Estimates of global mean surface temperature change (ΔGMST) for the 8.5–4.5 Ma interval following the methods of Clark et al. (2024, doi:10.1126/science.adi1908 )with slight modification to address the smaller quantity of data, lower number of records (n = 25, Fig. S7), and generally poorer age control in the available Late Miocene and Early Pliocene sea surface temperature (SST) records. Since it was impossible to align records based on benthic δ18O age models, data were binned into 500,000-year wide bins with a 250,000-year shifting bin window instead.

Methods:For comparison with our data from Lake Baikal, we generated new estimates of global mean surface temperature change (ΔGMST) for the 8.5–4.5 myr interval following the methods of Clark et al. (2024) with slight modification to address the smaller quantity of data, lower number of records (n = 25), and generally poorer age control in the available late Miocene and Early Pliocene sea surface temperature (SST) records. Our ΔGMST estimates differ somewhat from those recently generated by Brown et al. (2024) for the 7–4.5 myr interval in that our ΔGMSTs are calculated from an area weighted mean while those of Brown et al. (2024) are not. We took the area weighted approach for the sake of comparability with the ΔGMST dataset of Clark et al. (2024).All alkenone datasets were recalculated with BAYSPLINE (Tierney and Tingley, 2018); a prior standard deviation of 5°C was used for records with UK'37 values > 0.95 during the 8.5–0 Ma period while a prior standard deviation of 10°C was used for all other records per the recommendation in Tierney and Tingley (2018). We use the Mg/Ca sea surface temperature estimates (re)calculated by Martinot et al. (2022) and Liu et al. (2022) as published. Records were grouped into 15° latitude bands and were converted from absolute temperature estimates to change in temperature (i.e., anomaly) relative to the 3.1-2.9 Ma interval as in Clark et al. (2024). Since it was not possible to align records on the basis of benthic δ18O-based age models, we instead binned data into 500,000-year wide bins with a 250,000-year shifting bin window.This approach is like that of Herbert et al. (2016; 2022) but with a larger bin size to accommodate intercomparison to our less densely sampled Lake Baikal data. For records where data was unavailable from the 3.1–2.9 myr interval, we created a preliminary stack for that latitude bin and then mean shifted the binned sea surface temperature data from the record without data from 3.1–2.9 myr to match the mean of the preliminary stack for their overlapping interval, following Clark et al. (2024).Stacks for each 15° latitude band were then calculated by taking the mean of the sea surface temperature anomalies (Clark et al., 2024). Uncertainty for each binned SST record was calculated as follows:1σ Uncertainty = √((S.E.M. bin SST)2 + (1σ proxy uncertainty)2)Uncertainty for each stack was calculated as the mean uncertainty of each record. A global sea surface temperature stack was calculated from the 15° latitude band stacks as a weighted average based on the surface area of the earth represented by each latitude band (Clark et al. 2024). Global mean surface temperature was then calculated from the global mean sea surface temperature stack using a polynomial transfer function developed from global climate model output presented by Clark et al. (2024).ΔGMST = -0.37629 + 1.603508(ΔSST) -0.058842(ΔSST)2The standard error of this relationship, 0.3238°C, was then added in quadrature to the uncertainty in global mean sea surface temperature change to estimate the uncertainty in ΔGMST (Clark et al. 2024).Further funding information: PR 1414/1-1 Deutsche Forchungsgemeinshaft Priority Program "ICDP" 1006 Geological Society of America Continental Drilling Science Division Graduate Student Grant 13282-21* Sigma Xi Grants in Aid of Research G20211001-101