Local hydroclimate alters interpretation of speleothem δ¹⁸O records

Abstract

Oxygen isotopes (δ¹⁸O) are the most commonly utilized speleothem proxy and have provided many foundational records of paleoclimate. Thus, understanding processes affecting speleothem δ¹⁸O is crucial. Yet, prior calcite precipitation (PCP), a process driven by local hydrology, is a widely ignored control of speleothem δ¹⁸O. Here we investigate the effects of PCP on a stalagmite δ¹⁸O record from central Vietnam, spanning 45 – 4 ka. We employ a geochemical model that utilizes speleothem Mg/Ca and cave monitoring data to correct the δ¹⁸O record for PCP effects. The resulting record exhibits improved agreement with regional speleothem δ¹⁸O records and climate model simulations, suggesting that the corrected record more accurately reflects precipitation δ¹⁸O (δ¹⁸O_p). Without considering PCP, our interpretations of the δ¹⁸O record would have been misleading. To avoid misinterpretations of speleothem δ¹⁸O, our results emphasize the necessity of considering PCP as a significant driver of speleothem δ¹⁸O.

This study finds that in-cave processes affect speleothem oxygen isotope records. Correcting for these processes improves agreement with other regional records and climate models, providing a more accurate reflection of past hydroclimate change.

Introduction

The effects of karst and in-cave processes on speleothem oxygen isotopes (δ¹⁸O) have been largely overlooked when using speleothems for the reconstruction of past climate variability. When precipitated under equilibrium conditions, the primary controls on the δ¹⁸O of speleothem calcite are cave temperature and drip water δ¹⁸O, which typically reflects the δ¹⁸O of precipitation (δ¹⁸O_p) falling above the cave¹. Karst hydrology can complicate these signals, but these complexities can often be constrained through hydrological modeling and cave monitoring, and typically only introduce drip water δ¹⁸O variability of < 1 ‰^2,3.

In addition to karst hydrology, in-cave processes affect the fractionation of δ¹⁸O between drip water and calcite^4–8, resulting in the disequilibrium precipitation of most speleothem calcites⁹. Among these processes is prior calcite precipitation (PCP), which refers to all calcite precipitation from infiltrating water prior to speleothem formation. A recent modeling study demonstrated that PCP can theoretically offset speleothem δ¹⁸O by several ‰⁴. This finding raises concerns that PCP could obscure temperature- and δ¹⁸O_p-driven speleothem δ¹⁸O variability, making records susceptible to misinterpretation. Given the multitude of published speleothem δ¹⁸O records worldwide¹⁰, there is a critical need to further study this phenomenon.

PCP encompasses all calcite precipitation from infiltrating water prior to the drip water reaching the stalagmite. This includes calcite precipitation in the epikarst, on the cave ceiling, and on an overhanging stalactite or soda straw (Fig. 1)^11,12. Crucially, PCP changes the geochemical signature of drip water. When CO₂ degasses from drip water, lighter isotopes (¹²C and ¹⁶O) preferentially degas, enriching the remaining drip water HCO₃⁻ pool in heavier isotopes (¹³C and ¹⁸O). Cave-analog laboratory experiments show that the calcite that precipitates from the HCO₃⁻ pool reflects this enriched isotopic signature (higher δ¹³C and δ¹⁸O values)^5,13. Similarly, Mg²⁺ and other trace elements (TEs) with a partition coefficient < 1 are preferentially excluded during calcite precipitation¹⁴, with Mg/Ca and other TE/Ca ratios increasing alongside δ¹⁸O and δ¹³C in the remaining solution (Fig. 1). Dry conditions enhance PCP because more air in the karst and slower infiltration rates increase the duration of calcite precipitation, leading to stronger enrichment of heavier isotopes and trace element concentrations in the drip water (higher δ¹⁸O, δ¹³C, and Mg/Ca), and hence the speleothem calcite that precipitates from that water. Thus, PCP-sensitive geochemical proxies often reflect local hydroclimate change.

Fig. 1. Schematic of prior calcite precipitation (PCP) in a cave system.

The left panel shows the flow path of drip water and calcite precipitation along the infiltration pathway. Green arrows denote CO₂ fluxes, and blue dashed lines denote water flow. The right panel shows detailed changes in isotopes (δ¹⁸O and δ¹³C) and trace elements (Mg/Ca) during PCP in (a) the epikarst and (b) the cave ceiling.

While PCP is a well-known control of speleothem δ¹³C and Mg/Ca records^12,14, it is generally not considered when interpreting speleothem δ¹⁸O. Traditionally, H₂O exchange - the isotopic re-equilibration between HCO₃⁻ and the ambient H₂O reservoir - is thought to eliminate the degassing signal (referred to hereafter as the PCP signal) in drip water HCO₃⁻ δ¹⁸O prior to speleothem formation. With enough time, H₂O exchange, also known as buffering, reestablishes the initial pre-PCP δ¹⁸O value of the drip water HCO₃⁻ pool, thereby erasing the PCP signature¹⁵. However, laboratory experiments have demonstrated that H₂O exchange takes much longer than previously estimated¹⁶. Using the revised estimate of the exchange time, recent modeling studies suggest that speleothem δ¹⁸O does preserve a PCP signal when H₂O isotopic re-equilibration in the drip water is incomplete^4,8. Importantly, ref. ⁴ demonstrates that incomplete H₂O exchange is most likely to be preserved by speleothems when PCP occurs on the cave ceiling and/or stalactite. This is because the residence time of drip water in the epikarst far exceeds the time necessary for re-equilibration, whereas the residence time of drip water on the cave ceiling or stalactite may be sufficiently short to maintain isotopic disequilibrium during calcite precipitation.

Although geochemical models successfully simulate the effects of PCP on calcite δ¹⁸O, these effects have only recently been observed in speleothem δ¹⁸O records^7,17. Using a correction method to remove the PCP signal from a speleothem δ¹⁸O record of the Holocene from central Vietnam, ref. ¹⁷ found that PCP obscured the δ¹⁸O_p signal in portions of their δ¹⁸O record. They reinterpreted their record using the corrected δ¹⁸O, which resolved disagreements in their data and led to a deeper understanding of the local and regional components of the monsoon system in Southeast Asia during the Holocene, a relationship still unresolved over glacial time periods.

Here, we build upon this work by examining the impact of PCP on another previously published δ¹⁸O record from central Vietnam¹⁸. Using a geochemical model to remove the PCP signal from the speleothem δ¹⁸O record (see “Methods”), we demonstrate that PCP overprints the expected δ¹⁸O_p signal in the record. After removing the PCP signal, the corrected δ¹⁸O record shows substantially improved agreement with regional δ¹⁸O speleothem records and climate model output, providing a more accurate estimation of past δ¹⁸O_p variability in central Vietnam from 45 – 4 ka.

Results/discussion

Controls on δ¹⁸O_p in central Vietnam

Central Vietnam receives rainfall from two isotopically distinct monsoon systems, the southwest monsoon (more negative δ¹⁸O_p, 30% of annual rainfall from June-August) and the northeast monsoon (more positive δ¹⁸O_p, 47% of annual rainfall from September-November) (Fig. 2)¹⁸. Importantly, large-scale atmospheric processes, rather than local rainfall amounts, control the δ¹⁸O_p of both monsoon systems. Upstream rainout of moisture sourced from the Bay of Bengal and Indian Ocean drives southwest monsoon δ¹⁸O_p^19,20. While northeast monsoon δ¹⁸O_p is not correlated with local rainfall amount²¹, the large-scale mechanisms driving its variability are less understood. Annual δ¹⁸O_p, the signal recorded by the majority of speleothem records in central Vietnam, reflects a combination of rainfall from these two monsoon systems (seasonality), but not local rainfall amount²¹. On modern timescales, the proportion of southwest vs northeast rainfall amount, which is driven by the timing of the SW-NE wind direction shift during the southward migration of the ITCZ, controls interannual δ¹⁸O_p variability²¹. However, less is known about central Vietnam δ¹⁸O_p variability on longer timescales.

Fig. 2. Modern and pre-industrial iTRACE climatology.

a Map of total annual precipitation in Mainland Southeast Asia derived from Asian Precipitation-Highly Resolved Observational Data Integration Towards Evaluation (APHRODITE)⁵⁹. Arrows show wind direction of the SW monsoon and NE monsoon with average δ¹⁸O_p values of JJA and SON precipitation. The orange square denotes the study site location. The average seasonal cycle of (b) rainfall from APHRODITE (dark blue) and iTRACE Pre-industrial simulations (light blue) and (c) δ¹⁸O_p measured from rainfall collected at Phong Nha-Ke Bang Park Headquarters (orange) and from iTRACE Pre-industrial simulations (red). APHRODITE and iTRACE time series taken from grid cell encompassing the study site.

PCP-corrected δ¹⁸O

In this study, we revisit a δ¹⁸O record (~ 50 yr resolution) from a 3.7 m long previously broken stalagmite (HH-1) collected from Hoa Huong cave in Phong Nha-Ke Bang National Park in central Vietnam (Supplementary Figs. S1, S2; 17.5°N, 106.2°E). HH-1 grew continuously from 45 – 4 ka, and the Mg/Ca and δ¹³C records were previously used to investigate local hydroclimate change in central Vietnam¹⁸. A strong correlation between Mg/Ca and δ¹³C indicated an important role of local water balance (precipitation minus evapotranspiration) in driving speleothem geochemistry through the PCP mechanism. Notably, the authors also found that HH-1 δ¹⁸O strongly correlated with both δ¹³C (r = 0.79, p < 0.01) and Mg/Ca (r = 0.87, p < 0.01) records (Fig. 3), suggesting that a common mechanism drove variability in all three proxies. This result was unexpected because δ¹⁸O_p variability in central Vietnam does not reflect local rainfall amount, but rather regional processes such as upstream rainout and seasonality²¹. From this observation, the authors hypothesized that PCP, the primary control on the HH-1 δ¹³C and Mg/Ca records, likely influenced the HH-1 δ¹⁸O record. We test this hypothesis by using the HH-1 Mg/Ca record, cave monitoring data, and a geochemical model to remove the PCP signal from the HH-1 δ¹⁸O record. The corrected δ¹⁸O record (δ¹⁸O_corr) should reflect δ¹⁸O changes driven by regional hydroclimate change (δ¹⁸O_p) rather than local water balance (PCP), giving a history of central Vietnam δ¹⁸O_p variability from 45 – 4 ka.

Fig. 3. Stalagmite (HH-1) proxies from 45 – 4 ka.

a HH-1 δ¹⁸O (orange) and corrected δ¹⁸O (δ¹⁸O_corr, black). The black line is the correction using the best choice parameter configuration (see Supplementary Table S1) and α_{calcite-water} from ref. ⁵³ (see “Methods”). The gray shading denotes the standard deviation of the 1000 Monte Carlo simulations from the sensitivity testing. b HH-1 Mg/Ca (red) and δ¹³C (pink). The black dashed line shows the Mg/Ca threshold used to determine Mg/Ca_i. Beige shading denotes the Younger Dryas and Heinrich Stadials 1–4. The Mg/Ca, δ¹³C, and δ¹⁸O curves are from ref. ¹⁸.

Portions of the HH-1 δ¹⁸O and δ¹⁸O_corr records are notably different (Fig. 2a and Supplementary Fig. S2), with corrections reaching > 3–4 ‰ in places. Due to uncertainties introduced by choices we make in the correction method, including values of the model parameters, choice of temperature-dependent fractionation factor (α_{calcite-water}), and assumptions about the cave environment (see “Methods”), we focus our analyses on large changes (> 1 ‰) in the HH-1 δ¹⁸O_corr record. The original δ¹⁸O record ranges from − 3 to − 10 ‰, whereas the δ¹⁸O_corr record has a smaller range of − 6 to − 10 ‰. Because the correction removes the PCP signal, the largest changes occur during periods of high PCP (high δ¹⁸O, δ¹³C, and Mg/Ca values), while the smallest changes occur during periods of low PCP (low δ¹⁸O, δ¹³C, and Mg/Ca values). Generally, PCP is highest during the sea-level lowstand of the last glacial period and lowest during the sea-level highstands of the Holocene and late MIS 3. Low sea level exposes land adjacent to the study site (Gulf and Tonkin and South China Shelf), which reduces autumn moisture delivery to central Vietnam¹⁸. Ultimately, this results in smaller corrections during 45 – 35 ka and 13 – 4 ka, and larger corrections during 35 – 30 ka and 25 – 14 ka. The largest exceptions to this sea-level mechanism are two low PCP excursions from 30 – 25 ka and at 16 ka.

Notably, several features that the original δ¹⁸O record shares with the HH-1 Mg/Ca and δ¹³C records are no longer apparent. This includes the increasing trend starting at 35 ka and the abrupt shift to lower values at 14 ka. In the Mg/Ca and δ¹³C records, both of these features are interpreted to reflect changes in local rainfall amount driven by sea level change¹⁸. The correction also removes the anomalous negative excursions from 30 – 25 ka and at 16 ka. Features preserved in the δ¹⁸O_corr record include positive excursions during the Younger Dryas and Heinrich Stadials 1, 3, and 4. Overall, the δ¹⁸O_corr record no longer shows a relationship with sea level change at our site, but the δ¹⁸O_corr record remains sensitive to changes in Atlantic Meridional Overturning Circulation (Younger Dryas and Heinrich Stadials). These multi-proxy results now reveal the important differences between local hydrology (Mg/Ca and δ¹³C) and large-scale hydroclimate (δ¹⁸O_corr) that were not evident in the original uncorrected data¹⁸.

To test whether HH-1 δ¹⁸O_corr tracks the changes in δ¹⁸O_p, we compare the HH-1 δ¹⁸O_corr record to regional speleothem δ¹⁸O records and isotope-enabled climate model output. Since there are no long speleothem (> 8000 years) δ¹⁸O records from Mainland Southeast Asia, we compared HH-1 δ¹⁸O_corr to a composite δ¹⁸O stalagmite record from China, which reflects large-scale Asian summer monsoon strength²². Even though these records are somewhat distant from each other, the controls on summer monsoon δ¹⁸O_p are similar (upstream rainout of moisture sourced from the Bay of Bengal and the Indian Ocean)^19–21, which suggests speleothem δ¹⁸O records may covary across the greater region. The China composite shows far better agreement with the HH-1 δ¹⁸O_corr record than the original HH-1 δ¹⁸O record (Fig. 4 and Supplementary Table S2). They have similar values and are strongly in phase from 45 – 11 ka. Both show positive excursions during Heinrich Stadials and the Younger Dryas, and negative excursions during some Dansgaard Oeschger Events. Since δ¹⁸O_p is the primary driver of the China composite δ¹⁸O record, its similarities to the HH-1 δ¹⁸O_corr indicate that HH-1 δ¹⁸O_corr also reflects δ¹⁸O_p variability. Interestingly, these records are antiphased during the early- and mid-Holocene. Since the HH-1 record ends at ~ 4 ka, we are unable to determine whether this antiphasing continues during the late Holocene.

Fig. 4. HH-1 compared to Chinese stalagmites.

a Time series of HH-1 δ¹⁸O (orange), HH-1 δ¹⁸O_corr (black), and China composite δ¹⁸O (blue)²². Scatter plots of China composite δ¹⁸O vs. b HH-1 δ¹⁸O and c HH-1 δ¹⁸O_corr. The Pearson correlation coefficient (r) and Spearman’s rank correlation coefficient (ρ) are displayed in the bottom right corner of panels (b) and (c). For the correlation coefficient values, the China composite record was interpolated to the time steps of the HH-1 record, and Holocene values (gray circles) were excluded. The p-values for all coefficients are < 0.01. See Supplementary Table S2 for r values using δ¹⁸O_corr derived with different α_{calcite-water}.

Given the lack of Late Pleistocene terrestrial δ¹⁸O records from Vietnam, we compare the HH-1 δ¹⁸O record to simulated central Vietnam δ¹⁸O_p from iTRACE²³, an isotope-enabled transient climate model spanning the deglaciation, 20 – 11 ka (see “Methods”). Like the comparisons to regional δ¹⁸O records, the JJA (southwest monsoon), SON (northeast monsoon), and annual iTRACE δ¹⁸O time series from the grid cell encompassing the study site better correlate with the HH-1 δ¹⁸O_corr record than the original δ¹⁸O record (Fig. 5 and Supplementary Table S2). When converted to δ¹⁸O of calcite (δ¹⁸O_c), the simulated iTRACE δ¹⁸O_c time series are more negative than the HH-1 δ¹⁸O_corr time series (Fig. 5a). This difference is likely driven by the offset between iTRACE δ¹⁸O_p and δ¹⁸O_p measured from rainfall collected ~ 15 km from the study site (see “Methods”), where iTRACE δ¹⁸O_p is ~ 1–2 ‰ more negative than observed from July-January (Fig. 2c). Notably, δ¹⁸O_corr computed with a cave-derived α_{calcite-water} (see “Methods”), shows a smaller offset (Supplementary Fig. S3), suggesting the choice of α_{calcite-water} may be important when determining the absolute value of the correction. While the correlation with JJA is highest (Fig. 5e and Supplementary Table S2), the HH-1 δ¹⁸O_corr time series better matches the δ¹⁸O magnitude of change in the annual and SON time series (Fig. 5f, g). Nevertheless, the strong agreement between these records suggests that iTRACE correctly simulates δ¹⁸O_p variability in central Vietnam and provides further evidence that the HH-1 δ¹⁸O_corr record reflects changes in δ¹⁸O_p.

Fig. 5. HH-1 compared to iTRACE simulations.

a Time series of HH-1 δ¹⁸O (orange), HH-1 δ¹⁸O_corr (black), iTRACE JJA δ¹⁸O_c (purple), iTRACE SON δ¹⁸O_c (light blue), and iTRACE annual δ¹⁸O_c (red). iTRACE δ¹⁸O_c values (VPDB) calculated from mean weighted iTRACE δ¹⁸O_p (VSMOW) and temperature output using α_{calcite-water} from ref. ⁵³. For visual clarity, iTRACE curves are smoothed with a 100-year running mean. Scatter plots of iTRACE δ¹⁸O_p vs (b–d) HH-1 δ¹⁸O (orange) and (e–g) HH-1 δ¹⁸O_corr (black). iTRACE output was smoothed with a 50-year running mean. The Pearson correlation coefficient (r) and Spearman’s rank correlation coefficient (ρ) are displayed in the bottom right corner of each panel. The p-values for all coefficients are < 0.01. See Supplementary Table S2 for r values using δ¹⁸O_corr derived with different α_{calcite-water}.

Central Vietnam δ¹⁸O_p variability over the last 45,000 years

We interpret HH-1 δ¹⁸O_corr as a record of weighted annual mean δ¹⁸O_p in central Vietnam. Understanding this record is challenging because both southwest monsoon (JJA) and northeast monsoon (SON) rainfall, specifically the relative contribution of each monsoon system, drive modern annual δ¹⁸O_p variability in central Vietnam²¹. This means that HH-1 δ¹⁸O_corr variability could reflect changes in JJA δ¹⁸O_p, SON δ¹⁸O_p, simultaneous changes in both seasons’ δ¹⁸O_p, and/or changes in precipitation seasonality (e.g., the relative proportion of moisture delivered in autumn vs summer). For this reason, we consider both the southwest and northeast monsoons when interpreting the HH-1 δ¹⁸O_corr record.

We first examine δ¹⁸O_p variability during the deglaciation using iTRACE simulations. The similarity between the simulated JJA and SON δ¹⁸O_p time series and their strong correlation with the HH-1 δ¹⁸O_corr record makes disentangling the JJA and SON signals difficult (Fig. 5). A previous analysis of iTRACE simulations found that changes in seasonality (the amount of summer vs. autumn rainfall) minimally contributes to annual δ¹⁸O_p change across South China and Southeast Asia²⁴. Instead, they show that changes in the isotopic composition of JJA rainfall control annual δ¹⁸O_p variability during the deglaciation, with SON δ¹⁸O_p playing a minor role. However, iTRACE simulations underestimate the SON rainfall amount in central Vietnam (Fig. 2b), suggesting that the model may not capture the full influence of SON rainfall on the annual δ¹⁸O_p signal. Ultimately, the HH-1 δ¹⁸O_corr record most closely approximates the values of the annual δ¹⁸O_p simulation (Fig. 5, Supplementary Fig. S3 and Supplementary Table S2), suggesting that our stalagmite likely records the combined δ¹⁸O_p signal of both monsoon seasons.

When we consider the forcings individually, iTRACE δ¹⁸O_p simulations reveal that meltwater forcing drives the majority of δ¹⁸O_p variability during the deglaciation (MWF curve in Supplementary Fig. S4). This includes positive excursions during the HS1 and YD meltwater events, features also present in HH-1. In addition to HS1 and the YD, the HH-1 δ¹⁸O_corr record shows positive excursions during Heinrich Stadials 3 and 4, indicating central Vietnam δ¹⁸O_p responds somewhat consistently to meltwater events. The absence of HS2 in the HH-1 record suggests that HS2 is potentially weaker compared to other instances, which aligns with observations made in other paleoclimate records^25,26. The signature of meltwater events on the Asian summer monsoon is well documented in speleothem δ¹⁸O records^22,27. During meltwater events, the southward shift of the ITCZ and westerly jet weakens moisture transport²⁸, which reduces upstream rainout and increases δ¹⁸O_p values across the region¹⁹. Conversely, the meltwater-driven mechanism on δ¹⁸O_p during SON is currently unknown.

Deciphering the drivers of Holocene δ¹⁸O_p variability from the HH-1 δ¹⁸O_corr record is challenging. During the Holocene, speleothem δ¹⁸O records of Asian summer monsoon intensity reach their minimum δ¹⁸O values in the early Holocene (Fig. 4), lagging Northern Hemisphere Summer Insolation by ~ 3 kyrs^22,29. A recent synthesis of Holocene (8 – 0 ka) Southeast Asian speleothem δ¹⁸O records found the same relationship between the Southeast Asian summer monsoon (southwest monsoon) and summer insolation (PC1 from ref. ¹⁷). As for the Southeast Asian northeast monsoon (PC2 from ref. ¹⁷), the same study shows that speleothem δ¹⁸O (including a record from central Vietnam) tracks autumn insolation, which peaks in the mid-Holocene. Interestingly, the HH-1 δ¹⁸O_corr record resembles neither the southwest nor northeast monsoon but rather has an increasing trend from 11 – 7 ka and then becomes relatively stable from 7 – 4 ka (Supplementary Fig. S5). Notably, the PCP correction is minimal during the Holocene (Fig. 3a), so the uncertainty introduced from the correction method does not explain these differences. It is possible that the correction method underestimates PCP during the Holocene, which would cause δ¹⁸O_corr to be too high. While we cannot rule this out, the comparatively low and stable values of the Mg/Ca record indicate that PCP was low during the Holocene, making a large PCP-driven shift in speleothem δ¹⁸O unlikely. In fact, a recent modeling study found that rate-dependent fractionation when the PCP duration is small may decrease calcite δ¹⁸O by up to 0.5 ‰³⁰. If this is indeed the case during the Holocene, then PCP would reduce the difference between HH-1 and regional records rather than increase it. Thus, this effect also cannot explain our observed differences. The δ¹⁸O_corr record suggests that insolation does not directly drive changes in the HH-1 record δ¹⁸O during the Holocene. Instead, changes in seasonality (JJA versus SON precipitation amount), and changes in the isotopic composition of SON rainfall could both explain HH-1 δ¹⁸O variability during the Holocene (see Supplementary Text).

In summary, during the Late Pleistocene, meltwater-forced changes in δ¹⁸O_p drive central Vietnam δ¹⁸O_p variability. With the HH-1 δ¹⁸O_corr record deviating from other regional δ¹⁸O records during the Holocene, our understanding of Holocene δ¹⁸O_p variability in central Vietnam is less clear. Moving forward, more δ¹⁸O records from this region, particularly of the Holocene, and more in-depth investigations into autumn monsoon δ¹⁸O_p dynamics are essential.

Implications

Our study emphasizes the potential effects of PCP on a speleothem δ¹⁸O record, which provides insights into our understanding of hydroclimate variability in central Vietnam δ¹⁸O_p from 45 – 4 ka. Without correcting for the effects of PCP, we would have severely misinterpreted the HH-1 δ¹⁸O record. Furthermore, this would have led us to misdiagnose model-proxy disagreement with iTRACE, despite the model performing well. This has important implications, as speleothem δ¹⁸O records are increasingly being used to test paleoclimate model performance^31–33. While some uncertainties in speleothem δ¹⁸O records are inevitable, it remains essential to address and minimize them to continue improving the utility of speleothem δ¹⁸O records.

Although PCP only affects speleothem δ¹⁸O under certain conditions (when PCP occurs on the cave ceiling or stalactite), future studies should screen for PCP-influenced δ¹⁸O when generating new speleothem records. This could be done through the generation of multiproxy stable isotope and trace elements records from the same speleothem. How widespread the effects of PCP on speleothem δ¹⁸O are is currently unknown; however, PCP is a well-documented driver of many existing δ¹³C and trace element records. Generating and interpreting speleothem δ¹³C and trace element records and other proxies sensitive to PCP (e.g., Ca isotopes), alongside δ¹⁸O records, could identify whether PCP occurs in drip waters feeding the speleothem sample. As shown in this study, strong similarities between speleothem δ¹⁸O and these proxies are an indicator of PCP-influenced δ¹⁸O. Replication with speleothem δ¹⁸O records from the same or nearby caves is another effective screening method since individual drip pathways would not likely be subject to identical PCP histories. If multiple δ¹⁸O records show similar variability, it is unlikely that PCP substantially affects speleothem δ¹⁸O.

Furthermore, we show that by utilizing Mg/Ca data and knowledge of the cave environment, it is possible to remove the PCP signal from a speleothem δ¹⁸O record. This method builds upon well-established geochemical relationships and the strong similarity of corrected data with existing records, suggesting it can be reliably used to constrain δ¹⁸O_p. However, as detailed in the Methods, our model depends on cave monitoring measurements and makes several critical assumptions that may limit its applicability at certain cave sites. While not hugely impactful on the final correction, measurements of cave temperature, epikarst pCO₂, cave pCO₂, drip water δ¹⁸O, and the drip interval from an overlying or nearby dripsite are important for constraining the values of the corresponding model parameters. Nevertheless, when cave monitoring data is unavailable or difficult to measure (e.g., epikarst pCO₂), we show that values sourced from the literature and climate model simulations may be suitable substitutes. Although not done in this study, measuring additional proxies, such as fluid inclusion δ¹⁸O^34,35, fluid inclusion microthermometry³⁶, and TEX₈₆³⁷, would provide estimates of how cave temperature and drip water δ¹⁸O change through time. Of all the parameters in the model, the correction is most sensitive to the initial Mg/Ca value of dripwater (Mg/Ca_i). Since we approximated Mg/Ca_i from a low Mg/Ca threshold in the speleothem Mg/Ca time series, Mg/Ca_i estimated from speleothem records without a clear low-end member may add substantial error to the correction. The choice of α_{calcite-water} also impacts the correction. In future applications of this method, we recommend performing a sensitivity analysis that includes changing the model parameters and α_{calcite-water}.

The model assumes that PCP is the only driver of speleothem Mg/Ca variability and that all PCP occurs on the cave ceiling. For these reasons, our method may not be suitable for cave sites where there are multiple controls on speleothem Mg/Ca or where the majority of PCP occurs in the epikarst. Ultimately, further model development and testing in well-monitored caves is needed before widespread application to other PCP-affected speleothem records. Nonetheless, our findings confirm that PCP on cave ceilings can substantially obscure the δ¹⁸O_p signal, potentially leading to the misinterpretation of the widely-used speleothem δ¹⁸O proxy.

Methods

Rainfall δ¹⁸O measurements

A total of 83 rainwater samples were collected from January 2019 - August 2022 at the Phong Nha-Ke Bang National Park headquarters (17.6°N, 106.3°E) using collection methods outlined in ref. ²⁶. We analyzed all samples for δ ¹⁸O (VSMOW): 50 at Chapman University (Orange, California) using cavity ring-down spectroscopy (Picarro L2130-i), and 33 at Northumbria University (Newcastle, UK) using off-axis integrated cavity output spectroscopy (Los Gatos Research LWIA-24EP) (Table S3). The long-term standard deviation of an independent quality control standard for each instrument is 0.11 ‰ δ ¹⁸O (Picarro) and 0.20 ‰ δ ¹⁸O (Los Gatos).

iTRACE

iTRACE is a water isotope-enabled transient simulation of the last deglaciation (20 to 11 ka) performed in iCESM1.3^23,38. iCESM1.3 is comprised of the Community Atmosphere Model version 1.3 (CAM5.3), the Community Land Model version 4 (CLM4), Parallel Ocean Program version 2 (POP2), and Los Alamos Sea Ice Model version 4 (CICE4). The land and atmosphere resolution is 1.9° x 2.5° (latitude and longitude), with 30 vertical levels in the atmosphere. iTRACE uses four forcing factors: ICE: continental ice sheets that are changed every 1000 years following the ICE-6G model³⁹ and KMT ocean bathymetry (changed at 14 and 12 ka), ORB: orbital forcing, GHG: greenhouse gas concentrations from ice core reconstructions^40–42, and MWF: meltwater fluxes following TRACE-21ka⁴³. These forcings are applied additively to create four parallel simulations (ICE, ICE + ORB, ICE + ORB + GHG, and ICE + ORB + GHG + MWF). The effect of a single forcing can be approximated by subtracting simulations from each other (e.g., ORB = (ICE + ORB) – ICE). Refer to ref. ²³ for additional details on the experiment setup and the forcings.

Preindustrial iCESM 1.3 simulations largely capture seasonal changes in rainfall and δ¹⁸O_p in central Vietnam¹⁸. However, iCESM underestimates autumn rainfall amount by 2–5 mm/day and δ¹⁸O_p by 1–2 ‰ for July-January (Fig. 2b, c). Strong agreement between simulated iTRACE precipitation and the HH-1 Mg/Ca and δ¹³C records indicates that iTRACE correctly models centennial-millennial-scale hydroclimate change in central Vietnam across the deglaciation¹⁸.

δ¹⁸O PCP correction

Correction procedure and model description

We corrected the HH-1 δ¹⁸O record for PCP using the HH-1 Mg/Ca time series and following methods from ref. ¹⁷. We used a Rayleigh distillation model that simulates δ¹⁸O, the evolution of HCO₃⁻, and calcite precipitation (represented by the evolution of calcium concentration in the drip water) through time^8,44–46. This model has three sinks (H₂O, CO₂ degassing, and CaCO₃ precipitation), and can account for oxygen isotope exchange between drip water HCO₃⁻ and the water reservoir (H₂O exchange). The evolution of calcium concentration in the drip water is approximated by exponential decay⁴⁷.

Next, we outline the general procedure of the PCP correction method.

1. Estimate PCP duration. We first estimated PCP duration, defined as the length of calcite precipitation (in seconds) in the drip water prior to it reaching the speleothem, for each time point in the Mg/Ca time series with a corresponding δ¹⁸O measurement (n = 768, Supplementary Fig. S6a). To do this, we modeled the evolution of calcite precipitation over time, which requires estimates of initial Ca²⁺ concentration (Ca_i) and equilibrium Ca²⁺ concentration (Ca_eq) of the drip water, which are determined by the pCO₂ in the soil or karst for Ca_i and the pCO₂ in the cave for Ca_eq. For Ca_eq, we used measurements of the cave air pCO₂ from Hoa Huong cave (see Model Parameters section). For Ca_i, we chose the minimum value necessary to simulate the range of measured Mg/Ca values from HH-1. We started the simulation at an estimated initial Mg/Ca value (Mg/Ca_i, the Mg/Ca of calcite precipitated when the duration of calcite precipitation from the drip water is 0 s, i.e., PCP = 0) and simulated the Mg/Ca evolution over time using a Rayleigh distillation model, with a partition coefficient calculated following ref. ⁴⁸. We progressively increased Ca_i until the model simulated the Mg/Ca change necessary to explain the maximum Mg/Ca value of the HH-1 record (Supplementary Fig. S7). Thus, the modeled Mg/Ca evolution covers the whole range of Mg/Ca values measured from the HH-1 record and enables us to deduce the PCP duration for each Mg/Ca time point. We give further details on the Mg/Ca_i, Ca_i, and Ca_eq values, as well as the partition coefficient, in the Model Parameters section. In the Rayleigh model, longer PCP durations lead to increased Mg concentrations in the drip water and the subsequently precipitated calcite, resulting in larger deviations of the corresponding speleothem Mg/Ca from the Mg/Ca_i.

2. Calculate the PCP-induced change in δ¹⁸O. We then estimated the PCP-induced change in δ¹⁸O of drip water HCO₃⁻ for each timepoint using the Mg/Ca-based estimate of PCP duration. To do this, we modeled the oxygen isotope evolution of the HCO₃⁻ in the drip water and then determined the amplitude of δ¹⁸O change for each sample based on the estimated PCP duration. Similar to the increase in Mg/Ca over the course of calcite precipitation, longer PCP durations will correspond to larger increases in HCO₃⁻ δ¹⁸O in the Rayleigh model.

3. Calculate δ¹⁸O_corr. The PCP-induced change in δ¹⁸O for each sample, as calculated in the previous step, was then subtracted from the corresponding measured speleothem δ¹⁸O value to produce the final δ¹⁸O_corr time series (Fig. 3a and Supplementary Fig. S2). Ultimately, δ¹⁸O_corr should approximate the initial δ¹⁸O value of calcite precipitated in equilibrium with the drip water (when PCP = 0) which is equivalent (in equilibrium with) to the drip water δ¹⁸O. In the case of stalagmite HH-1, we expect drip water δ¹⁸O to reflect δ¹⁸O_p.

We make two important assumptions in this approach. First, we assume PCP is the only control on drip water Mg/Ca variability. Second, we assume all PCP occurs on the cave ceiling and/or overhanging stalactite. There may be other factors that substantially influence speleothem Mg/Ca (e.g., karst hydrology) and PCP may occur in both the epikarst and on the cave ceiling. We further explore these concepts and how they apply to the correction of the HH-1 δ¹⁸O record in the Model uncertainty and sensitivity section.

Model parameters

The model we used for the PCP correction requires a number of parameters as inputs (Mg/Ca_i, cave temperature, Ca_i, Ca_eq, drip interval, and drip water δ¹⁸O; Supplementary Table S1). When possible, we used direct measurements of the cave environment collected from March 2020 – March 2023 and further described in ref. ¹⁸. Otherwise, we used iTRACE model output and values from the literature. To assess the uncertainty of our correction, we also performed sensitivity testing by varying some of the model parameters within realistic ranges. Here, we describe the choices for each parameter.

We used a constant value of 2.23 mmol/mol for Mg/Ca_i (black dashed line in Fig. 3b). We derived this value by averaging two portions of the HH-1 Mg/Ca record (45 – 37 ka, and 13 – 4 ka). For these periods, we assume that little to no PCP occurred based on the comparatively low and relatively stable Mg/Ca values. For sensitivity testing, we varied Mg/Ca_i between 1.96 – 2.5 mmol/mol, based on the standard deviation of the record for the two time periods.

Since the temperature was not constant during the last 45,000 years, we approximated a temperature time series using a combination of cave monitoring data and iTRACE climate model output (Supplementary Fig. S8). We assumed a constant temperature of 21 °C during the Holocene (11 – 4 ka in the HH-1 record), which is an intermediate value between the modern annual temperature of Hoa Huong cave collected via data logger from March 2020 – August 2022 (~ 20 °C¹⁸) and Holocene (11.1 – 11 ka) iTRACE surface temperature values (~ 21.5 °C). From 20 – 11 ka, we varied temperature based on a 50-year running mean of simulated iTRACE surface temperature from the grid cell encompassing the study site. We used a constant temperature of 18.5 °C for the time period of 45 – 20 ka, based on the annual iTRACE simulated surface temperature from 20 ka. For sensitivity testing, we varied the temperature +/− 2 °C, the current seasonal range in Hoa Huong cave¹⁸.

The model also requires inputs of drip water Ca_i and Ca_eq. For our model to explain the full range of HH-1 Mg/Ca values, we had to use a minimum Ca_i value of 3.4 mol/m³, which is equivalent to a pCO₂ of 49,000 ppm at 21 °C (we assume the same temperature for all the following pCO₂ values). While this value exceeds the majority of soil pCO₂ measurements, pCO₂ levels in the epikarst can surpass those observed at the surface and in the soil. Notably, studies have recorded epikarst pCO₂ values as high as 60,000 – 100,000 ppm^49,50, and studies utilizing geochemical models to approximate epikarst pCO₂ found similarly high levels^51,52 We conducted sensitivity analyses using a range of Ca_i values between 3.4 and 4 mol/m³ (equivalent to 49,000 – 79,500 ppm) to test how higher epikarst pCO₂ values would affect our results. For Ca_eq, we used a value of 0.8 mol/m, which corresponds to a pCO₂ value of 635 ppm, an average of two modern pCO₂ values measured in Hoa Huong Cave, 820 ppm (March 2020) and 510 ppm (March 2023). For the sensitivity testing, we used a range of 0.6–1.4 mmol/m³, which corresponds to pCO₂ values of 270 – 3400 ppm. Our low estimate of 270 ppm reflects atmospheric conditions during the last glacial maximum⁴¹ and also assumes that the cave is fully ventilated. We based our highest estimate of 3400 ppm on the highest measured value of pCO₂ recorded in August 2022 in Hoa Huong cave.

Drip interval affects speleothem Mg/Ca, and thus, also the PCP duration we estimate with our approach. Since there is no active drip at the collection site, we were unable to measure the drip interval and assume a constant drip interval of 1 drip/second for the method. Even though drip interval may have changed through time, previous work has found changes in drip interval had a smaller effect on the isotopic composition of calcite when compared to the effects of PCP⁸.

The model uses an infiltrating water parameter (drip water δ¹⁸O) as the initial HCO₃⁻ δ¹⁸O value. We used iTRACE simulations of weighted annual mean δ¹⁸O_p to estimate the infiltrating water δ¹⁸O parameter (drip water δ¹⁸O). We assumed drip water values approximate the weighted annual mean of δ¹⁸O_p because we have limited observations of drip water δ¹⁸O at the study site. From 20 – 11 ka, we varied temperature based on a 50-year running mean of simulated precipitation weighted annual mean δ¹⁸O_p from the grid cell encompassing the study site. We used a constant δ¹⁸O_p value of − 9.4 ‰ for the time period of 45 – 20 ka and − 9.5 ‰ for the time period 11 − 4 ka, based on the iTRACE weighted annual values from 20 ka and 11 ka, respectively. We did not perform sensitivity testing because when PCP duration is sufficiently short (Supplementary Fig. S6a), the effects of the drip water δ¹⁸O value are minimal⁸.

We used laboratory-based measurements to calculate the partition coefficient of Mg (D_(Mg))⁴⁸. Since D_(Mg) is temperature-dependent, we calculated D_(Mg) at each time step using the approximated cave temperature record (Fig S8). The model uses temperature-dependent equilibrium fractionation factors (see ref. ⁸).

Model uncertainty and sensitivity testing

For the sensitivity analyses, we performed 1000 Monte Carlo simulations and determined the standard deviation for each parameter. To assess the uncertainty of the full model, we randomly varied all parameters within the chosen ranges (Supplementary Table S1 and Supplementary Fig. S9a, gray shading on δ¹⁸O_corr in Figs. 3–5). To assess the uncertainty of some parameters, we randomly varied a single parameter (Mg/Ca_i, cave temperature, Ca_i, and Ca_eq) while keeping the remaining variables constant (Supplementary Fig. S9b–e). For the full model run, the average 1 sigma uncertainty was +/− 0.21 ‰, a relatively small value when compared to the magnitude of variation in the HH-1 δ¹⁸O_corr record. Our simulation also revealed that the Mg/Ca_i parameter causes nearly all of the variability in the simulations (Supplementary Fig. S9b). The variability of the remaining variables (temperature, Ca_i, and Ca_eq), had negligible effects on the results. This suggests an accurate estimate of Mg/Ca_i is crucial for future use of this correction method.

Given that we do not fully understand how the cave environment changes over time, we assume constant values for several model parameters: Mg/Ca_i, Ca_i, and Ca_eq. This assumption is likely incorrect, particularly between glacial and interglacial climate conditions. However, the insensitivity of the PCP correction to changes in Ca_i and Ca_eq suggests using constant values for these parameters is reasonable (Supplementary Fig. S9d, e). Conversely, the correction is very sensitive to changes in Mg/Ca_i. Our sensitivity testing accounts for some of this uncertainty, but this is currently a limitation that warrants further investigation. To avoid introducing additional uncertainty to the correction by choosing arbitrary changes in Mg/Ca_i, we use our best estimate of Mg/Ca_i for the entire record.

The sensitivity of the correction to the Mg/Ca_i parameter highlights the importance of the model assumption that PCP is the only driver of drip water Mg/Ca variability. As previously stated, the strong similarities between the HH-1 Mg/Ca and δ¹³C records indicate that PCP is the primary control on drip water Mg/Ca¹⁸ (Fig. 3). While we cannot entirely eliminate the influence of other controls on drip water Mg/Ca and whether these controls change through time, the relative stability of the low/no PCP periods in the HH-1 Mg/Ca record during 45 – 37 ka and 13 – 4 ka suggests that factors other than PCP have a negligible effect on Mg/Ca variability.

We test the difference of the PCP corrected δ¹⁸O record when using the fractionation factor (α_{calcite-water}) derived from laboratory experiments⁵³ against three cave-derived fractionation factors^9,54,55. The fractionation factor from ref. ⁵³ results in the largest corrections, and the ref. ⁹ fractionation factor the smallest (Supplementary Fig. S2), and the maximum difference between the resulting δ¹⁸O_corr time series is ~ 1 ‰ (Supplementary Fig. S10). Since the different corrections largely do not change the outcome of our interpretations (Supplementary Table S2), we choose to focus our discussion primarily on the PCP correction obtained using the laboratory-derived fractionation factor⁵³.

We also found that using a version of the model that includes H₂O exchange has a small impact on the correction (Supplementary Fig. S6b). This is because PCP duration for most data points is substantially shorter than the time required for complete H₂O exchange (Supplementary Fig. S6a). For example, only 6 of the 768-time points exceed a PCP duration of 1000 s (orange dashed line in Supplementary Fig. S6a), where H₂O exchange only reduces PCP-induced δ¹⁸O change by ~ 0.35 ‰ (orange dashed line in Supplementary Fig. S6c). In fact, for the majority of the record, the duration of PCP is sufficiently short to render the effects of H₂O exchange negligible.

Importantly, our approach only accounts for PCP occurring immediately prior to speleothem precipitation (i.e., PCP on the cave ceiling or stalactite) and excludes PCP in the epikarst. Since PCP in the epikarst should not affect δ¹⁸O⁴, we expect the PCP impact on the speleothem δ¹⁸O record to originate from calcite precipitation on the cave ceiling or overlying stalactite. Conversely, PCP in the epikarst does affect drip water Mg/Ca as there is no exchange process that counterbalances PCP for Mg/Ca. In cases where PCP occurs in the epikarst, we would overestimate PCP-induced increase in speleothem δ¹⁸O and bias δ¹⁸O values of the corrected speleothem δ¹⁸O record toward low values. Our correction approach cannot currently disentangle epikarst PCP from in-cave PCP. While there is no stalactite or active drip above HH-1, we do see evidence of a previously connected soda straw on top of HH-1 and some secondary calcite deposits on the cave ceiling, indicating that some form of calcite precipitation likely occurred above the growing stalagmite (Supplementary Fig. S1b). To test the assumption that all PCP occurs on the cave ceiling or in stalactites, we compare our corrected δ¹⁸O record against other regional records less likely to be affected by PCP and climate model output.

Author contributions

Fieldwork: E.W.P., A.W., K.R.J., M.L.G., T.N.B., M.X.T., T.H.D., Q.D., and V.E. Water isotope analyses: V.E., G.R.G. δ¹⁸O correction model: V.S. Climate model analyses: E.W.P. Visualization: E.W.P., A.W., and V.S. Supervision: K.R.J., M.L.G., and D.M. Writing—original draft: E.W.P., V.S., and A.W. Writing—review & editing: E.W.P., V.S., A.W., M.L.G., D.M., T.N.B., M.X.T., T.H.D., Q.D., G.R.G., V.E., and K.R.J.

Peer review

Peer review information

Nature Communications thanks Valdir Novello and the other anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Data availability

All corrected δ¹⁸O time series and rainwater δ¹⁸O data from Phong Nha-Ke Bang National Park Headquarters are archived at Zenodo (10.5281/zenodo.13768949)⁵⁶. The corrected δ¹⁸O time series and the original HH-1 δ¹⁸O, δ¹³C, and Mg/Ca time series can be found at the National Oceanic and Atmospheric Administration National Centers for Environmental Information Paleoclimatology archive (10.25921/643e-eb71 and 10.5281/zenodo.13768949)^57,58.

Code availability

The code used for the PCP removal method, including instructions and example data, is accessible via Zenodo (10.5281/zenodo.13768862)⁵⁶.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-024-53422-y.