During the last ice age, the well-known millennial-scale events, including the Dansgaard–Oeschger (DO) cycles and Heinrich Stadial (HS) in the North Atlantic, and the Antarctic Isotopic Maxima (AIM) events in the Antarctica1,2,3,4,5, are anti-phased related, and likely driven by changes in the intensity of the Atlantic Meridional Overturning Circulation (AMOC), as conceptualized in the bipolar seesaw mechanism6. However, their characteristics and mechanisms remain in hot debate, including the apparent hemispheric asymmetry characteristic of DO-AIM events4,7, uncertain lead-lag relationships of the Northern Hemisphere (NH) and Southern Hemisphere (SH)5, external forcing of volcanism and solar activities8,9, roles of oceanic/atmospheric circulations in transferring the climatic signals10,11, etc. With the emergence of high-resolution geological records, detailed characteristics in the Greenland ice cores are found distinctive from records in the Antarctica and other regions. Few counterparts and abrupt changes in the Greenland ice cores could be found in archives from the Antarctica and other regions7,12, indicating that Greenland-type changes are regional while the Antarctic-type variability is likely of global significance13,14. Furthermore, fast shifts in greenhouse gases (CH4 and CO2) within millennial-scale HS events are observed, indicating mechanisms involving centennial-scale oscillations in regulating millennial climates15,16,17. Although those millennial-scale changes that occurred across the last ice terminations (T1) were extensively regarded as important forcings18,19,20, a further investigation on nature and mechanism of millennial-scale change will help to understand mechanisms behind the glacial-interglacial transition.

The analogs of the millennial-scale changes between different glacial-interglacial cycles provide critical constraints on their mechanisms. Similar frequency and amplitude of millennial events between the penultimate and the last glacial period have been observed, suggestive of an importance of orbital control21,22. Furthermore, throughout the Late Pleistocene, analogous nature and structures of millennial-scale monsoonal events across the last four terminations23 and mega-weak monsoon intervals associated with each termination18 indicate that the millennial-scale events are likely the pacemakers for the transition from glacial to interglacial periods. During the last glacial period, temperature changes in the AIM events are proportional with Antarctic termination during the first 1500 years, hence the southern warming is supposed to be an initial trigger for the terminations24. Previous study compared the intensity of ocean circulation in the SH during the last 40 ka and highlighted the importance of vigorous circulation in triggering T125. This point of view is further supported by ref. 26. Changing sequences of ice sheets in the New Zealand and the Asian summer monsoon similar to that of the T1 were found at the transition of Marine Isotope Stage 4/3 (MIS4/3t), namely the unfinished termination18,27,28. These studies display the strong self-similarity and possible recurrences of the Earth’s climate system, although analogs in ancient climates were not that perfect.

Asian Monsoon changes, as revealed by radiometric-dated cave speleothems, are characterized by a typical pattern of precessional cycles which follow changes in NH insolation18,21. Superimposed on the long-term insolation changes are series of millennial signals which can be related to both climate changes at the Greenland and the Antarctica28,29,30,31,32. A dominant SH control on millennial-scale monsoon variability was found documented in Hulu stalagmites during glacial times and a NH control during deglacial and interglacial times32. Hence, millennial-scale monsoon changes on the precessional climate background provide a potential to reveal an analog between different precessional cycles, which assists understanding of the millennial-scale oscillations and the glacial-interglacial theory.

Here we present data from Yongxing Cave, China, to demonstrate an analogous feature of millennial-scale changes during the last glacial period. Previous publications from Yongxing Cave, including detailed information on the DO-AIM events imprinted on East Asian summer monsoon (EASM)30, decadal-to-centennial-scale changes during the last interglacial33 and annually-resolved data across the MIS1134, highlight that speleothem records from this cave are optimal to reconstruct high-resolution Asian monsoon changes. More importantly, our record has comparable counterparts with those in Antarctic/Greenland ice core records on the millennial to sub-centennial timescales, indicating a global feature imprinted in the monsoon changes. Further comparison of climatic sequences on the rising limbs of the NH insolation indicates recurrences of the millennial-scale events in transitions on different timescales.

Results and discussion

Reconstructed YX δ18O record and its climatic interpretation

We determined 9 dates using 230Th dating techniques and established the chronologies of all samples by using an objective age model algorithm (see Supporting Materials and Methods, Fig. S1); and in total, we measured 826 stable isotopic values. Our high-resolution record then covers the latter part of the Marine Isotope Stage (MIS) 3, including DO events 6 to 9 and HS4 as found in Greenland ice cores (Fig. 1; Supplementary Data 1). During the entire growth interval, the average of δ18Ocalcite values is −8.6‰, featured with a relatively positive-valued period from 39.8 to 38.1 ka BP that outstands the whole record. δ18Ocalcite record has a 4‰ range from −10.1‰ to −6.1‰, superimposed by large-amplitude millennial-scale fluctuations and sub-millennial-scale oscillations, strikingly similar to other speleothem records from the same cave and other caves in Southern China (Fig. S2). This indicates that speleothem δ18Ocalcite is consistent with insignificant kinetic fractionation35 in Yongxing Cave. Although we have only 9 dating results, HS4 event could be well-constrained. Compared with most of the previously-published records (Fig. S2), our record has a long and continuous temporal coverage, high resolution and good dating constraints, displaying good potential in investigating climatic dynamics.

Fig. 1: Climate indices from 42 to 32 ka BP during the MIS3.
figure 1

a δ18Oice record from NGRIP ice core on the GICC05 chronology (purple)1,86. b Pa/Th ratios and error bars (SD) from the North Atlantic, indicating the AMOC strength (brown)62. c Our δ18Ocalcite record (orange) overlapped with δ18Oice record from the Antarctic EDC ice core on the AICC2012 chronology (blue)7,87. Numbers indicate DO events, AIM events and HS4. Orange dots indicating 230Th dating results and SD errors. Pearson correlation coefficient is calculated between our δ18Ocalcite and EDC δ18O record, denoted on the bottom-left corner.

We here regard the speleothem δ18Ocalcite as a proxy of the EASM intensity following several lines of reasoning. Firstly, a 3-year monitoring work in Yongxing Cave shows that cave dripwater and modern speleothem calcite mainly inherit δ18O signals in precipitation, especially the wet-season recharge from May to September36 when ~70% of the annual precipitation is received (Fig. S3a). Correlation between the amount-weighted annually-averaged precipitation δ18O and the annual summer rainfall ratio at the study site reaches 0.37 (n = 38, p < 0.01) (Fig. S3b), also indicating that δ18O at the cave site is modulated by the EASM-related rainfall. Secondly, comparison of the YX92 δ18O record37 from the same cave with historical flood/drought index series reconstructed in the Yangtze River Valley38 as well as the long-term EASM index record39 shows good correspondence and correlations during the overlapped interval extending back to 1785 AD, with lighter δ18O consistent with more flood events and higher EASM index (strong monsoon), and vice versa (Fig. S4). In sum, from the view of modern climatology, speleothem δ18O is controlled by summer monsoon rainfall δ18O to a large extent36. Furthermore, both geochemical proxies and petrographic analyses of YX55 sample from the same cave, which grew during 65 and 40 ka BP, provide strong evidence that positive shifts in calcite δ18O correlate to dry conditions controlled by weak monsoon40. It is found that during HS intervals, the deposition rate of stalagmite was relatively low, along with strong prior calcite precipitation processes (indicated by high trace metal ratios), and the corresponding lithological section is composed of dense and fibrous minerals; these features are opposite from those during the DO Interstadials40. Besides, our previous review41 on the cave records in Asia and modeling shows that two mechanisms affect stalagmite δ18O: changes in the fraction of monsoon rainfall in annual totals and changes in the amount of rainout between tropical sources and cave sites. The former is caused by changes in the seasonal migration of the sub-tropical jet and the dominant latter one involves changes in rainout from both the Pacific and the Indian Ocean sources. Therefore, we suggest that speleothem δ18O represents the EASM intensity and the related precipitation, as suggested in previous studies18,21,23,41.

Coupling of monsoon and Antarctic climates

The structure of YX δ18Ocalcite record depicts a variability that captures both Greenland and Antarctic climate signals (Fig. 1), and the millennial-scale relationship has been reported30. The structure of HS4 (AIM8) event is taken as an example here (Fig. 2) because both of its onset and ending are constrained by two 230Th dates. We used the Rampfit method42 to determine the inflections of HS4 transitions (red points in Fig. 2c). The fast 60-year enrichment centering around 39.9 ka BP and the abrupt negative jump around 38.2 ka in our cave record (both δ18O shifts >1‰) can be aligned to the rapid decadal-scale cooling and warming of HS4 in NGRIP δ18Oice record. However, these rapid shifts in the δ18Ocalcite reached only half of the entire amplitude in the processes (Fig. 2a, b). In contrast, changes in the EASM intensity mimic the “trapezoid-shaped” Antarctic temperature profiles (Fig. 2g). At the onset of HS4, a gradually positive shift in δ18Ocalcite sustained until 39.5 ka BP, alike the gradual warming in Antarctica which lasted for ~500 years7. Within the HS4, the EASM was weak though, it was recovering at a slow pace, exceeding 1/3 the intensity of the Interstadial 8 onset (Fig. 2c). This precursor recovery supports the finding43 that a CH4 response in boreal wetland regions precedes the rapid onset of DO8. The weakening and recovery processes of the EASM can also be supported by consistent changes in Ca2+ concentration (Fig. 2b), which are tightly related to monsoon intensity via the jet stream44. Meanwhile, Antarctica warmed up at a slower rate (Fig. 2f), entering the second warming phase7. In detail, several centennial-scale changes between monsoon weakening and Antarctic warming are correlated within this stage. At the end of HS4, monsoon experienced twofold intensification around 38.2 ka BP, with a fast process at first (<60 years, δ18O shift >1‰) and a much gradual interval lasting for ~240 years (Fig. 2c). The Antarctic temperature came to a “breakpoint” after the 1700-year warming, as described in ref. 7 (Fig. 2f). The onset of Interstadials in both NGRIP and YX records can be aligned to the “breakpoint” of the Antarctic AIM events7, even if we take the 200-year lead of the Greenland into account5. The similar detailed changes in the δ18Ocalcite are also evidenced by the extensive cave records such as Hulu, Zhangjia, Xianyun and Mawmluh in Asian monsoon domain (Fig. S2). The structures in speleothem records are also captured by the NEEM 17O excess data45 representing low-latitude hydroclimate changes (Fig. 2d).

Fig. 2: Detailed view of the HS4 (AIM8) Structure.
figure 2

a δ18Oice (purple)1 and b dust Ca2+ (yellow)44 records from NGRIP ice core on the GICC05 chronology86. c YX175 δ18Ocalcite record and dating results with errors (orange). d NEEM 17O-excess record (blue line and dots)45. e CH4 records from WDC ice core on the WD2014 (green)5,51. f CO2 record from Siple Dome (brown)74. g δ18Oice records from EDC ice core (navy blue)7 on the AICC2012 chronology87 and WDC ice core5 (light blue) on the WD2014 chronology51. Black dashed lines indicate the onset and the end of the HS4 in NGRIP δ18Oice record, and the red one indicates the abrupt shifts in proxies. Five bold black polylines depict the structure of HS4 (AIM8). Phase division in Antarctic records in f follows7. Duration of rapid shifts in proxies are denoted.

In general, our monsoon record exhibits the same duration, pattern and relative magnitude as the Antarctic record from AIM6 to AIM8 (Fig. 1), and the “trapezoid-shape” structure of HS4 (AIM8) is also recognized in AIM events 6 and 7 in YX record (Fig. 3), indicating that even a weak AIM can affect the EASM to some extent. Besides, YX record reproduces the Antarctic temperature variations on finer timescales, featured by the striking similarity at around 33.4, 33.7, 34.9, 35.7, 36.1, 38, and 39.9 ka BP on the speleothem chronology (yellows bars in Fig. 3). This finding could shed light on a close dynamic teleconnection between the SH oceanic-atmospheric dynamics and the NH monsoon intensities on different timescales. The Intertropical Convergence Zone (ITCZ), with the northerly ITCZ corresponding to the stronger monsoon status46, is prone to move towards the warmer hemisphere in response to changes in cross-equatorial temperature gradients47 caused by the AMOC intensity48,49 as well as the Southern Ocean temperature50. During Stadials, Antarctic warming is coherent with temperature increase over the SH, thus leading to a southerly shift of the ITCZ and the weakening EASM. Besides, large amounts of moisture and latent heat export into inland China from the remote South Indian Ocean rely on the cross-equatorial airflow between the Mascarene High in the South Indian Ocean and low-pressure system over the Asian continent51. The cross-equatorial airflow carries 87% of the latent heat from the South Indian Ocean into the Indian sub-continent52, and further northward into southern China, as a primary moisture source (53%) for summer precipitation at Yongxing site (Fig. S5). Being the most dominant extratropical annular mode in the SH, the Antarctic Oscillation (AAO) controls the interannual variability of the Mascarene High53. The AAO involves a seesaw pattern between the pressure levels in the SH midlatitude and polar regions; and a positive AAO is characterised with low-pressure anomalies over the south pole and high-pressure anomalies over middle latitudes54. There are strong correlations between the AAO, the Mascarene High and the EASM53,54. Generally, a positive AAO in boreal spring is correlated with temperature increases in Antarctica and an intensification in the Mascarene High during boreal spring through summer, which is followed by a weakened EASM53,54. This seasonal sense of AAO variability could be extrapolated to the Antarctic Centennial Oscillation (ACO) which is a paleoclimate precursor of the contemporary AAO55. Were it so, the atmospheric propagation from the SH could cause similar variations in our EASM record even on the centennial timescales, operating in a way similar to the present day.

Fig. 3: The “Trapezoid-style” EASM changes during the AIM events.
figure 3

We aligned the weak monsoon events in YX (orange lines) and AIM events (including AIM6, AIM7 and AIM8 events) in the EDC δ18Oice (ac) and WDC δ18Oice (df) records (blue lines). The comparison is within the AICC2012/WD2014 chronology’s uncertainties, 500~900 years for AICC2012 and 350~400 years for WD20145,7,51,87. Strong positive Pearson correlation coefficients are found of YX with EDC record (r = 0.55, n = 750, p < 0.01, Fig. 1c, with the EDC record tuned older by 340 years), and with WDC record (r = 0.74, n = 788, p < 0.01, Fig. 1c, with the WDC record tuned older by 200 years). Yellow bars indicate centennial to sub-centennial fluctuations in three records.

Despite a confounding effect of the two polar climates on the EASM pattern, the SH climates could shape tropical hydroclimatic changes, including sub-tropical monsoons, into the SH pattern on the millennial to sub-centennial timescales. Hence our cave record captures global signals on the millennial to sub-centennial timescales, given that Antarctic climates have wider imprints on geological archives13,14,30.

Climate analogs and implications for ice-volume terminations

During the last glacial period, three rising limbs of the NH insolation at around 15, 39, and 60 ka BP56 correspond to T1, MIS4/3t and HS4/DO8t, respectively. Previously, a climate sequence similar to that of the T1 is observed at MIS4/3t27,28. Regardless of different changing amplitudes in orbital configurations (including ice volume and insolation, Table S2), we find similar climatic fluctuations and feedbacks in the glacial-interglacial transition (T1), the sub-glacial and sub-interglacial transition (MIS4/3t) and millennial-scale sequence (HS4/DO8t) (Fig. 4 and Table S2). Bipolar temperature changes were all anti-correlated and phase-locked on the millennial timescale in the Stadial-Interstadial-Stadial sequences (Fig. 4c, h). Monsoonal records also show a strong resemblance of the millennial sequence during the HS4/DO8t and the MIS4/3t to the HS1, Bølling-Allerød and Younger Dryas events in the T1, and the sub-millennial oscillations superimposed on the three Interstadials (Fig. 4a). These are supported by atmospheric CH4 changes which are mechanistically-related with low-latitude hydroclimate17,57 (Fig. 4d). After removing temperature- and seawater-related deviations, we find nearly-similar millennial-scale δ18Ocalcite amplitudes across T1 (3.4‰), MIS4/3t (3‰) and HS4/DO8t (3.7‰) (Fig. S6 and Table S2), indicating recurrences of monsoonal variabilities in spite of the changing boundary conditions. In particular, the EASM variations during the HS4/DO8t are highly analogous to those in the T1 in terms of general trend and internal structures (Fig. 4a). For instance, the EASM intensity decreased during the early phase of HS1 and HS4, and then gradually increased during the second phase. Along with the phase transition, rapid monsoon weakening at 16.1 and 39.5 ka BP occurred within 20 years41,58 and 25 years, respectively. The rapidity of the 25-year shift by 1‰ is supported by nearby annual layers at around 39 ka BP (Fig. S1 and Methods), as well as the 0.7‰ increase in 10 years induced from the highly-resolved Xianyun Cave record59 (Fig. S3f). The similar feature is also found in DO8 Stadial and Younger Dryas, in terms of the internal centennial-scale oscillations and the much gradual onsets relative to their ends.

Fig. 4: Geological records across T1, the MIS4/3 transition and the HS4/DO8 transition.
figure 4

From top to bottom are: a Chinese stalagmite δ18Ocalcite records27,29 and this study (orange), b solar insolation at 65°N (grey)56, c NGRIP δ18Oice record (purple)1, d atmospheric CH4 concentrations from WDC ice core (green)5,16, e Pa/Th record from the North Atlantic (gold)62,88, f Opal record from Southern Ocean (blue)63, g atmospheric CO2 concentrations from ice cores in Antarctica (brown)16,73,74, h Antarctic temperature record from EDC ice core (blue)7,87 and i Northern Red Sea sea level record (black)89. Periods of millennial climatic events are denoted by colored bars. Three episodes of Bølling (B), Old Dryas (OD) and Allerød (A) within the Bølling–Allerød period are marked, and corresponding periods are coded for the MIS4/3t (middle panel) and the HS4/DO8t (right panel). Colored triangles mark centennial-scale abrupt jumps in CO2 and CH4. Arrows below (a) and (h) indicate changing trends of the EASM and the Antarctic temperature. The timing and duration of HS1, 4 and 6 are from ref. 90 and references therein. HS1 and HS4 include two bars, indicating periods of initial meltwater pre-events from the European ice sheet (light gray bar, early HSs) and final NH ice sheet collapse (gray bar, late HSs), respectively66,67.

Analogous millennial-scale variations in trend, amplitude and internal centennial-scale structures of the monsoonal activities indicate the same mechanism operating throughout, which are likely related to behaviors of the ice sheet and the oscillator mode of the AMOC. During the insolation rise, extensive ice sheet collapsed (Fig. 4i) due to its inner instability or changes in the ocean subsurface temperature60,61, leading to the stagnation of the AMOC as indicated by low Pa/Th values62 (Fig. 4e). This causes a southward movement in ITCZ, NH monsoonal weakening (Fig. 4a) and an intensification of the upwelling63 around the Antarctica (Fig. 4f), accounting for the CO2 release from the Southern Ocean. However, NH monsoon intensified and the EASM-related rainfall increased upon the recovery of the AMOC, possibly due to the Agulhas salt leakage64 or changes in the Atlantic salinity65. Meanwhile, opal flux was reduced and atmospheric CO2 was decreased by the decelerated SO upwelling during the Interstadials (Fig. 4f, g). Furthermore, the meltwater pulse from the European ice sheet during the early portion of HSs and the subsequent NH ice sheet collapse during the late portion of HSs66,67, which contributed to a much weaker AMOC during the second phase of HSs than during the early phase (Fig. 4e), could have caused the similarly twofold monsoonal phases and humidity changes during the HS141 and HS4 (Fig. 2). Our comparison in Figs. 3 and 4 show similar climate sequences during the last glacial period and the last deglaciation, indicating that climatic self-similarity is also applicable for glacial climate changes on different timescales (Fig. 4). Strong coupling of the climate system helps to disentangle the complex mechanism for the ice-age terminations.

Over the past decade, millennial-scale climate oscillations have been proposed to provide additional forcing to promote deglaciation by supplementing the astronomical forcing18,19,23,24,25. One recent study even addresses that millennial feedbacks in the atmosphere/ocean control the timing, structure and evolution of glacial terminations20. We agree with the view that the bipolar seesaw mechanism6 responsible for the changes between Interstadials and Stadials also operated across the termination19,23. This mechanism exerts strong influence on the EASM, with weak AMOC status causing decreased fraction of summer monsoon rainfall in the East Asia, decreased amount of rainout from sources to cave sites and less moisture transport to the high latitudes. According to calculations of δ18Ocalcite values by Rayleigh equation41,68 and assumptions similar to those in Fig. S6, the remaining fractions of the original oceanic water vapor were 51% during HS1 and 42% during the Bølling period for Hulu site, 56% during HS4 and 39% during DO8 for Yongxing site, 49% during HS6 and 37% during DO17 for Wulu Cave, respectively. These results indicate that rainfall from the tropical ocean to the cave sites are relatively lower by ~20% during HSs than DO warm phases. Hence, as revealed by Chinese speleothem δ18O records, the millennial-scale component might only contribute to T1 by ~60%, indicating that mechanisms other than bipolar seesaw might operate. Generally, the theory of bipolar seesaw mechanism focuses on two aspects of the ocean’s role in the climate system, including the transport of heat in the Atlantic Ocean and the storage of heat in the deep ocean6, and the transitions between different overturning states occur over a multi-centennial timescale69. However, this theory could only explain the smooth transitions in climates related with the oceanic processes, but could not answer the abrupt changes on the sub-centennial timescales. CO2 and its positive feedbacks are extremely important in amplifying the initial warming during terminations18,19,70. Large short-term oscillations of CO2 could enable tipping points to be reached rapidly, causing severe changes in steady-state climate70. Previous studies find that three remarkable CO2 jumps contributed to more than 50% of the entire CO2 rise (80 ppm) during the last deglaciation16, and suggest the potential role of the AMOC, terrestrial carbon reservoir, continental shelf flooding in influencing these sudden changes15,16,71,72. Here, we suggest that movements of ITCZ, changes in monsoon intensity and continental components could also contribute.

During T1 and HS4/DO8t, several sub-centennial abrupt CO2 increases with amplitudes over 10 ppm occurred around 11.6, 14.7, 16.2, 38.2, and 39.5 ka BP (Fig. 4g, triangles), although they have different amplitudes in various ice cores35,72,73. Two CO2 shifts at 16.2 and 39.5 ka BP were consistent with the decadal monsoon weakening, and left no imprint in the Antarctic temperature or the AMOC records (Figs. 2 and 4). The synchronicity of the 25-year monsoonal deterioration, the 43-year and 60-ppmv CH4 overshoot, and the abrupt CO2 jump within 60 years in Byrd, Siple Dome and WDC ice cores73,74 around 39.5 ka BP (Fig. 2c, e, f) implies a fast atmospheric circulation reorganization on the decadal timescale, including monsoon rainfall decrease in δ18Ocalcite by 1‰, a southward movement of the ITCZ and a strengthening of the SH westerly jet46,74. Analogous relationship is also found during HS1 around 16.2 ka58. Global model results show that the strengthening of SH westerlies could induce a multi-decadal CO2 outgassing from the Southern Ocean, as suppressed by the southerly ITCZ and the weakened Hadley Cell75,76. Variations in continental ice volume and vegetation could also be the possible agents for these two multidecadal-scale CO2 shifts. As the large reservoir of carbon, forests store 90% of the total carbon in terrestrial ecosystems77 and a substantial global terrestrial carbon sink would slow the rate of CO2 increase and thus climate change78. Thirty-one percent of Earth’s total forest area is found in Asia77 and thus influenced by the Asian Monsoon. The vegetation degradation during the HSs, especially due to centennial rapid NH ice sheet collapses41 and the deteriorated monsoon conditions, could have hampered the CO2 uptake by photosynthesis. Instead, the other three rapid CO2 jumps were connected to monsoon intensification with increased rainfall, the abrupt warming in the NH and increases in CH4 (Fig. 4). The abrupt warming in the NH was suggested to lead to massive permafrost thawing, activating a long-term immobile carbon reservoir72. But a recent study shows that climate wetting reduces permafrost thermal responses to warming, which is especially evident in the arid and semi-arid zones79. Therefore, the role of permafrost thawing still needs to be verified and the monsoon circulation needs to be considered. In short, while the AMOC shutdown leads to the millennial-scale CO2 outgassing during the HSs, climatic components on the continent, including ice sheets, monsoon rainfall and forests, induce multi-decadal atmospheric CO2 changes which should not be neglected when considering the causes for ice-volume terminations and future simulation work could help in testing their roles.


230Th dating

Yongxing Cave is ownerless and not located within a protected nature reserve, and no federal or municipal permissions were required for collection. Sample YX175 was found broken in the cave at collection. It has a length of 165 mm, and was halved along the growth axis and polished before measurements. Nine powder subsamples (each ~100 mg) were obtained by drilling on the polished section along the growth axis with a carbide dental burr for 230Th dating. 230Th dating work was performed at the Isotope Laboratory, Xi’an Jiaotong University. We followed chemistry procedures to separate U and Th for dating80. An isotope-dilution method using in-house 229Th–233U–236U spike was employed to determine U–Th isotopic ratios and concentrations. Nine sets of U and Th solutions were analyzed on a multi-collector inductively coupled plasma mass spectrometry (MC-ICP-MS) (Neptune-plus; Thermo-Finnigan) with an Aridus II desolvating nebulizer. Ion beams were measured in peak-jumping mode on the secondary electron multiplier behind the retarding potential quadrupole and followed similar procedures as described in previous study81. The instrumentation, standardization, and half-lives were reported82,83. Uncertainties in U–Th isotopic data were calculated at the 2σ level, including corrections for blanks, multiplier darknoise, abundance sensitivity, and contents of the same nuclides in the spike solution. Corrected 230Th ages assume an initial 230Th/232Th atomic ratio of 4.4 ± 2.2 × 10−6 and the values for a material at secular equilibrium with the bulk earth 232Th/238U value of 3.8. Most samples have high 230Th/232Th ratios and thus minor corrections. These 230Th dates have typical age uncertainties of less than 112 years (Table S1), all in stratigraphic order.

Oxygen isotope analysis

Subsamples (each 50 μg) for stable isotope analysis were shaved at a resolution of 0.1 mm from along the central growth axis with a knife. Every second subsample (n = 826) was measured, using a Finnigan-MAT 253 mass spectrometer fitted with a Kiel Carbonate Device at the School of Geography, Nanjing Normal University, China. CO2 was released from the carbonate by reaction under vacuum with 105% H3PO4 at 90 °C. Each gaseous sample was scanned six times, and calibrated against a reference CO2 gas. All results are reported in parts per mil (‰) relative to the Vienna Pee Dee Belemnite (VPDB) standard and have 0.06‰ at the 1σ level. Repeated analyses of an international standard (NBS19; δ18O = −2.20‰; VPDB) between every 10 subsamples was used to check the long-term reproducibility.

Chronology and annual layer counting

An age model was then developed using Oxcal Project due to its better fitting with 230Th dates and the estimated errors84,85 (Fig. S1), yielding a time range from 40.89 to 33.08 ka BP (present = 1950 AD). The average temporal resolution reaches approximately 10 years. Furthermore, between 81 mm and 82 mm depth and 130 mm and 131 mm depth of our sample, 52 ± 5 and 55 ± 3 of couplets of light/dark layers can be identified per millimeter under optical microscope. If these are annual bands, they would indicate a growth rate of 1.7–2.1 mm/century (Fig. S1), which agrees reasonably well with our estimate of 18.2–19.5 mm/ka from the bounding U/Th dates (at depths of 61 mm, 86 mm, 124.5 mm and 141.5 mm).