Climate Dynamics

, Volume 32, Issue 7–8, pp 1035–1054 | Cite as

Response in atmospheric circulation and sources of Greenland precipitation to glacial boundary conditions



The response in northern hemisphere atmospheric circulation and the resulting changes in moisture sources for Greenland precipitation to glacial boundary conditions are studied in NCAR’s CCM3 atmospheric general circulation model fitted with a moisture tracking functionality. We employ both the CLIMAP SST reconstruction and a modification thereto with reconstructions of glacial ice sheets and land masks. The individual components of the boundary conditions are added first one at a time and, finally, together. These steps show the atmospheric circulation to respond approximately linearly to the boundary condition changes, and the full glacial change may thus be decomposed into contributions from SST and topography changes, respectively. We find that using the CLIMAP SST reconstruction leads to a shift from Atlantic toward Pacific source regions not found with the modified reconstruction having cooler tropics and less sea ice. The occurrence of such a shift depends chiefly on the SST reconstruction and not on the existence of the large northern hemisphere glacial ice sheets. The influence of these circulation changes on important factors for ice core interpretation such as precipitation seasonality, condensation temperatures and source temperatures are assessed.

1 Introduction

Ice cores from the Greenland ice sheet are outstanding archives of past northern hemisphere atmospheric conditions. During the past decades, cores have thus been drilled in several Greenland locations in order to study past climatic conditions (Dansgaard and Johnsen 1969; Langway et al. 1985; Johnsen and Dansgaard 1992; Dansgaard et al. 1993; Mayewski et al. 1994; NGRIP members 2004). The abundance of one or both of the stable water isotopes (SWIs), H218O and HDO, are routinely measured throughout the depth of the core. Traditionally, the SWIs (typically written in δ notation relative to Standard Mean Ocean Water) are used as a proxy for local temperature at the time of deposition (e.g., Dansgaard and Johnsen 1969; Johnsen et al. 2001), while the so-called deuterium excess (δD−8δ18O) is interpreted as a proxy for conditions at the moisture source (e.g., Johnsen et al. 1989). However, factors such as wind speed, relative humidity and temperature at the time of evaporation and mixing and phase changes during transport complicate these relationships (e.g., Johnsen et al. 1989; Masson-Delmotte et al. 2005b).

In more recent ice coring efforts, SWIs are just a few out of a range of constituents that are measured (e.g., NGRIP members 2004), but they are nevertheless among the most important climatic proxies retrieved from ice cores (Johnsen and Vinther 2007). Despite the close connection observed between Greenland temperatures and δ18O during the past 100-200 years (Vinther et al. 2003, 2006), the nature of millennial scale δ18O variability is still poorly understood. Investigations of the Greenland ice cores that also included the available information from deuterium excess, have failed to produce δ18O based temperature reconstructions that are consistent with Greenland paleotemperatures derived from borehole thermometry (Masson-Delmotte et al. 2005b). Given our reliance on SWIs as a proxy for past temperatures, there is an urgent need to improve our understanding of the full range of mechanisms behind the observed SWI variability in the Greenland ice cores.

In this study we wish to dissect the relative roles of topography, sea surface temperatures (SSTs), sea ice, atmospheric transports, moisture sources and precipitation seasonality on Greenland precipitation-weighted temperatures. These are used here and by others (e.g., Krinner et al. 1997; Krinner and Werner 2003) as a proxy for SWIs. We employ a compartmentalization of the experiments reminiscent of those used by Rind (1987) and Yin and Battisti (2001). In the former, the LGM components are added in succession (SST, SST and flat ice sheets, SST and full-height ice sheets), while in the latter, different combinations of topographies and tropical and extratropical SSTs are used. We add topography and SST changes one at a time and finally together to yield the full set of LGM boundary conditions, and this permits us to assess the different contributions of the two components to the full change. To determine the sources of Greenland precipitation and monitor their changes with changing boundary conditions we follow the prescription by Joussaume et al. (1986) for moisture tracking (see description in the next section). This method has been used for over two decades (e.g., Koster et al. 1986, 1992; Charles et al. 1994; Armengaud et al. 1998; Delaygue et al. 2000; Werner et al. 2001; Noone and Simmonds 2002) in different models for both present day and glacial conditions, but we are unaware of any previous attempts to discern the relative roles of SSTs and topography in determining moisture sources and paths for Greenland precipitation. Comparisons of isotope records from multiple Greenland ice cores reveal that different locations are sensitive to different factors. For instance, Masson-Delmotte et al. (2005b) conclude that the NGRIP site may have a more stable moisture source during the Holocene than GRIP and NGRIP δ18O could therefore be the most reliable proxy for local temperature. GRIP deuterium excess may, on the other hand, to greater extent reflect North Atlantic conditions. Motivated by such inter-core differences, we have chosen to focus on six different Greenland ice core locations.

Atmospheric circulation changes stand out as a complicating factor for the simple temperature–SWI relationship. In a study by Schmidt et al. (2007) it was concluded that on modern-to-mid-Holocene timescales, local SWI signals at both low and high latitudes reflect non-local climate parameters rather than local ones. Even on present day time scales, interannual variability in the atmospheric circulation (such as the North Atlantic Oscillation, NAO) has influence on Greenland SWI composition unrelated to local temperatures (Sodemann et al. 2008b). Werner and Heimann (2002) found that precipitation-weighted inversion height temperatures, sea level pressure over Spain and relative humidity in a narrow mid-latitude Atlantic region constitute the best subset of climate variables to explain interannual δ18O variability. These findings are supported by the discovery of a significant NAO signal in δ18O from several Greenland ice cores (Vinther et al. 2003). This NAO imprint is in the companion papers by Sodemann et al. (2008a, b) shown to be a combined effect of variability in moisture sources and transport conditions. Cole et al. (1999) find that, on a world-wide scale, variations in moisture sources exert leading order influence on interannual δ18O variability. They speculate that this may be different for the larger temperature variations on ice age timescales, on which the traditional interpretation of δ18O reflecting local temperature and deuterium excess reflecting source regions probably is more valid. On orbital time scales, for instance, obliquity variations seem to influence the deuterium excess values in both Antarctic (Vimeux et al. 2001) and Greenland (Masson-Delmotte et al. 2005a, b) precipitation. One explanation for this obliquity imprint on deuterium excess is that obliquity variations dominate mean annual insolation at a given latitude (e.g., Loutre et al. 2004) and may thus induce changes in meridional SST gradients, meridional atmospheric moisture transports and, in turn, moisture sources.

Glacial atmospheric circulation changes arise as a combined response to topographic (massive ice sheets) and thermal (cooler SSTs and high-albedo continents) forcing (Rind 1987; Yin and Battisti 2001). Cook and Held (1988) compare stationary wave responses in a GCM and a linearized model, and conclude that the last glacial maximum (LGM) circulation was chiefly a linear response to the added ice sheets. Addition of the Laurentide ice sheet over northern North America induces a high pressure anomaly (relative to present day) centered slightly upstream of its peak (e.g., Rind 1987, his Fig. 22). This glacial anticyclonic circulation (as reported also by Cook and Held 1988; Krinner and Genthon 1998, and many others) advects cold air from the north onto the eastern flank of the ice sheet (near the east coast of northern North America) which influences the mass balance of the ice sheet itself (Roe and Lindzen 2001) but also tends to dampen the amplitude of the anomalous stationary wave (Cook and Held 1988). In this manner, the background meridional temperature gradient (and, by the thermal wind relation, the zonal wind) impacts the strength of the ice sheet-atmosphere interaction.

The influence of low-latitude SSTs on mid-to-high latitudes under LGM conditions has been investigated in several studies. Some conclude that mean tropical temperatures have a decisive impact on the radiative balance over the ice sheets (e.g., Rodgers et al. 2003). Yin and Battisti (2001), however, find that it is the tropical SST patterns rather than mean temperatures that dominate the high-latitude response. In fact, they conclude that low-latitude SST patterns play a role comparable to that of the ice sheet topography in determining the LGM stationary wave structure. Studying a subset of the Paleoclimate Modeling Intercomparison Project (PMIP, Joussaume and Taylor 2000) models, Kageyama et al. (1999) find that both the North Atlantic and North Pacific storm tracks shifted downstream in response to LGM conditions. This resulted in a reduction of the influence of North American continental conditions on the North Atlantic storm tracks relative to present day. Instead, sea ice conditions became the central influence on storm track position and extent (Toracinta et al. 2004; Byrkjedal et al. 2006).

Considerable uncertainty remains about the SST and sea ice distribution during the last glacial (see the discussion thereof in the next section). For a LGM scenario with a rather modest extent of sea ice in the Nordic Seas region, Byrkjedal et al. (2006) find that the Icelandic Low is much closer to its present-day state than in scenarios with more extensive sea ice. This central role of the sea ice thermal forcing on the regional atmospheric circulation is also called upon by Li et al. (2005) who conclude that North Atlantic/Nordic Seas sea ice reductions and concomitant shifts in precipitation seasonality are sufficient to explain the Dansgaard–Oeschger event temperature anomalies inferred from Greenland ice cores. They did, however, also speculate that these changes are not alone in determining the response; transport routes and moisture sources may also contribute.

This article is organized as follows: Section 2 describes the experimental configuration, the moisture tracking technique and the boundary conditions. In Sect. 3, results are presented and we show the changes in the regional atmospheric circulation, the changes at six Greenland ice core sites and the changes at the source areas for Greenland precipitation. Section 4 discusses the findings in relation ice core data and previous modeling studies, before conclusions are offered in Sect. 5.

2 Experimental configuration

In this study, the National Center for Atmospheric Research’s CCM3 (Kiehl et al. 1996, 1998) is used with a series of different combinations of present day (PD) and LGM boundary conditions. The CCM3 is an atmospheric general circulation model and is run with a spectral resolution of T42 (corresponding to approximately 2.8° × 2.8°) and 18 levels in the vertical. The CCM3 is distributed with a land surface model (Bonan 1998) with non-interactive, user-specified geographies of surface and vegetation types. In our experiments, the model system is run in specified SST mode, in which the user specifies mid-monthly climatologies of sea surface temperatures and sea ice distributions which are then interpolated to each time step. It is important to note that, in specified SST experiments like these, only ocean grid point temperatures are specified; those of land points are prognosed by the model’s land module.

The experiments were run for 20 years after an initial spin up period. All results have been compared with calculations based on the first and last 10-year periods, and all features discussed here are robust across the three calculations.

Seven different experiments were run with different combinations of PD and LGM (21 kyr BP) settings. The imposed boundary conditions in terms of SSTs, topographies, land masks, orbital parameters and greenhouse gas concentrations are discussed in the following along with a description of the changes to the standard distribution of the model performed in order to track moisture from source region to precipitation site.

2.1 Sea surface temperatures

Present day SSTs are taken from the Shea et al. (1992) global climatology giving monthly 2° × 2° SSTs based primarily on the 1950–1975 period. For the LGM, obtaining a reliable SST data set is less straightforward. The Climate: Long-range Investigation, Mapping and Prediction project (CLIMAP 1981, 1994) provided the first global systematic reconstruction which was based on the transfer function method on marine sediment faunal fossils. This data set displays only modest LGM cooling in the tropics compared to today (∼1.5 K cooler) and very extensive high-latitude sea ice with perennial coverage in the North Atlantic reaching as far south as the British Isles. Both of these features have been contested.

As reviewed by, for instance, Crowley (2000) and Toracinta et al. (2004) studies of terrestrial tropical snow lines, corals and noble gas solubility in groundwater indicate more substantial, for example, on the order of 5 K, cooling at low-latitudes, while marine evidence points to a more intermediate cooling of 2–4 K. Specifically, newer reconstructions reduce the extent of the areas within the North and South Pacific gyres in which CLIMAP displayed warmer-than-modern SSTs (e.g, Hostetler et al. 2006). At high latitudes, the low diversity of planktonic foraminifera reduces the reliability of the transfer function method employed by CLIMAP, and large parts of the Nordic Seas are now believed to have been at least seasonally ice free (e.g., Meland et al. 2005).

All our experiments involving LGM SSTs have been doubled, using both the CLIMAP reconstruction and a data set with appropriate low- and high-latitude corrections. For the CLIMAP reconstruction which only provides February and August climatologies, we followed the recommendations of the Paleoclimate Modeling Intercomparison Project (PMIP, Joussaume and Taylor 2000) and used a sinusoidal interpolation between these two, taking them as the summer and winter extremes. Currently, no common reconstruction like CLIMAP has been made with the updates necessary to match the current state of knowledge. There is still some debate (Crowley 2000) and corrections tend to be created from study to study where LGM SST boundary conditions are needed (e.g, Krinner and Genthon 1998; Delaygue et al. 2000; Werner et al. 2000, 2001; Crowley 2000; Yin and Battisti 2001; Toracinta et al. 2004; Bromwich et al. 2004; Li et al. 2005; Hostetler et al. 2006; Byrkjedal et al. 2006). Rather than create yet a correction, we have chosen to follow the directions of Toracinta et al. (2004). The corrections performed therein are not necessarily better than the others, but they are simple and have also been re-used in another study (Bromwich et al. 2004).

The directions for the “Toracinta et al. (2004)” LGM SST data are as follows: Annual average CLIMAP temperatures are calculated (from the February and August fields) and 4 K is subtracted between 30°S and 30°N with a linear ramping between 30° and 40°. From this resulting corrected annual average LGM field is subtracted the annual average PD field from Shea et al. (1992). This gives a corrected annual average LGM-PD difference which is then added to each month of the PD dataset. This gives the cooler LGM low latitudes and preserves the present day seasonal cycle leading to reduced high-latitude sea ice cover. The result is shown in Fig. 1, where annual averages are displayed along with the maximum and minimum sea ice extents.
Fig. 1

Annual average SSTs for a PD (Shea et al. 1992), b LGM (CLIMAP 1994) and c LGM with the corrections by Toracinta et al. (2004). In all panels, the thick white and gray lines show the February and August sea ice extents, respectively

2.2 Experiments and boundary conditions

For orbital parameters and greenhouse gases, guidelines from the Paleoclimate Modeling Intercomparison Project phase II (PMIP2, e.g., Crucifix et al. 2005) were used: Orbital parameters were taken for 0 kyr BP (eccentricity 0.016724, obliquity 23.446°, longitude of perihelion 102.04°) and 21 kyr BP (ecc 0.018994, obl 22.949°, lop 114.42°), and for greenhouse gas concentrations, pre-industrial (CO2 280 ppm, CH4 760 ppb, N2O 270 ppb) and LGM (CO2 200 ppm, CH4 350 ppb, N2O 190 ppb) values were specified. The solar constant was set to 1,365 Wm−2 in both cases.

Topographies change significantly between LGM and PD due to the enormous changes in continental ice sheets. Following again the recommendations of PMIP2, the ICE-5G reconstruction of Peltier (2004) was used. Figure 2 shows this reconstruction for the LGM in the North American region. Comparison of the two panels demonstrates the effect of smoothing associated with the T42 model resolution: The structure of the Laurentide ice sheet is reasonably conserved but the Rockies have undergone significant smoothing. The Brooks Range in northern Alaska is represented but only by one latitude band. The ICE-5G reconstruction was also used to determine the land masks (the difference in ice mass between the LGM and PD reconstructions corresponds to a sea level rise of approximately 125 m). To avoid further complexity and due to the lack of unequivocal data, LGM surface type and vegetation was only changed from that of present day where ice was present.
Fig. 2

Demonstration of the smoothing of topography associated with the T42 model resolution. Shown is the Peltier (2004) LGM topography at a 1° × 1° resolution and b model physical grid resolution after having been truncated to T42. Present day coastlines are shown to ease readability

With these components of PD and LGM boundary conditions, seven combinations were chosen (summarized in Table 1):

The control experiment with PD SSTs, topography, orbital parameters, greenhouse gas (GHG) concentrations and land mask.


Orbital parameters, GHGs and land mask were changed to LGM values. This case was included to provide a control against which only SSTs and topographies were changed.


As in CTRLLND, but with topography (and surface type in glacier points) changed to LGM values.


As in CTRLLND, but with SSTs changed to CLIMAP LGM values.


Both topographies and SSTs changed to LGM values.


As SST but with LGM SSTs following the directions of Toracinta et al. (2004)


As LGM but with LGM SSTs following the directions of Toracinta et al. (2004).

The experiments were compartmentalized to facilitate identification of the contributions to the full LGM changes by topography and SST changes, respectively. Relative to the CTRLLND case, it may loosely be said that “SST+TOPO=LGM” and that “SSTT + TOPO = LGMT”.
Table 1

Overview of the combinations of boundary conditions for the seven experiments. See text for details





Land mask


Shea et al. (1992)





Shea et al. (1992)





Shea et al. (1992)





CLIMAP (1994)





CLIMAP (1994)





Toracinta et al. (2004)





Toracinta et al. (2004)




2.3 Moisture tracking

For this study we have fitted the NCAR CCM3 with a moisture tracking functionality like those employed earlier in both the NASA GISS (by, e.g., Koster et al. 1992; Charles et al. 1994; Armengaud et al. 1998; Delaygue et al. 2000) and the ECHAM-4 (Werner et al. 2001) GCMs. The Earth’s surface is divided into a number of regions (in our case 17), and moisture evaporated from each of these is treated as a separate tracer until it leaves the atmosphere (due to, for instance, precipitation or condensation onto the land or ocean surface). During their lifetime in the atmosphere, the transports and phase changes of the tracers parallel those of the model’s moisture field. The tracers are thus subject to the same tendencies as the total moisture field and since the source regions cover the whole globe, the sum of the tracers equals, at each time step and in each grid box, the model’s moisture content. In this manner, precipitation at any point can be decomposed into its contributing source regions. Further details of the moisture tracking are given in the Appendix.

The source region division is shown in Fig. 3a and was aimed at yielding information on which seas, oceans and land areas contribute to Greenland precipitation. Resolution is thus best in the Northern Hemisphere and especially around Greenland. The source regions are: Labrador Sea/Davis Strait (2), Greenland-Iceland-Norwegian Seas (3), northern North Atlantic (4), North Atlantic (5), South Atlantic (6), Arctic Ocean (7), northern North Pacific (8), North Pacific (9), South Pacific (10), Indian Ocean (11), Southern Ocean (12), Mediterranean Sea (13), “other continents” (14), Greenland (15), ice core grid cells (16), North America (17) and Eurasia (18). The regions are held fixed, and shifts in source areas under changing boundary conditions may be monitored. When the land mask changes from present to LGM conditions, points that before were ocean points will be included in the nearby land regions. This, however, has no effect on the results shown here, since (as described next) changes are shown relative to the CTRLLND experiment which uses the LGM land mask.
Fig. 3

a The source regions into which the Earth’s surface is divided. Tracer numbers start from 2 (since tracer 1 is the total moisture field). To allow discounting precipitation stemming from local sublimation, a source area (no. 16) of the grid cells containing the ice core locations has been considered. b The locations of the six ice core sites which will be considered (crosses). The four dots surrounding each site show the four grid points used in the bilinear interpolations used to determine the site values

3 Results

This section will be organized as follows. First, the overall Northern Hemisphere circulation, the changes therein and the changes in precipitation patterns will be described. This will be followed by a detailed look at the consequences for precipitation at six Greenland ice core locations (see Table 2; Fig. 3b) and the changes at the source regions for Greenland precipitation.
Table 2

The names, abbreviations and latitude/longitude positions of the six ice core locations considered. See map in Fig. 3b



Lat (°N)

Lon (°W)

Camp Century
























For the results presented here, the differences between the CTRL and the CTRLLND experiments are small compared to the changes associated with the other boundary condition variations. This is due to the fact that the SSTs are fixed to the present day values; if SSTs were allowed to adjust, the drop in greenhouse gas concentrations would cause the global climate to cool. In fact, the climate sensitivity (equilibrium response in global mean surface temperature to a doubling of CO2) of the CCM3 at the present resolution coupled to a slab ocean model is 2.3°C (Kothavala et al. 1999). However, since the CTRL and CTRLLND climates are so alike, we will in the following present all results relative to the CTRLLND experiment. This has the advantage that, when viewing results from, for instance, the TOPO or SST experiments, only topographies or SSTs have been changed; there are no land mask, orbital parameter or greenhouse gas changes.

3.1 Atmospheric circulation

Figure 4 displays annual average 500 hPa height changes in the experiments relative to CTRLLND. The two bottom panels (Fig. 4f, g) show the sum of the changes in the TOPO and the two SST experiments. Since the boundary condition changes in these experiments add up to the full LGM boundary condition changes, these panels indicate the degree to which the system responds linearly in the 500 hPa height field. The result is convincing: The full LGM pattern (Fig. 4d) does not resemble the TOPO (Fig. 4a) nor the SST (Fig. 4b) pattern, but their sum (Fig. 4f) is very close. The same may be said for the two modified glacial temperature experiments: Added to the TOPO results, the SSTT (Fig. 4c) experiment yields a pattern (Fig. 4g) which is quite close to that of the full LGMT (Fig. 4e) experiment. The linearity displayed by the atmospheric circulation in terms of the 500 hPa height field is, however, not surprising given the success of linearized, steady-state models (as in, e.g., Hoskins and Karoly, 1981) in producing the steady response to thermal and orographic forcings. It is an extremely useful feature allowing us to attribute certain characteristics of the atmospheric response to certain components of the forcing. Plots similar to Fig. 4 for summer and winter (not shown) reveal that the approximate linearity holds also for these individual seasons. The winter, however, has the largest response and thus also displays the largest deviations from linearity.
Fig. 4

Changes in annual average 500 hPa height fields (in m) relative to CTRLLND in experiments a TOPO, b SST, c SSTT, d LGM and e LGMT. f, g are the sums (a) + (b) and (a) + (c), respectively. The resemblance of these sums to the full LGM runs (d, e) demonstrates the linearity of the response in this field to the boundary conditions

The topographic forcing (Fig. 4a) contributes mainly with an increase in the 500 hPa height, especially in the Canadian Arctic and the Beaufort Sea region. The CLIMAP temperatures (Fig. 4b) contribute with a lowering of the 500 hPa height due to the colder temperatures, located mainly around high latitude ocean areas (where CLIMAP SSTs display the largest decreases). This lowering of the 500 hPa height is maximum near the present-day Aleutian Low. The Toracinta SSTs (Fig. 4c) yield a more uniform lowering of the height field, in line with the greater low-latitude cooling and smaller high-latitude cooling. The continental interiors are least affected by the SST changes.

In the mean sea-level pressure (not shown), the most pronounced difference between the two SST experiments is due to the differences in the location of the North Atlantic sea ice lines (Fig. 1). Typically, CLIMAP SSTs yield a weakening of the winter season Icelandic Low relative to present day (e.g., Rind 1987), but the much larger turbulent heat fluxes to the atmosphere in the SSTT and LGMT experiments fuel cyclogenesis in the region and yield a reduced weakening of the Icelandic Low compared to that seen with CLIMAP SSTs [as noted by Toracinta et al. (2004) and Byrkjedal et al. (2006)].

The changes in 500 hPa height shown in Fig. 4 impact the stationary wave pattern upstream of Greenland. To more clearly demonstrate the flow in this area, Fig. 5 shows the winter 700 hPa height field. This somewhat lower level is shown since we are interested in moisture transports which mainly occur closer to the surface than at the 500 hPa level. The pressure ridge forced by the Rockies apparent in the control (Fig. 5a) is strongly reinforced by the existence of the Laurentide Ice Sheet (Fig. 5b). The SST experiment (Fig. 5c) displays a shallowing of the trough downstream of the Rockies, probably as a result of anomalous Rossby wave trains set up (as in Hoskins and Karoly 1981) by the negative thermal forcing in the mid-latitude western Pacific and the positive low-latitude forcing associated with the warmer Pacific gyre. The former is weaker and the latter is absent in the SSTT experiment and the shallowing of the trough is less pronounced (Fig. 5d). The cross-Arctic flow in the SST experiment (Fig. 5c, arrow) is qualitatively different with the changed thermal forcing (Fig. 5d), and these differences carry over to the LGM and LGMT experiments (Fig. 5e, f). In LGM, the low-to-mid-tropospheric flow over Greenland is dominantly northerly with air masses originating over the North Pacific. A more local cyclonic circulation dominates the Greenland region in the LGMT experiment making the circulation here resemble that of CTRL. As will be shown later, these two experiments, in fact, also have rather similar moisture source distributions.
Fig. 5

The stationary wave pattern in the winter (DJF) 700 hPa fields (in m) over the North American continent in experiments a CTRLLND, b TOPO, c SST, d SSTT, e LGM and f LGMT. The dots indicate the position of the Greenland ice core sites considered here, the arrows indicate approximate circulation directions

The spring and autumn patterns qualitatively resemble those shown for winter in Fig. 5 and the circulation differences described above thus carry over to the annual mean in spite of a more sluggish summer circulation (none of these seasons are shown) without the clear differences. While Fig. 5 shows the time mean situation, the height field weighted by Greenland precipitation (in line with the precipitation weighted quantities shown in the following) could alternatively have been calculated, thus focusing on the circulation patterns associated with Greenland precipitation. The simpler time mean circulation patterns displayed are, nevertheless, consistent with the moisture source differences described in the following.

Turning to the precipitation fields (not shown), linearity is not expected to hold as well as it did for the pressure fields due, for example, to the non-linearity of the Clausius–Clapeyron relation. Both when looking at absolute and relative changes, the glacial reduction in precipitation is exaggerated in the linear estimates in areas where both the topographic and thermal forcings yield decreases, such as southwest of Greenland, south of Iceland and west of Norway. These maritime areas experience, in the LGM and LGMT experiments, an annual mean drying of up to 80% locally. The fact that even the relative changes display non-linearity indicates that it is not only the exponential form of the Clausius–Clapeyron relation which is responsible for the non-linearity. This agrees with the finding of Krinner and Genthon (1998) that the Laurentide ice sheet induces circulation changes which modify precipitation rates beyond what is dictated by atmospheric temperatures alone. To the degree that the LGM and LGMT changes can be decomposed into the topographic and thermal contributions, however, the SST and SSTT changes tend to dominate. This is not surprising, considering the decrease in evaporation and holding capacity of the lower troposphere associated with the large temperature changes the surface undergoes in these experiments.

3.2 Site conditions

3.2.1 Precipitation rates

Figure 6a–c shows precipitation rates at six Greenland ice core sites (Table 2) for each of the experiments in annual average, winter and summer. Table 3 compares the annual average values for present day to water equivalent accumulation rates inferred from ice cores (Johnsen et al. 1992a; NGRIP members 2004) .1 In this and all of what follows, core site values are calculated with a bi-linear interpolation using the four nearest grid points. With the exception of the Dye-3 core site, the numbers are in fairly good agreement. The Dye-3 site is the closest to the North Atlantic storm track, and small inaccuracies in the position of the latter, combined with the relatively steep local topography, may give rise to the problems encountered here. In general, however, the model seems to do a reasonable job of duly reproducing modern annual accumulation rates, and we may expect the paleo-accumulations to be equally reasonable—within the accuracy of the prescribed boundary conditions.
Fig. 6

Water equivalent accumulation rates at the six ice core sites (see Fig. 3b) in the different experiments for a annual average b winter and c summer. Note that since the numbers are accumulation rates (in units of cm/year), the DJF and JJA values do not yield the winter and summer contributions; these are obtained as 3/12 of the rates. The axis in b is half of those in a and c. d Ratio of DJF to JJA precipitation rates indicating the seasonality. A value less than that in CTRL denotes a shift towards larger relative summer precipitation and vice versa

Table 3

“Observed” present day annual average water equivalent accumulation rates inferred from ice cores (Johnsen et al. 1992a; NGRIP members 2004) compared to modeled precipitation rates (with standard deviation). Units are cm/year






31 ± 7



22 ± 4



16 ± 2



17 ± 2



52 ± 9



85 ± 9

aNEEM shallow core (A. Svensson, 2008, personal communication)

When going to LGM and LGMT conditions (Fig. 6a) all sites experience a drying, an effect which is most pronounced at the southeastern sites (RL and D3). There is, however, a qualitative difference between the changes at these sites and at the rest: Between the LGM and LGMT experiments RL and D3 encounter increases as opposed to the slight decreases seen at the other sites. The southeastern sites are more directly influenced by the cooler North Atlantic and reduced evaporation while the rest are governed by the changes in transport routes. The precipitation rates do not display linear responses to the boundary conditions as we saw for the 500 hPa height field. Especially for the RL and D3 sites, addition of topography alone has very limited effect (for ANN, DJF and JJA). Comparing SST and LGM, however, addition of topography has a marked drying effect. This can be seen from Fig. 9 (which will be discussed later) to be due to changes in the continental North American source: In both CTRL and TOPO, the sources are mainly Atlantic and addition of the Laurentide ice sheet has limited effect on the evaporative input relevant to Greenland. When Atlantic temperatures are lowered in the SST experiment; however, the North American source becomes more important and shutting this off by addition of the Laurentide in LGM has a large effect.

Figure 6d shows the seasonality of the accumulation rates by plotting DJF/JJA ratios. Validation of the present day ratio is not straightforward since there (to our knowledge) are no good, seasonally resolved observations of precipitation on the Greenland ice sheet. Steffen and Box (2001) report automatic weather station (AWS) estimates of monthly surface height changes from the period 1995–1999. These height changes are the combined result of precipitation, firn compaction, redistribution of snow by drifting and sublimation/condensation and can thus not be compared directly to our precipitation rates. For the Summit station, where the model (GR) gives a large summer dominance, the AWS shows no significant seasonality (although larger summer precipitation may be countered by firn compaction). At South Dome (∼200 km south of D3, where the model also shows larger summer rates), the AWS shows larger height changes during winter, but here the warm summer temperatures certainly give significant summer compaction contributions. Alternative information comes from Cappelen et al. (2001) who report monthly resolved climatologies from coastal stations around Greenland. All along the west coast, the stations have a summer maximum in precipitation. On the south-eastern coast, they have a winter maximum and on the mid-eastern coast (near RL), they show equal summer and winter rates. The modeled summer dominance at the CC, NE and NG sites thus resemble the west coast observations (although somewhat exaggerated). The RL site, showing almost equal summer and winter rates, is well in line with the observations. The summer dominance at the GR site is in line with the west coast observations but seems somewhat at odds with the flat AWS distribution of Steffen and Box (2001). The D3 site is close to both the west and east coasts showing summer and winter dominance, respectively, and it is thus unclear whether the modeled seasonality is realistic.

Going to the other experiments in Fig. 6d, a decrease in the DJF/JJA ratio relative to the CTRL value denotes a shift toward greater summer weighting and vice versa. Most sites encounter a shift toward greater summer weighting in most experiments. Exceptions are the northwestern sites, for example, CC, NE and NG, when CLIMAP SSTs are used (exps SST and LGM). These sites do not suffer under the vast winter North Atlantic sea ice cover, but are rather affected by the circulation changes and increased Pacific storminess which is most pronounced in winter.

3.2.2 Precipitation weighted temperature

Having no water isotope module in the model configuration, we perform an online calculation of precipitation weighted condensation temperature,
$$T_{pw}(x,y)=\frac{\sum_{z,t} T(x,y,z,t) P(x,y,z,t)}{\sum_{z,t} P(x,y,z,t) }, $$
where T is atmospheric temperature, P is precipitation and sums are taken layer-by-layer and timestep-by-timestep for each (x, y)-gridcell (as in, e.g., Krinner et al. 1997). This diagnostic is the time average temperature at which the precipitation condenses out of the cloud as recorded in the accumulated snow (although a caveat is that if moisture condensing at one level re-evaporates at a lower level, it will enter into the contribution from the original level and deduct from the contribution from the lower level). Tpw thus crudely parallels the local temperature recorded by δ18O in ice cores. Figure 7 shows Tpw for each core and each experiment. Due to the larger summer accumulation, the annual average picture (Fig. 7a) naturally resembles that of JJA (Fig. 7c) more than DJF (Fig. 7b). In the latter, the SST and LGM curves and the SSTT and LGMT curves are pairwise alike, indicating that topography plays less of a role for the temperature recorded, while the choice of SST-reconstruction is important. Conversely, for JJA (Fig. 7c) the SST and SSTT curves and LGM and LGMT curves are pairwise alike. This indicates that, in summer (which has the larger weight on the annual average), the two SST-reconstructions lower Tpw and addition of LGM topography lowers it even further. On annual average (Fig. 7a), the LGMTTpw is colder than that for LGM for all sites except the two northwestern-most ones (CC and NE). This difference between the core sites will be demonstrated to owe to changes in the seasonality of the precipitation rather than local condensation temperatures.
Fig. 7

Precipitation weighted condensation temperature (see text) at the ice core sites in the different experiments. a Annual average, b winter, and c summer

3.2.3 Moisture sources

The moisture tracking described in Sect. 2.3 (and the Appendix) enables us to decompose the precipitation at the sites into contributions from the prescribed source regions shown in Fig. 3a. Examples of this breakdown are shown for the control, LGM and LGMT experiments at the GRIP core site in Fig. 8. The error bars show one standard deviation and thus provide a measure of the magnitude of changes necessary for them to be discernible from noise. The DJF case (Fig. 8b) clearly demonstrates the consequences of the circulation differences seen in Fig. 5: While the control and LGMT experiments have major contributions from the North Atlantic (regions 3,4 and 5), the LGM experiment encounters a clear shift toward North Pacific (regs 8 and 9) moisture origin. The distributions are quite different in JJA (Fig. 8c) where the major changes are due to the addition of the Laurentide Ice Sheet which lowers the North American (reg 17) contribution and the effect of the general decrease in precipitation enhancing the role played by local Greenland sublimation (reg 15). The large contribution of continental water sources seen in our control is in agreement with the findings in similar moisture tracking experiments (e.g., Charles et al. 1994; Armengaud et al. 1998; Werner et al. 2001). Due to the large summer bias in the precipitation seasonality at GR, the annual mean picture (Fig. 8a) is dominated by the shift from North American to Greenland moisture. The spring and autumn distributions (not shown) resemble those of winter, however, and a LGM decrease in the northern North Atlantic source and increase in the northern North Pacific source are discernible over the noise level.
Fig. 8

The relative contributions to GRIP precipitation a annually, b in winter and c in summer from the pre-defined source regions in Figure 3a. Data is shown for the CTRLLND, LGM and LGMT experiments. Note that contributions from the “local” ice core sites (region 16) are excluded in all calculations. The error bars denote one standard deviation as calculated from the 20 model years

An overview of the moisture source changes is given in Fig. 9 for all sites and all experiments. Here, the 17 regions have been grouped into the North Atlantic (regs 2,3,4,5), the North Pacific (regs 8 and 9), Greenland (reg 15), North America (reg 17) and the rest of the world. Again, the DJF picture (Fig. 9, second row) clearly demonstrates the change to cross-Arctic flow in SST and LGM (Fig. 5c, e): all cores see a shift toward dominance of the North Pacific source. In the SSTT experiment, a slighter, but still discernible, shift to Pacific influence is seen at all sites due to the relatively larger winter sea ice advance in the Atlantic than in the Pacific. This shift is cancelled, however, in the LGMT experiment, where Fig. 5f displays a more closed, local Greenland–Atlantic cyclonic circulation. The source changes are much less clear in JJA (Fig. 9, third row) where the most prominent feature again is the reduction of the North American contribution in the TOPO, LGM and LGMT experiments. The mean annual picture in Fig. 9 (first row) retains both of these major features: Reduced contributions (1) from the North Atlantic in SST, SSTT and LGM (except at D3) and (2) from North America in TOPO, LGM and LGMT.
Fig. 9

Relative contributions of five major areas to precipitation by site, season and experiment. The areas are North Atlantic (gray, regs 2,3,4,5), North Pacific (white, regs 8, 9), Greenland (black, reg 15), North America (gray, reg 17) and the rest of the world (white). For each site-by-season cluster, the bars denote from left to right CTRLLND, TOPO, SST, SSTT, LGM and LGMT

One might suspect that the Pacific source and the cross-Arctic transport in SST and LGM could be an artifact of the topographic smoothing associated with the T42 model resolution, for example, that at higher resolution higher topography would more efficiently drain the moisture out of the atmosphere before reaching the Arctic Ocean and eventually Greenland. However, the height of the Brooks Range and the Mackenzie Mountains, which judged from the circulation patterns in Fig. 5c and e would constitute the prime blockers of the moisture transport, are not significantly lowered by the smoothing. Taking an approximate mean range height by averaging along latitude 68°N from 145–155°W, the Brooks Range is lowered from ∼1,400 m (at 1° × 1° resolution) to ∼1,200 m (at T42 resolution). For the Mackenzie Mountains we average, as an example, along longitude 129°W from 57 to 64°N, and for both resolutions the height is just under 2,600 m. Topographic smoothing thus seems incapable of playing the decisive role for the cross-Arctic transport in the present model configuration.

3.2.4 Contributions to changes in Tpw

Having shown the changes in source regions under the changing boundary conditions, it becomes interesting to return to the precipitation weighted condensation temperature, Tpw, but this time calculated region-by-region:
$$T_{pw}^r(x,y)=\frac{\sum_{z,t} T(x,y,z,t) P^r(x,y,z,t)}{\sum_{z,t} P^r(x,y,z,t) }, $$
where the r-superscript denotes the component of the total precipitation originating from the rth region. This quantity is shown for the CTRL, LGM and LGMT experiments at the GRIP site in Fig. 10a. A look at the CTRL curve gives a good indication of what this figure can tell us: The more distant a source region is, the higher (and thus colder) the level from which its contribution condenses. The North Atlantic (regs 2–5) and the Arctic Ocean (reg 7), for instance, are close and moisture is advected at lower levels to the GRIP site than is the case for the South Atlantic (reg 6), Pacific (regs 8–10), Indian (reg 11) and Southern Oceans (12). In the LGMT experiment, the low latitudes are 4 K colder than in the LGM experiment, and this has a considerable cooling effect on the high-latitude atmosphere. This was demonstrated in the study by Alexeev et al. (2005) where a uniform low-latitude surface temperature change was more efficient at altering 700–250 hPa temperatures at 80°N than a local high-latitude surface temperature change. This explains why condensation temperatures are generally lower in LGMT than in LGM. An exception to this is in winter (not shown) for precipitation originating from regions 3 and 4. This occurs predominantly when air masses are advected from these adjacent waters where the sea ice cover differs the most between the two experiments. Hence, in LGM, it will generally be colder air masses carrying this moisture than in LGMT.
Fig. 10

a Annual average precipitation weighted condensation temperatures at GRIP for the contributions from the different source regions. Although the regions are physically unconnected, thin connecting lines have been included to ease readability. b, c Annual average changes in precipitation weighted condensation temperature at all sites between CTRLLND and LGM (b) and LGMT (c) using only change in seasonality (dash-dotted), local temperature (dashed), source region distribution (dotted), and the full experiment (thick solid). The thin solid curve is the sum of the three former, and shows the degree to which a linear decomposition is valid. See text for details

Given the fractional contributions to total precipitation (as shown in Fig. 8), denoted in the following by fr, and the precipitation weighted condensation temperatures for each moisture source (as shown in Fig. 10a), Trpw, we may write
$$T_{pw} = \sum_{r} T^r_{pw}f^r.$$
If we consider this to represent the control experiment and then turn to, say, the LGM experiment where TrpwTrpw + δTrpw and frfr + δfr, we have
$$ \begin{aligned} T_{pw}+\delta T_{pw} &= \sum_{r} ( T^r_{pw}+\delta T^r_{pw})( f^r+\delta f^r) \\ &\simeq \sum_{r} (T^r_{pw}f^r +\delta T^r_{pw}f^r+T^r_{pw}\delta f^r), \\ \end{aligned} $$
where the second order term has been dropped. Having interest in the temperature signal that would be recorded in the accumulated snow pack, we need to consider the associated precipitation weighted annual means of these quantities. Our model outputs monthly mean values of the fields, and the monthly precipitation rates are then used in the weighting. Since the seasonality of precipitation changes between the experiments (cf. Fig. 6d), we must be careful to choose the right seasonal weighting in our averages. The change we would see in an ice core is, in fact,
$$\delta \langle T_{pw} \rangle = \langle T_{pw}+\delta T_{pw}\rangle_{\rm LGM} - \langle T_{pw}\rangle_{\rm CTRL}, $$
where the angled brackets denote annual means and the LGM and CTRL subscripts detail which month-to-month precipitation weighting is employed. Inserting our expression from above, shortening the subscripts to “L” and “C” and taking sums over the repeated r-indices as implicit, we obtain
$$ \begin{aligned} \delta \langle T_{pw} \rangle &\simeq \langle T^r_{pw}f^r +\delta T^r_{pw}f^r+T^r_{pw}\delta f^r \rangle_{\rm L} - \langle T_{pw}\rangle_{\rm C} \\ &= \underbrace{\langle T_{pw}\rangle_{\rm L} - \langle T_{pw} \rangle_{\rm C}}_{\rm Seasonality} + \underbrace{ \langle \delta T^r_{pw}f^r \rangle_{\rm L}}_{\rm Temperature} +\underbrace{ \langle T^r_{pw}\delta f^r \rangle_{\rm L} }_{\rm Src. distr.}.\\ \end{aligned} $$
Here the contributions to the change in annual average Tpw are from the change in seasonality, the change in the local condensation temperature and the change in the source region distribution. The former is simply the difference one obtains using control experiment Tpw but two different month-to-month weightings in the annual average.

This breakdown is shown for LGM and LGMT in Fig. 10b and c, respectively. The accuracy of our linear decomposition, indicated by the degree to which the thick and thin solid curves are equal, is seen to be quite convincing. The full LGM/LGMT curves (thick solid) are exactly the differences between the corresponding curves and the CTRL curves in Fig. 7a. The major contributor to these changes is in both cases the decrease in local condensation temperature. In LGMT (Fig. 10c), this cooling is offset everywhere by about 3 K due to the combination of seasonality and distribution changes. As seen in Fig. 6d, all cores experience a shift towards greater summer weighting in LGMT and this yields the warming effect shown by the dash-dotted curve. The small positive contribution from the distribution change owes mainly to the shift from the more distant North American source to the local Greenland source seen in all cores (Fig. 9, upper row).

In the LGM experiment (Fig. 10b), the results are less spatially uniform: The temperature contribution is smaller than in LGMT (due, as mentioned, to the much warmer low- to mid-latitudes) but has most weight closest to the heavily sea ice-influenced North Atlantic. This gradient is partially offset by an opposite gradient in the seasonality effect, which is positive in the southeast where there is a shift toward summer weighting and negative in the northwest where there is a shift toward winter weighting (Fig. 6d). In fact, this difference in the seasonality changes is exactly what allows the total Tpw cooling in LGM at the northwestern-most sites to be as large or larger than in LGMT despite the weaker direct cooling. The source distribution effect is small due to the competition between the cooling by the shift from Atlantic to Pacific sources and the warming by the shift from North American to Greenland sources. In both Fig. 10b and c, the changes at the NG and GR sites are almost equal, and in both cases is this due to a cancellation between temperature and seasonality effects.

3.3 Source conditions

Analogously to the precipitation weighted condensation temperature discussed in the previous subsection, we also performed an online calculation of evaporation weighted surface temperatures,
$$T_{ew}(x,y)=\frac{\sum_{t} T_{S}(x,y,t) E(x,y,t)}{\sum_{t} E(x,y,t) }. $$
Within each source region, this quantity was then evaporation averaged (each point is weighted by its contribution to the region’s entire evaporation) to give monthly values of the regional evaporation weighted temperature, Tewr. This is the temperature signal, if any, which is carried with the water vapor. These temperatures can be combined with the source region distribution for precipitation for each core site to give the regionally weighted evaporation temperature,
$$T_{src} = \sum_{r} T^r_{ew}f^r.$$
To the degree that ice core deuterium excess records contain source region temperature information [as discussed by Sodemann et al. (2008b), at least for inter-annual variability, this may not be the case], it is this weighted evaporation temperature (as also calculated by, e.g., Delaygue et al. 2000) being recorded. Annual, winter and summer (precipitation weighted) mean results are shown in Fig. 11 for CTRL, LGM and LGMT. In the latter case, Tsrc changes by approximately 8–10 K at all sites rather homogeneously over the year. The annual mean LGM changes are considerably smaller with only small winter changes and much larger changes in summer.
Fig. 11

Source temperatures calculated from source distributions and evaporation weighted surface temperatures at the sources for CTRLLND, LGM and LGMT. a Annual average, b winter and c summer

A linear decomposition like the one performed for Tpw in Eq. (5) has also been attempted for the source temperatures. Results are shown in Fig. 12 and non-linear effects clearly have larger impact in this case. The cooling due to temperature changes is again dominant and the seasonality effects also resemble those for Tpw. In LGMT, the shift from North American to Greenland influence has a cooling effect, which in LGM is partially offset by the shift from Atlantic to Pacific influence. While regions 4 and 5 have the approximate same evaporation temperatures as regions 8 and 9, the loss of moisture from the cold regions 2 and 3 has a warming effect.
Fig. 12

Annual average changes in source temperature for the ice core sites between CTRLLND and LGM (a) and LGMT (b) using only change in seasonality (dash-dotted), temperatures at the sources (dashed), source region distribution (dotted), and the full experiment (thick solid). The thin solid curve is the sum of the three former, and shows the degree to which a linear decomposition is valid. See text for details

Especially in the LGM experiment do non-linear effects play a role; the linear decomposition is incapable of capturing the much smaller cooling in LGM than in LGMT seen in Fig. 11a. The difference is largest in LGM because an important non-linear contribution derives from the decrease in moisture from an area of large cooling (Atlantic) combined with a similar increase from an area that cools much less (Pacific). With subscripts A and P for the two oceans, respectively, we can approximately write for the non-linear term,
$$ \begin{aligned} \delta T_{NL} &\simeq \delta T_{A} \delta f_{A}+ \delta T_{P} \delta f_{P} \\ &\simeq \delta f_{P} ( \delta T_{P}-\delta T_{A}). \\ \end{aligned} $$
Since both δfP and δTP − δTA are positive, this non-linear contribution has a warming effect. It is strongest in winter where the shift toward Pacific moisture origin is most pronounced and is the reason for the different LGM winter behavior seen in Fig. 11.

4 Discussion

The temperature/δ18O relationships reviewed by Johnsen et al. (2001) differ by approximately a factor of 3 depending on whether condensation temperatures or time mean surface temperatures (as inferred from borehole thermometry) are considered. This agrees well with our results in which both glacial runs produce condensation temperature cooling of 6–8 K (Fig. 10b, c) and time mean surface temperature cooling of up to 20 K (not shown). Krinner et al. (1997) and Werner et al. (2000) have previously found that much of the discrepancy between present day-tuned temperature/δ18O relationships and those inferred from boreholes stems from seasonality changes rather than atmospheric inversion height temperature changes. In our case, the “Temperature” curves in Figs. 10b and c show only a cooling of ∼10 K (rather than the 15–20 K at the surface) and the seasonality bias is only a few degrees. The “Temperature” curve carries the combined effects of condensation height temperature changes and the short-term (sub-monthly) precipitation bias; temperatures are recorded only during snow events and the very cold spells dragging down time-mean temperatures are not recorded. Our results are thus consistent with the suggestion of Johnsen et al. (2001) that “precipitating clouds retain much of their warmth, even during times of cold glacial climate”.

Based on GRIP deuterium excess and δ18O, Masson-Delmotte et al. (2005a) infer present-day to glacial source temperature cooling of about 6 K. This is much like the number we get for LGM (Fig. 12a), while the LGMT number is slightly larger 8.5 K (Fig. 12b). It is worth noting, however, that our source temperature calculations are limited by the source region resolution. We are thus reluctant to claim that this lends credence to one simulation over the other, but remark that both results are roughly in accordance with ice core inferences. This source region cooling is opposite to the GCM results of Delaygue et al. (2000) for Antarctica in which source temperatures increase when using CLIMAP SSTs due to sufficient equatorward shifts of source regions. This effect is absent when they use a cool-tropics version of CLIMAP.

Contrasting the results for the LGM and LGMT runs, it is intriguing that the differences in source areas for the precipitation arriving in Greenland are so large. In the LGM run most ice core sites on the Greenland ice sheet receive significantly increased relative amounts of precipitation from Pacific sources as compared to the CTRL run (Fig. 9). This change toward a Pacific source region is not seen in the LGMT run that shows unchanged or even decreasing amounts of precipitation from Pacific sources when compared to the CTRL run. This difference between source areas in the LGM and LGMT runs is important because a marked depletion of stadial stable isotope data from the Camp Century and NGRIP ice cores points to increased precipitation in central and northern Greenland stemming from a northerly (i.e., Pacific) source during the LGM (Johnsen and Vinther 2007).

A more direct comparison between model results and ice core δ18O data is also possible. Greenland ice core δ18O can to first order be interpreted as a proxy for condensation temperature at the drill site. It is therefore instructive to extract the changes in condensation temperature between the CTRL run and the two LGM runs. Figure 13a presents the changes in precipitation weighted condensation temperature at selected ice core drill sites for the LGM and LGMT runs. The latter have been corrected for artificial height changes introduced by the model resolution: Since the spectral T42 resolution versions of the topographies are different from those tabulated by Peltier (2004), the topography change between CTRL and LGM (and LGMT) also differs therefrom. With no discussion of the degree to which the Peltier (2004) changes are correct, we use a lapse rate of −6.5 K/km to correct the temperatures thus ensuring that our results reflect the Peltier (2004) changes. For the Renland drill site, corrections have been made demanding zero height change. This correction is necessary because the Renland ice cap is not resolved by the elevation model, but is instead treated as a part of the margin of the vast Greenland ice sheet. The elevation model thus predicts large changes in elevation as the margin advances during the glacial, while in reality the small and isolated Renland ice cap cannot have changed its elevation significantly (Johnsen et al. 1992b). Comparing the pattern of changes seen in Fig. 13a to the observed changes in δ18O at the drill sites (Fig. 13c), the large decrease at the CC site is only seen in the LGM run. In the LGMT run, the CC site experiences a weaker cooling than the central Greenland sites. The northwestern-most sites CC and NE (no ice core glacial information yet) seem to be where the signs of a shift in circulation are most pronounced.
Fig. 13

a Changes in precipitation weighted condensation temperature for the ice core sites between CTRL and LGM (dashed) and LGMT (dotted). Notice that compared to earlier plots, comparison is with CTRL and data has been corrected for spectral smoothing of topography height using a lapse rate of −6.5 K/km (see text for details). b As in a but with the corresponding change in source temperature subtracted. c LGM–PD changes in annual average δ18O from ice cores. Note that data for the NE site are not yet available

Realizing that precipitation δ18O is also influenced by changes in temperatures in the source area (as the fractionation is a process that is dependent on the temperature change from the source area to the condensation area) we have also calculated the changes in temperature difference between source area temperatures and condensation temperatures. The resulting patterns of change can be seen in Fig. 13b. Again, the LGM run is most similar to the observed δ18O changes in the northwestern region. With or without taking source temperatures into account the main result of the comparison between modeled proxy and observed changes at the drill sites is, however, the same. A large increase in Pacific moisture at the CC, NE and NG sites is seen even in the annual average in the LGM experiment (Fig. 9) and yields larger decreases in the proxy to the northwest. This supports the suggestion made by Johnsen and Vinther (2007) that the observed pattern of Holocene to LGM δ18O changes in the ice cores are best explained by Pacific moisture transported to the northern and central parts of the Greenland ice sheet by more northerly winds.

Such shifts in Greenland moisture sources in response to CLIMAP sea surface temperatures and glacial topography have been found in some, but not all, similar simulations. Charles et al. (1994), for example, find that addition of the Laurentide ice sheet gives a large drop in North American contribution and the extensive sea ice gives a large drop in the Nordic Seas contribution. These changes give a zonal change throughout Greenland: North Atlantic sources dominate southern Greenland while North Pacific sources dominate northern Greenland much like in the present study. They find that North Pacific moisture arrives in Greenland much more depleted (order of 15‰) in δ18O and the shift would thus (even without changes in temperature) yield significant drops in δ18O. In contrast, Werner et al. (2001), also using CLIMAP SSTs, find no shift in source regions toward Pacific influence. Moreover, they find no significant differences when modified CLIMAP SSTs (with cooler tropics and warmer North Atlantic) are used. Their modeled LGM—present change in δ18O is on the low side, but although one possible reason could be the lack of a shift toward Pacific moisture, this would, they argue, not resolve their problem of a positive deuterium excess change (rather than the observed decrease). They do find an increase in the northerly component of the mean flow over central Greenland, but this leads only to a cooling and drying but no increase in Pacific moisture. They speculate that different model sensitivities to SST forcing or different ice cap specifications are possible explanations. Based on our compartmentalized experiments, however, where the increased Pacific influence on the northwestern cores was seen with and without changed topography (Fig. 9) we are reluctant to ascribe such inter-model differences to different specifications of ice sheets. In their high-resolution regional model, Bromwich et al. (2004) find that the existence of a pronounced split jet stream hinges also on model physics, resolution and glacial SST reconstruction.

Resolution may, in fact, be a part of the reason for our apparently contradictory result that the CLIMAP reconstruction yields results better in agreement with ice core inferences than does the modified and supposedly improved reconstruction. Sodemann et al. (2008a) find with their Lagrangian backtrajectory technique no Pacific moisture contribution to present-day Greenland winter precipitation, regardless of site, in a simulation with considerably higher-resolution ERA-40 data (T159). Thus, if the present-day Pacific influence in our study and those of Charles et al. (1994) and Werner et al. (2001) is merely an artifact of resolution, the shift in this quantity may also be so. This does not, however, explain the ice core indications of a changed source. Alternatively, the tropical cooling of 4 K in the modified reconstruction is too strong (e.g., Crowley 2000) or the tropical SST pattern is more important than the absolute tropical temperature (Yin and Battisti 2001; Hoerling et al. 2004). Perhaps a different tropical pattern (with larger-than-CLIMAP cooling) could give the right changes in moisture sources. Hostetler et al. (2006) suggest a different, oceanographically based LGM reconstruction which in the GENESIS model yields temperatures and precipitation changes which (to the degree paleodata exist) match better than CLIMAP. Finally, it may be that the models’ responses in atmospheric circulation simply are too weak as found by Hurrell et al. (2004) for 20th century North Atlantic Oscillation changes.

5 Conclusions

Using the NCAR CCM3 atmospheric GCM fitted with a moisture tracking functionality we have studied the changes in atmospheric circulation and sources for Greenland precipitation resulting from different components of glacial boundary conditions. For the glacial SSTs we have used both the CLIMAP reconstruction and the modified version thereof by Toracinta et al. (2004) displaying 4 K cooler tropics and less extensive sea ice cover, especially in the North Atlantic. The ICE-5G reconstruction of Peltier (2004) was used to determine ice sheets, topography and land mask. Before changing both SSTs and topography to glacial conditions, a series of intermediate steps was taken, in which the two components were changed alone. This allows us to monitor the contributions of each of them to the full glacial changes.

The changes in the 500 hPa height field was found quite accurately to respond linearly to the changing boundary conditions: For both of the SST reconstructions, the change due to adding only topography (TOPO–CTRL) plus the change due to SST changes (SST–CTRL and SSTT–CTRL, respectively) equals approximately the full glacial changes (LGM–CTRL and LGMT–CTRL, respectively). This linearity permits us to ascribe certain changes in the full glacial experiments to contributions from the individual factors. We find, for instance, a shift toward a cross-Arctic flow at the 700 hPa level and an associated shift in sources of Greenland precipitation from the Atlantic to the Pacific as a result of the CLIMAP SSTs. This shift occurs with or without changing the topography and does not occur in any case with the modified temperature reconstruction. Our experiments thus indicate that the temperature reconstruction is more important than the ice sheet reconstruction for inducing such shifts.

Although the glacial cooling tends to lead to a drying of the climate, the northwestern-most of the Greenland cores, which in SST and LGM acquire a new Pacific source of moisture, do not encounter as massive a drying during winter as do the more Atlantic-influenced cores. This leads to two different directions of change in seasonal weighting for the two groups, which in turn leads to two different signs of the seasonal bias in precipitation weighted temperatures. Interestingly, stable isotope data from the northwestern-most Greenland ice core (Camp Century) lend support to the proposition that a Pacific source played an important role for northwestern Greenland precipitation during the glacial (Fig. 13).

During winter when continental sources are insignificant, addition of the ice sheets (yielding no decisive circulation changes) does not make a large difference and the SST reconstruction is most important for determining the change in condensation temperature (Fig. 7b). In summer, where the North American contribution is important and the shift toward Pacific moisture is less pronounced, addition of the Laurentide Ice Sheet becomes the dominant factor in determining the change in condensation temperature (Fig. 7c).

In the annual mean, we find that it is possible to decompose the LGM and LGMT changes into linear contributions from temperature changes, seasonality changes and source distribution changes (Fig. 10b, c). In both cases, the direct source distribution effect is rather small and the warm bias from the seasonality change tends to counter some of the direct cooling. The exception to this is for the northwestern cores in the LGM case, where the opposite seasonality effect tends to lower the condensation temperatures even further. The low-latitude cooling in LGMT yields a cooling of the high-latitude atmosphere which is re-found as a larger drop in the condensation temperatures relative to the LGM case—again, except for the northwestern cores where the seasonality biases are opposite.

The source temperatures generally decrease and they do so more for LGMT than for LGM (Fig. 11). Shifts toward more equatorward source regions are insufficient to give any visible warming effect on the resulting source temperatures. When the circulation changes and the Pacific becomes more influential during winter in the LGM case, a non-linear effect of a shift from a region of large cooling (the Atlantic) to a region of much smaller cooling (the Pacific) yields a warming.

Our key finding is that SSTs seem to be more important than topography for qualitatively changing moisture sources for Greenland precipitation. In fact, the apparent dominance of the SST effects—be it global mean temperature, tropical SST patterns or meridional gradients—suggests that also the Eemian may have been characterized by qualitatively different source patterns. Since at least the glacial changes are more pronounced the more northwesterly the location of the core, this may become particularly important when it comes to interpretation of the (not yet drilled) NEEM core which is expected to give an undisturbed Eemian record.


  1. 1.

    Note that precipitation rates and accumulation rates are not exactly the same since the latter also includes effects of sublimation.



Both authors are supported by the Carlsberg Foundation. This work was supported in part by a grant of HPC resources from the Arctic Region Supercomputing Center (ARSC) at the University of Alaska Fairbanks as part of the Department of Defense High Performance Computing Modernization Program. We thank Harald Sodemann and an anonymous reviewer for comments and suggestions that have significantly improved the manuscript.


  1. Alexeev VA, Langen PL, Bates JR (2005) Polar amplification of surface warming on an aquaplanet in “ghost forcing” experiments without sea ice feedbacks. Clim Dyn. doi:10.1007/s00382-005-0018-3
  2. Armengaud A, Koster RD, Jouzel J, Ciais P (1998) Deuterium excess in Greenland snow: analysis with simple and complex models. J Geophys Res 103:8947–8953CrossRefGoogle Scholar
  3. Bonan GB (1998) The land surface climatology of the NCAR Land Surface Model coupled to the NCAR Community Climate Model. J Climate 11:1307–1326CrossRefGoogle Scholar
  4. Bromwich DH, Toracinta ER, Wei H, Oglesby RJ, Fastook JL, Hughes TJ (2004) Polar MM5 simulations of the winter climate of the Laurentide ice sheet at the LGM. J Clim 17:3415–3433CrossRefGoogle Scholar
  5. Byrkjedal Ø, Kvamstø NG, Meland M, Jansen E (2006) Sensitivity of last glacial maximum climate to sea ice conditions in the Nordic Seas. Clim Dyn 26(5):473–487CrossRefGoogle Scholar
  6. Cappelen J, Jørgensen BV, Laursen EV, Stannius LS, Thomsen RS (2001) The observed climate of Greenland, 1958–99—with climatological standard normals, 1961–90. Technical Report 00-18, Danish Meteorological InstituteGoogle Scholar
  7. Charles CD, Rind D, Jouzel J, Koster RD, Fairbanks RG (1994) Glacial–interglacial changes in moisture sources for Greenland—influences on the ice core record of climate. Science 263:508–511CrossRefGoogle Scholar
  8. CLIMAP (1981) Seasonal reconstruction of the Earth’s surface at the last glacial maximum. Map and Chart Series 36. Technical report, Geological Society of AmericaGoogle Scholar
  9. CLIMAP (1994) CLIMAP 18K Database. IGBP PAGES/World Data Center—a for Paleoclimatology Data Contribution Series no. 94-001. Technical report, NOAA/NGDC Paleoclimatology Program, Boulder CO, USAGoogle Scholar
  10. Cole JE, Rind D, Webb RS, Jouzel J, Healy R (1999) Climatic controls on interannual variability of precipitation δ18O: simulated influence of temperature, precipitation amount, and vapor source region. J Geophys Res 104(D12):14223–14235CrossRefGoogle Scholar
  11. Cook KH, Held IM (1988) Stationary waves and the ice age climate. J Clim 1:807–819CrossRefGoogle Scholar
  12. Crowley TJ (2000) CLIMAP SSTs re-visited. Clim Dyn 16:241–255CrossRefGoogle Scholar
  13. Crucifix M, Braconnot P, Harrison SP, Otto-Bliesner B (2005) Second phase of Paleoclimate Modeling Intercomparison Project. EOS 86(28):264CrossRefGoogle Scholar
  14. Dansgaard W, Johnsen SJ (1969) A flow model and a time scale for the ice core from Camp Century, Greenland. J Glaciol 8:215–223Google Scholar
  15. Dansgaard W, Johnsen SJ, Clausen HB, Dahl-Jensen D, Gundestrup N, Hammer CU, Hvidberg CS, Steffensen JP, Sveinbjörnsdóttir AE, Jouzel J, Bond G (1993) Evidence for general instability of past climate from a 250-kyr ice-core record. Nature 364:218–220CrossRefGoogle Scholar
  16. Delaygue G, Masson V, Jouzel J, Koster RD, Healy RJ (2000) The origin of Antarctic precipitation: a modelling approach. Tellus 52B:19–36Google Scholar
  17. Hoerling MP, Hurrell JW, Xu T, Bates GT, Phillips AS (2004) Twentieth century North Atlantic climate change. Part II: Understanding the effect of Indian Ocean warming. Clim Dyn 23:391–405CrossRefGoogle Scholar
  18. Hoskins BJ, Karoly DJ (1981) The steady linear response of a spherical atmosphere to thermal and orographic forcing. J Atmos Sci 38:1179–1196CrossRefGoogle Scholar
  19. Hostetler S, Pisias N, Mix A (2006) Sensitivity of last glacial maximum climate to uncertainties in tropical and subtropical ocean temperatures. Quat Sci Rev 25:1168–1185CrossRefGoogle Scholar
  20. Hurrell JW, Hoerling MP, Phillips AS, Xu T (2004) Twentieth century North Atlantic climate change. Part I: assessing determinism. Clim Dyn 23:371–389CrossRefGoogle Scholar
  21. Johnsen SJ, Dansgaard W (1992) On flow model dating of stable isotope records from Greenland ice cores. NATO ASI Ser I 2:13–24Google Scholar
  22. Johnsen SJ, Vinther BM (2007) Stable isotope records from Greenland ice cores. Encyclopedia of Quarternary Sciences, pp 3–29. doi:10.1016/B0-444-52747-8/00345-8
  23. Johnsen SJ, Dansgaard W, White JWC (1989) The origin of Arctic precipitation under present and glacial conditions. Tellus 41B:452–468Google Scholar
  24. Johnsen SJ, Clausen HB, Dansgaard W, Fuhrer K, Gundestrup N, Hammer CU, Iversen P, Jouzel J, Stauffer B, Steffensen JP (1992a) Irregular glacial interstadials recorded in a new Greenland ice core. Nature 359:311–313CrossRefGoogle Scholar
  25. Johnsen SJ, Clausen HB, Dansgaard W, Gundestrup NS, Hansson M, Jonsson P, Steffensen JP, Sveinbjörnsdóttir AE (1992b) A deep ice core from east Greenland. Medd Grønland 29:3–29Google Scholar
  26. Johnsen SJ, Dahl-Jensen D, Gundestrup N, Steffensen JP, Clausen HB, Miller H, Masson-Delmotte V, Sveinbjörnsdóttir AE, White J (2001) Oxygen isotope and palaeotemperature records from six Greenland ice-core stations: camp century, Dye-3, GRIP, GISP2, Renland and NorthGRIP. J Quat Sci 16(4):299–307CrossRefGoogle Scholar
  27. Joussaume S, Taylor KE (2000) The Paleoclimate Modeling Intercomparison Project. In: Braconnot P (ed) Proceedings of the third PMIP workshop, WCRP-111, WMO/TD-1007, pp 9–25Google Scholar
  28. Joussaume S, Sadourny R, Vignal C (1986) Origin of precipitating water in a numerical simulation of the July climate. Ocean Air Interact 1(1):43–56Google Scholar
  29. Kageyama M, Valdes PJ, Ramstein G, Hewitt C, Wyputta U (1999) Northern Hemisphere storm tracks in present day last glacial maximum climate simulations: a comparison of the European PMIP models. J Clim 12:742–760CrossRefGoogle Scholar
  30. Kiehl JT, Hack JJ, Bonan GB, Boville BA, Briegleb B P, Williamson DL, Rasch PJ (1996) Description of the NCAR Community Climate Model (CCM3). Technical Report TN-420, CGD, National Center for Atmospheric ResearchGoogle Scholar
  31. Kiehl JT, Hack JJ, Bonan GB, Boville BA, Williamson D L, Rasch PJ (1998) The National Center for Atmospheric Research Community Climate Model: CCM3. J Clim 11:1131–1149CrossRefGoogle Scholar
  32. Koster R, Jouzel J, Suozzo R, Russell G, Broecker W, Rind D, Eagleson P (1986) Global sources of local precipitation as determined by the NASA/GISS GCM. Geophys Res Lett 13(1):121–124CrossRefGoogle Scholar
  33. Koster RD, Jouzel J, Suozzo RJ, Russell GL (1992) Origin of July Antarctic precipitation and its influence on deuterium content: a GCM analysis. Clim Dyn 7:195–203CrossRefGoogle Scholar
  34. Kothavala Z, Oglesby RJ, Saltzman B (1999) Sensitivity of equilibrium surface temperature of CCM3 to systematic changes in atmospheric CO2. Geophys Res Lett 26(2):209–212CrossRefGoogle Scholar
  35. Krinner G, Genthon C (1998) GCM simulations of the last glacial maximum surface climate of Greenland and Antarctica. Clim Dyn 14:741–758CrossRefGoogle Scholar
  36. Krinner G, Werner M (2003) Impact of precipitation seasonality changes on isotopic signals in polar ice cores: a multi-model analysis. Earth Planet Sci Lett 216:525–538CrossRefGoogle Scholar
  37. Krinner G, Genthon C, Jouzel J (1997) GCM analysis of local influences on ice core δ signals. Geophys Res Lett 24:2825–2828CrossRefGoogle Scholar
  38. Langway CC, Oeschger H, Dansgaard W (1985) The Greenland ice sheet program in perspective. Greenland Ice Cores: geophysics, geochemimistry and environment. AGU Monogr 33:1–8Google Scholar
  39. Li C, Battisti DS, Schrag DP, Tziperman E (2005) Abrupt climate shifts in Greenland due to displacements of the sea ice edge. Geophys Res Lett 32:L19702. doi:10.1029/2005GL023492 CrossRefGoogle Scholar
  40. Loutre M-F, Paillard D, Vimeux F, Cortijo E (2004) Does mean annual insolation have the potential to influence the climate?. Earth Planet Sci Lett 221:1–14CrossRefGoogle Scholar
  41. Masson-Delmotte V, Jouzel J, Landais A, Stievenard M, Johnsen SJ, White JWC, Werner M, Sveinsbjornsdottir A, Fuhrer K (2005a) GRIP deuterium excess reveals rapid and orbital-scale changes in Greenland moisture origin. Science 309:118–121CrossRefGoogle Scholar
  42. Masson-Delmotte V, Landais A, Stievenard M, Cattani O, Falourd S, Jouzel J, Johnsen SJ, Dahl-Jensen D, Sveinsbjornsdottir A, White JWC, Popp T, Fischer H (2005b) Holocene climatic changes in Greenland: different deuterium excess signals at Greenland Ice Core Project (GRIP) and NorthGRIP. J Geophys Res 110. doi:10.1029/2004JD005575
  43. Mayewski PA, Meeker LD, Whitlow SI, Twickler MS, Morrison MC, Bloomfield P, Bond GC, Alley RB, Gow AJ, Grootes PM, Meese DA, Ram M, Taylor KC, Wumkes MA (1994) Changes in atmospheric circulation and ocean ice cover over the North Atlantic during the last 41,000 years. Science 263:1747–1751CrossRefGoogle Scholar
  44. Meland MY, Jansen E, Elderfield H (2005) Constraints on SST estimates for the northern North Atlantic/Nordic Seas during the LGM. Quat Sci Rev 24:835–852CrossRefGoogle Scholar
  45. NGRIP members (2004) High-resolution record of Northern Hemisphere climate extending into the last interglacial period. Nature 432:147–151Google Scholar
  46. Noone D, Simmonds I (2002) Annular variations in moisture transport mechanisms and the abundance of δ18O in Antarctic snow. J Geophys Res 107. doi:10.1029/2002JD002262
  47. Peltier WR (2004) Global glacial isostasy and the surface of the ice-age Earth: The ICE-5G (VM2) model and GRACE. Ann Rev Earth Planet Sci 32:111–149CrossRefGoogle Scholar
  48. Rind D (1987) Components of the ice age circulation. J Geophys Res 92(D4):4241–4281CrossRefGoogle Scholar
  49. Rodgers KB, Lohmann G, Lorenz S, Schneider R, Henderson GM (2003) A tropical mechanism for Northern Hemisphere deglaciation. Geochem Geophys Geosyst 4(5):1046. doi:10.1029/2003GC000508 CrossRefGoogle Scholar
  50. Roe GH, Lindzen RS (2001) The mutual interaction between continental-scale ice sheets and atmospheric stationary waves. J Clim14:1450–1465CrossRefGoogle Scholar
  51. Schmidt GA, LeGrande AN, Hoffmann G (2007) Water isotope expressions of intrinsic and forced variability in a coupled ocean-atmosphere model. J Geophys Res 113(D10103). doi:10.1029/2006JD007781
  52. Shea DJ, Trenberth KE, Reynolds RW (1992) A global monthly sea surface temperature climatology. J Clim 5:987–1001CrossRefGoogle Scholar
  53. Sodemann H, Schwierz C, Wernli H (2008a) Interannual variability of Greenland winter precipitation sources: Lagrangian moisture diagnostic and North Atlantic Oscillation influence. J Geophys Res 113(D03107). doi:10.1029/2007JD008503
  54. Sodemann H, Masson-Delmotte V, Schwierz C, Vinther BM, Wernli H (2008b) Inter-annual variability of Greenland winter precipitation sources. Part II: Effects of North Atlantic Oscillation variability on stable isotopes in precipitation. J Geophys Res (in press)Google Scholar
  55. Steffen K, Box J (2001) Surface climatology of the Greenland ice sheet: Greenland Climate Network 1995–1999. J Geophys Res 106(D24):33951–33964CrossRefGoogle Scholar
  56. Toracinta ER, Oglesby RJ, Bromwich DH (2004) Atmospheric response to modified CLIMAP ocean boundary conditions under the last glacial maximum. J Climate 17:504–522CrossRefGoogle Scholar
  57. Vimeux F, Masson V, Delaygue G, Jouzel J, Petit JR, Stievenard M (2001) A 420,000 year deuterium excess record from East Antarctica: information on past changes in the origin of precipitation at Vostok. J Geophys Res 106(D23):31863–31873CrossRefGoogle Scholar
  58. Vinther BM, Johnsen SJ, Andersen KK, Clausen HB, Hansen AW (2003) NAO signal recorded in the stable isotopes of Greenland ice cores. Geophys Res Lett 30(7):1387CrossRefGoogle Scholar
  59. Vinther BM, Andersen KK, Jones PD, Briffa KR, Cappelen J (2006) Extending Greenland temperature records into the late eighteenth century. J Geophys Res 111:D11105. doi:10.1029/2005JD006810 CrossRefGoogle Scholar
  60. Werner M, Heimann M (2002) Modeling interannual variability of water isotopes in Greenland and Antarctica. J Geophys Res 1107(D1). doi:10.1029/2001JD900253
  61. Werner M, Mikolajewicz U, Heimann M, Hoffmann G (2000) Borehole versus isotope temperatures on Greenland: seasonality does matter. Geophys Res Lett 27(5):723–726CrossRefGoogle Scholar
  62. Werner M, Heimann M, Hoffmann G (2001) Isotopic composition and origin of polar precipitation in present and glacial climate simulations. Tellus B 53:53–71CrossRefGoogle Scholar
  63. Yin JH, Battisti DS (2001) The importance of tropical sea surface temperature patterns in simulations of last glacial maximum climate. J Clim 14:565–581CrossRefGoogle Scholar

Copyright information

© Springer-Verlag 2008

Authors and Affiliations

  1. 1.Centre for Ice and Climate, Niels Bohr InstituteUniversity of CopenhagenCopenhagen ODenmark

Personalised recommendations