A post-wildfire response in cave dripwater chemistry

23 Cave environments are sensitive because surface environmental changes can affect both the 24 isotopic composition and solute concentration of infiltrating cave dripwaters. These changes 25 subsequently affect the speleothem geochemical record. One such agent of change is wildfire, 26 however its effect on karst processes remains poorly understood. Using dripwater data from a 27 shallow cave, at a forested site in southwest Australia, we provide a unique analysis of the post28 wildfire effects on dripwater O and solute concentrations. We analyse how wildfires affect 29 on local controls, i.e. vegetation cover, evapotranspiration and carbonate water-rock 30 interactions, effects cave dripwater hydrology and geochemistry. We compare our post31 wildfire data with modelled drip water O, regional groundwater chemistry, and a second 32 cave dripwater dataset, to determine the extent to which wildfire affects cave dripwater 33 composition. We find in our forested, shallow cave site, by effecting surface evaporation and 34 transpiration rates wildfire can have a multi-year impact on subsurface hydrology and 35 dripwater chemistry. Here we open a new avenue for speleothem science in fire-prone regions, 36 focusing on the geochemical records of speleothems as potential paleo-fire archives. 37


Introduction
Caves are observatories, that preserve invaluable geochemical archives of past-climates; in the form of speleothems (stalagmites, stalactites and flowstones).The existing paradigm in speleothem science has largely focused on establishing paleoclimate proxies in stalagmites (e.g.McDermott et al., 2001;Treble et al., 2008;Woodhead et al., 2010).While these proxies are useful for reconstructing paleoclimates, their interpretations may hold a predisposed bias towards using these proxies as indicators of paleoclimate only.
To avoid this bias, we need to consider the sensitivity of these proxies to the effects of local environmental factors like in our case, fire.This is especially important as incorporating this perspective may not only be used to correct the climate proxy interpretation, but also yield novel information about paleo-environments.Paleo-environmental proxies are verified by conducting process-based in-cave monitoring studies.However, in-cave monitoring has predominantly focused on understanding the extent to which dripwater  18 O (Lachniet, 2009), dripwater solute concentrations (Fairchild and Treble, 2009), speleothem calcite growth (Wong et al., 2011) and cave CO 2 processes (Breecker et al., 2012), are affected by climate.Further, such studies have largely been restricted to mid to high latitude climate regions where precipitation (P) is larger than actual evapotranspiration (AET), and climate is likely to be a major control on dripwater composition.
In water-limited regions, dripwater chemistry is influenced to a greater extent by environmental factors such as evaporation (E) (Pape et al., 2010;Cuthbert et al., 2014;Rutlidge et al., 2014) and transpiration (T), (Tremaine and Froelich, 2013;Treble et al., 2016).Wildfires, common in water-limited regions, are agents of change than can dramatically alter evaporation and transpiration rates by destroying vegetation.The potential impacts of vegetation loss from fire are both short-term and long-term.The short-term impacts could include: (1) an increase in evaporation rates due to changes in albedo and/or lack of shading (Silberstein et al., 2013); (2) a reduction in transpiration from reduced tree water use; (3) a reduction in soil microbial and root CO 2 production (Coleborn et al., 2016); (4) a decrease in cave CO 2 due to the destruction of vegetation (Wong et al., 2010), which could influence in-cave prior calcite precipitation (PCP); (5) the addition of plant ash to the soil profile, increasing concentrations of Ca, K, Mg, and S (Grove et al., 1986;Yusiharni and Gilkes, 2012a); and (6) altered infiltration patterns (González-Pelayo et al., 2010).While long-term impacts include: (1) the spatial redistribution of nutrients (Abbott and Burrows, 2003); (2) regrowth impacts on water balance and nutrient flux (Treble et al., 2016); and (3) a reduction in total soil CO 2 due to the destruction of CO 2 sequestering microbial communities and plant roots, both significant sources of soil CO 2 (Coleborn et al., 2016).Despite the fact that wildfires regularly affect water-limited regions, their impacts on  18 O and solute concentrations in cave dripwater have not been reported.
We analyse the composition of cave dripwater over five years (August 2005-March 2011) of cave monitoring in Yonderup Cave, a shallow cave system, in southwest Australia.Our monitoring followed an intense wildfire in February 2005 that burnt 1200 ha of Yanchep National Park.The fire was hot enough to calcine and fracture the limestone observed at the caves entrance (Supplementary Fig. 1).We compare our monitoring data to the regional groundwater geochemistry and published monitoring data (Treble et al. 2015) from Golgotha Cave in southwest Australia (lat.36.10 o S, long.115.05 o E).Our analysis provides one of the first analyses of the response of dripwater  18 O and solute concentrations to post-wildfire conditions in shallow caves located in the tree rooting zone.
Our study was conducted in Yonderup Cave in Yanchep National Park (lat 31.5475o S, long 115.6908 o E), 20 km north of Perth, southwestern Australia (Fig. 1A).This region has a Mediterranean climate characterised by dry hot summers and cold wet winters with a 25-year (1990 to 2015) average annual surface temperature of 15.1 o C and rainfall of 664 mm with 85% of rainfall falling between May and October.Yonderup Cave is located in the young Quaternary Tamala Limestone Formation, a porous, partially lithified calcareous coastal dune sand.This karst process is said to be "syngenetic" with karstification occurring simultaneously with lithification of the host rock (Jennings, 1964;Fairchild and Baker, 2012).
Yonderup Cave is situated in a tuart forest (Eucalyptus gomphocephala), with mature tuart trees 30 m high, and an understory of shrubs and trees standing 5-10 m high, Sheoak trees (Allocasuarina fraseriana) approx.5-15 m high, and Balga trees (Banksia attenuata, Banksia menziesii, Banksia grandis, Allocasuarina fraseriana, Xanthorrhoea pressii).Tree roots are exposed in the cave, both in the roof (fine roots), and cave floor (thick tap roots).In February 2005, the area above the cave was burnt in an intense wildfire (Fig. 1B, Department of Parks and Wildlife, pers. comm., 2015), substantially modifying vegetation above the cave including the death of mature trees and complete removal of canopy and understorey.
Over the period of August 2005 to March 2011, two drip sites in Yonderup Cave (Site 1a and Site 2a), were monitored for their chemical and hydrological variations.These two sites are 22.8 m apart (Site 2a east of Site 1a, ~ 1 m slope towards the East), located at similar depths below the surface (~ 4 m) within the same chamber (~ 7 m height) and partially separated by a large boulder fall-in.We use an existing cave survey to determine the location of each cave drip site relative to the ground surface (Fig. 1C).A soil depth survey was conducted within 5 m of each site (Supp.Table 3), along with visual vegetation/ground surface observations post-fire.Soil depths were measured every meter with a dynamic soil penetrometer in north, south, east and west directions and averaged soil depth above each site were calculated.Site 1a, 30 m from the cave entrance, has a drip source within a large cluster of soda-straw stalactites known as the 'Wheatfield' (Supp.Figure 2B).This circular feature is approximately 1 m across and as it appears in an otherwise very sparsely decorated part of the ceiling, suggesting that it represents a focused flow path into the cave.The land surface above this site is flat with 70% coverage by shallow soil (average 124 mm thickness) and the remaining surface is exposed bedrock (approx.30%).A tuart tree, located directly above Site 1a, burnt and collapsed during the 2005 wildfire which resulted in the entire removal of canopy cover above Site 1a.No other trees are close enough to provide shade on the surface above Site 1a.
In contrast, Site 2a situated 50 m from the cave entrance is in a highly decorated part of the cave known as the 'Cathedral' characterised by large icicle shaped stalactites.Above Site 2a, the soil cover is thicker (200 mm) and more homogenous with no bedrock exposure, and no trees directly above, however there is a partial canopy cover from adjacent trees ~15 m away.

Data collection
Cave dripwater was collected from 1L high-density polyethylene (HDPE) collection vessels at the two sites Site 1a and Site 2a between August 2005 and March 2011 (~5.5 years) at approximately bi-monthly intervals.The water was separated into three aliquots: two aliquots were filtered with 0.45 µm mixed-cellulose filters into two 50 ml polypropylene bottles for major and minor ion determination; the third was stored with zero-headspace in a 12 ml amber glass bottle for stable isotopes.All aliquots were refrigerated below 5°C until analysis.
Anion concentrations (Cl and SO 4 ) were determined using a Dionex DX-600 ion chromatograph with self-regenerating suppressor on one aliquot.The second aliquot was acidified to 2% HNO 3 in the collection bottle and used for cation concentrations (Ca, K, Mg, Na, Si and Sr) using a Thermo Fisher inductively coupled plasma-atomic emission spectrometer (ICP-AES) ICAP7600 at the Australian Nuclear Science and Technology Organisation (ANSTO) facility.An internal standard with concentrations approximating the cave waters was included in each cation batch to check for between-run reproducibility.
Dripwaters collected between August 2005 and May 2008 were analysed for  18 O using Isotope-Ratio Mass Spectrometry (IRMS) at the Australian National University (see Treble et al., 2013 for method).The remaining dripwaters were analysed for  18 O and  2 H at ANSTO using the Cavity Ring Down Spectroscopy (CRDS) method.Additionally, as there was sufficient remaining water in the stored aliquots analysed by IRMS for Site 2a, these were also re-analysed using CRDS to obtain a complete time series for  2 H.After Jan 2007 dripwater volume at Site 1a became insufficient to collect all three aliquots.Collections of aliquots were prioritised in the following order: 1) stable isotopes; 2) cations and; 3) anions.
At each cave visit for dripwater sampling, drip rates were manually recorded using a stopwatch and the level of water accumulated in the bottles was recorded to the nearest 100 ml.Weekly discharge was estimated using a drip volume of 0.2 ml per drip (Collister and Mattey, 2008).When timing drip intervals became impractical, only the bottle level was recorded.Thus in order to represent the data in common units we needed to use the Collister and Mattey, 2008 drip volume in order to convert all our discharge data into volume data.We use both sets of measurements from the overlapping period to convert volume to discharge for when direct measurements for drip interval using the stop watch were lacking.The calculations are provided in the supplementary info as excel sheets.
To distinguish dry and wet periods we applied a residual mass curve (RMC), (Hurst, 1951) to monthly P -AET data.The RMC is the cumulative sum of the monthly anomaly calculated from the 22 year mean and used to generate a time series of cumulative potential water surplus or deficit starting from Jan 2000, highlighting trends in above average or below average P -AET, we refer to this calculation as cumulative water balance (CWB) throughout the rest of this paper.
Unpublished monthly  18 O and  2 H rainfall data (2005 -2011) from Perth were obtained from ANSTO.We used modelled regional precipitation (P) and actual evapotranspiration (AET/F WE is the sum of soil evaporation and transpiration by vegetation based on Priestly-Taylor equations) from the Australian Water Availability Project (AWAP) (Raupach et al., 2009;Raupach et al., 2011) with monthly parameters, to determine P -AET.AWAP precipitation (P), actual evapotranspiration (AET/F WE ) and rainfall  18 O data were then used as input to the forward model (detailed in the next section) to predict cave dripwater  18 O composition under various hydro-climatic scenarios.Predictions are based solely on P -AET data which are then compared to the dripwater observations.Monthly rainfall  18 O and  2 H compositions were amount weighted and fitted with a linear regression (Hughes and Crawford, 2012) and compared to the long-term groundwater mean obtained from Turner and Thorpe (2001) and the cave dripwater to determine whether evaporation has affected cave dripwater isotopic composition (see section 4).
Post-fire solute and  18 O data from Yonderup Cave dripwater are also compared to other relevant published data.These include, long term Perth rainfall  18 O from Turner and Thorpe (2001), and local Yanchep rainfall solute data from Hingston and Gailitis, (1976), and published dripwater data from Golgotha Cave, located 300 km south of Yanchep.Golgotha Cave has been monitored since 2005 (Treble et al. 2013;2015;2016;Mahmud et al 2015).
The climate at Golgotha Cave is also Mediterranean, but receives annual mean rainfall of approx.795 mm, which is 23% higher than Yanchep.Both caves are located within the Tamala Limestone Formation, however the caves vary in depths: Golgotha Cave is significantly deeper than Yonderup approx.30 -35 m.Golgotha Cave is covered by a more extensive forest of mixed marri/karri (Eucalyptus calophylla / Eucalyptus diversicolor) trees and this site has not experienced an intense wildfire since 1992 and no prescribed burns since 2006.However, the prescribed burn at Golgotha Cave was much less intense and it was more controlled than the fire that is reported in our study.

Forward model
We use the forward model employed by Baker et al., (2010).This model uses monthly rainfall  18 O, monthly (P -AET), from 2003 to 2011 (we use 2003 to 2005 data as a 'warm up' period to avoid edge effects), and adjustable bedrock flow thresholds for seepage flow and fracture flow to predict dripwater  18 O based on hydro-climatic influences.Seepage flow and fracture flow thresholds are hydrological P -AET thresholds that are required for infiltrating water to enter seepage or fracture reservoirs (for further details see Baker and Bradley, 2010).The Tamala Limestone, retains high primary porosity thus seepage flow is likely to be dominant whilst fracture flow is less dominant and only likely to be activated during high infiltration (Treble et al., 2013;Mahmud et al., 2015).
Water that enters the seepage reservoir is modelled as a Gaussian distribution.A maximum residence time of 3 years is set; this reflects the dominating seepage or matrix flow type at our site, the shallow depth (4 m) and the potential for capillary barrier effects to impact hydrology in this region (Mahmud et al., 2015).Further a minimum residence time of 10 months is required, to maintain the observed year round discharge at both sites.The model allows for the mean and standard deviation to be specified for these functions.Being conservative we specify the minimum residence time of 10 ± 2 months.In contrast, the fracture-fed flow is instantaneously passed through the system (i.e. with a travel time of less than one month).In the model we can adjust the (P -AET) thresholds required for flow into the seepage reservoir and the threshold required for it to spill into the fracture flow.The seepage and fracture-fed components are mixed in the overlying bedrock reservoir, before predicting dripwater δ 18 O composition.By request the authors can supply the forward model as a spread sheet.
We tested a full range of seepage and fracture possibilities.This suite of model runs helps to place constraints on δ 18 O variability that can be explained by hydro-climatic variability alone.
We compare these scenarios to the observed dripwater δ 18 O at our sites, to assist in our interpretation of the post-fire dripwater  18 O response.
A time series of monthly P -AET, cumulative water balance (CWB), discharge, dripwater  18 O, and ion concentrations for Sites 1a and 2a from August 2005 -March 2011 are shown in Figure 2.

Water balance
Firstly, we observe a distinct seasonality in the water availability (P -AET) (Fig. 2A), where winter months generate an excess (P > AET), while summer months generate a deficit (P < AET).Further, CWB shows three distinct trends throughout the monitoring period: 1) a decline over the period of January 2006 to June 2006, consistent with very low excess in P -AET; 2) an overall rise from June 2006 to February 2010; 3) a decrease in P -AET from February 2010 to September 2010.Site 1a and 2a display moderate and similar discharge rates, at the start of the monitoring period that continue until July 2006; Site 1a an average of 90 ml ± 21 per week and Site 2a an average of 92 ml ± 23 per week.This coincides with infiltration indicated by positive CWB (Fig. 2B).In July 2006, Site 1a dramatically increased discharge five-fold to 468 ml/week on one cave visit, but had decreased to 55 ml ± 3 ml per week on the subsequent visit two weeks later and was completely dry, three months later.This site has not re-activated since (Department Parks and Wildlife, pers.comm.).Site 2a shows much less variation in discharge overall, but contains smooth long-term trends.Two periods of higher discharge are observed in August 2005 to May 2006 (average 92 ml ± 23 ml per week) and April 2008 to February 2009 (average 93 ml ± 29 ml per week), both coinciding with positive trends in CWB.

Water isotopes
Dripwater  18 O from Site 1a (Fig. 2C) shows no seasonal pattern but we see a steady increase of 1‰ to January 2007, then a further steeper rise of 1.5‰ in June 2007, after which the drip ceases.Dripwater  18 O from Site 2a presents an overall increasing trend rising from -3‰ to +0.7‰ over the monitoring period with a 6-month quasi-seasonal signal (approx.2‰ range) that peaks in cooler months (June to October) generally coinciding with months when infiltration from rainfall occurs.We hypothesise that the P < AET environment in drier summer months isotopically enriches soil water, but this only arrives at the cave when seepage thresholds are exceeded in periods of P > AET (winter months).
We forward modelled our rainfall isotopic data in order to predict drip-water  18 O under various hydro-climatic scenarios (Fig. 3).Our sensitivity analyses of hydrological residence times and thresholds showed that seepage residence times, less than 10 months resulted in the seasonal cessation of dripwater, which is not observed at our sites.Therefore, a minimum seepage residence time is required to match our observations.Further, seepage threshold values greater than 40 mm (P -AET) also resulted in the cessation of our drip site.Thus seepage threshold must be below 40 mm (P -AET) to match our observation.Next we varied the fracture threshold between 15 mm and 1000 mm, the wide range reflecting our uncertainty over this parameter.However, we know that seepage flow is dominant at these sites (Mahmud et al., 2015).This suggests two things, first, the seepage threshold is low, second, the threshold required for water to 'overflow' from the seepage reservoir to fracture reservoir must be significantly higher than the seepage threshold.We note that scenarios with a lower fracture threshold (10 -15 mm) show high variability in comparison to sites with a seepage dominated flow and no fracture flow (10 -1000 mm).Based on the variable morphology of stalactites and stalagmites at our sites we interpret discharge to be a combination of seepage and fracture flow, but with seepage clearly dominating.Hence we chose the 15 -100 mm scenario to represent the hydrology at our cave site (Fig. 3).Our forward-modelled dripwater  18 O mean is -4.1‰, slightly less than the mean of Perth rainfall (-3.1‰).The time series of modelled dripwater  18 O (Fig. 2C In all meaningful modelled scenarios i.e. ones that have full year flow and test the full range of hydrological variability, estimated dripwater  18 O cannot replicate the higher observed dripwater  18 O which are +1‰ to +3‰ higher compared to modelled (Fig. 3).This clearly suggests another factor is affecting dripwater  18 O composition: likely near-surface evaporation.
To investigate an evaporation effect, we plot cave dripwater along the local meteoric water line (LMWL, weighted LSR) to test for isotopic enrichment (Fig. 4). Figure 4 shows that while the least squares regression (LSR) for cave dripwater falls within the standard error (± 0.45‰) of the slope for the local meteoric water line (LMWL, weighted LSR), drip water isotopic composition is concentrated towards heavier δ 18 O and δ 2 H.These results are consistent with evaporation in a high humidity environment as has been observed in semiarid cave environments elsewhere (e.g.Cuthbert et al., 2014).Adopting Cuthbert's classification, our data falls under a type 1 scenario reported in Cuthbert et al., (2014).In the type 1 scenario, δ 18 O and δ 2 H do not deviate from the LMWL but are shifted along the LMWL towards higher values, as is the case with our data (Fig. 4).This means that our data are similarly impacted by evaporation occurring in a high humidity environment.

Water Solutes
There are significant differences in solute concentrations and trends between the two sites (Fig. 2E Trends in SO 4 and K are more subtle than for other solutes: at Site 1a, K shows a slight decline from the beginning of the monitoring until early 2007 and then has a small rise prior to drips ceasing.Although harder to judge in the shorter SO 4 time series, SO 4 also shows a small rise before drips cease, similar to K. Trends in K and SO 4 for Site 2a are more subtle, although they both increase slightly over time.K and SO 4 concentrations are, two to three times higher at Site 1a versus Site 2a and are considerably higher than those recorded at Golgotha Cave (Table 1).We also note that initial Cl and other solute concentrations at Site 1a are twice that at Site 2a.

Post-fire hydrology
Discharge at Site 1a is inconsistent with CWB: discharge rose as rainfall fell below the longterm mean (P < AET) (Fig. 2A and 2B), suggesting that Site 1a received a localised increase in discharge despite the declining input from rainfall.In contrast, discharge at Site 2a is more closely related to the cumulative water balance (CWB), with higher discharge coinciding with periods of higher water surplus and lower discharge with lower water surplus.
Chloride is a chemically conservative and highly soluble solute (Graedel and Keene, 1996), and its concentrations in dripwater will therefore reflect concentration/dilution effects (Tooth and Fairchild, 2003;Tremaine and Froelich, 2013).Chloride concentrations at Site 2a increase during the period of declining CWB (2006CWB ( to mid-2008) ) suggesting that evaporation is concentrating Cl.Rising δ 18 O and other solutes over this period are also consistent with increased evaporation.From mid-2008 onwards, when CWB is positive (P > AET), Cl decreases, consistent with an increase in infiltration and thus dilution (Fig. 2E).
At Site 1a higher discharge also coincides with falling Cl concentrations also suggesting dilution (Fig. 2B, 2D).However, we note this coincides with a highly negative CWB i.e. drier than normal conditions.This suggests in this case, a non-climatic driver has influenced infiltration.We propose that a reduction in localised transpiration, following the 2005 fire, may be driving this.Deeply-rooted trees within the area have been reported to produce high Cl concentrations in the unsaturated zone (Turner et al., 1987).Site 1a had a tuart tree directly above it and tree roots are visible above Site 1a in the cave, but not at Site 2a.The proximity of the tree to Site 1a is the most likely explanation for the higher solute concentrations here (Treble et al., 2016).The death of the tree in the 2005 fires would remove the previous transpiration demand and hence result in effective dilution of the solutes during infiltration, as observed.However, this reduction in transpiration would have been abrupt but we observe a response lasting 1.5 years after the fire.This could be due to a number of reasons; firstly, the minimum residence time is 10 months (for a year of continuous discharge) so a delay in the response is to be expected.Second, this occurred during a period in which the soil moisture deficit would have been larger than average, so a larger volume of cumulative infiltration would be needed to overcome this deficit and move the more dilute solute into the cave.
It is also possible the decrease in concentrations reflect the diminishing element concentrations after an immediate flush of the more soluble ash-derived material (i.e. the tail of a solute pulse).However, post-fire, highly soluble solutes like Cl, will still reflect dilution due to increased discharge.So, it is likely that we are seeing a decline in these elements due to a combination of the removal of these nutrients from the surface and subsurface, and dilution.
In the broader context we look at the differences in Cl at Yonderup Cave vs Golgotha Cave.
Both caves are ~5 km from the coastline, so they likely have a similar amount of Cl aerosol deposition.Yet, post-fire Cl values at Yonderup Cave, are more than double the Cl concentrations from Golgotha Cave (Table 1).Further, within cave variability in Cl concentrations at Yonderup Cave are twice that at Golgotha Cave (Table 2).While variations in vegetation density (along with wildfire history) may have some role to play in the difference in mean dripwater Cl at each location, the higher within cave variability at Yonderup Cave (Table 2), suggests a post-fire setting increases variability in dripwater chemistry.The impact of this is discussed further and a conceptual model (Fig. 6) devised later in section 6.3.

Post-fire carbonate chemistry
Similar to Cl, concentrations of carbonate metals (Mg, Ca and Sr) at Site 1a also decrease; this reflects solutes being diluted due to reduced tree water-use.However, we note that at Site 1a, Ca, for example, declines twice as much (in concentration, ~75 %) in comparison to Cl (~30%).Thus for Ca, another mechanism along with dilution is required to explain its nonlinear step-like decline (Fig. 2G).
There are a number of mechanisms that could influence post-fire Ca concentrations.First, we consider increased near-surface evaporation inducing prior calcite precipitation (PCP).
Increased evaporation, can saturate solutes relative to calcite in karstic waters and promote degassing.Further, evaporation will slow the flow increase water-rock interaction times in the remaining water.Both of these conditions are ideal for PCP.Both our sites show evidence of PCP: the ln(Sr/Ca) vs ln(Mg/Ca) slopes in our data agree with the diagnostic range for PCP (a slope of + or -0.88; Fig. 5B) (Sinclair et al., 2011).Expressed as a time series (Fig. 5A), we see ln(Sr/Ca) and ln(Mg/Ca) increase simultaneously with δ 18 O and Cl (Fig. 2), suggesting that evaporation is indeed the common driving mechanism and is inducing PCP at Site 1a.For further information on PCP processes we recommend the reader to Fairchild et al., 2000;Sinclair 2011 andTreble et al., 2015.A second mechanism influencing post-fire Ca concentrations may be the addition of plant ash (Yusiharni and Gilkes, 2012a) and highly soluble CaO (produced by the burning of exposed surface rock to the fire; Yusiharni and Gilkes, 2012b).Further, it is possible that the Ca decline may also reflect a decrease in Ca being leached post-fire that may have followed an earlier spike in Ca concentrations from the above.The extent to which each process is affecting the Ca concentration is difficult to assess in our data, especially since our monitoring did not commence until 6 months after the fire, and processes such as PCP and the addition of Ca from plant-ash are difficult to constrain.
We now consider other carbonate metals: Mg and Sr, at Site 1a to further constrain our interpretation.We find Mg and Sr decline by ~30% (in terms of relative concentration).Thus it is likely that the same process affecting Cl is also affecting Mg and Sr, that is, dilution, a decline in leaching of biomass-sourced ash, or some combination of both.Ca concentrations decline by a relatively larger amount (75%) suggesting that additional processes are specifically affecting Ca.The rise in Mg/Ca at Site 1a strongly suggests that the remaining portion of the Ca decline may be attributed to PCP (Fig. 5A).
At Site 2a, a rising trend, reflects the concentration of solutes due to a rise in post-fire evapotranspiration, evidenced by increasing δ 18 O and Cl, evaporation of near-surface water stores (Fig. 2F, 2G and 2H respectively), and possibly to some extent, an increase in transpiration from vegetation recovery for Cl (Treble et al., 2016).Additionally, Ca concentrations also show a quasi-seasonal response, interpreted from the Mg/Ca time series to be driven by PCP, possibly due to seasonal P -AET (supported by similar seasonality in dripwater δ 18 O; Fig. 2C) although in-cave PCP could also be contributing (Treble et al., 2015).
Sulphate and K post-fire at Site 1a are abnormally high in concentration, approximately three times higher in comparison to Site 2a (Figs.2H, 2I; Tables 1, Table 2).This is counterintuitive to the initial dilution signal (owing to a decrease in tree water-use) interpreted for the other solutes.While at Site 2a, SO 4 and K increase similar to other solutes, consistent with evaporative concentration (from post-fire conditions) and an increase in transpiration from (vegetation recovery).These observations suggest that there was an increase in the availability of SO 4 and K after the fire at Site 1a despite a decrease in treewater use here (Fig. 2H and 2I).
We note the majority of aboveground SO 4 is predominantly stored within the lower to middle storey of the forest (O'Connell and Grove, 1996), and post-fire soils contain 23% more S and 16% more K than pre-fire soils due to biomass-sourced ash deposition (Grove et al., 1986).
So SO 4 and K concentrations at each site may respond differently since the amount of available SO 4 and K above each site is influenced by the amount of biomass burnt above the site.Further, the dissolution rates of ash minerals containing these elements could also affect the rate at which these nutrients are leached from the surface and subsurface and subsequently their concentrations in dripwaters.We propose that the large amount of biomass burnt above Site 1athe tuart tree -is responsible for the much higher concentrations of SO 4 and K at Site 1a dripwater relative to Site 2a.We also propose that the increase in nearsurface evaporation from 2007 onwards drives even higher concentrations of SO 4 and K (Fig 2H and 2I).Site 2a, which has much less pre-fire biomass, has much lower SO 4 and K concentrations, consistent with our argument.Here, these solutes show a steady increasing trend over the monitoring period.This is consistent with increased evapotranspiration postfire, which is also evident in the other solutes.Now, we compare our Yonderup Cave results to those of Golgotha Cave to put SO 4 and K concentrations into context.Golgotha Cave, last experienced a wildfire in 1992 and a controlled low-temperature prescribed burn in 2006, while Yonderup experienced a high intensity burn in 2005.Firstly, we see five and ten times higher within cave differences in SO 4 and K (respectively) at Yonderup Cave than we do at Golgotha Cave (Table 2).Also, concentrations at Yonderup Cave are up to four and three times higher in SO 4 and K (respectively) at Yonderup Cave than at Golgotha Cave (Table 1).These results coupled with the difference between Cl compositions at Yonderup vs. Golgotha presents a clear case for more variability in burnt sites in comparison to unburnt sites.

A multi-proxy fire signal in dripwater
Here we propose post-fire scenarios for both sites at Yonderup Cave (refer to conceptual model Fig. 6) that account for the altered dripwater chemistry that is observed post-wildfire.
A straight-forward relationship between cumulative water balance (CWB), discharge and Cl concentrations at Site 2a, suggests increased concentration of solutes (Ca, Mg, Sr, K and SO 4 ) in response to an increase in near-surface evaporation (Fig. 2C and 2D).
In contrast solutes such as Cl, Mg, Sr, Ca at Site 1a show a declining trend.We have argued that this declining trend in these solutes is due to two underlying processes.First, a decrease in tree water-use (transpiration) due to the death of the tuart tree in the wildfire.Tuart trees are deeply-rooted, having adapted their root systems to access water from both the surface and subsurface.These roots have been found to generate potential energy between the tree and the soil to extract water and nutrients.Specifically, mature tuart trees generate a pressure gradient ranging from -0.86 ± 0.11 MPa (summer) to -0.35 ± 0.02 MPa (winter) (Drake et al., 2011).So the death of the tuart tree in the wildfire had a significant local effect on hydrology at Site 1a resulting in an increase in discharge and dilution of solutes (Cl, Mg, Sr and Ca).
Our study suggests that the consequent reduction in transpiration may not be immediately detected in dripwater, owing to transit time through the limestone and the requirement for overcoming a soil moisture deficit during drier than average climatic conditions.Hydrological effects such as 'capillary barriers' can also slow down vertical transport of infiltrating waters at our site (Mahmud et al., 2015).
Second, the decline in solutes at Site 1a could be the result of a gradual return to pre-fire concentrations following a pulse of increased input of solutes from ash (e.g.Site 1a scenario).
A contribution of SO 4 and K from burnt biomass may explain their relatively high concentrations.
Further, isotopic composition at Site 1a show that δ 18 O is offset from modelled hydroclimatic δ 18 O by ~+1‰ suggesting increased near-surface evaporation post-fire which we attribute to the reduction of shading from the tuart tree post-fire -even in P > AET periods (Fig. 2C).And post-2007, when P < AET conditions arrive, δ 18 O rises even higher and discharge declines.Eventually, due to the persistent duration of P < AET conditions Site 1a ceased dripping owing to eventual depletion of the near-surface reservoir feeding this drip.
Site 2a also shows higher δ 18 O, during the post-fire period, which we attribute to an increase in near-surface evaporation as a result of low albedo and reduced vegetation cover.The case for evaporation at this site is supported by the rise in Cl and other solutes.The possibility of an immediate spike in solutes from ash and a long-term decline from leaching, as was discussed for Site 1a, is limited here, as there was less biomass available to burn above Site 2a.Therefore, we interpret increased evaporation and transpiration from regrowth post-fire to be the dominant forcing at this site, similar to the findings of Treble et al. (2016).
We propose that differences in surface vegetation above sites can influence site specific drip chemistry.For example, we interpret Site 1a was influenced by a reduction in transpiration after the fire, due to the forcing biomass above the site; which may also have been a source of post-fire ash at this site.At Site 2a the response to the fire was primarily an increase in nearsurface evaporation owing to changes in surface albedo.This variability within the cave response at Yonderup Cave is significant, so too is the comparison between Yonderup Cave drip chemistry and the drip chemistry at Golgotha Cave.The latter more generally highlights that an intense wildfire has variable, but multi-year effects on dripwater composition in shallow caves.
From this we propose that post-fire condition persist up-to 5 -10 years' post-fire, affecting dripwater  18 O and solute concentrations.We would expect a full recovery, of  18 O and solute concentrations back to pre-fire levels within 10 -20 years as a result of revegetation growth (Treble et al., 2016) and re-establishment of vegetation cover and pre-fire albedo.
7 Application for a speleothem paleo-fire signal Our post-wildfire dripwater response from  18 O was a 2‰ increase above that predicted by a hydro-climatic model, and measured regional groundwater and Golgotha Cave  18 O data.If this signal is preserved at equilibrium in speleothems this is equivalent to some of the largest interpreted climatic changes seen in the Quaternary record.This highlights the significance of the findings in our study, which suggests a fire signal could in fact be misinterpreted as climate variability.Furthermore, the impact of the decrease in Ca dissolution from the limestone bedrock could have a significant effect on speleothem growth rate.
However, before attributing  18 O and growth rate abnormalities to fire, we must remember there are a number of processes that effect speleothem  18 O.Thus it is important that a multiproxy approach, which uses isotopic composition as well as a suite of trace elements (sourced from both soil and bedrock), is used to separate fire from other forcings such as climate and other local factors.Further, our study was conducted in a shallow cave environment, where perhaps the overlying vegetation can exert a more dominant forcing on dripwater hydrology and chemistry relative to deeper caves.Deeper caves have more complex hydrology (McDonald and Drysdale, 2007); this involves mixing with other flow paths, which are possibly not fire affected.This may result in the smoothing of the fire signal; making it harder to isolate.Further, a fire signal in cave dripwater and stalagmites may be much subtler in grassland environment in comparison to a forested environment as changes in the biomass would be smaller and vegetation recovery presumable faster (Coleborn et al., 2016).We recommend searching for fire signals in shallow cave environments in the tree rooting zone in forested areas.
One further approach that may help to differentiate fire and climate signals in a stalagmite would be to use multivariate statistical techniques such as principal component analyses (PCA).Using this technique, we would expect one component to reflect a bedrock/hydroclimatic signal and another to preserve a local soil/vegetative forcing.The soil/vegetative component could preserve the impact of a fire on stalagmite composition, in trace elements like S, and K. Further, it is also possible an immediate spike in solutes (Mg, Ca, Cl, Sr, S, K and P) from post-fire ash may be preserved in stalagmites and colloid associated metals, such as Al, Fe and Cu from an increase in discharge post-fire.Future studies of this kind will open a new avenue in speleothem research; speleothems as archives of paleo-fire.
We isolate a post-wildfire response by comparing a recently burnt cave monitoring site with forward modelled δ 18 O, which predicts δ 18 O based on hydro-climatic factors, and nearby cave monitoring and groundwater data.We provide a novel analysis of the multi-year impacts wildfire has on cave dripwater.Our analysis shows a strong hydrologic relationship between surface environments and shallow caves that are located within the tree rooting zone.This finding is especially important in water-limited environments (P < ET) as the overlying vegetation can exert controls on the cave hydrogeochemical environment.
A post-wildfire dripwater response is clearest in  18 O and Cl due to their sensitivity to variation in near-surface evaporation (both  18 O and Cl) and transpiration (Cl).Cl is a conservative ion and hence is driven mainly by by dilution/evaporation. Cl declines post-fire at Site 1a which we interpret as dilution of the water store that the dead tree previously exploited.In contrast, Cl increases at Site 2a, sympathetically with δ 18 O, consistent with an increased evaporative demand on shallow water stores driven by post-fire reduction in shading and reduced albedo.SO 4 and K are also important as at sites with abundant biomass they can be leached at high concentrations as they are made more abundant in post-fire soils due to the ash generated from a fire.Other solutes such as Mg, Sr and Ca support the dominant local forcing at the site post-fire and can be extremely powerful when using a multi-proxy approach.
We propose a conceptual model for a multi-year post-wildfire cave dripwater response in forested water-limited regions.This involves a 5 -10 yr response of: 1) higher  18 O and Cl in cave dripwater due to increased evaporation and decreased shading after the wildfire; 2) increased K and SO 4 due to the leaching of biomass-sourced ash, particularly in areas with large biomass; and 3) increased variability in Mg, Sr and Ca due to changes in evaporation, transpiration and water-rock interactions post-fire.We may expect a recovery within 10-20 years after the wildfire and a restore to pre-fire isotopic and trace element concentrations as a result of increased bio-productivity from forest regrowth and a re-establishment of canopy cover.
caves.We see that Cl, SO 4 , K and δ 18 O values, at both sites are distinctly different.Specifically, the solutes have higher concentrations and δ 18 O 703 is higher at Yonderup Cave in comparison to Golgotha Cave.

A B
Figure 6) Evaporation (red) increases post-fire at both sites due to a reduction in albedo and vegetation cover while precipitation (blue) remains the same and initial transpiration (green) decreases, but recovers over time.Site 2a shows higher δ 18 O and an increase in concentrations of solutes including SO 4 and K (lime) due to evaporation and slow increase in transpiration due to vegetation recovery, with cumulative water balance (CWB) remaining the same.While Site 1a shows, higher δ 18 O in response to increased evaporation and a decline in solute concentrations in response to increased discharge and a decrease in transpiration and removal of nutrients from the surface and subsurface.However, since SO 4 and K are from biomass-sourced ash, the tuart tree above this site acts as a source of increased SO 4 and K. Discharge increased immediately (blue).But the drip became inactive one year after the fire due to an increase in evaporation, which outweighed the reduction in transpiration (green), leading to depletion of the near-surface reservoir feeding Site 1a and an in active drip site.
) starts and remains at ~ -4.2‰ until February 2006 where it dips slightly before rising sharply to -3‰ where it remains steady until February 2007.Here it begins a step-wise decline; declining from February to March 2007 by 0.5‰ and remaining stable again until February 2008.It then shows a further step-decline in March 2008 to -4.5‰, where it remains at approximately this value, albeitwith a few small variations on timescales of months, until the end of the monitoring period.
, 2F, 2G and 2H).Solute concentrations are typically higher at Site 1a versus Site 2a and they demonstrate opposite trends post-fire.At Site 1a, Cl, Ca, Mg and Sr decline overall, although this trend is step-wise for Ca, and reverses for Cl ~ 6 months before the drip ceases.The trends in these solutes at Site 1a are inconsistent with the declining CWB during this period (Fig 2A), as we would expect a drying trend reflected through the evaporative concentration of solutes.In contrast at Site 2a, Cl and other solute concentrations show a direct relationship to CWB (i.e.increasing solute concentration with decreasing CWB from 2006 until mid-2008 followed by decreasing solute concentrations with increasing CWB.

Figure 1 .Figure 2 .
Figure 1.Geographical location of our study site (A), a post fire photograph of the area (B)

Figure 3 .
Figure 3. Modelled dripwater  18 O outputs under varying thresholds in our forward model

Figure 5 .
Figure 5. (A) A time series of ln(Sr/Ca) vs. ln(Mg/Ca).Site 1a shows a clear enrichment in