An intermediate complexity climate model is used to simulate the collapse of the Barents Ice Sheet during Marine Isotope Stage 6 (MIS 6; 140 ka B.P) with the purpose of investigating whether a mass input of freshwater from the collapse could have affected the convection and deep water formation in the North Atlantic Ocean. Further experiments used a coupled dynamic and thermodynamic iceberg model to determine the effects of deep-draft icebergs, rather than freshwater alone, on the ocean circulation. The results predict that the collapse of the Barents Ice Sheet had a significant impact on the meridional overturning circulation in both the Atlantic and Pacific oceans. Freshwater fluxes have more of an impact on the Atlantic overturning circulation during the actual release period compared to icebergs, but the bergs induce effects over longer time scales even after the pulse is removed. Freshwater fluxes of 0.15 sverdrup (Sv) and iceberg surges of 0.1 Sv trigger significant changes in the global patterns, particularly in the North Pacific where there is strengthening of the overturning circulation at the expense of that in the North Atlantic, and associated increases in Pacific sea surface temperatures. These results highlight the importance of simulating not only the correct flux but also the form of the freshwater input from ice sheet collapses appropriately.
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.
 The oceanic meridional overturning circulation (MOC) is the major global transporter of heat and therefore a key element in the global climate system (see Rahmstorf  for a review). It is sustained by a continuous input of mechanical energy which induces vertical mixing and thus balances the deep water formation at high latitudes [Wunsch and Ferrari, 2004], although large-magnitude freshwater fluxes from melting ice sheets may hamper the formation of deepwater [e.g., Broecker and Denton, 1990; Lenderink and Haarsma, 1994; Rahmstorf, 2002; Green et al., 2009]. A sudden major input of freshwater from the collapse of a large ice sheet in the northern part of the Barents Sea (hereafter the Barents Ice Sheet; see Figure 1) may therefore have had a significant effect on the oceanic and atmospheric circulation, hydrological balance, biology and sedimentation in the Arctic Ocean at the end of Marine Isotope Stage 6 (MIS 6; the penultimate glaciation, ∼140 ka). There is evidence from plow marks and seafloor erosion for the existence of very large icebergs in the Arctic during MIS6, consistent with such a collapse [e.g., Dowdeswell et al., 1993; Syvitski et al., 2001; Kristoffersen et al., 2004; Kuijpers and Werner, 2007; Metz et al., 2008]. The impact of icebergs from the Laurentide Ice Sheet on the ocean and climate has been investigated previously [MacAyeal, 1993; Alley, 1998; Levine and Bigg, 2008], but studies of the effect of freshwater inputs from the collapse of the Barents Ice Sheet during MIS 6 are lacking, despite its potential consequences for the MOC and the possibility to investigate abrupt climate change for another time period than the LGM.
 Here the response of the Atlantic meridional overturning circulation (MOC) to freshwater fluxes added either as meltwater pulses or as deep-draft icebergs calved from the ice edge of the glacial Barents Ice Sheet is investigated. The idea is that the MOC is less sensitive over longer periods to a meltwater pulse compared to icebergs, but that a meltwater pulse will have a larger initial impact on the circulation than do icebergs during the actual discharge [see Green et al., 2009, 2010]. The impact of freshwater fluxes added to the North Atlantic has been investigated in a number of papers over the last few years [e.g., Zhang and Delworth, 2005; Stouffer et al., 2006; Levine and Bigg, 2008]. Within the World Climate Research Program Coupled Model Intercomparison Project/Paleo-Modeling Intercomparison Project, the performance of a number of both intermediate complexity models and fully coupled atmosphere-ocean general circulation models was tested (see Stouffer et al.  for a summary). The model considered here is somewhere in between these models as it is of intermediate complexity but consisting of a full ocean model coupled through wind- and heat-flux feedbacks to a simplified atmospheric model.
 Generally, the Atlantic MOC experiences a weakening, but not a shut down, in response to freshwater inputs of 0.1 sverdrup (Sv) (1 Sv = 106 m3 s−1) in the North Atlantic [e.g., Zhang and Delworth, 2005; Swingedouw et al., 2009]. A flux between 0.3 and 1 Sv is enough to shut down and even reverse the Atlantic MOC, with the exact flux needed depending on the model and release location [Stouffer et al., 2006; Bigg et al., 2010; Okumura et al., 2009]. A freshening in the northern North Atlantic may thus lead to a switch in deepwater formation from the Atlantic the North Pacific [e.g., Seidov and Haupt, 2005]. This is because the global salinity conveyor is critically dependent on the salinity difference between the Atlantic and Pacific Oceans, and a reduced MOC can thus be associated with a reduced salinity difference [Stouffer et al., 2007], but the exact quantitative response depends on the mean background climate state [e.g., Swingedouw et al., 2009].
 Due to the majority of the Barents Ice Sheet melting over a short time period of about 500 years [e.g., Svendsen et al., 2004], the deglaciation can be described as a collapse. Studies of ice sheet collapse tend to focus on freshwater surges; however, an important component of the deglaciation of the Eurasian ice sheets was the calving of icebergs along the marine margins, as well as surface melting along the southern land-based margins of the ice mass [Siegert and Dowdeswell, 2002; Green et al., 2010]. During calving events the majority of freshwater does not enter the ocean immediately but is instead released gradually as the icebergs melt. Such processes have been modeled for other release points during the Last Glacial Maximum [Levine and Bigg, 2008; Bigg et al., 2010], but not MIS 6. The significance of the Barents Ice Sheet for climate is that the cold temperatures and subsequent slow melting could enable some of the icebergs from its collapse to drift into the North Atlantic before melting [e.g., Green et al., 2010], providing a potential mechanism for transporting freshwater through the Fram Strait, directly into the Nordic Seas and toward the area of deep water formation. Here we employ FRUGAL (the Fine ResolUtion Greenland sea And Labrador Sea model), a global intermediate complexity coupled atmosphere ocean model, set up for the MIS 6, to simulate the impact of meltwater and icebergs on the MOC [Bigg et al., 1998; Bigg and Wadley, 2001]. Forcing consisted of coupled heat and freshwater fluxes and climatological wind fields, and the simulations included addition of a freshwater pulse, either in the form of meltwater or as seeded icebergs, along the Barents Sea coastline to represent a collapsing ice sheet. We begin by introducing the FRUGAL model and the iceberg trajectory model in section 2, where we also discuss the experiments, forcing, and freshwater fluxes. Section 3 contains a discussion of the response in global transports and upper-ocean stratification to the experiments, section 4 contains the discussion and summary, and section 5 contains the conclusions.
2. The Model System
2.1. The Model System
 FRUGAL has a resolution of 1° × 1° and realistic representation of processes in the central Arctic, yet does not require inefficiently large computational power, despite being a global model. This is achieved by using a variable-resolution orthogonal curvilinear coordinate system with the grid North Pole located in Greenland (72.5°N, 40°W) and a lower resolution (6° × 4°) in the Southern and Pacific Oceans (see Figure 1c). The ocean model in FRUGAL has 19 vertical layers with thicknesses between 30 and 500 m, a free surface, and is coupled to thermodynamic and dynamic atmosphere [Fanning and Weaver, 1996], sea ice [Parkinson and Washington, 1979], and iceberg models [Bigg et al., 1997]. FRUGAL has been set up to employ a method for reducing time step length at selected points within the grid [Wadley and Bigg, 1999]. This variable time stepping method enables increasing numbers of shorter time steps to be taken in regions of higher resolution across the model domain, while maintaining a synchronous integration and achieving numerical stability throughout the model. FRUGAL is combined with a reduced complexity atmosphere model, which includes parameterization of clouds, mountains, land ice and land hydrology, based on Fanning and Weaver , but it has been modified to include advection of water vapor and simple wind feedback [Bigg and Wadley, 2001]. This coupling provides a grid point to grid point interactive exchange of heat and freshwater between the atmosphere and ocean. FRUGAL has been used successfully to model present-day [Bigg et al., 2005] and Last Glacial Maximum conditions [Bigg et al., 1998; Wadley and Bigg, 2002; Green et al., 2009], although prior to this study it has only been employed to simulate the MIS 6 by Green et al. .
 The thermodynamic and dynamic iceberg model is described by Bigg et al.  and includes drag forcing from the ocean, atmosphere and sea ice, in addition to the Coriolis force, pressure gradients and wave radiation. Icebergs are seeded individually at the ice edge and then allowed to drift through the domain until they melt or become beached. To determine the forcing for each iceberg, the bergs are assumed to occupy a single point in space (a valid assumption with the current model resolution). The main thermodynamical processes acting on the icebergs are basal melting, buoyant convection, wave erosion, and to a lesser extent sensible heating and sublimation at the ice-air interface. At each main ocean time step information about sea ice thickness, surface ocean velocities, wind velocities, sea surface temperature, sea surface salinity, and sea ice velocity are exchanged between the ocean and iceberg trajectory models. This information is used to determine iceberg trajectories and melting rates, which are subsequently passed to the ocean model to provide information on freshwater and heat fluxes in the ocean. Note that the iceberg drift patterns have been shown to be satisfactory using only surface current (as they usually dominate the current field), so this approach is used here to keep the computational cost down [Bigg et al., 1997; Gladstone et al., 2001]. The impacts of the iceberg on atmospheric variables are small and thus there is no direct coupling between the atmosphere component of the ocean model and the iceberg model. The iceberg model has been used previously within FRUGAL by Levine and Bigg  and Green et al. , the latter of whom used it to study the drift patterns of icebergs released from the collapse of a MIS 6 Barents Ice Sheet. For space reasons within the present paper, readers are referred to Bigg et al. , Levine and Bigg , and Green et al.  for details.
2.2. Model Setup
 Most investigations into the timing and duration of freshwater pulses, either from ice streams or as meltwater pulses, deal with the Last Glacial Maximum (LGM), 18–21 ka B.P. (see the summary by Bischof ), and we have therefore used some knowledge from MIS2 and applied it to MIS 6 whenever information is lacking.
 The maximum extent of the Fennoscandian Ice Sheet during MIS 6 is based on Svendsen et al.  (Figure 1) as they significantly redefined the extent of the ice sheet. It was found to be smaller than implied by previous reconstructions [e.g., Grosswald and Hughes, 2002], but sedimentary sequences along the northern margin of Siberia show no evidence of being overridden by an ice sheet and thus that these areas were ice free during MIS 6 (and the more recent MIS 2) [Svendsen et al., 2004]. The reconstructions indicate that the northern extent of the ice sheet over Eurasia was essentially the same during both MIS 2 and 6, with the ice front along the Barents Sea margin reaching the shelf break along the Norwegian Sea and Arctic Ocean [Lubinski et al., 1996; Polyak et al., 1997, 2002; Mangerud et al., 1998]. The ice sheet extended further northwestward toward Greenland during the MIS 6 glaciation than during the last glaciation, leading to a narrower Fram Strait. Also, during the earlier glaciation the ice sheet extended further east over the Kara Sea region to the Siberian Mountains and further south over the West Siberian Lowlands [Spielhagen et al., 2004; Svendsen et al., 2004]. The subsequent location of the Barents Ice Sheet as used in FRUGAL is shown in Figure 1.
 The ice sheet growth during MIS 6 resulted in a fall in global sea level due to the transfer of water from the ocean to the ice sheets. Few studies have been conducted which focus on a longer history of sea level [e.g., Mix and Ruddiman, 1984], and the most recent estimates suggest that sea level was 125–128 m below present-day levels during MIS 6 [Rohling et al., 1998; Rabineau et al., 2006]. Here we use the higher of these estimates, using the geographical pattern provided by the LGM database from Peltier  and with sill depths adjusted from Thompson  (note that the Bering Strait was closed during MIS6). Although this latter database may not be entirely correct for MIS 6, the sea level variations were roughly similar and it is currently the best available estimate.
 The other variables which were altered in the setup were related to the atmospheric CO2 concentration and variations in solar insolation due to variation in the Earth's orbit and rotation. The latter consist of the precession, obliquity and eccentricity, and all are set to represent those appropriate for MIS 6 (see Berger and Loutre  for details and Fanning and Weaver  for the implementation).The atmospheric CO2 concentration was set to 194 ppm based on data from the Vostok Ice cores [Raynaud et al., 1993; Petit et al., 1999; Siegenthaler et al., 2005]. Note that during MIS 2 the CO2 concentration was 181 ppm, the PD CO2 concentration is 350 ppm, and that the temporal accuracy of the MIS 6 estimate is ±15 kyr.
2.3. Experiments and Ice Sheet Collapse
 The model spin-up was created by running the model, using the described MIS 6 forcing and conditions, for 100 years in robust diagnostic mode from a state of rest. The relaxation was then suppressed and the integration continued with surface forcing until the model's natural variability showed no general trend and the Atlantic MOC had reached a quasi-steady state. The time taken to reach this state in the model simulations was less than 6000 years. Following this spin-up, the control run was created by continuing the integrations for a further 3000 years. Note that in the following MIS 6 conditions are being described unless specifically noted. In the perturbation experiments, a freshwater pulse was added after the spin up (i.e., at year 6000) to test the sensitivity of the MOC to amplitude and form of the prescribed freshwater pulses. The perturbation runs were then continued for 1500 years, as the model runs had mostly recovered or reached a new steady state at this time. Note that in the following all time indications are given as time elapsed from the start of the perturbation (i.e., year 6000; this holds for the control run as well). Two types of perturbed freshwater experiments were performed: the first involved simulating the collapse as a freshwater surge (henceforth denoted FW), and the second involved modeling implicit icebergs being calved from the circumference of the Barents Ice Sheet (ICE in the following).
 The actual mechanism for collapse of the Barents Ice Sheet during MIS 6 is unknown, but indications exist that the MIS2 ice sheets, especially the Laurentide, may have been subject to intense tidally driven straining and erosion [Griffiths and Peltier, 2009]. Because the extents of the ice sheets were roughly similar for both LGM and MIS 6 [e.g., Svendsen et al., 2004] we use the estimates of ice volume and melt rates from Bischof  for MIS 2 and apply them to this investigation. To account for uncertainties a number of experiments with different freshwater volume fluxes, both as meltwater and icebergs, were done. In all simulations, the pulses began at 140,000 calendar year B.P (model year 6000; after the spin-up using nonchanging MIS 6 conditions).
 The majority of both FW and ICE simulations involved a meltwater surge being added over the Barents Sea region for a period of 500 years, in compliance with the duration of extremely high deposition rates of IRD in the Norwegian Greenland Sea during the LGM [Bischof, 1994, 2000]. Further model simulations were conducted in order to investigate the sensitivity of the model to duration of meltwater additions, with FW pulses being simulated over time periods within the range of 100–1000 years. The pulses were applied equally at all ocean grid points covering the Barents Ice Sheet. A range of different magnitudes of meltwater pulses were applied, in accordance to estimated fluxes required to melt the Barents Ice Sheet over 500 years, ranging from 0.01 to 0.3 Sv, although focus here is on the 0.1, 0.15, and 0.3 Sv cases.
 During iceberg seeding experiments similar volumes of freshwater were added, but in the form of individual icebergs released at the seaward margin of the Barents Ice Sheet following the methodology used by Bigg et al.  and Green et al. . Icebergs were seeded at the end of the spin-up model runs for a typical period of 500 years, although again the perturbation periods ranged from 100 to 1000 years. A range of values for the flux of ice released from the Barents Ice Sheet was distributed between 10 classes of initial sizes of icebergs proposed by Bigg et al. . The dimensions of the bergs are based on observations of icebergs in the Southern Hemisphere by Morgan and Budd  and Scoresby Sund, East Greenland, by Dowdeswell et al. . The observations suggested that icebergs have a lognormal length distribution and there is a sharp decline in the number of bergs above 1 km in length [Dowdeswell et al., 1992]. The icebergs here have a width-length ratio of 1:1.5, and the draft equals the width up to 300 m. A ratio of draft to freeboard of 6:1 is used. This is based on 13% of the berg being freeboard, due to the density of pure ice (here taken to be 917 kg m−3), but takes into account observations of 18% of the berg being out of the seawater. Based on the findings of Orheim  and Dowdeswell et al. , Bigg et al.  proposed a maximum iceberg draft for large bergs of 300 m, but due to the evidence that icebergs with drafts exceeding the 850 m existed in the Arctic, the draft size in class 10 was altered to account for deep draft iceberg. These were presumed to have dimensions of draft and width of 900 m, allowing for some melting and thus reduction in draft between release site and the Lomonosov Ridge [see Green et al., 2010]. A freeboard that is 16.7% of the total length of the berg gives an equivalent freeboard of 190 m of a berg with draft 900 m, and a width:length ratio of 1:1.5 results in a length for class 10 bergs of 1350 m. As a comparison, class 1 bergs have a draft of 67 m draft, class 5 have 300 m draft, and class 9 have 300 m draft. The lengths increase from 100 m for class 1 to 1350 m for class 10. Icebergs were released from each class size along the entire margin of the MIS 6 Barents Ice Sheet four times a year to account for seasonal effects on trajectories. The equivalent volume flux of freshwater in the icebergs was distributed evenly between the seeding locations and the flux was distributed among the 10 class sizes of icebergs so that the total number of bergs released at each location satisfies the statistical distribution determined from observations (see Bigg et al.  for details). The total number of bergs released at each seeding occasion was 1460, 15% of which belonged to class 1, 8% to class 5, and 5% each to classes 9 and 10.
3.1. Control Runs
 There are no major deviations from the expected picture in either global fields of sea surface salinity (SSS, Figure 2) or sea surface temperature (SST, Figure 3), except for a very high SSS in the upper part of the water column in the North Pacific. The Arctic, Atlantic and Antarctic oceans are very cold, with extensive ice covers whereas the tropical Indian and Pacific oceans are quite warm. The control run has an average Atlantic MOC of 17 Sv (Figure 4 and auxiliary material), which is a reasonable value for glacial states [e.g., Seidov and Maslin, 1999; Green et al., 2009]. The overturning estimates for other basins than the Atlantic are surprisingly energetic in the present control run. The Pacific MOC of 21 Sv exceeds that of the Atlantic (Figure 4), suggesting a state with a weakened Atlantic and an increased Pacific MOC. This is reflected in the surface stream function (Figure 5), where we see a strong Pacific subpolar gyre circulation but weak Atlantic gyres, and further highlighted in the transects of salinity and meridional velocity (Figure 6).
 The MIS6 and LGM oceanic and states differ on a few significant points. Compared to the LGM, the Nordic Seas are slightly fresher whereas the North Pacific is significantly saltier and slightly warmer during MIS 6, but there are otherwise no major differences in SST [see, e.g., Green et al., 2010]. The conveyor circulation is stronger during MIS 6, including the Atlantic MOC [e.g., Seidov and Maslin, 1999; Green et al., 2009], but the latter is weaker than the Present-day circulation [Bigg et al., 1998; Green et al., 2009]. Furthermore, the strong Pacific MOC is not recognized in investigations of previous time periods, and the sum of the Atlantic and Pacific MOCs are nearly doubled compared to that simulated during the LGM [e.g., Levine and Bigg, 2008]. These effects are present in the MIS 6 simulations despite the CO2 concentration, sea ice extent, sea level, and orbital configurations being only marginally altered. There are some (minor) differences in the atmospheric variables between MIS 6 and the LGM (not shown), but a full climatic description of the MIS6, or a comparison of the MIS6-LGM-PD, is left for a future paper (C. L. Green et al., Comparison of the impacts of a Barents Ice Sheet collapse during MIS 6 and MIS 2, manuscript in preparation, 2011). We can conclude, however, that the lack of major differences in SST or in the atmospheric forcing between the two periods and the previously mentioned North Pacific SSS maximum imply that the main reason for the response of the MOC in the perturbation experiments are due to changes in the Pacific salinity transports [see also, e.g., Seidov and Haupt, 2005].
3.2. Freshwater Pulses
 Throughout the perturbation run there is a significant cooling and freshening in the equatorial Pacific, a cooling in the Indian Ocean, and an increase in the Southern Ocean SST, all of which persist long after the pulse has ended (Figures 2 and 3). Figure 3 shows no cooling in the Arctic or North Atlantic because they are ice covered in Figure 3 since we have chosen to show a winter situation to highlight the other changes which are most prominent during boreal winter. During northern summers the previously reported North Atlantic cooling does take place [e.g., Stouffer et al., 2006], but is not shown here to save space (although the main perturbations seen persist during summer as well). Note that the freshwater perturbation in the Arctic Ocean visible during the pulse has vanished at year 1000, that is, within 500 years from the end of (Figure 2). In the 0.3 Sv simulation an initial weak reduction in SSS and SST is present in the northeast Pacific, but with an increase of both SST and SSS postpulse (Figures 2 and 3; again we focus on Northern Hemisphere winter conditions). The rest of the global ocean experiences long-term decreases of SST in this experiment, but the salinity effects are mainly confined to the northeast Pacific. The corresponding long-term increase in temperature (Figure 3) is associated with an increase in air temperature of more than 8°C in the atmosphere over the Pacific and an even larger decrease in air temperature in the North Atlantic (not shown).
 The impact on the Atlantic MOC during the MIS 6 shows a reduction in strength during the freshwater surges but it recovers to almost original strength for the 0.1 Sv FW experiment (Figure 4; note the difference between the responses to FW and ICE (to be discussed in section 3.3)). In the 0.3 Sv FW experiment the MOC does not recover at the end of the perturbation, and the change in the equatorial Pacific upwelling seen in the 0.1 Sv case is not as pronounced. From the stream function (Figure 5) we can see that the North Atlantic subpolar gyre has weakened dramatically, whereas the subtropical North Atlantic is strengthened compared to the control to accommodate the increased volume flow due to the flushed freshwater (which at this point is mixed into the ocean surface layer). The Pacific gyres are only slightly stronger than during the control (approaching 100 m2 s−1), but their horizontal structure is more regular.
 The changes seen in the MOC for the 0.3 Sv case correspond to the changes in the ocean SST, which in turn is linked to a switch in large-scale conveyor from the North Atlantic to the North Pacific. However, whether the increase in SST (and air temperature and humidity) in the Pacific is initially a cause or effect of the overturning switch is unclear. Other experiments show that the switch between warming and cooling occurs just before fluxes of 0.15 Sv. Thus, freshwater surges of 0.15 Sv or greater, which are within the predicted estimates of fluxes from the Barents Ice Sheet collapse, lead to a significant alteration in the global overturning circulation, with the overturning reducing in the North Atlantic Ocean and increasing in the North Pacific Ocean. This is not the case for the LGM [e.g., Green et al., 2009], which means the switch must be related to the MIS 6 forcing; this is discussed further in section 4. The warming of the upper ocean to the west of North America accompanies a significant increase of nearly 10 Sv in the Pacific Overturning during the simulation by year 1000 (Figure 4), and there are also increased volume transports in both the Pacific subpolar and subtropical gyres (see auxiliary material). The general overturning in the Southern Hemisphere increases whereas it decreases in the Northern Hemisphere (and the main formation site for the Northern Hemisphere is thus the North Pacific). This is a typical bipolar seesaw effect, in which cooling in the Northern Hemisphere is balanced by warming in the Southern Hemisphere [e.g., Caillon et al., 2003; Dokken and Nisancioglu, 2004]. The cause for the reduction in North Atlantic overturning in these MIS 6 FW experiments is that the freshwater from the Barents Ice Sheet collapse forms a cap over the North Atlantic (see the SSS in Figure 2), leading to a reduction in NADW formation, which is compensated for by the overturning in the Pacific Ocean. There are indications from both models and observations that the North Pacific MOC behave inversely to the North Atlantic MOC, with the latter increasing when conditions support a more globally distributed pattern of the conveyor circulation (see, e.g., Rahmstorf , de Boer et al. , and Okumura et al.  for (LGM) modeling references or Brunelle et al.  for MIS6 observational evidence). This is supported by cross sections of salinity and meridional velocity through the Atlantic (Figure 10) after 400 years of a 0.3 Sv freshwater pulse. These show a significantly fresher upper ocean than for the control and the anomalous meridional surface velocity is stronger southward, with a stronger northward component beneath it.
 To identify the impact of having a more staggered collapse of the Barents Ice Sheet a simulation was run with 0.2 Sv of freshwater pulse being added at the location of the Barents Ice Sheet for 500 years, followed by a flux of 0.025 Sv for a further 1000 years. The simulation was then continued for another 1000 years with no freshwater perturbation. The results are compared to a 500 year flux of 0.25 Sv (which has a response similar to that of the 0.3 Sv case and is thus not shown), since the same overall amount of freshwater is added during experiments. The longer, stepped simulation had less impact on the Atlantic overturning strength between model years 0–1400 but then showed a higher than original overturning flux at the end of the perturbation (Figure 4). Together with the reduced Atlantic overturning this implies a switch from a state where deepwater is formed in the Atlantic to one where the Pacific dominates. The long-term impact is thus greater than the 500 year pulse simulation and again stresses the importance of knowing the duration of ice sheet collapse in order to determine its impact on the ocean. During the second stage, the very weak flux being added is not enough to prevent deepwater formation but it does increase the circulating volume in the overturning system and it allows for compensation within the MOC to reach the prepulse equilibrium. This is not seen in the 0.3 Sv experiment, which enters a new stable equilibrium state, nor in the reduced flux experiments (e.g., 0.1 Sv). In the latter case this is so because the overall perturbation induced is weaker in terms of stratification, although sufficient to hamper deepwater formation (just as during the first part of the staggered release experiment).
3.3. Iceberg Seeding
 An iceberg seeding surge equivalent to 0.1 or 0.3 Sv of freshwater results in a significantly reduced SSS in the subpolar Atlantic during the 0.1 Sv flux experiments (Figure 7), which is different to the results from the FW experiment, where only changes of this magnitude were observed for fluxes equal to and greater than 0.15 Sv. The long-term response of the ocean is thus more sensitive to iceberg inputs than freshwater fluxes, whereas the freshwater pulses generally have a larger immediate effect (see Green et al.  for details, including a description of the iceberg trajectories). In the ICE experiments, increases in SST of up to 5°C in the North Pacific occur by year 400 for both the 0.1 Sv and 0.3 Sv additions (Figure 8). Again this is an impact of the increased overturning in the Pacific Ocean which brings warm waters to the region.
 There is a reduction of 80% for ICE 0.1 and 75% for ICE 0.3 of the Atlantic overturning strength (Figure 4). Similar changes in major transports and stream functions to those from FW again reflect the changes in pattern of the global overturning circulation during the ICE experiments (Figure 5). In fact, the stream function for ICE 0.3 is very similar to that of FW 0.3. The increase in the Pacific overturning observed during the iceberg trajectory experiment involving an iceberg flux of 0.3 Sv (Figure 4) is of a similar scale to the equivalent freshwater surge experiment, although the long-term increase in overturning strength is of a lower magnitude. The strength of the Pacific overturning at year 1000 is 22 Sv for both runs, some 12 Sv higher than in the control and 2 Sv higher than for the equivalent FW experiments. Note that the overturning and gyres have not reached full equilibrium at year 1500, but a trial run showed that this happens within 200 years and is not pursued further. The result indicates a significantly increased recovery time in the ICE experiments [see Green et al., 2010].
 These effects are forced by the gradual melting of the icebergs and the subsequent release of the freshwater over a larger area than in the FW experiments (Figure 9) [Green et al., 2010]. This leads to a more pronounced long-term effect in the ICE simulations. The lifetime of the largest bergs is about 20 years and some bergs have then traversed the entire Arctic and will be close to Iceland. In practice, iceberg releases thus have a similar effect to what could be expected from smaller freshwater releases from multiple locations.
 The transects exhibit somewhat reduced changes compared to the equivalent FW (Figures 10c and 10d; see also Figure 6), and in ICE 0.3 the salinity increase in the Pacific is confined to the surface rather than extending throughout the deep water (Figures 11c and 11d). The Beaufort Sea/Bering Strait area (y grid 35) is significantly fresher in the FW run compared to the control, whereas the anomaly in ICE is almost negligible. This implies that freshwater accumulates in the area, but cannot exit to the Pacific since the Bering Strait is closed. Icebergs, however, do not enter this part if seeded from the Barents Ice Sheet (see Figure 9) [Green et al., 2010] and instead exit through the Fram Strait into the northern North Atlantic as they drift with the currents and only have a minor effect on the dynamics. A freshwater pulse, on the other hand, can alter the upper ocean dynamics and may spread over the entire basin.
 The different freshwater content in the Arctic between the runs is also associated with an alteration of the sea surface height in the region (Figure 9, right). This shows the expected result with a larger sea surface gradient from the Barents Ice Sheet toward the North Atlantic in the 0.3 Sv compared to the 0.1 Sv simulations. This is hardly surprising, and more interesting is the larger sea surface elevation in the gyres in the Pacific during ICE 0.1 compared to ICE 0.3. This response is the opposite of that in the Atlantic, where the elevation is proportional to the freshwater flux. This explains the modified gyre circulations in the runs, and it is very likely that the response of the Atlantic gyres is due to them absorbing the freshwater volumes. This explains the fate of the freshwater: it is absorbed into the (wind-driven) large-scale upper ocean circulation or evaporating to the atmosphere (not shown, but there are significant changes in humidity over the North Atlantic). The freshwater content in the water column is on the order of tens of meters which when mixed over the upper km or so will only provide minor changes in the SSS [see also Green and Bigg, 2011].
 The different methods for simulating the collapse of the ice sheet provide different results, with a freshwater surge (experiments FW) producing a larger impact than explicit icebergs on the MOC. The lifespan of the largest iceberg is on the order of 20 years, and this gradual melting and addition of freshwater over a larger area explains the different responses. Ice sheet collapses are often simulated as freshwater surges [e.g., Stouffer et al., 2007; Green et al., 2009], which thus may overestimate the short-term impact that the collapses have on the MOC and the climate. In terms of the upper ocean circulation, the stream functions are roughly similar between the two perturbation experiments but very different to the control (Figure 5). Furthermore, even 500 years after the pulse has ended the patterns remain although with slightly different structures.
 A mass input of freshwater from the collapse of the MIS 6 Barents Ice Sheet is modeled to have an immediate effect of reducing the salinity of the surrounding surface waters. The freshwater spreads through the Arctic Ocean and some of the pulse moves through the Fram Strait into the North Atlantic, where the northern arm of the meridional overturning circulation previously formed NADW. The subsequent reduction in salinity leads to increased stability of the upper layers, which prevents convection and deepwater formation. Even small volumes of added freshwater (0.05 Sv, not shown) at the Barents Ice Sheet cause a weak reduction in Atlantic Overturning, and larger freshwater pulses of course have greater impacts. The reduced overturning strength leads to reduced heat transport to the Arctic Ocean and North Atlantic, resulting in a reduction in SST and leading to reductions in atmospheric temperature and humidity (not shown). This connection between reduced production of NADW and cold temperatures has been observed in planktonic foraminiferal assemblages [Oppo and Lehman, 1995]. However, the postpulse Atlantic MOC partially recovers to control strength. This is not the case for the simulations with freshwater surges of 0.15 Sv or greater, which are within the predicted estimates of fluxes from a collapsing MIS 6 Barents Ice Sheet [Bischof, 2000], and lead to a significant alteration of the meridional overturning circulation strengths. The reason for the different responses is that smaller fluxes can be flushed out of the system more rapidly, and do not cause a significant alteration in the hydrography at the location of deep water formation. However, fluxes of 0.15 Sv or greater form a freshwater cap over large parts of the Arctic and North Atlantic, which significantly retards the formation of NADW. The differences between 0.2 and 0.3 Sv, and 0.3 and 0.5 Sv (0.2 and 0.5 Sv results are not shown) are relatively small. Inputs of 0.5 Sv are approaching unrealistically large fluxes, whereas at the other end of the spectra a freshwater input of 0.05 Sv shows only minor impacts. Furthermore, a total collapse of the Atlantic MOC occurs for fluxes over 0.5 Sv in the present model; see also Bigg et al.  and Green et al.  for further discussions related to other time periods. The duration of freshwater addition is also an important component of simulating ice sheet collapse. Adding the same total volume of freshwater over a longer period has a smaller effect on Atlantic MOC during the pulse, although the long-term ability to recover is not clear and may need further investigation.
 Similar responses are found when the freshwater is added to the Arctic in the form of icebergs instead of a freshwater pulse, although the freshwater surge has a greater impact on the Atlantic Overturning than the iceberg experiments during the time of input [see also Stouffer et al., 2007; Green et al., 2010]. The addition of freshwater to the ocean as the icebergs melt is more gradual than the freshwater surges, and since the cold temperatures of the Arctic allow the icebergs to travel significant distances and add meltwater over a larger area, the ocean is more able to adjust to the freshwater additions. It is thus probable that simulating an ice sheet collapse as a freshwater surge overestimates the impact that the collapse has on ocean and climate. This is an important finding as ice sheet collapses are often simulated as freshwater surges, which may overestimate the impact that the collapses have on the MOC and climate. For fluxes of 0.1 Sv a similar impact is observed on the strength of the Atlantic Overturning during both the freshwater surge and iceberg calving experiments, but for larger fluxes the same flux results in less of an impact during the iceberg seeding simulations. As with the freshwater surges, icebergs affect the physical properties of the upper ocean, particularly temperature, salinity and density. Consequently, accurately simulating the Arctic and North Atlantic Ocean currents and including collapsing ice sheets in the correct form is of importance in any (paleo)climate investigation.
 The location of MIS 6 deep water formation for both types of freshwater releases switches from the North Atlantic to the North Pacific Ocean. This is in agreement with a few other investigations, albeit for the LGM [e.g., Mikolajewicz et al., 1997; Seidov and Haupt, 2005; Okazaki et al., 2010], and has been explained previously by the dependence of the MOC on the Atlantic-Pacific salinity difference [Seidov and Haupt, 2005; Stouffer et al., 2007]. It has also been shown that the response of the Atlantic MOC to freshwater inputs is not linear and depends on the mean climate state [e.g., Swingedouw et al., 2009] and we find similar results here (although other models do give different responses [see, e.g., Okumura et al., 2009]). This goes someway to explain the different long-term effects after the pulse has been removed. In the present investigation we obtain a prolonged collapse after removal of the pulse – a result not found using FRUGAL in investigations of the PD or MIS 2 [e.g., Green et al., 2009; Bigg et al., 2010].
 This raises two significant questions: can we trust the model results, and why is the (modeled) response of the ocean during MIS 6 significantly different to the LGM? The proxy evidence from the MIS 6 supports our model results with regard to the large-scale pattern. However, since this is (to our knowledge) the first modeling attempt of this form of MIS 6, we can only assume that the model gives reasonable results because it does so when set up for PD conditions (compared to direct observations) and for LGM/MIS 2 conditions (compared to proxies); see the discussions by Bigg et al.  and Levine and Bigg  for details. There may be an issue with the resolution of the North Atlantic being finer than in the Southern and Pacific Oceans, which may affect the deepwater formation ratio between the North Atlantic and Southern Ocean. It is thus likely that the North Atlantic MOC is overestimated and the Southern Ocean and Pacific MOC are underestimated, and there are plans to run a high-resolution version of model to quantify the effect of the resolution (if any). Nevertheless, this has not been a major issue in past FRUGAL simulations of the PD or LGM. Hopefully, similar experiments with other models will soon follow and allow for model-model comparisons. While there is some evidence for a catastrophic Barents Ice Sheet collapse during MIS 6 such as that assumed here [Kristoffersen et al., 2004; Green et al., 2010] it remains to be seen whether further observational evidence will support the switch of deep convection locations.
 The answer to the second questions (why is the MIS 6 responding differently than the LGM/MIS2?) is less obvious. One explanation is that the mean climate state is slightly different to the PD and MIS2, and this is known to affect the response of the MOC to perturbations, but part of the answer lies within the control simulation. The control MIS 6 Pacific overturning strength is similar to that of the Atlantic (Figure 4), and the Pacific is relatively salty at the surface (Figure 2), promoting deep water formation there [Seidov and Haupt, 2005]. There is evidence, based on seawater δ18O values, of a larger Pacific-Atlantic surface salinity gradient during MIS 6 compared to the LGM [McManus et al., 1999; Lea et al., 2000], as well as support for a shallow thermocline and significant upwelling in the MIS 6 western Pacific [Li et al., 2010]. Both these observations are compatible with the North Pacific being more susceptible to convection during MIS 6, and so are consistent with the MIS 6 model results presented here.
 The response of the ocean to a combined collapse of the Barents and Laurentide ice sheets would most likely lead to a catastrophic shut down of the Atlantic MOC, as the system can barely deal with one of these pulses on its own. That said, when the pulses have been flushed out the Atlantic MOC recovers somewhat, and a more detailed investigation is left for future studies. The same is true for a detailed comparison of MIS 2 and 6 with PD conditions.
 Focus here has been on freshwater perturbations in the Arctic Ocean during MIS 6, whereas focus previously have been on additions from the Laurentide and Fennoscandian ice sheets during the LGM, mainly in the form of meltwater additions. As shown by Bigg et al. , there is significant variance in the response of the MOC due to location of the pulse, with freshwater additions to the Arctic inducing larger reductions of the Atlantic MOC than pulses from the other ice sheets. Here, it is also verified that simulating the collapse as icebergs, which is the probable mechanism from the Barents Ice Sheet [Kristoffersen et al., 2004], gives a different response to simulations with meltwater pulses, which is probably most accurate for simulating, for example, the Laurentide during the Younger Dryas. The Atlantic MOC in the MIS 6 state also seems to be less sensitive to freshwater pulses (in the Arctic) than MIS 2 and the reason for this is under investigation. These results highlight the importance of simulating not only the correct flux but also the form of the freshwater input from ice sheet collapses appropriately, regardless of which area or time period is under investigation. The model also produces results supporting MIS 6 proxies with a more saline Pacific, which aids in setting up a dominating Pacific MOC. This suggests that periods with a weakened Atlantic-Pacific salinity gradient may experience a different MOC pattern.
 Funding was provided by the British Natural Environmental Research Council through grants NERC/S/R/2005/13953 (e-science studentship to C.L.G. through G.R.B.) and NE/F014821/1 (Advanced Fellowship for J.A.M.G.). J.A.M.G. also acknowledges funding from the Climate Change Consortium for Wales (C3W). The manuscript was greatly improved by comments from the editor (Rainer Zahn) and from three reviewers (Igor Polyakov and two anonymous reviewers).