This study analyzes a new transient simulation result based on CESM, named NNU-Hol (Nanjing Normal University-Holocene), which considers more comprehensive external forcings for the Holocene climate change. CESM is a fully coupled, global climate model, which was launched by the National Center for Atmospheric Research (NCAR) in June 2010. It is an Earth system model developed on the basis of CCSM4. CESM includes components for atmosphere, ocean, sea ice, and land surface and considers atmospheric chemistry, biogeochemistry, and anthropogenic forcing. It is widely used to study the mechanism of the changes in climate and the environment and the interaction between natural and anthropogenic forcing for the climate and for scenarios of future climate change. CESM is coupled with several advanced modules, including CAM5 (Community Atmosphere Model 5) used by the atmosphere module, POP2 (Parallel Ocean Program 2) used by the ocean module, CLM4 (Community Land Model 4) used by the land module, CICE (the Los Alamos National Laboratory sea ice model) used by the sea ice module, and CISM2.0 (The Glimmer Ice Sheet Model 2.0) used by the land ice module. CISM2.0 is deactivated in the NNU-Hol simulation. For a more detailed introduction, please visit the official website of CESM (http://www.cesm.ucar.edu/models/cesm1.0/notable_improvements.html, last access: 6 March 2023)).

In NNU-Hol, CESM 1.0.3 was configured to simulate the transient climate evolution of the Holocene period at a horizontal resolution of 3.75∘×3.75∘, forced by several external forcings (orbital parameters, solar irradiance, volcanic eruptions, greenhouse gases, and land use and land cover) accelerated by a factor of 10. With this acceleration method (Lorenz and Lohmann, 2004), climate trends and feedbacks from the past 11.95 kyr to 1990 CE, imposed by the external-forcing-driven changes, are represented in the experiments with 1199 model years to save computation resources. The solar irradiance forcing comes from the reconstruction of Vieira et al. (2011), aggregated to 10-year average timescale of solar forcing and prescribed in the simulation. The volcanic eruption comes from the ice-core-based reconstructions of Wan et al. (2020) and Sigl et al. (2015). For the volcanic forcing, the volcanic events during the 10-year period were integrated into one volcanic eruption event. On the basis of this assumption, the horizontal diffusion of lower-stratospheric aerosols was calculated using the stratospheric transport parameters. Based on the stratospheric–tropospheric folding and BD (Brewer Dobson) circulation theory, latitude- and time-dependent functions were used to describe aerosol production and deposition (Grieser and Schonwiese, 1999; Holton et al., 1995). The greenhouse gas forcing data use the reconstruction based on ice cores in Joos and Spahni (2008). The data for land use and extra land cover forcing comes from the HYDE 3.2.1 (the History Database of the Global Environment, referred to as HYDE version 3.2.1) (Klein Goldewijk et al., 2017) data set. The orbital parameters come from Berger (1978). The external forcing time series used in the NNU-Hol simulation is shown in the Supplement. Wan et al. (2020) found that the global annual average temperature in the NNU-Hol “All” forcing simulation and reconstruction records from Marcott et al. (2013) has similar trends and strengths, decreasing by about 0.5 K during 5.0 to 0.15 ka. We explore the characteristics of changes in Arctic temperature during the Holocene and try to understand its underlying mechanism based on the All forcing (AF) and orbital forcing (ORB) simulations in NNU-Hol.

There are, to our knowledge, five sets of climate simulations published so far covering the entire Holocene period (0–11 700 ka), namely ECBilt-CLIO (Timm and Timmermann, 2007), FOAM (Kutzbach et al., 2008), TraCE-21ka (Liu et al., 2009), FAMOUS (Smith and Gregory, 2012), and LOVECLIM (Timmermann et al., 2014). Except for TraCE-21ka and one of the ECBilt-CLIO simulations, these simulations are accelerated by different factors. Additional un-accelerated simulations, such as the simulations based on MPI-ESM (Bader et al., 2020) and IPSL (Braconnot et al., 2019) covering only part of the Holocene period (0–8 and 0–6 ka, respectively) have also been published in recent years. The external forcings considered in these simulations are generally a part of the combination of orbital forcing (ORB), greenhouse gases (GHGs), continental ice sheets (ICE), meltwater flux (MWF), volcanic forcing, land use forcing, and ozone forcing.

In order to better understand the Holocene evolution of the earth system, a comprehensive database of paleoclimate records was compiled by Kaufman et al. (2020), which is named the Temperature 12k database.

The Temperature 12k database is a global compilation of good-quality, published, temperature-sensitive proxy records (such as lake sediment, marine sediment, peat, glacier ice, and pollen) throughout the entire Holocene period. The data are mainly collected from previously published research, containing 1319 records (1162 from the Northern Hemisphere and 157 from the Southern Hemisphere), distributed across 679 sites (including 470 terrestrial and 209 marine sites), where the time period covers at least 4000 years. For our study we only selected records with a resolution finer than 400 years: 15 % of these records have a resolution of 50 years or finer, 39 % have a resolution of 51 to 150 years, and 21 % have a resolution of more than 250 years.

The data are mainly based on a collection of published reconstructions of Holocene temperature especially in the Northern Hemisphere (Routson et al., 2019; Marcott et al., 2013; Sundqvist et al., 2014; Chen et al., 2008; Wanner et al., 2011). Part of the global paleotemperature data are collected from the PAGES 2k Consortium 9 database. The compilation of Marsicek et al. (2018) provides most pollen-based paleotemperature records, and the other part comes from public repositories (such as PANGAEA and the World Data Service for Paleoclimatology, NOAA). The paleotemperature records at 209 marine sites are mainly from the US-based Data Assimilation for Deep Time (DADT) project and the compilations of the German Climate Modeling Initiative (PalMod) (Jonkers et al., 2020). In this database, uncertainties are estimated by various methods. Some studies characterize uncertainty based on calibration and proxy bias or measurement error, while others represent uncertainties after rigorous cross-validation. The uncertainties in most of these paleotemperature records from North American and European pollen and most of the marine sediment-based records (except for those from microfossil assemblages) are calculated using the Bayesian procedure in early studies (Malevich et al., 2019; Tierney and Tingley, 2014, 2018; Tierney et al., 2019). For the other proxy types, most of the paleotemperature values and their uncertainties are based on multigenerational analysis and calibration methods. Specific information about the uncertainty for each proxy recorded is available in Supplement Table 2 of Kaufman et al. (2020) as well as in their paper itself. More detailed information can be found at https://www.ncei.noaa.gov/access/paleo-search/study/27330 (last access: 6 March 2023).

In order to investigate the changes in temperature over the Arctic area during the Holocene, we selected records from the site, which are located above 60∘ north latitude, and eliminated some records with shorter time series (missing more than 2500 years during the Holocene), which leaves us with 58 records.

We focus on long-term temperature changes in the Arctic during the Holocene. The significance test used in this study was calculated according to a two-tailed Student's t test at the 90 % (α = 0.1) or 95 % (α = 0.05) confidence level. Student's t test was used to compare the means of two groups and determine if the difference in means is statistically significant and was also used to test the statistical significance for each grid point in the figures in this paper. The sample size of the Pacific Arctic region in the temperature proxy data is small and thus has a small degree of freedom. The potential impact of temporal and spatial correlation is not taken into account in the analyses. We apply empirical orthogonal function (EOF) analysis, also known as principal component analysis (PCA), to sea level pressure changes in the Northern Hemisphere. EOF analysis is a standard analytical technique used in climate science to study patterns of spatial variability. The EOF is obtained by computing the eigenvectors and eigenvalues of the spatially weighted covariance matrix of the temperature field. Equation (24) from North et al. (1982) was used as a statistical test to evaluate the separation of EOF eigenvalues (leading modes). Applying the EOF to the Northern Hemisphere sea level pressure is a common method to study the Arctic dipole mode. The objective is to show the variation in the Arctic dipole mode during different periods of the Holocene (0–2 and 5–8 ka). As described in Sect. 3 below, the second mode of the EOF for the Holocene 5–8 ka period explains 11.5 % and that for the 0–2 ka period explains 16.3 % of the sea level pressure variation.

In order to demonstrate asymmetric temperature changes in the Arctic during the Holocene, we analyzed the Temperature 12k database (Kaufman et al., 2020). The results are shown in Fig. 1, where the circles represent the site locations of the 52 selected records. The proxy records are mainly concentrated on the Atlantic coast and in Europe, Greenland, and northern Canada. The average temperature across the Arctic showed a cooling trend during the Holocene, which is consistent with previous findings (Marcott et al., 2013; Kaufman et al., 2004). To study the asymmetric changes in temperature, we divided the Arctic region into two parts: the Pacific Arctic (>60∘ N, 90∘ W–59∘ E) and the Atlantic Arctic ( >60∘ N, 60∘ E–91∘ W). The proxy data show that there is an asymmetric cooling between the two regions, with an average cooling of -1.54 ∘C in the Atlantic Arctic and -0.61 ∘C in the Pacific Arctic, showing significance at the 90 % confidence level (p<0.10) on Student's t test. The degrees of freedom of the temperature proxy data in the Pacific and Arctic region are small. The histogram shows that only two proxy sites have extreme temperature variations in the Atlantic Arctic region. However, combined with the box plot of the temperature changes (Fig. S1 in the Supplement), the asymmetric temperature variation in the two regions is still considered robust. To further study the robustness of this asymmetric change in temperature, we likewise analyze it with model data, allowing us to also investigate the seasonal changes.

Here we use the model data based on CESM and the NNU-Hol, and we choose data from the All forcing simulation (AF) and the orbital forcing simulation (ORB) in order to assess the most realistic temperature changes (AF) and the role of the orbital forcing (ORB) in the changes during the Holocene. As shown in Fig. 2, to better understand the asymmetry, we further analyze the temperature changes in two seasons: DJF (December, January, and February) and JJA (June, July, and August). The results from AF and ORB show a similar pattern. In the JJA, the Arctic exhibits a comparable cooling trend in the Atlantic sector and the North Pacific sector, with cooling centers in Eurasia and northeastern America. However, there is significant regional asymmetric cooling in the DJF (-0.67 ∘C for Atlantic Arctic average cooling and 0.09 ∘C for the Pacific Arctic average warming in AF; the difference is statistically significant at the 95 % confidence level on Student's t test), with the Arctic Ocean, the North Atlantic Ocean, and Greenland cooling while the land and North Pacific Ocean are warming. Areas that are dotted are significant at the 90 % confidence level in Fig. 2. Northern Canada, where numerous proxy sites are located, is the center of the temperature warming, with the cooling center changing from the Norwegian Sea in the AF to the North Atlantic in the ORB. We can see more clearly the difference in temperature asymmetry in the two seasons through Fig. 3, and the intensity of the asymmetry is more pronounced at the proxy sites. In the DJF of AF, the average cooling is -0.44 ∘C for the Atlantic Arctic proxy site and the average warming is 0.85 ∘C for the Pacific Arctic proxy site, while in the ORB it is -0.25 ∘C and 1.61 ∘C, respectively. The difference in the mean temperature changes in the two regions is significant at the 90 % confidence level in both AF and ORB. Figure 3 shows that the temperature difference between the Pacific and Atlantic Arctic is neglectable for either the regional average or the sites' station average in JJA. Therefore, the asymmetric temperature changes in the Atlantic and Pacific Arctic are dominated by the changes in DJF.

Past research has shown that sea ice is always an important factor when we discuss Arctic climate change (Jenkins and Dai, 2021). Arctic temperature changes are often tightly linked to sea ice changes, with temperature causing changes in the expansion of sea ice cover and therefore changes in surface albedo, further amplifying climate change in the Arctic region (Wohlfahrt et al., 2004; Renssen et al., 2005; Braconnot et al., 2007). The simulations show that the temperature asymmetry is closely related to the asymmetric change in sea ice pattern. The difference in March sea ice area between the two periods (0–2 ka and 5–8 ka) is fairly consistent in the AF and ORB simulations. Consistent with previous studies of Holocene sea ice concentration proxies, sea ice concentrations were low in the early to mid-Holocene and increased in the mid- to late Holocene during the Neoglacial (Jennings et al., 2002; de Vernal et al., 2005; Müller et al., 2012). For the Atlantic Arctic, the sea ice concentration of the Chukchi Sea was not always low during the Early Holocene, while there were millennial oscillations and minimum values in the Chukchi Sea during the Neoglacial (de Vernal et al., 2005). Müller et al. (2012) showed that for the North Atlantic sea ice reconstruction P25 decreased significantly in the early Holocene, while in the middle Holocene (7000–3000 years BCE), sea ice gradually increased. The maximum value was gradually reached in 3000–300 years. In our model, sea ice increased throughout the Arctic during the Holocene, especially around the Norwegian Sea and Davis Strait, with the exception of the Barents Sea, where sea ice area declined significantly (significant areas are dotted in Fig. 4). The large expansion of sea ice in the North Atlantic and the slight decrease in sea ice in the North Pacific leads to strong differences in surface albedo between the two regions, with more reflection of radiation in the North Atlantic and more absorption in the North Pacific, which contributes to the stronger asymmetry during winter (Fig. 3).

To understand the dynamical aspects of the stronger cooling in the North Atlantic, we investigate the regional-scale atmospheric circulation and surface wind variations related to sea ice transport. We focus our analysis on the December–February season because these months yield the largest impact of temperature asymmetry. In DJF, the sea level pressure (SLP) has a distinct dipole distribution. Sea level pressure increases in the North Atlantic and decreases in the North Pacific. The low pressure in the North Pacific intensifies during the late Holocene. Based on data from the past decades, many studies (Wu et al., 2006; Niebauer et al., 1999; Stabeno et al., 2001; Rodionov et al., 2005) have demonstrated that the stronger low pressure transported warm air from the south to the North Pacific, resulting in higher temperatures in the North Pacific, offsetting parts of the Holocene cooling trend in the North Pacific and causing sea ice expansion in the North Atlantic due to atmospheric circulation. We suspect that this contributes to the regional asymmetry in Arctic temperatures. Compared with the AF results, the ORB simulation shows that the sea level pressure changes in the two regions contrast more and the difference is larger. This suggests that orbital forcing plays a dominant role in generating this asymmetry.

Previous studies have established that atmospheric circulation anomalies in the Arctic atmosphere, dipole structure anomalies, are closely related to sea ice (Wu et al., 2006; Watanabe et al., 2006; Choi et al., 2019). The Arctic dipole strongly influences sea ice movement and sea ice export. The empirical orthogonal function (EOF) method has commonly been used to study the Arctic dipole in past studies (Wu et al., 2006; Skeie, 2000; Wang and Ikeda, 2000). The EOF's first leading mode of the monthly mean SLP north of 60∘ N in winter corresponds to the Arctic Oscillation pattern, and the second leading mode of the EOF corresponds to the Arctic dipole anomaly. In order to confirm our conjecture that changes in SLP and sea ice play a role in promoting regional temperature asymmetry, we analyzed the Arctic dipole, and the results are shown in Fig. 6. The first and second leading modes are statistically independent at the 95 % confidence level on the North significance test (North et al., 1982). In the AF simulation, the EOF pattern of the SLP is similar for both periods (0–2 and 5–8 ka), with the second mode showing an Arctic dipolar distribution. There are two action centers with opposite signs: one center is located over the North Pacific and the other exists over northern Eurasia (significant at the 90 % confidence level), accounting for 11.5 % (5–8 ka) and 16.3 % (0–2 ka) of the variance. Figure 6c and d show that the PC2 (second principal component) time series of the EOF in the mid-Holocene oscillated repeatedly, while the PC2 in the late Holocene showed a strengthening trend. Combined with the stronger SLP of the late Holocene shown in Fig. 5, it can be concluded that the late Holocene has a stronger Arctic dipole pattern.

Regressing on the PC2 of the EOF allows us to understand how the Arctic dipole pattern affects the changes in sea ice and wind, and the corresponding sea ice field and wind field can explain the physical mechanism of the Arctic dipole effect on temperature asymmetry. To explore our conjectures, we compared two spatial patterns of sea ice and U and V wind field regressions to the PC2 (Fig. 7). By regressing the sea ice distribution onto the second principal component time series (Fig. 7), a pattern consistent with past sea ice reconstructions (de Vernal et al., 2005; Müller et al., 2012; Jennings et al., 2002), we show that the Arctic dipole in the late Holocene had a greater role in influencing sea ice. In the late Holocene, the Bering Strait, Chukchi seas, and East Siberia were influenced by warm winds from the south, promoting strong ice melting, pushing the ice away from the coast and increasing the temperature in these areas. On the other hand, the pressure pattern causes the Arctic Ocean sea ice to expand to the North Atlantic Ocean through Fram Strait transfer. This leads to more sea ice in the Atlantic Ocean and, therefore, more radiant heat reflectance and surface temperature cooling in the Atlantic Arctic due to the sea ice feedback. This pattern contributes to the asymmetry of temperature variations between the North Atlantic and the North Pacific. In the late Holocene, the wind field characteristics are more pronounced than those in the early to mid-Holocene, and thus the temperature asymmetry is more significant.

Figure 8 shows the average change in the regression of temperature and sea ice onto the PC2, and the differences in sea ice and temperature between the two regions are mainly contributed by the late Holocene period. The regression of sea ice on the PC2 of SLP increased by 1.71 % in the North Atlantic and 0.44 % in the North Pacific during the late Holocene, while it decreased by -1.60 % and -0.91 % during the early to mid-Holocene (significant at the 90 % confidence level). The corresponding asymmetric cooling in the early to mid- to late Holocene was -2.74 ∘C in the North Atlantic and -1.83 ∘C in the North Pacific (significant at the 90 % confidence level). The results of the ORB simulations in terms of asymmetry are similar to the AF simulations but stronger. Figure 7 shows that in the ORB simulation, the late Holocene had stronger warm southerly winds and more open water in the North Pacific, while sea ice expanded more widely in the North Atlantic, resulting in stronger asymmetric temperature. The regression of temperature shows a cooling of -3.37 and -0.05 ∘C in the North Atlantic and North Pacific (significant at the 90 % confidence level), respectively, during the late Holocene. Therefore, the regression of the U and V wind, the sea ice, and the temperature on the PC2 show the contribution of the Arctic dipole to the temperature asymmetry, The result indicates that it is an important factor affecting the temperature asymmetry in the Arctic.