Sensitivity of mid-Pliocene climate to changes in orbital forcing, and PlioMIP’s boundary conditions

. In this study, we compare results obtained from modelling the mid-Pliocene warm period using the Community Earth System Models (COSMOS, version: COSMOS-landveg r2413, 2009) with the two different modelling methodologies and sets of boundary conditions prescribed for the two phases of the Pliocene Model Intercomparison Project (PlioMIP), tagged PlioMIP1 and PlioMIP2. Boundary conditions, model forcing, and modelling methodology for the two phases of PlioMIP differ considerably in palaeogeography, in particular with regards to the state of ocean gateways, ice-masks, treatment of vegetation 5 and topography. Further differences between model setups as suggested for PlioMIP1 and PlioMIP2 consider updates to the concentration of trace gases: atmospheric carbon dioxide (CO 2 ), is speciﬁed as 405 and 400 parts per million by volume (ppmv) for PlioMIP1 and PlioMIP2, respectively. There are also minor differences in the concentrations of methane (CH 4 ) and nitrous oxide (N 2 O) due to changes in the protocol of the Paleoclimate Model Intercomparison Project (PMIP) from phase 3 to phase 4. Employing a single model across two phases of PlioMIP enables a better understanding of the impact that the various 10 differences in modelling methodology between PlioMIP1 and PlioMIP2 have on model output. Yet, a dedicated comparison of COSMOS model output of PlioMIP1 and PlioMIP2 is not in the curriculum of model analyses proposed in PlioMIP2. Here, we bridge the gap between our contributions to PlioMIP1 (Stepanek and Lohmann, 2012) and PlioMIP2 (Stepanek We highlight some of the effects that differences in the chosen mid-Pliocene model setup (PlioMIP2 vs. PlioMIP1) have on the climate state as derived with the COSMOS, as this information will be valuable in the framework of the model-model and 15 model-data-comparison within PlioMIP2. We evaluate the model sensitivity to improved mid-Pliocene boundary conditions using PlioMIP’s core mid-Pliocene experiments for PlioMIP1 and PlioMIP2, and present further simulations where we test model sensitivity to variations in palaeogeography, orbit and concentration of CO 2 . Firstly, we highlight major changes in boundary conditions from PlioMIP1 to PlioMIP2 and also the challenges recorded from the initial effort. The results derived from our simulations show that COSMOS simulates a mid-Pliocene climate state that is 20 0.29 K colder in PlioMIP2, if compared to PlioMIP1 (17.82 ◦ C in PlioMIP1, 17.53 ◦ C in PlioMIP2, values based on simulated surface skin temperature). On one hand, high-latitude warming, which is supported by proxy evidence of the mid-Pliocene, is underestimated in simulations of both PlioMIP1 and PlioMIP2. On the other hand, spatial variations in surface air temperature (SAT), sea surface temperature (SST) as well as the distribution of sea ice suggest improvement of simulated SAT and SST in PlioMIP2 if employing the updated palaeogeography. Our PlioMIP2 mid-Pliocene simulation produces warmer SSTs in the Arctic and North Atlantic Ocean than derived from the respective PlioMIP1 climate state. The difference in prescribed CO 2 accounts for 1.1 K of warming in the Arctic, leading to an ice-free summer in the PlioMIP1 simulation, and a quasi ice-free summer in PlioMIP2. Beyond the ofﬁcial set of PlioMIP2 simulations, we present further simulations and analyses that sample the phase space of potential alternative orbital forcings that have acted during the Pliocene and may have impacted on geological records. Employing orbital forcing, which differ from that proposed for PlioMIP2 (i.e. corresponding to Pre- 5 Industrial conditions) but falls into the Mid-Pliocene time period targeted in the PlioMIP, leads to pronounced annual and seasonal temperature variations, which are not directly retrievable from the marine and terrestrial reconstruction of the time-slice. 2 sensitivity study within the framework of PlioMIP1 and PlioMIP2 shows that a difference of 5 ppmv between both phases of PlioMIP causes appreciable changes in SAT over land and oceans. The increase in CO 2 produces warmer oceans in the PRISM4-based model setup, 20 except for parts of the North Atlantic and the Arctic as outlined above. As in the PRISM4 COSMOS model setup the effect of increased CO 2 is opposite to the pronounced mPWP warming simulated in the North Atlantic, that is found in PlioMIP2 simulation Eoi400 in comparison to PlioMIP1 simulation PlioM1, we conclude that the warmer North Atlantic simulated by us in PlioMIP2 is not inﬂuenced so much by changes in greenhouse forcing as it is by the collective contribution of all boundary conditions. As a matter of fact, the effect that relatively small changes in CO 2 have in the PlioMIP2 mPWP setup of COSMOS 25 on the temperature of the North Atlantic Ocean leads us to the inference that the gateway effect on North Atlantic Ocean warmth is modulated in our model by greenhouse forcing, and that our PlioMIP2 model setup does not provide a ﬁnal state-ment with regard to the strength of the gateway effect that may have impacted on the mPWP temperature signals interpreted from the geologic recorder. Other model conﬁgurations with altered greenhouse gas forcing may well produce a larger temperature anomaly than the COSMOS PlioMIP2 core simulation Eoi400. If we extend our focus beyond model performance 30 within PlioMIP2, then we ﬁnd that a PlioMIP2 COSMOS simulation with the higher volume mixing ratio of 405 ppmv of CO 2 is actually in better agreement with the PlioMIP1 simulation PlioM1 in parts of the North Atlantic Ocean (where that simulation suggests colder SST than the PlioMIP2 core simulation Eoi400). Consequently, our results show that a slightly higher CO 2 forcing provides larger disagreement between modelled and reconstructed SST, as both model setups of PlioMIP1 and PlioMIP2 are, in comparison with temperature data derived from proxy records, too cold.

study would be a promising approach to identify orbital-forced warm peaks, but is beyond the capabilities of the modelintercomparison. Most of the mis-matches recorded between mid-Pliocene model simulations and interpretation of data stored in geological archives have been attributed to the choice of orbital forcing prescribed for model simulations Haywood et al. (2013b); Prescott et al. (2014). The need for a more discrete time slice for simulating the mPWP towards an improved model-data comparison has been stated . Thus, the KM5c time-slice has been selected, partly on the 5 basis of a strong similarity of the orbit at that time to the modern orbital configuration, that is useful towards interpreting paleoenvironments in the context of future warming (Haywood et al., 2016) that is based on antropogenic activity and will obviously be set in a near-modern orbital configuration. While acknowledging the utility of the KM5c orbit for the scientific aims of PlioMIP2, here we go beyond the simulation of the KM5c and quantify the effect of an alternative orbital configuration.
We create model setups where the prescribed PlioMIP2 model setup for the COSMOS employs an orbital configuration that is 10 representative of MIS K1. The MIS K1 time-slice is one of the lightest isotope excursions found during the mPWP with the total global annual mean insolation being is approximately 0.5 W/m 2 higher than modern (Prescott et al., 2014). The effect of orbital forcing on the mPWP has been studied by Prescott et al. (2014), with emphasis on Surface Air Temperature (SAT).
The novelty in our approach is that we employ the updated PlioMIP2 model setup and test the impact of orbital configuration also on the ocean state, in particular on SST and sea ice. Furthermore, we put differences in orbital forcing into context with 15 varying concentrations of CO 2 .

Model description COSMOS
The coupled atmosphere-ocean model used in producing simulations for this study is the COSMOS, which was developed by the Max Planck Institute for Meteorology (MPI) in Hamburg, Germany. The COSMOS consists of four major components, namely the ECHAM5 atmosphere model (Roeckner et al., 2003), the MPI-OM ocean model (Marsland et al., 2003), and 20 the land-vegetation and carbon cycle model JSBACH (Raddatz et al., 2007). For a description of the coupled setup and an evaluation of its performance please refer to Jungclaus et al. (2006). The ocean biogeochemistry model HAMOCC, introduced by Maier- Reimer (1993), is as well a part of the COSMOS, but is not used in the production of PlioMIP simulations. The COSMOS has already proven to be a valuable tool for the study of paleoclimate, also beyond the Pliocene epoch. The various time-slices studied by means of the COSMOS include, but are not limited to, the Last Millennium (Jungclaus et al., 2010), 25 warm climates of the Miocene (e.g., Knorr et al., 2011), the mid-Pliocene (Stepanek and Lohmann, 2012), as well as glacial (e.g., Gong et al., 2013;Kageyema et al., 2013;Zhang et al., 2013) and interglacial climates (e.g., Pfeiffer et al., 2016;Varma et al., 2012;Wei and Lohmann, 2012). A detailed description of the COSMOS model components is given by Stepanek and Lohmann (2012).

Experimental designs 30
The aim of this study is to identify and discuss differences in the modelled climate that occur between the COSMOS simulations contributed to PlioMIP1 and PlioMIP2, and to study the effect of orbits warmer than the configuration applied for PlioMIP.
With this aim in mind, we have applied a methodology for setting up simulations as described below. In general, experiments were carried out following PlioMIP1 and PlioMIP2 protocols (Haywood et al., 2010(Haywood et al., , 2011(Haywood et al., , 2016, respectively. Yet, for the purpose of this study, small modifications to the proposed official PlioMIP model setups are necessary. Three different orbits are employed here (Table 1). Two of them are similar and representative of modern, respectively Pre-Industrial, conditions -the only difference between them being small deviations that were introduced into the reference setup of the COSMOS between PlioMIP1 and PlioMIP2 via adaptation of the Pre-Industrial orbit in COSMOS according to the PMIP4 protocol (Otto-Bliesner 5 et al., 2017). The third employed configuration represents the orbit of MIS K1, the values of orbital parameters being based on the astronomical solution by Laskar et al. (2004). Earth's orbital parameters are prescribed as constant values of eccentricity, obliquity and the longitude of perihelion as outlined in Table 1. The orbit employed for simulating K1 is consistent with the configuration chosen by Prescott et al. (2014).
Simulations are classified in four different categories namely, Pre-Industrial, standard PlioMIP1 set-up, standard PlioMIP2 10 set-up, and modified PlioMIP2 set-up with adapted (MIS K1) orbit (PlioMIP2_K1). First in the order as shown in Table 1 is the PI category, which consists of the PI-control simulations (PI_1 and PI_2) for PlioMIP1 and PlioMIP2, respectively. Ocean bathymetry and land-sea mask are taken from the standard modern setup of the COSMOS. As described by Jungclaus et al. (2006), this set-up has been generated based on the Earth Topography Five Minute Grid (ETOPO5, National Geophysical Data Center, 1988;Edwards et al., 1992). An identical CO 2 concentration of 280 ppmv is prescribed for PI_1 and PI_2, but 15 both experiments differ slightly with respect to orbital parameters and volume mixing ratios of trace gases N 2 O and CH 4 (see Table 1). The constant volume mixing ratios of 270 ppbv N 2 O and 760 ppbv CH 4 are prescribed for PI_1, while the corresponding values for PI_2 are set to 273 and 808 ppbv, respectively. Chlorofluorocarbons on the other hand are constant across all simulations and are absent. Category PlioMIP1 consists of simulation PlioM1, which is the COSMOS mid-Pliocene experiment within the framework of PlioMIP1 (Stepanek and Lohmann, 2012) and utilizes the PRISM3D land-sea mask, 20 orography and ice-mask. Category PlioMIP2 consists of mid-Pliocene simulations based on PRISM4 boundary conditions with slight differences in either orbital forcing or concentration of atmospheric CO 2 . Simulation Eoi400 is the COSMOS mid-Pliocene experiment for PlioMIP2 (Stepanek et al., 2020) in which CO 2 is prescribed to be 400 ppmv, while simulation Eoi405 is derived from Eoi400 in that the concentration of carbon dioxide is set to the PlioMIP1 CO 2 forcing of 405 ppmv.
In comparing them, both simulations enable us to study the impact of the difference of CO 2 forcing between the two phases 25 of PlioMIP on achieved results. Furthermore, simulation Eoi400_ORB employs the orbital forcing utilized by COSMOS in simulating the mid-Pliocene for PlioMIP1, while retaining other boundary conditions and forcing as prescribed for PlioMIP2.
Direct comparison between simulations Eoi400 and Eoi400_ORB will give an indication of the influence of the slight change in orbital forcing on mid Pliocene climate as simulated with COSMOS for PlioMIP1 and PlioMIP2. This is important within the framework of PlioMIP, where orbital forcing is described as similar to modern day, but specific Pre-Industrial orbital parameters 30 differ in the case of the COSMOS between PlioMIP1 and PlioMIP2. Simulation Eoi405_ORB of category PlioMIP2 differs from simulation PlioM1 (Stepanek and Lohmann, 2012) in that it employs the paleoenvironmental reconstruction of PRISM4, and also employs different trace house gas concentrations for methane and nitrous oxide. It differs from simulation Eoi400 on the other hand in both orbital configuration and the prescribed concentration of carbon dioxide (see Table 1). This choice of modelling methodology enables us to infer the relative effects of the improved representation of mid-Pliocene geography from 35 6 https://doi.org/10.5194/cp-2020-5 Preprint. Discussion started: 23 January 2020 c Author(s) 2020. CC BY 4.0 License. PlioMIP1 to PlioMIP2 on our model, while honouring the presence of other differences in the model setup. Furthermore, in order to study the effect of an alternative orbit on mid-Pliocene warmth, simulation Eoi400_K1 in category PlioMIP2_K1 is consistent with the standard mid-Pliocene set-up prescribed for PlioMIP2 and employed in simulation Eoi400, but employs a different orbital forcing which is representative of MIS K1. The choice of our experimental design, enabling a direct comparison between simulations Eoi400 and Eoi400_K1, provides an indication of orbitally influenced climatic variability within the 5 mid-Pliocene. Ultimately, the total effect of all the improved boundary conditions in PlioMIP2 is examined by comparing simulations PlioM1 and Eoi400.

Results
This section present results of the mid-Pliocene and PI simulations listed in Table 1, and attempt to investigate and quantify the differences in the mid-Pliocene climate simulations by COSMOS for PlioMIP1 and PlioMIP2, respectively. Furthermore, we 10 present results from our sensitivity experiments which show the relative contributions of newly prescribed PlioMIP2 boundary conditions (Haywood et al., 2016), with respect to that of PlioMIP1 (Haywood et al., 2010, 2011 in the context of deviations of the COSMOS model setup between PlioMIP1 and PlioMIP2 that are not due to the PlioMIP2 protocol. We also show simulated seasonal variability which may occur due to orbital forcing prescribed in simulating the mid-Pliocene climate. This is examined by a direct comparison of climate forced by two distinct orbital configurations representative of two discrete time-15 slices within the mid-Pliocene, namely MIS K1 and the PlioMIP2 reference orbit of MIS KM5c. The MIS KM5c is selected for simulations within the framework of PlioMIP2 due to it's strong orbital similarity to present-day (Haywood et al., 2016) and thus present-day orbital configuration is adopted (Otto-Bliesner et al., 2017). We analyse and report in this section annual and seasonal means of SAT, SST, and sea ice distribution. Anomalies are tested with regard to significance in the context of internal variability in the contributing simulations by means of the autocorrelation method by Matalas and Dawdy (1964).

SAT
The physical quantity SAT is defined in this study as the temperature of air near the surface (below 2 meter) of the Earth.
At large scale, the COSMOS simulate fairly similar patterns of mid-Pliocene SAT in response to boundary conditions for PlioMIP1 and PlioMIP2 core experiments. Yet, there are noteworthy differences in details of mid-Pliocene SAT anomalies between PlioMIP2 and PlioMIP1. The most pronounced annual average SAT anomaly of mPWP relative to PI occurs in both 25 phases of PlioMIP at high latitudes and in polar regions, providing evidence of a similar, albeit not identical, level of polar amplification in PlioMIP1 and PlioMIP2 mPWP model setups. On average, the COSMOS simulate a mid-Pliocene climate that is 0.29 K colder in PlioMIP2 with the global average SAT reaching 290.97 K for PlioMIP1, while the corresponding value for PlioMIP2 is 290.68 K. It is important to note that the values computed for PlioMIP1 are recomputed based on averaging over 100 years, while Stepanek and Lohmann (2012) have, in agreement with the PlioMIP1 analysis protocol, provided aver-30 ages over 30 years. Anomalies between coupled ocean-atmosphere simulations averaged over 100-year period for PlioMIP1 and PlioMIP2, namely PlioM1 and Eoi400, show that SAT anomalies are fairly similar in PlioMIP1 and PlioMIP2 over the equatorial oceans, but that there is substantial deviation of SAT over continents and polar regions if results from PlioMIP1 and PlioMIP2 are compared with each other. The most pronounced warming from PlioMIP1 to PlioMIP2 is seen over Greenland and Antarctica. On the other hand, gradual cooling is present over the Arctic from PlioMIP1 to PlioMIP2. Over the oceans in PlioMIP2, COSMOS simulates warmer SAT over the North Atlantic, while SAT over the Indian Ocean, South Atlantic Ocean, and low-latitude Pacific Ocean is largely unchanged. Generally, both PlioMIP1 and PlioMIP2 mid-Pliocene simulations suggest 5 that land masses are warmer than the ocean, and furthermore that there is presence of substantial polar amplification; the latter is more pronounced in PlioMIP1 simulation PlioM1 (compare Figure 2a, 2b and 3b). The strong regional warming simulated over Greenland and Antarctica is linked to changes in albedo and orography over these regions (confer Figure 1a and 1b). As the magnitude of changes in albedo and elevation over Greenland are more pronounced in PlioMIP2 than in PlioMIP1 ( Figure   1a,b) due to reduction in the spatial extent of Greenland ice-sheet from PRISM3D to PRISM4 (Dowsett et al., 2016), also the 10 regional warming signal is more pronounced in PlioMIP2 ( Figure 3b). For simulation PlioM1, a warming of about 15 K is found over Greenland, while the corresponding value is about 21 K for Eoi400. This illustrates that the mPWP SAT anomaly simulated with the COSMOS is regionally increased in PlioMIP2 in comparison to PlioMIP1, even though the global average mPWP temperature anomaly is smaller in PlioMIP2. Antarctic ice-sheet estimates, based on results produced with the British Antarctic Survey Ice Sheet Model utilizing a climate simulation produced with PRISM boundary conditions (Haywood et al.,15 2010), remain the same for both phases of PlioMIP (Dowsett et al., 2016;Haywood et al., 2016). Over Antarctica, there are strong regional mPWP temperature anomalies, while cooling is evident in the South Pacific, between 60 • S -70 • S, extending from 65 • W -150 • W for both simulations. This South Pacific cooling is about -1.2 K on average in PlioMIP1 (PlioM1 with respect to PI_1), but more intense in PlioMIP2 (Eoi400 vs. PI_2), where the temperature anomaly locally reaches -4 K and is characterized by a wider spatial extent. In both PlioMIP1 and PlioMIP2, the Indian sector of the Southern Ocean experiences 20 intense warming that extends from 60 • E into the South Pacific Ocean. Extreme values of SAT anomalies vary from 25.2 K in PlioM1 to 23.1 K in Eoi400 over the adjacent Antarctic landmass. Moving the focus again to the Northern Hemisphere, we highlight that both simulations PlioM1 and Eoi400 are characterized by land-sea masks in which the modern Hudson Bay is absent, while that region is part of the oceans in our PI simulation setups. The change in the land sea mask introduces an annual mean warming in the mid-Pliocene in comparison to Pre-Industrial that is fairly similar in PlioMIP1 (simulation 25 PlioM1) and PlioMIP2 (simulation Eoi400), but slightly stronger in PlioMIP1.If we study the influence of modified radiative forcing on mPWP SAT, that is introduced by differences in prescribed CO 2 between PlioMIP1 and PlioMIP2, and we find a contrasting effect over the Arctic and Antarctica Figure 4a. The prescribed 5 ppmv difference between simulations Eoi405 and Eoi400 causes cooling of the Arctic region and a dipole of warming and cooling over the North Atlantic, while the rest of the ocean surface warms up. Testing the impact of the small deviation in orbital forcing on the mPWP climate we find that, 30 despite prescribing orbital forcing that is representative of modern day for both PlioMIP1 and PlioMIP2, small changes in this forcing reflect different patterns of warming in the polar regions, as well as in mid-to-high latitudes. The PlioMIP2 core experiment Eoi400, if forced in simulation Eoi400_ORB with orbital configuration utilized by (Stepanek and Lohmann, 2012) in PlioMIP1, shows fairly constant SAT in the Equatorial Pacific, while it is relatively warmer in the Atlantic and Indian Ocean (see Figure 4b). Analyzing the effect of a different orbital configuration that is known to have been present during the mPWP we find via a comparison between simulations Eoi400_K1 and Eoi400 that specifying orbital parameters, representative of MIS K1, will produce a mid-Pliocene climate that is warmer at low-and-high latitudes and in the Arctic, but that is colder in the Southern Hemisphere polar region (see Figure 5a). While almost all land masses are warming under the influence of MIS K1 orbital forcing, cooling is evident over the southern part of the modern United States, north Greenland, northern Australia and some 5 parts of Eurasia. With the exception of the impact of eccentricity, orbital forcing causes a redistribution of incoming solar radiation across latitudes and seasons, rather than a change of the overall input of solar radiation into the climate system. Hence it is of particular interest to study the seasonality of the climatic effect. We find that the impact of MIS K1 orbital forcing is stronger at seasonal time scale than in the annual mean. We show time-mean SAT anomalies for boreal summer (JJA) and boreal winter (DJF) in order to analyse the orbital control on seasonality of the mid-Pliocene (Figure 5b and c). Our finding is 10 that in simulating the mid-Pliocene, the choice of orbit has great influence on seasonal patterns of SAT. During boreal winter, the K1 orbit is warmer with respect to KM5c, with a maximum SAT anomaly of 3.5 K in comparison to the simulation with KM5c orbit, while averages over boreal summer months provide a colder Northern Hemisphere for Eoi400_K1. Substantial winter warming is present over Eurasia, North and South America as well as South-West Africa. SAT over Greenland shows pronounced seasonal variation, with warming during winter and cooling in summer. We note potential implications of relative 15 summer cooling for the state of the Greenland Ice Sheet during the mPWP.
With respect to the impact of K1 orbital forcing on mPWP climate in the Southern Hemisphere we find that the seasonal dependency of the temperature anomaly is different for Antarctica, where cooling is dominant all year round. SAT changes in the Southern Hemisphere during summer are not statistically significant in many regions, based on the significance test to account for effective degrees of freedom by means of the autocorrelation method of Matalas and Dawdy (1964). This indicates 20 similarity between summer SAT simulated in this region for both K1 and KM5c (see Figure 5b). We suggest that changes in palaeogeography between the two phases of PlioMIP may be the main contributor to the differences between mid-Pliocene SAT for PlioMIP1 and PlioMIP2, as simulated by the COSMOS. Simulations conducted to account for the influence of the PRISM4 geography show interestingly a similar pattern with the total boundary condition effect, the latter deduced from the difference between simulations Eoi400 and PlioM1 (compare Figure 3a and 3b). The major difference between the effect of 25 changed palaeogeography and the total effect (including orbit) is noticed in the Arctic region, where cooling of a relatively higher magnitude is observed for the total effect. Furthermore, there are significant differences between these simulations especially in regions where obvious changes in orography are applied.

30
In mid-Pliocene simulations for both PlioMIP1 and PlioMIP2, the equatorial warm pool extends across all ocean basins of the world ( Figure 6). This warm pool is characterized as a pattern of warm water at the surface around the equator, and defined as the region where absolute SSTs exceed 28.5 • C (Watanabe, 2008). In COSMOS simulations we find no change in the pattern of the equatorial warm pool between the mid-Pliocene simulations of PlioMIP2 and PlioMIP1, as both realizations of the 9 https://doi.org/10.5194/cp-2020-5 Preprint. Discussion started: 23 January 2020 c Author(s) 2020. CC BY 4.0 License. mid-Pliocene climate state show warm pools of almost identical spatial extent (cf. Figure 6). Beyond the extent of the equatorial warm pool we find similarity of SST anomalies obtained from the comparison of mid-Pliocene simulations of PlioMIP1 (PlioM1) and PlioMIP2 (Eoi400) to their respective PI control runs, that show in many regions of the world very similar cooling and warming signals. The Northern Hemisphere surface ocean is -with few exceptions of regional cooling -characterized by regions of increased SST, a pattern that is more pronounced in the Pacific. Evidence from PlioMIP1 (Dowsett et al., 2013) sug-5 gests that mid-Pliocene North-Atlantic SST, as derived from geological records, is consistently underestimated by the models.
The COSMOS PlioMIP2 mid-Pliocene simulation produces a warmer North-Atlantic than the respective PlioMIP1 simulation (compare Figure 6 a,b). As a result, the model-data-discord in that region, that has been identified by Dowsett et al. (2013), is mitigated to some extent. This is confirmed by obtaining the Root Mean Square Deviation (RMSD) between COSMOS simulated SSTs for both PlioMIP1 and PlioMIP2 and the SST reconstruction by Dowsett et al. (2013) values taken from their 10 supplementary table S1), respectively. A RMSD of 5.14 is evident between reconstructed and modelled mid-Pliocene North Atlantic SSTs in PlioMIP1 (Stepanek and Lohmann, 2012), while a corresponding RMSD of 2.96 is estimated in PlioMIP2 (Stepanek et al., 2020). A better agreement between reconstructed and simulated North Atlantic SSTs for PlioMIP2 is linked to the strength of the Atlantic Meridional Overturning Circulation (AMOC) at different depths. Increased SSTs are accompanied by enhanced AMOC in the upper cell at about 1,000 m depth, which transports more heat to high latitudes of the Northern 15 Hemisphere (cf. Figure 7 a,b). Furthermore, the inflow of deep-water from the South-Atlantic, more precisely the transport of the Antarctic Bottom Water, is stronger in our PlioMIP2 simulation: AMOC at depths shallower than 3000 m is slightly enhanced from PlioMIP1 to PlioMIP2 (see Figure 7a and b). Similarly, RMSD difference of 0.51 is obtained in favor of Eoi400 when compared with Eoi400_K1, observing a clearer signature of MIS KM5c than K1 in the available mid-Pliocene SST reconstruction.

20
Similar to SAT, COSMOS simulates almost the same SST pattern in PlioMIP1 and PlioMIP2 when comparing the influence of PRISM4 geography to the total boundary condition effect (constituting geography, orbit, and atmospheric trace gases). Both mid-Pliocene simulations show the expected warming in the Arctic Ocean, North Atlantic Ocean, and also in the Indian and Atlantic sector of the Southern Ocean. In contrast to the total effect of PlioMIP2 boundary conditions, PRISM4 geography produces a warmer Equatorial Pacific when implemented instead of PRISM3D geography together with other PlioMIP1 boundary 25 and initial conditions (see Figure 8a and b). The relative contribution of CO 2 difference between PlioMIP1 and PlioMIP2 generally produces a warmer ocean surface in PlioMIP1 in comparison to the setup of PlioMIP2, which is more pronounced in the Southern Ocean (see Figure 9a). In the North Atlantic, a dipole of warm and cold ocean surface prevails as seen in SAT, with the maximum SST increase amounting to 2.9 • C, while maximum cooling reaches -1.3 • C as a result of increased CO 2 .
The North Atlantic pattern is largely similar when considering the effect of changes in orbital forcing on mid-Pliocene SSTs 30 (Figure 9b), with the obvious difference being a modification of the warming/cooling pattern and decreased magnitude of SST anomalies. If we turn our attention towards the influence of K1 orbital forcing on simulated SSTs of the mPWP, we find that a maximum anomaly of 2.2 • C is obtained from the difference between simulations Eoi400_K1 and Eoi400 in the mid-latitudes ( Figure 10). Furthermore, simulation Eoi400_K1 with K1 orbital forcing produces warmer Northern Hemisphere oceans than that of simulation Eoi400 that is based on a KM5c orbit -with the few but obvious exceptions being the Arctic Ocean beyond 35 10 https://doi.org/10.5194/cp-2020-5 Preprint. Discussion started: 23 January 2020 c Author(s) 2020. CC BY 4.0 License. the Barents Sea, Greenland Sea, Hudson Bay, as well as the central North Pacific Ocean, where (regional) cooling is observed.
The Southern Ocean on the other hand shows mixed signals of warming and cooling under the influence of K1 orbital forcing, with the most pronounced SST increase noticed to the south east of Australia and South America, which is probably related to a shift of the polar fronts, and also between 100 • E and 160 • E off the Antarctic coast ( Figure 10). The effect of improved mid-Pliocene palaeogeography seems to be most dominant when comparing SATs obtained from simulations Eoi405_ORB 5 and PlioM1 (Figure 3a). On the contrary for SSTs, the combined effect of all the improved boundary conditions produces a warmer North Atlantic (compare Figs. 8a and b). This implies that the effect of geography on Arctic temperatures is different for atmosphere and ocean realms.

Sea ice
Generally, annual mean sea ice is strongly reduced for both PlioMIP1 and PlioMIP2 mid-Pliocene simulations if compared to 10 the respective PI simulation. The general pattern of mid-Pliocene sea ice is rather similar, in that sea ice cover retreats towards the North Pole (see Figure 11).
For seasonal analysis of sea ice, the definition of seasons is different from the conventional December to February (boreal winter) and June to August (boreal summer) time period. In this study, with regard to sea ice, winter is rather defined as February to April (FMA), and summer rather as the months from August to October (ASO). According to Howell et al. (2016), 15 these are the three-month-periods in which more than half of the PlioMIP1 ensemble simulations show the highest and lowest mean sea ice extent, respectively. There are slight differences in prescribed orbital parameters, both for PI simulations and the respective mid-Pliocene simulations considered in this study that employ an orbital forcing similar to the Pre-Industrial's. We find that differences in the orbital forcing have no effect on the large scale seasonal pattern of Pre-Industrial Arctic sea ice, as our results show for respective simulations similar spatial extent for both summer and winter (compare Figure 11a and c, 20 11b and d). With the mPWP boundary conditions prescribed for PlioMIP1 and PlioMIP2, COSMOS simulates a considerably smaller sea ice extent for the mid-Pliocene with respect to Pre-Industrial simulations (compare Figure 11e, f, g, h to Figure   11a, b, c, d). The most obvious loss of sea ice in the mPWP is present around the Hudson Bay, which is represented as land for mid-Pliocene simulations. In contrast, in our Pre-Industrial simulations, the Hudson Bay is totally ice-free during summer, but sea ice compactness at this location reaches a maximum during winter. The Canadian Arctic Archipelago is totally free of sea 25 ice for mid-Pliocene simulations with PRISM4 geography. In addition to these trivial changes in sea ice, little or no sea ice is present around the Bering Strait during mid-Pliocene boreal summer.
The combined effect of mid-Pliocene geography, orbit and CO 2 (derived from simulations PlioM1 and Eoi400) shows that the mid-Pliocene Arctic ocean is ice free during summer. Sea ice compactness drops below 15% in our PlioM1 simulation, while for simulation Eoi400 we find that summer sea ice is quasi absent (compare Figure 11f and 11h). If we allow as an additional 30 degree of freedom variation of the orbital configuration within the limits of plausibility during the mPWP, and choose the K1 configuration in simulation Eoi400_K1 rather than the KM5c configuration in PlioM1, Eoi400, and derived simulations, we find a larger sea ice extent and compactness during boreal summer as a result of the K1 orbital configuration ( Figure   12). Enhanced warming in the Northern Hemisphere polar regions during summer is simulated for MIS KM5c (Figure 10).
Hence, colder conditions with increased summer sea ice prevail for K1 (see Figure 12b and 12d. Simulation Eoi400 shows considerable amount of boreal winter sea ice around the pole and a gradual reduction towards the continents. This is also the case for simulation Eoi405, despite the 5 ppmv difference between both simulations (compare Figure 11g and 12a). Simulation Eoi400_ORB shows a similar spatial sea ice pattern as simulation Eoi400 (compare Figures 12e, f to 11g, h), which implies that small changes in orbital forcing between both COSMOS simulations have little or no effect on sea ice. Generally, all 5 simulations with PRISM4 geography produce sea ice with larger spatial extent and compactness during summer months than it is the case for their PRISM3D counterpart PlioM1, irrespective of the specified concentration of atmospheric CO 2 (compare Figure 12b,d,f,h with 11f). Therefore, the change in CO 2 between PlioMIP1 and PlioMIP2 does not affect large-scale patterns of summer sea ice in the Arctic.

10
When comparing large scale patterns of mPWP climate simulated by us with COSMOS in the framework of PlioMIP1 and PlioMIP2, we unsurprisingly find that also in our contribution to PlioMIP2 the mPWP offers a glimpse into a climate state that is overall warmer than the conditions humankind is currently experiencing. Noteworthy is that relative warmth in the mPWP is possible with a prescribed CO 2 forcing that is actually below the current volume mixing ratio in the atmosphere -400 ppmv in PlioMIP2's mPWP, about 407 ppmv for 2018 (Friedlingstein et al., 2019, and one reference therein). Yet, we also infer that 15 updates of the model setup from PlioMIP1 to PlioMIP2 lead to both a global and a regional modulation of the overall warmth simulated for the mPWP with respect to the Pre-Industrial.
One of the main objectives of the PlioMIP is to determine the dominant components of mid-Pliocene warming, derived from the imposed boundary conditions (Haywood et al., 2016). For the COSMOS simulation in response to PlioMIP1 boundary conditions (simulation PlioM1), the most pronounced warming is evident over areas where changes in albedo and orography 20 have been implemented (Stepanek and Lohmann, 2012). This is also the case for PlioMIP2 simulations with the COSMOS, above all for the PlioMIP2 core simulation Eoi400. However, the increased number of simulations with dedicated sensitivity studies in PlioMIP2, in comparision to the approach in PlioMIP1, that considered only one simulation based on the best knowledge on mPWP boundary conditions (Haywood et al., 2010(Haywood et al., , 2011, allows a proper inference of the main drivers of mid-Pliocene warmth. While the study by Stepanek et al. (2020) aims at unravelling the various contributions of PRISM4 25 boundary conditions to the mPWP climate anomaly as simulated with COSMOS, with this study we employ further sensitivity experiments that go beyond the PlioMIP2 modelling protocol in order to determine the relative contribution of the improved boundary conditions from PlioMIP1 to PlioMIP2. Our aim is to bridge the gap between COSMOS contributions to PlioMIP1 and PlioMIP2, that are based on model setups that differ beyond the discrepancies between boundary conditions outlined in protocols for PlioMIP1 (Haywood et al., 2010, 2011) and PlioMIP2 (Haywood et al., 2016. Our extended modelling approach 30 allows inference into the contributions of different components to the results achieved with COSMOS in PlioMIP2. Our respective inferences are outlined below. Generally, we find that the effects of changes in boundary and initial conditions are pronounced in the higher latitudes, while SAT and SST are largely unchanged in the lower latitudes. Hence, for the COSMOS the impact of updates of the modelling methodology from PlioMIP1 to PlioMIP2 (encompassing implementation of PRISM4 boundary conditions with slightly reduced CO 2 , increased detail of mPWP land sea mask and gateway configuration, resolution of climate-vegetation feedbacks via the use of the model's dynamic vegetation module) modifies the polar amplification in the simulated mPWP climate.

5
While regionally the PlioMIP2 core simulation Eoi400 is warmer than the respective climate state of PlioMIP1 (simulation PlioM1), in particular where the ice sheet reconstruction is updated in PRISM4, the Arctic is generally cooler in our contribution to PlioMIP2. Lower temperature anomaly in the Arctic imprints on sea ice conditions in high latitudes of the Northern Hemisphere and the global average temperature simulated in PlioMIP2, where a small reduction of the modelled temperature anomaly is evident from PlioMIP1 to PlioMIP2. Generally, we find that in relation to updated gateways there is a pronounced 10 regional increase of SST in the North Atlantic Ocean, while reduced carbon dioxide leads, as expected, to an overall cooling of the mPWP climate in COSMOS simulations of PlioMIP2. Yet, a comparison of SAT simulated in the framework of a sensitivity study, where the PlioMIP2 core simulation Eoi400 is repeated with the higher CO 2 forcing employed in PlioMIP1 simulation PlioM1, reveals that the impact of radiative forcing by increased CO 2 is hemispherically dependent. In the PlioMIP2 model setup with increased CO 2 (as in PlioMIP1), the North-Atlantic and large parts of Mediterranean and Arctic are actually cooler 15 than in the standard PlioMIP2 model setup with 400 ppmv CO 2 . In contrast, for Southern Hemisphere, North Atlantic equatorward of 30°N, and most of the North Pacific, a model setup with the higher PlioMIP1 CO 2 forcing would indeed lead to a warmer mPWP state than what is simulated in our PlioMIP2 core simulation Eoi400. Our CO 2 sensitivity study within the framework of PlioMIP1 and PlioMIP2 shows that a difference of 5 ppmv between both phases of PlioMIP causes appreciable changes in SAT over land and oceans. The increase in CO 2 produces warmer oceans in the PRISM4-based model setup, 20 except for parts of the North Atlantic and the Arctic as outlined above. As in the PRISM4 COSMOS model setup the effect of increased CO 2 is opposite to the pronounced mPWP warming simulated in the North Atlantic, that is found in PlioMIP2 simulation Eoi400 in comparison to PlioMIP1 simulation PlioM1, we conclude that the warmer North Atlantic simulated by us in PlioMIP2 is not influenced so much by changes in greenhouse forcing as it is by the collective contribution of all boundary conditions. As a matter of fact, the effect that relatively small changes in CO 2 have in the PlioMIP2 mPWP setup of COSMOS 25 on the temperature of the North Atlantic Ocean leads us to the inference that the gateway effect on North Atlantic Ocean warmth is modulated in our model by greenhouse forcing, and that our PlioMIP2 model setup does not provide a final statement with regard to the strength of the gateway effect that may have impacted on the mPWP temperature signals interpreted from the geologic recorder. Other model configurations with altered greenhouse gas forcing may well produce a larger temperature anomaly than the COSMOS PlioMIP2 core simulation Eoi400. If we extend our focus beyond model performance 30 likely at least partly a result of modulation of internal variablity in the model that is triggered by alteration of the model forcing.
While PlioMIP2 already made a major leap forward by increasing the analysis time period, from 30 years in PlioMIP1 to 100 years in PlioMIP2, this finding highlights that in future intercomparisons it may be worthwhile to further increase the analysis time period towards multicentennial time scales. Our suggestion is motivated by the observation that beyond the common short term climate variability, that occurs for example in the form of the El Nino Southern Oscillation (Christensen et al., 2013, 5 see their Figure 14.13) and North Atlantic Oscillation (Hurrell, 1995, see their Figure 1A), there are certain modes of internal variability that act at ocean basin scale with much slower periodicity. If the PlioMIP analysis period were at multicentennial time scales, such slow modes of internal variability would be more likely suppressed by averaging over multiple realizations of climate patterns that are related to opposite phases of these modes, hence leading to a clearer emergence of a mean (that means, predominant across time periods of centuries) mPWP climate state.

Summary and conclusions
In this study, we compared the results of mid-Pliocene simulations within the frameworks of PlioMIP1 and PlioMIP2, that have been carried out with the COSMOS at the Alfred Wegener Institute in Germany. We have tested to which degree changes between the COSMOS model setups of PlioMIP1 and PlioMIP2 are able to create different realizations of simulated mPWP climate, and how different these alternative mPWP climate states are from PlioMIP2 core simulation Eoi400 and PlioMIP1 15 simulation PlioM1, both produced with the same climate model as employed in this study. Overall, we find that the global patterns of mPWP climate as expressed in SST and SAT are similar in PlioMIP2 and PlioMIP1. Yet, in PlioMIP2 we simulate certain differences that impact in particular sea ice conditions in the Arctic and the amplitude of mPWP temerpature anomaly in the North Atlantic Ocean. Beyond small changes of the Pre-Industrial orbit reference configuration in COSMOS and the modification of CO 2 , that arises from the transition of PRISM3D to PRISM4 boundary conditions, we carried out sensitivity 20 experiments that address the climatic impact of plausible orbital configurations across the mPWP (MIS KM5c and K1) to assess variability across different discrete time-slices within the mid-Pliocene warm period, using boundary conditions prescribed for PlioMIP2.
From our results, we conclude that palaeogeography is the dominant component of the imposed mid-Pliocene boundary conditions, influencing the climate across the warm period. Its effect on atmosphere and ocean differs as seen in the different 25 pattern of warming and cooling achieved for SAT and SST. We also conclude that the response of atmospheric and oceanic variables differ distinctly when exposed to the same geography. The difference in CO 2 between PlioMIP1 and PlioMIP2 simulations does not change the general impression of large scale mPWP climate patterns, but produces warmer oceans especially in high latitudes of the Northern Hemisphere. A similar effect we find if we produce simulations with slight alterations of the Pre-Industrial surrogate of KM5c orbital forcing. The impact of a small amplitude of change in orbital forcing leads us to the 30 suggestion that the simulated PlioMIP climate state may be subject to appreciable internal variability that -while not changing the overall impression on the characteristics of simulated mPWP climate -provides regional patterns of SST and SAT that are worthwhile to be studied in future phases of PlioMIP based on a prolonged multicentennial analysis period. Furthermore, the newly prescribed PlioMIP2 boundary conditions play a significant role in achieving warmer North Atlantic SSTs, that were modelled too cold during PlioMIP1, if compared to reconstructions. On the other hand, the PlioMIP2 mPWP reference state simulated with COSMOS (simulation Eoi400) is globally slightly cooler, in agreement with reduced concentrations of CO 2 , and also expresses itself via reduced warmth and resultantly increased summer sea ice conditions in the Arctic Ocean. When considering various modifications of mPWP boundary conditions in our simulations, we find that there is the 5 potential for pronounced variability of Arctic summer sea ice conditions during the mPWP. Enhanced AMOC in PlioMIP2 is achieved through the combined effect of PlioMIP2 boundary conditions, contributing to increased meridional transport of warm water from low to mid latitudes in the North Atlantic. We note that details of the model forcing, in particular the exact level of carbon dioxide and the realization of the KM5c orbit surrogate, have an impact on the amplitude of North Atlantic Ocean warming in comparison to results derived with COSMOS in PlioMIP1. With respect to Pre-Industrial conditions, ex-10 tended equatorial warm pool, shown in COSMOS simulation PlioM1 for PlioMIP1, is also present in our PlioMIP2 simulation Eoi400 with similar spatial pattern and extent. This is in line with our finding that the effect of changes in mid-Pliocene boundary conditions from PlioMIP1 to PlioMIP2 is minor in low latitudes, and more pronounced in high-latitudes. The magnitude of Arctic warming simulated for PlioMIP1 is reduced in our PlioMIP2 simulation, while increased warming over Greenland and Antarctica is a direct effect of changes in reconstructed orography and albedo from PlioMIP1 to PlioMIP2.