We introduce explicit icebergs from a dynamic and thermodynamic iceberg model into an intermediate complexity climate model, which includes the coupled atmosphere-ocean system. This modeling approach allows iceberg meltwater to be injected into the ocean on the basis of thermodynamical considerations along the iceberg trajectories. Icebergs are seeded from known ice sheets in both hemispheres. Adding icebergs to the present-day climate model has a minimal impact, but during the Last Glacial Maximum (LGM), Atlantic overturning strength is reduced by a third, while producing a model state that is consistent with a steady state climate. We test the sensitivity of the model at the LGM to additional Heinrich event-scale fluxes of icebergs from three possible sources: Hudson Strait, the Gulf of Saint Lawrence, and the Norwegian Channel Ice Stream (NCIS). The sensitivity of the ocean is similar for all locations, with differences dominated by the length of the iceberg meltwater pathways to the main ocean convection region. The NCIS events result in more variability and a distinctly different, more northerly, salinity anomaly. We compare these results to a more typical modeling approach, whereby meltwater is injected directly into the ocean at the iceberg source locations, and find that these floods overestimate the oceanic response compared to the iceberg events. Our results suggest that 0.3–0.4 Sv of additional freshwater flux, either as icebergs or freshwater, is required to shut down the North Atlantic meridional overturning, a larger freshwater flux than sometimes suggested because of the localized nature of the release of the freshwater.
 Heinrich events are large sustained released of icebergs from ice streams, and are thought to be an important mechanism for delivering meltwater to the ocean. Evidence of these events can be seen in extensive records of ice-rafted debris (IRD) peaks in deep sea cores from the last glaciation over much of the North Atlantic [Heinrich, 1988; Dowdeswell et al., 1995]. The deposition of these layers correlates with cold periods in the North Atlantic, which is often attributed to a shutdown of the ocean thermohaline circulation [e.g., Sarnthein et al., 1995]. This circulation, which plays an important role in transporting heat from the equator to high latitudes, is sensitive to freshwater inputs that can be provided in sufficient quantity and rapidity by former ice sheets [MacAyeal, 1993; Marshall and Clarke, 1999; Broecker, 2003].
 The peaks of IRD are commonly ascribed to episodic releases from the Hudson Strait Ice Stream of the Laurentide Ice Sheet, set off by binge-purge oscillations within the ice sheet [MacAyeal, 1993]. Lithological evidence exists to connect many of them with North American sources [Grousset et al., 1993; Gwiazda et al., 1996a]. The case for meltwater sourced from the Fennoscandian and British ice sheets is less well known than for the Heinrich events originating from North America, and yet good evidence exists for increases in IRD from these sources during some Heinrich events in the marine sedimentary record [Gwiazda et al., 1996b; Grousset et al., 2000]. This is particularly strong for H3, about 29 ka BP. However, work by Lang Farmer et al.  suggests the Gulf of Saint Lawrence is also a possible source for H3 IRD.
 Previous modeling efforts looking at the impact of abrupt freshwater fluxes on the glacial ocean have not included an explicit iceberg model, but have rather simulated Heinrich events by directly introducing freshwater into the North Atlantic Ocean [e.g., Ganopolski and Rahmstorf, 2001; Prange et al., 2004]. Another approach is taken by Seidov and Maslin , who modeled these events by replacing Last Glacial Maximum (LGM) boundary conditions with Heinrich event sea surface conditions from proxy data. Both approaches result in a significant climatic impact through cessation of the Atlantic Meridional Overturning Circulation (MOC), with the results of Seidov and Maslin  suggesting it is difficult to distinguish between the impact of Laurentide and European sources.
 In the experiments presented here we have coupled a dynamic and thermodynamic iceberg model to an intermediate complexity climate model, in order to model Heinrich events with the release of explicit icebergs. The inclusion of a realistic iceberg model is likely to have a different impact on the MOC response to Heinrich events than the direct release of freshwater at iceberg source locations, as the iceberg trajectories will determine whether the meltwater is gradually released in, or upstream of, the deep convection regions. Icebergs are likely to have an impact on the temperature and freshwater distribution in the ocean, even during a steady state climate, as has been shown to be the case in the present-day (PD) Southern Ocean by Gladstone et al. .
 Although the contribution of iceberg meltwater release can be indicated by available IRD data, it has been shown in a model study by De'Ath et al.  that the IRD signature from the icebergs is more likely to be confined to the coast, while the meltwater is released further offshore. Therefore by coupling the iceberg and ocean models we aim to get a more realistic view of iceberg impact on the ocean. Iceberg calving events are modeled from three source regions that have been put forward by paleo-observations: Hudson Strait, the Gulf of Saint Lawrence, and the Norwegian Channel Ice Stream (NCIS).
 In the next sections we discuss the intermediate complexity climate model, followed by details of the iceberg model and the coupling strategy, including the global iceberg seeding approach. First steady states of the coupled iceberg-atmosphere-ocean model are described for both PD and LGM periods that include icebergs that are seeded from all global ice sheets. Then simulations of the impact of iceberg surges from the three source locations are described. These are then compared to the more typical modeling approach of Heinrich events as freshwater floods, whereby the meltwater is injected directly into the ocean at the iceberg source locations.
2. Climate Model
 The ocean component of the model is the same as that described in the PD model study by Wadley and Bigg , and uses a curvilinear coordinate system [Madec and Imbard, 1996]. It has a free surface formulation for the barotropic mode [Webb, 1996] and 19 levels in the vertical, varying in thickness from 30 m near the surface to 500 m at depth. Tracer (temperature and salinity) mixing has components in the horizontal, vertical and along isoneutral surfaces [Griffies et al., 1998]. The values of the mixing coefficients are taken from England . The topography is constructed from the ETOPO (1986) data set, with sill depths taken from Thompson .
 The model's curvilinear grid places the North Pole in Greenland. This enhances model resolution in the North Atlantic and Arctic, and in particular in the Greenland and Labrador seas. Thus in the Nordic Seas the horizontal resolution is 1–2°, whereas in the Southern Hemisphere it is 6–8°. Time step length is a function of grid spacing, to allow efficient integration of the variable resolution grid [Wadley and Bigg, 1999].
 The ocean model is coupled to a simple radiative advective atmosphere, which is an adaptation of the energy moisture balance model as used in the UViC Earth Systems Model [Fanning and Weaver, 1996] that allows for advection of water vapor. The atmosphere model is adapted to use the same grid as the ocean model. The model includes a thermodynamic sea ice component, which is advected by the surface ocean currents. The representation of the atmosphere includes parameterizations of clouds, mountains, land ice, and land hydrology. For the PD the wind stress is taken from climatological values [Hellerman and Rosenstein, 1983], which is adjusted for the LGM by (LGM-PD) wind stress from atmospheric GCM simulations by Dong and Valdes .
 For the LGM integrations the ice sheet topography is prescribed from Peltier , and the topography is altered further by lowering the sea level 120 m compared to the present day to be consistent with the reconstructed sea level of Fairbanks . In order to get a state with intermediate water formation in the North Atlantic, as is believed to be consistent with proxy data [Weinelt et al., 1996; Curry and Oppo, 2005], an additional freshwater flux of 1.5 mm/d is added in a latitude band of 60–75°N over the Atlantic Ocean. Also included at the LGM is a scheme allowing the freshwater flux over land to flow into the ocean via the steepest gradient. This scheme is not included in the PD simulation as it is found that this leads to too much coastal freshening in the ocean when added to the original freshwater fluxes. The base model setup does not include a representation of icebergs, however at the LGM this is represented by the additional freshwater flux of 1.5 mm/d over the northern North Atlantic.
 The PD and LGM models were initialized from a uniformly cold ocean (T = −2°C) with global mean salinity (S = 34.73 psu). The PD and LGM models reach an equilibrium state after approximately 1500 and 3000 years respectively. Table 1 shows the main transports and sea ice volumes for the models. The peak North Atlantic overturning of 27 Sv and 14 Sv occur at approximately 50°N and 40°N for PD and LGM respectively. The peak strength of the PD overturning is quite high, although the amount of North Atlantic Deep Water (NADW) that flows southward across the equator contributing to the global thermohaline circulation is only 13 Sv, which is reasonably consistent with observations [Gordon, 1986; Schmitz, 1995].
Table 1. Model Transports and Sea Ice Areas for Control Runs With and Without Icebergsa
Transports (Sv)/Sea Ice Areas (106 km2)
PD Without Icebergs
LGM Without Icebergs
PD With Icebergs
LGM With Icebergs
The errors indicate the variance of the annual mean time series. Unit is Sv for model transports and 106 km2 for sea ice areas.
27.2 ± 1.0
13.5 ± 0.6
28.1 ± 0.3
9.6 ± 0.3
Atlantic subpolar gyre
20.4 ± 1.8
3.9 ± 0.1
20.5 ± 0.2
4.1 ± 0.1
Atlantic subtropical gyre
30.7 ± 2.5
12.1 ± 0.2
11.7 ± 0.1
Drake Passage transport
57.5 ± 5.0
92.5 ± 11.2
63.5 ± 4.2
88.6 ± 2.3
15.3 ± 0.5
9.3 ± 0.5
15.7 ± 0.1
9.6 ± 0.4
Northern Hemisphere sea ice
5.4 ± 0.4
25.6 ± 0.2
5.4 ± 0.2
25.8 ± 0.2
Southern Hemisphere sea ice
1.8 ± 0.1
8.7 ± 1.8
1.7 ± 0.1
9.3 ± 0.9
 In the PD run the main deep convection occurs at 60°N in the Greenland Sea, where NADW sinks down to 4500 m, which is consistent with observations [e.g., Marshall and Schott, 1999]. In the LGM run the main convection area is at 45°N in the central and eastern Atlantic, with water sinking down to 1800 m. This is consistent with other studies suggesting the shallower convection of the last glacial occurred south of Iceland [Weinelt et al., 1996, Seidov and Maslin, 1999] and the reduced production of NADW at the LGM, with the dominance of the AABW as bottom water in the whole Atlantic [Boyle, 1995; Curry and Oppo, 2005]. There is another, more localized, convection site in the Norwegian Sea that produces less dense water than in the Atlantic.
 Historically there has been large uncertainty over the state of the ocean thermohaline circulation at the LGM, with different interpretations of the available paleo-observations. Three conceptual models that have been put forward have been summarized by Boyle  and Yu et al. . The scenarios differ in the formation rate of North Atlantic Deep Water, and the resulting ocean states have also been discussed and modeled by Bigg et al. [1998, 2000]. A southern sinking state (SSS), with little NADW influence and where Antarctic Bottom Water (AABW) is the dominant water mass in the deep Atlantic, has been suggested by Oppo and Fairbanks  on the basis of nutrient-related interpretations of benthic δ13 C. This scenario has also been implied from reconstructions of relatively shallow convection in the North Atlantic by Boyle and Keigwin .
 An alternative interpretation of these data by Duplessy et al.  suggests intermediate water was formed in the North Atlantic, and then spread out into all other oceans except the Southern Ocean (an intermediate sinking state, ISS). There it is proposed AABW formation occurred, which spread into the deep levels of the other oceans.
 A third scenario, based on radiocarbon ages of LGM deep water, contradicts the 2 previous scenarios, and suggests that the ventilation of Atlantic and Pacific Oceans occurred at rates similar to PD [Shackleton et al., 1988; Broecker et al., 1990], a northern sinking state (NSS).
 Recent work, based on temperature and salinity reconstructions from deep ocean cores and paleonutrient tracers, has strengthened the case for the ISS case with increased influence of AABW in the deep North Atlantic at the LGM compared to PD [e.g., Adkins et al., 2002; Curry and Oppo, 2005; Marchitto and Broecker, 2006]. However, inferring the strength of the NADW cell from these types of measurements is more difficult [LeGrand and Wunsch, 1995]. The strength of the glacial overturning that has been implied from sedimentary records varies a great deal. Export rates of North Atlantic water into the Southern Ocean have been implied by sedimentary records to be similar or slightly higher to PD [Yu et al., 1996], while McManus et al.  infer a reduction of the Atlantic overturning by 30–40% (although probably less).
 Our model for the LGM also predicts the dominance of AABW in the deep Atlantic, and suggests an ocean state with NADW formation up to intermediate depths, with reduced strength maximum Atlantic overturning of about 50% at the LGM (13.5 Sv) compared to our PD value (27 Sv). This seems to be within the uncertainty ranges provided by paleo-observations as discussed above, therefore this model provides a suitable tool for this sensitivity study of the impacts of Heinrich events. The uncertainty of the strength of the MOC at the LGM is also seen in other coupled models of the LGM, while producing similar climates for the PD. Weber et al.  find that of 9 coupled models of the LGM only 1 model shows a similar strength MOC to PD, while 4 models show a decrease of 20–40%, and 4 models show an increase of 10–40%.
 Sea surface temperature (SST) and sea surface salinity (SSS) fields of both PD and LGM model runs for January are shown in Figure 1, together with PD climatological SST [Levitus et al., 1994] and SSS [Levitus and Boyer, 1994] for the same period. The PD sea surface properties compare reasonably well with the climatological values, although the gradients associated with the Gulf Stream are not fully resolved. The modeled meridional temperature gradients across the Southern Ocean are less well resolved because of the relatively coarse resolution of the grid in this region. The tropical surface air temperatures and SSTs are several degrees too low, which results in reduced evaporation and fresher tropical sea surface conditions than observed, with salinity anomalies over 1 psu. The temperature and salinity distribution in the northern North Atlantic and the Nordic seas corresponds quite well to the climatology, and results in realistic NADW formation.
 For the LGM the SSTs remain fairly close to the CLIMAP reconstructions [CLIMAP Project Members, 1984], which are taken as boundary conditions in the AGCM of Dong and Valdes  that is used for calculating the LGM wind stress. The reconstructions of the MARGO project [Kucera et al., 2005] provide more recent reconstructions of Atlantic SSTs at the LGM, suggesting the CLIMAP values (and our modeled values) are too cold in the northern North Atlantic and Nordic Seas. The CLIMAP LGM time slice has not been precisely defined and possibly includes Heinrich events, while the MARGO data is more accurately defined as from the 19–23 ka BP period, and therefore excludes any influence of Heinrich events. The possible cold bias in the CLIMAP reconstructions can influence our model results through the wind stress. This has been seen in the PMIP project, where the use of CLIMAP SST data as a boundary condition in atmospheric GCMs leads to sea surface temperatures (from slab ocean models) that are more consistent with CLIMAP than MARGO data [Kageyama et al., 2006].
 Our modeled temperatures in the tropics and subtropics are lower than CLIMAP, indicating an adjustment toward a state with a stronger meridional heat transport in the Atlantic. However, the winter limits of near freezing ocean surface temperatures in the model between 40–50°N are still further south than in the MARGO reconstructions. The modeled sea ice distribution produces sea ice all year round in the Nordic Seas, while this area is thought to have been seasonally ice-free [Pflaumann et al., 2003; Kucera et al., 2005], which also indicates our model is too cold in the northern North Atlantic. Other coupled atmosphere-ocean models from the PMIP project have similar problems obtaining LGM-PD temperature anomalies similar to those of MARGO [Kageyama et al., 2006]. In these PMIP simulations the main area of LGM-PD cooling is not over Europe as is the case in the MARGO reconstruction.
 Reconstructions of SSS conditions in the northern North Atlantic by de Vernal et al.  have a pattern of relatively fresh waters in coastal regions adjacent to the large ice sheets due to the meltwater input. There is a general freshening at high latitudes compared to the PD, while salinity maxima are found in the northeastern North Atlantic. Although these reconstructions have been questioned by Telford , this corresponds well with the relatively fresh waters of 29–32 psu in the Labrador Sea and Nordic Seas in the model, and the relatively fresh waters around the American, Canadian and Scandinavian coasts north of 45°N. The reduced northward salt transport in the model compared to the PD in the northern North Atlantic is a result of the reduced strength of the North Atlantic Current, which is consistent with reduced Atlantic overturning.
3. Iceberg Model Coupling
 The intermediate complexity climate model is coupled to an iceberg model to investigate the impact of iceberg surging during Heinrich events on the LGM ocean. The resulting coupled model allows iceberg meltwater to be gradually injected into the ocean on the basis of thermodynamical considerations along the iceberg trajectories. A similar coupling has recently been performed independently by Jongma et al. , who focus on the PD period. The dynamic and thermodynamic iceberg model we use was first introduced by Bigg et al.  and modified by Gladstone et al. . It has been used to calculate iceberg trajectories and meltwater fields that are forced by off-line ocean and atmospheric fields of momentum and temperature from GCMs. The dynamical processes included in the model are ocean/atmosphere/sea ice drags, the Coriolis force, pressure gradients in the surrounding ocean, and wave radiation. The main thermodynamical processes in the model are basal melting, buoyant convection, wave erosion, and several smaller terms due to sublimation and latent heat transfer.
 Information of surface ocean and wind velocities, atmospheric surface and sea surface temperatures, sea surface salinity, and sea ice thickness and velocity, is exchanged between the climate model and the iceberg model for each main ocean time step. This data is used to calculate the iceberg trajectories and melt rates, which are fed back into the climate model to alter the surface freshwater and heat fluxes into the ocean.
 The iceberg model has typically been used for decadal time scale simulations with only local iceberg seeding, and uses a time step of 120 s for integrating the iceberg equations of motion. This time step is 360 times shorter than the ocean tracer time step, and would result in severe increase in the model run time. Here we intend to use the coupled iceberg-ocean model for centennial to millennial-scale simulations, with icebergs seeded from all global ice sheets. This leads to a much larger number of iceberg computations, and requires some compromise in accuracy to limit the impact on the coupled iceberg-ocean model. Therefore we have chosen to increase the iceberg time step to 10800 s, the same value as used for the main ocean time step. This has required a change to an implicit numerical integration scheme to reduce the chance of numerical instability, for which we have chosen the Backward Euler method [e.g., Atkinson, 1989]. Time step sensitivity studies show that the iceberg trajectories are affected slightly in regions with strong currents, such as off the coast of East Greenland, as the trajectories are calculated within each ocean model grid box. However, the resulting impact of the iceberg meltwater on the ocean does not change significantly because of the relatively coarse resolution of the ocean model grid.
 The original basal melting scheme for the iceberg thermodynamics [see Bigg et al., 1997] has been replaced by the three-equation formulation of Holland and Jenkins  for the turbulent transfer of heat beneath ice shelves. This formulation has been shown to produce more realistic iceberg meltwater fields in the Southern Ocean by Silva et al.  in iceberg model experiments. In the Northern Hemisphere this change is not found to lead to major qualitative changes in the distribution of iceberg meltwater, although the meltwater is found to spread slightly further from its sources. The impact of the icebergs on the heat budget of the ocean is also taken into account, whereby the temperature of the iceberg meltwater is taken as 0°C. In the model the wind fields are prescribed, however the changes in the heat and freshwater fluxes due to the introduction of icebergs influence the ocean surface circulation, allowing for interactive coupling between the ocean and the iceberg trajectories.
 A new control state of the ocean circulation is created with the coupled iceberg-ocean model for both PD and LGM periods, whereby icebergs are seeded globally from ice streams in all major ice sheets. Estimated iceberg discharges for the Northern Hemisphere in a steady state climate for both PD and LGM periods have been calculated by Bigg and Wadley . These calculations were based on balancing atmospheric precipitation fields from the AGCM runs for the LGM by Dong and Valdes  against iceberg calving, under the assumption that the LGM ice sheets [Peltier, 1994] were in a steady state. The locations of the coastal ice streams were then determined using a steepest gradient algorithm over the ice sheet topography. This method has been found to reproduce realistic ice streams for the PD Greenland ice sheet [Bigg and Wadley, 2001].
 For the Southern Hemisphere we have used climatological iceberg fluxes from Gladstone et al.  that are based on PD mass balance calculations for the Antarctic ice sheet. Even though the Antarctic ice sheet has decreased in volume since the LGM, there is evidence from cores taken at various latitudes in the South Atlantic that IRD delivery is at a minimum at periods surrounding the LGM and during the Holocene [e.g., Kanfoush et al., 2000]. We assume the Antarctic ice sheet to be in a steady state for both PD and LGM simulations and thus use the PD iceberg fluxes for both PD and LGM simulations. The resulting iceberg fluxes for the basic coupled runs are shown in Table 2, with the seeding locations shown in Figure 2.
Table 2. Base Iceberg-Ocean Runs: Iceberg Seedinga
Base Iceberg Fluxes
PD: Northern/Southern Hemisphere
LGM: Northern/Southern Hemisphere
The release sites of the PD in the Northern Hemisphere have been adapted to observed ice streams; therefore the lower number of release sites for the LGM is due to the relatively coarse resolution of the model combined with more limited knowledge of ice stream locations.
Total iceberg flux (km3/a)
Number of release sites
 The icebergs are divided into 10 size classes, which are based on PD observations of icebergs in the Southern Ocean [Bigg et al., 1997]. One iceberg from each size class is released from each seeding location, and the meltwater release of each berg is then calculated using a scaling factor such that the total number of icebergs released from each location satisfies the statistical distribution as given by Bigg et al. . The icebergs are released 4 times a year in order to include seasonal effects on the trajectories. There is no seasonal bias included in the iceberg calving as sensitivity studies show this has little impact on the resulting ocean circulation.
 The increase in computations due to the iceberg coupling depends on the number of icebergs present in the ocean at any time. With the global iceberg seeding approach the number of icebergs present in the model at one time varies between 5000 and 8000 bergs in our simulations. This number is of the same order as the number of horizontal grid points in the ocean model, leading to an increase in the number of computations per ocean time step of up to 1.5–2, making millennial time scale simulations still possible.
 At some seeding locations icebergs tend to get stranded at the coast for unrealistically long periods, because of unfavorable surface ocean currents. This is mainly due to the relatively coarse resolution of the ocean model grid, which does not allow for mesoscale features, such as coastal currents, to be resolved. When the icebergs have been modeled in a standalone (uncoupled) iceberg model, where the forcing was provided by GCM model output, it has been found that icebergs are less likely to get grounded with a higher-resolution grid and forcing fields [e.g., Bigg et al., 1997]. In the simulations described in the following sections we have allowed the icebergs to melt instantaneously in such situations. This approach was found to give the most realistic and computationally economical solution, as we discuss in the next sections.
4. Iceberg-Atmosphere-Ocean Model Control Runs
 The icebergs were added to the steady states of the control runs of the climate model at year 3000 for PD, and year 5000 for LGM. The coupled iceberg-atmosphere-ocean model run was found to stabilize well within 500 years, and was continued for a total period of 1500 years. The steady state iceberg fluxes do not have much impact on the major transports in the ocean for the PD, as can be seen in Table 1. At the LGM the peak Atlantic overturning is reduced to 9.6 Sv, because of the increased surface freshening of the North Atlantic, while other transports are not significantly affected. The location and depth of convection in the Atlantic is unaffected by the inclusion of the icebergs for both the PD and LGM. The resulting coupled iceberg-atmosphere-ocean states for PD and LGM are therefore still as consistent with (paleo-) observations as the model without icebergs.
 The resulting PD and LGM iceberg meltwater distributions in the ocean are shown in Figure 3. As found previously [e.g., Bigg et al., 1997] the icebergs are mainly transported in the direction of the ocean currents. In the PD Southern Ocean the icebergs reach slightly further northward than observed iceberg limits of the Soviet Antarctic Survey . Also the meltwater pattern does not reproduce a minimum in the Indian Ocean sector of the Southern Ocean. We attribute this to the relatively coarse resolution of our model grid in the Southern Ocean as this is better represented by higher-resolution standalone (uncoupled) iceberg model simulations by Gladstone et al. . In the Northern Hemisphere the greatest impact of the PD icebergs is the melt delivery to Baffin Bay, while there is also significant impact around the whole Greenland coast. The icebergs from southwest Greenland are carried southward because of the large influence of the Labrador Current and then entrained in the North Atlantic Drift. Therefore the modeled icebergs reach further east than reconstructed iceberg limits for the Northern Hemisphere by Bigg et al. , as is also found by Jongma et al. . The model does not resolve the Labrador coastal current sufficiently to keep icebergs entrained in this near the land, as occurs in reality.
 The LGM bergs in the Southern Ocean reach further north than the PD bergs, despite the LGM iceberg seeding from Antarctica being the same as for the PD. This is a result of the colder conditions at the LGM. In the North Atlantic the meltwater from the Gulf of Saint Lawrence icebergs has the greatest influence on the ocean surface of all the steady state iceberg fluxes, with significant amounts of meltwater reaching the eastern Atlantic boundary. These bergs flow south of the main convection area, and therefore do not directly impact the North Atlantic Deep Water (NADW) formation, although their meltwater is able to mix further northward. A relatively large number of Hudson Strait bergs, which contain the most meltwater, get stuck at the coast because of the unfavorable surface ocean currents, although the resulting freshwater input into the ocean is able to spread toward the central Atlantic. Carbonate concentrations and the magnetic signature in non-Heinrich IRD in the northern Atlantic are consistent with little input from the Hudson Strait area [Watkins et al., 2007], which fits the low rate of escape from the Labrador Sea for modeled Hudson Strait bergs, even though the coastal current is not well resolved here.
 Other sources of icebergs at the LGM have relatively small contributions to the meltwater field affecting the North Atlantic. The icebergs seeded from the NCIS area tend to flow northward toward Spitsbergen. However, before reaching Fram Strait they are carried back in the ocean recirculation toward the Nordic Seas and the Atlantic, where they release a significant amount of their meltwater. Therefore it is likely these ice streams can have a significant impact on the glacial convection if the calving rate is increased.
 In the next sections we describe simulations where Heinrich-type iceberg surges are introduced into this base LGM iceberg-atmosphere-ocean state, in order to assess the likelihood of three hypothesized sources for Heinrich events.
5. Simulation of Heinrich Events as Iceberg Surges
 The duration of the iceberg surges during Heinrich events are estimated to be of the order of several hundred years, with Hemming  suggesting a period of 500 ± 250 years. A reconstruction of the duration and flux of H4 has been performed by Roche et al. , who compare measurements of oxygen isotope data with results from an intermediate complexity climate model. This results in an estimate for the freshwater flux of 0.29 ± 0.05 Sv for the duration of 250 ± 150 years. Other modeling results of surges in the Laurentide Ice Sheet by Calov et al.  have suggested Heinrich fluxes are of the order of 0.1 Sv for periods of several hundred years.
 In our simulations we take the duration of the iceberg surges during Heinrich events as 500 years, which lies within the range of uncertainty, and allows the major impacts to develop, which occur mainly within 200 years. We then test the sensitivity of the model to iceberg fluxes equivalent to 0.1, 0.2, and 0.4 Sv of freshwater. These fluxes are added for 500 years to the base iceberg fluxes in the steady state of the LGM coupled iceberg-atmosphere-ocean model, at year 5500. The model is then allowed to recover by continuing the simulation for another 500 years with only the base iceberg fluxes. To assess the impact of the iceberg coupling we compare the results of these simulations to the impact of freshwater floods of the same magnitude in the next section.
 We consider three sources for the Heinrich events: Hudson Strait, the Gulf of Saint Lawrence, and the NCIS. There are four ice streams that correspond to the Hudson Strait outflow area. These are ice streams A, B, C, D as seen in Figure 2b. Table 3 shows the original and maximum strength flux of each of these ice streams. The combined original flux for the Hudson Strait ice streams is approximately 0.04 Sv, and therefore the maximum flux is only a tenfold increase in flux.
Table 3. Original and Maximum Heinrich Event Fluxes for Three Sources
Original Flux (km3/a)
Maximum Flux (km3/a)
Hudson Strait Ice Streams
1297 km3/a (eqv. 0.04 Sv)
12969 km3/a (eqv. 0.4 Sv)
Gulf of Saint Lawrence Ice Streams
672 km3/a (eqv. 0.02 Sv)
12969 km3/a (eqv. 0.4 Sv)
European Ice Streams
49 km3/a (eqv. 0.0016 Sv)
12969 km3/a (eqv. 0.4 Sv)
 The Gulf of Saint Lawrence outflow area corresponds to ice streams E, F, G, H. The original flux for these ice streams is approximately 0.02 Sv, and therefore the maximum flux is a twenty fold increase in flux. The NCIS outflow area corresponds to ice streams I and J, which have a combined original flux of 0.0016 Sv. This is a similar order of magnitude to the maximum strength of the NCIS (0.003 Sv) reconstructed from sediment deposits between 20 and 19 ka BP [Nygård et al., 2007]. Therefore the NCIS requires the largest increase from its steady state strength to reach Heinrich size fluxes. Simple reconstructions of ice stream fluxes, based on typical ice stream velocities and the cross section of the outflow area, suggest the sum of all ice streams from the Fennoscandian and Barents Sea Ice Sheets (if all operating simultaneously) would have been in the range of 0.003–0.1 Sv (A. Nygård, personal communication, 2008). Therefore the 0.2 and 0.4 Sv events from the NCIS should be seen as sensitivity experiments.
 In these enhanced flux experiments a large number of Hudson Strait icebergs still become grounded at the coast because of unresolved small-scale coastal currents. This is unrealistic for Heinrich events as IRD with a Hudson Bay signature has been found far out in the Atlantic for H1 and H2 [Dowdeswell et al., 1995]. When the Hudson Strait bergs are allowed to melt in situ it is found that the melt rate of the icebergs is unrealistically slow. This results in almost no oceanic impact for Hudson Strait Heinrich events, even for the maximum flux that is equivalent to 0.4 Sv of freshwater. This also causes an enormous number of icebergs to build up in the ocean, significantly slowing down the model performance. Therefore for the present study we have allowed the icebergs to melt instantaneously in such situations, which was found to give the most realistic and economical solution. Introducing a size threshold for this process had little impact, while a time delay merely slowed the eventual effect, so all grounded bergs are melted instantaneously. Using this approach the Hudson Strait 0.4 Sv Heinrich event leads to a shutdown of the MOC, with only a partial collapse of the MOC for smaller iceberg fluxes.
Figure 4 shows time series of the peak Atlantic overturning for the Heinrich simulations associated with each release site. The results show that in general the impact on the ocean increases with the size of the event. A complete shutdown of the MOC is seen for fluxes of 0.4 Sv followed by a complete recovery, with a partial collapse of the MOC for smaller fluxes. The impact time is less than 30 years for the Hudson Strait and Gulf of Saint Lawrence events, while the NCIS events are more variable with the main impact within 100 years. The recovery back to the normal LGM range of MOC strength occurs within 500 years. The impact is largest for the Hudson Strait and the Gulf of Saint Lawrence events, with the NCIS events leading to a more variable response but a slower recovery. Note that this additional net freshwater can alter the new mean MOC value by 10–20%.
 The melting of the coastal Hudson Strait bergs causes a large rise of the sea surface at the inflow area, which dominates the pressure field at the surface, and drives a relatively large southward surface flow from the northwest Atlantic. This has the greatest impact on the Labrador Sea, and also increases the surface return flow of the Atlantic subtropical gyre. This impact is seen in the sea surface salinity anomaly shown in Figure 5a. Although the Gulf of Saint Lawrence icebergs are able to flow out into the Atlantic, the ocean response is similar to the Hudson Strait event, but with the impact of meltwater seen further south (Figure 5b).
 The NCIS bergs release a large part of their meltwater after recirculating toward the Atlantic in the Greenland and Norwegian seas. This area is very sensitive to salinity changes, and this leads to a great deal of variability in the currents in the Nordic Seas and the subpolar North Atlantic. The result of this is seen in the salinity anomalies in Figure 5c, with a much patchier field than for the other events. This also shows that for similar sized events the NCIS bergs are at least as likely as other sources to influence the main North Atlantic convection region, which is at approximately 45–50°N in the central and eastern Atlantic. These results show that while the meltwater input from Hudson Strait and Gulf of Saint Lawrence events leads to a similar signature in the ocean, the salinity signature of the NCIS events is significantly different.
 The different salinity signatures lead to different ways in which the MOC recovers. After the Gulf of Saint Lawrence events the ocean returns almost exactly to its preperturbed state, with any remaining freshwater slowly being mixed out of the North Atlantic gyre circulations. The additional freshwater from the Heinrich events leads to slightly reduced overturning, where the reduction increases with the size of the event. The Hudson Strait events lead to greater changes in the post-Heinrich ocean compared to the preperturbed state. While the location of convection is still the same, the large freshening of the northwestern North Atlantic remains because of the relatively weak local circulation, and also leads to reduced densities over other parts of the North Atlantic. Although the maximum of the Atlantic overturning increases compared to the preperturbed state after the 0.2 and 0.4 Sv events, this also coincides with a more localized NADW cell. The exact value for the maximum of the NADW cell after recovery tends to be quite random, showing the sensitivity of this region of the ocean at the LGM.
 The greatest impact is seen after the NCIS events. The longer recovery periods show the freshwater is slowly flushed through the ocean thermohaline system. In this case the freshwater enters the ocean in a very sensitive location, and hits the convection area more directly than the other sources. The more northerly salinity anomaly in this case leads to southward shift of the convection area, and a more limited northward extent of the North Atlantic current, which is maintained after the Heinrich event. During the Heinrich event this creates more northerly cooling of the surface ocean, and together with the sea-ice-albedo feedback causes an increase in sea ice and much more northerly extent of atmospheric cooling than any of the other events, reaching as far north as Greenland. This can be seen in Figure 6, which shows January surface atmospheric temperature anomalies 300 years into the Heinrich event.
 The Gulf of Saint Lawrence event shows a more easterly atmospheric temperature anomaly, with greater cooling near the west European coast. In this case the local pressure increase around the source area due to the increased input of freshwater into the ocean leads to circulation changes, forcing a strong southwest flow in the northern part of the subtropical gyre, and limiting the northward extent of the North Atlantic Current. This leads to cooling of the sea surface and an increase in sea ice formation in the eastern part of the North Atlantic, which results in widespread atmospheric cooling in this region due to the sea-ice-albedo feedback. The northerly extent of the atmospheric cooling for the Hudson Strait event is limited. In this case the sea-ice-albedo effect provides the opposite effect, as the combined impact of the introduction of relatively warm freshwater (0°C) with no change in the convection area leads to some melting of sea ice.
 The shutdown, or partial collapse, of the MOC leads to a similar pattern of sea surface temperature (SST) anomalies in all cases. The SST anomalies for all events are shown in Figure 7. These anomalies are similar in size for the Hudson Strait and Gulf of Saint Lawrence events. The cooling in the NCIS events is slightly less pronounced because of the lesser impact on the MOC. There is cooling over much of the subtropical Atlantic, while the subpolar regions are not affected as much, because of the large-scale presence of sea ice. The SST anomalies in the simulations correspond reasonably well to available paleodata. The cooling off the Iberian margin corresponds well with Bard et al.'s  reconstructions from marine sediment cores for the last 3 Heinrich events. Cooling in the subtropical North Atlantic during H1 has been found in SST reconstructions from marine cores by Chapman and Shackleton . The warming of the Benguela current in the South Atlantic during H1 as found by Kim et al.  is also present in our simulations, although the main area of warming is in the Southern Ocean. Reconstructions by Cortijo et al.  of the H4 period (40–30 ka BP) show cooling of up to 2°C between 40 and 60°N in the North Atlantic, while salinity anomalies were restricted to south of 50°N. Our model results show similar sized cooling, however for the Hudson Strait and Gulf of Saint Lawrence events the cooling is limited to south of 50°N. The NCIS events do show more northward cooling reaching toward the Nordic Seas. However, these results also show that the model is too cold in the northern North Atlantic, where the cold SSTs cannot cool further.
6. Simulation of Heinrich Events as Freshwater Release
 To compare the impact of direct addition of freshwater to that of icebergs we introduced freshwater floods to the steady atmosphere-iceberg-ocean state, with the steady state icebergs. In this approach the meltwater is injected into the ocean directly as freshwater at the base of the ice streams, using the same fluxes as in the previous experiments (0.1 Sv, 0.2 Sv, and 0.4 Sv). The freshwater is added to the same model grid boxes that were used to seed the icebergs in the iceberg simulations. Time series of the MOC for these events from Hudson Strait, the Gulf of Saint Lawrence, and the NCIS are shown in Figure 8.
 The results are similar to the iceberg events, although the oceanic impacts are significantly larger than for iceberg events. The main differences occur for the NCIS events, where the larger floods (>0.1 Sv) cause numerical instability due to the large input of freshwater in a very sensitive area of the ocean. In this case the large increase in the sea surface height, and the pressure field, leads to numerical instabilities and a model crash for the 0.4 Sv event. In the case of the 0.2 Sv event, while the model recovers from an initially sharp oscillation in overturning, it is most likely that the associated numerical problems cause the rise in overturning following the removal of the flood after 500 years. To maintain comparability with other simulations the time step was not altered in these cases. The qualitative behavior of the Hudson Strait and Gulf of Saint Lawrence floods on the MOC are similar, causing a complete shutdown for 0.4 Sv, and partial collapse for smaller fluxes, with slightly shorter impact times than in the iceberg runs. Here the ocean circulation is less sensitive than the Nordic Seas, where the NCIS freshwater enters the ocean. The resulting sea surface salinity anomalies after 300 years are seen in Figure 9, showing a similar pattern, with larger anomalies, to the iceberg events. There are now very large increases in the sea surface height at the source regions, leading to large southward surface flows. Figure 9 also highlights the much larger surface impact of NCIS freshwater floods compared to the floods from the other sources.
 In this paper we have discussed the coupling of an iceberg model, with global iceberg seeding, to an intermediate complexity climate model for the simulation of Heinrich events. This approach allows iceberg meltwater to be gradually injected into the ocean along the iceberg trajectories. The resulting coupled model produces model states that are reasonably consistent for PD, and within the uncertainty bounds for the LGM period, especially in the Northern Hemisphere. It is also found that the inclusion of icebergs, from steady state ice sheets, does not have significant impact on the main ocean transports compared to model simulations without icebergs, as the associated fluxes of freshwater are relatively small. However, during Heinrich events the surface circulation, and therefore the iceberg trajectories, are significantly changed because of changes in the surface pressure field that are associated with the additional freshwater and changes in the heat fluxes.
 Heinrich events from all three sources require a flux of at least 0.4 Sv to completely shutdown the Atlantic MOC, for both iceberg surging and freshwater flood events. We have shown the ocean is almost as sensitive to events from European sources as Laurentide sources, for similar sized fluxes. However, it is unlikely that the NCIS produced fluxes as large as 0.4 Sv during the last glaciation [Nygård et al., 2007]. This flux is approximately equivalent to a sea level rise of 3.5 m per century. The collapse of the Atlantic MOC occurs within a century for all events, and therefore there is potentially a large enough supply of ice from Northern Hemisphere ice sheets [Peltier, 1994] to cause such a collapse within a century. However, it should be noted that the full 500 year duration of our simulated Heinrich events corresponds to approximately 17.5 m of sea level rise, which would likely require almost all of the ice contained within the Fennoscandian Ice Sheet to enter the ocean.
 The impact of the Hudson Strait and Gulf of Saint Lawrence events is found to be relatively similar. The impact of the NCIS events on the MOC is similar to the other events, though with somewhat slower and weaker impact. However, they lead to a significantly different anomaly pattern in the sea surface salinity field. This is due to increased variability in the ocean response, as much of the iceberg meltwater is released directly upstream of the Atlantic convection regions. The SST anomalies are similar for all events, and compare reasonably well to the pattern of Northern Hemisphere cooling from observations during Heinrich events.
 We compared the results of the iceberg surging events to similar freshwater floods, whereby the meltwater is injected directly into the ocean at the iceberg source locations. The latter approach is similar to typical Heinrich modeling studies that add freshwater directly to large areas of the North Atlantic [e.g., Ganopolski and Rahmstorf, 2001; Prange et al., 2004]. In our case we find that although the floods give qualitatively similar results, the impact is larger than for the iceberg events, with significantly more weakening of the MOC and colder SST anomalies. Therefore it is likely that modeling Heinrich events by directly releasing freshwater into the ocean will overestimate the impact, by neglecting the effect of gradual iceberg melt. The base model state with icebergs tends to be especially sensitive to large (>0.1 Sv) freshwater releases in the Nordic Seas region when compared to additional iceberg releases, with the model exhibiting instability behavior for the large freshwater floods from the NCIS region.
 The general sensitivity of our model to freshwater fluxes depends on the input location. In all the experiments in this study the freshwater and iceberg fluxes have been added to the ocean at the base of ice streams. Therefore all the meltwater does not reach the Atlantic convection area simultaneously. This reduces the impact on the MOC compared to less realistic experiments, whereby the freshwater flux is prescribed over a larger area of the North Atlantic. The fluxes needed to halt the MOC completely in other glacial climate models with such freshwater hosing lie in a range of 0.15–0.5 Sv [Ganopolski and Rahmstorf, 2001; Prange et al., 2004]. In our experiments we need a relatively large freshwater flux of 0.4 Sv to cause a complete shutdown of the MOC, because of our more realistic localized freshwater and iceberg input. This study thus suggests that MOC shutdown may be harder to induce than previously suggested.
 This work was funded as part of the NERC Rapid Climate Change program through grant NE/C509523/1. We would like to thank Tiago Silva for helpful discussions on the basal melting scheme in the iceberg model. We would also like to thank the editor and two anonymous referees who helped us to a more rigorous manuscript.