Corresponding author: A. Jahn, National Center for Atmospheric Research, 1850 Table Mesa Drive, Boulder, CO 80305, USA. (email@example.com)
 Using CCSM4 climate simulations for 1850–2300 with four different future forcing scenarios, we show that the maximum strength of the Atlantic meridional overturning circulation (MOC) decreases proportionally to the applied CO2 forcing. This weakening of the overturning is caused by a reduction or shut down of North Atlantic (NA) deep convection due to a surface freshening. In the Labrador Sea, the surface freshening is caused by strongly increased liquid freshwater exports from the Arctic, which are largely due to the decrease in the Arctic sea ice cover. In the strongest forcing scenario (RCP8.5), the Arctic becomes summer ice-free by the end of the 21st century and year-round ice-free by the end of the 23rd century. As a result of the associated freshening, all NA deep convection ceases by 2145, which leads to a 72% (18 Sv) decrease of the MOC strength by the end of the simulation in 2300.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 To explore the long-term climate impacts of the different emission scenarios (called representative concentration pathways or RCPs) used in the Coupled Model Intercomparison Project Phase 5 (CMIP5), some modeling centers extended the CMIP5 simulations to the year 2300. The CO2 forcing used in these simulations is shown in Figure 1a (see Meinshausen et al. , for details on the forcing of these RCP extension simulations). In several of the CMIP5 models, the strength of the MOC in these RCP simulations shows a reduction in response to the increasing CO2 forcing, but no abrupt shut down [Weaver et al., 2012] Similar results are found in the Community Climate System model version 4 (CCSM4), which shows a reduction of the MOC in all RCPs, with a reduction of up to 72% by 2300 in the strongest emission scenario, the RCP8.5 [Meehl et al., 2012]. This reduction occurs at atmospheric CO2 levels of 1962 ppm, with globally averaged surface temperatures 8°C above 1986–2000 levels and a year-round ice-free Arctic Ocean [see Figures 9, 15, and 19 in Meehl et al., 2012]. Here we investigate the reasons for the simulated MOC reductions and link it to the simulated changes in the Arctic sea ice cover in the CCSM4 RCP simulations and their extensions to 2300, demonstrating the important potential downstream impacts of changes in the Arctic Ocean under the CMIP5 forcing scenarios.
 We use CMIP5 ensemble simulations with the CCSM4 model, which is a configuration of the Community Earth System Model (CESM) family of models [Gent et al., 2011]. CCSM4 is a fully coupled climate model with a nominal resolution of 0.9° × 1.25° in the atmosphere and land models and 1° in the ocean and sea ice models. Its components and simulations have been described in a large collection of articles, many of them contained in a special collection in the Journal of Climate (http://journals.ametsoc.org/page/CCSM4/CESM1). In particular, it has been shown that the CCSM4's representation of late 20th century Arctic atmospheric [de Boer et al., 2012] and sea ice processes [Jahn et al., 2012; Kay et al., 2011] compares reasonably well with observations. Note that the CCSM4 simulations do not include an active ice sheet model, so impacts of ice-sheet melting in Greenland are not accounted for.
 The simulations used here consist of six ensemble members for the historical period (1850–2005), six ensemble members for 2005–2100 for each RCP (2.6, 4.5, 6.0, 8.5), and one ensemble member for each extended RCP scenario (for 2101–2300). The global climate in these CCSM4 simulations has been analyzed by Meehl et al. . Note that for RCP8.5, data are missing due to a damaged data storage tape between 2235 and 2241 (shown as gap in the graphs). Freshwater (FW) fluxes and storage are calculated relative to a mean salinity of 34.8 practical salinity unit (psu), and details on the FW calculations and the Arctic FW budget in the CCSM4 20th century simulations can be found in Jahn et al. . The Atlantic MOC index is calculated as the maximum of the annual-mean Atlantic stream function below 500 m in the northern Hemisphere, including both the Eulerian mean and the additional components. The sea-ice extent is calculated as the area covered by at least 15% ice concentration.
3 Arctic Changes
3.1 Arctic Sea Ice
 The Arctic sea ice extent decreases proportionally to the radiative forcing in all RCPs (once interannual variability is smoothed out through a 20 year running mean, see Figure 1d). September ice-free conditions of less than 1 million km2 are simulated at the end of the 21st century in RCP8.5 and in the early 22nd century in RCP6.0 (see Figure 1b). At less than 1 million km2 of sea ice extent, the Arctic Ocean is ice-free for all practical purposes, with the remaining sea ice located north of Greenland and in the narrow channels of the CAA. Wintertime sea ice extents also decline over the 21st century in all RCP scenarios, but only RCP8.5 with its increasing radiative forcing past 2150 shows a drastic decline in the winter sea ice extent, leading to year-round ice-free conditions by the end of the 23rd century (Figure 1c). We find that the climatic evolution of sea ice and many other properties are largely determined by the CO2 level, not the trajectory by which that level was reached. As such, a similar transient climate response is seen for all RCPs when assessing changes as function of the increasing CO2 level. Once the CO2 levels stabilize or begin to decline, the climate begins to adjust to the new stable conditions and departs from the transient climate state proportional to the CO2 level. As our focus is mainly on the transient response of the climate to the increasing CO2 forcing in the CMIP5 scenarios, we focus on results from the RCP8.5 scenario runs in the following, as these simulations prescribe CO2 levels that encompass and far exceed the CO2 levels in the other scenarios.
3.2 Arctic Freshwater Export
 The large simulated changes in the Arctic sea ice extent and volume have a wide range of impacts, but here we concentrate on the resulting changes in the Arctic FW export and the downstream impacts on the deep water formation (section 4). As shown in Figures 1e–1g, the impact on the simulated FW export to the NA through the two main pathways (Fram Strait and CAA) is a shift toward more and more FW export in the liquid phase, as well as a general increase in the total FW export. The simulated phase change of the FW export is important for downstream effects, as the FW in liquid and solid form reaches different regions (see section 4).
 The large increase in the liquid FW export in RCP8.5 reflects a significant freshening of the outflow between the early 21st century and the early 23rd century (by 2.5 psu in Fram Strait and 4.5 psu in the CAA). Between 2005 and 2200, this freshening of the Arctic Ocean (Figure 1h) is caused by increased sea ice melt within the Arctic and reduced sea-ice formation and export (53%), as well as by increased FW input from rivers (28%), an increase in the FW import through Bering Strait (16%), and increased net precipitation (2%) due to a generally enhanced hydrological cycle in a warmer climate.
 While the salinity of the outflow decreases, the volume of the liquid FW export through the CAA and Fram Strait in RCP8.5 declines by up to 50% during the late 21st and the mid 22nd century, respectively. For the CAA, the impact of the decrease in the volume export can be seen around 2100 as an intermittent reduction in the RCP8.5 liquid FW export in Figure 1g. It is caused by the shutdown of the deep convection in the Labrador Sea at this time (see section 4) and the associated rise in sea surface height in the Labrador Sea, which reduces the sea surface height gradient to the Arctic that drives this export (see Jahn et al.  and Houssais and Herbaut  for details of this mechanism). This decrease in the CAA liquid FW export leads to a sharp increase in the Arctic liquid FW storage after 2100 (Figure 1h). For Fram Strait, the volume flux decrease is fully compensated by the freshening of the outflow, so it cannot be seen in Figure 1e. It occurs around 2145, after the shutdown of the deep convection in the Nordic Seas, and is caused by the associated sea surface height gradient change across the East Greenland current after deep convection ceases (see Jahn et al. , for details of this mechanism).
4 Impacts on Deep Convection and MOC
4.1 Deep Convection Impacts
 The large increase in the liquid FW export from the Arctic at the end of the 21st century in RCP8.5 leads to a shutdown of February–April (FMA) deep convection in the Labrador Sea deep convection region at the start of the 22nd century (shown as white box in Figure 2a, maximum depth of deep convection is shown in Figure 2b). Before a complete shutdown of Labrador Sea deep convection occurs in RCP8.5, several abrupt reductions of deep convection occur in all six ensemble members of RCP8.5 at the end of the 21st century. They occur in response to decreased sea surface salinity (SSS) in the same region (see Figures 3b and 3c), with a correlation between the maximum depth of deep convection and the SSS in the same region of between 0.90 and 0.95 in the different ensemble members (p > 0.95, see Figure 3d). We find that deep convection always ceases when the FMA SSS drops below 33.8–34 psu (see Figures 3b, 3c, and 4e), while the FMA sea surface temperatures during these abrupt reductions of deep convection vary between 1°C and 5°C. We find that the Labrador Sea SSS changes are controlled to a large extent by the liquid FW export through Fram Strait, which affects the Labrador Sea SSS with a lag of 1 year. This is shown by a lagged-correlation analysis with correlations at 1 year lag between −0.59 and −0.87 for the different ensemble members (p > 0.95; see Figure 3d). The solid Fram Strait FW export does not have a strong impact on Labrador Sea SSS, as most of it does not reach the interior of the Labrador Sea.
 The FW export through the channels of the CAA does not show an impact on the SSS or deep convection in the Labrador Sea (not shown). This is in line with results from other model studies that showed that the FW export from the CAA does not reach the interior of the Labrador Sea [e.g., Myers, 2005; Houssais and Herbaut, 2011]. We do, however, find a positive correlation of up to 0.77 (p > 0.95) between the CAA FW export and the Labrador Sea deep convection at zero-year lag. This correlation is caused by the impact of sea surface height variations in the Labrador Sea on the volume flux through the CAA, as described in section 3.
 Deep convection south of Iceland and in the Nordic Seas shuts down during the first part of the 22nd century in RCP8.5, but shows no large changes in the other RCPs (Figures 2c and 2d). In RCP8.5, the reduction of deep convection to below 200 m again occurs when the local SSS drops below 33.8–34 psu (see Figure 4), as already seen in the Labrador Sea. The deep convection south of Iceland stops first (around 2120), followed by the deep convection in the Nordic Seas about 25 years later (around 2145, see Figures 2c, 2d, and 4).
 In contrast to the Labrador Sea, there is no significant correlation between the SSS and the FW export from the Arctic. Instead, the surface freshening occurs as a result of a decreased northward flux of salty Atlantic water. This is due to the changes in the NA ocean circulation after the Labrador Sea deep water formation (and later the deep water formation south of Iceland) stops and the MOC index weakens (Figure 2e). This reduced northward flux of salty water is consistent with previous work that describes the impact of changes in the subpolar gyre circulation on the inflow into the Nordic Seas [Hátún et al., 2005]. It is also consistent with the general shift toward a more positive Northern Annular Mode during the course of the RCP8.5 simulation [Meehl et al., 2012], which has been linked to a freshening in the Nordic Seas due to an intensification of the subpolar gyre [Richter and Maus, 2011].
4.2 MOC Impacts
 As the MOC represents the integrated effects of deep convection from different sites and of the large-scale ocean circulation in general, the Atlantic MOC index shows no abrupt changes, despite the abrupt shutdown of deep convection in some of the NA deep convection sites (Figure 2e). Instead it gradually weakens to 7 Sv (28% of its 20th century value of 25 Sv) by 2200, as deep convection in all three NA convection sites ceases and the MOC spins down. This decrease of the MOC strength is again proportional to the transient CO2 forcing once interannual variability is smoothed out (Figure 2f). A reduction in MOC strength is therefore observed for all RCPs while the CO2 increases. In RCP2.6, RCP4.5, and RCP6.0, the MOC index recovers within one to three decades after the CO2 forcing decreases or stabilizes and the salinity anomalies are flushed out of the deep convection regions. In RCP8.5, however, the stabilization in 2250 does not lead to a similarly fast recovery, as the water column in the deep convection regions is too stable to erode quickly.
 The results from the RCP ensemble simulations with the CCSM4 and their single-ensemble member extensions to 2300 show that the large simulated changes in the Arctic sea ice have impacts on the NA deep convection and the strength of the MOC, through changes in the Arctic FW export. We find that the response of the simulated sea ice cover, FW export, Labrador Sea deep convection, and Atlantic MOC index is proportional to the CO2 forcing in the RCP simulations (until stabilization of the CO2 forcing), with similar values obtained across the different scenarios when they reach the same CO2 level. This indicates that the rate of CO2 changes does not strongly influence the transient climate response. We therefore focused our analysis on the strongest forcing scenario, the RCP8.5 simulation. It shows a summer ice-free Arctic by the end of the 21st century and a year-round ice-free Arctic by the end of the 23rd century. The associated freshening of the Arctic Ocean and the loss of the Arctic sea ice cover lead to an increased liquid FW export, while the solid FW export decreases strongly. The associated freshening of the Labrador Sea causes a shutdown of Labrador Sea deep convection by the end of the 21st century. The deep convection in the other NA deep convection sites ceases by 2145, leading to a slow reduction of the Atlantic MOC to 7 Sv (28% of its 20th century value). The resulting decrease of the northward oceanic heat transport causes a 2°C cooling over the NA between the end of the 21st and the end of the 22nd century [Figure 11 in Meehl et al., 2012]. A net cooling over land in northern Europe is not simulated in RCP8.5, due to the very strong global warming simulated in the extended RCP8.5 simulation, which offsets most of the cooling associated with the reduced ocean heat transport to high latitudes. While these results are from only one model, a large reduction of the MOC index and the absence of abrupt changes in the MOC are seen in RCP8.5 extension simulations with at least two other CMIP5 models [Weaver et al., 2012].
 We thank G. Strand for making the RCP extension output available. Helpful comments from P. Gent (NCAR) and two anonymous reviewers are appreciated. A. Jahn was supported by a postdoctoral fellowship from ASP at NCAR. NCAR is sponsored by NSF. The CESM project is supported by NSF and BER at DOE. Computing resources were provided by CSL at NCAR's CISL, sponsored by NSF and other agencies. The extended RCP simulations were run at NCCS at ORNL and at NERSC, which are supported by BER at DOE under contracts DE-AC05-00OR22725 and DE-AC02-05CH11231, respectively. Figures were plotted using NCL (http://dx.doi.org/10.5065/D6WD3XH5).