For PlioMIP2, the boundary conditions for the modern-day and the “enhanced” variant of the Pliocene reconstruction are provided on regular 1∘ grids held within NetCDF files . For convenience, we shall refer to the PlioMIP2 enhanced boundary condition as PRISM4. The modern geography is provided to facilitate the anomaly method of boundary condition generation. The LSM is created by computing the anomaly of PRISM4 Pliocene minus PRISM4 modern (at 1∘ resolution) and regridding using bilinear interpolating to the 3.75∘×2.5∘ model grid and then applying the anomaly to the model's pre-industrial LSM. This is so that the final reconstruction is consistent with both the original pre-industrial model setup and the PRISM4 LSM. Finally, a number of manual corrections were applied to the resulting 3.75∘×2.5∘ PRISM4 LSM to ensure that the underlying character of the PRISM4 reconstruction is represented as best as reasonably practicable at the model's resolution. For consistency with the pre-industrial boundary conditions developed by MOHC, we remove Svalbard and Novaya Zemlya, despite their subaerial extension within PRISM4. Similarly, we keep the Pliocene LSM in the Persian Gulf region the same as pre-industrial despite a withdrawal of the Persian Gulf within PRISM4. This choice was made as the Persian Gulf within the pre-industrial LSM is represented by an inland sea (due to inadequate spatial resolution) and so further changes would be difficult to interpret. At model resolution, the Pliocene Strait of Gibraltar is identical to the pre-industrial and so the diffusive pipe is incorporated.

The resulting PRISM4 LSM was used to constrain the generation of the Pliocene orography and bathymetry (which was generated using area-weighted regridding and then applied as an anomaly to the existing HadCM3 pre-industrial orography and bathymetry). River basins and outflow points were derived from the pre-industrial routing scheme (Sect. ) but corrected in regions of LSM, topographical and ice–bedrock change using a model-resolution river routing model based on the D8 method . This was then followed by manual correction in regions where model resolution fails to capture important orography, or where the regridded Pliocene orography is flat. The PRISM4 vegetation scheme (represented by BIOME4 biomes) was regridded by combining a BIOME4-to-MOSES2 lookup table with an area-weighted survey of underlying biomes. The vegetation scheme is then held fixed within each experiment. A similar area-weighted regridding was conducted for the lake field. We chose not to generate the lake field as an anomaly from the modern lake distribution, as land surface change since the pre-industrial would be imprinted on the model's lake distribution.

Rigid-lid Bryan–Cox-type models, such as the ocean of HadCM3, require islands (and by extension, continents) to be identified so that a net non-zero barotropic flow (depth independent) can be achieved around the line integral (streamfunction non-zero). The default pre-industrial configuration of the model has six islands defined and is shown within Fig. . For consistency, aforementioned (Sect. ) manual corrections to both LSM and bathymetry have allowed islands to be specified that are consistent with the E280 experiment but also reflect the key palaeogeographic changes presented by the PRISM4 palaeogeography. In particular, western Iceland and East Greenland land cells were adjusted to ensure that Iceland could be defined as a streamfunction island (Fig. ), and hence we could fully represent the East Greenland Current. The island to the west of the Antarctic Peninsula body lies outside the island definition of the main Antarctic continent and therefore the circulation between the two is not fully resolved (only the baroclinic flow is resolved fully). Figure  compares the pre-industrial and PRISM4 Pliocene HadCM3 island specification. It can be seen that the six islands in the pre-industrial configuration have been increased to eight islands in the Pliocene.

It is noted that within the pre-industrial HadCM3 model setup the Bering Strait barotropic component of throughflow is unresolved and both the Makassar Strait and the Canadian Archipelago are spatially unresolved (Sect. ). This poses a conceptual problem in the interpretation of the Pliocene experiments with respect to the pre-industrial, as the PRISM4 Pliocene geography has these throughflow regions closed. Therefore, our simulations do not resolve the full climatic response of these regional palaeogeographic changes. A pre-industrial experiment with a fully resolved Bering Strait and Canadian Archipelago would partially address these problems but would then force a divergence away from the previous HadCM3 descriptions and evaluations, as well as from past and current CMIP/PMIP and PlioMIP1 model implementations. These problems are likely to arise in all rigid-lid streamfunction ocean models that have insufficient spatial resolution to fully resolve these gateways and inherently cannot resolve line integrals around bounding land masses. Ocean models that have explicit or implicit free-surface schemes with sufficiently high horizontal spatial resolution may reduce these issues.

Modelled mean annual 1.5 m surface air temperatures (hereafter MASAT) are detailed within Table  and corresponding Pliocene anomalies are shown within Fig. . Relative to the pre-industrial control (E280) temperatures are generally warmer within the Pliocene experiments. Very high differences in MASAT of up to 31.3 ∘C are reached over regions of Greenland and Antarctica, where the elevation of Pliocene ice sheets has been changed with respect to the present. Typically, warming is greatest over land, although in ocean regions at or near Antarctic LSM change (pre-industrial grounded ice to Pliocene ocean) warming is significant. This pattern of warming is similar to results derived with HadCM3 within PlioMIP1 under PRISM3 boundary conditions (Exp. 2 of ).

The Pliocene cooling in the Barents Sea is statistically significant and persistent through the model integration (Fig. ). It coincides with an increase in Pliocene winter and spring sea ice concentration driven by palaeogeographic terrestrial winter cooling in the circum-Arctic (Pliocene subaerial Barents and Baltic Sea). This cooling is potentially driven by the partial suppression of northward heat transport (in the Norwegian Current) by the subaerial extension of Ireland and Scotland within the model.

The Eoi400–E280 MASAT anomaly of 2.9 ∘C (Table ) is lower than the 3.3 ∘C of HadCM3 within PlioMIP1 and lies within the PlioMIP1 model ensemble range of 1.84–3.60 ∘C . The MASAT anomaly also lies between the PlioMIP2 studies of (2.4 ∘C) and (3.8 ∘C), although note that this comparison is not exhaustive as PlioMIP2 is incomplete at the time of press. Table  also presents MASAT data for the equatorial (between 30∘ S and 30∘ N) and polar regions (latitudes greater than 60∘). The resulting polar amplification factors for the Pliocene control (Eoi400) relative to the pre-industrial control (E280) are 1.7 for the Northern Hemisphere and 2.2 for the Southern Hemisphere.

Figure  shows the annual and seasonal temperature anomalies for Eoi280 and Eoi400 (against E280). Terrestrial regions that are exposed only within the Pliocene, such as the Hudson Bay and the Baltic Sea regions, are up to 10 ∘C warmer (colder) during the summer (winter) seasons, due to land–ocean heat capacity contrast. It is unclear how much of this seasonal temperature response in the Baltic Sea region (exposed during the Pliocene) is a driver of persistent cooling within the Barents Sea region.

From the results in Table , it is possible to diagnose the factors that contribute to Pliocene warming relative to the pre-industrial (E280). Considering the Pliocene control experiment (Eoi400), we find that the change in palaeogeography (Eoi280–E280) accounts for a temperature change of 1.4 ∘C, whilst the increase in CO2 (Eoi400–Eoi280) accounts for a further 1.5 ∘C of warming. Considering uncertainty in Pliocene CO2 level, we find temperature changes of 0.9 and 2.0 ∘C for Eoi350–Eoi280 and Eoi450–Eoi280, respectively. The PlioMIP2 experimental design provides a second pathway to examine Pliocene palaeogeographical and CO2 forcing (e.g. Eoi400–E400 and E400–E280). Within this pathway, the Pliocene geography (Eoi400–E400) accounts for 1.1 ∘C of warming and the increase in CO2 (E400–E280) accounts for 1.8 ∘C of temperature increase. These differences highlight that there are non-linearities within the climate system's response to changes in boundary condition.

The climate system's sensitivity to a doubling of CO2 (climate sensitivity; CS) is 3.5 ∘C for the pre-industrial (derived from E560 and E280) and 2.9 ∘C for the Pliocene (derived from Eoi400 and Eoi280 and scaled by 1.94 (=log⁡(560/280)/log⁡(400/280)). The pre-industrial CS is consistent with the 3.3 ∘C for HadCM3 within CMIP3 . The Pliocene CS is similar to the 3.1 ∘C for HadCM3 and lies at the lower end of the 2.7–4.1 ∘C ensemble range of PlioMIP1 experiment 2 . When we approximate Earth system sensitivity (ESS) using Eoi400 and E280 (with ESS =1.94xΔTEoi400–E280), we obtain ∼5.6 ∘C. Subsequently the ESS / CS ratio is ∼1.9, which lies at the higher end of the 1.1–2.0 range of the PlioMIP1 ensemble in which HadCM3 had a ratio of 2.0. It must be noted, however, that such a comparison of CS and ESS is only meaningful when one assumes that the PlioMIP2 enhanced boundary condition approximates the equilibrated Earth system under a contemporary doubling of CO2, which is a reasonable position since the changes in non-glacial elements of the PRISM4 retrodicted palaeogeography are relatively small.

The globally integrated mean annual precipitation (MAP; Table ) is influenced by both Pliocene geography and CO2 changes. Pliocene geography acts to increase globally integrated MAP, although this appears sensitive to the background CO2 level (e.g. Pliocene geography increases MAP by 0.07 and 0.05 mm d-1 at 280 and 400 ppm, respectively). The Eoi400–E280 MAP anomaly of 0.11 mm d-1 (Table ) compares with the 0.17 mm d-1 from HadCM3 within PlioMIP1 and sits at the lower end of the ∼0.09–0.18 mm d-1 range of the PlioMIP1 model ensemble .

The geographical distribution of MAP change can be seen within Fig. . Northern Hemisphere land masses generally see increased precipitation within the Pliocene, although this effect is minimal in the continental interiors. In the Southern Hemisphere, much of South America and south Africa receive less precipitation, whilst Australia and northern Greenland see an increase in precipitation during the Pliocene. Increasing Pliocene CO2 generally intensifies the precipitation anomaly, which is most apparent in the tropics. Regions that receive little precipitation within E280; e.g. north Africa and the East Antarctic Ice Sheet have little (<0.1 mm d-1) change in precipitation under increasing Pliocene CO2.

Seasonal plots of precipitation change between the Pliocene (Eoi400) and the pre-industrial (E280) control experiments are shown in Fig. . During the Pliocene, we see wetter summers over much of North America and northern Europe. Regions experiencing reduced precipitation in western North America as well as central and western Europe are a consequence of weakened westerlies (not shown). As can be seen within Fig. c–f, the Pliocene geography and land surface change drive an intensification of precipitation associated with the Intertropical Convergence Zone (ITCZ), although changes in seasonal latitudinal distribution are not evident. The South Pacific Convergence Zone, extending from the western Pacific warm pool (WPWP) southeastward to the south central Pacific, extends ∼15∘ further east in E280 than Eoi400 and Eoi280.

The time-averaged zonal mean meridional mass transport streamfunction for the atmosphere is shown within Fig. . Clearly distinguished are the Hadley, the Ferrel and the polar cells. Taking the maximum of the meridional streamfunction as a measure of the Hadley cell strength, we find that the Pliocene geography acts to weaken (intensify) the Hadley cell within the Northern (Southern) Hemisphere. Looking at E280, we find the northern cell is stronger (+10.8 %) than the southern cell, which is in contradiction with observational and reanalysis data that consistently show the southern cell being stronger than the northern cell. With increasing Pliocene CO2, the southern cell intensifies and becomes stronger than the north (+19 % in Eoi280 and +42 % in Eoi400). This intensification (weakening) of the Hadley cell under changed land surface and geography should be driven by steepening (shallowing) of the tropical meridional temperature gradients in the tropics south (north) of the ITCZ. Coincident with the change in land surface and geography (Eoi280–E280) is a weakening of the combined annual mean overturning within the two Hadley cells (191 and 180×109 kg s-1 for E280 and Eoi280, respectively).

The wintertime subtropical jet (StJ; also known as the midlatitude jet) and polar jet (PJ) are shown within Fig. . We characterise the mean spatial envelope of the jet path by deriving, from 50 years of daily data, the days per season in which the mean mass-weighted flow speed integrated over 400–100 hPa (∼7–16 km) exceeds 30 ms-1. For both E280 (Fig. a–d) and E400 (not shown), we obtain a seasonal jet stream configuration which is consistent with the ERA-40 and derived results of . The PJ and the StJ streams can be difficult to differentiate as the former is latitudinally irregular, so following we use normalised wind shear as a height differentiator. The StJ stream path is more persistent and stable and so is characterised by the mean latitude of the StJ core which is shown within Table . The change in geography (Eoi280–E280) drives a poleward shift of the mean StJ latitude of ∼1.6∘ in the Northern Hemisphere (both seasons) and 2.2∘ in the Southern Hemisphere summer. The response to Pliocene CO2 (Eoi400–Eoi280) increase is weaker with a 0.8∘ poleward shift of the mean StJ latitude in the Northern Hemisphere (both seasons). The Southern Hemisphere mean StJ appears only weakly poleward shifting in response to Pliocene CO2 increase. Regionally, jet behaviour deviates from the global mean view. Within the North Atlantic, the PJ moves equatorward in response to the change in palaeogeography (Eoi280–E280) moving the jet stream mean path from northern to southern Europe (Fig. b vs. f). Synoptic storms grow and propagate along jet stream axes, and so this equatorward shift in the PJ likely contributes to the increase in rainfall seen in southern Europe during Pliocene wintertime (Fig. e vs. f).

Modelled mean annual SSTs (MASSTs) are detailed within Table  and Pliocene anomalies are shown within Fig. . We see a 0.8 ∘C warming due to the change in palaeogeography (Eoi280–E280) and a further 1.0 ∘C of warming due to the change in Pliocene CO2 (Eoi400–Eoi280). With increasing levels of CO2, regional patterns of MASST change due to palaeogeography are overprinted by CO2-induced warming which is most evident in the midlatitudes. The greatest warming occurs within the North Atlantic subpolar gyre where Eoi400–E280 reaches 9.3 ∘C. In the vicinity of the modern Gulf Stream and North Atlantic Drift, we find a cooling during DJF and MAM seasons of up to -4.9 ∘C within Eoi280–E280 (not shown). Investigation of surface ocean velocity vectors (not shown) suggests an intensification of the North Atlantic wind-driven subpolar gyre which appears to disrupt western intensification and the path of the Gulf Stream. The westerlies in the region appear to intercept the remnant Gulf Stream and divert it from a northeasterly to a more eastward path. This is seen as the warm tongue south of the extant Gulf Stream (Fig. ). A similar expression of MASST within the North Atlantic was seen by , and characteristic signatures may be present within other PlioMIP1 experiments (e.g. Fig. 1 of ). A persistent cooling is also found within the Barents Sea region coincident with the surface air temperature anomalies discussed within Sect. .

Table  also details the size of the global and component equatorial warm pools within the pre-industrial and Pliocene experiments. We see an expansion of the globally integrated warm pool with the change in palaeogeography (Eoi280–E280). This is evident in both the Western Hemisphere warm pool (WHWP) and Indo-Pacific warm pool (IPWP) regions. As expected, increased CO2 drives warm pool expansion under both modern and Pliocene geographic conditions.

A complex picture emerges in the sensitivity of seasonal sea ice distribution to geographic and CO2 changes as shown within Fig. . Within the Northern Hemisphere winter, the palaeogeography changes drive an equatorward expansion of sea ice in the Greenland Sea region. Increasing CO2 from 280 to 400 ppm counteracts some of this expansion. In the Southern Hemisphere, the palaeogeographical changes suppress sea ice extent significantly within the Weddell Sea and also eastward towards the Davis Sea in both summer and winter. Coincident with this suppression is an equatorward expansion of sea ice within the Bellingshausen Sea region. As we increase CO2, we see a general reduction in the sea ice extent and concentration in both summer and winter months. Within Eoi400 boreal summer, the Arctic is largely ice-free and the ice that is present is mostly <50 % concentration. During austral summer, the concentration of sea ice within the Pliocene is reduced in extent and more zonally asymmetric, concentrated within the Amundsen and Ross seas.

The mixed layer depth (MLD) for E280, Eoi280 and Eoi400 is shown within Fig. . We focus on deep convection, the principle mechanism of deep-water formation. Deep convection is highly localised, and therefore model representation is only suggestive. Nevertheless, E280 represents reasonably well the modern open-ocean deep convection that occurs within the Weddell and Ross seas (which form the main formation sites of AABW) and in the Labrador, Irminger and Greenland seas. All Pliocene experiments exhibit more widespread deep convection particularly within the Labrador and Norwegian seas, and near the Antarctic Peninsula island. In contrast to , we do not model any significant increase in Pliocene North Pacific MLD and hence no subsequent intensification of North Pacific Deep Water (NPDW) formation (Table  and Fig. ).

The Atlantic meridional overturning circulation (AMOC) streamfunctions for E280, Eoi280 and Eoi400 are shown within Fig.  and detailed within Table . The pre-industrial experiment E280 has a maximum AMOC strength at 26.5∘ N of 13.4±1.1 Sv. This compares reasonably well with the estimate of 17.2±4.6 Sv derived by using measurements from the RAPID array between April 2004 and October 2012. The all-latitude maximum in AMOC strength (AMOCmax) within E280 occurs at ∼650 m depth at 33.75∘ N with a strength of15.7±1.2 Sv.

We find an AMOC which is more intense in the Pliocene than in the pre-industrial, which is attributed to the Pliocene palaeogeography (Table ). The AMOCmax of Eoi400 is 19.6 ± 1.0 Sv and occurs at ∼650 m depth at 33.75∘ N. Multidecadal to centennial fluctuations within the spin-up phase are present within the Pliocene experiments but not within the pre-industrial experiment. In all Pliocene simulations, AMOCmax occurs within the 25–33.75∘ N zonal envelope and at a depth of ∼650 m. The Eoi400 AMOCmax lies within the 10–24.6 Sv range of PlioMIP1 , whilst the Eoi400–E280 AMOCmax anomaly of 3.9±1.6 Sv (Table ) lies at the upper end of the PlioMIP1 ensemble range of -0.9–3.6 Sv. Despite an intensification of the AMOC within the Pliocene experiments, we find that the overturning strength reduces poleward of ∼40∘ N driven by the changed land surface and bathymetry (Eoi280–E280). This is seen within cooling evident in Gulf Stream MASSTs of Fig. . Under increasing Pliocene CO2, the midlatitude overturning intensifies with a corresponding decrease in the Gulf Stream MASST cold anomaly. The overturning within the polar region is evidence of bottom water formation within the Nordic Seas. In E280, overturning extends to ∼80∘ N but is weaker than in the Pliocene experiments (which extends to ∼75∘ N). This is reflected within the geographic extent and intensity of deep convection shown within Fig. .

The Pacific meridional overturning circulation (PMOC) streamfunction is shown within Fig.  and detailed within Table , in which PMOC+ve reflects the strength of the subtropical gyre circulation, whilst PMOC-ve reflects the strength (and depth) of the Pacific Deep Water (PDW) and NPDW. Pliocene palaeogeography (Eoi280–E280) drives an intensification of both the subtropical gyre and PDW overturning, whilst increasing CO2 acts to weaken them. The Pliocene subtropical gyre (PMOC+ve) and PDW (PMOC-ve) overturning are stronger regardless of CO2 level (e.g. within Eoi400, PMOC+ve and PMOC-ve are 22 % and 6 % stronger than E280). With the change in palaeogeography (Eoi280–E280), the PDW shoals (from ∼4 to 3 km) and with increasing Pliocene CO2 the NPDW overturning reduces in northward reach, associated with the warming of North Pacific MASST (Fig. ).

The Antarctic Circumpolar Current (ACC) strength is detailed within Table  and shown within Fig. . We calculate the volumetric flow of the ACC at the Drake Passage across a 64.4–56.9∘ S, 65∘ W, transect using the positive aspect of the U component (zonal) of the total (barotropic and baroclinic) velocity. We find an overly intense ACC within E280 and E400 when compared against recent observations of 134–164 Sv . The overly intense ACC within HadCM3 has been identified previously. compared CMIP5 historical experiments to observations and found the model's ACC flow at the Drake Passage transect of 244.5±4.0 Sv compared unfavourably to observations and 155±51 Sv of the CMIP5 multi-model mean. This unrealistic intensity appeared to be driven, or at least connected to, an overly strong salinity gradient across the ACC, particularly towards low latitudes . This could be a consequence of the artificial freshwater correction field used within the CMIP5 historical and piControl experiments and E280 here.

Modelled ACC strength appears significantly reduced within the Pliocene experiments. Westerlies intensify in the Southern Hemisphere within the Pliocene but mostly in regions poleward of the Sub-Antarctic front (poleward of the ACC). The weakened Drake Passage throughflow is mirrored within the vertically integrated barotropic streamfunction. Care must be taken when interpreting ACC strength in situations of changed palaeogeography and island specification. The ACC is weakly stratified and vertically coherent and so is dominantly barotropic in nature. Within the Pliocene boundary conditions (Sect. ), the island peninsula is defined as a separate barotropic island (from the Antarctic continent), and this may be driving the Pliocene reduction in ACC strength. Also, given a more complex line-integral configuration, the model's barotropic solver may not be converging fully towards a solution. The change in island specification may also be responsible for the change in ACC geographical extent shown within Table . Defining the streamfunction cross section by the latitudes of the centroid and upper 50 % of zonal transport, we see that the change in geography (from E280 to Eoi280) drives a general latitudinal thinning of the ACC extent and an equatorward shift of its centroid.

Within the Pliocene experiments, the ACC runs mostly between the surface and sea floor between 60 and 57∘ S, whilst a deeper countercurrent is present closer to the peninsula. In the Pacific, a pronounced thinning of the ACC latitude extent is observed whereby the Sub-Antarctic front moves equatorwards (the subtropical front is mostly unchanged). With the Pliocene geography, there are suggestions that the Antarctic Coastal Current (the countercurrent to the ACC) flows between the peninsula island and the Antarctic land mass. There is uncertainty as smaller islands in this region are unrepresented within the model. Figure  also suggests a more continuous coastal current with the Pliocene palaeogeography, particularly between 180 and 90∘ E.

Here, we examine the sensitivity of the Pliocene climate to a different choice of orbital configuration (e.g. modern (default) vs. KM5C at 3.205 Ma). For Eoi400, there is no meaningful difference in global means (Table  MASAT, Table  MAP, Table  MASST and warm pool areal extent).

There is a statistically significant difference between orbEoi400 and Eoi400 AMOCmax (t(98)=7.20, p≪0.0001) and AMOC at 26.5∘ N (t(98)=11.36, p≪0.0001) using a two-sample t test assuming unequal variance (null hypothesis being there is no difference in the two time series of annual means). With regards to PMOC+ve, orbEoi400 and Eoi400 are deemed equivalent (t(98)=0.62, p=0.54), whilst for PMOC-ve, the two experiments are equivalent at the 95 % confidence level (t(98)=1.93, p=0.06). Centennial-scale fluctuations in Pliocene AMOCmax could potentially account for statistical differences between the climatological mean periods of orbEoi400 and Eoi400, as AMOCmax differences could simply reflect a lack of coherence introduced since the year 2000 fork point.

Section  identified the possibility of different TSI values being used within PlioMIP2 climate models. Here, we determine the sensitivity of HadCM3 within E280 and Eoi400 experiments to changing the TSI parameter. Reducing total solar insolation from 1365 to 1361 Wm-2 (-0.3 %) reduces the mean incoming solar (SW) radiation averaged over the entire Earth's surface by 1 Wm-2 (from 341.25 to 340.25 Wm-2). Table  accumulates climatological indices from E280 and Eoi400 under these two TSI values. Figure  shows the spatial pattern of climatological differences (Pliocene minus pre-industrial) for simulations based upon 1365 and 1361 Wm-2 for MASAT, MAP and MASST. Overall the patterns of climatological anomalies for the experiments using TSI of either 1361 or 1365 Wm-2 are very similar. In this sense, comparison of model temperature anomalies to proxy temperature anomalies should not generally be influenced by the choice of TSI.

However, in a similar way to the orbital configuration, AMOCmax does appear sensitive to TSI when we compare Eoi400 against 1361Eoi400 (t(98)=-13.3, p≪.0001) and E280 to 1361E280 (t(98)=2.47, p=0.015). It is possible that this sensitivity to TSI could be a consequence of the previously described AMOC cyclicity and lack of coherence between Eoi400 and 1361Eoi400.