How well can we simulate past climates? Evaluating the models using globalpalaeoenvironmental datasets

Global syntheses of palaeoenvironmental data are required to test climate m odels under conditions different from the present. Data sets for this purpo se contain data from spatially extensive networks of sites. The data are ei ther directly comparable to model output or readily interpretable in terms of modelled climate variables. Data sets must contain sufficient documentat ion to distinguish between raw (primary) and interpreted (secondary, tertia ry) data, to evaluate the assumptions involved in interpretation of the dat a, to exercise quality control, and to select data appropriate for specific goals. Four data bases for the Late Quaternary, documenting changes in lak e levels since 30 C-14 kyr BP (the Global Lake Status Data Base), vegetatio n distribution at 18 C-14 kyr and 6 C-14 kyr BP (BIOME 6000), aeolian accum ulation rates during the last glacial-interglacial cycle (DIRTMAP), and tro pical terrestrial climates at the Last Glacial Maximum (the LGM Tropical Te rrestrial Data Synthesis) are summarised. Each has been used to evaluate si mulations of Last Glacial Maximum (LGM: 21 calendar kyr BP) and/or mid-Holo cene (6 cal. kyr BP) environments. Comparisons have demonstrated that chang es in radiative forcing and orography due to orbital and ice-sheet variatio ns explain the first-order, broad-scale (in space and time) features of glo bal climate change since the LGM. However, atmospheric models forced by 6 c al. kyr BP orbital changes with unchanged surface conditions fail to captur e quantitative aspects of the observed climate, including the greatly incre ased magnitude and northward shift of the African monsoon during the early to mid-Holocene. Similarly, comparisons with palaeoenvironmental datasets s how that atmospheric models have underestimated the magnitude of cooling an d drying of much of the land surface at the LGM. The inclusion of feedbacks due to changes in ocean- and land-surface conditions at both times, and at mospheric dust loading at the LGM, appears to be required in order to produ ce a better simulation of these past climates. The development of Earth sys tem models incorporating the dynamic interactions among ocean, atmosphere, and vegetation is therefore mandated by Quaternary science results as well as climatological principles. For greatest scientific benefit, this develop ment must be paralleled by continued advances in palaeodata analysis and sy nthesis, which in turn will help to define questions that call for new focu sed data collection efforts. (C) 1999 Elsevier Science Ltd. All rights rese rved.