Reconstructed Deglacial Sea Surface Temperatures

Reconstructed SST estimates were compiled that spanned time periods between the LGM (23 to 19 ka) to Holocene time periods, including temperature reconstructions based on the alkenone  $U_{37}^{K\text{'}}$  index, Mg/Ca in planktic foraminifera, TEX₈₆, and planktic foraminiferal assemblages, generally following the same criteria, calibrations, and methods in ref. 41 (link). If a site had multiple proxy estimates, SST anomalies were averaged. Additional SST estimates using the BAYSPLINE  $U_{37}^{K\text{'}}$  calibration (49 ) are also included in the Dataset S4. Age models were recalibrated using the Marine20 calibration curve (93 ) with the Calib 8.2 software (94 ) whenever possible (i.e., when age model data were accessible). In most cases, the most recent age model was adopted, with attempts to retain prior correlation datums (such as tephra or geophysical correlations), along with the recalibrated radiocarbon dates. Generally, the authors’ original suggested marine reservoir corrections were used, except in the cases when the original corrections were <550 y, in which case the new default marine reservoir correction of 550 y in Marine20 (93 ) was used. In some cases, the original age models were retained if there were not sufficient data available to update datasets (i.e., no age model data or depth in core data provided for the SST estimates). Additional information and references for various age models are provided in the supplementary data files (Datasets S2 and S4).
Deglacial SST anomalies for various time slices were calculated for records that had average sample spacing finer than 400 y between 10 and 18 ka. Additional lower resolution sites were included for estimates of LGM-Holocene temperature anomalies. SST anomalies were calculated for the following climate intervals: LGM relative to the early Holocene (23.0 to 19.0 ka − 11.5 to 11.0 ka), Siku Event 1 relative to the LGM (18.0 to 16.5 ka − 23.0 to 19.0 ka), the pre-Bølling period relative to Siku Event 1 (16.4 to 15.0 ka − 18.0 to 16.5 ka), Bølling–Allerød relative to the pre-Bølling period (14.6 to 13.0 ka − 16.4 to 15.0 ka), Younger Dryas relative to Bølling–Allerød (12.7 to 12.0 ka − 14.6 to 13.0 ka), and early Holocene relative to the Younger Dryas (11.5 to 11.0 ka − 12.7 to 12.0 ka). We plot our proxy SST anomalies for the deglacial climate intervals with annual SST estimates from the transient model output of iTRACE (37 ) (Fig. 4). SST anomalies for the LGM relative to the late Holocene (0 to 3 ka) were also calculated (Dataset S4), but due to few records that extend from the LGM to late Holocene, we opted for the better spatial coverage provided by the LGM-early Holocene anomalies.
An averaged record of high-resolution (~100 y average) Northeast Pacific SST records was also produced, similar to that presented in (41 (link)), but the record we present here includes an additional record (a Mg/Ca-based SST reconstruction on the thermocline dwelling Neogloboquadrina pachyderma sinistral from core MD02-2496; ref. 63 ) along with the following   $U_{37}^{K\text{'}}$ records, which were included in the original average: EW0408-85JC (54 (link)); EW0408-66JC & EW0408-26JC (95 ), JT96-09PC (96 ), ODP1019 (61 , 62 ), and the Mg/Ca-based SST reconstruction on the planktic species Globigerina bulloides from core MD02-2496 (63 ). All records were linearly interpolated on a 100-y time step and averaged for overlapping time intervals, with a minimum of two records required. As fewer high-resolution records are available beyond 20 ka, the number of records contributing to the stack is reduced going back farther in time, and thus more susceptible to site-specific variability rather than regional trends. An average SST record utilizing two additional, lower resolution  $U_{37}^{K\text{'}}$  records from the Gulf of Alaska (EW0408-87JC; ref. 41 (link) & U1419; ref. 31 ) was also produced to increase the number of records in the stack (SI Appendix, Fig. S7). For this stack, records were linearly interpolated on a 200-y time step and averaged for overlapping time intervals. A normalized stack was also constructed with all cores on a 200-y time step. Each record was normalized to its mean and SD and then averaged. All versions of the Northeast Pacific stack show similar trends.