Extended Data Fig. 2: Relationship between estimated mean-annual and October δ¹⁸O_sw, and the overall seasonal range at Site U1446.

Scatterplot of SSS-derived mean-annual δ¹⁸O_sw (using a region-specific equation on reanalysis SSS data¹⁴) with (a) October δ¹⁸O_sw (blue triangles) and (b) its seasonal range (blue circles; maximum minus minimum values in a year) at Site U1446. Lines of best fit were built using (solely) the SSS-derived δ¹⁸O_sw values on both plots (dashed blue lines), incorporating bivariate uncertainty in the parameters (slope, intercept, and correlation coefficients are provided at the bottom⁶⁵), where 1σ errorbars are based on average error propagation for δ¹⁸O_sw inversion (based on 225 downcore points), including analytical and sampling error¹⁷. The resultant covariability indicates that October δ¹⁸O_sw minima drive yearly-averaged δ¹⁸O_sw values, which are in turn, highly correlated with the seasonal range of δ¹⁸O_sw at Site U1446; 1σ errorbars for the seasonal range were derived from bootstrap resampling (n = 480). This relationship thereby allows us to estimate past seasonality based on our mean-annual δ¹⁸O_sw reconstruction. For (b) we also plot available observations of δ¹⁸O_sw near the Mahanadi Basin (black square) and reconstructed δ¹⁸O*_sw for the late Holocene (red square), early Holocene (green square), and Heinrich Stadial 1 (purple square). Note that these reconstructed points of mean-annual δ¹⁸O_sw are plotted on the line of best fit to estimate changes in seasonality during that interval. For example, our δ¹⁸O_sw record indicates a value of ~0.55 ‰ (VSMOW) for HS1, which yields a seasonal range of ~0.7 ‰ (VSMOW) — a ~ 45% decrease relative to the late Holocene seasonal range of ~1.45 ‰.