We conducted our simulations with version 1.2 of the three-dimensional Earth system model of intermediate complexity LOVECLIM (Goosse et al., 2010), in which the components of the atmosphere, ocean including sea ice, vegetation, ice sheets and carbon cycle are dynamically included. However, in our version, the components for the ice sheets and the carbon cycle were not activated. Therefore, the ice-sheet evolution and greenhouse gases were prescribed in our present study. The atmospheric component is the quasi-geostrophic model ECBilt, which consists of three vertical layers and has T21 horizontal resolution (Opsteegh et al., 1998). CLIO is the ocean component, which consists of a free-surface, primitive-equation oceanic general circulation model (GCM) coupled to a three-layer dynamic–thermodynamic sea-ice model (Fichefet and Maqueda, 1997). The ocean model includes 20 vertical levels and a 3∘ × 3∘ latitude–longitude horizontal resolution (Goosse and Fichefet, 1999). These two core components were further coupled to the biosphere model VECODE, which simulates the dynamics of two main terrestrial plant functional types, trees and grasses, in addition to desert (Brovkin et al., 1997). More details on LOVECLIM can be found in Goosse et al. (2010).

The LOVECLIM model is a useful tool to explore the mechanisms behind climate change, and it has made critical contributions to our understanding of the climate history and variabilities observed in proxy records (Renssen et al., 2005, 2006, 2010). For example, it has helped with investigations of the potential forcings behind abrupt climate events (Renssen et al., 2002; Wiersma and Renssen, 2006; Renssen et al., 2015) as well as understanding the role of the decaying LIS and GIS in temperature evolution over the last 9 kyr (Renssen et al., 2009; Blaschek and Renssen, 2013). Moreover, the LOVECLIM model simulates a reasonable modern climate (Goosse et al., 2010). It also simulates the meridional overturning stream function reasonably well and reproduces a large-scale structure of atmosphere circulation that agrees with observations and with other models (Goosse et al., 2010). In addition, the model's sensitivity to freshwater perturbation is reasonable compared to that of other models (Roche et al., 2007), and its sensitivity to a doubling of atmospheric CO2 concentration is 2 K, which is at the lower end of coupled general circulation model (GCM) estimates (Flato et al., 2013).

We included the major climatic forcings in terms of greenhouse gases (GHGs) in the atmosphere, astronomical parameters (orbital forcing, or ORB) and decaying ice sheets. In all simulations, the solar constant, aerosol levels, the continental configuration and bathymetry were kept fixed at preindustrial values. We based the concentrations of CO2, CH4 and N2O on ice core measurements for GHG forcing (Loulergue et al., 2008; Schilt et al., 2010). The radiative GHG forcing anomaly (relative to 0 kyr) in W m-2 (Ramaswamy et al., 2001), representing the overall GHG contribution, at first showed a rapid rise with a peak of -0.3 W m-2 at 10 kyr, which was followed by a slight decrease towards a minimum at 7 kyr, and gradually increased towards 0 kyr (Fig. 1). The astronomical parameters (eccentricity, obliquity and longitude of perihelion) determine the incoming solar radiation at the top of atmosphere and were derived from Berger (1978). An example of the resulting change in insolation is shown as the anomaly for June at 65∘ N in Fig. 1, which shows the gradual decrease over the course of the Holocene. While the global annual mean insolation stayed at almost the same level (not shown), both changes in obliquity and precession are resulting in insolation variations on the multi-millennial timescale of the Holocene. At the beginning of the Holocene (11.5 kyr), the orbitally induced insolation anomaly in the NH was positive in summer and negative in winter (Fig. S1 in the Supplement). Overall, this setup of GHG and ORB forcing is in line with the PMIP3 protocol ((http://pmip3.lsce.ipsl.fr), except that our simulation excluded the increase in GHG levels during the industrial era (Ruddiman, 2007). Accordingly, the terms preindustrial (era) and 0 kyr are considered equivalent in the present text and indicate modern conditions without anthropogenic impacts.

We took three aspects into account concerning the ice-sheet forcing, namely their spatial extent, their thickness and their meltwater discharge. The reconstructions of ice-sheet spatial extent were based on the dating of geological features and on the correlation of these geological data sets between different regions (Dyke et al., 2003; Svendsen et al., 2004; Putkinen and Lunkka, 2008). According to these reconstructions, the FIS at 11.5 kyr covered most of Fennoscandia except for southern Scandinavia and eastern Finland (Svendsen et al., 2004; Putkinen and Lunkka, 2008; Clark, unpublished data). The LIS occupied most of the lowland area north of the Great Lakes region and filled the whole Hudson Strait (Licciardi et al., 1999; Dyke et al., 2003; Occhietti et al., 2011). The thickness of LIS was up to 2000 m, and for the FIS this thickness was only about 100 m (Ganopolski et al., 2010), which is comparable with the ICE-5G reconstruction (Peltier, 2004). Both the spatial extent of the ice sheets and their thickness were updated every 250 years in our transient experiments, and they decreased rapidly during the earliest Holocene, followed by a more gradual deglaciation rate from 8 kyr onward (Fig. 2a).

We applied the meltwater release for 1200 years in our equilibrium experiments for 11.5 kyr by adding 0.11 Sv (1 sverdrup is 106 m3 s-1) of freshwater at the St. Lawrence River, 0.05 Sv at Hudson Strait and Hudson River, 0.055 Sv coming from the FIS, and 0.002 Sv from the GIS (Licciardi et al., 1999; Jennings et al., 2015). The total freshwater volume added to the oceans in our transient experiments was about 1.46 × 1016 m-3 in the first 4700 years (Fig. 2b), which roughly matches the estimated ice-sheet melting volume during the early Holocene (Dyke et al., 2003; Ganopolski et al., 2010; Clark, unpublished data). The volume of meltwater was slightly lower than the volume of the estimated 60 m sea level rise that took place during the early Holocene (Fig. 2c) (Lambeck et al., 2014), which suggests a coeval Antarctic melting contribution that is not considered here. Given the lack of a direct imprint left by meltwater on terrestrial records and hence the relatively large uncertainty, we used two versions of the freshwater flux (thick dashed lines and solid lines with symbols in Fig. 2b) that represent two possible deglaciation scenarios of the GIS and FIS, named FWF-v1 and FWF-v2. The GIS FWF_v1 scenario is derived from the ICE_5G reconstruction, and FWF_v2 is based on the reconstruction of Vinther et al. (2009), which suggests a faster GIS thinning. The two FIS freshwater flux (FWF) scenarios are based on two estimations of the FIS melting, since the recent cosmogenic dating (FWF_v2) supports a faster melting (Clark, unpublished data) than previously thought (FWF_v1). However, we kept the freshwater discharge from LIS the same as in version 1, since the LIS deglaciation has been relatively well studied and we are more certain about its contribution.

We performed two types of experiments: equilibrium and transient simulations. First, the equilibrium experiments of OG11.5 and OGIS11.5 with boundary conditions for 11.5 kyr (Table 1) were designed. The OGIS11.5 experiment included ice-sheet forcing, whereas no ice sheets were included in OG11.5 (Table 2). Each of these experiments was initiated from the model's default modern condition and was run for 1200 years, of which the last 200 years of data was used for the analysis. Renssen et al. (2006) demonstrated that a 1200-year spin-up is sufficient to reach a quasi-equilibrium in all components of the model.

The data from the end of the 1200-year equilibrium run were then taken to initialize the transient experiments that covered the last 11.5 thousand years. In the first transient simulation named ORBGHG, both GHG and ORB varied on an annual basis. In the second simulation OGIS_FWF-v1, the ice-sheet topography (Fig. 2a) and FWF_v1 (thick dashed lines in Fig. 2b) were additionally included. A third experiment (named OGIS_FWF-v2) was performed with the freshwater version 2 (solid lines with symbols in Fig. 2b) and with the same ice-sheet topography as in OGIS_FWF-v1 to further investigate the climate response to the relatively uncertain freshwater forcing. Both OGIS_FWF-v1 and OGIS_FWF-v2 were initialized from the OGIS11.5 experiment. A preindustrial simulation (PI) was run for 1200 years from the model's default (representing modern conditions) with the boundary conditions that are shown in Table 1 and, similarly to the other equilibrium experiments, the results of the last 200 years were used as a reference. These simulations and their forcings are summarized in Table 2. All temperature values in this study are shown as deviations from the PI simulation (indicates the climate at 0 kyr). Temperatures presented here are simulated near-surface temperature values without the environmental lapse rate corrections to the sea level temperature, which imply approximately a 0.5 ∘C cold bias over ice-sheet covered regions when compared with site-specific proxy records.

It is clear in our analysis of Sect. 3.1 that the climate showed different responses in the following areas: the Arctic, northwestern Europe, northern Canada, Alaska and Siberia (marked in Fig. S2). Therefore, these areas were selected for special examination and the temperature evolutions of these regions will be shown. Our major focus was on millennial-scale temperature trends; therefore, we applied a 500-year running mean to our simulated time series that effectively filtered out high-frequency variability.

At the onset of the Holocene, the overall cool climate indicated by the reconstructions generally matches that of our OGIS11.5 simulation, which shows lower annual temperatures at 11.5 kyr than the PI. Climate reconstructions based on proxy data generally show a cooler early Holocene over northern Europe than at 0 kyr both in the summer and winter (Heiri et al., 2104; Mauri et al., 2015). Terrestrial and ocean sediment data also suggest a cooler early Holocene climate over eastern Siberia (Klemm et al., 2013; Tarasov et al., 2013) and slightly lower SSTs over the North Atlantic Ocean (Came et al., 2007; Berner et al., 2008). Cooler conditions over the Barents Sea and Greenland are also indicated by multiple proxies (Peros et al., 2010; de Vernal et al., 2013; Vinther et al., 2008). Therefore, these proxy data agree with simulated lower temperatures over these areas.

However, there is less agreement with proxies in places where the reconstructions are sparse. The only available pollen-based reconstruction from the western side of the Ural Mountains suggests similar early Holocene summer temperatures (within 1 ∘C anomaly) compared to the preindustrial era (Salonen et al., 2011), whereas OGIS11.5 indicates that summer temperatures were slightly higher at 11.5 kyr over most areas. At high latitudes, the sea-ice cover reconstructions serve as an indirect paleotemperature proxy due to the scarcity of temperature records, and reveal an inconclusive temperature signal over the Canadian Arctic (de Vernal et al., 2013), whereas our simulation reflects an overall warmer climate in the west and cooler conditions in the east.

Proxies indicate significantly different climate patterns over the east and the west of northern America. The later initiation and termination of HTM over northern Canada imply lower temperatures during the early Holocene in the east (Kaufman et al., 2004). However, the higher-than-present early Holocene temperatures over central Beringia and Alaska are reflected by peat accumulation and by northward expansion of animal species (Kaufman et al., 2004; Jones and Yu, 2010). This thermal contrast agrees with those simulated patterns in the OGIS11.5 simulation, which indicates warmer temperatures for Alaska and a much cooler climate over Canada. However, this interpretation of high temperature was recently challenged by Kaufman et al. (2016), who argued that the highest summer temperature in Alaska occurred as late as 8–6 kyr. Hence our simulation agrees better with the interpretation of Kaufman et al. (2004). In general, our simulation with full forcings was able to capture main temperature features indicated in proxy-based reconstructions.

Shakun et al. (2012) and Marcott et al. (2013) stacked multiple proxies to construct a record of temperatures since the LGM. Both above stacked reconstructions and our simulation OGIS_FWF-v2 show that the Holocene was generally characterized by an initial warming and subsequent Holocene warm period over the NH extratropics, which indicates the broad consistency between simulation and proxy data. However, there are some disagreements related to seasonality (Fig. 9). Marcott et al. (2013) interpreted the stacked temperature reconstruction as representative of the annual mean climate, whereas it shows a better agreement with our simulated summer temperature than with annual mean value (Fig. 9). One potential explanation for this seasonal mismatch is that some proxy records have seasonal bias toward summer conditions, as has been suggested recently for many marine-based SST reconstructions from high latitudes (Lohmann et al., 2013). Further region-by-region comparisons of these warming rates with proxy records are beyond the focus of this work and will be dealt with in a future publication.

It is clear from our data that the spatial patterns of climate response at the onset of the Holocene can be attributed to the variation in the dominant forcings prevailing in the different areas. Orbital-scale insolation variations are important driving factors for the early Holocene climate. For instance, higher temperatures in Alaska could be attributed to the orbitally induced positive insolation anomaly in combination with an anomalous atmospheric circulation caused by the remnant LIS. The air descended over the cold LIS surface, which created a high surface pressure anomaly that produced a clockwise flow anomaly at the surface, as indicated by the 800 hPa geopotential height (Fig. 10). This induces stronger southerly winds over Alaska, which advected relatively warm air from the south. A potentially different early Holocene atmospheric circulation near the North Atlantic was also found in a proxy record of Steffensen et al. (2008), who reported an abrupt transition of deuterium excess that indicates a temperature change of precipitation moisture sources and is thus indirectly connected to atmospheric circulation changes.

The strong influence of the ice sheets on early Holocene temperatures has been found in previous studies (Renssen et al., 2009, 2012). Simulated lower summer temperatures over northern Canada and northwestern Europe in our OGIS11.5 simulation were the result of such ice-sheet-induced cooling, which would have fully overwhelmed the warming effect of the positive summer insolation anomaly. The ice-sheet cooling effect could partly be explained on a local scale by the enhanced albedo over the ice sheets and by the climate's high sensitivity to albedo change (Romanova et al., 2006). Indeed, the summer surface albedo over the ice sheets was much higher (up to 0.8) than over ice-free surfaces, where the values varied from only 0.1 to 0.5, depending on the vegetation type and the fractional snow cover (Fig. 11). Temperatures could be further reduced by the ice-sheet orography impact. The elevation of ice sheets introduced descending air over the ice-sheet surface, which caused locally cooler conditions. There was also an approximate 0.5 ∘C cold bias induced by the lapse rate effect when compared with the site-based records.

Changes in vegetation and land cover during the early Holocene contributed to climate change as well, especially over ecotonal regions. Modeling studies suggest that deforestation in boreal regions could decrease regional temperatures by up to 1 ∘C due to an increase in surface albedo and related positive feedbacks (Levis et al., 1999; Claussen et al., 2001; Liu et al., 2006). Taking Siberia as an example, the insolation-induced warming was partially offset by the overall higher summer albedo (Fig. 11) induced by the southward expansion of the tundra and/or bare ground and related feedbacks at 11.5 kyr, resulting in a minor warming in summer. The albedo-related feedbacks and the smaller annual insolation anomalies jointly result in a 0.5–2 ∘C cooler in annual climate at 11.5 kyr. We are aware of the potential role of permafrost at high latitudes; however, the discussion of the impact of permafrost thaw is hindered by the fact that our model version did not include a dynamic permafrost module. A version of LOVECLIM that is coupled to a permafrost module (VAMPERS) is currently in development (Kitover et al., 2015), and should enable us to quantify the role of permafrost in a future study.

Meltwater release and sea-ice-related changes also had a footprint in the early Holocene climate. The OGIS11.5 simulation produces a sluggish AMOC in the North Atlantic with the largest decrease being more than 3 Sv. It was also reflected in a shallower overturning circulation at 11.5 kyr compared to the PI simulation as a response to meltwater release (Fig. 12). This slowdown also coincides with the foraminifera data from the Arctic Ocean and the Fram Strait that suggest a reduced northward oceanic heat transport (Thornalley et al., 2009). The slowdown and reduced heat transport led to slightly lower temperatures at high latitudes (western Arctic Ocean) at 11.5 kyr than that at 0 kyr. Likewise, after the meltwater fluxes of the LIS diminished around 7 kyr, strong intensification of the AMOC followed. This sudden intensification of AMOC would explain the rapid Arctic temperature increase that occurred at this time (Fig. 4). However, it is important to note that the temperature decrease was not simply inversely linear with the amount of northward transport of heat since the sea-ice feedbacks further reinforce this change (Roche et al., 2007). In fact, sea-ice coverage in the OGIS11.5 simulation was much more extensive over the Davis Strait (northern Labrador Sea) than the corresponding value in OG11.5 (Fig. 13). This extended sea-ice cover in this region was stronger than the direct cooling effect of the reduced oceanic heat transport. Such an anomaly might be explained by positive feedbacks involving sea ice being active (Renssen et al., 2005). The Greenland Sea warming could be attributed to enhanced convective activity that releases more oceanic heat into the atmosphere. This enhanced convective activity was caused by the shift of deep water formation from the eastern Greenland Sea to the west, which was initially induced by the freshwater discharge from ice-sheet melting. The net response of the climate reflects the impact of a combination of forcings and feedbacks, which showed a high temporal–spatial variability.

The simulation of OGIS_FWF-v2 indicates a stronger cooling (before 9 kyr) in the Arctic and northwestern Europe than those found in the OGIS_FWF-v1 with the strongest temperatures reduction at around 10 kyr. The enhanced freshwater influx from the GIS and the redistributed meltwater from the FIS caused an alteration in the surface ocean freshening in the Nordic Seas, which reduced convective activity (Renssen et al., 2010; Blaschek and Renssen, 2013). Indeed, this reduction led to a further slight reduction of the northward heat transport by the Atlantic Ocean, which was associated with a further AMOC weakening (by 1–2 Sv than in FWF_v1): this in turn produced a slightly stronger cooling at 10 kyr (Fig. S3) and a sea-ice expansion over the Denmark Strait (Figs. 14 and S4). However, the efficiency of the above meltwater flux freshening effect is determined by multiple aspects. The most important factor is the maximum flux of meltwater that was added to the ocean, while the total freshwater amount had only a second-order effect (Roche et al., 2007). Numerous investigations on the behavior of the coupled atmosphere–ocean system suggest that the application of freshwater will not lead to a disruption of the North Atlantic Deep Water production (NADW) as long as a certain threshold is not crossed (Ganopolski et al., 1998; Rahmstorf et al., 2005). Apart from the intensity and duration, the ocean circulation response to freshwater also depends on the location where this freshwater is released. For instance, it is more sensitive to the release of freshwater in the eastern Norwegian Sea than at the St. Lawrence River outlet since the former is closer to the main site with NADW formation (Roche et al., 2010). This is consistent with a previous study by Blaschek and Renssen (2103), who found that freshwater from the GIS did have a tangible impact on the Nordic Seas, even though the total amount was minor. Since the second freshwater scenario (OGIS_FWF-v2) includes a slightly larger FWF from the GIS (compared to that in OGIS_FWF-v1) and the FWF was released in a sensitive area, the location-dependent sensitivity could also partially explain further AMOC weakening in the OGIS_FWF_v2 simulation compared to OGIS_FWF-v1.

The OGIS_FWF-v1 simulation indicates two peaks in the temperature evolution over northwestern Europe, at around 10 and 7 kyr. High temperatures at 7 kyr are recorded in proxy-based reconstructions as well. However, no warm peak at 10 kyr was observed in pollen-based reconstructions, which actually suggests a cooler climate prevailed at 10 kyr than in the preindustrial Europe (Mauri et al., 2015). In contrast to the climate simulated in OGIS_FWF-v1, the simulation with updated freshwater (OGIS_FWF-v2) produced a warming trend that is consistent with a highest temperature around 7 kyr. Moreover, the OGIS_FWF-v1 produced a temperature decrease between two peaks, whereas the proxies indicated a rapid temperature increase at the beginning followed by a more gradual warming (Brooks et al., 2012). Therefore, from the viewpoint of temperature evolution in northwestern Europe, the OGIS_FWF-v2 represented a more realistic climate than OGIS_FWF-v1 did, which implies that the existing uncertainties in the reconstructions of ice-sheet dynamics can be evaluated by applying different freshwater scenarios. Further comparison with proxy data and with other model transient simulations will be conducted in a future paper.