Modelling a Modern-like-pCO2 Warm Period (MIS KM5c) with Two Versions of IPSL AOGCM

The mid-Piacenzian warm period (3.264 to 3.025 Ma) is the most recent geological period with a present-like atmospheric pCO2 as thereby it is expected to have exhibited a warm climate similar to or warmer than present day. On the 15 basis of understanding that has been gathered on the climate variability of this interval, a specific interglacial (marine isotope stage KM5c, MIS KM5c, 3.205 Ma) has been selected for the Pliocene Model Intercomparison Project phase 2 (PlioMIP 2). We carried out a series of experiments according to the design of PlioMIP2 with two versions of the IPSL AtmosphereOcean Coupled General Circulation Model (AOGCM): IPSL-CM5A and IPSL-CM5A2. Compared to PlioMIP 1 experiment, run with IPSL-CM5A, our results show that the simulated MIS KM5c climate presents enhanced warming in mid-to-high 20 latitudes, especially over oceanic regions. This warming can be largely attributed to the enhanced Atlantic Meridional Overturning Circulation caused by the high latitude seaway changes. The sensitivity experiments, conducted with IPSLCM5A2, show that besides the increased pCO2, both modified orography and reduced ice sheets contribute substantially to mid-to-high latitudes warming in MIS KM5c. When considering the pCO2 uncertainties (+/-50 ppmv) during the Pliocene, the responses of the modelled mean annual surface air temperature to changes to pCO2 (+/-50 ppmv) is not symmetric, 25 which is likely due to the non-linear response of the cryosphere (snow cover and sea ice extent). By analysing the Greenland Ice Sheet surface mass balance, we also demonstrate its vulnerability under both MIS KM5c and modern warm climate.


Introduction
The mid-Piacenzian warm period (MPWP; 3.264 to 3.025 Ma) is the most recent geological period with a present-like pCO2 concentration and exhibited significant warming relative to today. This interval has been intensively studied during the past three decades as this time period is generally considered to be a potential analogue for the future warmer climate. There is an abundance of marine and terrestrial data that allow us to reconstruct the ocean/land temperatures, soil and vegetation 5 conditions for this period. The reconstructed pCO2 for the MPWP ranges from 350 to 450ppmv (Bartoli et al., 2011;Pagani et al., 2010;Martí nez-Botí et al., 2015), which bracket the present-day level. The MPWP is thought to be globally warmer by 2-4C than preindustrial climate (e.g., Dowsett et al., 2009). A large warming amplification of 7-15C is estimated in arctic regions derived from terrestrial proxies from the lake El'gygytgyn in NE arctic Russia (Brigham-Grette et al., 2013) and Ellesmere Island in North Arctic circle (Rybczynski et al., 2013). The meridional SST gradient is reduced compared to 10 the present day due to the amplified warming in the high latitudes. The zonal SST gradient is much weaker than present day (Wara et al., 2005;Ravelo et al., 2006;Fedorov et al., 2013). Different causes have been investigated for this weaker zonal SST gradient during the Pliocene. Brierly et al (2009) argue that the ocean warm pool expansion over most of the tropics can be responsible for the reduced zonal SST gradient. Some researchers argue that a reduction in the meridional gradient of cloud albedo can sustain the reduced zonal and meridional SST gradient (Burls and Fedorov 2014). Reconstruction of 15 vegetation distribution indicates a northward shift of boreal forest at the expense of tundra regions due to the warmer conditions (Salzmann et al., 2008). Associated with this strong warmth, the eustatic sea level is estimated to have been 22(+/-10m) higher (between 2.7 and 3.2 Ma) than present (e.g., Miller et al., 2012) suggesting a complete disintegration of Greenland ice sheet and a significant collapse of the West Antarctic Ice sheet as well as unstable regions of East Antarctic (Hill, 2009;Dolan et al., 2015;Koenig et al., 2015). 20 An early motivation for studying this period was to apply the knowledge thus gained to the issue of the ongoing climate change. However, considering the non-equilibrium state of the present and the future climate due to the continuously changing anthropogenic forcing factors, the simulated quasi-equilibrium MPWP may not be directly regarded as an analogue for future warming (Crowley, 1991). The importance of studying the MPWP nowadays is to investigate the abilities of 25 climate models to produce warm climates and to study the relative impacts of forcings and feedbacks of internal climate components under warm conditions, and which can assist in developing future climate projections. In Pliocene Model Intercomparison Project phase 1 (PlioMIP1), 11 models conducted the MPWP experiments. Among these models, there is agreement with regards to surface temperature change in the tropics but a lack of agreement on temperature changes at high latitudes as well as total precipitation rate in the tropics (Haywood et al., 2013). The modeled Atlantic Meridional 30 Overturning Circulation and the associated ocean heat transport for this interval in models are not very different compared to modern conditions (Zhang et al., 2013). However, when comparing to proxy data of sea surface and surface air temperature, climate models uniformly underestimate the warming in the high latitudes (Dowsett et al., 2012, Haywood et al., 2016b. Reasons for this discord between data and model are complex, but they can be attributed to three main aspects: boundary conditions uncertainty, modeling uncertainty (e.g., the model bias, annual variability in the produced climatology fields) and data uncertainty (Haywood et al., 2013). In PlioMIP1, the MPWP is regarded simply as a stable interval despite the climate variability existing over a 300-kyrs time slab due to the climate sensitivity and orbital parameters' change, thus the boundary conditions are made as an averaged condition over this long interval, whereas proxy data are representative of 5 some orbital conditions inside this time slab. This boundary conditions uncertainty is thus considered as the main contributor to this data-model discrepancy (Haywood et al., 2016a). Therefore, the PlioMIP phase 2 (PlioMIP2) switched to choosing a representative interglacial during the MPWP interval: marine isotope stage KM5c (MIS KM5c; 3.205 Ma). Thus, boundary conditions (known as PRISM4; Dowsett et al., 2016) have been updated for PlioMIP2, which include a new paleogeography reconstruction containing ocean bathymetry, and land/ice surface topography, which represent closure of Bering Strait and 10 North Canadian Archipelago region and a reduced Greenland ice sheet by 50% in comparison to PlioMIP1. Besides, extra information of lake distribution and soil types (Pound et al., 2014) are also provided, but not used in this paper.
This study is conducted in the framework of PlioMIP2. Here we employ the new PRISM4 boundary conditions to conduct the MPWP experiments by using two French AOGCM models: IPSL-CM5A and the updated IPSL-CM5A2. The purpose of 15 this study is to better understand the warm climate of the MPWP and to study the sensitivity of the IPSL model to changes of boundary conditions, such as changes of land-sea mask and pCO2. As IPSL AOGCM model has participated in PlioMIP1 (Contoux et al., 2012), we also compare the modelling results of PlioMIP2 with those of PlioMIP1 to quantify the impact of the high latitude seaways' changes on the climate system.

2 Model Descriptions
To accomplish the modelling work, we have employed the two versions of Institute Pierre-Simon Laplace (IPSL) coupled atmosphere-ocean general circulation model (AOGCM): IPSL-CM5A and IPSL-CM5A2. IPSL-CM5A is a low resolution coupled model which has been applied in CMIP5 for historical and future simulations (Dufresne et al., 2013) as well as for 25 Quaternary and Pliocene paleoclimate studies (Kageyama et al., 2013;Contoux et al., 2012). in prep) is an updated version of IPSL-CM5A. Critical changes from IPSL-CM5A include (i) technical developments to make IPSL-CM5A2 run faster (64yrs/day in CM5A2 instead of 8 years per day in CM5A), (ii) updates of the versions of components and (iii) a major re-tuning of the cloud radiative forcing to correct the cold bias in the mid and high latitudes that is known to be present in CM5A. Thus, to compare with PlioMIP1 results (Contoux et al., 2012), we carried out

developed at Laboratoire de Mé té orologie Dynamique in
France. This is a complex model that incorporates several processes decomposed into a dynamic part that calculatesthe numerical solutions of the general equations of atmospheric dynamics, and a physical part, calculating the details of the climate in each grid point and containing parameterizations processes such as the effects of clouds, convection, orography 5 (LMD Modelling Team, 2014). Atmospheric dynamics are represented by a finite-difference discretization of the primitive equations of motion (e.g., Sadourny and Laval, 1984) on a longitude-latitude Arakawa C-grid (e.g., Kasahara, 1977). The horizontal resolution of the model is 96x95, corresponding to an interval of 3.75 degrees in longitude and 1.9 degrees in latitude. There are 39 vertical levels, with around 15 levels above 20 km. This model has the specificity to be zoomed (the Z of LMDZ) if necessary on a specific region and then may be used for regional studies (e.g., Contoux et al., 2013). In IPSL-10 CM5A2, re-tuning of the model has been done by altering the cloud radiative effect to decrease the cold bias of the model.
More details can be found in Sepulchre et al (in prep).

15
The land component in both CM5A and CM5A2 is ORCHIDEE (Organizing Carbon and Hydrology In Dynamic Ecosystems (Krinner et al., 2005) which is comprised of three modules: hydrology, vegetation dynamics and carbon cycle.
The hydrological module (Ducoudré et al., 1993) describes the exchange of energy and water between the atmosphere as well as the biosphere, and the soil water budget (Krinner et al., 2005). Vegetation dynamics parameterization is derived from the dynamic global vegetation model LPJ (Sitch et al., 2003;Krinner et al., 2005). The carbon cycle model simulates plant 20 phenology and carbon dynamics of the terrestrial biosphere. Vegetation distribution is described using 13 plant functional types (PFTs) including agricultural C3 and C4 plants, which are not present in the MPWP simulations. In this case, hydrology and carbon modules are activated, but vegetation is prescribed as the PlioMIP1 study by Contoux et al. (2012), using 11 PFTs, derived from the PRISM3 vegetation dataset (Salzmann et al., 2008). Therefore, soil, litterfall, and vegetation carbon pools (including leaf mass and thus LAI) are calculated as a function of dynamic carbon allocation.

Ocean and sea ice
The ocean model included in IPSL-CM5A is NEMOv3.2 (Madec, 2008) which includes three principle modules: OPA (for the dynamics of the ocean), PISCES (for ocean biochemistry), and LIM (for sea ice dynamics and thermodynamics). The 30 configuration of this model is ORCA2.3 (Madec and Imbard, 1996), which uses a tri-polar global grid and its associated physics. The average horizontal resolution is 2° by 2°, which increases to 0.5° in the tropics and there are 31 layers in 5 vertical. Temperature and salinity advection are calculated by a total variance dissipation scheme (Lé vy et al., 2001;Cravatte et al., 2007). The mixed layer dynamics is parameterized using the Turbulent Kinetic Energy (TKE) closure scheme of Blanke and Delecluse (1993) improved by Madec (2008). The sea ice module LIM2 is a two-level thermodynamic-dynamic sea ice model (Fichefet and Morales Maqueda 1997). Sensible heat storage and vertical heat conduction within snow and ice are determined by a three-layer model. OASIS coupler (Valcke, 2006) is used to interpolate and exchange the variables and 5 to synchronize the models. This coupling and interpolation procedures ensure local energy and water conservation. New version NEMOv3.6 is included in IPSL_CM5A2 in which the river runoffs are now added through a non-zero depth and have a specific temperature and salinity. The coupling system has been switched from OASIS3.3 to OASIS3-MCT (for Model Coupling Toolkit). More details are provided by Sepulchre et al (in prep).

3 Experiment Design
This section describes the boundary and the initial conditions imposed in our experiments. Here, the experiment names are generally consistent with the design of PlioMIP2 (Haywood et al., 2016a) and they are referred to by the abbreviated form E(x)(c), where c is the concentration of atmospheric CO2 in ppmv and x represents boundary conditions that have been 15 changed from the pre-industrial (PI) conditions, such that x can be absent for cases in which no boundary conditions have been modified or it can be "o" for a change in orography and/or "i" for a change in land ice configuration. Because we report on experiments performed with two versions of the IPSL model, we indicate the experiment conducted using the updated version of the model by the suffix "_v2". 20

Pre-industrial experiments
The pre-industrial control simulation in IPSL-CM5A was performed as required by CMIP5/PMIP3 by the LSCE modelling group. It is a 2800-years simulation, which already started from equilibrium conditions. The pre-industrial control simulation in IPSL-CM5A2 was conducted by Sepulchre et al., (in prep) forced by CMIP5 pre-industrial boundary conditions and has 25 3000-years integration length.

Pliocene experiments
We have conducted six AOGCM experiments for PlioMIP2, they are the core experiment Eoi400 using the IPSL-CM5A 30 model and the core experiment Eoi400_v2 as well as four tiered experiments E400_v2, Eoi450_v2, Eoi350_v2, Eo400_v2 using the IPSL-CM5A2 model. The pCO2 concentration in each experiment is indicated in the experiment name as mentioned above. Other greenhouse gases and orbital forcing are kept the same as in the IPSL PI control run (Table 1).
Vegetation is kept the same as in the PlioMIP1 AOGCM simulation by Contoux et al. (2012). Soil patterns and river routing are kept the same configurations with PI control, except the regions where the changes to the topography modify the river routing and the estuaries. Land sea mask in these experiments is modified from present, only by closing the Bering Strait and the North Canadian Archipelago region, and by modifying the topography in the Hudson Bay ( Figure 1). Ice sheet mask is changed to PRISM4 dataset  except in Eo400_v2 experiment, in which the modern ice sheet is imposed.
Topography in these five experiments is calculated based on modern topography used in the IPSL model, on which the 5 anomaly between the PRISM4 reconstructed topography and the modern topography provided by PlioMIP2 database (Haywood et al., 2016a) is superimposed. Wherever the resulting topography was lower than zero, it was replaced by the absolute PRISM4 topography. Figure 1 shows the resulting topography anomalies in our PlioMIP2 experiments respectively compared to PI and PlioMIP1 experiments. The initial sea surface temperature and sea ice in Eoi400 and Eoi400_v2 are derived from the IPSL PlioMIP1 AOGCM simulation (Contoux et al., 2012). Eoi400 has run for 800 modelling years and 10 the initial condition is from the equilibrium state of PlioMIP1 experiment (Contoux et al., 2012), which has 650-years integration length. Eoi400_v2 has run for 1500 modelling years. Average climatologies for these two experiments are calculated over the last 50 years. Four tiered experiments: E400_v2, Eoi450_v2, Eoi350_v2, Eo400_v2 are conducted based on the equilibrium state of Eoi400_v2 core experiment and have 400 years of integration length. Average climatologies for these four experiments are calculated over the last 30 years. Table 2 summarizes the aforementioned information. Figure S1 15 shows time series of surface air temperature and deep ocean temperature at around 2.3km depth. For both core simulations, the trend in both the global mean surface air temperatures (< 0. 18C century -1 ) and the deep ocean temperature (< 0.05C century -1 ) over the final 50 years of integration are small. The tiered experiments also show relatively stable trends over the last 30 years of integration (< 0.2C century -1 and < 0.08C century -1 in surface air temperature and sub-surface ocean temperature respectively). Therefore, we conclude that model runs have reached a quasi-equilibrium state.

20
Although a standard pCO2 of 400ppmv is selected for the Pliocene core experiments, the pCO2 records during this interval mostly range from 350 to 450ppmv. Thus, the tiered experiments Eoi450_v2 and Eoi350_v2 are conducted to investigate the impact of pCO2 uncertainty on the modelled Pliocene climate. The tiered experiments E400_v2 and Eo400_v2 combined with the core experiment Eoi400_v2 and PI control are used to quantify the relative importance of pCO2, land ice and 25 orography in the PlioMIP2 warmth. Because of the limited computational resources, we apply the linear decomposition for the forcing factors as: dTCO2 = E400_v2 -E280_v2 (1); dTorography = Eo400_v2 -E400_v2 (2); dTland_ice = Eoi400_v2 -E400_v2 (3)T = dTCO2 + dTorography + dTland_ice (4).  Figure 2 shows the anomalies of global mean annual near surface air temperature (SAT, i.e. temperature at 2 meters), precipitation rate and sea surface temperature (SST) between PlioMIP experiments and pre-industrial control with IPSL-CM5A. The global mean annual SAT in Eoi400 is 14.4C which is 2.3C warmer than that of pre-industrial. The warming in Northern Hemisphere (NH) high latitudes (>50N) (4.2C) is higher than that in the tropics (1.8C). The magnitude of the 5 warming for Eoi400 is slightly larger than that for PlioMIP1 experiment, which shows a global warming by 2.1C. The major differences in SAT between Eoi400 and PlioMIP1 are found respectively in mid-latitude Eurasia and arctic regions due to the change of regional topography and high latitude seaways as well as the reduced Greenland ice sheet. Thus, Eoi400

Results and Discussion
shows a reduced meridional temperature gradient than that in PlioMIP1 experiment.

10
The global mean annual precipitation rate increases by 0.14 mm/d in Eoi400 due to the vapour varying capacity of the warmer atmosphere and the increase is mostly confined to the global monsoon regions and tropical oceans. The increase in global mean precipitation rate as well as the monsoon area index ( Figure S2, calculated based on the method of Wang et al (2008)) in Eoi400 compared to PI is similar to that in PlioMIP1. However, regional discrepancies still exist between these two experiments: the precipitation rates in Eoi400 in the tropics and NH high latitudes are higher than those in PlioMIP1 by 15 0.03 -0.05 mm/d because of the increased warming in Eoi400 in these regions. Regional differences also exist over mountainous regions (e.g., the Andes, the Rockies, Tibetan Plateau, the Himalayas and the Ethiopian Highlands) since the elevation over these regions is modified largely in PlioMIP2 compared to PlioMIP1 (Figure 1). In East Africa, Eoi400 simulates an intensified precipitation than PlioMIP1, which is better consistent with proxy data from East Africa inferring a wet vegetation condition and hydrological systems during this period (Drapeau et al.,2014;Bonnefille 2010). Apart from the 20 high latitude seaways' change, the regional difference in topography between PlioMIP2 and PlioMIP1 can also contribute to the rainfall change. Further sensitivity studies are needed to verify it.

25
The global mean annual SST of Eoi400 is 1.7C warmer compared to the pre-industrial. It is 0.3C warmer than PlioMIP1 and the warming is largely confined to the mid to high latitude oceans of the Northern Hemisphere. The warming in Eoi400 relative to PlioMIP1 can be attributed to the closure of the Bering Strait and the Canadian Archipelago, which is the major difference in the boundary conditions between these two experiments. In the preindustrial control run (Figure 3a . This is because the simulated modern AMOC (11 Sv) with this model is much weaker than the observations. Moreover, the simulated AMOC in PlioMIP1 with our model is also weaker than other models (Zhang et al.,2013). As shown in Figure 6, the total heat transport in PI control, Eoi400 and PlioMIP1 simulations is similar. The stronger AMOC in Eoi400 indeed strengthens the northward heat transport in the Atlantic Ocean, while the weakened Pacific meridional ocean circulation in Eoi400 (PMOC, Figure S3), which contrasts 20 with the data-based findings by Burls et al (2017), decrease the northward heat transport, thus leading to very slight change in total ocean heat transport. This compensation was also found by Chandan and Peltier (2017).
The simulated warm conditions in high latitudes prevent sea ice from expanding during winter season and increase sea ice melt during summer season (Figure 7). When compared to the PI condition, sea ice extent in the Eoi400 decreases by 5.4 25 Mkm 2 and 3.8 Mkm 2 respectively for the winter and summer season in the NH. In the Southern Hemisphere (SH), sea ice extent reduces by 8.8 Mkm 2 for the winter season and is nearly extinct during the summer. In comparison with PlioMIP1, NH sea ice cover in Eoi400 reduces by 2.1 Mkm 2 and 0.8 Mkm 2 respectively for cold and warm season but there is no large difference in SH between these two experiments. The largely decreased sea ice extent can amplify the warming in the high latitudes, through its role as an insulation between the ocean and the atmosphere as well as positive albedo temperature 30 feedback (Howell et al.,2014;Zheng et al., 2019). Reconstructed data in the Arctic Basin suggest the presence of seasonal rather than perennial sea ice in the Pliocene Arctic (Polyak et a., 2010;Moran et al.,2006), indicating a less or diminished summer sea ice cover. However, our IPSL model as well as half of participating models in PlioMIP1 cannot predict sea ice-free conditions during the summer season (Howell et al., 2016). Reasons for that are discussed in Howell et al (2016), which demonstrate the unreasonable sea ice albedo parameterization for the warmer condition.

Pliocene runs with IPSL-CM5A2
5 4.2.1 Results from the core experiment Eoi400_v2 Figure 8 shows the anomalies of global mean annual near SAT (2-meter temperature), precipitation rate and SST between Eoi400_v2 and pre-industrial control with the CM5A2model. The global mean SAT in Eoi400_v2 is 15.3C, which is 2.2C warmer than pre-industrial conditions with greater warming in high latitudes. It should be noted that the absolute SAT in 10 Eoi400_v2 is greater than that obtained in Eoi400, while the SAT anomaly in Eoi400_v2 is lower than that in Eoi400. This is due to the cold bias correction in the new version of the IPSL model. IPSL-CM5A2 simulates a pre-industrial that is warmer (Sepulchre et al., in prep) than the IPSL-CM5A pre-industrial by 1.1C. The global mean annual precipitation rate increases by 0.13 mm/d in Eoi400_v2 compared to PI, which is comparable to the results obtained with IPSL-CM5A. In Eoi400_v2, the changes in the ocean conditions relative to the pre-industrial control are similar to changes seen with Eoi400. The global 15 mean annual SST in Eoi400_v2 is 0.7C warmer than Eoi400, the AMOC strength ( Figure S4) in Eoi400_v2 is 17.9 Sv which is 2.2 Sv larger than Eoi400, while AMOC anomaly is about 4.7 Sv relative to its pre-industrial level of 13.2 Sv. The magnitude of this anomaly is close to the result obtained withIPSL-CM5A, indicating a coherent response of the AMOC to the same changes of boundary conditions. The sea ice cover is also largely decreased due to the warming in high latitudes   Figure 9 shows the relative contribution of various boundary conditions (CO2 (a), orography (b) and land ice (c)) to the warming during MIS KM5c as obtained using the linear decomposition method. Among these forcings, the increase in pCO2 25 by 120 ppmv (from 280 to 400 ppmv) plays the most important role in both the annual (+1.4C) and seasonal SAT (+1.38C and +1.48C respectively during the summer and winter). The changes to orography in PlioMIP2 also exert an important influence on the annual mean warming (+0.51C), especially in the north Atlantic and Barents Sea regions. However, changes to the orography decrease the temperature in the NH mid-to-high latitude inland regions, which may result from changes in North Pacific circulation. Seasonally, the orography changes contribute more to the warming in summer 30 (+0.65C) than that in winter (+0.38C). The impact of smaller ice sheets is largely restricted to the high latitude regions and is less important than the other two forcing factors in North polar region but plays the key role in the warming of South polar region. The mean annual warming resulting from the smaller ice sheets is about 0.25C which is close to contribution in both summer and winter seasons, indicating that the ice sheet contribution is seasonally invariant. The residual impact besides the pCO2, orography and land ice forcings is relatively small and negligible when making the linear decomposition of the forcing factors. These results are in agreements with those of Chandan and Peltier (2018)

Greenland ice sheet instability under MIS KM5c warmth
To understand the extent to which the Greenland ice sheet (GrIS) could be sustained under the warmth of MPWP, we impose the modern GrIS into the Pliocene simulation (Eo400_v2). In comparison with PI control (Figure 10a), the mean annual surface mass balance (SMB) in Greenland in Eo400_v2 (Figure 10b) is strongly negative around the coastal regions, 10 indicating vulnerable conditions along costal ice sheet and ice shelves. This negative SMB condition largely results from the increased summer temperature which leads to enhanced ablation in these regions ( Figure S5). The mean annual SMB condition in Eo400_v2 is similar to that in modern condition (E400_v2, Figure10c). However, the warmer condition in Eo400_v2 bring more precipitation in the South and Northwest Greenland, leading to enhanced accumulation ( Figure S5), thus we observe increased SMB in these areas as compared to the PI control condition. In E400_v2, we also have increased 15 SMB in these regions but much weaker than that in Eo400_v2, due to the different paleogeography settings as discussed earlier. Although these snapshot results cannot quantify the impact of the warm climate on the modern GrIS extent, which needs another series of climate model-ice sheet model experiment, the results we get here can also herald the vulnerability of GrIS under such warm climate condition.  Figure 11 depicts the anomalies of global mean annual SAT, precipitation rate and SST in Eoi450_v2 and Eoi350_v2 compared to the core experiment Eoi400_v2. By increasing pCO2 of 50ppmv in Eoi450_v2, global climate gets slightly warmer (+0.48C) and the warming in high latitudes is larger (+0.7C). However, when lowering pCO2 by 50ppmv in 25 Eoi350_v2, the change of climate is more important than that in Eoi450_v2, since we observe a global cooling of 0.71C and cooling of 1.29C over NH high latitudes. This asymmetric pattern in temperature response to change of pCO2 largely results from changes to surface albedo associated with snow cover (not shown here). In Eoi450_v2, the mean annual snowfall decreases by 6% between 40N and 80N when comparing to Eoi400_v2, while Eoi350_v2 shows an increase by 30% (not shown). The asymmetric pattern between Eoi450_v2 and Eoi350_v2 is also found in the changes of precipitation 30 rates: global climate gets slightly moister with an increased global precipitation rate by 0.02mm/d (+15%) in Eoi450_v2, while in Eoi350_v2, the global precipitation rate reduces by 0.04mm/d (-31%) and this reduction is more important in the tropical regions. Thus, our results can also show that the response of IPSL coupled model to changing pCO2 from 350 to 400 ppm is larger than from 400 to 450 ppm.

pCO2 uncertainties in MIS KM5c warmth
However, in the ocean, the increase or decrease in SSTs resulting from increasing or lowering pCO2 by 50ppmv is nearly the same magnitude. The AMOC strengths are also similar between Eoi450_v2 (17.4 Sv) and Eoi350_v2 (17.6 Sv) ( Figure S4).
Nevertheless, the changes of sea ice cover in these two experiments are unlike from each other (Figure 7). As in Eoi450_v2, 5 the sea ice covers decrease slightly relative to Eoi400_v2 for both hemispheres (decreased by 0.2-0.5 Mkm 2 during cold season and decreased by 0.01-0.2 Mkm 2 during warm season). Whereas in Eoi350_v2, the sea ice cover expands for both hemispheres, especially during the warm season in the NH (+1.7 Mkm 2 ).

Model-Data Comparison
10 Figure 12 shows the simulated mean annual SST anomalies (relative to PI experiments) of both core experiments (Eoi400， Eoi400_v2), together with the reconstructed SST (3.20 -3.21Ma, Foley and Dowsett 2019) anomalies relative to near preindustrial data (1870( -1900( , Rayner et al.,2003. The simulated SST anomalies in both core experiments are generally in phase with the reconstructed data. Some extremely warm sites are in disagreement with model results (e.g., Drilling sites in 15 North Greenland Sea, in Mediterranean and Benguela current region). Overall, the simulated MIS KM5c SSTs generally underestimate the warming that is inferred from proxies, especially for the sites showing warming higher than 4C ( Figure   12b). Amongst the three experiments (PlioMIP1, Eoi400, Eoi400_v2), both Eoi400 and Eoi400_v2 show increased warming in the mid-to-high latitudes as compared to the PlioMIP1 result. However, despite the increased warming exhibited by our PlioMIP2 simulation, there is still an obvious disagreement between model and proxy for which model performance is partly 20 to blame. However, the interpretation of the reconstructed data can also affect the data-model comparison. Conventionally, SSTs are reconstructed from 37 ′ paleothermometry assuming they represent annual mean values, whereas it has been shown that they can represent seasonal temperatures, for example representing the warmest summer month in the North Atlantic (NATL) (Leduc et al., 2017) and in the Benguela (Leduc et al., 2014). If we compare the reconstructed SST anomalies with modelled SST anomalies for the warmest summer month rather than the mean annual anomalies for the NATL and the 25 Benguela region ( Figure S6), the discrepancies between model and data is reduced. To well understand the discord, more studies are needed with regards to data interpretation as well as the multi-model comparison.

30
In this paper, we describe the results of modelling the warm interglacial MIS KM5c (3.205 Ma), located in MPWP interval of (3.0-3.3 Ma), while driving the model with the new PRISM4 boundary conditions . Two versions of the core Pliocene experiments namely Eoi400 and Eoi400_v2 are conducted on two versions of the IPSL coupled model: Eoi450_v2,Eoi350_v2,Eo400_v2) are also conducted with IPSL-CM5A2 to study the relative contribution of various forcing factor towards the warming of climate. The new PRISM4 boundary conditions produce an enhanced global warming in MIS KM5c, especially over the mid-to-high latitudes oceans compared to the results from PlioMIP phase 1. The enhanced warming can be largely attributed to changes to high latitude seaways which strengthen AMOC and transport more heat to high latitudes, and to the reduction in the spatial extent 5 of ice sheets and sea ice which decreases the outgoing shortwave radiation. The warming in MIS KM5c simulated with either of our models is weaker than those found in other studies (e.g. Kamae et al., 2016;Chandan and Peltier, 2017). In both our two core experiments, AMOC strength increases remarkably (+4.7 Sv) in comparison to PI controls due to the closure of the Bering Strait and the North Canadian Archipelago regions. This result agrees with other studies (Kamae et al., 2016;Hu et al., 2015), but the extent of the increase of the AMOC highly depends upon the processes included in the ocean models.     Figure 1: Anomalies of the PlioMIP2 topography relative to PI control (upper) and PlioMIP 1 (lower).

5
Data availability: Climatological averages of each simulation in NetCDF format will be uploaded to the PlioMIP2 data repository soon (sftp://see-gw-01.leeds.ac.uk). Specific data requests should be sent to the lead author (ning.tan@mails.iggcas.ac.cn). All PlioMIP2 boundary conditions are available on the USGS PlioMIP2 web page (http: //geology.er.usgs.gov/egpsc/prism/7_pliomip2/).

10
Author Contributions: N. T., G. R. and C. C. designed the study. N. T. conducted the model set-up, spin-up and major data analysis and wrote the manuscript. Y. S., C. C., C. D. and Z. G. contributed to discuss the data analysis and the structure of this work. P. S. provided the IPSL-CM5A2 information and its related control run simulation. All co-authors helped to improve this manuscript. Correspondence and requests for materials should be addressed to N. T.

15
Competing interests: The authors declare that they have no conflict of interest.