Figure 3: Comparison of Δ₄₇-derived and δ¹⁸O-derived temperature records.

Horizon mean Δ₄₇-derived temperature shown in black with grey error bars (same as Fig. 1a). Horizon mean was calculated as the inverse variance weighted mean, to account for variable reproducibility of samples, with error bars showing the inverse variance weighted error on the mean (see Methods). Line connecting horizon means is dashed to represent extrapolation of temperature trends between measured horizons. Samples within ±3.5 m stratigraphic position of another sample were combined into a single horizon for horizon means (see Methods). δ¹⁸O-derived temperature calculated from horizon mean δ¹⁸O (this study) shown in dark turquoise circles with error bars representing 1 s.d. on δ¹⁸O, propagated through the δ¹⁸O-T-δ¹⁸O_w relationship¹². Published δ¹⁸O-derived temperature record from Tobin et al.¹⁰, with age model updated to match this study, shown in light green triangles with error bars representing 1 s.e. on the mean of all sample replicates within a given time bin. Both δ¹⁸O-derived temperature records assume a δ¹⁸O_w value of −1.0‰ (ice-free, globally constant)²⁹. If δ¹⁸O_w was instead assumed to be −1.2‰ (ice-free, latitude-adjusted)³⁰, the shape of the δ¹⁸O-derived curves would be identical but shifted ∼1 °C colder. The two δ¹⁸O-derived temperature curves miss the million-year cool/warm/cool temperature structure, underestimate the magnitude of the pre-KPB warming event and infer an earlier timing for the onset of warming, and do not capture the full range of temperatures seen in the Δ₄₇-derived temperature record. Vertical grey dashed line represents the KPg boundary (labelled KPB)³⁴ and orange dotted line represents the timing of the first extinction event¹⁰. Age model construction is described in Methods.