Introduction

A basic understanding of the timing and magnitude of natural variability of tropical rainfall in the past is critical to place present trends in context and to validate climate model performance1. Much of modern society directly or indirectly relies on the consistency of seasonal rains for agriculture in tropical monsoon settings2,3. Reliable and well-constrained hydroclimate records from the tropics, and from the heavily populated Australasian monsoon domain in particular, are necessary for understanding processes driving monsoon strength and deep convection over the Indo-Pacific Warm Pool over multiple timescales. While ice cores, lake sediments, and tree-rings make up a large proportion of terrestrial paleoclimate records at the higher latitudes, speleothems provide exceptionally long, highly resolved and well-dated land-based archives of glacial–interglacial (G–I) rainfall variability at tropical latitudes4,5,6,7,8,9,10,11,12,13. At present, over 350 speleothem δ18O records have been published for monsoon domains in Australasia and the Americas14 (Fig. 1).

Fig. 1: Global distribution of speleothem δ18O and Mg/Ca records and season-specific rainfall.
figure 1

The global map is accompanied by insets: a South American monsoon region and b Australasian monsoon region. These areas were selected to illustrate speleothem δ18O record density and Mg/Ca presence. The yellow star marks the location of the Sulawesi speleothem δ18O and Mg/Ca records for this study. Global speleothem locations are from the SISAL V2 database14. The number of speleothem entities associated with each site is indicated by colour and size of circles. Black dots denote sites with Mg/Ca data reported in SISAL V2. Because the northern Borneo sites are in close proximity to each other, resulting in overlapping dots, the number of stalagmite records is indicated with a line to each site (inset, b). In addition, a Mg/Ca marker was added over northern Borneo as the data exists85 but is not recorded in the SISAL V2 database. Tropical rainfall for June–August (JJA, green scale) and December–February (blue scale) represents corresponding regional summer monsoon precipitation. Darker shades signify average monthly rainfall exceeding 500 mm. Precipitation data are from WorldClim 2.1 and span 1970–2000 at 5 min resolution86. The map was created using QGIS software and country boundaries are based on the geometry from EBM v2020 scale 1:60 million (EuroGeographics and UNFAO87).

Full size image

Early speleothem δ18O records for the core area of the Australasian monsoon were generally interpreted to reflect rainfall amount and relevant moisture sources (e.g., summer monsoon intensity and El Niño Southern Oscillation), consistent with the amount effect7,9,15,16. As the number of paleoclimate records for near-equatorial monsoonal settings continues to grow, alongside ongoing monitoring programmes and climate modelling, it is clear that speleothem δ18O and rainfall δ18O in these areas are influenced by site-specific factors and integrated changes in atmospheric circulation that may not always reflect rainfall amount17,18,19,20,21,22. Therefore, in tropical monsoon settings, the ideal situation would be to utilise a multi-proxy approach capable of separating the amount effect from rainfall δ18O influenced by source moisture and oxygen-isotope fractionation along transport pathways. A better understanding of the utility of speleothem δ18O, by far the most commonly used proxy for paleomonsoon reconstructions, will improve our ability to identify the drivers of past monsoon rainfall variability. Glacial–interglacial transitions provide an opportunity to understand how globally significant monsoon systems respond to periods of rapid warming. Information gained from tropical speleothem δ18O combined with proxies sensitive to local rainfall can assist in attributing major changes in rainfall amount to a particular shift in the regional hydrological system.

Tandem measurements of speleothem Mg/Ca and δ18O offer the possibility of decoupling local rainfall amount from variations in rainfall δ18O due to changes in regional atmospheric circulation and moisture source4,23,24,25. Prior calcite precipitation (PCP) is recognised as a primary control on speleothem Mg/Ca variability and is controlled by effective infiltration (i.e., karst infiltration rate and local rainfall amount)26,27 and by the pCO2 content of cave air28,29,30. During periods of reduced rainfall, the infiltration rate is reduced, and water remains in the karst zone for extended periods. These conditions can cause increased void space in the karst and CO2 degassing from seepage waters, which enhances the likelihood of calcite precipitating along flow pathways prior to the dripwater reaching a stalagmite26,31. The pCO2 content of the air in the cave and void spaces drives the degassing process and is influenced by soil pCO2, atmospheric pCO2, and cave ventilation30,32. Since calcite precipitation discriminates against both Mg and Sr in favour of Ca (i.e., Mg and Sr partition coefficients are less than one26), PCP causes Mg/Ca and Sr/Ca in seepage waters upstream of the stalagmite to increase. Mg/Ca and Sr/Ca ultimately precipitated on the stalagmite will be elevated accordingly. The preference for Ca to precipitate out of solution during PCP drives coherency among stalagmite Mg/Ca and Sr/Ca and the slope of the ln(Sr/Ca) versus ln(Mg/Ca) regression line can serve as a test for the occurrence of PCP33,34,35. As speleothem δ13C has been shown to record changes in vegetation and soil pCO236,37, comparison with speleothem δ13C can further distinguish primary environmental or climatic signals preserved in each proxy. Given the complexity of these interconnected processes, it is optimal to use multiple speleothems and a multi-proxy approach to produce accurate palaeohydrological records.

Few studies have reported on paired records of δ18O, δ13C, and trace elements for speleothems that span multiple G–I cycles4,24,38. Our study site in southwest Sulawesi, Indonesia, located in the core area of the Australasian monsoon, is ideally positioned to test the utility of speleothem δ18O as a tropical rainfall indicator on G–I timescales (Fig. 1). Here, we use Mg/Ca (and Sr/Ca) in four southwest Sulawesi stalagmites spanning two G–I cycles to determine the suitability of δ18O as a tropical rainfall indicator. We show that Sulawesi stalagmite Mg/Ca and Sr/Ca are primarily sourced from the limestone bedrock and subsequently altered by PCP. Variations in Mg/Ca and Sr/Ca of Sulawesi stalagmites with relatively slow growth are most likely driven by rainfall amount (i.e., PCP driven by infiltration, water residence time and void-space availability), whereas fast-growing stalagmites appear to be more strongly influenced by soil-karst processes (i.e., PCP driven by the influence of soil pCO2 on upper void spaces and conduits exposed to root respiration). The coherent pattern of variability between the slow-growing Mg/Ca and all four Sulawesi speleothem δ18O records makes it possible to robustly interpret changes in local rainfall through major climatic transitions. Our approach confirms a major increase in rainfall amount in the Australasian monsoon domain at glacial transitions that coincides with the largest shift in speleothem δ18O.

Study site and speleothem analysis

Stalagmites were collected from Abadi Cave in southwest Sulawesi, Indonesia (lat. 5.0°S, long. 119.7°E; elevation ~300 m above sea level). This region of southwest Sulawesi has highly seasonal rainfall, with the Indo-Australian summer monsoon delivering ~80% of annual rainfall between November and March. Mean annual temperature is 26.5 °C. High-elevation topography to the southeast of the study site, including Mount Lompobatang (elevation 2,874 m), prevents much of the austral winter monsoon rainfall (June–August), associated with southeast trade winds, from reaching the cave site12.

Abadi Cave is ~320 m long and one of many systems carved within the Eocene to middle Miocene Tonasa Limestone Formation that covers much of the southwestern region of Sulawesi39. The ground surface above the cave is characterised by karst and lowland intermediate forest with thin soil profiles40. Abadi Cave has minor vertical changes of ±15 m and variable ceiling height. Although the cave’s chambers are extensively decorated, active stalagmite growth has yet to be identified.

All stalagmite samples (Fig. 2) were collected ~250 m from the cave entrance. Stalagmites AC11-03 and AC11-02 grew in a large ‘basement chamber’ (~20 m across), while AC09-04 and AC11-06 were collected from a small ‘side chamber’ nearby and 10 m above the ‘basement chamber’. Both chambers are isolated from the main entrance by a narrow crawl-through passageway. Stalagmite AC09-04 was found broken at the time of collection in 2009. The other three stalagmites were collected in situ in 2011 based on results from reconnaissance U-Th dating of mini-cores41 drilled in 2009. Limestone bedrock material was collected near the entrance of Abadi Cave in 2011 (Supplementary Methods).

Fig. 2: Abadi Cave stalagmites analysed for this study.
figure 2

Photos of polished surfaces with sample milling tracks (black lines). U-Th date locations (black circles) are shown for each stalagmite. Note the uniform growth structure and clearly visible laminae on the slower-growing stalagmites, AC09-04 and AC11-06. The faster-growing stalagmites, AC11-02 and AC11-03, tend to be broader and less uniform (varying diameter), with subtle and intermittent laminae. AC11-03 contains growth unconformities (yellow bars); these segments are excluded from the analysis. Three U-Th dates associated with these unconformities (yellow circles) are excluded from the final age-depth model.

Full size image

Age models

The age-depth profiles of the Abadi Cave stalagmites are constrained by 34 high-precision U-Th dates ('Methods' and Supplementary Data 1 and online repository). Collectively, the geochemical records encompass ~100 ky of past climate, dating between 381 to 330 ky BP and 232 to 173 ky BP. Initial age-depth models for each stalagmite were constructed with the U-Th dates and Bacon42, a Bayesian age-depth modelling tool implemented in geoChronR43 to produce weighted mean age-depth models with 95% confidence intervals for each stalagmite (see 'Methods' and Supplementary Figs. S1 and S2).

The new stalagmite δ18O timeseries for Sulawesi is well replicated, with common patterns clearly identified across sections of overlap (Fig. 3 and Supplementary Fig. S2). Therefore, the Bacon age models were optimised using linear interpolation between visual tie-points in the δ18O records. Sections of the Bacon age models were aligned by taking the average age of shared excursions and incorporating these sections into the original Bacon age-depth model. Importantly, all averaged tie-points and optimised age models fall within the age-depth uncertainties established using Bacon (Supplementary Fig. S2). Because the individual stalagmite δ18O, δ13C, Mg/Ca and Sr/Ca records are generated from the same sample aliquot, these age-depth models apply to each geochemical record described below. The age models show that growth rates for AC11-06 and AC09-04 (from the ‘side chamber’) are relatively slow (0.017 and 0.011 mm/y, respectively). In contrast, AC11-03 and AC11-02 (from the ‘basement chamber’) grew about four times faster (0.042 and 0.064 mm/y, respectively) with maximum growth occurring at deglacial transitions.

Fig. 3: Stalagmite δ18O, Mg/Ca, Sr/Ca, δ13C and growth rates.
figure 3

a Stalagmites AC11-06 and AC11-03 spanning Sulawesi termination IIIa (S-TIIIa) and b AC09-04 and AC11-02 spanning Sulawesi termination IV (S-TIV). Geochemical data for fast-growing stalagmites (AC11-03, AC11-02) are shown in green, slow-growing stalagmites (AC11-06, AC09-04) are shown in blue. Stalagmite δ18O has been corrected for the effect of ice volume45 and stalagmite δ13C has been corrected for the effect of atmospheric CO253; the uncorrected records are shown in grey (see Supplementary Methods for details of these corrections). Sulawesi glacial terminations (determined by prominent shifts in Sulawesi stalagmite δ18O, Supplementary Methods) are indicated by blue bars. Colour-coded circles at the top of each panel show U-Th dates and their uncertainties (2σ). The y axes for geochemical profiles are inverted to show wetter conditions upward.

Full size image

Coupled analysis of δ18O, δ13C, Mg/Ca and Sr/Ca

A total of 1253 stable-isotope measurements (δ18O and δ13C) and 683 paired Mg/Ca and Sr/Ca measurements were made on aliquots of calcite powders micro-milled along the central growth axis of each stalagmite (see 'Methods'). Mg/Ca and Sr/Ca were measured via inductively coupled plasma atomic emission spectroscopy on every second sample for AC09-04 and AC11-06 and nearly every sample for AC11-02 and AC11-03.

Measurements of Mg/Ca and Sr/Ca were made on a sample of surface bedrock limestone to determine the relationship between bedrock and stalagmite values (see Supplementary Methods). Four spot-check analyses of Mg/Ca in the bedrock sample range from 12.1 to 13.2 mmol/mol, which is ~20 times higher than in the stalagmite samples. Bedrock Sr/Ca ranges from 0.24 to 0.28 mmol/mol, ~10 times higher than in the stalagmite samples. The large difference is due to the strong discrimination against Mg and Sr during the precipitation of stalagmite calcite from dripwaters31.

Results and discussion

Geochemical and growth-rate profiles for the four Sulawesi stalagmites are presented in Fig. 3. Stalagmite δ18O values range between −10.2‰ and −4.9‰ after correction for the effect of ice volume44,45 (see Supplementary Methods). The effect of temperature change on speleothem δ18O at our site is deemed to be negligible because the range of G–I change in δ18O is much larger than can be explained by temperature12. The older stalagmites, AC09-04 and AC11-02, have δ18O values that are consistently ~0.8‰ lower than the younger stalagmites, AC11-06 and AC11-03. Stalagmite δ13C values range between −14.0‰ and −5.2‰ after correction for atmospheric pCO2 (see Supplementary Methods).

Mg/Ca and Sr/Ca for the younger stalagmites (AC11-06, AC11-03) range from 0.25 to 1.0 mmol/mol and 0.009 to 0.025 mmol/mol, respectively. The older stalagmites (AC09-04, AC11-02) have a slightly larger range of variability, with Mg/Ca values between 0.32 and 1.14 mmol/mol and Sr/Ca values between 0.013 and 0.044 mmol/mol. Due to the low concentrations of Sr in our samples, nearing instrumental detection limits, as well as limited consensus around Sr partition coefficients in the literature, we focus on Mg/Ca for comparative analyses with δ18O.

Karst recharge and PCP

Speleothem Mg, Sr and Ca are typically derived from the dissolved limestone bedrock and secondary minerals along flow pathways, but can also be sourced from the soil and atmosphere26,32. Bedrock dissolution, and associated incorporation of Mg, Sr and Ca into the seepage water, is driven by soil CO2 and groundwater infiltration rates46,47. Mg/Ca and Sr/Ca in speleothems are sensitive to changes in Ca from the feeding dripwaters through processes associated with soil and cave air CO2 (soil-karst interactions) and water–rock interactions along flow pathways33,34,48, referred to here as infiltration-PCP. If PCP occurs upstream of the cave drip site, the Mg/Ca and Sr/Ca will increase in the dripwater as Ca is preferentially lost from solution. Calcite precipitating from these waters on the speleothem surface will also be similarly co-enriched in Mg and Sr. PCP can occur when infiltration rates are low, drip intervals are long, or when cave CO2 is low; these processes and combinations of processes can differ depending on the individual dripwater flow pathways feeding a stalagmite30,46,49,50.

Sulawesi stalagmite Mg/Ca and Sr/Ca positively and significantly co-vary (r2 = 0.57, P < 0.001, n = 651) and demonstrate a constant slope when compared in cross-plot. These features indicate that PCP is a primary driver of Sulawesi stalagmite Mg/Ca and Sr/Ca variability27,34. Using k-means cluster analysis, we identify two significant groupings of the Mg/Ca and Sr/Ca data for the Sulawesi stalagmites (Fig. 4). The Cluster 1 group has relatively low Mg/Ca and Sr/Ca values, supporting reduced PCP, while Cluster 2 has higher values, suggesting enhanced PCP. Cluster 1 mainly comprises the fast-growing stalagmite data and younger section of AC11-06 (212–173 ky BP), where Mg/Ca and Sr/Ca deviate from the rest of the stalagmite record (Fig. 4). This offset in AC11-06 may derive from wetter conditions that reduce the occurrence of PCP and potentially increase the chance of mixing water reservoirs with different trace element ratios (e.g., soil chemistry, re-routing of drip pathway). Thus, Mg/Ca is less effective as a hydroclimate indicator at times when the epikarst is nearing a saturated state26,31,32.

Fig. 4: Mg/Ca and Sr/Ca cluster analysis and ln–ln PCP test.
figure 4

Elemental ratios for a slow-growing stalagmites and b fast-growing stalagmites. c Comparison of ln(Mg/Ca) and ln(Sr/Ca) with regressions and bedrocksp alongside the theoretical PCP slope of Sinclair et al.34. The Sulawesi average PCP slope is calculated using the average of each stalagmite regression. Slopes shown for individual records were calculated from resampled data. Regressions on original (unevenly spaced) data made no significant difference in slopes (see Supplementary Fig. S3 and Supplementary Table S1). Data pairs in a section of AC11-06, with offset Sr/Ca (grey asterisks), show limited PCP and are not included in the average Sulawesi PCP regression. k-means cluster analysis of Mg/Ca and Sr/Ca for all Sulawesi stalagmites: Cluster 1 (blue ellipse, 1.5σ) grouping near average bedrocksp (yellow square with uncertainty range) is interpreted as minimal PCP while Cluster 2 (orange ellipse, 1.5σ) has evolved away from bedrocksp towards higher values, indicating enhanced PCP. Large uncertainties on bedrocksp correspond with the uncertainties in the Sinclair PCP slope equation used to determine the Sulawesi partition coefficient, KdSr (Supplementary Methods). K-means cluster analysis (k-means clustering), ellipse determination (ELLIPSATE) and linear regression (Linear Regression plot with Confidence Intervals) were performed and plotted in Matlab (https://au.mathworks.com/matlabcentral/fileexchange/).

Full size image

Measured Mg/Ca and Sr/Ca values for Abadi Cave bedrock and estimated partition coefficients34,35,51 are used to calculate approximate Mg/Ca and Sr/Ca values for stalagmite calcite sourced entirely from the host rock, defined here as bedrocksp (see Supplementary Methods and Supplementary Fig. S3). Calculated bedrocksp values plot alongside stalagmite Mg/Ca and Sr/Ca ratios associated with Cluster 1 (Fig. 4). Proximity of Cluster 1 data to the calculated bedrocksp implies limited evolution away from the original bedrock composition and thus minimal PCP. Cluster 2 contains higher Mg/Ca and Sr/Ca values, which have evolved further away from the predicted bedrocksp. These higher values indicate increased PCP, likely in response to reduced groundwater recharge and rainfall amount at the surface of the cave.

Given that Mg/Ca and Sr/Ca increase predictably as PCP proceeds27,31, the ln–ln relationship between Mg/Ca and Sr/Ca data can be used as an additional test for the occurrence of PCP (Fig. 4). Sinclair et al.34 developed a mathematical model of PCP showing that ln(Sr/Ca) versus ln(Mg/Ca) produces a positive vector of constant slope that falls within a range of 0.75 to 1.01 (i.e., 0.88 ± 0.13). The average ln–ln slope of Sulawesi Mg/Ca and Sr/Ca data is 0.84 ± 0.13, which is within range of Sinclair’s theoretical PCP slope and well within other reported ranges35. Each Sulawesi stalagmite regression is slightly offset, indicating that the Mg and Sr content of individual stalagmites may be influenced by distinct dripwater flow pathways46. The average ln–ln regression of Sulawesi Mg/Ca and Sr/Ca was calculated using resampled data (determined by the median resolution of each record) to obtain evenly spaced increments. The average regression slope for the raw data is also within the resampled PCP range (Supplementary Table S1). The younger section of AC11-06 is excluded from these analyses due to the notable Sr/Ca offset occurring within the same stalagmite.

PCP determined by flow pathways

The two main drivers of PCP are (1) groundwater infiltration and availability of air-filled voids within the karst aquifer (infiltration processes) and (2) pCO2 content of air in the cave and void spaces (soil-karst processes)28,30,31,47. Individual drip pathways and flow rates can determine which of these processes prevail for site-specific dripwaters46,48,49. While rainfall amount and infiltration through the karst influence the extent of void space where PCP can occur, large G–I shifts in global temperature and CO2 likely had a major effect on soil and cave air pCO2 via vegetation productivity, microbial activity, and plant root respiration37,52,53. Modern studies show that elevated mean annual temperature54 and atmospheric CO2 accelerate biogeochemical cycles and increase soil pCO255,56.

Fracture and conduit flow pathways allow for seepage waters to be in contact with void air pCO2 that may be more directly influenced by soil pCO231,57. Here, the soil-influenced pCO2 gradient between seepage water and the surrounding air of upper void spaces may drive PCP (Supplementary Fig. S4). For example, depending on the saturation state of seepage waters, void spaces with higher pCO2 relative to seepage waters would prevent the occurrence of PCP, while void spaces with pCO2 lower than seepage waters would increase PCP. In addition, because Ca prefers to be in a solid phase during dissolution and precipitation, corrosive seepage waters associated with higher soil pCO2 may prompt greater dissolution of the bedrock and increase Ca ion release into solution30,57—this process is linked to trace element partition coefficients and thus ratios evolve in a fashion that resembles PCP33,51. Seepage waters flowing to stalagmites along a more direct fracture flow pathway would be more likely to retain this higher Ca (and thus lower trace element ratios), while longer diffuse flow pathways would provide greater opportunity for PCP (Ca loss), especially when infiltration rates are low.

Soil-karst PCP processes are likely to be recorded in Mg/Ca and Sr/Ca of stalagmites that formed from dripwaters with a more direct path from soil to stalagmite. Infiltration-driven PCP processes, resulting from residence time of seepage water and contact with cave and void-space air, are more likely to be imprinted in stalagmites fed by longer dripwater flow pathways. The δ13C signal from root respiration and vegetation productivity is likely to be retained in stalagmites regardless of the flow pathway, especially if this signal is much larger than the PCP influence on δ13C.

Optimal stalagmites for recording hydrology

Although both fast and slow-growing stalagmites have a PCP slope, Mg/Ca in fast-growing stalagmites remain relatively low, near bedrocksp (Cluster 1), indicating limited PCP. The fast-growing specimens (~0.04-0.06 mm/y), are characterised by flat-topped growth laminations and relatively wide diameters that vary with height. In addition, both fast-growing specimens have ‘coffee-coloured’ calcite, indicative of organics and greater connectivity to the soil58. Irregular features along the flanks of these stalagmites indicate high dripwater Ca concentration and high drip rates58. Given the above, these stalagmites are likely fed by variable drip rates, with a combination of fracture flow and diffuse flow48. Flow pathways for the fast-growing stalagmites may be subject to higher hydraulic loading, fewer air-filled voids and less water–rock interactions, resulting in a stronger chemical imprint from the soil and vadose zone.

The slow-growing stalagmites (AC09-04 and AC11-06) demonstrate a clear trend away from bedrocksp towards more highly evolved Mg/Ca and Sr/Ca (spanning both Clusters 1 and 2). This suggests that PCP plays a larger role in their trace element chemistry than for fast growers. The slow-growing (~0.01–0.02 mm/y) Sulawesi stalagmites are uniformly narrow in diameter with conical growth laminae and likely fed by seepage flow with relatively slow and regular drip intervals48. Dripwaters feeding the slow-growing stalagmites are hypothesised to travel through longer flow pathways, with increased residence time and exposure to void space, leading to enhanced PCP during periods of low infiltration (Supplementary Fig. S4).

When the Mg/Ca and Sr/Ca timeseries are considered for slow and fast-growing stalagmites that grew over the same time period, there are distinct differences in trends, indicating that different processes are impacting site-specific flow pathways (Fig. 5). We note that the coeval stalagmite δ18O (and δ13C) records show excellent reproducibility and appear to be unaffected by differences in site-specific flow pathways. The slow-growing stalagmite Mg/Ca timeseries share the G–I baseline shift with stalagmite δ18O, transitioning to lower values near the end of glacial terminations. Conversely, Mg/Ca values in the faster-growing stalagmites are more coherent with stalagmite δ13C, atmospheric CO2 fluctuations, and global temperature59,60,61, where lower Mg/Ca and δ13C coincides with higher CO2 and temperature (Fig. 5).

Fig. 5: Sulawesi stalagmite proxies driven by rainfall amount versus temperature and soil CO2.
figure 5

a Sulawesi stalagmite δ18O records (corrected for ice volume) and slow-growing stalagmite Mg/Ca. b Fast-growing Sulawesi stalagmite Mg/Ca and all stalagmite δ13C records compared with Vostok ice core measurements of atmospheric CO2 (grey) and EPICA Dome C deuterium (black) plotted on the AICC chronology59,60,61 (lower panel). The gradual trend in the fast-growing stalagmite Mg/Ca (green) corresponds with changes in stalagmite δ13C (corrected for atmospheric CO2 effect), atmospheric CO2 and global temperature/δD (grey arrows). This Mg/Ca response is attributed to soil-karst processes, driven by vegetation changes (green bars). The step-change in the slow-growing stalagmite Mg/Ca (blue) corresponds with the deglacial transition in δ18O and is attributed to water-karst PCP processes driven by infiltration. Note the pulses during Sulawesi glacial terminations S-TIIIa (215–212 ky BP) and S-TIV (337–335 ky BP), interpreted as a response to increased infiltration (see text for details). The durations of S-TIIIa (215–212 ky BP) and S-TIV (337–335 ky BP) are based on all four stalagmite δ18O change points and coinciding change points in slow-growing stalagmite Mg/Ca (blue bars, Supplementary Fig. S5).

Full size image

Given the faster growth rates and shared trends with temperature and atmospheric CO2, we postulate that fast-growing stalagmite Mg/Ca was predominantly imprinted by large (G–I) changes in soil CO226,30,48. The shared trends between fast-growing stalagmite Mg/Ca and coeval δ13C likely exist because both proxies are influenced directly or indirectly by soil pCO2. Carbon in speleothem calcite is sourced primarily from soil CO2 and bedrock carbonate62. Thus, speleothem δ13C (that has been corrected for the effect of atmospheric pCO2 on 13C/12C fractionation in C3 plants53) can be interpreted to reflect variations in vegetation type, productivity, and root respiration above the cave37,53,62. Although beyond the scope of this study, integrated modelling coupled with in-cave monitoring experiments would provide more insight into these soil-karst processes.

Because PCP associated with the slow-growing Sulawesi stalagmites is more likely to be influenced by rainfall amount and effective recharge, Mg/Ca in these stalagmites is better suited for hydrological reconstructions and comparison with δ18O. The exceptional length, reproducibility, and covariation of the slow-growing stalagmite records allow us to assess the validity of the heavily relied upon δ18O rainfall proxy across G–I transitions. We therefore focus on the slow-growing stalagmites for comparison of Mg/Ca and δ18O.

Coupled Mg/Ca- δ18O assessment of G–I transitions

The Mg/Ca values of the slow-growing stalagmites (AC11-06, AC09-04) share a prominent shift to lower values that coincides with the G–I transitions to lower δ18O. The good agreement between the paired proxies provides evidence for the preservation of a locally-relevant climatic signal63. Additionally, the agreement between all four Sulawesi δ18O records and PCP-sensitive Mg/Ca of the slow-growing stalagmites indicates that both proxies are responding to rainfall amount at the Sulawesi glacial termination26. The onset of major G–I shifts in Sulawesi δ18O and Mg/Ca are identified using Bayesian change-point analysis64 (Supplementary Fig. S5). Stalagmite δ18O undergoes a 3‰ shift to lower values over a 2–3 ky period, from ~337 to 335 ky BP and ~215 to 212 ky BP. These shifts occur in the later stages of global deglaciation, after weak monsoon intervals (WMIs) identified in China5,6, and are referred to here as Sulawesi glacial terminations: S-TIV and S-TIIIa. The Mg/Ca in the slow-growing stalagmites also undergoes a baseline shift to lower values during the δ18O transitions. The response of Mg/Ca, as δ18O begins to decrease, is characterised initially by brief pulse(s) to higher values, followed by a rapid transition to low interglacial values, coinciding with the completion of the δ18O shift. The common trend of the two proxy datasets is consistent with drier glacial conditions with enhanced PCP, followed by increased rainfall and reduced/absent PCP during interglacials.

There is an obvious lag between the Mg/Ca and δ18O transitions over the Sulawesi glacial terminations recorded by the slow-growing stalagmites. Using cross-correlation analysis, we find that the correlation between the two proxies, across the deglacial transitions, is optimised when Mg/Ca is shifted to older by a few hundred years. The Mg/Ca glacial transition lags δ18O by approximately ~200–300 years across S-TIV (342–332 ky BP) and ~500–700 years across S-TIIIa (220–210 ky BP) (see Supplementary Fig. S6). Because the Mg/Ca and δ18O measurements were made on the same sample aliquots, the lag cannot be due to chronological offsets between the two records. The most likely explanation for a lag between the two proxies is that it takes longer to fill voids and saturate diffuse flow pathways (indicated by Mg/Ca) than for seepage water to reach the stalagmite surface itself (indicated by δ18O)26,65. For example, if a relatively dry karst network experiences a dramatic increase in recharge due to increased rainfall, back-filling and saturation of the epikarst may take hundreds of years to measurably suppress the PCP system and dripwater Mg/Ca, whereas dripwater δ18O would be influenced quickly by a change in the δ18O of rainwater. When the karst network finally recovers from an extended glacial period of low infiltration, the replenished water moving through the system is likely to entrain trace metals that have built up along flow pathways during the dry periods and deposit them onto stalagmite surfaces26. In addition, cation exchange (Na for Mg) and the release of Mg from soils above the cave has been observed when rainfall increases following a prolonged dry period66. Thus, the brief pulse(s) towards higher Mg/Ca values occurring over both Sulawesi terminations may signal the initial back-filling of the epikarst and/or cation exchange from soils. The final Mg/Ca shift to interglacial values at each termination coincides with the conclusion of the δ18O transition, however, the Mg/Ca pulses over S-TIIIa are more apparent and occur over a longer period of time (~2.5 ka) compared to S-TIV. This extended delay, prior to Mg/Ca achieving interglacial values, could be due to the amount and regularity of increased rainfall over the transition at S-TIIIa or a more extensive flow pathway network feeding the stalagmite.

Sulawesi terminations and global implications

Changes in deep atmospheric convection over the maritime continent are hypothesised to dominate hydroclimate over the last glacial period, resulting in reduced convection associated with the Australasian monsoon during the last glacial, followed by rapid monsoon strengthening in the final stages of deglaciation12. The Sulawesi paired stalagmite δ18O and Mg/Ca records spanning TIV and TIIIa, combined with the last glacial Sulawesi stalagmite δ18O record12, provide strong evidence for reduced rainfall and a weakened Australasian summer monsoon spanning glacial maxima, deglacial onset, and WMIs in China5,6. These prolonged periods of reduced rainfall in the northern (e.g., WMIs) and southern hemisphere sectors of the Australasian monsoon also correspond with glacial drying in northern Borneo7,8,10. These periods of reduced rainfall occur despite rising global temperatures and warm pool sea surface temperatures and under various phases of insolation forcing (Fig. 6 and Supplementary Fig. S7). Previous work has proposed a southward shift in the Australasian monsoon rainbelt and ITCZ during cold intervals in the northern hemisphere and glacial terminations5,6,7,8,10,11,12,13,67. This southward shift is attributed to the collapse of the northern hemisphere ice sheets, associated slow-down of Atlantic meridional overturning circulation, increased interhemispheric temperature gradients, and a poleward shift of the austral westerlies. Together, these events prompt CO2 upwelling from the Southern Ocean and enhanced heat transport across the planet68,69,70,71.

Fig. 6: Comparison between Australasian monsoon and glacial–interglacial climate transitions.
figure 6

a EPICA Dome C (EDC) estimated temperature difference (relative to the last 1000 years)60,61. b Sea surface temperature (SST) from the Ontong Java Plateau (ODP 806B, cyan)88 and West Pacific Warm Pool (MD97-2140, purple)89. c Global sea level stack90. d Suborbital Asian Monsoon (AM) component (Δδ18O, detrended from the composite stalagmite δ18O record from China by subtracting summer insolation)6 showing weak monsoon intervals (WMIs, yellow bars). e Composite speleothem δ18O records from China5,91,92,93 and 21 June insolation at 65oN94. f Speleothem δ18O records from Sulawesi: S-TIV, S-TIIIa (this study) and S-TI12 and 21 January insolation at 5oS94. g Composite speleothem δ18O record from Flores11,13 h Mg/Ca records from slow-growing Sulawesi stalagmites (this study). S-TIIIa and S-TIV (blue bars), are based on all four Sulawesi stalagmite δ18O change points and coinciding change points in slow-growing stalagmite Mg/Ca.

Full size image

Reduced rainfall in Sulawesi throughout WMIs is surprising since the study site is located 5oS of the equator, where an increase in rainfall might be expected as the Australasian monsoon shifts southward. Instead, it appears that the tropical rainbelt and associated summer monsoon was likely shifted further south beyond Sulawesi and/or convection was greatly reduced over the IPWP. Sea level-driven changes in landmass configuration of the maritime continent are thought to contribute to a reduction in IPWP convection (drying) during the glacial maxima72,73,74, thus it is possible that both a southward shift in the rainbelt and reduced convection influenced Sulawesi rainfall. A southward shift of the tropical rainbelt in response to interhemispheric temperature gradients at the onset of WMIs in China5,6 is supported by stalagmite records spanning the IPWP and Australasian monsoon region, where wetter conditions develop in Flores11,13 and northern Australia67 and drying occurs in Borneo7,8,10 and Sulawesi (this study and ref. 12). Model and proxy results indicate that a southward shift of an anomalous anticyclone during the austral summer, as a result of the slow-down of the Atlantic meridional overturning circulation, resulted in drying over the IPWP maritime continent during Heinrich Stadial 1 (at the onset of WMI-I)75. The increase in Sulawesi rainfall and return of IPWP hydroclimate to interglacial conditions coincides with rising sea levels, likely as a result of increased convection as the maritime continent floods72,73. The establishment of interglacial conditions in the Australian summer monsoon also coincides with the stabilisation of interglacial SSTs and global temperature, therefore a sea level and/or temperature threshold may be required to trigger reinvigoration of IPWP convection and associated rainfall.

The Sulawesi stalagmite δ13C and fast-growing stalagmite Mg/Ca support an increase in vegetation productivity at the onset of global deglaciation and WMIs that corresponds with changes in temperature and atmospheric CO2. This transition occurs ~4–7 ky earlier than Sulawesi rainfall intensification. Similar disconnects between rainfall amount and vegetation change were identified in New Zealand over the Last Glacial Maximum, where multi-proxy stalagmite data support an increase in rainfall above the cave occurring 2 ky after a substantial post-glacial increase in forest extent65. A stalagmite record from northern Borneo also shows evidence for decoupling between G–I temperature and hydroclimate over the last glacial termination, where cave temperatures track changes in ice core CO2, corresponding to Antarctic warming, whereas regional hydroclimate responds more strongly to millennial-scale northern hemisphere cooling events76. A disconnect between tropical temperature and hydroclimate across G–I transitions highlights the importance of distinguishing the prevailing climatic driver when interpreting paleoclimate proxies from these regions.

There is a notable mismatch in the Sulawesi stalagmite Mg/Ca and δ18O from 375–355 ky BP when PCP is present (Supplementary Fig. S8). The mismatch in the coupled record may indicate changes in the seasonal balance of rainfall, where variations in δ18O are driven by changes in rainfall source and trajectories, rather than rainfall amount77,78. Seasonal changes in the position of the ITCZ and Australasian monsoon moisture-transport pathways are accompanied by large latitudinal gradients in the δ18O of rainfall16,22. The unexpectedly high rainfall δ18O in Sulawesi around 375–355 ky BP corresponds to lower amplitude changes in summer insolation and relatively high monsoon rainfall in China5. Thus, a northward bias in the position of the ITCZ or reduced seasonality (a change in the dominant source of Sulawesi rainfall) could decrease the contribution of isotopically light austral summer rainfall and tip the oxygen-isotope water balance in southwest Sulawesi toward isotopically heavier austral winter rainfall. In addition, variability in the Pacific Walker Circulation could influence regional rainfall δ18O through changes in zonal moisture-transport pathways and rainfall amount78. Therefore, unexpectedly high rainfall δ18O in Sulawesi over this period could also be attributed to a weaker or displaced Pacific Walker Circulation that does not necessarily reflect changes in local rainfall amount10,17.

Conclusions

The results presented here demonstrate the potential for multi-proxy records of stalagmite δ18O, Mg/Ca, Sr/Ca and δ13C to provide tighter constraints on interpretations of hydroclimate variation in the deep tropics, where records of δ18O alone can be ambiguous. A key finding from the coupled Mg/Ca and δ18O records for Sulawesi is that glacial cycles in the core of the Australasian monsoon domain are characterised by a prominent and relatively rapid shift from dry glacials, coinciding with WMIs in China, to wet interglacials during the final stages of global deglaciation. This increase in Sulawesi rainfall is decoupled from global temperature and tropical SSTs and occurs ~4–7 ky after temperatures begin to rise. Differences between Mg/Ca and δ18O in sections of the record where PCP is evident indicate that Sulawesi δ18O may not always strictly record rainfall amount. At these times, rainfall δ18O likely retains a combined signal of rainfall amount and the source or regional system from which that rain formed. With further development, discrepancies in coupled records of Mg/Ca and δ18O for near-equatorial settings could become a powerful tool for distinguishing the relative contributions of summer and winter rainfall (with contrasting δ18O) to the monsoon water balance.

An important goal for the paleoclimate community will be to establish criteria for identifying cave sites containing speleothems with PCP signals driven by groundwater infiltration. Our results indicate that slow-growing and uniform stalagmites, located in sections of the cave with diffuse flow, are the most likely candidates to contain PCP signals indicative of rainfall amount. Our results highlight that soil processes influencing pCO2 of void space can drive Mg/Ca variability in fracture-fed stalagmites, thus recording climatic changes that are unrelated to infiltration-driven PCP. Stalagmite δ13C, responding to vegetation productivity and root respiration above the cave, may co-vary with Mg/Ca that has also been influenced by soil-related processes. An international effort to perform trace element analysis on existing high-quality, well-dated and relatively slow-growing stalagmites that have corresponding δ18O measurements could provide a wealth of knowledge on tropical climate dynamics.

Methods

Sample preparation

Following collection, stalagmites were cut in half and slabbed at ~15-mm thickness along the central plane of growth. Our method for sequential extraction of carbonate micro-samples from the specimens was adapted from existing techniques32,79. Briefly, calcite powders for stable isotopes (δ18O, δ13C) and minor elements (Mg/Ca, Sr/Ca) were milled from the central growth axis of each stalagmite using a Sheridan GCM Micromill. A 1-mm-diameter end-mill bit was used to collect high-resolution samples and a 2-mm-diameter end-mill bit for low-resolution samples. High-resolution samples were milled continuously from stalagmites AC09-04 and AC11-06 at 0.8 mm and 1.2 mm increments, resulting in sample resolutions of 90–130 years and 40–180 years, respectively. The low-resolution samples were milled at 10 and 30 mm increments, yielding a sample resolution ranging between 80–740 years for AC11-03 and 460–1890 years for AC11-02. High-resolution sampling was not performed on AC11-03 and AC11-02, as these samples were primarily intended for replication and high resolution was not necessary. Four spot samples were milled for Mg/Ca and Sr/Ca analysis from clean, pre-milled surfaces of a single piece of surface limestone bedrock material.

Uranium–thorium dating

Material for uranium–thorium (U-Th) dating was extracted from pre-cleaned surfaces of slabbed stalagmites using an impulse high torque air-drill fitted with a miniature circular saw blade. The saw blade was continuously lubricated with demineralised water while cutting. For each date, four cuts were made into the material allowing for the extraction of a solid block weighing 130–200 mg. Numerous reconnaissance U-Th dates for this work were measured at the National Taiwan University’s High-Precision Mass Spectrometry and Environment Change Laboratory (HISPEC). The majority of the 34 dates for the age models used in this study were performed at the University of Minnesota Trace Metal Isotope Geochemistry Group (TMIG). The methodology for measuring U and Th isotopes by multi-collector inductively coupled plasma mass spectrometry (MC-ICP-MS) is described in refs. 80,81 and ref. 82. All U-Th isotope results and age determinations can be found in Supplementary Data 1 and in the online repository. U-Th dates, relative to CE 1950, were corrected using the initial 230Th value that minimised age reversals (Supplementary Fig. S1). Age errors (2σ) on U-Th dates range from 0.5 to 1.8%.

Age models

We used Bacon software42, implemented through geoChronR43, to construct initial age models for the stalagmite records in this study (Supplementary Figs. S1 and S2). Bacon is an age-depth modelling approach that uses Bayesian statistics to reconstruct accumulation histories based on U-Th dates and uncertainties. Millions of Markov Chain Monte Carlo (MCMC) iterations are used to estimate accumulation rates between depths, using information provided by U-Th dates and input parameters, such as estimated mean accumulation rates, memory, and sample thickness/slices. Bacon output consists of a weighted mean age-depth model with 95% confidence intervals. The Bacon age models were then optimised using linear interpolation between visual tie-points in the δ18O records (Supplementary Figs. S1 and S2). The tie-points and optimised age models for each stalagmite fall within the 95% confidence intervals of the original Bacon age-depth models.

Stable-isotope analysis

Measurements of stalagmite δ18O and δ13C were made at the Australian National University (ANU) using Finnigan MAT-251 and Thermo MAT-253 mass spectrometers, each coupled to a Kiel microcarbonate preparation device. To ensure consistency among runs, in-run measurements of the NBS-19 standard (δ18O = −2.20‰; δ13C = −1.95‰) were made for every 5–8 samples, complemented by less frequent measurements of NBS-18 (δ18O = −23.0‰; δ13C = −5.0‰). Results are reported relative to Vienna Peedee Belemnite (VPDB). The in-run analytical error for NBS-19, averaged for each run, is ±0.05‰ for δ18O and ±0.02‰ for δ13C (n = 375, 1σ). The mean standard error for duplicate analyses on sample aliquots was ±0.03‰ for δ18O and ±0.01‰ for δ13C (n = 64). Quantifying sample measurement reproducibility demonstrates that variability down to around 0.1‰ in the stalagmite δ18O and δ13C profiles is not due to measurement uncertainty. The powder samples for AC09-04 were alternately measured between instruments (i.e., odd-numbered samples on the MAT-253, even-numbered samples on the MAT-251). No significant offsets for δ18O and δ13C measurements were detected between the instruments for the purposes of this study.

Minor element measurements

Measurements of Mg/Ca and Sr/Ca in stalagmite samples were measured on the same sample powders used for the stable-isotope analyses using Varian Vista AX inductively coupled plasma atomic emission spectroscopy (ICP-AES) at the ANU. For the stalagmites milled at high resolution, every other sample was measured on the ICP-AES to preserve half of the sample material for future analyses. The ICP-AES analysis was based on methods described in Schrag (ref. 83) where a ~1 mg aliquot of carbonate powder was dissolved in 5 mL of 2% HNO3. Each sample solution underwent ten replicate analyses, with the final mmol/mol ratio reported as the average of all ten replicate ratios. The internal precision of replicate measurements averaged over all runs is ±1.0% for Mg/Ca (mmol/mol) and ±1.7% for Sr/Ca (mmol/mol).

To standardise the ICP-AES data and account for potential instrument drift, standard-sample-standard bracketing was employed. ANU coral bracketing standards (Mg/Ca = 5.754 mmol/mol, Sr/Ca = 8.983 mmol/mol) were prepared from high purity CaCO3, SrCO3, and an AccuTrace 1000 ppm Mg standard, dissolved in 2% HNO3. Bracketing standard trace element ratios were previously validated by thermal ionisation mass spectrometry (TIMS) against an accepted ANU in-house standard. Bracketing standards had an average in-run precision of ±0.45% for Mg/Ca and ±0.29% for Sr/Ca (RSD, average of 64 analyses per run) over the course of 5–10 h runs. Because the coral bracketing standard has higher concentrations of Sr than the Sulawesi stalagmite calcite, instrument rinse times between samples were increased to 180 s to avoid memory effects.

An additional synthetic standard with low Sr concentrations similar to those observed for southwest Sulawesi stalagmites (Sr/Ca = 0.02 mmol/mol), was used to test instrument detection limits. The precision for Sr/Ca ratios from the low Sr standard measurements was ±0.43% (n = 20, 1 RSD), thus demonstrating that the instrument is capable of precisely measuring low Sr concentrations. In-run measurements of the JCp-1 standard were also made to check the reproducibility among runs. Measurements for Mg/Ca and Sr/Ca ratios of JCp-1 (Mg/Ca = 4.125 ± 0.028, Sr/Ca = 8.766 ± 0.032, n = 11) are within error of those reported in ref. 84 (Mg/Ca = 4.199 ± 0.065, Sr/Ca = 8.838 ± 0.089), with a measurement precision of ±0.25% (n = 11, 1 RSD).