the surface freshwater flux into the northern North Atlantic increases under abrupt4 Â CO 2 ( figure 5(a)). Here the water flux includes the contribution from precipitation minus evaporation, rivers and sea ice thermodynamics. According to Kravitz et al (2013), precipitation minus evaporation in the region increases by about 0.4-0.8 mm day À1 under abrupt4 Â CO 2 . Additionally, the March sea ice concentration for abrupt4 ÂCO 2 relative to piControl reduces 20% to 30% at the Nordic Seas (Moore et al 2014), both of which lead to lower surface water densities.
Thus reduced heat loss, aligned with increased freshwater input in the northern North Atlantic, would act to reduce the strength of the AMOC. Despite a reduction in the strength of the AMOC, nowhere is there a decreasing sea surface or nearsurface temperatures under abrupt4 Â CO 2 , illustrating the dominance of greenhouse gas radiative forcing over heat transport by the AMOC.   The decline of the AMOC under G1 is relatively small but still recognizable. There is almost no change in the total water flux (figure 5(b)) in the northern North Atlantic under G1, but there is still a weakening of 1.3 Sv in the AMOC. To show the roles of temperature and salinity in manipulating the change of AMOC under G1, we calculate the contribution of each to the density at the northern North Atlantic. As shown in figure S6, the model ensemble temperature anomaly under G1 relative to piControl is 0.45 ± 0.01 K (uncertainty is the standard error of the mean) and the salinity anomaly is 0.03 ± 0.01 psu. Neither of the anomalies have a significant time trend. The salinity and temperature at the northern North Atlantic are 33.15 psu and 3.19°under piControl which results a density of 1026.39 kg m À3 there. A temperature rise of 0.45 K under G1 relative to piControl without the salinity change, induces a reduction of 0.04 kg m À3 in density. A salinity rise of 0.03 psu under G1 relative to piControl without the temperature change, induces an increase of 0.03 kg m À3 in density. The density under G1 accounting for both temperature and salinity results in a reduction of 0.02 kg m À3 in density compared with the piControl value. This hints that the reduction in the AMOC is driven mostly by the warming of the sea surface in the northern North Atlantic. Figure 4(b) shows a 10 Wm À2 increase of heat flux into the ocean in the northern North Atlantic under G1 demonstrating decreasing heat loss from ocean to atmosphere. Meanwhile figure S5b, d, f show reductions of about 0.5 K, 0.2 K and 0.5 K in the sea-air temperature contrast in the Labrador Sea, the Irminger Sea and the Nordic seas respectively, which would support the decreased heat loss from ocean to atmosphere under G1. On the contrary, the small increase of salinity prevents a larger reduction of the AMOC under G1.
Sea ice growth during winter is a significant influence on the salinity of water mass in the deep convection regions via the salt rejection mechanism (Muthers et al 2016). The March sea ice concentration anomaly for G1 relative to piControl features an increase of 5% to 10% in the Nordic Seas, but a reduction of 5% to 10% in the Labrador Sea. At least six out of the eight models the study used agreed on the signs of the changes (Moore et al 2014); While for the Irminger Sea, there were no clear changes These mixed results indicate possible sea ice induced strengthening of deep convection in the Nordic Sea but a weakening in the Labrador Sea, producing little overall change, consistent with figure 5(b).

Discussion and conclusions
Substantial weakening of the AMOC under abrupt4 Â CO 2 and associated warming of both the Atlantic and Southern Oceans below a depth of 1000 m can clearly be mitigated under G1. However, surface and subsurface warming of the northern North Atlantic and Southern Ocean remains under G1. This predicted warming of the ocean surface and subsurface would increase ice loss from Greenland and Antarctic ice sheets via the interaction of warm water and floating ice. However, this effect would be much stronger under greenhouse gas forcing alone.
Ocean surface warming in the northern North Atlantic in G1 can be explained by changes in heat exchange between ocean surface and atmosphere. Substantial warming of the atmosphere above the northern North Atlantic is simulated under the G1 scenario (Kravitz et al 2013). However, the ocean warms less than the atmosphere, reducing air-sea temperature contrast and inhibiting ocean-atmosphere heat exchange in the North Atlantic (figure 4). Thus reduced heat loss from ocean to atmosphere further induces reduced density of the northern North Atlantic, and suppresses the sinking of surface waters under abrupt4 Â CO 2 and G1. Changes in freshwater flux are an essential requirement for large changes in  (50°N-60°N, 20°W-40°W) of (a) abrupt4 Â CO 2 , (b) G1 relative to piControl. The total water flux includes the contributions from precipitation, evaporation, rivers, icebergs and sea ice thermodynamics.
Environ. Res. Lett. 12 (2017) 034009 the AMOC. Precipitation minus evaporation increases by 0.4-0.8 mm day À1 and the total water flux increases by 0.02 Sv under abrupt4 Â CO 2 but is unchanged from piControl under G1 meaning that the main cause of the small AMOC change under G1 is the change in the heat flux at the ocean surface. The AMOC effectively warms the high latitude climate as it transfers heat from south to north. A weakened AMOC under greenhouse gas forcing tends to reduce high latitude surface warming, whereas a relatively less weakened AMOC under the geoengineering discussed here would moderate the cooling effects.
As the experiment of G1 is only 50 years in duration, the behavior of the AMOC over the longer term can be suggested by the 1000 year geoengineering simulation using the HadCM3L model (Cao et al 2016), where AMOC variability under G1 type geoengineering is similar as simulated under piControl, and much less than under abrupt4 Â CO 2 . As the cooling of the polar regions continues to be smaller than the mid-latitudes under G1, the surface density gradient will become larger between the northern North Atlantic and the southern North Atlantic which could strengthening the current flowing from south to north. Additionally, as the temperature gradient becomes smaller between sea surface and the air at the northern North Atlantic as the ocean warms up slowly, the ocean will lose more heat at the convection zones which will strengthen the deep convection. Both of the effects mentioned above would strengthen the AMOC in a longer term. But whether the mechanism discussed here is correct or not needs further modeling studies.