Climates of the past, lessons for the future: emerging results from the CMIP5 model evaluation exercise

Abstract:

Past climates provide an out-of-sample test of the climate models used for future climate projections. We evaluate palaeosimulations of the Last Glacial Maximum (21,000 years before present) and Mid-Holocene (6000 years before present) from the fifth phase of the Coupled Model Intercomparison Project (CMIP5), focusing on features that are also present in simulations of the response to raised CO₂. Several features of the simulations are common to both past and future simulations, including the ratio of land-sea temperature contrast, high-latitude amplification of temperature changes, winter amplification of temperature responses to year-round forcing, precipitation scaling with temperature, and differential precipitation/temperature scaling over land and ocean. The mechanisms giving rise to these phenomena are similar in past and future climates. Comparison with palaeoclimatic reconstructions show these features are simulated realistically in the past, and presumably therefore in the future simulations. There are also parallels in the regional response to changes in forcing in the past and future simulations, including changes in monsoon strength and in mid-continental climates. Comparison with palaeo-reconstructions shows the models capture the sign of such changes correctly but predict the magnitude poorly, and thus future projections are also likely to underestimate the strength of regional responses. LGM climates have been used to constrain estimates of climate sensitivity, on the assumption that the model that best reproduces reconstructed LGM climates is likely to have the most realistic climate sensitivity. However, global biases in simulated LGM temperature are only weakly related to climate sensitivity. Furthermore, ocean temperature biases are globally low and land temperatures globally high compared with reconstructions. The current generation of climate models includes improvements in resolution, complexity and parameterizations and produce better simulations of modern climate but evaluations show there has been no improvement in the simulation of past climates. Since past forcing and climate changes are of comparable magnitude to those expected in the 21^st century, this emphasizes the need to focus on further improving the simulation of regional climate changes.