Southern Ocean marine sediments provide an important archive of Quaternary glacial-interglacial climate changes. Sea surface temperature (SST) reconstructions in the Southern Ocean depend exclusively on the fossils or geochemical signatures of planktic organisms, but the strengths of these SST proxies remain poorly quantified in this region. To improve confidence in Southern Ocean paleoclimate reconstructions, we first evaluated the reliability of SST proxies employed at Quaternary glacial-interglacial time scales, focusing on three key potential problems: advection/dispersion, seasonality, and non-thermal influences. On that basis, we selected foraminifera assemblages, long-chain alkenones (UK'37 index), diatom assemblages, and the Globigerina bulloides Mg/Ca ratio as the most reliable SST proxies in the Southern Ocean. We also revised calibrations where appropriate, using core-top sediment databases. Overall, Southern Ocean SST reconstructions using these recommended proxies and calibrations should be robust when averaging across multiple sites and proxy types, but should be treated with caution when analysing spatial variability, a small number of sites, or a single proxy type. We have applied these recommendations to a synthesis of Southern Ocean SST over the penultimate glacial and last interglacial (LIG). Similar to previous studies, we find that LIG warming at 40°S to 60°S reached 1.6 ± 1.1°C (annual mean) or 1.9 ± 1.3°C (austral summer: JFM) relative to present. Annual / summer cooling in the penultimate glacial maximum reached -3.6 ± 1.0°C / -4.0 ± 1.2°C, similar to the last glacial maximum. Compared with the previous LIG SST syntheses, our reported uncertainties more strongly reflect geographic variability and dating errors, as we have reduced errors in the individual temperature reconstructions and do not date records by aligning peaks in their SST.

This dataset contains individual SST records for the region south of 40°S, resampled to 2000 year resolution following Chandler and Langebroek (QSR, in review). Time slices without a temperature within +/-1000 years are assigned empty rows. Age models are based on benthic d18O where possible, with planktic d18O as a second choice. d18O is aligned with the respective regional stack (Lisiecki and Stern, 2016). Records aligned by other methods (for example, aligning by SST), are not included.Temperatures based on alkenones were calculated using annual SST (°C) = 30.3(UK'37) - 1.30 or summer SST (°C) = 29.7(UK'37) + 0.30. The former is the Prahl et al. (1988) calibration; the latter a revised calibration (Chandler and Langebroek, QSR, in review).Temperatures based on G. bulloides Mg/Ca were calculated using annual SST (°C) = 13.0ln(1.23Mg/Ca), based on a revised calibration (Chandler and Langebroek, QSR, in review).Temperatures based on diatom assemblages were taken as published by the original authors.Temperatures based on foraminifera assemblages were calculated using the modern analogue technique and Haddam et al. (2016) core-top database. For full details, see Chandler and Langebroek (QSR, in review).