Introduction

Understanding the variability and spatial distribution of precipitation is not only significant for agricultural production and hence human settlements, but also essential to help project climate change futures1,2. Precipitation conditions in the Chinese monsoon region exist in a “sandwich” pattern, which is characterized by more arid conditions in central China, with wetter conditions in north and south China, and vice versa3. Estimating the geographical range of the sandwich pattern extent under different climate regimes is crucial for further understanding of precipitation variations and their forcing mechanisms. However, it is challenging since existing modern meteorological observed data cannot provide strong enough constraints to reflect the natural variability of precipitation because of their short duration (i.e., <70 years). On the other hand, there is still a lack of regional paleo-hydroclimate records that span larger spatial and temporal scale to expand the range of variability.

Yunnan Province, in southwest of China, is a gate way for the Indian monsoon incursion into China. Some studies investigate regional hydroclimatic changes, using various geological archives, such as speleothems and from lacustrine sediments, and their results show that the Indian summer monsoon dominates the precipitation regime in Yunnan1,4,5. However, north of Yunnan, near the upper reaches of the Yangtze River (Fig. 1), there is still lack of paleo-hydroclimatic records, which limits understanding of the full range of variations and distribution patterns of precipitation in the southwest of China.

Fig. 1: Modern climatology and the locations of other study sites discussed in this test.
figure 1

MJJAS rainfall anomalies (mm/day) during the decaying phase of the 1983,1998 and 2016 strong El Niño events. The monthly spatial rainfall amount data (during the period 1979–2019) were obtained from the ERA5 reanalysis (https://cds.climate.copernicus.eu/datasets/reanalysis-era5-single-levels-monthly-means?tab=overview) with a grid resolution of 0.25° × 0.25°. The scale on the right shows the anomaly represented by different colors. The red star indicates Yingpan Cave, and the dark blue circles represent other study sites mentioned in this study, Xiaobailong Cave1, Shuiyuan Cave34 in central-south Yunnan, and Haozhu Cave8 and Dajiuhu peat9 in mid-lower reaches of the Yangtze River Valley. The dash grey lines in upper and lower indicate Yellow River and Yangtze River respectively.

Full size image

The last deglaciation was a period of major climate changes, including climate evolution from glacial to interglacial stages and included some abrupt climate changes6. Importantly, the period also provides an opportunity to understand variations and distribution patterns of hydrological conditions under significant climate transitions. To this day, some scientists have conducted studies investigating the spatial heterogeneity of precipitation on different time scales during the last deglaciation in the East Asian monsoon region, and reveal opposite patterns between mid-lower reaches of the Yangtze River valley in central China and northern and southern China on millennial to long-term time scales7,8,9. However, these studies on paleoprecipitation anomalies in central China are limited to the area around Dajiuhu peat region in the mid-lower reaches of Yangtze River. Whether a phenomenon similar to the precipitation anomalies in central China exists in other regions during the last deglaciation, is unclear and requires further high-resolution hydroclimatic records for its validation.

In this study, we present speleothem δ18O, δ13C, and Mg/Ca records across the last deglaciation from the Yingpan Cave, located in northeast of Yunnan, southwest of China (Fig. 1). This site on the northern Indo-China Peninsula, has its precipitation dominated by the Indian summer monsoon1, and mainly falls during the summer monsoon season, with 68% of precipitation delivered between June and September. Based on a detailed chronology, this study reconstructs the variations of precipitation δ18O, i.e. Asian summer monsoon intensity, spanning the last deglaciation, and a regional hydroclimate record for southwest China. We aim to use this data to investigate regional hydroclimate changes, connections with the monsoon climate and its possible driving forces, as well the spatial heterogeneity of hydroclimate in southwest China across the last deglaciation.

Results

Chronology

Seven 230Th dates were attained from stalagmite LD1. The dating results show that the concentration of uranium in stalagmite LD1 ranges from 40.9 ppb to 127.9 ppb and the concentration of thorium ranges from 1127 ppt to 17279 ppt (Table S1). Notably, most dates are in stratigraphic order but with large dating uncertainties (Fig. 2A). Therefore, to establish the age model, an integrated method was applied.

Fig. 2: Study sample and obtained data.
figure 2

A The section of stalagmite LD1 from Yingpan Cave. 230Th dates are indicated in red next to the blue drilled slits for U-Th dating and the Libby 14C dates are indicated in yellow next to the purple drilled slits for 14C dating. B The records of LD1, including speleothem δ18O, δ13C and Mg/Ca. C Dating results and age-depth model for LD1. All ages are plotted with 2σ uncertainty as calendar years. The 14C0 ages here represent 14C ages without the “dead carbon” correction, and the 14C10 and 14C50 ages indicate the corrected 14C ages assuming the dead carbon proportion is 10% and 50%, respectively. StalAge models are shown by lint pink (assuming reversal age at the depth of 90 mm) and purple (assuming reversal age at the depth of 68 mm) dash line respectively, and the corresponding shaded areas denote the upper and lower 95%-confidence limits. Tuning point ages and age models are indicated by green dash line and derived from isotope wiggle-match.

Full size image

Five 14C ages were also obtained, and the results were in stratigraphic order within the error terms (Table S2). All the 14C ages were calibrated using CALIB Rev. 8.1.0 by Stuiver and Reimer (2020)10, based on the IntCal20 Northern Hemisphere radiocarbon age calibration curve11. The uncertainties of calibrated ages are no more than 150 years (2σ), and most were only of a few decades (Table S2).

Stable isotope composition and trace elements

According to sampling strategy and multi-method established age models (see below), the mean temporal resolution of our record is ~150 years. The δ18O values of LD1 vary between −7.7‰ and −14.9‰ with a mean value of −11.5‰ during the growth period (Fig. 2B), showing a decreasing long-term trend superimposed by several fluctuations during the growth period. Generally, δ18O oscillated around −10.8‰ from 112 to 154 mm depth. Then, the values increased abruptly at the depth of 101.5 mm, and a heaviest value appeared at the depth of 100.5 mm, and this positive interval spans 4 mm. From there, the record was interrupted by a reduction in values followed by a distinct increase at the depth of 72 mm, and decreased again at the depth of 57 mm.

The δ13C values varied between −6.6‰ and −10.9‰ with a mean value of −9.2‰. Contrary to the stalagmite δ18O record, the δ13C record shows a long-term increasing trend. Nevertheless, we note some significant enrichments and depletions of δ13C were correlated with δ18O variations (Fig. 2B), implying the contemporaneous responses of δ13C to these environmental changes.

As with the stable isotope measurements, the Mg/Ca record also yielded a mean temporal resolution of ~150 years. Notably, the Mg/Ca (mmol/mol) values demonstrated an apparent increasing trend and they fluctuated between 12.88 and 35.03 with several shifts (Fig. 2B), which were similar with changes of stalagmite δ13C values.

Chronology constrained by 230Th and 14C results

The U-Th dating results exhibit significantly larger uncertainties, particularly in the upper portion of the study sample, owing to low concentrations of Uranium and relatively high concentrations of Thorium (Table S1). These factors complicate the establishment of a robust age model and may introduce considerable uncertainty in the initial 230Th correction. Previous studies of speleothems from Bailong (White Dragon) and Xiaobailong (Little White Dragon) caves in Yunnan Province, which were formed during the last deglaciation, have demonstrated that initial 230Th can substantially impact dating results due to the low Uranium and relatively high Thorium concentrations, potentially affecting the fidelity of records on interdecadal timescales12,13. In this study, the measured 230Th/232Th activity ratio in stalagmite LD1 ranges from 14 × 10−6 to 318 × 10−6, and the uncertainties in the uncorrected ages may not be sufficient to cover the actual ages14, indicating that initial Thorium correction for this sample cannot be neglected13,14.

After correcting the measured dates using an initial 230Th/232Th ratio of 4.4 ± 2.2 × 10−6, corresponding to the bulk earth 232Th/238U value of 3.8, most dates are in stratigraphic order within dating uncertainties except for one anomalous age, which should occur either at the depth of 68 mm or 90 mm (Fig. 2C). We employed the StalAge model for age model establishment, as it is the preference for those results with reversed dates and/or where dating uncertainties are relatively large15. At the same time, given the difficulty in determining the exact depth at which the age reversal occurs, two age models were generated for consideration. (Fig. 2C). The results show that LD1 grew from 18785 to 10665 a BP assuming there is a reversal at the depth of 68 mm, and between 18792 and 9784 a BP if reversal occurs at the depth of 90 mm (Fig. 2C). Both models show the sample grew during the last deglaciation. Additionally, as the dating results show that the maximum and minimum of the uncorrected 230Th ages respectively are 13629 ± 399 a BP and 19655 ± 127 a BP, suggesting LD-1 was possibly deposited after 19655 a BP and suggests that the sample ceased growing after 13629 a BP, which are consistent with StalAge results. However, the inability to clarify where the reversal occurs and the large variations in growth rates, indicating that StalAge models are unable to provide an explicit model and are difficult to clarify the existence of growth hiatuses. Determining the precise age model requires integration with other dating methods, such as 14C dating and the wiggle-match of speleothem δ18O records, as demonstrated in previous study13.

Speleothem 14C ages could be affected by dead carbon incorporation from cave bedrock and aged organic matter in the overlying soil system16,17, and the dead carbon proportion (DCP) generally varied along the growth period. As an extreme condition, DCP could range from 0 to 50%16,18. Assuming our study sample was not affected by dead carbon, the calibrated 14C0 ages are 11316, 11881, 12952, 18222 and 18827 a BP. On the other hand, the calibrated 14C50 ages are about 4927, 5415, 6237, 10510 and 11344 a BP, if the DCP reached 50% (Fig. 2C). As such, according to the results of 14C calibrated ages, we assume our study sample was deposited later than 18827 a BP and the deposition layer with the youngest 14C ages was not later than 4927 a BP. Although it is still difficult to anchor the accurate chronology, the possible growth period can be identified as, approximately between 18827 and 4927 a BP, i.e. from the late Last Glacial Maximum (LGM) to the Holocene. Importantly, 14C ages also provide an important clue, that is the real age must be younger than or equal to 14C0 ages. Obviously, the median values of StalAge models almost are older than 14C0 ages, indicating the results of StalAge models are older than actual if the confidence limits were not considered. Notably, two StalAge models were obtained only based on the U-Th dates which were corrected by the initial 230Th with the average 232Th/238U value of 3.8, hence the age model still needs to be further determined in combination with other methods.

The establishment of chronology using wiggle match of speleothem δ18O

As previous work shows, speleothem δ18O records can be a benchmark to tune a chronology model of some speleothem samples which grew during periods, such as the deglacial and with notable abrupt changes13. Because there is a good replication of speleothem δ18O in Asian monsoon regions at millennial to orbital time scales, especially for the deglacial periods, where speleothems were deposited under isotopic equilibrium conditions6,19,20,21. However, before tuning the age model using speleothem δ18O identifying the deposition conditions are prerequisite, to further ensure the fidelity of speleothem δ18O changes for paleoclimate reconstructions.

Following a ‘Hendy Test’22, we drilled 26 sub-samples from four layers and inspected the characters of the δ18O values. Stable values of δ18O are observed on same layer analyses and there is no obvious progressively positive trend from the central growth axis to outer edges (Fig. S1A). In addition, we also calculated the correlation of δ18O and δ13C along the growth axis. The R value shows that δ18O values are negatively correlated with δ13C (r = −0.51, n = 75, p < 0.01) for LD1 (Fig. S1B). The above lines of evidence suggest that the stalagmite LD1 from Yingpan Cave was formed under (near) isotopic equilibrium conditions, and δ18O of LD1 values primarily reflect changes of drip water δ18O and cave temperature (which is close to the annual mean surface temperature)23. However, the impact of temperature on the δ18O of LD1 (the amplitude of ~7.2‰) is not significant enough to be considered a major factor. This is because the temperature-dependent fractionation between calcite and water are about −0.23/°C24,25 and −0.177/°C26, and the amplitude of temperature in the regions within 87°E-116°E and 17°N-33°N is about 3 °C over the past 150,000 years27. Therefore, temperature would only account for that less than 0.7‰ of calcite δ18O changes, and the δ18O of drip water (controlled by local precipitation δ18O in general28) mainly controls the stalagmite δ18O record. The climate at our study site is primarily controlled by the Indian summer monsoon1. Although multiple factors have been employed to interpret values in Chinese speleothem δ18O1,6,29,30,31,32,33, their variations across the Asian monsoon region show a spatial consistency, and reflect summer monsoon circulations19. Hence, the δ18O record of LD1 mainly reflects variations in Indian summer monsoon.

To tune the chronology of LD1, we compared our δ18O with speleothem δ18O records in Yunnan and surrounding areas within the Indian monsoon region (Fig. 3). Using the chronology framework based on the U-Th and 14C dating results, the comparison also shows that our speleothem likely grew from the late LGM to the early Holocene. This is because there is a two-step jump from the old to young parts with an amplitudes of ~2‰ and ~4‰ respectively (i.e. from the most positive position to the second positive excursion, and from the second positive excursion to the most negative part), and is similar with the structure and ranges of those speleothem δ18O records from Xiaobailong Cave1 and Shuiyuan Cave34 in Yunnan Province, and these two positive excursions are said to correspond to two cold periods Heinrich 1 event (H1)35,36 and the Younger Dryas (YD) events (Fig. 3). At the same time, the absolute values of LD1 δ18O during the Holocene, at around ~14‰, also are consistent with those in these three caves during the Holocene (Fig. 3).

Fig. 3: The comparison of speleothem δ18O records in Indian monsoon region.
figure 3

A speleothem δ18O from Xiaobailong Cave in central Yunnan1; B speleothem δ18O from Southeast Yunnan34; C speleothem δ18O from Mawmluh Cave in India37; D speleothem δ18O from Yingpan Cave in northeast of Yunnan (this study); E the tuned δ18O time series of LD1 (this study, the dark blue thick line indicates the 5-points smoothed averages) and the δ18O record from Mawmluh Cave37.

Full size image

Specifically, the transitions during the last deglaciation are used as the anchor to tune the speleothem δ18O record (Fig. 3). Meanwhile, considering the lower resolution of speleothem δ18O records obtained from Xiaobailong and Shuiyuan caves with their larger age uncertainties, the record of speleothem δ18O from Mawmluh Cave (dating uncertainties are lower but with a higher resolution δ18O record, Fig. S2) located in Indian monsoon region37 is used as the benchmark to calibrate our δ18O record. During the last deglaciation, there were three notable abrupt events, i.e. H1, YD and Bølling–Allerød (BA), and make it easier to tune the δ18O from Yingpan Cave. We show five black dotted lines to highlight the anchor points (Fig. 3C), which are the onsets and terminations of those three abrupt events and hence tune the chronology. In addition, an age marked blue in Fig. 3D shows the oldest 14C0 age point (18827 a BP at the depth of 142 mm). The main reasons for choosing these points are: (1) for the five transition points, H1, BA, YD are three successive abrupt events, using the four transition points as the anchors can be effective to distinguish different time periods, and whether or not growth hiatuses occur within the internal of those events would not affect the reliability of distinguishing the time period of these three events; the youngest anchor highlights the onset of the most negative values during the Holocene. (2) the oldest 14C0 age point, was selected as a possible oldest limit to tune, and the other ages were reconstructed by a linear interpolation. Consequently, the growth period of LD-1 is constrained within ca. 18900 to ca. 9000 a BP.

It is noticeable that, due to the larger age uncertainties and the low resolution of our record, establishing a highly resolved chronological framework and identifying potential growth hiatuses is challenging (Fig. 3). Although there seems to be a hiatus at the depth of 100 mm (between ca. 18000 and ca. 14500 a BP) (Fig. 2C), further specific wiggle matching and values of speleothem δ18O, show that the growth rate is merely slow. Meanwhile, the tuning results at a depth of 106 mm is approximately 60 years older than the oldest possible age (14C0), due to the temporal resolution of speleothem δ18O from Mawmluh Cave is about 170 years before 17000 a BP (Fig. S2), and results in a deviation. Nevertheless, since our study primarily focuses on the overall characteristics of the proxies and climate changes on long term trends during the last deglaciation, and given that the wiggle match of speleothem δ18O span all key stages of this period (i.e. early Holocene, YD, BA, H1 and the late LGM), these uncertainties can be neglected. Therefore, LD1 records span from the late LGM to the Holocene, and record abrupt events (at least some parts of each abrupt event) within the last deglaciation.

The interpretation of speleothem δ13C and Mg/Ca

Speleothem δ13C is generally used to reflect the contribution proportion of inorganic carbon contribution from different sources16, changes in types and density of vegetation conditions38,39,40 and local hydroclimate variations41,42. Previous studies showed that the abundance of C4 vegetation with relative positive δ13C values decreased in south China, including Yunnan, from the LGM to the early Holocene43. However, our δ13C increased during this period, indicating vegetation was not the main driver in regulating speleothem δ13C of LD1, hydroclimate conditions should be considered. It has been suggested that a significantly positive correlation between δ13C and trace element ratios (such as Mg/Ca) generally has been used as a criterion to verify the occurrence of prior calcite precipitation (PCP), which is the occurrence of calcium carbonate deposition in seepage system before percolating water reaches the cave due to CO2 degassing and long residence times of percolating water that is associated with dry hydroclimate condition42,44. Under these circumstances, the happens of PCP will cause an obvious reduction of Ca than other trace elements in percolating water and escape of CO2 with lighter isotopic composition42,43, thus increasing δ13C and Mg/Ca values. Therefore, speleothem δ13C and trace element ratios can be used to infer the regional hydroclimatic changes, i.e. higher values indicate dry conditions and vice versa44,45,46. As showed in Figs. 2B and S3, speleothem δ13C positively correlated with Mg/Ca values (r = 0.76, n = 92, p < 0.01), indicating PCP dominates the changes of δ13C and Mg/Ca values. Although the effects of water-rock interaction (WRI) on the variations of δ13C and Mg/Ca cannot be ruled out, the impacts of PCP and WRI are in the same direction47. Longer water-rock interaction enhances the dissolution of host rock and simultaneously causes calcite precipitation before percolating water reaches the stalagmites, and hence increase of δ13C and Mg/Ca values48. Therefore, speleothem Mg/Ca and δ13C reflect regional hydroclimate changes.

Although temperature potentially impact the variations of speleothem Mg/Ca because of the distribution coefficient of Mg is temperature dependent, the large magnitude changes of speleothem trace elements are ultimately controlled by the parent solution composition42, and co-variations of δ13C and Mg/Ca suggest that they are mainly controlled by common factors. Therefore, the effects of temperature on Mg/Ca values can be neglected, because increased temperature generally causes the decrease of speleothem δ13C49. In addition, the appearance of debris could lead to the enrichments of speleothem δ13C and Mg/Ca, however this was not observed (Fig. S4).

Consistent Asian summer monsoon variations

The comparison shows similarities of speleothem δ18O records from Yingpan and Mawmluh37 caves in the Indian monsoon region, and they broadly correlate with records from several thousand kilometers away in the East Asian Monsoon domain, such as Haozhu Cave50, Hulu Cave6 (Fig. 4).

Fig. 4: The comparisons between the speleothem δ18O records in this study site and other records.
figure 4

A speleothem δ18O (Yingpan Cave, this study, the dark blue thick line indicates the 5-points smoothed averages) and the insolation of July at 65°N (grey curve)71; B speleothem δ18O (Mawmluh Cave)37; C speleothem δ18O (Haozhu Cave)50; D speleothem δ18O (Hulu Cave)6; E 231Pa/230Th and εNd profiles from core GGC5 from the Bermuda Rise, 4541 m51,52, and F Greenland average temperature reconstruction ±1 SD uncertainty53.

Full size image

All the speleothem δ18O records show a long-term decreasing trend and are punctuated by two rapid millennial-scale climate events (Fig. 4) which correspond to H1 and the YD. Thus, during the late LGM, from ca. 19 to ca. 17.7 ka BP, the δ18O values indicate a weak Indian summer monsoon. Then, an increase in speleothem δ18O with a heaviest value suggest a weakest monsoon, corresponds to the H1 event. Thereafter, the speleothem δ18O record has a negative shift followed by a positive excursion from 12.9 to 11.7 ka BP, namely, the BA, and YD, respectively. These large changes suggest abrupt Asian summer monsoon events (Fig. 4). The depleted speleothem δ18O values show a strengthened summer monsoon during the Holocene. These traits are also subtly reflected in our record which is dominated by the Indian summer monsoon (Fig. 4A).

Decreasing of speleothem δ18O records broadly follow the increase in Northern Hemisphere summer insolation across the last deglaciation, and the similar rhythm suggests that summer insolation drives the Asian summer monsoon changes on orbital time scales (Fig. 4). Meanwhile, the lag and decoupling of speleothem δ18O with insolation changes during abrupt event episodes, have been generally ascribed to the impacts of ice sheet and Atlantic meridional overturning circulation activities (Fig. 4)2,30,33,37,51,52,53,54,55,56. How this broad consistency in large-scale spatial climate patterns is manifested in local precipitation still needs more evidence to be further clarified.

Hydroclimate changes in southwest China across the last deglaciation

Mg/Ca and δ13C values gradually increased from the late LGM (ca. 19 to ca. 17.7 ka BP) to the Holocene, indicating a long-term drying trend (Fig. 5D, E). In addition, Mg/Ca and δ13C records were significantly punctuated from ca. 15.5 to ca. 11.7 ka BP, indicating that monsoonal events affected hydroclimate conditions in this region. This drying during the last deglaciation has been recorded in other regions, including trace element ratios from the Haozhu Cave8 and the Dajiuhu peat record9 in the mid-lower Yangtze River valley in central China. A recent study for quantitative rainfall reconstruction based on elevated shorelines showed that lake level of Chenghai lake57, in the north of Yunnan, gradually fell from ca. 15 ka BP to the early Holocene (Fig. 5F). All these lines of evidence show a drying climate during the last deglaciation in northeast Yunnan and the mid-lower Yangtze River valley.

Fig. 5: The comparisons of regional hydroclimate in central China and Yunnan.
figure 5

A, B Rb/Sr and LOI by 550 °C/% records of the core LC1 from Dajiuhu Basin9; C Speleothem trace element records from Haozhu Cave8; D, E speleothem δ13C and trace elements records from Yingpan Cave (this study, the red and pink curves indicate the 5-point smoothed averages); F reconstructed lake level of Lake Chenghai57, G black carbon contents from Lake Tengchongqinghai58; H Ti counts from Lake Tengchongqinghai5; I speleothem δ13C record from Xiaobailong Cave1.

Full size image

Since our results align with the dry-wet fluctuations in the mid-lower Yangtze River8,9 (Fig. 5A–E), hints that the precipitation anomaly in the mid-lower reaches of the Yangtze River in the ‘sandwich’ hydroclimatic structure may exist on a larger spatial scale, potentially reaching the northeast of Yunnan during the last deglaciation. However, due to the limitations of our age model, we are unable to examine the characteristics of our records in conjunction with hydrological records from the mid-lower Yangtze River precisely. To fully understand the spatial extent of the ‘sandwich’ structure, additional evidence from this region is needed.

Notably, this gradually dry condition is not common in the Yunnan of southwest China (Fig. 5D–F). The percentage of black carbon and the Ti content in lake sediments from Tengchongqinghai Lake indicate that the hydroclimatic conditions there were gradually wetter from the LGM to the Holocene5,58, and the speleothem δ13C record from Xiaobailong Cave suggests the climate gradually became warmer and wetter (possibly more vegetation cover and more active microbial activity) during the last deglaciation1 (Fig. 5G–I). Thus, a dipolar hydroclimatic pattern is evident between the north and central-southern Yunnan (Fig. 5).

Discussion

The regional hydroclimate records imply that the strength of the Indian summer monsoon was not the main driver of local hydroclimate. Indeed, the hydroclimatic records resemble the variations of equatorial Pacific gradients of Pacific Sea Surface Temperature (SST)59 (Fig. 6A) which could affect the Walker circulation4,60. Thus a decreased zonal SST gradient (El Niño condition) weakened Walker circulation and the descending branch which is generally distributed over northeast India and southwest China, suppressed Indian monsoon rainfall, and vice versa4,60,61. The zonal Pacific SST gradient increased from LGM to the early Holocene (Fig. 6A). However, our hydrological record and the existing records indicate a shift from wet to dry across the last deglaciation.

Fig. 6: The comparison between different records.
figure 6

A Tropical Pacific zonal SST gradient59; B Lake Qinghai Westerly climate index (WI, flux of >25 μm fraction)65; C speleothem δ13C record (Yingpan Cave, this study, the red curve indicates 5-point smoothed averages); D speleothem Mg/Ca record (Yingpan Cave, this study, the pink curve indicates the 5-point smoothed averages); E Speleothem trace element records from Haozhu Cave8.

Full size image

Modern observations and geological records show that the changes of zonal Pacific SST gradient have a significant impact on the spatial distribution of different hydrological conditions in the subtropical Asian monsoon region via the movements of the west Pacific subtropical High (WPSH)57,62. In particularly, paleoclimate studies suggest that the movements of WPSH, in relation to the zonal Pacific SST gradients (ENSO conditions), played a significant role in contributing to a gradual drying in the Dajiuhu Basin9 and stranded lake levels in north Yunnan57 during the last deglaciation. As Fig. 6 shows, wet conditions during the late LGM inferred from lower speleothem δ13C and Mg/Ca corresponding to a decreased SST gradient, and gradually increased speleothem δ13C and Mg/Ca synchronized with elevated SST gradient (Fig. 6A, C and D) during the last deglaciation. It is likely that increased (decreased) zonal SST gradients caused stronger (weaker) equatorial easterly winds, corresponding to higher (lower) SSTs in the tropical west Pacific, which can drive an anomalous anticyclone over the western North Pacific. Further, the WPSH shifted northeastward (southwestward) and resulted in depleted (plentiful) precipitation in our study site and at Dajiuhu Basin.

The shifts of westerlies play an important role in driving precipitation in central China, especially on millennial scale during the last deglaciation8,9,63,64. The changes in temperature gradient between high and low latitude regions can largely regulate the shifts of northerlies and westerlies, as well the duration of Meiyu, and eventually impact the precipitation in mid-lower reaches of the Yangtze River valley8,63,64. Our regional hydroclimate records and those in mid-lower reaches of the Yangtze River valley are broadly consistent with the changes in westerlies on long term trend65, indicating their role in regulating precipitation. However, it was found that the intensity of westerlies slightly enhanced from the late LGM to BA (Fig. 6B), was not synchronous with the changes in the records from our study site and Haozhu Cave (Fig. 6C–E). Thus, the westerlies may not play a key role in regulating the hydroclimate changes on long term trend in our study site.

The regional hydroclimate contrast between the center-south and the north part in Yunnan Region can be ascribed to the different responses to ENSO conditions. The regional hydroclimate changes in central-southern Yunnan are likely controlled by the Indian summer monsoon rainfall associated with ENSO conditions4, however, those in northern Yunnan are primarily modulated by the zonal SST gradient in tropical Pacific water and shifts of WPSH, as well the shifts of westerlies. This could explain why dryer conditions appear in central-southern Yunnan under El Niño-like conditions, due to decreased Walker circulation and the location of the descending branch over northeast India and southwest China, while wetter conditions could be observed in northern Yunnan due to decreased zonal SST gradient (El Niño-like) and a southwestward movement of WPSH.

Methods

Study area and climatology

The Yingpan Cave (27°7′N, 103°23′E) is located approximately 20 km southwest of the Ludian County, Yunnan Province, southwestern China (Fig. 1). The cave formed in the Permian limestone of the Qixia-Maokou Group. According to data from nearest Ludian meteorological station (1981-2010AD), the mean annual precipitation is 825.8 mm, and the mean annual temperature is about 13.4 °C.

Chronology

U-Th dating

The stalagmite LD-1 was collected from deepest part of the Yingpan Cave in 2021, it is ~17 cm in length and ~8 cm in diameter (Fig. 2A). The stalagmite was halved along the growth axis by using a diamond saw and then polished. About 80–100 mg powered subsamples were drilled parallel to the growth layers along the growth axis, and prepared for 230Th dating. The chemical procedures used to separate the uranium and thorium for 230Th dating are similar to those described in Edwards et al. 66, and the measuring protocol is the same as that used by Cheng et al. 67 using a ThermoFisher NEPTUNE multi-collection inductively coupled plasma mass spectrometer (MC-ICP-MS) at the Isotope Laboratory, Xi’an Jiaotong University67. A 230Th/232Th atomic ratio of 4.4 ± 2.2×10−6 was used to correct for the initial 230Th. U-Th isotopic and concentration data, the decay constants and the calculated ages are listed in Table S1 with 2σ error.

14C dating

Five carbonate powder subsamples of approximately 12 mg were drilled from LD-1 using a hand-held dental drill with a carbide bit of 0.5 mm diameter, and the sampling positions of these subsamples are at the depth of 23, 50, 68, 106 and 142 mm. Specifically, the carbonate powders were converted to CO2 using 100% H3PO4 under vacuum at 70 °C, and the resulting CO2 was then converted to graphite with an Fe catalyst68,69 and measured at the Xi’an AMS center facility for 14C analysis70.

Isotopic analyses

The subsamples for stable isotopic composition analysis were drilled at intervals 0.5 mm and 2 mm by using a handheld carbide dental drill, adjusted according to deposition rates. Specifically, the sampling intervals used were 0.5 mm between 94 and 103 mm, and between 58.5 mm and 60.5 mm, and otherwise at 2 mm intervals. In total, 94 sets of oxygen and carbon stable isotope analyses were performed at the Institute of Earth Environment of the Chinese Academy of Sciences. For the stable isotope values of those samples from four individual laminae, seven subsamples were collected from each of three layers, with the exception of the 30 mm depth, where five subsamples were collected due to the limited width of the growth layer. All oxygen and carbon isotope results are reported in delta (δ) notation. The standard results show that the precision of δ18O and δ13C analyses are better than 0.15 and 0.12 (2σ), respectively.

Trace elements analysis

Sampling strategy of those subsamples for trace elements measurements on the same subsamples for stable isotope analysis, and were drilled at the positions parallel to those points used for calcite δ18O and δ13C analysis. After sampling, the subsamples for Mg/Ca analysis were dissolved in about 1 mL of 1 mol/L HNO3, and then analyzed for Ca and Mg by ICP-OES at the Institute of Earth Environment, Chinese Academy of Sciences. Three spectral lines, namely, Ca (317.433 nm), and Mg (279.553 nm) were used in this study. The lab standard N15 was repeatedly measured to monitor the status of the instrument and the results showed the external precision to be 1% for Mg/Ca (8.90 ± 0.09 mmol/mol, n = 138).