The climate of the Last Glacial Maximum (LGM; ∼ 21 000 years ago) has been a focus of the Paleoclimate Modelling Intercomparison Project (PMIP) since its inception. It is the most recent global cold extreme and as such has been widely documented and used for benchmarking state-of-the-art climate models (Braconnot et al., 2012; Harrison et al., 2014, 2015). The increase in global temperature from the LGM until now (∼ 4 to 6 ∘C; Annan and Hargreaves, 2015; Friedrich et al., 2016) has been the same order of magnitude as the increase projected by 2100 CE under moderate-to-high emission scenarios. The LGM world was very different from the present one, with large ice sheets covering northern North America and Fennoscandia, in addition to the Greenland and Antarctic ice sheets still present today. These additional ice sheets resulted in a lowering of the global sea level by ∼ 120 m, which induced changes in the land–sea distribution. The closure of the Bering Strait and the exposure of the Sunda and Sahul shelves between southeast Asia and the Maritime Continent are the most prominent of these changes in land–sea geography. Atmospheric greenhouse gas (GHG) concentrations were lower than pre-industrial (PI) values, leading to cooling in addition to that induced by the large ice sheets. The cooling is more pronounced in the high latitudes than in the tropics and greater over land than ocean. The polar amplification and the land–sea contrast signals simulated by the previous generation of paleoclimate simulations (PMIP3 – Coupled Model Intercomparison Project; PMIP3-CMIP5) are similar in magnitude (although opposite in sign) to the signals seen in future projections and have been shown to be consistent with climate observations for the historic period and reconstructions for the LGM (Braconnot et al., 2012; Izumi et al., 2013; Harrison et al., 2014, 2015). However, while the models are able to represent the thermodynamic behaviour that gives rise to these large-scale temperature gradients, they underestimate cooling on land, especially winter cooling, and overestimate tropical cooling over the oceans (Harrison et al., 2014). Thus, one question to be addressed with the new PMIP4-CMIP6 simulations is whether there is any improvement in capturing regional temperature changes. The large temperature changes during the LGM compared to the pre-industrial period make this interval a natural focus for efforts to constrain climate sensitivity but attempts to do this using the PMIP3-CMIP5 simulations were inconclusive (Schmidt et al., 2014; Harrison et al., 2014), in part because of the limited number of LGM simulations available and in part because of the limited range of climate sensitivity sampled by these models. Changes in model configuration have resulted in several of the PMIP4-CMIP6 models having substantially higher climate sensitivity than the PMIP3-CMIP5 versions of the same models, and thus the range of climate sensitivity sampled by the PMIP4-CMIP6 models is much wider. This provides an opportunity to re-examine whether the LGM could provide a strong constraint on climate sensitivity (Renoult et al., 2020; Zhu et al., 2021).

The atmospheric general circulation was strongly modified from its modern-day conditions by changes in coastlines at low latitudes (DiNezio and Tierney, 2013) and by the presence of the Laurentide and Fennoscandian ice sheets (e.g. Laîné et al., 2009; Löfverström et al., 2014, 2016; Ullman et al., 2014; Beghin et al., 2015; Liakka and Löfverström, 2018). These changes in circulation had an impact on precipitation, which was reduced globally (Bartlein et al., 2011) but increased locally, for example, in southwestern North American and in the Mediterranean region (e.g. Kirby et al., 2013; Beghin et al., 2016; Goldsmith et al., 2017; Lora et al., 2017; Lora, 2018; Löfverström and Lora, 2017; Löfverström and Liakka, 2016; Löfverström, 2020; Rehfeld et al., 2020). The interplay between temperature-driven and circulation-driven changes in regional precipitation during the LGM represents a test of the ability of state-of-the-art models to simulate precipitation changes under future scenarios, where both thermodynamic (e.g. related to the Clausius–Clapeyron relationship) and dynamic (e.g. related to changes in the position of the storm tracks and extent of the subtropical anticyclones) effects contribute to changes in the amount and location of precipitation (e.g. Boos, 2012; Scheff and Freirson, 2012; Lora, 2018). Evaluation of the PMIP3-CMIP5 simulations showed that models underestimate the LGM reduction in mean annual precipitation over land (Harrison et al., 2014), reflecting the underestimation of temperature changes in the simulations (Li et al., 2013). This resulted in an underestimation of the observed aridity (precipitation minus evapotranspiration). While the models reproduced circulation-induced changes in precipitation in western North America, they showed no increase in precipitation south of the North American ice sheet and only limited impact on the precipitation of the circum-Mediterranean region (Harrison et al., 2014; Lora, 2018; Morrill et al., 2018). Thus, one question to be addressed with the new PMIP4-CMIP6 simulations is whether there is any improvement in capturing regional precipitation changes. One complication here is that most of the reconstructions used to evaluate the PMIP3-CMIP5 simulations were pollen based and relied on statistical approaches that do not account for the direct impact of low CO2 on water-use efficiency (Prentice and Harrison, 2009; Gerhardt and Ward, 2010; Bragg et al., 2013; Scheff et al., 2017) and could therefore be dry biased. However, new methods have been developed that account for this effect (Prentice et al., 2017), and thus it is possible to determine whether accounting for the effect of low CO2 resolves model–data mismatches in regional precipitation at the LGM.

The LGM boundary conditions also had a strong impact on ocean circulation, as documented via multiple tracers (e.g. Lynch-Stieglitz et al., 2007; Jaccard and Galbraith, 2011; Böhm et al., 2015), which suggest a shallower North Atlantic Deep Water (NADW) cell and expanded Antarctic Bottom Water (AABW). In addition, Gebbie (2014) used a combination of synthesis of multiple tracers measured in sediment cores for the LGM and a global tracer transport model to show that these tracers are compatible with a vertical distribution of NADW and AABW similar to today but that the core of the NADW water mass shoals by 1000 m. None of these proposed reconstructions of glacial circulation are consistent with the PMIP3-CMIP5 model results (Muglia and Schmittner, 2015), which all show a deepening of the Atlantic Meridional Overturning Circulation (AMOC), with NADW reaching the ocean floor in the northern North Atlantic for some models. Previous studies show that this increase in AMOC is related to changes in northern extratropical wind stress due to the presence of the high ice sheets (Oka et al., 2012; Muglia and Schmittner, 2015; Klockmann et al., 2016; Sherriff-Tadano et al., 2018; Galbraith and de Lavergne, 2019). Thus, the simulation of the AMOC, and ocean circulation in general, during the LGM could be highly sensitive to the ice-sheet reconstructions used as boundary conditions (see, e.g. Ullman et al., 2014; Beghin et al., 2016). There is still some uncertainty about the height and shape (although not the extent) of the LGM ice sheets, so the protocol for the LGM PMIP4-CMIP6 experiment takes this uncertainty into account by allowing for alternative ice-sheet configurations (Kageyama et al., 2017) in order to test the sensitivity of LGM climate and ocean circulation to ice-sheet configuration. The PMIP4-CMIP6 LGM experimental protocol also includes changes in other forcings, including vegetation changes and changes in atmospheric dust loadings and their uncertainties. Thus, the new PMIP4-CMIP6 simulations provide opportunities to examine the response of the climate system to multiple forcings, to calculate the impact of individual forcings through sensitivity experiments and to investigate how these forcings combine to produce circulation and climate changes in the marine and terrestrial realms.

In this paper, we present preliminary results from the PMIP4-CMIP6 LGM simulations, compare them to the PMIP3-CMIP5 results (Sect. 3) and evaluate their realism against a range of climatic reconstructions (Sect. 4). We focus on temperature and precipitation, extratropical circulation, energy transport and the AMOC.

The evaluation of the PMIP3-CMIP5 LGM simulations showed that large-scale climate features, such as the ratio of changes in land–sea temperature, high-latitude temperature amplification and precipitation scaling with temperature, were broadly consistent with modern observations (Braconnot et al., 2012; Izumi et al., 2013; Harrison et al., 2014, 2015).

All PMIP3 and PMIP4 models simulate larger cooling over land than over oceans, on average for the tropics and for the globe. Figure 10 shows averages of model output sampled at sites for which there are reconstructions compared to the averages of the reconstructed values. Since the different reconstructions do not cover the same sites, the averages of the model values at reconstruction sites differ slightly for each dataset. However, for all datasets, the multi-model relationship between the average cooling over land and that over the ocean is approximately linear. Figure 10 allows a comparison between model output and reconstructions averaged over land and over oceans, as well as a comparison of the ratio of the land cooling over the ocean cooling. Although the Cleator et al. (2020) dataset has a larger spatial coverage than the Bartlein et al. (2011) dataset, there is no significant difference between the two datasets for most of the temperature variables across common grid cells (Fig. 9). However, the new reconstructions have a reduced range at the warm end, especially between 0 and 40∘ N, so that for the averages over the tropics, most simulations are recorded as within or warmer than the land-based reconstructions, while they are within or colder than the Bartlein et al. (2011) reconstructions (Fig. 10, left-hand side). The results for the global averages are fairly consistent for both land-based reconstructions (Fig. 10, right-hand side) but the uncertainty is smaller for the Cleator et al. (2020) dataset. All in all, there are as many simulations within the range of globally averaged reconstructed temperatures of Cleator et al. (2020) as that of Bartlein et al. (2011) but the models outside this range tend to be on the warm side for the Cleator et al. (2020) dataset and both on the warm and cold sides for the Bartlein et al. (2011) dataset.

The reconstructions of the LGM–Late Holocene SST anomalies provided by Tierney et al. (2020) are colder than the MARGO reconstructions in the tropics, and although this removes the apparent cold-bias shown by some simulations, this results in some simulations falling outside the window of reconstructed SSTs at the warm end. This is even more the case if we compare the results to the data-assimilated product from Tierney et al. (2020), which has a global coverage (Fig. 10, left-hand side, bottom line). The results from this latter dataset contrast the results from other global products, as shown in Supplement Fig. S3. These other global datasets (Annan and Hargreaves, 2013; Kurahashi-Nakamura et al., 2017; GLOMAP2020 from Paul et al., 2021) are all derived, at least in part, from the MARGO Project Members (2009) dataset, which might explain their overall consistency with the averages estimated by MARGO Project Members (2009). We have added the warmest estimate (GLOMAP2020) which includes uncertainties in Fig. 10 to obtain a more complete view of the available reconstructions as of 2020, keeping in mind that the authors of the GLOMAP2020 reconstruction estimate that it could be biased by 0.5 to 1.0 ∘C in the tropics. The simulated mean annual surface air temperature decreases over the tropical oceans stand between these two extremes. This illustrates that model-related uncertainties are comparable with the uncertainties raising from the multiple approaches taken to reconstruct both the continental and oceanic temperatures.

The ratio for the land–sea difference in changes in mean annual temperature in the tropics in the PMIP4 simulations is compatible with the ratio reconstructed from the Bartlein et al. (2011) and MARGO Project Members (2009) datasets. This is also the case if we consider the more recent reconstructions by Cleator et al. (2020) and Tierney et al. (2020), although the multi-model land–sea ratio appears to be smaller than that suggested by the reconstructions. This is the case for both the tropical and global averages. However, it would not be compatible with a land–sea contrast based on the Cleator et al. (2020) dataset and the GLOMAP2020 dataset, even if the warm bias pointed by its authors is taken into account. We are therefore left with large uncertainties on the topic of LGM cooling over land and oceans, from the reconstructions as well as from the models. The uncertainties based on the ensemble of model results and on the ensemble of continental and marine reconstructions are actually very similar.

The amplification of temperature changes at high northern latitudes compared to the tropics is apparent over both the land and the ocean domains, although the amplification appears to be smaller in the new data syntheses (Fig. 11), except of the Tierney et al. (2020) data-assimilated product. For the ocean domain, this could reflect the influence of seasonal production on the extratropical sites, with indicators being more sensitive to summer changes or to changes in the seasonal production cycle. Comparisons of the amplification over land areas with the Bartlein et al. (2011) dataset suggest that the simulated tropical cooling is too large in the PMIP3-CMIP5 simulations, whereas the extratropical cooling was both larger and smaller than that suggested by the reconstructions in both ensembles. Simulated tropical temperatures are more consistent with or warmer than the Cleator et al. (2020) reconstructions, suggesting that the apparent overestimation of tropical cooling in the PMIP3-CMIP5 simulations over land may reflect the paucity of tropical data points in Bartlein et al. (2011). However, the discrepancies between the simulated and reconstructed extratropical land temperatures are still present: there are several PMIP3 and PMIP4 simulations that are much colder than the reconstructions and many which are warmer than the reconstructions. Although polar amplification is more muted over the ocean domain, the comparisons show a similar picture to the land-based comparisons. Simulated tropical ocean temperatures are more compatible with the Tierney et al. (2020) than the MARGO Project Members (2009) synthesis. Simulated extratropical temperature changes in the PMIP3-CMIP5 ensemble mean are considerably colder than those shown by either of these syntheses, but most tend to be on the warm side of the Tierney et al. (2020) data-assimilated product.

The LGM climate is characterised by an increase in temperature seasonality in extratropical regions, with larger changes in winter than in summer (Izumi et al., 2013). This is confirmed by the Cleator et al. (2020) reconstructions. In general, this change in seasonality is reproduced by the models, although the ranges of PMIP4 results for winter are less distinct from their summer counterparts than for the PMIP3 models. The multi-model average seasonality is, however, increased for both ensembles. The simulated cooling in winter temperature is smaller than that indicated by the Bartlein et al. (2011) reconstructions (Fig. 12, top line). This is not the case compared to the Cleator et al. (2020) reconstructions, with which more models are in agreement, except for western Europe, which remains a region of model–data discrepancy. The magnitude of the summer cooling is more consistent between the PMIP4 simulations and the Cleator et al. (2020) reconstructions than between the PMIP3 simulations and the Bartlein et al. (2011) reconstructions in North America, Europe and extratropical Eurasia. Finally, the North Atlantic mean annual cooling simulated by the PMIP4 models spans a larger range that of the PMIP3 ensemble. While the PMIP3 ensemble mean showed temperatures within or above the reconstructed ranges from all oceanic datasets, the PMIP4 ensemble average stands within the range of the reconstructions, with six individual models being within this range, four above and three below. It is therefore quite difficult to determine the cause of the discrepancy in western Europe winter temperatures, which was previously assigned to an underestimation of the North Atlantic cooling. Some of the PMIP4 simulations are in fact much colder over both the North Atlantic and western Europe, and could be studied to further disentangle this model–data disagreement.

Regional changes in the tropics (Fig. 12, bottom line) are more muted than those in the northern extratropics, and seasonality differences are small. We therefore base our comparisons on the mean annual temperature and mean annual precipitation. Both PMIP3 and PMIP4 multi-model averages underestimate MAT cooling over tropical America, which is consistent for both reconstructions. More PMIP4 results stand within the reconstructed range over tropical America. Over tropical Africa, the PMIP3 models were broadly consistent with the Barltein et al. (2011) reconstructed MAP anomaly but underestimate this cooling if we refer to the more recent Cleator et al. (2020) reconstructions. This is also the case for the PMIP4 models, but four simulations (IPSLCM5A2 and the three HadCM3 simulations) are now within the reconstructed range. This is probably related to the simulated tropical SSTs being colder in these simulations. The reconstructed changes in tropical precipitation over America are larger in the Cleator et al. (2020) dataset than in Bartlein et al. (2011), and both PMIP3 and PMIP4 models underestimate the reconstructed drying (Fig. 12). The PMIP4 models, however, all simulate the correct negative sign of the reconstructed precipitation change. There is a large difference between the estimates of precipitation change given by the Bartlein et al. (2011) and the Cleator et al. (2019) datasets for tropical Africa, with the Cleator et al. (2020) reconstructions reducing the drying reconstructed by Bartlein et al. (2011). The ranges of PMIP3 and PMIP4 results are broadly similar over this region, and there are the same number of models (four) within the reconstructed range of Cleator et al. (2020), while no model result was compatible with the Bartlein et al. (2011) reconstructions. All other models underestimate the change or even simulate an increase in precipitation. All in all, the simulated changes in precipitation are therefore more consistent with the newer dataset. Thus, there is no systematic improvement in the simulation of tropical climates between the PMIP4 and PMIP3 ensembles.

The six data syntheses can be used to try and constrain the global MAT change from LGM to PI. There is a good correlation between the change in global average MAT over the reconstruction grid points and computed taking all the model grid points into account (Fig. 13). The idea here is therefore to take advantage of this relationship to obtain a range in the global MAT anomaly from the reconstructions. There are models with results below, within and above the average of all of the reconstructions, except MARGO Project Members (2009) and GLOMAP2020, for which no model simulates MAT LGM–PI anomalies above the reconstructed range of values. Retaining only the models which produce changes in MAT consistent with the reconstructions (and reconstruction uncertainty), the globally averaged change in MAT is between -5.7 and -3.7 ∘C using the Bartlein et al. (2011), between -6.7 and -4.6 ∘C using the Cleator et al. (2020) datasets, between -4.7 and -3.3 ∘C for the Tierney et al. (2020) dataset and above -3.9 and -4.4 ∘C for the MARGO Project Members (2009) and GLOMAP2020 datasets, respectively. Taken altogether, these estimates span a larger range than previous estimates, which indicate changes in MAT of between 4 and 6 ∘C (Annan and Hargreaves, 2015; Friedrich et al., 2016).

The results from the PMIP4 models differ from those of the PMIP3 ensemble in several ways. The multi-model global cooling is similar in both ensembles but the PMIP4 ensemble range is larger, with four simulations showing colder results than the coldest PMIP3 model. This feature mainly arises from the Southern Hemisphere extratropics, and it is currently difficult to disentangle whether it is due to the PMIP4 model ensemble being largely different from the PMIP3 ensemble or the changes in protocol from PMIP3 to PMIP4 (see also Zhu et al., 2021). The change in the ice sheets appears to have an impact on atmospheric circulation over North America and the North Atlantic. The AMOC increases less in the PMIP4 than in the PMIP3 simulations, and the depth of the NADW cell remains more stable, except for two models, in contrast with more than half the models of the PMIP3-CMIP5 ensemble, which simulated a large deepening of this cell. This could be due to the changes in atmospheric circulation over the North Atlantic, as well as changes in the North Atlantic freshwater balance. Changes in precipitation are generally similar for the PMIP3 and PMIP4 ensembles and characterised by less precipitation overall. Reduced evaporation due to colder temperatures partially compensates for the reduction in precipitation, so that areas of negative and significant LGM–PI anomalies in net precipitation (i.e. precipitation minus evaporation) are larger than areas with positive LGM–PI precipitation anomalies. However, both precipitation and net precipitation changes show large spatial heterogeneity and different regional-scale patterns of change between the PMIP4 and PMIP3 ensembles, which appears to be related to the performance of the model ensemble for PI. Additional sensitivity experiments are needed to separate the effects of changes in model configuration and sensitivity on general circulation features, such as the position of the jet streams, from the effects of differences in boundary conditions, such as the improved realism of the ice-sheet configuration.

The PMIP4-CMIP6 ensemble confirms that the models simulate large-scale thermodynamic behaviour common to historical and future simulations, such as land–sea contrast and polar amplification. The results from PMIP3 and PMIP4 align on the same relationships for these large-scale characteristics of climate change. The new reconstructions of Tierney et al. (2020) and Cleator et al. (2020) are in better agreement than with the reconstructions from Bartlein et al. (2011) and MARGO Project Members (2009) used to evaluate these features previously (Braconnot et al., 2012; Izumi et al., 2013; Harrison et al., 2014, 2015) for the tropical and global averages. However, global reconstructions of the surface ocean temperatures, such as the Tierney et al. (2020) data-assimilated product, the GLOMAP2020 data by Paul et al. (2021) and the reconstructions by Annan and Hargreaves (2013) and Kurahashi-Nakamura et al. (2017), show a wide range of results in terms of tropical temperatures (from -1 to -4 ∘C), which prevents firm conclusions on the model–data comparisons based on such global reconstructions. Interestingly, evaluating the uncertainty on tropical cooling from the PMIP model ensemble on the one hand and from the ensemble of continental and marine reconstructions on the other yields very similar results.

The simulated global change in MAT averaged over all the grid cells where reconstructions are available is well correlated with the global average on all model grid points, providing a constraint on the value of the global LGM cooling compared to PI. Using the terrestrial datasets as a constraint indicates a global cooling between -6.7 and -3.7 ∘C, while using Tierney et al. (2020) as a constraint indicates a global cooling of -4.9 to -3.2 ∘C, and using the MARGO Project Members (2009) and GLOMAP2020 datasets constrains the global average to be above -3.9 and -4.4 ∘C, respectively. The constrained range (-6.7 to -3.2 ∘C) is larger than previous estimates (-4 to -6 ∘C).

There is no obvious improvement in model performance at a regional scale between the PMIP3 and PMIP4 ensembles. In some cases (e.g. summer temperature over western Europe and extratropical Asia), the PMIP4 ensemble demonstrates a better ability to capture the changes depicted by the reconstructions; in some others (e.g. winter temperatures over Europe, mean annual precipitation over tropical America), the PMIP4 ensemble is still far from the reconstructed values.

Our analyses present a first picture of the PMIP4 LGM experiments. Results from CMIP6 models with high climate sensitivity have only recently become available (Zhu et al., 2021) and will need to be considered in a full assessment of the PMIP4-CMIP6 simulations. Sensitivity experiments, for example, to different ice-sheet configurations, are needed to disentangle the impact of model improvements from those related to using more realistic boundary conditions. Additional planned simulations will also help to disentangle the impacts of changes in vegetation and aerosol loading on the LGM climate. A more systematic evaluation of the simulated climates, using a wider range of paleoenvironmental data, will be helpful in understanding why there are persistent mismatches between the simulations and reconstructions at a regional scale. Nevertheless, this preliminary analysis demonstrates the utility of the PMIP4-CMIP6 simulations in addressing questions about the response of climate to large changes in forcing and illustrates the need to investigate the causes of inter-model differences in these responses.