Fig. 5: Global atmospheric teleconnections during stage III.

a NGRIP deuterium-excess record⁶⁹ on the GICC05 chronology and shifted by +320 years. b–g speleothem δ¹⁸O records from Yongxing Cave (YX-51, this study), Cherrapunji Cave (Cherrapunji, this study), Larga Cave (PA-LA-1⁹⁶), Marota Cave (MAG, this study), Paixão Cave (PX-07, this study), and Botuverá Cave (BTV-4C, this study), respectively. h Antarctica 5-core averaged d_ln anomaly¹⁶. i Antarctica WDC ice-core non-sea-salt Ca²⁺ [nssCa²⁺] record⁷⁴ and overlain by the thick line (high-frequency variability (>1 cycle per 300 year) removed by a low-pass Butterworth filter, same as in ref. 74). Antarctic ice-core records in (h) and (i) are on the WD2014 chronology and shifted by +400 years. The yellow dashed lines in (c–f) are generated via the BREAKFIT algorithm⁴⁵. Black dots and associated error bars in (c–f) depict the timing of breakpoints and combined uncertainties (2σ) (Supplementary Table 2). Latitudes and longitudes for the various records in (a–g) and (i) are shown. Note the inverted y-axis in (a–d) and (h). j Probability density functions of spatial age offsets between the start of AHP2 and SAHP2 as well as the start of the large shift in Antarctic ice-core d_ln record (Southern Hemisphere records minus Asian summer monsoon record); shadings indicate the 95.4% intervals (2σ uncertainty), the uncertainties are the combined uncertainties (Supplementary Table 2). k Laser scanning confocal microscopy image of speleothem Cherrapunji−2 between 316–329 mm, the green curve is the plot of Cherrapunji-2 δ¹⁸O versus depth; the timing of change points (bottom) and lamina counting results (top) are shown, the laminae were counted as ten between adjacent yellow dots, and the horizontal double-sided arrow depicts the rapid reversal which occurred within decades; the uncertainties of the change points in (k) depict the age model uncertainty (2σ) (Supplementary Data 2). AHP2 Asian Heinrich Period 2, SAHP2 South American Heinrich Period 2.