Clim. Past, 3, 331–339, 2007 www.clim-past.net/3/331/2007/ Climate © Author(s) 2007. This work is licensed of the Past under a Creative Commons License. How cold was Europe at the Last Glacial Maximum? A synthesis of the progress achieved since the first PMIP model-data comparison G. Ramstein1, M. Kageyama1, J. Guiot2, H. Wu2,3, C. Hely´ 2, G. Krinner4, and S. Brewer2 1Laboratoire des Sciences du Climat et de l’Environnement, UMR CEA-CNRS-UVSQ 1572, CE Saclay, L’Orme des Merisiers, Bat.ˆ 701, 91191 Gif-sur-Yvette Cedex, France 2CEREGE, BP 80 Europole Mediterran´ een´ de l’Arbois, 13545 Aix-en-Provence cedex 4, France 3SKLLQ, Institute of Earth Environment, Chinese Academy of Sciences, Xi’an 710075, China 4Laboratoire de Glaciologie et Geophysique´ de l’Environnement, 54, rue Moliere,` 38402 Saint Martin d’Heres` cedex, France Received: 1 December 2006 – Published in Clim. Past Discuss.: 22 January 2007 Revised: 3 May 2007 – Accepted: 31 May 2007 – Published: 18 June 2007 Abstract. The Last Glacial Maximum has been one of 1 Introduction the first foci of the Paleoclimate Modelling Intercomparison Project (PMIP). During its first phase, the results of 17 at- The aim of the first phase of the Paleoclimate Modelling In- mosphere general circulation models were compared to pa- tercomparison Projet (hereafter PMIP1) was to assess the leoclimate reconstructions. One of the largest discrepancies sensitivity of the atmosphere general circulation models in the simulations was the systematic underestimation, by ◦ (AGCMs) used to predict future climate change, to very dif- at least 10 C, of the winter cooling over Europe and the ferent conditions. The Last Glacial Maximum, which oc- Mediterranean region observed in the pollen-based recon- curred 21 000 years ago, was chosen as a test for extremely structions. In this paper, we investigate the progress achieved cold conditions. For this period, a relatively large number of to reduce this inconsistency through a large modelling ef- paleoclimate reconstructions exist, against which model re- fort and improved temperature reconstructions. We show that sults can be compared. For example, across Europe and west- increased model spatial resolution does not significantly in- ern Siberia, models and reconstructions could be compared crease the simulated LGM winter cooling. Further, neither in terms of changes in Mean Annual Temperature (MAT), the inclusion of a vegetation cover compatible with the LGM Mean Temperature of the Coldest Month (MTCO), total an- climate, nor the interactions with the oceans simulated by nual precipitation (TAP) and moisture index (ratio of the the atmosphere-ocean general circulation models run in the mean annual actual evaporation to the mean annual poten- second phase of PMIP result in a better agreement between tial evaporation). These bioclimatic parameters were com- models and data. Accounting for changes in interannual vari- puted from pollen assemblages using a transfer function re- ability in the interpretation of the pollen data does not result lating modern distributions of Plant Functional Types with in a reduction of the reconstructed cooling. The largest re- climatic data (Peyron et al., 1998; Tarasov et al., 1999). For cent improvement in the model-data comparison has instead instance, the modern analogues found for the Western Eu- arisen from a new climate reconstruction based on inverse rope pollen assemblages in the original climatic reconstruc- vegetation modelling, which explicitly accounts for the CO2 tions from Peyron et al. (1998) were located in tundra or decrease at LGM and which substantially reduces the LGM very cold steppe environments, which resulted in very cold winter cooling reconstructed from pollen assemblages. As a MTCO reconstructions. result, the simulated and observed LGM winter cooling over Western Europe and the Mediterranean area are now in much In order to involve a large number of groups and mod- better agreement. els that had not been previously used with paleoclimates, the LGM experimental design in PMIP1 was designed to be as simple as possible (Joussaume and Taylor (1995) and http://pmip.lsce.ipsl.fr/). The boundary conditions were as follows: the CO2 atmospheric concentration was fixed at Correspondence to: M. Kageyama 200 ppm following the Vostokmeasurements (Raynaud et al., ([email protected]) 1993) and the orbital parameters were set to their 21 ky BP Published by Copernicus Publications on behalf of the European Geosciences Union. 332 G. Ramstein et al.: Europe LGM winter temperatures from models and data 2 Increasing the spatial resolution of the models Paleoclimatic reconstructions such as those based on pollen are representative of a small area around the sites where they have been retrieved, compared to the usual resolution (i.e. a few hundred km) of the AGCM used in PMIP1. Therefore, the spatial resolution of the PMIP models may be a problem when comparing model results to data, particularly in areas of complex coastlines and topography such as southern Eu- rope and around the Mediterranean Basin. For sites situated in or close to mountain ranges such as the Pyrenees and the Alps, the local climate can be very different from the climate simulated in the corresponding grid box of the models. Dur- ing the LGM, these mountain ranges were partly covered by large glaciers, which can affect the local atmospheric circu- Figure 1 lation but are not represented in the GCMs. Increasing the Fig. 1. (a) MTCO anomalies between the LGM and CTRL results models’ resolution should improve the representation of a for the PMIP1 models, compared to the MTCO anomalies (triangles given climate. However, it is unclear that the sensitivity of +90% confidence intervals) reconstructed from pollen as described the models to changes in their boundary conditions will be in Peyron et al. (1998). (b) Same MTCO anomalies but the data set affected by changing their resolution. has been updated as described in Jost et al. (2005) (squares +90% This question has been investigated by comparing the sim- confidence intervals). ulations of three AGCMs at low and high resolution (Jost et al., 2005), with the increase in resolution achieved through three different methods. All simulations were run accord- ing to the PMIP1 protocol. CCSR1 provided a run with a values (Berger, 1978); the ice-sheet elevation and extension, global T106 resolution (low resolution: T21), whereas the as well as the land-sea mask were prescribed using the ICE- HadRM was nested within the HadAM global AGCM over a 4G reconstruction (Peltier, 1994); eight simulations used pre- domain including the North Atlantic and Europe, and LMDZ scribed Sea Surface Temperature (SST) and sea-ice extension used a stretched grid version, with higher resolution over Eu- derived from CLIMAP (1981), while the eight others used a rope (low resolution: 72 points in longitude × 46 points in slab-ocean model to compute SST and sea-ice cover. latitude, high resolution 144×108, the resolution reaching The comparison between these model results and pollen- 60 km over Paris). The CCSR and LMDZ experiments are based reconstructions over Europe and western Siberia all 11-year long, the last 10 years being used to compute the showed a relatively good agreement for MAT, but large dis- climatological averages. The HadAM and HadRM results crepancies were found over western Europe and the Mediter- are averages for the last 6 years of a 14-year-long present- ranean area for MTCO and TAP (Kageyama et al., 2001). day simulation and of the last 5 years of a 9-year-long LGM Since then, the modern data base used for the calibration simulation. All these models have a spatial resolution of of the Peyron et al. (1998) and Tarasov et al. (1999) trans- around 50–100 km over Europe. However in terms of MTCO fer function method has been updated (Peyron et al., 2005) (Fig. 2), there is not any convincing improvement for CCSR1 and new reconstructions for western Europe and the Mediter- and LMDZ, while HadRM simulates cooler winter temper- ranean area are slightly warmer (Jost et al., 2005). How- atures than the low resolution model, in better agreement ever, a comparison of the PMIP1 model results with this up- with the reconstructions. However, the latter model simu- dated MTCO reconstruction (Fig. 1) shows that the model- lates significant increases in precipitation over western Eu- data disagreement remains, even if one takes into account rope and the Mediterranean areas which are in total disagree- the large error bars due to a poorly diversified vegetation ment with the reconstructions. Therefore, Jost et al. (2005) (steppe-tundra) and a tolerance of this type of vegetation to argue that such increases in spatial resolution cannot explain large climatic amplitudes. This therefore raised a series of the model – data MTCO discrepancy over Western Europe questions from models and data point of view. In the present and the Mediterranean area. It remains to be investigated paper, we first review different model improvements: using whether a larger increase in the resolution and the inclusion a higher spatial resolution, investigating the role of missing of finer scale processes in the models would better resolve the interactions or feedbacks, such as from the biosphere and the circulation and climate around the fine topography/coastlines ocean. We then investigate factors that were not included in of the European and Mediterranean areas and yield a better the first pollen-based reconstructions: the potential impact of agreement of the model results with MTCO reconstructions. a different interannual variability and that of a much lower Other methods, such as statistical downscaling, could also be CO2 at LGM, via the use of an Inverse Vegetation Model. investigated. Clim. Past, 3, 331–339, 2007 www.clim-past.net/3/331/2007/ G. Ramstein et al.: Europe LGM winter temperatures from models and data 333 Tcold, LGM-CTRL 10 longitudes [-10:15], latitudes [30:50] Jost et al 2005 ccsr1_HR_22x18 0 LMDZ_HR_27x24 HADRM_HR_50x40 -10 LMDZ_LR_6x6 HADAM_LR_7x9 -20 ccsr1_LR_4x4 pmip1fix -30 temperature anomaly -40 -50 30 35 40 45 50 latitude Fig.