NorESM-L is a low-resolution of NorESM1-M (the version of NorESM that contributed to the Coupled Model Intercomparison Project Phase 5, CMIP5) and is developed for paleoclimate simulations (Zhang et al., 2012). The atmospheric component has a horizontal resolution of T31 (∼3.75∘) and 26 vertical levels. The ocean component employs a bipolar grid with a horizontal resolution of nominal 3∘ and uses 30 isopycnic vertical layers (Table 1). NorESM-L was used to simulate the Pliocene climate in PlioMIP1 (Zhang et al., 2012). Further information on NorESM-L is documented in Zhang et al. (2012).

NorESM1-F is assembled for long time simulations with relatively high resolutions and improved process representations and climate performance compared to NorESM1-M (Guo et al., 2019). In the model, the atmosphere component uses a horizontal resolution of 1.9∘ latitude and 2.5∘ longitude and uses 26 vertical levels. The ocean component employs a tripolar grid with a nominal 1∘ horizontal resolution and uses 53 vertical layers (Table 1).

In NorESM1-F, the change in ocean–sea-ice grid from bipolar in NorESM1-M to a tripolar configuration together with a reduction in model complexity by replacing the comprehensive aerosol–cloud process representations in NorESM1-M with the standard prescribed aerosol chemistry of CAM4 (as was done in NorESM-L) lead to large speedup of the model. In addition, the reduction in the coupling frequency between atmosphere–sea-ice and atmosphere–land and the dynamic subcycling of the sea ice is helpful to improve the computational performance with a relatively small effect on the modeled climate (Guo et al., 2019).

Compared to NorESM1-M, there are some updates in the ocean physics in NorESM1-F. NorESM1-F takes measures to reduce sea ice thickness biases in shelf regions and modifies the parameterization of oceanic mesoscale eddies and the vertical mixing (Guo et al., 2019). With those updates to the ocean physics, NorESM1-F provides reasonable simulations of sea ice and AMOC compared to NorESM1-M (Guo et al., 2019). In the atmosphere component, the model adopts the formulation for energy updates and energy conservation, changes the air–sea flux calculation, and modifies the calculation of the solar zenith angle (Guo et al., 2019). Compared to NorESM1-M, the seasonal cycle of sea surface temperature in the equatorial Pacific is markedly improved with NorESM1-F. There are also several important improvements on how precipitation is simulated, e.g., improvements in seasonality, reduced wet bias, and mitigation of the common double intertropical convergence zone (ITCZ) problem. Further details on the model performance of NorESM1-F can be found in Guo et al. (2019).

According to the PlioMIP2 protocol (Haywood et al., 2016b), we use modern geographic boundary conditions, including modern land–sea mask, topography, and ice sheets and vegetation for the year 1850, in the preindustrial control experiments (Table 2). For NorESM-L, we set atmospheric CO2, N2O, and CH4 levels to the preindustrial values of 280 ppmv, 270 ppbv, and 760 ppbv, respectively. The orbital parameters apply values for the year 1950. For NorESM1-F, the default preindustrial atmospheric CO2, N2O, and CH4 levels are 284.7 ppmv, 275.68 ppbv, and 791.6 ppbv, respectively.

The preindustrial experiment with NorESM-L was run for 2200 years, and the experiment with NorESM1-F was run for 2000 years. Climatological means of the last 100  years were analyzed in this study.

Following PlioMIP2 experimental guidelines (Haywood et al., 2016b), we close the Hudson Bay, Bering Strait, and straits through the Canadian Arctic Archipelago in the Pliocene land–sea configuration. The atmospheric CO2 concentration is set to 400 ppmv. Atmospheric N2O and CH4 concentrations, the solar constant, and orbital parameters are identical to preindustrial values (Table 2).

We use the “anomaly method” recommended in PlioMIP2 to create the paleogeography for the Pliocene experiment. We first calculate differences between the PRISM4 Pliocene and PRISM4 modern topography and interpolate these to a T31 resolution for NorESM-L and to a 1.9∘×2.5∘ resolution for NorESM1-F. Then, we add the interpolated topography anomalies to modern topography in the preindustrial experiment.

To create vegetation in the Pliocene experiment, we first interpolate PRISM4 Pliocene vegetation to the resolution for NorESM-L and for NorESM1-F. Then, we convert biome vegetation types to LSM (land system model) vegetation types following the procedure outlined by Rosenbloom (2009). Lakes for the Pliocene are prescribed by adding the PRISM4 lake area anomaly to modern conditions. Pliocene soil conditions remain the same as the preindustrial conditions.

With NorESM-L, the Pliocene experiment was run for 1200 years. With NorESM1-F, the Pliocene experiment was first spun up for 2000 years with atmospheric CO2 concentration set to 400 ppmv and then run for 500 years forced with all Pliocene boundary conditions. Climatological means of the last 100 years were used for analysis.

Relative to the preindustrial experiments, the simulated Pliocene climate is warmer according to both NorESM versions (Fig. 1). The global annual mean surface air temperature (SAT) increases by 1.7 and 2.1 ∘C under the NorESM1-F and NorESM-L Pliocene simulations, respectively. In particular, stronger warming appears at high latitudes (Fig. 1 and Table 3). The simulated Pliocene annual mean SAT increases by 5.2 ∘C (NorESM1-F) and 4.9 ∘C (NorESM-L) at the northern high latitudes and by 3.2 ∘C (NorESM1-F) and 7.6 ∘C (NorESM-L) at the southern high latitudes. Weak cooling appears in tropical Africa, India, northeastern Asia, northern Australia, and the southern Pacific close to western Antarctica, under the NorESM1-F and NorESM-L Pliocene simulations (Fig. 1a, b). Both NorESM1-F and NorESM-L simulate stronger warming over land than over ocean. Relative to the preindustrial period, the simulated Pliocene global mean surface air temperature (SAT) over land increases by 2.3 ∘C with NorESM-L and 2.0 ∘C with NorESM1-F, which is notably larger than the warming over ocean (2.0 and 1.6 ∘C for the NorESM-L and the NorESM1-F, respectively). This stronger warming over land is a common feature in most PlioMIP2 simulations. However, the simulated zonal mean SAT over land is nearly twice as large as in the ocean at the northern high latitudes (Fig. S1 in the Supplement).

Changes in seasonal SAT follow a similar pattern to those of the annual SAT. The two models generate strong seasonal warming in Circum-Arctic regions, e.g., Hudson Bay, the Canadian Arctic Archipelago, and Greenland, in the Pliocene experiment. However, NorESM-L produces larger Pliocene seasonal warming over the Southern Ocean and Antarctica than NorESM1-F.

The simulated Pliocene global mean annual precipitation increases by 0.14 mm d-1 according to the two NorESM versions (Fig. 2). The mean annual precipitation increases largely in the tropical regions and especially in monsoon regions of North Africa, Asia, and Australia. In subtropical regions, the zonal mean annual precipitation does not change markedly or slightly decreases. The changes in seasonal precipitation generally follow the pattern of annual precipitation. Both models suggest that the ITCZ shifts northward in the Atlantic and in Africa in the boreal summer but does not change considerably in the boreal winter.

Both NorESM1-F and NorESM-L simulate higher sea surface temperatures (SSTs) in the Pliocene experiments compared to preindustrial experiments (Fig. 3). The simulated Pliocene global mean annual SST is 1.2 ∘C (NorESM1-F) and 1.5 ∘C (NorESM-L) higher than preindustrial levels (Fig. 3 and Table 3). Large increases in SST appear at the high latitudes of both hemispheres (Fig. 3 and Table 3). The simulated Pliocene annual mean SST increases by 2.4 ∘C (NorESM1-F) and 2.1 ∘C (NorESM-L) in the middle and high latitudes of the North Atlantic (north of 30∘ N) and by 1.4 ∘C (NorESM1-F) and 2.4 ∘C (NorESM-L) in the Southern Ocean. A slight cooling occurs in the South Pacific close to western Antarctica according to the Pliocene experiments for both NorESM model versions. The seasonal change in SST is largely consistent with the annual pattern.

Changes in sea surface salinity (SSS) from the Pliocene experiments differ notably between NorESM1-F and NorESM-L. The global mean SSS decreases by 0.34 g kg-1 in the NorESM1-F Pliocene simulation, while it increases by 0.16 g kg-1 according to the NorESM-L Pliocene simulation (Fig. 4). With NorESM1-F, Pliocene SSS generally decreases in most oceans, except for Baffin Bay, the Labrador Sea, and the North Atlantic subpolar regions. With NorESM-L, Pliocene SSS increases slightly in most oceans, while it decreases in the Indian Ocean and in the Arctic.

The divergent responses in SSS in NorESM1-F and NorESM-L are most likely to be associated with the different vertical redistribution of salt in the two models, due to differences in, e.g., surface layer mixing, ocean ventilation, convection, and circulation. The two models have different vertical resolutions and horizontal or vertical mixing schemes. When the drift in global mean SSS is removed, NorESM-L and NorESM1-F show similar regional anomalies (Fig. S2). Both versions show that the SSS contrast among the Indian Ocean, the Arctic, and the rest of the oceans is intensified in the Pliocene experiment (Fig. S2).

The simulated Pliocene sea ice with NorESM1-F and NorESM-L is reduced both in terms of its thickness and its extent (Fig. 5). In the Northern Hemisphere, the simulated Pliocene sea ice thickness in the Arctic Ocean is reduced by 0.5 to 1 m in March and by more than 1 m in September. In September, NorESM1-F simulates an almost ice-free Arctic in the Pliocene experiment, while sea ice remains in the central Arctic according to the NorESM-L Pliocene simulation. In March, sea ice becomes thinner while still covering most of the Arctic Ocean in both models. In the Southern Hemisphere, although both models generate retreated sea ice extents for the Southern Ocean, NorESM-L simulates the larger sea ice responses. The Southern Ocean becomes almost ice-free specifically in March according to the NorESM-L Pliocene simulation.

The simulated Pliocene AMOC becomes stronger and deeper in both models compared to the preindustrial climate. With NorESM1-F, the maximum AMOC is 28.1 Sv in the Pliocene experiment, increasing by about 15 % (Fig. 6 and Table 4). With NorESM-L, the simulated maximum AMOC is 23.3 Sv in the Pliocene experiment, which is 2 Sv (about 9 %) larger than that in the preindustrial experiment. Both models suggest that the vertical extent of the AMOC cell penetrates deeper during the Pliocene relative to the preindustrial period (Fig. 6 and Table 4). Compared to the preindustrial period, Pliocene SSS increases in Baffin Bay, the Labrador Sea, and the North Atlantic subpolar gyre in both models (Fig. 4), reducing the surface stratification and tending to favor more open-ocean convection, thereby potentially contributing to the strengthened AMOC.

In the Pacific and Indian oceans, meridional overturning circulation is slightly stronger in the Pliocene experiments than in the preindustrial experiments. As for the shallower circulation in the North Pacific subtropical gyre, the maximum of the circulation is increased by 1.0 Sv with NorESM1-F (22.3 Sv for the preindustrial experiment vs. 23.3 Sv for the Pliocene experiment) and by 0.9 Sv with NorESM-L (30.7 Sv for the preindustrial experiment vs. 31.6 Sv for the Pliocene experiment) (Fig. 6 and Table 4). As for the deeper circulation associated with Pacific Deep Water, the strength is increased by 0.8 Sv with NorESM1-F (-13.6 Sv for the preindustrial experiment vs. -14.6 Sv for the Pliocene experiment) and by 4.6 Sv with NorESM-L (-17.0 Sv for the preindustrial experiment vs. -21.6 Sv for the Pliocene experiment) (Fig. 6 and Table 4). In the Pliocene experiments, both NorESM versions simulate an extended northward penetration of deep water.

Although the two versions of the NorESM simulate similar Pliocene climates, they still exhibit some differences. The most significant differences appear in the Southern Ocean (Fig. 1). SST increase over the Southern Ocean is ∼1 ∘C larger with NorESM-L than with NorESM1-F in the Pliocene experiments. This difference between the two model versions is likely to be associated with different responses in ocean heat transport and sea ice. Simulated Pliocene southward ocean heat transport to the Southern Ocean is reduced according to NorESM1-F but increased according to NorESM-L (Fig. 7a and c), which partly explains the reduction in the Southern Ocean sea ice extent being more pronounced for NorESM-L than it is according to NorESM1-F (Fig. 5). Pliocene (austral summer) sea ice is nearly absent according to NorESM-L, while it still covers part of the Southern Ocean according to NorESM1-F. On the one hand, the larger seasonal warming in the Southern Ocean favors less sea ice extent in the Pliocene experiment simulated with NorESM-L. On the other hand, the presence of less sea ice leads to a reduction in albedo and to a more active ocean–atmosphere interaction (e.g., ice–albedo feedback) and contributes to the higher levels of Southern Ocean warming in the Pliocene experiment simulated with NorESM-L.

NorESM-L also simulates increased ventilation in the Southern Ocean, while NorESM1-F does not. As is indicated by the changes in salinity, sea water over the Southern Ocean becomes less stratified according to the NorESM-L Pliocene simulation (Fig. 8). The weakened ocean stratification allows the Southern Ocean to be well ventilated. As a result, the simulated Pliocene deep water is much younger in the Southern Ocean (Fig. S3). This well-ventilated Southern Ocean also appeared from the PlioMIP1 simulation with NorESM-L (Zhang et al., 2013b). However, with NorESM1-F, simulated Pliocene Southern Ocean stratification appears similar to that simulated in the preindustrial experiment. Such divergent responses in Southern Ocean stratifications also appeared in the PlioMIP1 simulations (Zhang et al., 2013a). It remains difficult to fully explain the divergent responses. The explanation is likely related to the updated ocean physics and/or higher resolution in NorESM1-F when compared to NorESM-L.

The other remarkable differences in the Pliocene simulations with the two versions appear in the northern middle and high latitudes. Compared to NorESM-L, the increase in the Pliocene SST simulated with NorEMS1-F at the northern middle (high) latitudes is 0.2 ∘C (0.1 ∘C) larger (Fig. 3 and Table 3). The increase in the Pliocene annual mean SAT at the northern middle (high) latitudes with NorESM1-F is 0.2 ∘C (0.3 ∘C) larger (Fig. 1 and Table 3). The stronger Pliocene warming at the northern high latitudes is most likely related to the mechanism responsible for the larger responses in sea ice reduction with NorESM1-F since the clear-sky albedo, particularly in sea ice regions, dominates the high-latitude warming in the Pliocene (Hill et al., 2014). In association with the larger salinity increase in the northern North Atlantic (Fig. 4), the enhancement of AMOC is larger with NorESM1-F than with NorESM-L (∼15 % vs. ∼9 %), which favors the larger increases in the Pliocene northward ocean heat transport to the Atlantic with NorESM1-F (Figs. 6 and 7). Correspondingly, less sea ice simulated in the Pliocene experiment contributes to a larger warming at the high latitudes with NorESM1-F than with NorESM-L through the ice–albedo feedback (Figs. 1 and 3).

With NorESM-L, the simulated Pliocene surface temperatures in PlioMIP2 are slightly cooler than those simulated in PlioMIP1. Simulated increases in global annual mean SAT and SST are 1.1 and 0.43 ∘C less in PlioMIP2 than those generated in PlioMIP1 (Fig. 9). At the northern middle and high latitudes, the simulated increases in annual mean SAT (SST) are 1.7 and 5.0 ∘C (1.0 and 0.7 ∘C) weaker, respectively, in PlioMIP2 than in PlioMIP1 (Fig. 9, and Table 5). Simulated weaker levels of Pliocene warming also appear in the PlioMIP2 experiment with HadCM3, according to which the Pliocene annual mean SAT is 0.4 ∘C cooler in PlioMIP2 (Hunter et al., 2019) than in PlioMIP1 (Bragg et al., 2012). However, MRI-CGCM2.3, CCSM4, and IPSL-CM5A all simulate a larger warming in the Pliocene experiments in PlioMIP2 (Kamae et al., 2016; Chandan and Peltier, 2017; Tan et al., 2020) than those in PlioMIP1 (Contoux et al., 2012; Kamae and Ueda, 2012; Rosenbloom et al., 2013).

The simulated weaker Pliocene warming is attributable to the modified boundary conditions used in PlioMIP2. Relative to PlioMIP1, the Pliocene atmospheric CO2 is 5 ppmv less in PlioMIP2 (400 ppmv in PlioMIP2 vs. 405 ppmv in PlioMIP1). The slight reduction in Pliocene atmospheric CO2 is not likely to induce a significant effect on the simulated Pliocene warming weakness. Compared to PlioMIP1, the seaways at the northern high latitudes (including the Bering Strait and Canadian Arctic Archipelago straits) are closed under the PlioMIP2 boundary conditions. Previous sensitivity investigations have demonstrated that the closure of the Bering Strait favors the formation of North Atlantic deep water and the intensification of AMOC by reducing freshwater transport from the Pacific to the Arctic Ocean and the North Atlantic (e.g., Hu et al., 2010; Brierley and Fedorov, 2016; Otto-Bliesner et al., 2017). In contrast, the closure of Canadian Arctic Archipelago Straits tends to weaken AMOC, by increasing fresh water transport through the Fram Strait and thereby leading to the freshening and subsequent cooling of the Labrador and Greenland–Iceland–Norwegian seas (Otto-Bliesner et al., 2017). However, the combined effects of closing these two seaways are complicated and appear to be model dependent. For example, with the Bering Strait and Canadian Arctic Archipelago Straits closed together, Pliocene ocean surface warming in the North Atlantic high latitudes is enhanced under the CCSM4 (Otto-Bliesner et al., 2017) and IPSL-CM5A simulations (Tan et al., 2020) but is weakened according to NorESM-L.