Paleoclimate simulations provide an opportunity to validate model performance under substantially different conditions to the modern climate. Nevertheless, most effort in climate modelling still goes towards improving the models’ representation of the details of the present day climate. Recent ×CO2) in the work (Annan et al., 2005) has, however, shown that this is by no means a guarantee of success: it is possible to improve the representation of the present day (quasi steady-state) climate in a model while simultaneously decreasing the accuracy of its representation of climate change in response to substantial historical changes in boundary conditions (and therefore presumably worsening predictions of future change). Therefore, it is important to consider whether there are other ways of gaining confidence in, and improving the accuracy of, model predictions.
The last glacial maximum (LGM) epoch has long been recognised as a time which might provide useful information for inferring future climate changes (eg Manabe and Brocolli, 1985), due to the fact that it is the most recent time (and therefore the time for which paleoclimate data is available in some quantity and quality) when forcings, (including those from greenhouse gases), and the climate state itself, were significantly different from the modern era. Since the net forcing at that time was strongly negative, and includes large contributions from factors other than greenhouse gas levels (most notably, large ice sheets in the northern hemisphere), it is unclear as to how directly we can infer future climate changes based on the LGM state. Nevertheless, there is still useful evidence here, especially when considered in combination with other lines of evidence such as the modern warming trend, and the short-term response to volcanic forcing, which are individually somewhat weak but collectively rather more convincing (Annan and Hargreaves, 2006). Furthermore, even if paleoclimate simulations provide only limited validation of climate predictions, not undertaking such studies at all could hardly be argued to be a better strategy.
Annan et al. (2005) found a correlation between modelled LGM (global and tropical) 2mtemperature (T2) change and global T2 change (compared to the modern climate) for doubled atmospheric carbon dioxide (2 MIROC3.2 GCM (Hasumi and Emori, 2004). In that work the data used to validate the model’s LGM state were the PMIP1 Alkenone data (http://www-lsce.cea.fr/pmip/, Harrison, 2000) from from the tropical ocean region. These data 2 experiments. However, due to very limited2 states, they did2 states,2 integrations are×CO2 conditions×CO2 climates and therefore do have been widely used (eg Houghton et al., 2001, Chapter 8) and provide coverage over a substantial proportion of the Earth’s surface, so were therefore assumed to be reasonably representative of global climate change, but this question is still very much open. The availability and precision of regionally inhomogeneous data, the understanding of the forcings that dominate over particular geographical areas, and the confidence with which past and future changes can be linked are all factors which may affect which data are most useful for validating and improving model performance.
A recent examination of a multi-model ensemble from a range of different experiments (broadly PMIP1, PMIP2 and CMIP; http://www-lsce.cea.fr/pmip/, http://www-lsce.cea.fr/pmip2/ and http://www-pcmdi.llnl.gov/projects/cmip/index.php respectively) was undertaken by Masson-Delmotte et al. (2006) (hereafter MD06), with the focus of assessing the potential value of polar ice cores for providing “quantitative insights on global climate change”. Although their results were somewhat inhibited by small sample statistics, they concluded that there was a clear correlation between the global average and polar temperature changes compared to the control climates in the models for both the LGM and increased CO overlap between the model populations which were integrated for both the LGM, and increased CO not in fact analyse whether the polar or global LGM temperature changes were related, in the models, to the global or polar temperature changes for the increased CO although they considered their results to be consistent with the hypothesis that such a relationship does exist. Crucifix (2006) investigated this question with the set of 4 models for which both LGM and doubled CO available, and found no evidence of a relationship between global or tropical temperature changes. With only 4 coupled atmosphere-ocean model runs available which covered a modest range of climate sensitivity, it is not yet clear to what extent LGM simulations can help to narrow the rather wider range of model results that has sometimes been presented as plausible (e.g. Andronova and Schlesinger, 2001; Stainforth et al., 2005). Schneider von Deimling et al. (2006) found a strong relationship between LGM and 2 across an ensemble of intermediate complexity climate model with uncertain parameters allowed to vary, but Annan et al. (2005) found a rather weaker relationship with a more sophisticated model (state of the art AGCM with slab ocean) which has more sources of uncertainty in atmospheric feedbacks.
In this paper we extend our previous analysis and consider further the conclusions presented by MD06, by examining a large perturbed-parameter ensemble from one particular model (MIROC3.2 Hasumi and Emori, 2004). While we are able to integrate numerous pairs of identical model versions for both LGM and 2 not have such a severe problem with small sample statistics, our results are necessarily tentative because we show results from only one model, and as MD06 showed, results can vary considerably between different models. In addition, for computational reasons we are using the model in a slab ocean configuration, rather than the fully coupled model which is now state-of-the-art for PMIP2. However, our results suggestareas where further investigations may be worthwhile with a wider range of models. Also, where even a single-model ensemble generates negative or weak results, it seems unlikely that a multi-model ensemble, which introduces more sources of uncertainty, will generate anything more useful. We broaden the scope of the MD06 work, by considering not only annual average temperatures at the poles, but consider more broadly the zonal variation, the effects of land and ocean and also the seasonal variations. The main motivation for this is that data are available at a wide range of latitudes, and some are plausibly considered more directly representative of seasonal changes (e.g. precipitation-dependent proxies) rather than annual averages. In order to further explore the value of the LGM climate for estimating climate sensitivity we also compare the results from an experiment where we do not impose massive ice sheets or the insolation forcing of the LGM state, and thus the only change compared to the control run is that the levels of greenhouse gases (GHG) are changed to the LGM levels prescribed by PMIP2.
In Sect. 2 we outline the way the ensemble of model runs was formed and discuss the climate states that were modelled. In Sect. 3 we discuss the results focusing principally on a zonal analysis of the T2 temperature changes. In Sect. 4 we discuss the implications of our results for the calculation of climate sensitivity. In Sect. 5 we briefly touch on the complex issue of the attributing climate changes to variation in individual parameters, and then we conclude with an overview of the results and discussion of the wider implications.