Groundwater use is accelerating globally due to population growth1,2. Further reductions in water availability in many regions due to climate change3,4 are likely placing increasing pressure on groundwater resources and groundwater-dependent ecosystems5,6. It is critical to understand how drying in the mid-latitudes will impact rainfall recharge to groundwater resources globally, yet our current understanding of the sensitivity of groundwater systems to climate change is limited4,7,8,9. This is because water table monitoring using direct methods such as bores can be complicated by groundwater extraction and land use change which can obscure the climate change signal10, and the water table may be recharged by rivers or lakes11 in addition to local rainfall recharge. Furthermore, the hydraulic memory of groundwater systems can delay a response to climate change by decades or longer1. These factors mean that attributing changes in groundwater recharge to the impacts of climate change on rainfall and the surface water balance is difficult.

Caves offer a natural window to observe the movement of water from the surface through the vadose zone and into groundwater aquifers, and can thus indicate rainfall recharge to groundwater12,13. Rainfall recharge may be via preferential flow along fractures, fracture and fissure zones and conduits, or it can be diffuse porous flow through the matrix. In karst environments, recharge is likely to be a mixture of diffuse and preferential-flow types12,14. Similar to groundwater, the oxygen isotopic composition (δ18O) of cave drip waters and consequently speleothems, is largely determined by the recharge-weighted δ18O, particularly in water-limited climates15,16 where groundwater resources are increasingly under pressure. Accordingly, speleothems offer a highly temporally-resolved record of rainfall recharge to groundwater. As speleothems form in the vadose zone above the water table, they are not complicated by modern groundwater abstraction nor the longer-term processes such as changes in hydraulic boundary conditions. Modern speleothems formed since the early 20th century can be used to assess the response of rainfall recharge to groundwater during recently observed climate change, while older speleothem material can provide a longer-term context of the range of natural, pre-industrial variability of rainfall recharge to groundwater.

Rainfall in southwest Australia is highly seasonal and falls in the cooler months, typically from May to October. The southwest Australian region has experienced a prolonged decline in annual rainfall (Fig. 1) attributed to changes in Southern Hemisphere atmospheric circulation systems with a considerable contribution from anthropogenic forcing17,18. In response to anthropogenic changes in stratospheric ozone and then atmospheric greenhouse gases18, rainfall has decreased by 10–25% overall across the region since the 1960s19. This has resulted in an annual rainfall decrease of between 100 and 230 mm since 1971–2010 compared to 1911–1970 (Fig. 1). Several local studies have since recognised the effect of reduced rainfall and increased abstraction on water resources within southwest Australia, as direct rainfall recharge has generally been found to be the major recharge mechanism20,21. For example, the Rottnest Island freshwater lens (Fig. 1) has decreased in size by 40% primarily due to the rainfall decrease20. Likewise, water table decline within the Perth Basin has been identified as a result of decreasing rainfall21,22. However, quantifying the amount of water table decline attributable to decreased rainfall input is complicated by increasing groundwater extraction and land use changes over the same time period23. Climate models indicate, with high confidence, that rainfall in this region is likely to decline further throughout the 21st century18 which will continue to impact water availability in southwest Australia21,22. This will have a consequent impact on water-dependent ecosystems24 and water security for this region that currently depends on groundwater to meet 75% of its water demand21.

Fig. 1: Location of caves and regional rainfall decrease.
figure 1

The map shows the location of caves, rainfall samples and the total decrease in annual rainfall for 1971–2010 compared to 1911–197033. See Supplementary Note 1 for description of field sites and stalagmite samples. Inset shows the study area location, full extent of groundwater in the Perth Basin (hatched area) and extent of southwest Australian region (shown by red line joining 30°S, 115°E and 35°S, 120°E). Annual rainfall shown on the inset map shows that the study area is in an important rainfall area of the southern Australian region.

In this study, we constructed a composite speleothem δ18O record using seven speleothems from four caves across a 30-km transect in southwest Australia (Fig. 1; see Supplementary Fig. 1 for speleothem images). All caves are located in the Leeuwin-Naturaliste National Park and are developed in the Late Quaternary aeolian calcarenite that outcrops along the coast25. All caves are located in the vadose zone and receive recharge from rainfall only. No surface runoff is observed at the caves due to the free-draining quartz-rich sandy soil with infiltration also aided by seasonal rainfall occurring in the cooler months. See Table 1 for a summary of climate, vegetation and cave sampling locations.

Table 1 Summary of speleothem sites and sample information.

One of the caves, Golgotha Cave, has been extensively monitored between 2005 and 2020 including for stable isotopes of rainwater26 and cave drip water from the locations of four speleothems presented here27. Opportunistic campaign sampling of ~20 additional drip sites in Golgotha Cave occurred during five field campaigns conducted in 2012, 2013 and 2014 (see Supplementary Table 1)27. The flow classification of these drips was determined in refs. 12,28,29 using statistical analysis of stalactites mapped by lidar and automated drip rate loggers. Recent studies identified that rainfall recharge through Golgotha Cave occurs via drips that are a mixture of slow diffuse flow, predominantly through the matrix, and faster preferential flow focussed along fractures or conduits27. For brevity, we refer to these henceforth as diffuse versus preferential-flow paths. Deep percolation via rapid movement of water along preferential flow paths is supported by threshold increases in drip rates12 and water oxygen isotope studies27. See Supplementary Note 1 for detailed site descriptions.

The stalagmites used in this study have all grown through the period of recent regional rainfall decline (Fig. 1). The speleothem δ18O records, along with cave drip water monitoring, demonstrate a common rise, or uptick, in values beginning in the 1990s. The uptick is not driven by a change in the input source water (rainfall) δ18O values but rather is demonstrated to be a shift in the hydraulic flow properties in the karst aquifer in response to reduced rainfall recharge. We argue that the declining rainfall has reduced rainfall recharge to groundwater along preferential-flow paths. We also assess this in the context of the longer (paleo) speleothem record for the region and demonstrate that this uptick is unprecedented for the last 800 years, implying that the current reduction in rainfall recharge is also unprecedented over this period.

This study uses seven modern speleothems, as well as cave drip water, from multiple caves to examine the response of reduced rainfall recharge due to climate change in the modern record. In doing so, it also confirms that upticks in speleothem δ18O values primarily represent reduced water supply to the speleothem. This substantiates the interpretation of speleothem δ18O datasets as a proxy for hydroclimate, and importantly validates the role of karst flow paths in driving a non-linear response to hydroclimate forcing via control on recharging water δ18O values.

Results and discussion

Composite speleothem δ18O record

Seven speleothem records from four caves, as well as cave drip water δ18O data from one of the caves (Fig. 2), were used to construct a composite speleothem δ18O record (Fig. 3a). These records all demonstrate a rise or uptick in δ18O values during the late 20th Century and early 21st Century. The rising trend in the Golgotha Cave speleothem and drip water δ18O values in recent decades was first reported in ref. 27 and is now replicated here in additional speleothems from other caves in the Cape Leeuwin-Cape Naturaliste region. The rise in δ18O for speleothem MND-S1 in response to the first rainfall decrease was initially documented in ref. 30, although this record ceased in 1992 when the speleothem was removed, i.e., too early to record the full extent of the δ18O uptick. New data presented in this study from speleothems, CRY-S1 and LAB-S1, as well as GL-S4 from ref. 27, cover a longer period since the rainfall decrease (CRY-S1 and LAB-S1 ceased in 2009 CE while GL-S4 in 2012 CE; Fig. 2). These three records replicate the rising isotopic trend recorded in the drip water monitoring during the respective periods of overlap with the monitoring data (Fig. 2), confirming the robustness of the uptick as a regional signal (Fig. 3a). The beginning of the uptick varies between speleothem records which may be due to different hydraulic response times (i.e., cave flow properties) or speleothem age uncertainty, although the latter is small (see Methods). Nevertheless, the comparison of records indicates that the isotopic rise becomes a common signal by the 1990s. The speleothem records show that this rise is approximately 1‰, while our drip water data confirm that this rising trend has continued to the present, indicating an overall increase of 1.5‰ above the 1900–1968 normalised speleothem δ18O baseline (Fig. 3a).

Fig. 2: Modern speleothem and drip water records.
figure 2

Individual speleothem (stalagmite) δ18O records from Crystal Cave (CRY-S1), Golgotha Cave (GL-S1, GL-S2, GL-S3 and GL-S4), Labyrinth Cave (LAB-S1), and Moondyne Cave (MND-S1), as well as drip water δ18O expressed as calcite-equivalent values and farmed calcite δ18O from the Golgotha speleothem locations. Farmed calcite δ18O error bars represent ± 1 standard deviation of measurements from each site and the year error bars represent the sample collection time. All data were normalised by subtracting 1900–1968 mean speleothem δ18O values. Chronologies for all speleothems were published previously (see 'Methods').

Fig. 3: The uptick in cave δ18O data is too large to be driven by changes in mean annual rainfall δ18O and is interpreted to be a reduction in preferential flow reaching the cave.
figure 3

a Composite cave δ18O record that is the 5-year rolling average of the 1900–1968 mean subtracted speleothem and drip water records. Shaded area represents one standard deviation. Calculated 20thC warming effect on δ18O values also shown. b Rainfall amount for the southwest Australian region (southwest of the line joining 30°S, 115°E and 35°S, 120°E; Fig. 1, inset) and for the coastal limestone extent (Fig. 1) demonstrating that the coastal locations of the caves also experienced two decreases in regional rainfall. Means for the periods 1910–1968, 1969–1999, and 2000–2020 are represented by horizontal lines. c Five-year rolling averages of annual amount-weighted rainfall δ18O: measured at Calgardup and Perth; predicted from high quality long-term rainfall records at Boyanup and Cape Naturaliste; and predicted from IsoGSM257. Circles of the same colour correspond to the monthly amount-weighted rainfall δ18O. Also, range of δ18O measured in groundwater from the Leederville and Yarragadee aquifers of the Central and Southern Perth Basin58. d Number of days per year soil is saturated with horizontal bars indicating the average values across all of the cave sites for the periods 1910–1968, 1969–1999 and 2000–2020. Three model estimates were determined corresponding to the three 0.05° gridded cells covering the four cave sites in the AWRA-L model56. e Local water table level beneath Jewel Cave.

Speleothem δ18O is a function of multiple processes31,32, and can be altered due to changes in water–calcite reaction processes such as: (1) cave temperature, (2) isotopic disequilibrium and/or (3) kinetic fractionation due to precipitation rate; as well as a function of drip water isotopic composition. However, as shown below, neither changes in cave temperature nor other changes in reaction processes can account for the observed uptick.

The uptick in speleothem δ18O cannot be attributed to the temperature-dependent fractionation that occurs during calcite precipitation31 as recent warming33 would lower speleothem δ18O by an estimated 0.27‰ below pre-1960 values rather than raise speleothem δ18O values as observed (Fig. 3a). Enhanced isotopic disequilibrium processes31,34 (disturbance of the isotopic equilibrium between H2O and the carbonate species) were shown not to be dominating δ18O variability in the modern Golgotha Cave records, representing four of the seven stalagmites in our composite27. This was constrained numerically by demonstrating that the O isotopic fractionation factors (αcalcite-water) were spatially consistent across the four sites and temporally consistent over the period of rising drip water δ18O values (see ref. 27 their Table 2 and their Supplementary Note 2). Moreover, cave temperature has been monitored at the same site in Golgotha Cave in discontinuous intervals since 2005 and has not risen (see Supplementary Table 2). We further tested for changes in precipitation rate, which may drive the isotope fractionation factor further from predicted (kinetic isotope fractionation35), by calculating annual extension rate using laminae width for the three speleothem records with the longest datasets (CRY-S1, LAB-S1 and GL-S4) and the pre-uptick versus uptick mean annual extension rates were within ± 1 standard deviation, indicating that this is not the driver of the uptick. Hence water–calcite reaction processes do not account for the magnitude of the uptick and, furthermore, could not have produced the rising trend observed in the drip water δ18O values.

Table 2 A comparison of precipitation-weighted annual rainfall mean δ18O values before and after southwest Australian rainfall decreased.

The observed uptick is present in both the drip water and speleothem δ18O values. The rising trend in the Golgotha Cave speleothem and drip water δ18O values in recent decades was first reported in ref. 27 and is now replicated here in additional speleothems from other caves in the region. This implies a regionally consistent response of drip water δ18O to hydrological forcing persisting over several decades, and is examined further below.

Rainfall recharge and the drip water δ18O response

Southwest Australian rainfall δ18O

Long-term changes in mean rainfall δ18O values have the potential to drive similar changes in drip water δ18O values31. Rainfall δ18O and its relationship to climate in southwest Australia has been previously studied26, and these datasets can help constrain the impact of climate change on rainfall δ18O in the region26. The decline in southwest Australian winter rainfall has occurred in two steps occurring in 1968/1969 and 1999/200019 (Fig. 3b). This regional pattern of rainfall decline (Fig. 1) is also evident over the narrow coastal limestone belt where the caves are located (Fig. 3b). The rainfall decrease has been characterised as a reduction in total rainfall and heavy rainfall during winter, as well as a decrease in the total number of rain days36. Of these three factors, it is rainfall intensity that most impacts the δ18O value of southwest Australian rainfall (see ref. 26 which shows that more intense rainfall is associated with more negative δ18O values). However, using both measured and modelled rainfall δ18O values (Fig. 3c), we can quantify that the observed change in rainfall intensity accounts for between a − 0.3‰ to +0.1‰ change in precipitation-weighted mean annual rainfall δ18O values over the period of rainfall decline (Table 2). In contrast, the measured rise in the composite δ18O record of 1.5‰ is fifteen times greater (Fig. 3a). Thus, we argue that the overall rise of 1.5‰ in our composite δ18O record reflects a rise in drip water δ18O despite little change in the mean δ18O value of annual rainfall supplying these drips.

Considering that temporal variability in mean rainfall δ18O values cannot account for the observed rise in drip water δ18O, the possible remaining processes are (1) isotopic modification of the infiltrated rainwater by rising evapotranspiration demand, (2) increasing evaporation of soil water from within karst flow paths37 and/or (3) a shift in the dominant flow paths supplying drip water to a stalagmite, i.e., a change in the contribution of diffuse versus preferential flow27. These remaining processes are discussed below.

Evapotranspiration and soil/karst evaporation

Evapotranspiration demand is potentially important in this environment as evapotranspiration fluxes for Golgotha Cave are high due to water demand from the overlying forest, and the soil profile becomes dry during the warmer months38,39. Transpiration does not result in isotopic fractionation of soil water40 although it is plausible that transpiration could alter the fraction of rain events contributing to drip water, but this should favour the removal of smaller rain events (relatively higher δ18O values)26 during the dry summer months when vegetation water stress is highest. Thus, an increase in transpiration demand is likely to lower, rather than raise, the mean value of rainfall recharge δ18O, i.e., opposite to the observed change. Hence we rule out this process as the driver of the uptick.

To diagnose the effect of evaporation in the soil, Fig. 4 shows isotopic composition of rainfall and flow-classified campaign drip water data on a δD versus δ18O plot. Linear fits to each category: rainfall (also known as the local meteoric water line or LMWL), diffuse flow and preferential flow; have slopes which are indistinguishable when judged from the 95% confidence interval, although the slope for diffuse-flow drips appears to have a lower value than other categories (Fig. 4, inset). This indicates the possible incorporation of a small fraction of water affected by evaporation in the soil zone, preferentially into the diffuse-flow path (cf. ref. 37). For example, soil zone evaporation at Golgotha Cave was proposed by ref. 38 to explain a transient peak whereby drip water δ18O values rose 2.5‰ and returned to long-term mean values within 12 weeks. However, it was observed that the contribution of isotopically-modified soil water was small and was quickly diluted by the newly infiltrated water in the following recharge season. Thus soil water evaporation is a process which can affect drip water, at least transiently, but since it is short-lived and not affecting preferential flow δ18O values (Fig. 4, inset), it is unlikely to contribute enough volumetrically to drive the uptick in observed δ18O values. The overall similarity of the slope of values of the drip waters to the LMWL supports our interpretation that evaporation under low humidity (the soil zone) has not substantially modified the drip waters.

Fig. 4: Water isotope plot comparing drip water and local rainfall.
figure 4

Rainfall data are from ref. 26, and drip water are from the five sampling campaigns described in the text. The regression, or local meteoric water line (LMWL), was calculated using the precipitation-weighted method (WLS-LMWL) in ref. 26. Confidence interval is 95%. Large crosses are the distribution modes for preferential flow (blue; δ18O = −4.67 ± 0.06‰, δD = −19.32 ± 1.21‰) and diffuse flow (orange; δ18O = −4.12 ± 0.12‰, δD = −15.09 ± 0.79‰). Inset compares slope values and error bars represent ± 2 standard error. The similarity of the slope values of the drip waters and local rainfall suggests that evaporation under low humidity (soil zone) has not substantially modified the drip waters.

If evaporation in the soil zone is unlikely to drive the uptick, there is also the potential for the diffusional loss of water vapour from the deeper karst zone to dry the deep water stores and concurrently affect drip water isotopes. For this process, the δD versus δ18O slope values are not a helpful diagnostic as we can assume that liquid water below the surface is in equilibrium with the surrounding vapour, hence relative humidity is close to 100% and isotopic fractionation is governed by equilibrium processes so that isotopic values evolve along the same line as the LMWL (see ref. 37 for further discussion). The relative magnitude of this process on the drip water δ18O values may also be estimated by comparing the calculated drip δ18O using the regression relationships on Fig. 4 for preferential versus diffuse flow, and can be shown to be small. For example, using the precipitation-weighted δD mean value of rainfall (−20.2‰; ref. 26), the estimated increase in drip δ18O is 0.1‰ due to evaporation at high humidity. Thus, while this process may always be present, it also appears to be insufficient in magnitude to be responsible for the uptick. Nonetheless, we propose that less frequent recharge of the flow paths by preferential flow over time could increase the potential for evaporation under high humidity of slower percolating diffuse-flow water, and this is consistent with our interpretation outlined below that there has been a reduction in the preferential flow component owing to the decline in rainfall.

Flow paths supplying drip water

In the study of Golgotha Cave drip waters by ref. 27, it was shown that drip water influenced by preferential flow typically has lower δ18O values compared with drip water predominately supplied by diffuse flow. Figure 5 shows the distributions of drip water compared with rainwater δ18O values for these data. Based on the campaign data, which sampled the largest number of drip locations, the distribution mode of diffuse-flow dominated drips (−4.12 ± 0.12‰VSMOW) is closest to the precipitation-weighted mean of rainfall (−3.61 ± 0.21‰VSMOW), whereas the distribution mode of preferential-flow-influenced drips (−4.67 ± 0.06‰VMSOW) are 1‰ lower than rainfall (Fig. 5). The lower value for the preferential flow δ18O values is closer to rainfall when only heavier totals are included (−4.55 ± 0.40‰VSMOW, daily events >9 mm; Fig. 5). Drip water δ18O values lower than rainwater is characteristic of the typical recharge-bias response seen at a global scale in cave drips15,27. The isotopic distinction in drip flow type is also evident in two of the speleothem drip sites classified as predominately diffuse (GL-S1, mode = −3.99 ± 0.03‰VMSOW) and preferential flow influenced (GL-S3, mode = −4.60 ± 0.04‰VMSOW). We note here that the distribution of the diffuse-dominated drip waters appears bimodal, indicating that the majority of drips are a mixture of diffuse and preferential flow with the contributions of preferential flow varying over time due to the higher recharge threshold required to activate flow along the fractures. See ref. 27 for an example of flow activated along a fracture after high recharge, captured in the drip time series for speleothem GL-S4, that led to a 0.5‰ decline in drip water δ18O values.

Fig. 5: Distributions of δ18O values for Calgardup rainfall, Golgotha drip water and groundwater.
figure 5

Drip waters with a greater preferential-flow influence are 18O-depleted with respect to diffuse flow and indicate a recharge bias. For drip water, both the campaign dataset (~20 drip sites) and the long-term sampling (two drip sites: GL-S1 and GL-S3) are shown. The distribution of preferential-flow drip waters (blue/purple) are shifted to lower δ18O values compared with the distribution for diffuse-flow drips (orange/red). Preferential flow and groundwater δ18O (green) values closely agree indicating that infiltration to groundwater similarly has a recharge bias towards higher-intensity rainfall. Groundwater data are Holocene age data for the Southern Perth Basin (Leederville and Yarragadee aquifers) taken from ref. 58. Dashed lines indicate the distribution modes (values given in text). Precipitation-weighted monthly rain δ18O values were used for comparing with drip waters to approximate the sampling interval of the drip waters (4–6 weeks). Error bars show the mean ± 2 standard deviation of the precipitation-weighted monthly rain δ18O values. We expect that the distribution of drip water to be narrower since drip water is the averaging of rain events. The gaussian_KDE routine from scipy Python library was used for computing kernel density estimates. This employs a Gaussian kernel, with a bandwidth computed using the method from ref. 59.

Each speleothem in our composite record will have unique flow path properties and length, yet as with the drip water, all are likely influenced by a mixture of diffuse plus additional preferential flow. Fewer of these preferential-flow recharge events, resulting in a drift towards the higher isotopic value of the diffuse-flow dominated drips, is the most likely explanation for the 1–1.5‰ uptick in drip water and speleothem δ18O. We further interpret that the reduction in this preferential-flow component is due to the rainfall decrease that this region has experienced (Fig. 3b), owing to the impact on soil moisture balance which we outline below.

Movement of water along preferential-flow paths requires overlying soils to have become saturated41, consistent with observations at Golgotha Cave39. In a time-lapse electrical resistivity tomography study, it was observed that autumn and early winter rains saturate the soil and epikarst layers before downward percolation of water was observed. There can be delays of up to several months since the rain season began owing to the time it takes to saturate the soil which becomes dry each summer (see Supplementary Note 1 for further description). As seasonal saturation of the soil and epikarst zone is needed to initiate recharge of karst stores above the monitored drip sites, it follows that the opportunity for recharge each season will also depend on the number of days that the soil remains saturated each season39. We examined temporal variability in the frequency of days that the soil moisture was saturated by extracting modelled soil moisture values for the four cave sites for 1911–2019 from the Australian Bureau of Meteorology’s Australian Water Resources Assessment Landscape model33 (see 'Methods'). Prior to the first rainfall decrease, the average seasonal duration of soil saturation was 96 days per year (1910–1968; Fig. 3d). This decreased to 66 days per year (1969–1999) following the first rainfall decrease and to 55 days per year (1999–2020) after the second rainfall decrease (Fig. 3d). This represents an overall drop of 43% in the number of days that the soil is saturated.

As stated above, a reduction in the duration of the soil saturation will reduce the potential contribution of recharge from preferential flow, and consequently impact the δ18O values of the drip water. Our analysis of the modelled soil moisture data suggests that this seasonal window in which recharge from preferential flow is possible has almost halved since regional rainfall began declining. The shrinking of this seasonal window coincides with the period of the δ18O uptick (Fig. 3a, d) with the uptick accelerating in the last two decades (Fig. 3a). Thus, it appears that the majority of the isotopic uptick is in response to the second rather than the first rainfall decrease. This likely reflects the cumulative impact of reduced rainfall on soil moisture, whereby reduced saturation in the soil profile and the reduction in the duration when the soil/epikarst is recharging the karst stores, reduces the likelihood of preferential flow along fractures or conduits occurring. That is, we propose that preferential flow is still occurring but was reduced during the period of the first rainfall decrease (1969–1999), and was further reduced when the second rainfall decrease occurred after 1999. A further consideration is the lag for karst stores to drain. For example, the fracture activation event observed for GL-S4 drip water (described above and see ref. 27) took at least 3 years for drip water δ18O to return to values equivalent to diffuse-flow drips27. Thus we interpret that the impact of a reduced water balance has decreased recharge along preferential-flow paths in the karst, resulting in the uptick in drip water and hence speleothem δ18O values.

Implications for understanding rainfall recharge to groundwater

Preferential-flow paths with their higher drip rates will contribute more rainfall recharge to groundwater volumetrically. This is supported by ref. 12: the estimated total flux from the ceiling area of Golgotha Cave with a higher proportion of drips fed by preferential flow was 39% higher compared with the ceiling area predominately supplied by the diffuse flow. This is also supported by water table measurements at Lake Cave (2000–2022; Fig. 1 for location) overlapping the period of drip discharge monitoring (2005–2022; Golgotha Cave) in Fig. 6. For example, higher rainfall in 2013 (Fig. 6a) resulted in a rapid rise in discharge for drips with a greater preferential-flow influence (Fig. 6b) and this coincided with the local water table also rising approximately 0.5 m (Fig. 6c). In comparison, the discharge from diffuse-flow dominated drips showed little to no response (Fig. 6b). We also compared the isotopic value of drip water and groundwater, and this further supports that rainfall recharge to groundwater beneath these caves occurs primarily via preferential flow. i.e., Fig. 5 shows that the distribution mode of preferential-flow-influenced drip water δ18O values (−4.67 ± 0.06‰VMSOW; campaign data) closely agrees with that of Holocene groundwater δ18O values (mode: −4.67 ± 0.03‰).

Fig. 6: Rainfall recharge and the response of drip discharge and the local water table.
figure 6

Higher rainfall (e.g., green bar) increases flow along preferential-flow pathways and recharge to the local water table. a Monthly meteoric precipitation and their smoothed anomalies. b Drip discharge from four speleothem sites in this study. c Water table data beneath Lake Cave, 3.6 km from Golgotha Cave. Precipitation and drip discharge data are for Golgotha Cave. Drip discharge data were calculated using drip rates from the long-term monitored sites and assuming a drip volume of 0.1433 ml (as in ref. 12) and an average of 30.4 days per month. Monthly precipitation anomalies were calculated using the climatological means from 1950 to 2020 and applying an exponentially weighted moving average with a half-life parameter of 36.6 months. Water table level data are relative to the first measurement (June 1999).

A reduction in preferential flow will also lead to a volumetric decline in rainfall recharge to groundwater. This is observed in a dataset of longer water table observations from another cave system in the region (Jewel Cave, Fig. 1 for location) that spans the period of the rainfall decrease (Fig. 3e). This record began in the 1950s and confirms a decline in the volume of rainfall recharge to the local water table that began around 1980 (Fig. 3e), in agreement with the beginning of the δ18O uptick. This volumetric decline in rainfall recharge to groundwater, as well as drier catchments with reduced runoff and fewer major flooding events have also been recorded in the Perth Basin and throughout southwest Australia22,42.

Rainfall recharge in the paleo record

Our study offers a well-constrained processed-based relationship between rising speleothem δ18O values in response to a sustained decrease in rainfall. Importantly, it has been shown that the uptick in δ18O was in response to a reduction in the preferential-flow component to drip waters that is likely linked to the seasonal duration of soil moisture, rather than a shift in mean annual rainfall δ18O values. The large magnitude of the δ18O uptick highlights that this mechanism amplifies the response of the speleothem δ18O signal to hydrological forcing. Thus, the identification of replicated upticks in speleothems can be used to identify past reductions in rainfall recharge to groundwater in a region.

The speleothem record from Golgotha Cave27 provides a multi-century context to assess the recent δ18O uptick. Stalagmite GL-S4 grew continuously since 1200 CE with a robust chronology constructed from annual laminae counting (Fig. 7a). It was identified as fed by diffuse drip water supplied from the matrix but also by preferential flow along fractures that activate during wetter years27. The other three speleothems were supplied either predominately by diffuse flow (GL-S1) or by complex fractures (GL-S2 and GL-S3). Thus GL-S4 is the most suitable record to assess variability in preferential-flow contribution in the longer-term context. Speleothem δ18O values for GL-S4 are approximately −3.9‰VPDB or lower, consistent with preferential flow along a feeding fracture contributing to the drip water for the record from around 1200 CE until the late-1990s CE, when the trend to higher values indicates that the contribution of preferential flow has rapidly decreased and that the drip water has become predominately supplied by diffuse flow (Fig. 7a, b). The modal value for groundwater on the calcite-equivalent scale is –3.9‰VPDB, and is also compared with the longer speleothem record on Fig. 7a, b. This study has linked recharge to the caves along preferential flow paths to rainfall recharge of regional groundwater, both isotopically (Fig. 5) and volumetrically (Fig. 6). Thus the longer GL-S4 speleothem record presented here implies that the southwest Australian region is currently experiencing a reduction in rainfall recharge to groundwater that is unprecedented for the last eight centuries.

Fig. 7: Recent uptick in δ18O values indicates a disconnection in rainfall recharge (reduced preferential flow) to groundwater that is unprecedented in the last 800 years.
figure 7

Cave record compared with reconstructed annual rainfall for southwest Australia43,60 and southern Australia44. a Golgotha Cave δ18O record: speleothem GL-S427 (solid purple) plus drip water (dashed purple) and composite record (red). b Distributions of drip water and Holocene groundwater δ18O values on the calcite-equivalent (VPDB) scale for comparison with the speleothem record. c Southwest Australian precipitation (olive) reconstructed from snow accumulation rates at Law Dome, Antarctica, where higher ice accumulation is related to reduced rainfall in southwest Australia43,60 and southern Australian precipitation (teal) reconstructed from multiple paleoclimate proxies44. Rainfall reconstructions are shown as box and whisker plots with long-term mean values shown as dotted lines, grouped into 45-year bins, whiskers show min/max and boxes show 25/50/75th percentiles. Reference 60 identified 1971 as a statistically significant change in the Law Dome-based rainfall reconstruction.

Our cave-based reconstruction builds on evidence for the hydroclimate history of southwest Australia from other paleoclimate archives. Figure 7a, c compares the speleothem GL-S4 rainfall recharge δ18O record with two reconstructions of rainfall for southwest Australia based on: (1) the Law Dome ice core record from Antarctica17,43 and (2) multiple paleoclimate records from the wider southern Australian region44,45 (see Fig. 1 for locations). The recent rainfall decrease is highlighted as particularly dry in these longer reconstructions with the ice core-based reconstruction indicating that the decrease in rainfall is also unprecedented for the last eight centuries. Our study presented here confirms this interpretation with evidence directly from the southwest Australian region rather than distal reconstructions based on atmospheric teleconnections. Importantly, the cave study demonstrates that the impact of the rainfall decline has resulted in an unprecedented decrease in rainfall recharge to groundwater for the last eight centuries. The chronology for the Law Dome-based rainfall reconstruction for southwest Australia is longer than that for GL-S4 and suggests two possible earlier intervals of comparable rainfall decrease during 385–429 CE and 732–776 CE. Extending the southwest Australian speleothem record in future studies will assist in understanding whether comparable dry periods that are persistent enough to impact rainfall recharge exist in the local record. Additionally, tree-ring records from Lake Deborah45 identify evidence for two prolonged droughts in around 1755–1785 CE and 1828–1859. While these may have been of insufficient duration and/or intensity to drive sustained reductions in preferential flow and disconnection to groundwater, the speleothem record does display relatively higher δ18O values in these two periods, particularly the early-mid-1800s, lending support to a reduction in rainfall recharge to Golgotha Cave at these times also. In addition, the tree-ring records should reflect greater hydroclimate sensitivity as these records are located in more arid inland areas (Fig. 1), compared to the relatively wet coastal fringe for our cave sites.


In this study, the δ18O of drip waters and calcite deposited from them in caves located in the vadose zone, and where recharge can be identified unambiguously as being from rainfall, have provided clear evidence of decreased rainfall recharge to groundwater resources. We confirm that no replicated uptick events are seen in the last 800 years in the stalagmite record from the southwest Australian region27 suggesting that the current decoupling of rainfall to groundwater recharge is unprecedented for the last eight centuries. This has important implications for contextualising the threat to the biodiversity in groundwater-dependent ecosystems and sustainable use of groundwater in a region heavily dependent on it for agriculture, and where future use is anticipated to increase22. The recent speleothem and drip water dataset indicates that rainfall recharge to groundwater may no longer be reliably occurring and highlights the immediate threat of climate change to water security in a region heavily dependent on groundwater.

Our findings are relevant for future climate change also as there is debate over the modelled changes in rainfall intensity and whether that would amplify hydrological responses. Runoff (i.e., flooding) has been the focus of research and adaptation planning so far46, with little attention until now given to the implications of future climate change on groundwater recharge. This study shows that a network of monitored caves in other regions where groundwater is relied upon could serve as a direct and unambiguous measurement of reduced recharge, particularly in karstified carbonate rocks that have flow paths associated with both primary porosity and fractures (e.g., as also observed in an Ethiopian stalagmite47) and widespread rainfall declines are expected this century as a result of human-caused climate change48. Monitoring also provides valuable data for ground-truthing speleothem paleoclimate records, while upticks in the speleothem paleoclimate records provide a longer-term context for evaluating if current groundwater recharge changes are outside the range of natural variability, thus the risks to groundwater resources from natural and human-caused climate changes.


Speleothem chronologies

Chronologies for all speleothems were published previously. Briefly, all stalagmites had been actively forming at the time that they were removed from the caves and the date of removal from the cave is used as the youngest age. The chronologies for CRY-S149, GL-S3 and GL-S427 were determined from annual laminae counting of 2D maps of Sr concentration; MND-S1 from annual laminae counting of Ba concentration from duplicate transects50; LAB-S1 from the peak fitting method of ref. 51 utilising principal components analysis of duplicate transects of Ba, Sr and U concentrations; while GL-S1 and GL-S2 chronologies were determined from bomb pulse detection using radiocarbon analyses52 and U/Th disequilibrium dating27. Mean annual extension rates for CRY-S1, GL-S4 and LAB-S1 through time were compared by calculating the mean and standard deviation of annual laminae width for each decade.

Speleothem δ 18O

Data were previously published for all records27,30,53 with the exception of CRY-S1 and LAB-S1. Milling and isotope ratio mass spectrometry methods are as described in ref. 27. CRY-S1 δ18O measurements were made at the Mark Wainwright Analytical Centre at UNSW Sydney using a MAT-253 isotope ratio mass spectrometer with Kiel carbonate device (25–50 μg samples). LAB-S1 δ18O measurements were made at the Research School of Earth Sciences, ANU, using a Thermo MAT-253 isotope ratio mass spectrometer coupled to a Kiel IV carbonate device (110–130 μg samples). Data in both laboratories are normalised to the Vienna Pee Dee Belemnite (VPDB) scale using NBS19 (δ18O = − 2.20‰ and δ13C = +1.95‰) and NBS-18 (δ18O = − 23.0‰ and δ13C = − 5.0‰).

Golgotha Cave drip water and farmed calcite δ18O

Drip water data, sampled at 4–6 weekly intervals; and farmed calcites, in situ for 9–31 months, were previously published in ref. 27. Data and methods for converting drip waters from the VSMOW scale to a calcite equivalent on the VPDB scale were also previously published27 using mean annual cave temperature measured at Golgotha Cave (14.6 ± 0.1 °C) and the bulk empirical fractionation factor for Golgotha Cave (1.0318 ± 0.0004) determined therein.

Composite speleothem record

Speleothem, and drip water δ18O data (as calcite-equivalent values) were interpolated to be annually resolved to construct the composite record. For Golgotha Cave speleothem records this represented a 3–5× increase in resolution, and for records CRY-S1 and LAB-S1, this represented a 2–3× decrease in resolution relative to the original data. Alternative methods were investigated such as maintaining the sampled resolution and weighting each data point according to the time interval that it represents, and the uptick in δ18O values is still robust i.e., the choice of method for compositing the data did not affect the interpretation. We considered uncertainty in the age-depth models but this is least at the youngest end. i.e., 0% at the year of collection and still small at 1900 CE (range: ± 2 to 18%, 2σ). It was not necessary to account for analytical uncertainty in δ18O as these are small (0.05‰ for carbonates and 0.1‰ for water)27.

Rainfall δ18O data

Annual rainfall amount for southwest Australia (line joining 30°S, 115°E and 35°S, 120°E; Fig. 1, inset) and the extent of the Tamala Limestone (Fig. 1) was extracted from Bureau of Meteorology gridded spatial data33. Annual and monthly amount-weighted δ18O was measured at the Perth Global Network of Isotopes in Precipitation (GNIP) site54 and Calgardup26. Predicted amount-weighted annual δ18O were calculated from an empirical relationship between daily δ18O and daily precipitation, Eq. (3) in ref. 26, applied to Bureau of Meteorology rainfall observations from the high quality55 sites Boyanup and Cape Naturaliste.

Soil moisture data

Modelled soil moisture data were extracted for each cave site from the Australian Bureau of Meteorology’s Australian Water Resources Assessment Landscape model: AWRA-L version 656. Briefly, AWRA-L is a continent-wide water balance model with 0.05° gridded outputs of soil moisture, runoff, evapotranspiration and drainage. We utilised daily soil moisture in the lower (0.1–1 m; term Ss) and deep soil store (1–6 m; term Sd), as well as rainfall (term Pg) and actual evapotranspiration (term Etot)56. Model output (Ss), validated against field measurements of soil moisture at Golgotha Cave38 (see Supplementary Fig. 2) and the Sd output, corresponding to the observed depth of seasonal wetting prior to initiation of percolation into the host rock39 is used to calculate the number of days per year when soil is saturated. The best match between observed and modelled soil moisture data is 66%; however, observed recharge at Golgotha suggests a higher percentile (78th) would be representative of fracture activation necessary for recharge, and we apply this to all five sites. See Supplementary Fig. 2 for further details.