Harrie Wood Cave (35∘44′ S, 148∘30′ E) is located in a limestone belt ca. 14 km long and 1.5 km wide along the Yarrangobilly River in the north of Kosciuszko National Park, New South Wales, Australia (Fig. 1a). The Yarrangobilly Limestone formed in the upper Silurian period from a coral reef (Worboys, 1982). Harrie Wood Cave is located within the Yarrangobilly Caves system, which includes over 250 independent limestone caves that began to develop in the Pleistocene (Worboys, 1982). Harrie Wood Cave is hosted within a highly fractured hard massive limestone and the drip sites are in close proximity to a fracture contact zone. The host limestone bedrock contains red/brown palaeokarst features and little to no dolomite. The cave entrance is ca. 965 m above sea level (a.s.l.) on a north-dipping steep rocky gorge. Harrie Wood is a restricted access, medium-sized south-dipping cave. The cave chamber is 80 ± 2 m in length and 34 ± 1 m deep (Nicholl, 1974).

Surrounding vegetation consists of open snow gum (Eucalyptus pauciflora subsp. Pauciflora) and black sallee (E. stelullata) woodland dominated by a snow grass (Poa sieberi) understorey (Aplin et al., 2010). Above the cave, vegetation coverage is sparsely developed on shallow rocky soil that lacks clearly defined horizons, and are dominated by angular clasts (typically 2–10 cm size) indicating mechanical weathering processes. The surface over the cave was burnt by an intense wildfire in 2003 and the shrubby vegetation that is present over the cave shows evidence of regeneration post-fire (Coleborn, 2016). There is no evidence of a distinct zone of infiltration on the surface directly above and upslope of Harrie Wood Cave. There is also no evidence of surface runoff following rain events. We interpret from our field observations that infiltration is via pervasive cracks and fissures.

The Yarrangobilly Caves system is part of the Australian Alps bioregion and is dominated by a montane climate (Stern et al., 2000), being characterized by mild dry summers and cold wet winters (Fig. 1c). The median annual rainfall at the Bureau of Meteorology (BoM) weather station at Yarrangobilly Caves (BoM station 72141), calculated using the climatological median 1985–2013, is 1178 ± 29 mm with a winter maximum of 349 ± 15 mm. The two dominant synoptic weather categories delivering most of this cool-season rainfall to SE Australia are cut-off low-pressure systems out of the westerlies and systems from the mid-latitudes including mid-latitude storms and fronts (Chubb et al., 2011; Pook et al., 2014; Callow et al., 2014). Modelled total evapotranspiration (ET) is maximum in summer and minimum in winter, and mean annual ET is 838 ± 40 mm over the period 1961–1990. The site has a mean annual temperature of 10.5 ∘C, with mean maximum temperatures in January and mean minimum temperatures in July of 27.8 and -1.8 ∘C, respectively. The growing season is limited by cold winter temperatures.

The three drip-water monitoring sites (HW1–3) in this study are measured from active stalactites (Table 1) located centrally within the cave (Fig. 1b) at a depth of 38 m to the surface. HW1 and 2 were feeding actively forming stalagmites ca. 0.5 m apart on either side of the main path. These stalagmites were removed for palaeoclimate studies in 2006. In order to remove HW2, a small adjacent calcite column that had formed from a fused stalactite–stalagmite pair ca. 10–15 cm from HW2 (see Supplement S1 for photograph), had to be removed also. This resulted in re-invigoration of the drip point that had formed the column, which we included in our sampling program and refer to here as HW3.

Unsaturated zone hydrology of Harrie Wood Cave has recently been characterized (Markowska et al., 2015) and 5 drip-water regimes were identified. All flow types are fed by a theoretical soil storage and epikarst storage reservoir by fracture/fissure drainage. The five discharge flow types are as follows. Type 1 is designated mixed-flow/storage connectivity (low flow/high flow). Water at these discharge points is drained from a bulk homogenized epikarst storage reservoir. At the high-flow sub-type, during periods of water excess, the epikarst store is bypassed and water is routed directly from the soil storage reservoir. At the low-flow sub-type, discharge is from a pocket reservoir with a variable head within the epikarst storage reservoir. With Type 2 extreme events activated drip sites, a large intense infiltration event is required to initiate flow from the epikarst store. Type 3 are overflow sites, with discharge at the drip point fed by overflow from the pocket reservoir. Type 4 are non-linear flow sites; based on intra-karst dynamics, flow is intermittent between both storage reservoirs. Type 5 are underflow sites; during high infiltration, discharge is preferential underflow where both reservoirs are bypassed, and during base-flow conditions flow is sourced from the epikarst store. The sampling points in this study, HW1–3, are situated within the transect monitored by Markowska et al. (2015) (Fig. 1b).

Daily rainfall was measured from the BoM standard 203 mm rain gauge at 09:00 local time (LT) each morning. Provided precipitation was 2 mm or greater, an aliquot of this precipitation was collected in a 10 mL amber bottle, ensuring zero headspace, and stored for stable isotope analysis (δ18O, δ2H). Maximum and minimum air temperatures are also recorded at 09:00 LT using BoM standard procedures.

A network of two automatic weather stations were installed above Harrie Wood and Jillabenan Caves on the 14 October 2011 and 6 September 2012, respectively. Atmospheric measurements of pressure, humidity, rainfall, temperature and wind speed and direction are recorded by a Davis Vantage Pro2^™. Within the soil zone above each cave system, soil sensor probes were installed and buried in holes drilled with an auger to a depth of 25–30 cm at three localities (Fig. 1b). The Stevens Hydra Probe^® soil sensor measurements include temperature, soil moisture and electrical conductivity and complex dielectric permittivity (both corrected from 0 to 35 ∘C). All data are recorded every 15 min by a dataTaker DT80 data logger. A detailed description of the parameters measured by the various instrumentation and data available from the network are outlined in Supplement S2.

Drip water analysed for this study was collected approximately fortnightly over the period July 2006 to December 2013 (n = 468). The drip rate was measured manually as the time interval elapsed between two drips recorded using a stopwatch and the drips emanating from the stalactites accumulated in 1 L HDPE containers. Since March 2011 and sample volume permitting, in situ field measurements of electrical conductivity (referred to 25 ∘C; ±1 %), temperature (±0.1 ∘C) and pH measurements (±0.01 pH) were made using a TPH1-MyronL TechPro II handheld meter, which was calibrated using buffer solutions that were kept at cave temperature.

At each drip site, aliquots from the bulk water sample were collected and filtered through a mixed cellulose ester 0.45 µm filter and split into (i) two clean 50 mL polypropylene bottles for cation and anion analysis and (ii) a 10 mL amber glass bottle for δ18O and δ2H analysis. On exiting the cave, samples were refrigerated, transported to the laboratory in an insulated container and maintained at 4 ∘C until analysis.

Cation and anion analysis was conducted at the Environmental Research Chemistry Laboratory at the Australian Nuclear Science and Technology Organisation (ANSTO). Cation analysis was carried out on a Varian^™ Vista Pro AX ICP-AES; prior to analysis aliquots were acidified with 1 % HNO3. Anion analysis was conducted on an un-acidified aliquot using a Dionex 600 Instrument with an auto suppressor. Analytical error on cation and anion analyses was ≤ 5 %.

Representative limestone samples (n = 7) were collected from bedrock exposures above the cave and within Harrie Wood Cave. Freshly cleaved samples were dried at 40 ∘C and powdered using a ROCKLABS^® TC-40 tungsten carbide ball mill. Then, 0.2 g of sample was microwave digested at 180 ∘C for 15 min using aqua regia (HCl / HNO3 = 1 : 3) and analysed by ICP-AES at the Environmental Research Chemistry Laboratory at ANSTO.

The isotopic composition (δ18O and δ2H) of the rainfall and drip-water samples were performed on a LGR-24 d off-axis, integrated cavity output, cavity ringdown mass spectrometer at University of New South Wales (UNSW) Australia. δ18O and δ2H values are reported relative to V-SMOW2. Analytical precision for δ18O is 0.17 and 0.6 ‰ for δ2H.

To determine whether the geochemical evolution of the drip water was principally due to prior calcite precipitation (PCP), the hydrochemistry was assessed based on the accepted mathematical methods after Sinclair et al. (2012) and Tremaine and Froelich (2013). First, to compare between the two methods, ln(Sr / Ca) (mmol mol-1) vs. ln(Mg / Ca) (mmol mol-1) ratios in dissolved host bedrock endmembers overlayed with the drip-water ratios was graphed. The Mg / Ca and Sr / Ca ratios during the step change in 2007, wet period and high dry-season Ca values were differentiated from the complete dataset to isolate processes affecting the drip-water chemical evolution during these periods. Then, at each drip site the slopes of the linear regression of the ln(Sr / Ca) vs. ln(Mg / Ca) (weight ratio) graph was calculated based on the model suggested by Sinclair et al. (2012), where a slope of 0.709–1.003 is an indicator of water–rock interactions, i.e. PCP and incongruent dissolution-driven processes.

The Ca concentration and Mg / Ca and Sr / Ca ratios for the suite of bedrock samples are listed in Table 2. Bedrock Mg / Ca ratios range from 2.6 to 15.3 mmol mol-1 and Sr / Ca ratios lie between 0.03 and 0.32 mmol mol-1. These low ratios indicate the host limestone rock above and within Harrie Wood Cave are a low-Mg calcite type, indicating diagenesis of the original bedrock material. We observe three groupings of samples; Ca does not vary significantly between the samples but there is an observed difference in the Mg / Ca and Sr / Ca ratios between groups. The ratios for samples (R7–R10) are similar and there is no difference between colour variations, the limestone conglomerate (R12) has the highest Mg / Ca ratio and the palaeokarst samples (R13, R14) have the lowest Mg / Ca and Sr / Ca ratios.

The observed daily rainfall and monthly cumulative water balance (CWB) at the study site and the Southern Oscillation Index (SOI) are shown in Fig. 2. The CWB represents a residual mass curve to show trends in the cumulative monthly water budget trends, following the method of Hurst (1951). It is calculated as the cumulative sum of the monthly P–ET anomalies from the climatological mean (1961–1990). The earlier half of the record overlaps the latter 4 years of the “Millennium Drought”, including SE Australia's most persistent rainfall deficit that occurred between 1997 and 2009 (CSIRO and BoM, 2015). Hence, the site experienced below-average rainfall from 2006 to 2009, and for much of the preceding decade. For example, from May 06 to December 06 (Fig. 3a), annual rainfall totals were 52.6 % below the mean attributable to a weak El Niño event (BoM, 2015a). The years 2007–2009 were a period of generally dry conditions, where the annual total precipitation was on average 10.3 % below the 30-year average. There were exceptions of above-average monthly rainfall for February and May 2007, November 2007 to April 2008, July 2008 and in April, July and September 2009 (Fig. 3b; BoM, 2015b). However, the reduced annual totals until 2010 resulted in a decline in the calculated cumulative water balance (Fig. 2b).

By contrast, in 2010–2012, average annual rainfall at Yarrangobilly Caves (BoM station 72141) was 152, 135 and 115 % above the mean for these 3 years, due to the strong 2010–2012 La Niña event (BoM, 2015a). The wettest interval in our study occurred from July 2010 to March 2011 (Figs. 2b and 3c) during which the calculated cumulative water balance increased (Fig. 2b). The subsequent 6 months from May to October  2011 were relatively dry (Fig. 3d), followed by a 5-month period of above-average monthly rainfall from November 2011 to March 2012 (Fig. 3e).

Based on measurements of soil moisture saturation between 14 October 2011 and 9 January 2013, Markowska et al. (2015) interpreted that a daily rainfall threshold range between 13 and 31.4 mm for Harrie Wood Cave was required to initiate a discharge response at 14 monitored drip sites. During the drier 2006 to mid-2010 period of our study, there were 75 such events in total, representing 30 % of total rain days. By contrast, during the wetter mid-2010 to 2014 interval, there were 124 events, representing 37 % of rain days.

Daily rainfall, δ18O, monthly precipitation-weighted mean (PWM), as well as drip-water δ18O, are shown in Fig. 2. The monthly PWM of rain water δ18O values ranges from -13.6 to -2.1 ‰ over the sampling period, with an overall δ18O PWM of -6.9 ‰ of all rainfall events. At Harrie Wood Cave, rainfall events greater than 13 mm initiate recharge (Markowska et al., 2015); the δ18O PWM value from these recharge events (-6.7 ‰) is not significantly different than that of all rainfall events. The monthly δ18O PWM values show a large 5–10 ‰ intra-annual variation following the general seasonal trend of depleted isotopic values in winter and isotopically enriched values in summer, although the winter isotopic depletion is much less pronounced in 2008, 2009 and 2011. During the dry period prior to mid-2010, the PWM was enriched (-6.2 ‰) compared to the 2010/2012 wet period (-6.9 ‰).

With regards to drip water, there is ca. 0.5–1 ‰ variability over the observation period, which is largely dampened compared to rainfall. Site HW2 had a wider range of δ18O values (-5.4 – -7.7 ‰) than site HW1 (-6.0 – -7.4 ‰) and HW3 (-5.7 – -7.9 ‰). The δ18O arithmetic mean values, close to the PWM (-6.9 ‰), were -6.8 ± 0.3 ‰ for HW1 and 2 and -6.9 ± 0.3 ‰ for HW3, suggesting significant mixing of infiltration events and that evaporation of infiltrating water is not significant. There is no clear seasonality in drip-water δ18O, in contrast to the rainfall δ18O, and there appears to be no simple isotopic response to infiltration of winter rainfall (Fig. 2). Furthermore, at all sites the mean drip-water δ18O values are similar prior to mid-2010 during the dry period compared to the wettest interval between mid-2010 and March 2011 (HW1: -6.7 ‰ cf. -7.0 ‰; HW2: -6.6 ‰ cf. -6.9 ‰; and HW3: -6.9 ‰ cf. -6.9 ‰), although there is a subtle trend to lighter δ18O values.

Figure 4 shows δ2H vs. δ18O calculated local meteoric water line (LMWL; δ2H = (8.11 ± 0.08) × δ18O + (15.9 ± 0.5), n = 415) compared with the global meteoric water line (GMWL; Rozanski et al., 1993). The LMWL intercept is greater than the GMWL, but is in agreement with the LMWL established by Hughes and Crawford (2013) from a 3-year precipitation dataset from Big Hill in the Southern Highlands (δD = (8.10 ± 0.12) × δ18O + (16.3 ± 0.8); 188 km north-east (NE) of our site, 652 m a.s.l.). Crawford et al. (2013) attributed the high intercept value to a larger contribution of moisture recycling from the land surface to the local moisture budget.

Drip rates throughout the 7-year period demonstrate irregular multi-annual variations (Fig. 2d). All sites remained hydrologically active, with a mean drip rate of 2.2 ± 0.6, 2.2 ± 1.7 and 3.5 ± 1.9 × 10-3 mL s-1 at site HW1–3 respectively (Table 3). Prior to mid-2010, during the long phase of below average monthly rainfall and water deficit, a base-flow drip rate of 2.1 ± 0.5 × 10-3 mL s-1 for HW1, 1.9 ± 1.4 × 10-3 mL s-1 for HW2 and 3.1 ± 1.3 × 10-3 mL s-1 for HW3 was maintained, only increasing slightly in response to above-average monthly rainfall anomalies during this period (Sect. 4.2), therefore suggesting the drip sites receive flow from a storage reservoir. The increased rise in discharge is often only captured by one fortnightly sampling event, but is often present at all three sites. During the wettest interval between mid-2010 and March 2011, the mean drip rate increased to between 2.8 and 4.7 × 10-3 mL s-1 (Fig. 2d). Site HW2 is a slower dripping site compared to HW1, and HW3 is the fastest dripping stalactite. Drip rates at all three sites are mostly in phase, recording a similar pattern.

Concentrations of major cations and anions in the drip-water samples from the three drip sites are listed in Table 3. Drip-water samples from all three sites represent a Ca–HCO3 dominating water type (pH 6.8–8.3) (Drever, 1982); accordingly, Ca is the main cation in the drip-water solutions. The mean Ca concentrations observed at drip sites HW1–3 are 66.4, 56.4 and 63.0 mg L-1 respectively, and range from 31.1 to 84.1 mg L-1. Mean concentrations of other ions are similar across the three drip sites, further supporting that the drip water is sourced from the same well-mixed storage reservoir.

Time series of Ca, Mg / Ca, Sr / Ca and Cl are shown in Fig. 2. Broadly, Ca concentrations are declining from 80 to 60 mg L-1 at sites HW1 and 3 from the beginning of the study until early 2008 when they become more similar to one another, but still greater than HW2 (which remains lower at 50 mg L-1 through the same period). There is a prominent rise in Ca at all sites beginning in the summer of 2008/2009 and peaking in November 2009 at ca. 80 mg L-1, before falling back to ca. 60 mg L-1 in autumn 2010. After this time, and throughout the following wetter interval, Ca concentrations at all sites become closer in value and less variable overall with some excursions to lower Ca at individual drip sites over periods of 0.5–2 months.

The prominent Ca peak in 2009 also occurs during a dry period when Ca is increasing rather than decreasing. In this case, this prominent peak in Ca occurs during a persistent run of above-average surface temperatures dominating the entire year (Fig. 3). Specifically, an unusually warm winter was experienced in 2009 (Fig. 3f), followed by the hottest November on record (BoM; 2015b). We propose that, in this interval, higher than average drip-water Ca concentrations are driven by a temperature control on net soil CO2 due to respiration. We therefore propose that the warmer than average temperatures, particularly during the winter months, induced higher soil CO2 respiration and thereby increasing CaCO3 dissolution. Drip-water Ca returned to mean levels in January 2010 (Fig. 2f), consistent with cooler temperatures (minimum temperatures were 3.1 ∘C lower than the long-term average) restoring soil respiration conditions to previous levels (Fig. 3g).

Chloride concentrations show consistent trends through time across the three sites. Chloride concentrations are the highest at the beginning of our study (ca. 2.3 mg L-1) and decline steadily until mid-2010; at which time there is a more abrupt decrease in mean Cl at all three sites. After this transition, Cl concentrations are relatively stable at ca. 1.3 mg L-1. Chloride is our most conservative of the measured ions. During the transition from the relatively drier to the relatively wetter interval in our study, the abrupt decline in Cl, coinciding with an increase in discharge is consistent with dilution by recharge at the onset of the wetter interval (Fig. 2i). The earlier downward trend in Cl (2007–mid-2010) appears initially inconsistent with this, but we propose that reduced recharge events to the epikarst storage reservoir during this interval resulted in a decreasing supply of Cl from the soil zone (discussed further in Sect. 5.1).

The observations above suggest processes include the mixing and dilution of drip water as indicated by Cl concentrations. To diagnose water–rock interactions including PCP and mixing of waters, we used a number of approaches (Sect. 3.3). Figure 5 shows a cross-plot of ln(Mg / Ca) vs. ln(Sr / Ca) drip-water ratios presented together with the bedrock ln(Mg / Ca) and ln(Sr / Ca) ratios (after Sinclair et al., 2012, and Tremaine and Froelich., 2013). Since all three drip sites are influenced by the same processes, as the same Mg(Sr) / Ca trends are observed at all sites (Supplement S3), we present drip-water data at site HW2 for visual clarity as the processes are enhanced due to the slower discharge at this site. Many of the bedrock values are similar but lower than the drip-water ln(Mg / Ca) and ln(Sr / Ca) ratios (Fig. 5). The drip-water data themselves appear to fall into two groups: the pre-October 2007 group (red squares) with relatively lower Sr / Ca ratios, and the post-November 2007 group with relatively higher Sr / Ca ratios. If the geochemical evolution of the drip water was solely due to PCP, Sinclair et al. (2012) mathematically demonstrated that the drip-water ratios would produce a straight-line correlation of known slope through the whole dataset (Sect. 3.3). The observed step change indicates a change in the initial composition of the chemistry at equilibration. We interpret this to indicate a flow path change, which we discuss further in Sect. 5.1. All HW2 drip-water values are consistent with evolution from the bedrock values, notably R7 to R10.

Linear regressions are plotted for each group of data in Fig. 5. There is some scatter of the data around these lines suggesting that more than one process may be influencing the resulting drip-water Sr / Ca and Mg / Ca ratios. During low discharge and dry rainfall conditions the drip-water composition shifts to higher Mg / Ca and Sr / Ca ratios away from the bedrock plotting in the distal outer ends, representing progressive CO2 degassing due to increased PCP. During the transition period from dry to wet conditions in mid-2010 when discharge increased (blue filled circles in Fig. 5), drip-water Mg / Ca and Sr / Ca ratios plot closer to the bedrock region, indicating reduced PCP. During the high dry-season Ca values (November 2009, square window), Mg(Sr) / Ca ratios are close to the bedrock ratios. As drip water reached maximum Ca values, Mg(Sr) / Ca ratios evolved towards slightly lower values, due to dissolution of the host limestone, caused by increased soil CO2 from bioproductivity. An alternative hypothesis for the rise in Ca in late 2009 could be a flow path change; however, the Sr / Ca and Mg / Ca data show no evidence to support this.

The calculated drip-water ln(Sr / Ca) vs. ln(Mg / Ca) (weight ratio) correlation slopes of the two groups of data at each drip site range between 0.66 ± 0.17 and 1.01 ± 0.07, within the range of predicted slopes suggested by Sinclair et al. (2012). Although a slight slope change is observed between the two flow paths, this suggests that cave drip water evolves under PCP.

Unlike the late 2007 step change, we observe a rise in Mg(Sr) / Ca and a drop in Ca in 2011 and 2013 (arrows; Fig. 2). These events are short-lived and are most notable at one site only (HW2); there is insufficient data to evaluate this site-specific change and nonetheless are features that would not be preserved in the speleothem.

Excluding 2007, the long-term Mg values are increasing relative to Ca and there is a drift towards higher Mg / Ca values overall but less evident in the Sr / Ca time series (Fig. 2). There is a long-term rise in Mg concentrations of 0.15 ppm from 2008 to 2013. A more complete characterization will include an investigation of aerosols and soil in a future study.

Based on hydrogeochemical observations presented in Sect. 4, we suggest in Harrie Wood Cave the flow paths feeding stalagmites from drip sites HW1–3 are from a Type 1, mixed-flow/storage connectivity low-flow sub-type, i.e. fracture flow from a “pocket reservoir” in the well-mixed epikarst storage reservoir (Markowska et al., 2015; see Sect. 2.2). Our data confirm the drip water that precipitates speleothem calcite drains from bulk homogenized epikarst store water, the primary karst storage reservoir at Harrie Wood Cave. The results from the oxygen-isotope data indicate that infiltrating water is well-mixed, because the range of drip-water δ18O values is narrow in comparison to the rainfall, and the arithmetic mean of the drip water reflects the weighted mean of precipitation (Sect. 4.3); moreover, from the drip-water isotope time series there is a strong buffering of the extremely low δ18O winter rainfall values in 2007, 2010, 2012 and 2013 (Fig. 2). Furthermore, consistency in trace element trends among the sites (Table 3) confirms the karst waters are well homogenized.

A significant proportion of flow to the drip points is drained from a storage reservoir, owing to sustained base-flow discharge levels during the weak El Niño in 2006 and the low rainfall period between 2007 and 2009 (Sect. 4.4). Furthermore, as base flow is maintained and a large infiltration event or threshold level is not required to activate the drip site; this suggests the karst plumbing system feeding the stalagmites is not as described by Markowska et al. (2015), i.e. a Type 2 extreme event activated or Type 3 overflow site (see Sect. 2.2).

Dilution of the epikarst storage reservoir was observed during the transition to the La Niña phase in 2010/2011, consistent with the observation of a clear increase in discharge followed by a decrease in Cl concentrations. This further suggests that after recharge the epikarst storage reservoir volume reached close to maximum capacity maintaining “steady state”, supported by the drip-water Mg / Ca and Sr / Ca ratios reaching mean levels indicating minimum PCP. Additionally, this indicates that discharge at the drip sites are not from a Type 1 (high flow sub-type), Type 4 or Type 5 site, as during periods of water excess discharge at these sites receive flow directly from the soil storage reservoir, and therefore a rise in Cl would be anticipated due to higher ET to the system as a whole; however, the converse was observed.

During the dry period from 2007 to mid-2010, we suggest the decreasing trend from elevated Cl levels indicates limited recharge. During the dry period we observe a decline in the CWB and reduced drip rate, consistent with fewer recharge events from the soil–vadose zone to the epikarst store, and therefore decreasing drip-water Cl concentrations. Alternatively, it may also reflect higher ET in the soil–vadose zone, but this would also result in an enriched drip-water δ18O signal. ET is relevant at this karst site; however, we do not favour this explanation, as the similarity between drip-water δ18O values during the dry period compared to the wetter period suggests evaporation of the infiltrating drip water was not significant (Sect. 4.3).

A flow path change at October 2007 is inferred from the shift to higher Sr / Ca ratios (Sect. 4.6; Fig. 5). This process occurred during the drying conditions when the CWB was decreasing (Fig. 2b), but when base-flow levels were maintained at all drip sites during this period. We infer that a flow path re-direction through a higher Sr / Ca endmember occurred due to soil–vadose zone drying to sustain discharge. This indicates a non-linear response of the system to progressive drying. A possible explanation is that calcification of the flow path produced a threshold change causing re-routing through a higher Sr / Ca source.

The three drip-water monitoring sites (HW1–3) are located within a small area (Fig. 1b); however, there are differences in the chemistry, which is notable in 2007 during the earliest interval of the dry period. In particular a lower drip rate (Sect. 4.4) and a lower Ca concentration and higher Mg / Ca and Sr / Ca ratio (Sect. 4.5) of site HW2 compared to the neighbouring sites HW3 and 1, which are only 10–15 cm away. Equivalent Cl concentrations over this period (Sect. 4.5) suggest differences have subsequently arisen in the carbonate chemistry. A potential explanation for this is that all three drip sites are fed by the same chemistry, but the lower drip rate at site HW2 is consistent with greater in-cave PCP on the stalactite tip, inducing a lower Ca concentration and enhancing PCP (Treble et al., 2015).

Thus, we deduce that the studied drip water is from a Type 1 mixed-flow/storage connectivity low-flow sub-type, the system is open to CO2, and a ventilated air pocket with variable height and lower pCO2 provides the potential for degassing and calcite precipitation from the drip water (Tooth and Fairchild, 2003). Although we observed a hydrological flow path change, a distinction in drip-water Mg / Ca and Sr / Ca composition between drier and wetter periods is evident. During dry periods, the PCP mechanism was enhanced and the highest Mg / Ca and Sr / Ca ratios in this study are recorded during the El Niño and intervals of below average rainfall, as PCP is promoted due to a dewatering of spaces. In contrast, for the duration of the strong La Niña phase in 2010/11 and above-average rainfall, reduced PCP (mean Mg / Ca and Sr / Ca ratios) is noted (blue circles, Fig. 5) since a reduced reservoir head space limits degassing.

We also examine the role of calcite dissolution and soil zone CO2 in more detail. Drip-water Ca concentrations reached maximum values (80 mg L-1) in November 2009 (see Fig. 2f), as unseasonal temperatures increased. We suggest that soil microbial production increased the CO2 concentration of infiltrating waters driving CaCO3 mineral dissolution (square window, Fig. 5). Presumably vadose zone CO2 production also drives calcite dissolution during the dry period (Atkinson et al., 1977). During the dry period there is a long-term decline in the CWB and progressive draining of the epikarst store; therefore, more unsaturated zone is available for CO2 production, which may also lead to increased calcite solubility.

An alternative hypothesis is the Ca trend may be a fire-driven process as the site was affected by fire 4 years before the monitoring period, which may have decreased soil CO2 production, associated calcite dissolution and Ca concentrations (Coleborn et al., 2006). That effect is most likely in the first decade after the fire. However, a decreasing trend in Ca in the early monitoring period (the opposite of what might be expected) in combination with thin soil and sparse vegetation indicates that any fire-induced soil biogeochemical changes had relatively little impact on the drip-water signature. Also, in the short term, the concentration of elements in the soil can increase in response to a fire. Considering this, during the drying trend when there is reduced recharge and based on the observed trend of the conservative tracer Cl, the effect would be an analogous decrease in Mg and Sr concentrations in the drip water, whereas a long-term rise in Mg is noted (Sect. 4.6), as such we do not consider the 2003 fire event had an impact on the drip-water dataset.

Based on a 7-year observational study, we have unveiled a complex non-linear geochemical response, as depicted in Fig. 6. We now attempt to discern whether these long-term trends in the geochemistry are climate related. During the drying trend (2007 to mid-2010), we observed low discharge, a flow path change, decreased Cl, enhanced and increased Ca caused by increased CaCO3 solubility due to higher soil CO2 and bioproductivity. This is a response consistent with reducing recharge conditions. During the wetter period, we also observed trends consistent with increased recharge, increased discharge, dilution and reduced PCP. Therefore, we interpret these as evidence of climatic induced changes. Over the long-term, drip discharge and hydrogeochemical variations between the relatively dry and wet period are driven by variations in water availability due to ENSO and are of palaeoclimatic relevance.

We have confidently constrained the possible hydrological processes, which can occur in a karst system situated in a region greatly impacted by ENSO episodes. We attribute this to long-term monitoring that encapsulated the shift between the two extremes of ENSO, namely from dry El Niño to wet La Niña conditions. Based on our findings and with the knowledge that the impact of each ENSO episode on Australia varies, we have identified how Harrie Wood Cave responds to a changing climate signal (rainfall), and through geochemical proxies (δ18O, Ca, Cl, and Mg / Ca and Sr / Ca ratios) identified how this response has been transformed and is recorded in cave drip water over time. These results have important implications in informing palaeoclimate findings from speleothem archives, which we discuss further in Sect. 5.2.

Our results provide a foundation to inform speleothem palaeoclimate records where drip-water compositions are influenced by multiple processes. During our studied interval, element concentrations and ratios in the drip water were driven by climate (ENSO) and karst hydrological processes. Based on drip characterization of sites HW1–3, Mg / Ca variations in the conjugate stalagmite could in principle, be successfully applied to construct palaeorainfall conditions (Tooth and Fairchild, 2003; Markowska et al., 2015). We have demonstrated that drip-water Mg / Ca and Sr / Ca ratios are a relative measure of modern rainfall variability. Therefore, in a highly resolved speleothem time series, we anticipate displacements of [Mg / Ca]calcite and [Sr / Ca]calcite from a baseline ratio (bedrock) to higher and mean ratios will differentiate between dry and wet periods respectively. McDonald et al. (2004) also identified a relationship between drip-water Mg / Ca and Sr / Ca ratios and drought in Wombeyan Caves, New South Wales (NSW); however, their interpretation of the trace element stalagmite record varies slightly. At Wombeyan, PCP is a more dominant control on the karst hydrochemistry, although it is in the same climate region, but a warmer site with greater evapotranspiration and a longer growing season. Consequently, we would expect that the resultant drip water (and stalagmite calcite) would be enriched in Mg / Ca and Sr / Ca. As such we expect to observe a greater displacement of Mg / Ca ratios away from the bedrock composition during any given dry phase at Wombeyan Caves, in comparison to a corresponding record at Harrie Wood Cave. For wet periods, no La Niña events occurred during their study period (July 2001–January 2004); therefore, McDonald et al. (2004) could only speculate Mg / Ca calcite ratios would shift towards bedrock values due to a decreased effect of PCP. In contrast, extended monitoring through the La Niña mode of an ENSO cycle provides us with a more informed interpretation; whereby we anticipate La Niña events in the stalagmite trace element time series may be resolved from baseline ratios by a shift to mean ratios.