Using multiple methods to investigate the effects of land-use changes on groundwater recharge in a semi-arid area

Understanding the applicability and uncertainties of methods for documenting recharge rates in semi-arid areas is important for assessing the successive effects of landuse changes and understanding groundwater systems. This study focuses on estimating groundwater recharge rates and understanding the impacts of land-use changes on recharge rates in a semi-arid area in southeast Australia. Two adjacent catchments were cleared ∼ 180 years ago following European settlement, and a eucalypt plantation forest was subsequently established ∼ 15 years ago in one of the catchments. Chloride mass balance analysis yields recharge rates of 0.2 to 61.6 mm yr−1 (typically up to 11.2 mm yr−1). The lower of these values probably represents recharge rates prior to land clearing, whereas the higher likely reflects recharge rates following the initial land clearing. The low pre-land-clearing recharge rates are consistent with the presence of old groundwater (residence times up to 24 700 years) and the moderateto-low hydraulic conductivities (0.31 to 0.002 m d−1) of the aquifers. Recharge rates estimated from tritium activities and water table fluctuations reflect those following the initial land clearing. Recharge rates estimated using water table fluctuations (15 to 500 mm yr−1) are significantly higher than those estimated using tritium renewal rates (0.01 to 89 mm yr−1; typically < 14.0 mm yr−1) and approach the long-term average annual rainfall (∼ 640 mm yr−1). These recharge rates are unrealistic given the estimated evapotranspiration rates of 500 to 600 mm yr−1 and the preservation of old groundwater in the catchments. It is likely that uncertainties in the specific yield results in the water table fluctuation method significantly overestimating recharge rates, and despite the land-use changes, the present-day recharge rates are relatively modest. These results are ultimately important for assessing the impacts of land-use changes and management of groundwater resources in semi-arid regions in Australia and elsewhere.


Introduction
Groundwater is a critical resource for meeting the expanding urban, industrial and agricultural 25 water requirements, especially in semi-arid areas that lack abundant surface water resources (de Vries and Simmers, 2002;Siebert et al., 2010). Groundwater also makes a significant contribution to the streamflow of rivers in semi-arid areas. Land-use changes may modify groundwater recharge rates, which thus affect groundwater systems as well as groundwater resources (Foley et al., 2005;Lerner and Harris, 2009;Owuor et al., 2016). In many semi-arid 30 regions, there has been the conversion of native forests to agricultural land (Foley et al., 2005).
Deep-rooted trees generally return more water to the atmosphere via transpiration than shallowrooted crops and grasses (Hewlett and Hibbert, 1967;Bosch and Hewlett, 1982;Fohrer et al., 2001). In southeast Australia, the reduction in evapotranspiration following the land clearing has commonly resulted in a net increase in recharge and a rise of the regional water tables. In 35 turn, this has resulted in waterlogging and salinization of cleared lands and increased stream salinity (Allison et al., 1990). Eucalyptus tree plantations were subsequently initiated partially to reduce groundwater recharge and thus prevent the rise of regional water tables (Gee et al., 1992;Benyon et al., 2006). In order to assess the impacts of successive land-use changes on the groundwater and surface water systems, estimates of recharge are required. Estimation of 40 recharge rates is also important for groundwater modelling, because recharge represents the water flux used as a boundary condition at the water table.
Recharge is the water that infiltrates through the unsaturated zone to the water table and thus increases the volume of water stored in the saturated zone (Lerner et al., 1990;Healy and Cook, 2002;Scanlon et al., 2002;Moeck et al., 2020). A distinction between gross and net recharge 45 may also be made (Crosbie et al., 2005). The total amount of water that reaches the water table is the gross recharge, while the net recharge accounts for the subsequent removal of water from the saturated zone by evapotranspiration. In areas with shallow water tables and deep-rooted vegetation, this subsequent water loss can be considerable. Estimating groundwater recharge rates, in general, is not straightforward (Lerner et al., 1990;Healy, 2010;Moeck et al., 2020) 50 and recharge rates potentially vary in space and time (Sibanda et al., 2009).
Several techniques may be used to estimate groundwater recharge, including Darcy's Law, measuring water infiltration using lysimeters installed in the unsaturated zone, measuring and modelling soil moisture contents, use of heat flow calculations, catchment water budgets, remote sensing, numerical models, water table fluctuations, chemical (chloride) mass balance 55 calculations, and/or the concentrations of radioisotopes such as 3 H (tritium), 14 C (carbon), 36 Cl (chloride) or other time-sensitive tracers (e.g., chlorofluorocarbon) in groundwater (Scanlon et al., 2002(Scanlon et al., , 2006Healy, 2010;Doble and Crosbie, 2017;Cartwright et al., 2017;Moeck et al., 2020;Gelsinari et al., 2020). Different techniques estimate recharge over different spatial-temporal scales, and they may 60 thus yield different results (Scanlon et al., 2002). Because each technique has different uncertainties and limitations, it is recommended that multiple methods are used to constrain recharge Sophocleous, 2004;Scanlon et al., 2006). Understanding the broader hydrogeology also helps to understand recharge. For example, areas where recharge rates are high, should contain high proportions of young groundwater. Additionally, recharge 65 rates are likely to be low if evapotranspiration rates approach rainfall totals.
This study estimates recharge rates using Cl mass balance, water table fluctuations, and 3 H renewal rate methods in a semi-arid area that has undergone successive land-use changes. We evaluate the applicability and uncertainties of these commonly applied methods to determine the changes in recharge rates caused by these successive land-use changes. While based on a 70 specific area, the results of this study, in particular the comparison of present-day recharge rate estimates, will be applicable to similar semi-arid areas in southeast Australia and elsewhere.
Specifically, predicting the impacts of changes to land-use on recharge rates is required to understand and manage waterlogging and salinization of soils and streams. A brief description of the assumptions and limitations of these techniques is provided below. 75

Cl mass balance
The Cl mass balance (CMB) approach yields average regional net recharge rates (Bazuhair and Wood, 1996;Scanlon, 2000;Scanlon et al., 2002). The assumptions of this method are that all Cl in groundwater originates from rainfall and that Cl exported in surface runoff is negligible or well known. Under these conditions, the net groundwater recharge (R net in mm yr -1 ) is 80 estimated from: (1) (Eriksson and Khunakasem, 1969) where P is mean annual precipitation (mm yr -1 ), Cl p is the weighted mean Cl concentration in precipitation (mg L -1 ), and Cl gw is Cl concentration in groundwater (mg L -1 ). The CMB method estimates net recharge rates averaged over the time 85 that the Cl contained within the groundwater is delivered; this may be several years to millennia.
Uncertainties in the CMB method are mainly the long-term rate of Cl delivery and the assumptions that runoff has remained negligible over time.

Water table fluctuations
Water table fluctuations may be used to estimate gross recharge rates over the time period for 90 which groundwater elevation data are available. Because bore hydrograph data are abundant, this probably is the most common method of estimating present-day recharge rates. The water table fluctuation (WTF) method strictly requires the water table to be located within the screened interval of the bore; however, it can be used in bores screened within a few metres of the water table . The method assumes that: evapotranspiration from the 95 water table has not occurred; the rise in the water table is solely due to recharge following rainfall events; groundwater elevations are not influenced by pumping; and the water table falls in the absence of recharge. R gross is calculated from where S y is the specific yield (dimensionless) of the aquifer, and Δh/Δt is the variation in the 100 hydraulic head over the recharge event (mm yr -1 where there is an annual recharge event).
Despite its simplicity, there are several potential uncertainties in the WTF method. S y is not commonly measured, and most studies rely on typical values based on aquifer materials. More importantly, the retention of moisture in the unsaturated zone between recharge events reduces S y and results in S y being spatially and temporally variable (Gillham, 1984;Sophocleous, 1985;105 Healy and Cook, 2002;Crosbie et al., 2019). However, many recharge studies assume that S y is constant and close to the effective porosity. This may result in the WTF method significantly overestimating recharge rates (Gillham, 1984;Sophocleous, 1985;Crosbie et al., 2019). Other processes may also affect head measurements. These include entrapment of air during rapid recharge events (the Lisse effect) and the impacts of barometric pressure changes and ocean or 110 Earth tides, especially when the head is measured using sealed pressure transducers (Crosbie et al., 2005). The estimation of the recession curve of the groundwater hydrograph used to calculate h in Eq. (2) also involves some judgement.

3 H renewal rate
The 3 H renewal rate (TRR) method envisages that recharge mixes with pre-existing 115 groundwater in a discrete zone at the top of the aquifer with an equivalent amount of water from this upper zone displaced lower into the groundwater system. The renewal rate (R n ) represents the proportion of new water added in each recharge cycle. If there is an annual cycle of groundwater recharge, the 3 H activity of groundwater in year i ( 3 H gw i ) is related to R n by (3) 120 (Leduc et al., 2000;Le Gal La Salle et al., 2001;Favreau et al., 2002) where λ t is the radioactive decay constant for 3 H (0.0563 yr -1 ), and 3 H p i is the average 3 H activity of rainfall in year i (in Deleted: with an equivalent amount displaced lower into the groundwater system. Tritium Units, TU where 1 TU corresponds to 3 H/ 1 H = 1×10 −18 ). The application of the TRR 125 method requires the 3 H input function over the past few decades to be known. The 3 H activities of southern hemisphere groundwater recharged during the 1950s and 1960s atmospheric tests were several orders of magnitude lower than northern hemisphere groundwater (Morgenstern et al., 2010;Tadros et al., 2014). These 3 H activities have now decayed and are lower than those of present-day rainfall, which results in individual 3 H activities yielding a single R n 130 estimate (Cartwright et al., 2007(Cartwright et al., , 2017(Cartwright et al., , 2020; this is not yet the case in the northern hemisphere (Le Gal La Salle et al., 2001).
Groundwater recharge rates are related to R n by

Rnet = R n bn
(4) where b is the thickness of the upper part of the aquifer system that receives annual recharge 135 and n is the effective porosity. Uncertainties in the TRR estimates include uncertainties in the 3 H input function and having to estimate b and n, which may be variable and not well defined.
The recharge rates are net estimates averaged over the residence time of groundwater in the upper part of the aquifer, which in an ideal system is R n -1 .

140
Gatum is situated in western Victoria, southeast Australia (Fig. 1a). The native eucalyptus forests in this region were originally cleared for grazing following European settlement ~180 years ago (Lewis, 1985) and then partially replaced by eucalyptus plantation in the last ~15 years (Adelana et al., 2015). Gatum lies in the regional recharge area of the Glenelg River Basin to the south of the drainage divide between the Glenelg and Wannon Rivers, and surface 145 water drains to the Wannon River via the Dundas River (Dresel et al., 2012). The area is predominantly composed of fine-to coarse-grained weathered Early Devonian ignimbrites containing abundant large locally derived clasts near their base (Cayley and Taylor, 1997).
The study area consists of two catchments with contrasting land-use, one catchment is predominately dryland pasture used for sheep grazing, and the other is mostly occupied by plantation Eucalyptus globulus forestry. The pasture catchment is around 151 ha and is typical of the cleared land in this region. It is covered by perennial grasses with about 3% remnant 155 eucalyptus trees. The forest catchment is around 338 ha and comprises approximately 62% plantation forest, established in 2005, and 38% grassland (Adelana et al., 2015). The elevations of the pasture and forest catchments range from 236 to 261 m and 237 to 265 m AHD (Australian Height Datum), respectively (Fig. 2). The two catchments were subdivided into the upper slope, mid-slope and lower slope, based on the elevation of the study area; the drainage 160 zones are in the riparian zones of the small streams (Dresel et al., 2018). The catchments are drained by two small intermittent streams (Banool and McGill: Fig. 1a) that export ~8% of annual rainfall (Adelana et al., 2015;Dresel et al., 2018).
The regional groundwater is not extensively used in this area. However, the study area is one of many in southeast Australia that was identified as being impacted by dryland salinity due to 165 land clearing and rising water tables (Clark and Harvey, 2008). During the Millennium Drought in the first decade of the century, the water tables dropped considerably and the emphasis on dryland salinity diminished. The focus of water management in this area switched from salinity to water sustainability and the effect of land-use changes on the water balance of this area (Dresel et al., 2012). In addition to the regional groundwater system, shallow (1 to 4 m deep) 170 perched groundwater exists in the riparian zones (Brouwer and Fitzpatrick, 2002;Adelana et al., 2015).
The climate is characterized by cool, wet winters and hot, dry summers. From 1884 to 2018, the average annual rainfall at Cavendish (Station 089009) ~19 km southeast of Gatum ( Fig. 1a) was ~640 mm (Bureau of Meteorology, 2020), with most rainfall in the austral winter between 175 May and October (Fig. 3a). Average annual actual evapotranspiration across the two catchments between 2011 and 2016 was estimated at about 580 mm (Dresel et al., 2018). The mean concentrations of Cl in rainfall range from 2.2 mg L -1 at Cavendish (Hutton and Leslie, 1958) to 4.4 mg L -1 at Hamilton (~34 km southeast of Gatum, Fig. 1a: Bormann, 2004;Dean et al., 2014). Similar Cl concentrations were recorded in rainfall across much of southeast 180 Australia (Blackburn and McLeod, 1983;Crosbie et al., 2012).

Water sampling
There are 19 monitoring bores at different landscape positions sampling the regional groundwater in the pasture and forest catchments ( Fig. 1a) with sample depths ranging from 185 1.3 to 29.7 m (Supplementary Table S1). Hydraulic heads have been measured since 2010 at four hourly intervals using In Situ Aquatroll or Campbell CS450 WL pressure loggers corrected for barometric pressure variations using In Situ Barotroll loggers. Occasional spikes (generally resulting from the logger being removed from the bores) were removed. Twelve shallow piezometers (~1 m deep with ~10 cm wide screens at their base) were installed in 2018 near 190 the monitoring bores in the drainage zones and the lower slopes of the pasture and forest catchments (Fig. 1a). These piezometers intercept the riparian groundwater that in places is perched above the regional groundwater. Regional groundwater was sampled from bores (n = 24) and riparian groundwater from shallow piezometers (n = 24) between May and November 2018. The groundwater samples were collected from the screened interval using a submersible 195 pump or bailer following the removal of at least three bore volumes of groundwater or removing all water and allowing it to recover. Following sampling, hydraulic conductivities (K s : m day -1 ) were determined from the rate of recovery of the groundwater levels measured at 3-minute intervals using an In Situ Aquatroll pressure logger (Hvorslev, 1951). A one-year aggregated rainwater sample was collected in a narrow-mouthed container with an open funnel. 200 The sample was periodically removed from the container and aggregated into a single sample.

Analytical techniques
Geochemical data are presented in Table S1. Electrical conductivity (EC) was measured in the field using a calibrated hand-held TPS WP-81 multimeter and probe. Groundwater samples were collected in high-density polyethylene bottles and stored at ~4°C prior to analysis. 205 Alkalinity (HCO 3 ) concentrations were measured within 12 hours of sampling by titration.
Major ion concentrations were measured at Monash University. Cation concentrations were determined on filtered (0.45 µm cellulose nitrate filters) water samples that were acidified to pH <2 with double distilled 16 N HNO 3 using ICP-OES (Thermo Scientific iCAP 7000).
Concentrations of anion were determined on unacidified filtered water samples by ion 210 chromatography (Thermo Scientific Dionex ICS-1100). Based on replicate analyses, the precision of cation and anion concentrations are ±2%; from the analysis of certified standards, accuracy is estimated at ±5%. Total dissolved solids (TDS) concentrations are the sum of the cation and anion concentrations. 3 H and 14 C activities were measured at the Institute of Geological and Nuclear Sciences (GNS) 215 in New Zealand. Samples for 3 H activities were measured by liquid scintillation in Quantulus ultra-low-level counters following vacuum distillation and electrolytic enrichment as described by Morgenstern and Taylor (2009). The quantification limits are 0.02 TU and the relative uncertainties are typically ±2% (Table S1). 14 C activities were measured by AMS following Stewart et al. (2004). Dissolved inorganic carbon (DIC) was converted to CO 2 by acidification 220 with H 3 PO 4 in a closed evacuated environment. The CO 2 was purified cryogenically and converted to graphite. 14 C activities are normalised using the δ 13 C values and expressed as percent modern carbon (pMC), where the 14 C activity of modern carbon is 95% of 14 C activity of the NBS oxalic acid standard in 1950. Uncertainties are between 0.27 and 0.35 pMC (Table   S1). 225

Recharge calculations
Recharge rates were estimated using the methods discussed in sections 1.1-1.3. Net recharge rate estimates from the CMB (Eq. 1) utilised present-day average rainfall amounts (~640 mm) and Cl concentrations of 2.2 to 4.4 mg L -1 together with the measured Cl concentrations of groundwater (Table S1). Gross recharge rates were estimated using the WTF method (Eq. 2) 230 from the bore hydrographs that display seasonal variations in the water levels (Figs. 3b, 3c).
There is a single pronounced annual increase in the hydraulic head following winter rainfall, and Δh was estimated as the difference between the highest head value and the extrapolated antecedent recession curve . The effect of evapotranspiration on the magnitude of the hydraulic heads is assumed to be low, especially during winter when radiation 235 and temperature are lower. S y was assumed to be close to n (0.03 to 0.1: Adelana et al. 2015;Dean et al., 2015), which will be the case if the unsaturated zone dries up between recharge events (Sophocleous, 1985). The TRR calculations (Eq. 3) used 3 H activities in Melbourne rainfall as the input function (Tadros et al., 2014). The annual average 3 H activity of presentday rainfall in both Melbourne and Gatum is ~2.8 TU (Tadros et al., 2014; Table S1) and the 240 rainfall prior to the atmospheric nuclear tests was assumed to have had the same 3 H activity as present-day rainfall. n = 0.03 to 0.1 was again used and estimates of b are discussed below.

Mean residence times
The mean residence times (MRTs) and the covariance of 3 H and 14 C activities in groundwater were estimated via lumped parameter models (LPMs: Zuber and Maloszewski, 2001;Jurgen 245 et al., 2012). LPMs relate 14 C activity of water at time t (C out ) to the 14 C input during recharge over time (C in ) via the convolution integral (Zuber and Maloszewski, 2001;Jurgen et al., 2012) where q is the fraction of DIC derived from the rainfall or the soil zone, (t - m ) is the age of the water,  m is the MRT, λc is the decay 250 constant for 14 C (1.21×10 -4 yr -1 ), and g( m ) is the system response function that describes the distribution of residence times in the aquifer (described in detail by Maloszewski and Zuber, 1982;Zuber and Maloszewski, 2001;Jurgens et al., 2012). 3 H activities may be calculated from the input of 3 H over time in a similar way. Unlike 14 C, 3 H activities are not changed by reactions between the groundwater and the aquifer matrix; hence the q term is omitted. 255 There are several commonly used LPMs. The partial exponential model (PEM) may be applied for the aquifers where only the deeper groundwater flow paths are sampled. The dimensionless PEM ratio defines the ratio of the unsampled to sampled depths of the aquifer (Jurgens et al., 2012). This study used PEM ratios of 0.05 to 0.5 that cover the ratios of unsampled to sampled portions of the aquifers at Gatum. The dispersion model (DM) is derived from the one-260 dimensional advection-dispersion transport equation and is applicable to a broad range of flow systems (Maloszewski and Zuber, 1982;Zuber and Maloszewski, 2001;Jurgens et al., 2012).
The dimensionless dispersion parameter (DP) in this model describes the relative contributions of dispersion and advection. For flow systems of a few hundreds of metres to a few kilometres, DP values are likely to be in the range of 0.05 to 1.0 (Zuber and Maloszewski, 2001). Other 265 commonly applied LPMs, such as the exponential-piston flow or the gamma model, produce similar estimates of residence times (Jurgens et al., 2012;Howcroft et al., 2017). The longterm variations of atmospheric 14 C concentrations in the southern hemisphere (Hua and Barbetti, 2004;McCormac et al., 2004) were used as the 14 C input function, and 3 H activities in rainfall for Melbourne (Tadros et al., 2014) were used as the 3 H input function. 270

Hydraulic heads and properties
The hydraulic heads in regional groundwater from both pasture and forest catchments decrease from the upper to lower slopes implying that the regional groundwater flows southwards (Fig.   1b). In the pasture, the hydraulic heads in groundwater from all bores generally gradually 275 increase over several weeks to months following the onset of winter rainfall (Fig. 3b). The increase in hydraulic heads was higher in 2016, which was a year of higher than average rainfall (~800 mm: Bureau of Meteorology, 2020). This was especially evident at bore 63 (Fig. 3b). In the forest, groundwater heads from bores in the upper (3663 and 3665) and mid (3668) slopes decline uniformly over the monitoring period, and the groundwater head from bore 3658 near 280 the drainage zones does not show seasonal variations (Fig. 3c). However, fluctuations of the head from three bores near the drainage zones (3669) and the lower slopes (3656 and 3657) show seasonal variations similar to that of the groundwater in the pasture (Figs. 3b, 3c).
Values of K s range from 0.06 to 0.31 m day -1 in the pasture (Table S1, Fig. 2a) and from 0.002 to 0.18 m day -1 in the forest catchments (Table S1, Fig. 2b). The aquifers in the upper and lower 285 slopes of the pasture catchment have the highest K s values of ~0.31 m day -1 , whereas K s values of the aquifers in the forest are lowest on the lower slopes (Table S1, Fig. 2). The aquifers contain rocks from the same stratigraphic unit, and the heterogeneous hydraulic properties probably reflect the degree of weathering, cementation, and clay contents.

290
TDS concentrations of regional groundwater range from 282 to 7850 mg L -1 in the pasture catchment and 1190 to 7070 mg L -1 in the forest catchment (Table S1); the lowest salinity regional groundwater is from the upper slope of the pasture catchment. The TDS concentrations of the shallow riparian groundwater (≤1 m depth) are between 3890 and 8180 mg L -1 in the pasture and from 169 to 13600 mg L -1 in the forest (Table S1). Regional and riparian 295 groundwaters from both catchments have similar geochemistry. Na constitutes up to 67% of the total cations on a molar basis, and Cl accounts for up to 91% of total anions on a molar basis. Cl concentrations range between 45.2 and 8140 mg L -1 , which significantly exceed the mean concentrations of Cl in local rainfall (2.2 to 4.4 mg L -1 : Hutton and Leslie, 1958;Bormann, 2004;Dean et al., 2014). Molar Cl/Br ratios are between 180 and 884, with most in 300 the range between 450 and 830 (Fig. 4a), which spans those of seawater and coastal rainfall (~650: Davies et al., 1998Davies et al., , 2001. Cl/Br ratios are significantly lower than those that would result from halite dissolution (10 4 to 10 5 : Kloppmann et al., 2001;Cartwright et al., 2004Cartwright et al., , 2006 and do not increase with increasing Cl concentrations. These observations indicate that, as is the case throughout southeast Australia (e.g., Herczeg et al., 2001;Cartwright et al., 2006), Cl 305 is predominantly derived from rainfall and concentrated by evapotranspiration There is also no halite reported in the aquifers in this region. Cl concentrations of the shallow and the deeper groundwater overlap (Fig. 4b) and there is no correlation between Cl and 3 H (Fig. 4c). Ca and HCO 3 concentrations are uncorrelated (Fig. 4d) indicating that the dissolution of calcite is not a major process influencing groundwater geochemistry. 310

Radioisotopes
3 H activities of the regional groundwater are up to 1.48 TU (Table S1, Fig. 5). These are lower than the average annual 3 H activities of present-day rainfall in this region of ~2.8 TU (Tadros et al., 2014; Table S1). The highest 3 H activities (>1 TU) are from the regional groundwater in the upper slopes (15.5 m depth) and the drainage zone (~1.3 m depth) of the pasture catchment 315 and between 15.8 and 28.8 m depths in the forest catchment (Table S1). The regional groundwater from ≥28 m depth in the lower slopes of the pasture catchment and the drainage zones of the forest catchment locally have below detection (<0.02 TU) 3 H activities (Table S1).
The 3 H activities of the shallow riparian groundwater in the pasture vary from 0.26 to 0.79 TU with the highest activities from the lower slopes (Table S1, Fig. 5). The riparian groundwater 320 in the forest catchment has 3 H activities ranging from 2.01 to 4.10 TU (Table S1, Fig. 5), which are locally higher than the annual average 3 H activity of present-day rainfall (~2.8 TU). These high 3 H activities probably reflect seasonal recharge by the winter rainfall that in southeast Australia has higher 3 H activities than the annual average (Tadros et al., 2014). 14 C activities in the regional groundwater from the pasture and forest catchments range from 325 70.7 to 104 (pMC) and from 29.5 to 101 (pMC), respectively (Table S1, Fig. 5). The highest 14 C activities (>100 pMC) are from groundwater in the upper slopes of the pasture catchment and the lower zones of the forest catchment that also has high 3 H activities (Table S1). The lowest 14 C activities are from groundwater at 18 to 28.4 m depths in the mid-slope and the drainage lines of the forest catchment (Table S1). 14 C activities of the shallow riparian 330 groundwater are 85.5 to 102 pMC, with higher activities (>100 pMC) in the drainage zones of the forest catchment (Table S1, Fig. 5).

Discussion
The combined groundwater elevation and geochemical data allow residence times, mixing, and recharge rates at Gatum to be interpreted. 335

Mean residence times and mixing
3 H and 14 C activities help to understand water mixing within the aquifers (Le Gal La Salle et al., 2001;Cartwright et al., 2006Cartwright et al., , 2013 and the MRTs. The predicted 3 H vs. 14 C activities (Fig.   5) were calculated for all DIC being introduced by recharge (q = 1) and for 10% contribution of 14 C-free DIC from the aquifer matrix (q = 0.9). Mixing between older (low 3 H and low 14 C) 340 and recently-recharged groundwater (high 3 H and high 14 C) results in groundwater samples that plot to the left of the decay trends in Fig. 5. It is difficult to calculate MRTs for these mixed waters; however, it is possible to estimate MRTs from the 14 C activities for groundwater lying close to the predicted decay trends. The aquifers are dominated by siliceous rocks, and the major ion geochemistry implies little calcite dissolution. Similar values of q were estimated for 345 groundwater from other siliceous aquifers in southeast Australia (Cartwright et al., 2010(Cartwright et al., , 2012Atkinson et al., 2014;Raiber et al., 2015;Howcroft et al., 2017) and elsewhere (Vogel, 1970;Clark and Fritz, 1997). Much lower q values are precluded as samples cannot lie to the right of the 3 H vs. 14 C curves (Cartwright et al., 2006(Cartwright et al., , 2013(Cartwright et al., , 2017. This is because samples that are not a mixture of old and young groundwater, containing measurable 3 H will be less than 200 years 350 old. Over that time span, there has been negligible decay of 14 C, and the initial a 14 C of the sample is a 14 C/q (Clark and Fritz, 1997). If there were greater than 10% contribution of DIC from 14 C-free calcite dissolution, the estimated initial a 14 C would exceed the highest a 14 C recorded in soil CO 2 of ~120 pMC.
The calculated MRTs are up to 3,930 years in the pasture and up to 24,700 years in the forest 355 (Table 1, Fig. 6). While using LPMs is preferable to using a simple decay equation that assumes uncertainties in MRTs were up to 25%. While these are considerable, much of the regional groundwater undoubtedly have residence times of several thousands of years and were recharged prior to land clearing. These long residence times are consistent with the locally clay-rich nature of the aquifers and the moderate to low hydraulic conductivities.

Cl mass balance
Recharge rates calculated from the CMB method (Eq. 1) using total rainfall of ~640 mm yr -1 and Cl concentrations of 2.2 to 4.4 mg L -1 are similar between the pasture (0.3 to 61.6 mm yr -1 ) and forest (0.2 to 58.8 mm yr -1 ) catchments (Figs 2, 7a). The typical recharge rates for most of the regional groundwater are from 0.3 to 2.5 mm yr -1 in the pasture and 0.2 to 11.2 mm yr -1 in 370 the forest (Figs. 2, 7a). The Cl/Br ratios imply that the dissolution of halite is negligible, and all the Cl is delivered by the rainfall. Whether the rate of Cl delivery has been constant over long time periods is more difficult to assess; however, the rainfall Cl concentrations are typical of inland rainfall, and southeast Australia does not record major climate fluctuations such as glaciations or monsoons (Davies and Crosbie, 2018). 375 The CMB technique also assumes that the export of Cl by surface runoff is negligible. The streams at Gatum currently discharge ~8 % of local rainfall and much of the Cl that they export represents groundwater discharging into the stream (Adelana et al., 2015). This component of Cl does not impact the CMB recharge rate calculations. If some direct export of Cl has occurred, the recharge estimates would be slightly lower than estimated above. However, because the 380 initial land clearing has most likely increased streamflow in this region (Dresel et al., 2018), streamflows and the export of Cl would have historically been lower than the present-day.
Because Cl in groundwater accumulates over hundreds to thousands of years (Scanlon et al., 2002(Scanlon et al., , 2006, the CMB method generally yields longer-term recharge rates; these largely reflect pre-land clearing recharge in Australia (Alison and Hughes, 1978;Cartwright et al., 2007;Dean 385 et al., 2015;Perveen, 2016). This conclusion is consistent with the long 14 C residence times of much of the deeper regional groundwater at Gatum. The higher recharge rates (25.3 to 61.6 mm yr -1 ) are from the regional groundwater in the upper slopes of the pasture (bore 63) and the shallow riparian groundwater in the drainage zones (piezometer FD2) and the lower slopes (piezometer FB1) of the forest (Figs. 2, 7a). The groundwater at these sites has high 3 H and 14 C 390 activities, and the recharge rates from the CMB technique are likely to represent present-day recharge.

Water table fluctuations
The recharge rates were calculated using the WTF method (Eq. 2) from the bore hydrographs, which show seasonal head variations assuming S y = 0.03 to 0.1. The estimated recharge rates 395 range from 15 to 500 mm yr -1 (2 to 78% of rainfall) in the pasture and 30 to 400 mm yr -1 (5 to 63% of rainfall) in the forest (Figs. 2, 7b). As with the CMB estimates, the recharge rates are generally high at the upper slopes of the pasture catchment (Figs. 2, 7b). However, the highest recharge rates from the WTF method are unlikely given that evapotranspiration rates in this region approach the rainfall rates (Dean et al., 2016;Dresel et al., 2018;Azarnivand et al., 400 2020). The lower recharge rates estimated from the WTF method appear more reasonable but are still larger than most recharge rates estimated from the TRR method. The observation that much of the older saline groundwater has not been flushed from the catchments also implies that present-day recharge rates cannot be very high.
The WTF method requires the hydrograph recession curves to be estimated. There are 405 significant steep and straight recession curves in the bore hydrographs (Figs. 3b, 3c) that can lead to errors in recharge estimates. The WTF method may overestimate recharge due to air entrapped during recharge (the Lisse effect: Crosbie et al., 2005). However, this occurs during rapid recharge, which is not observed in the Gatum area. Dean et al. (2015) suggested that the high recharge rates estimated from the WTF method in the adjacent Mirranatwa catchments 410 might reflect focussed recharge from the streams. This is not the case at Gatum as high WTF recharge rates are recorded at all landscape positions and the streams only export ~8% of rainfall (Adelana et al., 2015). Because the WTF estimates gross recharge and geochemical methods estimate net recharge, there may be differences if the water is removed from the water table by evapotranspiration, especially in spring after the water tables reach their seasonal peak. 415 The plantation forest plausibly has high evapotranspiration rates (Benyon et al., 2006;Dean et al., 2015;Dresel et al., 2018); however, this explanation is unlikely in the pasture where water tables are locally several metres below land surface, and there is not deep-rooted vegetation.
It is most likely that the unrealistically high recharge rates estimated from the WTF method reflect an overestimation of S y due to the presence of remnant moisture in the unsaturated zone 420 between the recharge events (Gillham, 1984;Sophocleous, 1985;Crosbie et al., 2005Crosbie et al., , 2019.
While this is not unexpected, it is difficult to determine realistic values of S y to improve these estimates.

3 H renewal rate
The recharge rates for bores and shallow piezometers were estimated using the 3 H activities 425 and the TRR method (Eqs. 3, 4). These recharge rates were calculated for those groundwater samples which do not show the mixing of recent and older groundwater (Fig. 5). Regional groundwater from nested bores commonly has different TDS contents, EC values, 3 H and 14 C concentrations (Table S1), indicating that the groundwater is stratified. Much of the deeper groundwater has low 3 H and 14 C activities implying that it is not recently recharged. Based on 430 these differences in geochemistry (Table S1), b is estimated as being between 1 and 5 m in the regional groundwater. b values for the shallow riparian groundwater are estimated as 1 to 2 m, which is the approximate thickness of the shallow perched aquifers (Brouwer and Fitzpatrick, 2002). The estimated n values of 0.03 to 0.1 (Adelana et al., 2015;Dean et al., 2015) were used for these calculations. 435 Recharge rates from the regional groundwater are 0.5 to 14.0 mm yr -1 in the pasture and 0.01 to 59.5 mm yr -1 in the forest with most in the range of 0.01 to 0.6 mm yr -1 (Figs. 2, 7c). The higher recharge rates were from the upslopes of the pasture (14.0 mm yr -1 ) and the lower slopes of the forest (59.5 mm yr -1 ). The recharge rates in the riparian groundwater are from 0.05 to 0.5 mm yr -1 in the pasture and 13.3 to 89.0 mm yr -1 in the forest (Figs. 2, 7c). 440 The average annual 3 H activity in present-day rainfall at Gatum (~2.8 TU) is within the predicted range of the 3 H activities in present-day Melbourne rainfall (3.0 ± 0.2 TU), implying that the Melbourne 3 H input function is appropriate to use for this area. Assuming uncertainty in the 3 H input function of 5 to 10% (which is similar to the present-day variability of 3 H activities reported by Tadros et al., 2014) results in <5% uncertainties in recharge estimates. 445 The variation resulting from analytical uncertainties are lower than this. Recharge rates are most sensitive to the b values, which are not explicitly known and may be variable. However, b is unlikely to be >5 m based on the observed degree of chemical stratification. It may also be possible to estimate b from the fluctuation of the water table (on the basis that the rise in the water table corresponds to recharging water added to the top of the aquifer). If that is the case, 450 b values would be typically 1 to 3 m (Figs. 3b, 3c), which is within the range used in these calculations. There is also an assumption of a homogeneous aquifer. However, older water with low 3 H activities may locally be present in the zones of low hydraulic conductivity. Diffusion may reduce the 3 H activities in the more mobile groundwater adjacent to those zones (Sudicky and Frind, 1981;Cartwright et al., 2006;2017, 2020. Overall, the recharge rates from the TRR 455 method are again generally higher than those calculated using the CMB, which reflects the effects of the initial land clearing. However, despite both reflecting post-land clearing recharge, they are significantly lower than those estimated using the WTF.

Predicting the effect of land-use changes
In large regions of southeast Australia (including the study area), understanding whether and 460 by how much recharge increased following the initial land clearing is important in predicting the impact of a rising water table in causing salinization of the soils and the streams. For areas where plantation forests have been established, it is important to assess any subsequent impact of those plantations on recharge.
As expected, the recharge estimates from the CMB method are generally lower than those from 465 the WTF and TRR methods and largely reflect those prior to the initial replacement of native eucalyptus vegetation by pasture. Although both methods determine present-day recharge rates (Scanlon et al., 2002(Scanlon et al., , 2006, those estimated using the WTF method are significantly higher than the TRR estimates (Fig. 8). Having to estimate b represents a major uncertainty in the TRR calculations; however, b would have to be up to 50 m to achieve agreement between the 470 Deleted: T recharge estimates from these two methods. This is unlikely given the observations that major ion geochemistry, 3 H and 14 C activities of groundwater vary over vertical scales of a few metres (Table S1), implying that the groundwater is compartmentalised on those scales. It is also unlikely that b could be so large given the heterogeneous nature of the aquifers and the presence 475 of clay layers. It is most likely that the WTF method systematically overestimates recharge due to issues in estimating S y .
The recharge estimates from the TRR method differ little between the pasture and the forest; this is unexpected given that the establishment of plantation forests aimed to reduce the recharge rates. The evapotranspiration rates in the forest are also higher than in the pasture 480 (Adelena et al., 2015;Dresel et al., 2018) and the water levels are declining in some areas of the forest with no corresponding decline in the pasture (Figs. 3b, 3c), suggesting higher water use by the trees. The plantation covers ~62% of the forest catchment, and many of the bores are in cleared areas between the stands of trees (Fig. 1a). Thus, the recharge rates may not be representative of the forest as a whole. Additionally, the TRR averages recharge rates over the 485 timespan of the residence times of the aliquots of water contained in the water sample (Maloszewski and Zuber, 1982;Cartwright et al., 2017). If the zone at the top of the aquifer approximates a well-mixed reservoir, the timespan is 1/R n (Leduc et al., 2000;Favreau et al., 2002). R n values at Gatum are 3×10 -4 to 4×10 -1 , implying that recharge rates are averaged over decades to centuries. Thus, the recharge rates in the forest catchment may reflect those from 490 both before and following the recent reforestation.

Conclusions
As has been discussed elsewhere (Scanlon et al., 2002;Healy, 2010;Crosbie et al., 2010Crosbie et al., , 2019Cartwright et al., 2017;Moeck et al., 2020), estimating recharge rates can be difficult and a range of techniques together with other data (such as estimates of residence times) is required 495 to produce reliable results. By necessity, estimating pre-and post-land clearing recharge rates requires different methods. Both the CMB and WTF methods use data that are readily available (or is relatively low cost to attain). The uncertainties in the CMB estimates are relatively straightforward to address, and this represents a viable method of estimating historic recharge rates; however, the commonly-used WTF method may not be able to be applied in a 500 straightforward manner to estimate present-day recharge rates. Relatively high WTF recharge rates (up to 161 and 366 mm yr -1 ) were also calculated in adjacent catchments with similar land-use (Dean et al., 2015;Perveen, 2016). 3 H activities in groundwater from those catchments are similar to those at Gatum, implying that recharge estimates based on the TRR method would again be significantly lower. Cartwright et al. (2007) and Crosbie et al. (2010) also 505 reported that the recharge estimates from the TRR method and other geochemical tracers in semi-arid catchments elsewhere in Australia are lower than those from the WTF method. A similar observation was made for temperate catchments (Cartwright et al., 2020). Some of the discrepancy may be caused by the local presence of older water in lower permeability regions; however, this probably does not entirely account for the systematic differences across a range 510 of catchments.
Additionally, the recharge rates are likely to be spatially variable across both catchments, and even with a relatively high density of data such as at Gatum, it is difficult to estimate typical or area-integrated values. In the case of understanding recharge rates in the plantation forest, the necessity that bores are in cleared areas (between the stands of trees) also makes it 515 questionable whether the recharge rates are representative. Finally, all the geochemical techniques integrate recharge rate estimates over years to centuries and are thus ineffective at determining changes over short-timescale than this.
Detailed soil moisture measurements that would improve S y estimates and geochemical tracers, such as 3 H, may not always be available. Integrated surface and subsurface hydrogeologic 520 models, which simulate coupled groundwater, surface water and soil water fluxes, might Deleted: in the same region provide additional tools to estimate recharge rates that could be used to support the field and geochemical data (Scudeler et al., 2016;Daneshmand et al., 2019). With the increasing availability of soil moisture, evapotranspiration, rainfall, streamflow and groundwater 525 elevation data, catchment water balance models (e.g., Wada et al., 2010;Moeck et al., 2020) might also represent viable methods of estimating recharge, especially over large areas.
The results of this study inform the understanding of hydrogeological processes in this and similar semi-arid regions globally. The present-day recharge rates in the pasture, which is typical of cleared land in southeast Australia, are likely to be <10 mm yr -1 . Despite these being 530 significantly higher than the pre-land clearing recharge rates, they only result in the gradual replacement of the older saline water stored in these aquifers (as is implied by the trends of dbgs vs. Cl and 3 H vs. Cl: Figs. 4b, 4c). Additionally, while there has been a rise in the water table caused by increased recharge, and in some cases increased drainage in the streams, the magnitude of these changes will be limited by the modest recharge rates. The results also 535 indicate that care must be used in assigning recharge rates as boundary conditions numerical models.
Author contribution: Shovon Barua and Ian Cartwright conducted the sampling assisted by P. Evan Dresel and Edoardo Daly. Shovon Barua carried out the analytical work conducted at Monash University. P. Evan Dresel and Edoardo Daly manage the field site and provided pre-540 existing data. All authors were involved in writing the manuscript.
Competing interests: The authors declare that they have no conflict of interest.   symbols are for the regional groundwater. The single high 3 H activity possibly reflects recharge by winter rainfall. Samples lying to the left of the covariance curves probably record mixing between younger and older groundwater (see text for discussion).   : Comparison between recharge rates for the regional groundwater estimated from 935 WTF and TRR. Bars represent the ranges of calculated recharge values from Table 1.