Résumé
Past climate is an important benchmark to assess the ability of climate models to simulate key processes and feedbacks. Numerous proxy records exist for stable isotopes of water and/or carbon, which are also implemented inside the components of a growing number of Earth system model. Model-data comparisons can help to constrain the uncertainties associated with transfer functions. This motivates the need of producing a comprehensive compilation of different proxy sources. We have put together a global database of proxy records of oxygen (δ.sup.18 O), hydrogen (δD) and carbon (δ.sup.13 C) stable isotopes from different archives: ocean and lake sediments, corals, ice cores, speleothems and tree-ring cellulose. Source records were obtained from the georeferenced open access PANGAEA and NOAA libraries, complemented by additional data obtained from a literature survey. About 3000 source records were screened for chronological information and temporal resolution of proxy records. Altogether, this database consists of hundreds of dated δ.sup.18 O, δ.sup.13 C and δD records in a standardized simple text format, complemented with a metadata Excel catalog. For coherency with the Paleoclimate Modelling Intercomparison Project (PMIP), we focus on records spanning the past 200 years, the mid-Holocene (MH, 5.5-6.5 ka; calendar kilo years before 1950), and the Last Glacial Maximum (LGM, 19-23 ka), and those spanning the last interglacial period (LIG, 115-130 ka). For the LGM, a quality control flag was implemented to describe age markers and inform on chronological uncertainty. Basic statistics have been applied to characterize anomalies between these different periods. Most changes from the MH to present day, and LIG to MH appear statistically insignificant. Significant global differences are reported from LGM to MH with regional discrepancies in signals from different archives and complex patterns. This compilation effort highlights the need to homogenize the format of datasets and chronological information, and enhance the distribution of published datasets that are currently highly-fragmented and scattered. We also provide an online portal with an intuitive and interactive platform (