Reconstructing the global mean surface temperature of the last 130 thousand years

Global mean surface temperature (GMST) is a fundamental measure of climate evolution in both past and present and a key quantity to evaluate climate simulations. However, for paleoclimate periods, its reconstruction hinges on uncertain and indirect observations which are distributed sparsely and non-uniformly in both space and time. Large datasets of homogenised proxy records help to reduce the sparsity. Then, the records can be aggregated in an algorithm retrieving the GMST signal. Here, we build on the algorithm designed in Snyder 2016, and on a recent database of ocean temperature proxy records to reconstruct the GMST evolution over the last glacial cycle (the last 130 thousand years). First, we evaluate the algorithm and quantify the sources of uncertainty. This analysis draws on pseudo-proxy experiments using a range of simulations of the last glacial cycle. We find that the over-representation of some regions (e.g. coasts, the Atlantic), to the detriment of others (e.g. the central Pacific) significantly impacts the reconstructed temperature anomaly and its variations. Additionally, millennial and shorter timescale variability cannot be reconstructed by the algorithm, due to bioturbation and age uncertainty. However, these experiments also demonstrate the ability of our algorithm to reconstruct the amplitude and timing of GMST variations occurring at orbital timescale (>10.000 years). Second, we reconstruct the GMST evolution during the last glacial cycle. We compare our result to previous studies, and discuss the improvements coming from the use of the recent proxy database. The high number of proxy records allow us to additionally investigate smaller regions (e.g. hemisphere) and overall further our understanding of the driver of orbital-scale GMST variability.