Glacial ocean temperatures from ‘clumped isotope’ thermometry in foraminifera and coccoliths

A primary goal of paleoclimate research over several decades has been to establish the magnitude and spatial patterns of temperature change in the oceans since the Last Glacial Maximum (LGM; ~21 ka). We are using ‘clumped isotope’ thermometry to develop a global map of sea surface temperatures (SSTs) for the LGM. This thermometer is based on the principle that ordering, or ‘clumping’, of heavy isotopes into bonds with each other in molecules is temperature-dependent due to an internal isotope exchange reaction which requires no knowledge of seawater isotope composition. We have previously calibrated the clumped isotope thermometer in foraminifera and coccoliths and shown that that the proportions of ^(13)C-^(18)O bonds in these biogenic carbonates are not subject to non-equilibrium biological fractionations, and are independent of the composition of the water in which the mineral precipitated (Tripati et al., in revision). Here we present the first record of Pleistocene and Holocene temperatures that is purely thermodynamically based, based on clumped isotope measurements of samples from the West Pacific warm pool. The warm pool is the warmest oceanic region and a benchmark for climate models. Reconstructions of the glacial-interglacial change in SSTs in the region vary from 1 to 6°C, differing systematically between methods. At the LGM average SSTs in the region were 25.5°C (~3.9°C cooler than today), and the δ^(18)O_(SMOW) of surface waters was 0.9‰. Assuming these results are representative for the region, then SSTs derived from alkenones, transfer functions, multi-proxy syntheses (i.e. MARGO) are systematically underestimating glacial cooling in this region. Mg/Ca SSTs appear to be the most accurate empirical proxy of SST in the region. Salinity-corrected Mg/Ca SSTs, and some terrestrial reconstructions, are indistinguishable from clumped-isotope derived temperatures. The predicted LGM temperature difference, defined as the difference in annual mean SSTs between LGM and pre-industrial simulations, generated by PMIP2 models for the region ranges from 1.0 to 2.4°C. All models predict less warming in the warm pool than our reconstruction. This mismatch could arise if there are problems in modelling tropical processes (e.g. if the models were not simulating changes in ENSO correctly), if the LGM simulations were incomplete (e.g. not including dust changes), and/or if these models systematically underestimate climate sensitivity