PlioMIP2 simulations using the MIROC4m climate model

The second phase of the Pliocene Model Intercomparison Project (PlioMIP2) has attracted many climate modelling groups in its continuing efforts to better understand the climate of the mid-Piacenzian warm period (mPWP) when atmospheric CO2 was last closest to present day levels. Like the first phase, PlioMIP1, it is an internationally coordinated initiative that allows for a systematic comparison of various models in a similar manner to PMIP. Model intercomparison and model-data comparison now focus specifically on the interglacial at marine isotope stage KM5c (3.205Ma) and experimental design is not 10 only based on new boundary conditions but includes various sensitivity experiments. In this study, we present results from long-term model integrations using the MIROC4m atmosphere-ocean coupled general circulation model, developed at the institutes CCSR/NIES/FRCGC in Japan. The core experiment, with CO2 levels set to 400ppm, shows a warming of 3.1°C compared to the Pre-Industrial, with two-thirds of the warming being contributed by the increase in CO2. Although this level of warming is less than that in the equivalent PlioMIP1 experiment, there is a slightly better agreement with proxy sea surface 15 temperature (SST) data at PRISM3 locations, especially in the northern North Atlantic where there were large model-data discrepancies in PlioMIP1. Similar changes in precipitation and sea ice are seen and the Arctic remains ice-free in the summer. However, unlike PlioMIP1, the Atlantic Meridional Overturning Circulation (AMOC) is now stronger than that of the PreIndustrial, even though increasing CO2 tends to weaken it. This stronger AMOC is a consequence of a closed Bering Strait in the PlioMIP2 paleogeography. Also, when present day boundary conditions are replaced by those of the Pliocene, the 20 dependency of the AMOC strength on CO2 is significantly weakened. Sensitivity tests show that lower values of CO2 give a global SST which is overall more consistent with the PRISM3 SST field presented in PlioMIP1. Inclusion of dynamical vegetation and the effects of all realistic orbital configurations should be considered in future experiments using MIROC4m for the mPWP.

The model consists of an atmosphere-land-river component and a sea ice-ocean component, with the air-sea exchange of momentum, heat and water occurring at the air-sea ice interface. At ice-free grid cells, exchange still occurs via the sea ice subcomponent but the flux to the ocean remains unaffected by the sea ice. Below is a brief description and readers should refer to K-1 model developers (2004) and the references contained within for further details. 100 The atmospheric component is identical to the AGCM described in Numaguti et al. (1997), namely, CCSR/NIES/FRCGC AGCM5.7b. The horizontal resolution is set to T42, which corresponds to a grid size of approximately 2.8° longitude and latitude and the number of levels is set to 20 in the σ coordinate system where pressure at all heights is scaled with the surface pressure. USGS GTOPO30 is used to generate the surface elevation. The level 2 scheme of the turbulence closure model by Mellor and Yamada (1982) is used for sub-grid vertical fluxes of prognostic variables. A radiative 105 transfer scheme (Nakajima et al., 2000) based on the two-stream discrete ordinate and k-distribution methods is employed.
Other physical parameterizations include a prognostic Arakawa-Schubert cumulus scheme and a prognostic cloud water scheme for large-scale condensation (Le Treut and Li, 1991). Optical parameters for water cloud, ice cloud and five aerosol types -soil dust, black carbon, organic carbon, sulfate and sea salt -are included. Classification of aerosols is based on Spectral Radiation-Transport Model for Aerosol Species (SPRINTARS) (Takemura et al., 2000). Indirect effects of aerosols are 110 considered for condensation in stratus clouds. Monthly aerosol mass and particle number concentration used in radiative processes are prescribed off-line by SPRINTARS.
The land-surface model used is Minimal Advanced Treatments of Surface Interaction and Runoff (MATSIRO) (Takata et al, 2003), whose horizontal resolution is the same as that of the atmospheric component. Here, water and heat exchange between the land surface and atmosphere is computed. Within the same model, runoff on the land is also calculated and passed 115 over to a river routing model which transports the runoff water to the ocean model at river mouths. Prognostic variables include canopy water content, canopy temperature and soil moisture. Land-cover classification is derived from USGS GLCC (Global Land Cover Characterization). See Chan et al. (2011) for the present-day vegetation distribution.
The ocean component is basically version 3.4 of the CCSR Ocean Component Model (COCO) (Hasumi, 2000). The horizontal grid has 256x192 points so that each grid point is spaced equally at 1.40625° in the longitudinal direction. In the 120 latitudinal direction, resolution is highest in the tropics (0.56°) and lowest at the polar regions (1.4°). The are 43 vertical levels, including 8 σ levels near the sea surface. Here, σ denotes a normalised geopotential height, with a value of 1 at the free surface and 0 at a fixed depth above which the σ coordinate system is applied. The Bering Strait throughflow is fully represented but the Hudson Bay and the Mediterranean Sea are treated as isolated lakes with heat and salinity exchanged with the open seas by a 2-way linear damping. A simple, vertical adjustment is applied, whereby unstable water columns are homogenised 125 instantaneously. Vertical mixing of sea tracers and momentum use viscosity and diffusion coefficients calculated by Noh and Kim (1999). While there are no changes to the model as used in PlioMIP1, it should be noted that a larger Gent-McWilliams coefficient has been used in other more recent published work using MIROC4m (Obase andAbe-Ouchi, 2019, Sherriff-Tadano andAbe-Ouchi, 2020). This coefficient which refers to the horizontal diffusion of the isopycnal layer thickness is set to 300m 2 /s in the present study. 130 years are shown on the extreme left in Figure 2. For this study, these experiments are continued for another 1000 model years with the greenhouse gases changed to levels specified in PlioMIP2; the time series for these 1000 years are plotted on the righthand side of Figure 2.

Core Pliocene (Eoi 400 ) and related experiments with different CO2 levels (Eoi 280 , Eoi 350 and Eoi 450 )
With greenhouse gas levels initially set to their previous values, the core experiment, Eoi 400 , and the same experiment 175 with Pre-Industrial CO2 levels, Eoi 280 , start from E 280 and the model is integrated for 3000 and 1500 years, respectively. At the end of 3000 years, Eoi 350 and Eoi 450 branch off Eoi 400 ; the model for these two branches is integrated for 2000 years. Then as before, with greenhouse gas levels modified to PlioMIP2 values for these 4 experiments, the model is integrated for a further 1000 years.
The full, enhanced boundary conditions from Haywood et al. (2016) are employed, in particular, the Pliocene minus 180 Modern topography anomaly, as shown in Figure 1, is applied to the existing MIROC4m land elevation. The largest reductions in surface elevation can be found in Greenland and parts of Antarctica, whereas the largest increases are located over North America and in the interior of Antarctica. Note that the Pliocene surface elevations used in PlioMIP1 and PlioMIP2 differ from one another (Supplementary Figure 1a). In addition to a lower surface elevation in the northern half of Greenland for PlioMIP2 due to a smaller ice sheet, higher surface elevation is found in southern Greenland, much of North America, except 185 over the southern Rocky Mountains, northern Eurasia and the northern and central Andes Mountains. The land-sea mask is modified according to PlioMIP2 paleogeography. Modifications which did not exist in PlioMIP1 include the closure of the Bering Strait and the Canadian Arctic Archipelago Straits (CAAS), and the introduction of Pliocene lakes across Africa.
For consistency we apply the same vegetation distribution as that in the PlioMIP1 study by Chan et al. (2011), derived from Salzmann et al. (2008). The vegetation is extended in several regions where the land mask differs from that of PlioMIP1,190 for example, over an expanded Indonesia, over Beringia, resulting from a closed Bering Strait, and across parts of Greenland where the ice sheets are now specified to be smaller than that of PlioMIP1. In addition, several lakes have now been included on the African continent. The bathymetry is modified by the usual anomaly method, although a separate experiment was carried out without such modifications, resulting in no noticeable differences in the climate. Soil types are changed according to their texture. 195

Pliocene with PlioMIP1 boundary conditions (Eplio1)
For the lone experiment with PlioMIP1 boundary conditions and CO2 level set to 405ppm, the model is initialised with Pre-Industrial conditions and integrated for 3000 years before the other greenhouse gas levels are changed to PlioMIP2 values for a further 1000 years. As in our original PlioMIP1 experiment (Chan et al., 2011), there are no modifications made to the bathymetry, soil types or lakes. 200

Deleted:
The overall temperature increase in Eoi 400 is not as large as that in the PlioMIP1 experiment (Table 3), as in the case with the models HadCM3 (Bragg et al., 2012;Hunter et al., 2019), NorESM-L (Li et al., 2020) and COSMOS (Samakinwa et al., 2020), with the first model and MIROC4m both giving a PlioMIP2-PlioMIP1 global SAT difference of 0.4°C. Other 230 models, as noted by Li et al (2020), show the opposite result, ie. larger warmer in PlioMIP2. In our results, overall, the PlioMIP1 temperatures lie between those of PlioMIP2 with CO2 set to 400 and 450ppm. PlioMIP2 temperature increases in most regions of the northern high latitudes and the tropics are smaller and this may be at least partly a result of the increased SAT follows closely that of the elevation difference ( Supplementary Figures 1a and 1b). Conversely, SAT in Eoi 400 is much higher than that in Eplio1 across West Antarctica, northern Greenland, the southern Rocky Mountains and the Near East, the 250 last three of which also exhibit the same SAT differences in NorESM-L ( Figure 9a of Li et al., 2020).
The globally averaged SAT from PlioMIP has been used to estimate the Earth system sensitivity (ESS), which, unlike the climate sensitivity (CS), takes into account feedbacks operating over longer timescales (Lunt et al., 2009). From E 280 and E 560 , CS is estimated to be 3.9, and using Eplio1 and Eoi 400 , ESS is estimated to be 6.6 and 6.0, respectively, and the ESS/CS ratio 1.7 and 1.5, respectively. All these values compare quite well with the PlioMIP2 multi-model mean values (Haywood et 255 al., 2020). Figure 4 shows the zonal mean surface air temperature increase from E 280 . When only CO2 is changed, a small but gradual and near-linear increase in temperature anomaly starting from the southern mid-latitudes to the northern mid-latitudes is evident (see the red and purple lines between 45°S and 45°N). The larger anomaly in the northern half is simply a consequence of the larger land area there. Increase in CO2 is also accompanied by polar amplification of the warming. For 260 all other experiments, the increase in temperature anomaly from the tropics to the northern mid-latitudes is more pronounced.
In the northern polar region, peak warming is seen at around 75°N for all experiments. However, in the southern polar region, peak warming shifts from 65°S to 75°S with the inclusion of Pliocene land elevation and reduced ice sheets. For the Pliocene experiments, Eoi 350 , Eoi 400 and Eoi 450 , the increase in zonal temperature anomaly with an increase of 50ppm in CO2 is limited to less than 1°C at low and mid-latitudes, and at most 1.5°C at the polar regions. This increase is relatively small in comparison 265 to results from models like IPSL-CM5A2 which shows a fairly small uniform change from Eoi350 to Eoi400, except at the northern high latitudes where there is a sharp change of up to 2.5°C ( Figure 11c of Tan et al., 2020).
The seasonal SAT anomalies for Eoi 400 are shown in Figure 5. There is little seasonal change over north Africa and much of the oceans. Throughout the year, temperature increases over Greenland remain large. However, there are distinct seasonal changes at high latitudes elsewhere, for example, the small temperature increase in the Arctic region during the 270 summer, followed immediately by the extremely large temperature increase during the autumn. In the summer, there is very little sea ice in the Arctic in Eoi 400 and so the ocean warms up more from incoming insolation until the SST reaches a maximum.
As the summer ends, heat from the ocean is released back into the atmosphere. Since there is very little sea ice, more heat can be released, explaining the higher SAT in the Arctic during September to November. This was also seen in the study by Zheng et al (2019). The Hudson Bay shows a large temperature reduction during winter because it has been replaced by land which 275 cannot stabilise the surface air temperature as much as water can (Hunter et al., 2019). Conversely, by the same reasoning, a large temperature increase is seen over the Hudson Bay during the summer. Another region where the temperature reduces is the zonal strip in Africa at latitude 15°N during the summer. This is similar to what was seen in the early work of Chandler et al. (1996) (Table 3), slightly lower than that for the PlioMIP1 experiment (19.2°C).
The spatial changes in the PlioMIP1 experiment are similar to those in Eoi 400 , with the global changes in precipitation also being similar (Table 3). However, there are small differences in the amount of precipitation change (Supplementary Figure 1d). In many areas, precipitation increases and decreasing surface elevation and vice versa, for example, over Gr eenland and Antarctica. Eoi 400 precipitation is lower in the Indian Ocean, west of Australia, while it is higher in northern Africa and 325 the tropical Atlantic. This may be related to differences in the ITCZ.

Sea ice and ocean mixed layer depth
The total sea ice area in the polar regions is depicted in Figure 8. In the Arctic, during March when the sea ice area is at its greatest, there is a gradual decrease in area as CO2 is increased, whether Pre-Industrial or Pliocene boundary conditions are used. The sea ice area in Eplio1 is actually larger than any of the other Pliocene experiments because the land area around 330 the Arctic is smaller in Eplio1, as can be seen later in Figure 9 the Bering Strait is open, the Hudson Bay is still set as open water and the Labrador Sea is larger and connected to the Arctic Ocean. During September, when the sea ice area is at its smallest, a similar trend is observed. The Arctic Ocean is ice-free in Eplio1, Eoi 400 and Eoi 450 . In the Antarctic, during September, the decrease in sea ice area as CO2 increases is much more drastic. The Antarctic sea ice area when the CO2 level is doubled is also much lower than that of any of the Pliocene experiments. Unlike the Arctic, the sea ice area in Eplio1 and 335 Eoi 400 are similar as the Antarctic coastlines in the two cases do not differ by much. The same behaviour is observed in March, although unlike in the Arctic, the Antarctic is never ice-free. However, in both polar regions, the boundary of sea ice extent in Eoi 400 and Eplio1 are similar (Figures 9 and 10).
The sea ice extent in the Arctic and the mixed layer depth (MLD) in the surrounding regions during March are shown in Figure 9. The definition of the MLD follows that of Oka et al. (2006), i.e. the depth at which σθ, the potential density 340 anomaly, differs from the surface value by 0.1. The sea ice extent here is largely unaffected by the Pliocene boundary conditions alone and only recedes near the north-west Pacific coastline and in the Barents Sea as CO2 is increased. In E 280 , the MLD is large everywhere south of the sea ice extent in the North Atlantic, including the Labrador Sea and the Norwegian Sea, and to a lesser extent west of the British Isles. The Labrador Sea responds with a larger MLD, spread over a greater area directly south of Greenland to Pliocene boundary conditions and increasing CO2. when applied separately. However, when 345 both Pliocene boundary conditions and changes in CO2 are applied, as in Figures 9(e-h), it can be seen that CO2 has an opposing effect on the MLD, suggesting that deepwater formation in the Labrador Sea is greater in Eoi 280 than in Eoi 450 . In the Eplio1 experiment, lower SST in the Labrador Sea, as discussed previously, also contributes to greater deepwater formation so that the MLD is more comparable to that of Eoi 280 . In the Norwegian Sea, the areal extent of deep MLD is similar in all experiments, but only E 280 shows the maximum MLD of about 3000m, located at the northernmost ice-free region, at latitude 80°N. In all 350 experiments with higher CO2, the sea ice in the northeast of the Norwegian Sea retreats and exposes the Barents Sea, but the extent of the deepwater formation region in the Norwegian Sea remains unchanged. In the region to the west of the British Isles, regardless of CO2 levels, no change is seen when PlioMIP2 boundary conditions are applied. On the hand, there is little deepwater formation there in Eplio1 or when CO2 level is 400ppm or higher with present day boundary conditions.

Moved up [1]:
As with SAT, the globally averaged precipitation 355 in Eoi 400 is larger than that in E 280 , with CO2 contributing about twothirds of the total increase.

Deleted:
The MLD, on the other hand, also responds to changes in CO2 alone. In E 280 , the MLD is large west of the British Isles and everywhere south of the sea ice in the Atlantic. As CO2 is increased 360 (E 400 and E 560 ), the MLD decreases in the former but increases over a larger area south of Greenland. With Pliocene boundary conditions, a large MLD is also exhibited in the same area south of Greenland, even in Eoi 280 . However, the MLD west of the British Isles remains large while a large MLD is now seen south of Alaska. These results 365 suggest an increase in downwelling south of Greenland The sea ice extent and MLD in the Antarctic during September are shown in Figure 10. With present day boundary conditions (E 280 , E 400 and E 560 ), the MLD is extremely large in the Atlantic sector of the Southern Ocean, even below parts of the sea ice in E 280 , suggesting the formation of dense water due to brine rejection. In the Pliocene experiments, this large MLD is absent, while in the other region where the MLD is large with present day boundary conditions, i.e. the eastern South Pacific, 370 it is reduced.

Atlantic meridional overturning circulation
The Atlantic meridional overturning circulation (AMOC) is shown in Figure 11 and the AMOC index, defined as the maximum streamfunction value, in the last column of Table 3. Increasing the CO2 level alone has a tendency to weaken the AMOC. As the overturning cell becomes shallower, the underlying Antarctic Bottom Water extends further northward. In 375 addition, the anti-clockwise overturning cell north of 65°N strengthens, contributing to increased convection and deepwater formation in the Labrador Sea, as indicated by the MLD in Figure 9. A comparison of Eoi 280 with E 280 shows that similar changes occur when only the Pliocene boundary conditions are applied, except that the AMOC index increases by nearly 0.7Sv, despite shoaling of the AMOC. Thus, increasing CO2 and applying Pliocene boundary conditions have an opposite effect on the AMOC index, but up to 450ppm CO2, the indices for all the Pliocene experiments are still greater than that for 380 E 280 . Note that the degree of weakening in the AMOC as CO2 increases seem to be highly dependent on the boundary conditions. With present day boundary conditions, from E 280 to E 400 , the AMOC index decreases by 0.8Sv, whereas in the corresponding Pliocene experiments, Eoi 280 and Eoi 400 , there is only a decrease of approximately 0.2Sv (Table 3).
A comparison of Eoi 400 and Eplio1 shows that while the AMOC cell extends to similar depths and the circulation in the other two cells change little, the AMOC index with PlioMIP2 boundary conditions is larger than that with PlioMIP1, 20.0Sv 385 versus 17.8Sv. While we have not performed specific sensitivity experiments to see what difference between the PlioMIP1 and PlioMIP2 boundary conditions is exactly responsible for this difference in the AMOC index, we did perform some (not shown) with a Pre-Industrial background climate and looked at the effects of closing the Bering Strait and the CAAS closed.
We find that, as in Pliocene studies by Otto-Bliesner et al. (2017)  The transport peaks at around 0.7PW, near latitude 15°N. Figure 12(b) shows the difference between the same heat transport Pliocene experiments, there does not appear to be any general trend as the CO2 level is increased. For example, Eoi 450 has a 405 significantly lower heat transport compared to the others south of the equator, whereas at the northern low to mid -latitudes, Eoi 450 values are rather close to those of Eoi 400 and it is the Eoi 280 values that stray from the rest. It can also be seen that there is a marked difference between the heat transport in Eplio1 and those using PlioMIP2 boundary conditions. At all latitudes south of 30°N, heat transport in Eplio1 is about 0.1PW lower than that of E 280 . Irrespective of CO2 levels, this larger difference can mostly be attributed to the differences between the PlioMIP1 and PlioMIP2 boundary conditions, as can the difference 410 between the AMOC index of Eplio1 and those of the PlioMIP2 Pliocene experiments. However, as noted in the latter experiments, the northward heat transport decreases even though the AMOC index increases, albeit to a relatively small degree.
While these two properties are commonly thought of as being positively correlated to each other, Sévellec and Fedorov (2016) show that this is not always the case. Using NorESM-L, Eoi 400 exhibits the same features, ie. reduced meridional heat transport in the Atlantic Ocean but a stronger AMOC (Li et al., 2020), whereas its sister model, NorESM1-F, as well as many other 415 models, show both increased heat transport and a stronger AMOC (Kamae et al., 2016;Chandan and Peltier, 2017;Tan et al., 2020). Figure 13 shows the total meridional heat transport in a similar way, but for the whole climate system, combining atmosphere and ocean. The absolute values, in the Figure 13(a), confirm that, in general, heat is transported northwards in the northern hemisphere and southwards in the southern hemisphere. In some studies, the total heat transport in both hemispheres 420 in Eoi 400 either decreases (Chandan and Peltier, 2017) or increases (Feng et al., 2020). In contrast, for all Pliocene experiments in this study, northward heat transport is reduced in the northern hemisphere, while southward heat transport is increased in the southern hemisphere ( Figure 13b). This appears to be more akin to the results from IPSL-CM5A, as seen in Figure 6 The opposite trend is seen in the southern hemisphere as southward heat transport increases with CO2. Eoi 400 represents a halfway mark whereby the bold green line in Figure 13(b) is roughly symmetric across the equator. Secondly, Eplio1 aligns very closely with Eoi 400 in the southern hemisphere, whereas in the northern hemisphere, Eplio1 sets itself apart from the other Pliocene experiments and its heat transport reduces much less. Finally, we note that the largest anomalies occur at low 430 latitudes, a characteristic not evident in the studies mentioned above which tend to show greatest anomalies at the mid-latitudes where the absolute values are at their peak.

Comparison with surface air temperature proxy data
In Figure 14, the annual mean SAT from the Pliocene experiments and E 280 are compared with SAT estimates from vegetation reconstructions at marine sites near land, as compiled by Salzmann et al. (2013). The red symbols refer to the 435 Deleted: For Deleted: .

Deleted: proxy data
Deleted: certain locations Barents Sea are lower in Eoi400 compared to Eplio1, the Eoi 400 values at the PRISM3 sites are actually higher, and thus agree better with the proxy data, as shown by the yellow triangles. 475 We also include a similar comparison between model results and the newer PRISM4 proxy SST data sets (Foley and Dowsett, 2019) in Figure 17. The degree of warming in Eoi 400 is much less than that suggested by PRISM4 proxy data in the Atlantic sites, in particular, at northern high latitudes again, and also near southern Africa. Slightly higher warming in Eoi 400 is generally seen in the other sites, mostly located at low latitudes, but especially in the Caribbean Sea, which is not evident in the PRISM3 comparison ( Figure 16a). The PRISM4 data used here refer to the broader 30ka interval, but alternative data for 480 a 10ka interval give the same conclusions. Qualitatively speaking, at least, these results are similar to those obtained from the multi-model mean in Figure 8 Table 4. An important caveat here is that this global SST field is reconstructed 485 from data at a finite number of sites for February and August using interpolation and extrapolation, and that there are regions where data are sparse (Dowsett et al., 1999). Comparing Eplio1 and Eoi 400 , we see that the former matches the proxy data better at northern high latitudes, and even better at southern high latitudes (a difference of only 0.04°C). On the other hand, the latter shows a smaller discrepancy at the tropics and low latitudes (30°S-30°N) where the larger surface area means that, globally speaking, Eoi 400 gives a better fit (a difference of 0.76°C). Next, a comparison of Eoi 350 , Eoi 400 and Eoi 450 shows that, 490 while there is a trend in the model-data difference as CO2 is reduced, there is no particular level at which the difference is small at the three latitudinal ranges in Table 4. Since the global difference is determined more by the low latitudes, Eoi 350 gives the best global fit (a difference of 0.14°C). It is worth noting that not only does Eoi 280 give the best fit at low latitudes, but the discrepancy is smaller than that for Eoi 350 at northern high latitudes, despite the increasing trend in discrepancy from Eoi 450 to Eoi 350 . 495 For reference, the discrepancies between the model and proxy SST anomalies at the PRISM3 sites for different CO2 levels are shown in Supplementary Figure 2. Similarly, comparisons using PRISM4 data for different CO2 levels are shown in Supplementary Figure 3. There is no spatial reconstruction to accompany the data from PRISM4 sites, and so comparisons are based solely on these sites. Lower values of CO2 tend to give better agreement at low latitudes, except to the west of western Africa where Eoi 450 shows good agreement. However, even in Eoi 450 , Pliocene warming in the Greenland and 500 Norwegian Seas, the Northern North Atlantic Ocean and the coastal region near southern Africa is much lower than that suggested by PRISM4 data (blue triangles)..

More recent studies by Tierney et al (2019) have highlighted uncertainties in tropical SST estimates and showed good
agreement between their reduced space reconstruction and a Pliocene simulation run with CESM1. We carry out SST comparison ( Figure 18) between our Pliocene experimental results and their reconstruction, limited to the Pacific Ocean, using 505 alkenone proxy data and a probabilistic approach. In the Tropical Pacific, this proxy reconstruction exhibits higher temperature anomalies than those from PRISM3, leading to optimal agreement with Eoi400 (Figure 18c), which was warmer than PRISM3  Deleted: Figure 1 data by 1-3°C in that same region (not shown). Although the proxy reconstruction shows enhanced warming in the eastern equatorial Pacific and in the northern subtropical Pacific, this warming is underestimated by 1-3°C, compared to PRISM3. This is most evident in the northwest Pacific where the reconstruction has more limited data, and warming is weaker than that 515 in all our Pliocene experiments.

Summary and conclusions
In the present study, we have shown some basic results from the core PlioMIP2 experiment using the MIROC4m AOGCM, compared them to results from both the PlioMIP1 experiment and some other Tier 1 and Tier 2 PlioMIP2 sensitivity experiments. Additionally, we have evaluated the consistency between these experimental results and some temperature proxy 520 data from marine and terrestrial sources.
For the core experiment, PlioMIP2 boundary conditions produce a global temperature increase smaller than that with PlioMIP1 and it can be assumed that the CO2 level has little effect as it differs only slightly in the two phases of PlioMIP. The difference in the results from these two experiments is not uniform as greater warming is seen in PlioMIP2 in parts of the northern high latitudes and of the Southern Ocean. Moreover, PlioMIP2 SSTs actually reconcile better with proxy-derived 525 values at PRISM3 sites in the northern North Atlantic and Greenland Sea, albeit to a very small degree, although the large discord in the northern North Atlantic Ocean SSTs still remains. For SAT, both PlioMIP1 and PlioMIP2 values show fairly good agreement with proxy data from paleobotanical sites, although comparisons at the northern high latitude sites highlight the weaker polar amplification in model results. Northern polar amplification is slightly less in PlioMIP2, but nonetheless, zonal SAT increases are more than double that of the low latitudes. Our sensitivity experiments have only distinguished 530 between the two forcings from CO2 and Pliocene boundary conditions and we have not considered the effects from the ice sheets, orography and vegetation separately. CO2 accounts for two-thirds of the total surface air temperature and precipitation increase. Unlike PlioMIP1, the AMOC in PlioMIP2 is stronger compared to the Pre-Industrial for MIROC4m, which is in line with other model results published so far in PlioMIP2. The strengthening of the AMOC from PlioMIP1 to PlioMIP2 is tied to the closure of the Bering Strait. 535 We have also looked at the mid-Pliocene climate for a range of CO2 values. From these CO2 sensitivity experiments, we find that, not only does the AMOC strength decrease with increasing CO2, but that this dependency on CO2 is weaker when Pliocene boundary conditions are applied. While other expected trends are seen, such as the increase in global temperature and precipitation with CO2, of much importance is the comparison with proxy-derived data. Mismatches between Eoi 400 and proxy-derived SSTs at low and high latitudes are of the opposite sign, and data in both regions cannot be simultaneously 540 reconciled simply by changing the CO2 value. From a global perspective, a value below 400ppm leads to a better overall fit with PRISM3 data. However, the warming at many of the newer PRISM4 proxy data sites is too high to be reconciled with model data, even at higher CO2 levels. In the tropical Pacific, more recent reconstructions suggest that Eoi 400 does not overestimate the SST, as implied by PRISM3 data, thus reducing the global discrepancy. These results underscore firstly, the Deleted: We also include a similar comparison between model 545 results and the newer PRISM4 proxy SST data sets (Foley and Dowsett, 2019) in Supplementary Figure 2. The degree of warming in Eoi 400 is much less than that suggested by PRISM4 proxy data in the Atlantic sites, especially at northern high latitudes again, and also near southern Africa. Slightly higher warming in Eoi 400 is generally 550 seen in the other sites, mostly located at low latitudes. The PRISM4 data used here refer to the broader 30ka interval, but alternative data for a 10ka interval give the same conclusions. Qualitatively speaking, at least, these results are similar to those obtained from the multi-model mean in Figure 8 (Dolan et al., 2011;Feng et al., 2017). While using present-day orbital parameters for the KM5c interglacial peak appears valid, at least with fixed 570 vegetation (Hunter et al., 2019), an investigation of the mPWP as a whole necessitates more realistic orbital parameters, even when restricted to interglacial peaks (Prescott et al., 2018). This should be borne in mind when considering paleoclimate modelling experiments such as those for the mPWP.