Fig. 2: Model framework for interpreting CO₂ and δ¹³C-CO₂ data.

a Compilation of model results estimating the change in δ¹³C-CO₂ per unit change in CO₂ concentration due to different processes, indicated by shaded regions (see Supplementary information for details). The shading for the processes relevant to our interpretations in each interval is drawn on the following panels. b The temporal evolution of δ¹³C-CO₂ and CO₂ data is indicated by the color gradient on the markers. c–f δ¹³C-CO₂ and CO₂ data from each of the four intervals I–IV shown in Fig. 1. Note the axes are scaled differently for each panel. c δ¹³C-CO₂ and CO₂ change during Dansgaard-Oeschger (DO) 19 (interval I in Fig. 1). The data are most consistent with an increase in Southern Ocean air–sea gas exchange rates or a release of land carbon. d δ¹³C-CO₂ and CO₂ data for the negative isotope excursion and enrichment during the Marine Isotope Stage (MIS) 5-4 transition (interval II in Fig. 1). The negative excursion is consistent with a large pulse of land carbon combined with increasing efficiency of the biological pump. The growth of Antarctic sea ice and continued deep carbon storage could explain the following enrichment trend. e Oscillations in δ¹³C-CO₂ during MIS 4 were accompanied by very little change in CO₂ concentration (interval III in Fig. 1), perhaps due to fluctuations in Antarctic sea ice. f The δ¹³C-CO₂ and CO₂ change during Heinrich Stadial (HS) 6 (interval IV in Fig. 1). The large decrease in δ¹³C-CO₂ is consistent with decreasing Antarctic sea ice and increased air–sea gas exchange in the Southern Ocean. The youngest data (60.9–59.6 ka) are consistent with decreasing efficiency of the biological pump.