Using the comparison between the 00dhL and the noIS experiments, we first consider the impact of the LIS albedo on the 2 m summer air temperature over the Scandinavia–Barents–Kara (SBK) region (Fig. 3, 00dhL experiment, left column). The most striking feature is a cooling over the whole Eurasian continent with a maximum (more than 10 ∘C) centered over the Barents and Kara seas and a moderate cooling (2 to 4 ∘C) over Scandinavia. The area experiencing the largest cooling also corresponds to the area with the largest albedo (orange contours). This is a direct consequence of the LGM land–sea mask used in this study for which the Barents and Kara seas are represented as continental areas. This favors snow accumulation and therefore higher albedo values in this area, causing in turn a positive feedback on temperatures. Figure 3 also displays the 2 m mean summer temperature over the SBK region for different altitudes of the LIS. Here, we only give results of selected experiments to simplify the presentation of our results. It appears that the SBK cooling gets smaller when the LIS gets higher. Overall, the simulated summer surface air temperature evolves together with the LIS altitude, with a progressive reduction of the zonal and meridional extent of the largest cooling zone. When the LIS reaches its full LGM size (100dhL-noIS), the summer cooling over the Barents and Kara seas is reduced to 4–6 and to 2 ∘C at most over Scandinavia. The topography of the LIS has therefore a warming effect on the SBK region for the summer season.

In winter, the response of the temperature to the LIS albedo and its evolution with the LIS altitude are different from the summer response (Fig. 3, right panels). First, the LIS albedo induces warmer temperatures (+3 ∘C) over southern Scandinavia, the British Isles and the present-day North and Baltic seas, and cooler temperatures over an area in the Barents Sea expanding from Svalbard to Novaya Zemlya (3 ∘C). Surprisingly, in the 30dhL experiment, the amplitude of these temperature changes (with respect to noIS) is reduced, showing that the response of winter temperatures to changes in LIS altitude is far from being linear. Beyond 60dhL, the cooling zone progressively expands, and spreads across the entire British Isles and Scandinavia–Barents–Kara area when the LIS has its full LGM size. In contrast to the climatic response simulated for the summer season, the LIS topography leads to colder surface air temperatures in winter.

To understand the origin of these contrasted responses, we investigate the processes through which the changes in the altitude of the Laurentide ice sheet modify the atmospheric circulation.

The comparison of the zonal anomaly (00dhL vs. noIS) of the 500 hPa summer geopotential height shows that the sole albedo effect of the LIS is sufficient to drastically change the atmospheric circulation (Fig. 4, 00dhL and noIS, left column). The ridge over the Rockies, clearly visible in noIS, disappears in 00dhL, and the trough over the Labrador sector is more extensive. Ridges over the North Atlantic, the Greenland ice sheet and northern Europe are more developed in 00dhL (with respect to noIS). By contrast, the troughs over Iceland, the Norwegian and Barents seas appearing in noIS are weaker or even vanish in 00dhL. In noIS, the trough over Svalbard implies southerlies over the Barents Sea (Fig. 4, right column) and therefore, warm temperatures in this area (Fig. 3). The weakening of this trough in 00dhL as well as the stronger ridge over Greenland induces northerlies over Svalbard and the Barents Sea (Fig. 4) and therefore colder temperatures (Fig. 3).

The Labrador trough and the North Atlantic ridge become stronger when the LIS gets higher, until it reaches 60 % of its full size. Beyond 60dhL, the Labrador trough and the Atlantic ridge keep more or less the same amplitude. When the LIS gets higher, the Greenland ridge gets weaker, the European ridge remains almost unchanged (Fig. 4), and there is a return of the Iceland–Svalbard–Barents trough. This new trough, centered between Greenland and Svalbard, along with the weakening of the Greenland ridge, brings southerlies again over the northern Barents Sea. The Barents winds shift from northerlies to southerlies when the LIS is 60 % of its full size. The southerlies bring warmer temperatures, explaining the SBK warming with the increase of the LIS height (Figs. 3 and 4).

In winter (Fig. 5), the sole albedo effect of the LIS does not drastically change the geopotential anomaly. However, the meridional wind field shows a slight decrease of the southerlies over the Barents Sea, explaining the slight cooling over this region. There is also a slight decrease of the northerlies simulated over the British Isles, the eastern North Atlantic Ocean and southern Scandinavia, consistent with the moderate warming simulated over this region. In the 30dhL experiment, a strengthening of the southerlies is observed (Fig. 5) inducing a warming (with respect to 00dhL) over the SBK region (Fig. 3) and explaining the non-linear behavior on winter temperatures. When the altitude of the LIS further increases, there is a progressive emergence of a trough over the Scandinavian–Barents sector. This trough comes with northerlies over the Norwegian Sea and weaker southerlies over the Barents sea (Fig. 5, right column), consistently with colder temperatures over this area, as seen in Fig. 3.

This analysis explains why opposite sign temperature responses are obtained for the summer and winter seasons. Since ablation is rather sensitive to the summer season, more ice is expected over the Fennoscandian area when the sole albedo effect of the LIS is operating (i.e. 00dhL experiment). However, the development of an ice sheet in this area also depends on the amount of snow accumulation. Therefore, we next examine hereafter the LIS impact on the SBK total precipitation and snowfall.

In this section, we investigate the impact of the LIS on the mean annual precipitation and snowfall over northwestern Eurasia (Fig. 6). In response to the LIS albedo effect (00dhL-noIS panel) a slight excess of precipitation is simulated over the northeastern North Atlantic and the southern and western parts of Scandinavia, but less precipitation is simulated over the Barents–Kara sea area. However, over the whole SBK region, the snowfall anomaly is positive. This increase in snowfall is due to the summer signal (not shown) related to the cooling observed over this region (Fig. 3). This cooling is strong enough to allow for an increase of the fraction of solid precipitation and to maintain the snow cover.

When the LIS reaches its full LGM altitude (100dhL-noIS), the positive anomaly of total precipitation over Scandinavia simulated in the 00dhL experiment shifts southward, reaching even the French and the Iberian Peninsula Atlantic coasts. This results in dry conditions over the western flank of the Scandinavian region, both in terms of total and solid precipitation. Concomitantly, the negative anomaly of total precipitation over the Barents–Kara sea region simulated in 00dhL expands westward leading to a drier climate and to a decrease of snowfall over Scandinavia. These results suggest that the Barents and Kara seas and the Scandinavian regions are sensitive to different characteristics of the LIS; while the precipitation over Scandinavia appears to be sensitive to the LIS height, the precipitation anomaly over the Barents and Kara seas mainly appears to result from the LIS albedo effect and is rather insensitive to height.

To disentangle the mechanisms responsible for precipitation changes in these regions, we split up the total precipitation into its large-scale and convective components (Fig. 7). The large-scale component behaves similarly to the total precipitation shown in Fig. 6 with a positive anomaly in the 00dhL experiment simulated over Scandinavia and the eastern North Atlantic sector and the southward shift of this pattern simulated when the LIS gets higher. Hence the drier signal over the western flank of Scandinavia obtained with the full-LGM LIS (100dhL-noIS) can also be found in the large-scale component of the total precipitation. This strongly suggests that the Scandinavian precipitation, as expected, is driven by large-scale processes.

The negative anomaly of precipitation appearing in the Arctic Ocean north of the Barents–Kara area (Fig. 6) is also clearly visible in the large-scale component (Fig. 7); by contrast, the large negative precipitation anomaly south of the Novaya Zemlya clearly arises from convective processes. Over the Barents–Kara area, the origin of the precipitation signal is less clear but seems to be a combination of both convective and large-scale processes. The decrease of convective precipitation negative anomaly (with respect to noIS) over the entire SBK sector when the LIS is higher can be easily explained by the weakening of the summer cooling discussed in section 3.1.

To further investigate the mechanisms governing precipitation over Scandinavia and the Barents–Kara area, we examine the seasonality of precipitation. We first compute the average precipitation above Scandinavia (55–70∘ N : 0–20∘ E) (Fig. 8a) and the Barents–Kara region (65–90∘ N : 20–100∘ E) (Fig. 8b), then we compare summer and winter precipitation anomalies with the annual precipitation anomaly. Results displayed in Figure 8 show that over Scandinavia winter precipitation changes are clearly correlated with annual precipitation (r= 0.99). This means that the annual signal observed over Scandinavia (Figs. 6 and 7) is mainly due to winter precipitation. Conversely, the Barents–Kara sector is rather governed by summer precipitation (at least for LIS altitudes up to 60 % of the PMIP3 one), as shown by the positive correlation (r= 0.98) obtained for the summer season (Fig. 8b).

We therefore look for a mechanism prevailing in winter to explain the Scandinavian precipitation differences. Using the PMIP3 model outputs for the LGM and the PI time slices, Beghin et al. (2015) found a correlation between the southward shift of the North Atlantic jet stream from the pre-industrial to the LGM periods and the winter precipitation changes over the Iberian Peninsula. Following the same approach, we examine whether a similar mechanism to the one found in (Beghin et al., 2015) may explain the precipitation changes over Scandinavia when the LIS characteristics (albedo and topography) are modified.

Similarly to Chavaillaz et al. (2013), we use the 850 hPa zonal wind to define the jet stream. The position of the jet displayed in Fig. 9 corresponds to the position of the 850 hPa meridional wind maximum. The LIS albedo effect induces a slight northward shift (with respect to noIS) of the North Atlantic jet stream (00dhL-noIS panel). As the LIS becomes higher, the jet moves progressively southward and shifts southward of its noIS position between 30 and 60dhL. From 60dhL and beyond, the Scandinavian winter precipitation anomaly becomes more and more negative. This suggests a relationship between the shift of the North Atlantic jet stream and the precipitation over the Scandinavian region.

To confirm this assumption, we plot the shift of the jet (Δφ) as a function of the precipitation difference between the noIS and the xxdhL experiments over Scandinavia (Fig. 10). Δφ is defined as the difference of latitudes between noIS and xxdhL experiments where the 850 hPa zonal wind is at its highest value. The latitude of this maximum has been found by computing the zonal mean of the zonal wind over the Atlantic basin (50–10∘ W). To minimize the uncertainties due to model discretization, we used a parabolic interpolation around the grid point where the mean zonal wind is largest. The maximum of the corresponding parabola yields a better approximation of the exact latitude of the maximum jet wind. This method allows for a more accurate finding of the latitude of the jet maximum (Chavaillaz et al., 2013; Beghin et al., 2015). We find a good linear correlation between the shift of the North Atlantic jet and the Scandinavian precipitation anomaly during the winter season (Fig. 10). The quantification of the range of the precipitation anomalies and of the jet shift is obtained by bootstrapping: a sample of thousand values is obtained by calculating the average of 1000 randomly picked samples of 40-year duration in the noIS and the xxdhL original samples. Linear regressions of the 5th and 95th percentiles of the bootstrapping sample confirm the close link between the shift of the North Atlantic jet stream and the winter precipitation changes over Scandinavia due to the effect of LIS.

The 500 hPa geopotential height (z500) zonal anomaly over the East Laurentide sector (90–60∘ W) does not change significantly when a flat LIS is imposed (Figs. 5, 11). As the LIS gets higher, the Labrador trough expands southward: Fig. 11 shows that the zero isoline is located at ∼ 43∘ N in the noIS and 00dhL experiments and progressively reaches 37∘ N for a full-LGM LIS. Over the Atlantic sector (50–10∘ W) the Icelandic low/Azores high pattern remain similar in all experiments up to 60dhL and only show a significant change in the full LIS experiment (Figs. 5, 11). Indeed, in 100dhL, the Icelandic low's amplitude weakens but simultaneously expands southward (Fig. 11). Consistently with these changes in the z500 zonal anomaly, the 500 hPa westerlies shift southward over the whole Atlantic sector (Fig. 11, bottom). Again, the largest southward shift for this variable is between the 60 and 100dhL experiments, showing the non-linearity of the atmospheric circulation to the LIS height.

As previously mentioned in Sect. 3.2.2, the mean annual precipitation changes over the Barents–Kara area are dominated by the summer signal when the LIS remains at a relatively low level (i.e. up to 60dhL). This result is confirmed by the comparison of the patterns of annual (Fig. 6), summer and winter (Fig. 9) precipitation anomalies. The amount of summer precipitation is at its lowest level when the sole albedo effect of the LIS is accounted for. It increases with the LIS altitude in response to warmer temperatures (Fig. 3). Conversely, the summer snowfall amount decreases from 00 to 100dhL due to the progressive lowering of the fraction of solid vs. liquid precipitation.

Figure 14 displays the summer surface air temperature, which is the main driver of the ablation, for the 00dhL (flat LIS, no FIS), 00dhLFIS (flat LIS, high FIS), 100dhL (high LIS, noIS) and 100dhLFIS (high LIS, high FIS) experiments. The comparison between Fig. 14a and b and c and d illustrate the impact of the Fennoscandian ice sheet on the 2 m summer temperature. As expected, the presence of Fennoscandia induces a large cooling over the ice sheet itself, due to the temperature–altitude relationship and to the albedo effect. In previous sections, we have shown that the cooling observed over the Barents–Kara region is weakened when the LIS gets higher due to a modification of the large-scale atmospheric circulation (Figs. 3–4). A similar mechanism is observed when the Fennoscandian ice sheet is high (100dhLFIS experiment): the cooling over the ice sheet is reduced and the warming along the southern margin is amplified (Fig. 14d vs. 14b). These features result from an increase of warm air advection in the Barents–Kara sector and to a decrease of northerly heat flux in the south. The presence of the FIS also induces a split of the North Atlantic jet stream in two branches resulting in northward deflected heat fluxes along the western margin of the ice sheet and the Barents Sea sector (Figs. 15b, d and 16b, d). The heat advection modulates the ice-sheet altitude and albedo effects but is not sufficient to cancel out the cooling observed over the ice sheet.

Figure 17 displays maps of snow accumulation differences (after downscaling on the GRISLI grid) between the four experiments computed by GRISLI after downscaling. The presence of the Fennoscandian ice sheet results in an increase of precipitation and snow accumulation over its western flank, along the Norwegian Sea coast (Fig. 17a, c). This is due to the barrier effect of the ice sheet and to the subsequent increase of the vertical wind. This effect is less pronounced when the LIS is high (Fig. 17c) as a result of the southward displacement of the North Atlantic jet stream (Figs. 6, 11). A second striking feature is the increase of snow accumulation in the southern part of the Eurasian ice sheet with respect to the no FIS situation. This is mainly due to colder temperatures over the ice sheet, increasing the fraction of total precipitation which falls as snow. In case of a flat LIS, this accumulation zone is confined to the southern margin and drier conditions (with respect to the noFIS case) in the interior of the Eurasian ice sheet are observed, mainly located over the Kara Sea (Fig. 17a), and likely due to the strong barrier effect. By contrast, when the LIS is high (Fig. 17c), differences between FIS and no FIS experiments are larger and expand over more than the southern half of the FIS. This coincides with a larger temperature difference over the FIS area when the LIS is high (Figs. 14c, d) compared to the low LIS situation (Fig. 14a, b).

As mentioned above (Sect. 3.2.1 and Fig. 6), in the absence of FIS, the total precipitation is driven by large-scale processes over the Scandinavian region and by convection over the Barents–Kara sector. When the LIS is high, the southward shift of the winter jet stream (Fig. 9) and the increase of summer temperature lead to an overall decrease of snowfall over the Eurasian region. When the FIS is present, the situation is rather different and convective precipitation is negligible (not shown) due to the cold temperatures induced by the presence of the FIS. The differences between the 00 and 100dhLFIS experiments are mainly located over the western and southeastern parts of the ice sheet with respectively larger and smaller snowfall amounts simulated in the 00dhLFIS experiment (with respect to 100dhLFIS). The excess of precipitation (total and solid) in the flat LIS/high FIS case is a direct consequence of the increase of both zonal and meridional humidity transports in winter (not shown) favored by the enhanced strength of the 850 hPa horizontal winds (Fig. 18).

The climatic outputs simulated in the 00 and 100dhLFIS experiments have been used to compute the surface mass balance of the Eurasian ice sheet at the GRISLI resolution and to force the ice-sheet model with the downscaling procedure described in Sect. 4.1. Starting from the initial state provided by the 00dhL GRISLI run with flat (00dhLFIS) or full LGM LIS (100dhLFIS), GRISLI has been integrated during 100 kyr. At the end of the simulation, the simulated ice volume and ice covered area simulated in the 00dhLFIS experiment are respectively 10 and 11 % greater than those obtained with the 100dhLFIS simulation. The differences in simulated ice thickness between both simulations are displayed in Fig. 19 (00–100dhLFIS). In most parts of the simulated ice covered areas, ice thickness differences are fully consistent with the snow accumulation signal (Fig. 17). In particular, large positive anomalies (> 500 m) are found in the southwestern sector, that are greater than 1000 m near the ice sheet margin. In the northern part, lower ice thicknesses obtained in the 00dhLFIS experiment correspond to the area of lower snow accumulation (with respect to 100dhLFIS), as ablation is very small due to summer temperatures largely below 0 ∘C. Surprisingly, another large positive difference is found in the eastern part which is characterized by a lower amount of snow accumulation in case of flat LIS. (with respect to full LIS). This pattern results from warmer temperatures simulated in the southern and eastern parts of the 100dhLFIS Eurasian ice sheet in response to the decrease of cold air advection (see Sect. 4.2.1). Warmer temperatures found in the south and east of the FIS when the LIS has its full LGM size favor the westward and the northward retreats of the eastern and southern margins of the ice sheet. Since our approach does not account for the feedback of the simulated ice sheets on the climate, the amplitude of the different responses are likely to be underestimated due to an underestimation of the albedo effect. However, this analysis confirms that the growth of the FIS is slowed down when the LIS is high and demonstrates the self-consistency of the overall study.