Located in the northwest of the island (Figure 1), Welsby Lagoon is a shallow (∼1.2 m maximum water depth) palustrine wetland. The wetland lacks major stream inflow and outflow connections and is the surface expression of the perched freshwater aquifer of the Welsby drainage province, which drains an area of approximately 538 ha (∼2% of the island) (Leach, 2011). As is the case with many wetlands on the island, including Fern Gully Lagoon, the wetland is dominated by emergent aquatic plants including Baumea species and Lepironia articulata, along with submerged Cycnogeton procerum (Marshall and McGregor, 2011).

Fern Gully Lagoon is a shallow (1.2 m maximum depth) palustrine wetland. It lies within a shallow bowl of vegetated dunes at the north-western end of North Stradbroke Island in the Flinders Beach drainage province (Leach, 2011). The wetland has two above-ground outflows, the largest being to the eponymous Fern Gully, through which outflow has been reported by local residents only very occasionally though the period of this study.

Swallow Lagoon is a small, perched lake that sits high in the dunes. Swallow Lagoon has no inflow or outflow streams and, as such, is effectively the surface expression of a small aquifer contained within the surrounding sand. There is an outflow channel to the south-east of the lake but its thalweg is at least 0.7 m above maximum recorded lake elevation and was not observed flowing during the study period.

The largest lake on the island, Brown Lake (or Bummiera in the Quandamooka language) is a popular tourist destination and holds significant cultural value for the Quandamooka people. Except for very high precipitation events, the lake has no active surface inflow or outflow streams. An outflow channel to the south has been inactive during the monitored period. Previous studies have noted that the depth of Brown Lake can fluctuate substantially over a relatively short period of time (Laycock, 1975; Laycock, 1978). Laycock (Laycock, 1978) found lake-level fluctuation in the range of 3.49 m over “a number of years,” while Lee-Manwar et al. (Lee-Manwar et al., 1980) reported a fall of 3.9 m between May 1974 and July 1977. This is consistent with current monitoring which recorded a water level decline of 3.5 m between June 2012 and December 2019.

The bathymetries of Brown Lake and Swallow Lagoon were surveyed using a gridded network of depth, position and time every 2 s recorded by a boat mounted sonar and GPS system. Each survey consisted of a longitudinal section, a near-perimeter trace and 40 cross sections. Perimeters were hand drawn to encompass the survey points, and fit available base maps. Comparisons with a whole of NSI digital elevation model (DEM) showed they were compatible. Brown Lake was surveyed on 11 March 2011, when the water surface was 57.53 m relative to Australian Height Datum (AHD) and again on 18 May 2018, when the water surface was 54.75 m (AHD). Swallow Lagoon was also surveyed twice; on 20 October 2009, when the water surface was 154.21 m (AHD) and on 11 March 2011, when the water surface was 154.96 m (AHD). The two Brown Lake surveys were merged to create a final DEM (Supplementary Figure S1). The 2018 survey data were adjusted by adding 2.6 m to account for lower water levels at this time. This was calculated by averaging the difference in survey points from the two surveys that were within 0.5 m of each other, resulting in 100 such points. As a validation check, the wetland volume from a DEM generated from the 2018 survey alone was compared with that from the merged survey DEM at the same depth, and was found to be very similar. The same process was used to combine Swallow Lagoon surveys by adding 0.75 m to the 2009 data to correspond to the 2011 data. These pooled observations were used to generate DEMs of each wetland by linearly interpolating between survey points (ArcGIS 10.4.1, ESRI). Wetland volume and surface area were calculated and tabulated at 0.01 m depth intervals from full wetland extent to empty (ArcGIS, 10.4.1, 3d analyst tools, ESRI). To date, no bathymetric surveys have been completed for Welsby Lagoon and Fern Gully Lagoon, due to the logistical challenge of traversing and depth sounding the shallow, densely vegetated wetlands.

Wetland and aquifer water levels were monitored using a network of barometrically-compensated diver loggers (Van Essen, Delft) in hand-installed piezometers recording at 60-min intervals. Manual water level measurements were made every 1–3 months to validate the logger data. Regional groundwater table head heights were monitored at bore 14400099, which is located near to Brown Lake. This bore samples the regional, island overarching aquifer, and is slotted between 36.4 and 42.4 m below ground surface. Water measurements were recorded manually, on average 3–4 times each year during the period 2011–2020. Monitored water levels are reported relative to AHD to provide a common reference elevation across all locations. All surface and groundwater levels were laser-levelled from formal Queensland Government survey points to ensure that all were correct to within <5 mm.

Daily rainfall and evaporation data from the Bureau of Meteorology stations at Dunwich and Point Lookout, which are closest to the wetlands, were obtained from the SILO Patched Point Dataset, which combines observations and interpolations to provide daily data for a selected set of weather stations (DES, 2019).

Monthly rainwater samples were collected using a purpose-built rainfall sampler, consisting of a 90 mm diameter polycarbonate tube (400 mm length; total volume ∼2.54 L) with an inbuilt funnel and a tap in the bottom to collect the water sample [e.g. 40]. A non-water soluble evaporation barrier was used (silicone or paraffin oil), and replenished after each collection. Wetland surface waters were collected monthly in 50 ml high-density polyethylene (HDPE) bottles from the edge of each wetland at the locations marked in Figure 1. When the water level dropped beneath the sediment surface at Welsby Lagoon and Fern Gully Lagoon, sediment interstitial water was collected by digging a hole to 20 cm depth and collecting the water that subsequently filled the hole. This interstitial water is referred to as “peat water,” given the peaty nature of the surficial sediments at Welsby and Fern Gully Lagoons. Shallow groundwater samples were collected from piezometers installed beside Welsby Lagoon and Fern Gully Lagoon (Figure 1). Groundwater samples were collected from the piezometer pipe using a battery powered slimline centrifugal groundwater pump after initially purging three times the piezometer pipe volume to ensure the collection of fresh aquifer water. All water samples were collected in carefully sealed HDPE bottles which were completely filled to avoid any headspace. Samples were refrigerated at 4°C until analysis.

Oxygen and hydrogen isotope analysis was performed at Flinders Analytical, Flinders University, South Australia, on a Picarro L2130-I Isotope Water Analyser. Samples were calibrated against in-house laboratory standards consisting of desalinated seawater (δ²H = 0.95‰, δ¹⁸O = 7.4‰) and bottled water (δ²H = −73.8‰, δ¹⁸O = −10.36‰). A rainwater sample (δ²H = −52.1‰, δ¹⁸O = −8.49‰) was used as an internal laboratory quality control.

Oxygen and hydrogen isotope ratios are referred to using the standard delta notation: δ = ( R s a m p l e − R s t a n d a r d R s t a n d a r d ) . 1000 Where δ refers to either δ¹⁸O or δ²H in ‰, and R is the ratio ¹⁸O/¹⁶O or D/H respectively. All data are reported relative to the Vienna Standard Mean Ocean Water (V-SMOW).

Lake water mass balance was modelled for Swallow Lagoon and Brown Lake for December 2011 through to February 2020—the period for which lake depth monitoring data were collected. Models were developed using the EcoModeller software Waterhole Persistence Model plug-in (Lobegeiger et al., 2015). The model requires a detailed bathymetry of the wetland, relating wetland volume and surface area to depth at 0.01 m intervals, and calculates the mass balance of fluxes into (direct rainfall, runoff and groundwater) vs. out (outseepage, surface flow, evaporation) of the wetland to estimate change in wetland water depth through time. Data inputs to develop the model were daily rainfall (sourced from SILO) with evaporation estimated using Morton’s shallow lake model (Morton, 1983) (both for BoM station Dunwich 040537). A “surface flow” dummy variable was set to zero for the entire time series because of the absence of such flows during the monitored period. Parameters for runoff (catchment size) and outseepage were estimated as described below. The absence of bathymetric data prevented the modelling of Welsby Lagoon and Fern Gully Lagoon.

Water balance models were calibrated iteratively so that predicted water depth best matched the measured depth. Calibration was achieved by adjusting two key model parameters: the daily rate of seepage loss (seepage through the base and sides of the wetland in mm per day) and a catchment area multiplier to represent runoff and shallow throughflow. Seepage loss was initially assumed to be zero and increased so that water loss matched observed behaviour. Rainfall catchment area was initially set as the maximum area of the wetland from the DEM and adjusted accordingly to match observed increases in wetland depth associated with rainfall. Varying catchment area in the model is used as a proxy for variable catchment runoff (i.e., the fraction of catchment precipitation which is transported to the wetlands) which is assumed to be constant in the model. It should be stressed that we do not propose that the actual catchment area varied through time. The optimum parameter values for calibration were determined as the minimum Root Square Mean Errors (RSMEs) achieved. Calibration assumed no anthropogenic water extraction and no daily evaporation multiplier.

For both wetlands, initial calibration attempts were unable to match observed depth behaviour for the entire record and it was evident that different periods of the overall record required independent calibration. These periods were identified iteratively, based on dates where there were notable depth changes. In some cases these selections were suitable for accurate calibration, but in others they required further division. Eventually low RMSE (<0.1 m) calibrations for the entire record were achieved by dividing the Brown Lake record into five periods and the Swallow Lagoon record into eight, each with its own combination of model parameter values. The need to alter the model calibration parameters through time reflects complex changes in the efficiency of catchment runoff and wetland outseepage according to climate and the depth of the wetland. As a consequence, the model could not be validated against an independent test period and can only reliably serve to explain observations during the monitored period, rather than project into the future. The model was used to test the effect of a lengthy drought on water depths in Swallow Lagoon and Brown Lake by repeating the climate of the driest year on record (2019) for 22 years, with the seepage rate and catchment area parameters fixed as those identified to best fit the 2019 state of the wetlands. 2019 was identified as the driest year on record by calculating the annual totals of daily rainfall and evaporation from the Dunwich weather station, through the 1893 to 2019 period. Annual totals of each parameter were used to rank years from the lowest to highest rainfall and from the highest to lowest evaporation. A biplot of these ranks identified 2019 as having the least rain and most evaporation (Supplementary Figure S2). 22 year long pseudo-time series for both rainfall and evaporation were created by repeating the daily climate of 2019. Commencing from the wetland depths measured on 29 July, 2016, and again using the calibrated 2016–2020 models, the depth behaviour of Brown Lake and Swallow Lagoon was simulated under this scenario to assess if repeat occurrences of the driest year on record would result in the wetlands emptying and, if so, after how long.

Rainfall at Point Lookout varied markedly throughout the 4 years of monitoring (Figure 3). With the exception of 2015 (with 1,613 mm), the most recent 5 years of the monitoring period described here recorded below average rainfall with 2016 (868 mm) and 2019 (876 mm) being the two lowest annual rainfall totals since records began at Point Lookout in 1947. At Dunwich, on the west coast of NSI, the year 2019 experienced the lowest rainfall and the highest evaporation since records commenced in 1893 (BoM, 2020). Long term records on NSI indicate a dominance of summer rainfall; however, during the monitored period the month of June was consistently wet, with monthly precipitation ∼250 mm. Several years experienced annual maximum precipitation in February and March (250–350 mm), associated with late summer storms. The driest months were consistently August and September, which experienced ∼50 mm/month.

During the 2015–2019 isotope monitoring period, daily temperature varied over a consistent seasonal cycle between 12 and 28°C (Figure 3). Modelled evaporation also exhibited a distinct seasonal cycle, albeit with minima and maxima (50–300 mm/month) lagged behind the measured temperature by approximately 1–2 months (Figure 3).

Water depth logging commenced in December 2011, at Swallow Lagoon and Brown Lake and in July 2015, at Welsby Lagoon and Fern Gully Lagoon. For Swallow Lagoon, only surface water levels were monitored; however, at the other three sites both surface and groundwater depths were monitored. Wetland water depths declined from mid-2015 through to late 2019 (Figures 3, 4), although at Swallow Lagoon, for which a longer record is available, water depths had increased during the first half of 2015. The decline in water depth at Brown Lake stabilised by the beginning of 2017, fluctuating around a level of ∼54.6 m AHD except for a further short decrease in late 2018 (Figure 3). Water depth at Swallow Lagoon continued to fall throughout our study period (Figure 4). Groundwater level at Brown Lake, exhibited a similar pattern to surface water level, with both surface and groundwater levels being nearly identical during periods of low lake level, and with groundwater being higher by ∼0.2–0.3 m after periods of lake level rise (Figure 3).

Surface water levels at Welsby and Fern Gully Lagoons persisted at or just below the sediment-water interface during most of the monitoring period, although water could still be sampled from saturated lagoonal sediments. The water depth at these two sites fell rapidly between July and December 2015, after which their hydrological regimes were characterised by brief episodes of rapid re-filling after rainfall (e.g. March 2016 and May 2017) followed by subsequent declines in water level over the course of one to five months (Figure 3). Groundwater depth at both Welsby and Fern Gully Lagoons was found to be consistently lower than surface water level, ranging from ∼0.2 m beneath surface water level during periods of wetland inundation, and ∼0.5 m beneath surface water level during periods of wetland drying (Figure 3).

The δ¹⁸O of precipitation (herein referred to as δ¹⁸O_P), sampled at Welsby Lagoon, fluctuated seasonally around a mean of –2.9‰ with a standard deviation of 1.7‰ between August 2015 to May 2019 (Figure 4). Notable low δ¹⁸O_P values were recorded in June 2016 (–7.6‰), June-July 2017 (–6.4‰ and –5.5‰) and July 2018 (–5.4‰), all of which coincided with periods of high monthly precipitation (Figure 4). High δ¹⁸O_P values of +0.9‰ and +0.1‰ were recorded for November 2016 and August 2018 (Figure 4).

δ¹⁸O in wetland surface and interstitial peat waters (herein referred to as δ¹⁸O_L) at Welsby Lagoon fluctuated within a range of –6‰ to +7‰, with highs and lows coinciding with those in δ¹⁸O_P, albeit with a greater range (Figure 4). Two periods of markedly increasing δ¹⁸O_L at Welsby Lagoon coincided with periods of declining water level, between July 2015 and February 2016, and again between June 2016 and November 2016 (Figure 4). Interstitial peat waters collected between November 2016 and January 2017 had values consistent with the trend of the surface waters (Figure 4). The δ¹⁸O of water sampled at two shallow groundwater bores (δ¹⁸O_GW) at Welsby Lagoon exhibited markedly different temporal patterns, despite being located only ∼13 m apart. δ¹⁸O_GW variability at bore #1, located closer to the wetland, resembled a time-averaged surface water signal, with a period of gradually increasing δ¹⁸O_GW values from 0‰ in June 2015 to ∼ +3‰ in September 2016, followed by a decline to ∼ –2‰ by April 2017 and subsequent stabilisation around that value for the rest of the monitored period (Figure 4). By contrast, δ¹⁸O_GW at bore #2 was consistently lower than that at bore #1, except for a period during February–March 2017. δ¹⁸O_GW at bore #2 exhibited a high of ∼ +2‰ in June 2016, followed by a decline to ∼ –5‰ by September 2016. A subsequent increase to ∼0‰ in March 2018 was followed by an immediate decline and stabilisation at ∼ –4‰ for the rest of the record (Figure 4).

Variability in δ¹⁸O_L at Fern Gully Lagoon exhibited similar patterns to that at Welsby Lagoon, however within a narrower range, from –4‰ to +4‰ (Figure 4). The two groundwater bores sampled had water with similar, but not identical δ¹⁸O_GW to each other, except for May 2018. A marked decline in δ¹⁸O_GW occurred in May 2016, declining from –0.5‰ to –4‰, followed by a gradual increase in δ¹⁸O_GW through to late 2019. This variability in δ¹⁸O_GW at Fern Gully was generally 0–2‰ lower than surface waters throughout the record; however, during two notable periods, January to April 2016 and September 2016 to March 2017, marked 6‰ increases in δ¹⁸O_L were not mirrored by δ¹⁸O_GW, which remained relatively invariant (Figure 4).

No groundwaters were sampled at Swallow Lagoon or Brown Lake; however, δ¹⁸O_L exhibited similar patterns throughout the monitored period with a range of 0 to +6‰ at Swallow Lagoon and +1 to +5‰ at Brown Lake (Figure 4). δ¹⁸O_L in both lakes exhibited an increasing trend from July 2015 to February, 2017, punctuated by a period of lower δ¹⁸O_L from June 2016 to December 2016 (Figure 4). Subsequently, δ¹⁸O_L gradually decreased to ∼ +3‰ at both lakes, with a larger range of short term variability at Swallow Lagoon compared to Brown Lake, coinciding with overall lower and more stable water levels (Figure 4).

At all of the monitored wetlands, δ¹⁸O of surface waters correlated with equivalent δ²H with R ² exceeding 0.94 in all cases (Figure 5). Deviation from the δ¹⁸O-δ²H relationship defined by rainfall at Welsby Lagoon (a.k.a the Meteoric Water Line; MWL) is indicative of a local evaporation line (LEL), and the slopes of the LELs were consistent between Swallow Lagoon and Brown Lake, and similar between Welsby Lagoon and Fern Gully Lagoon (Figure 5). Where peat waters and groundwaters were sampled at Welsby Lagoon and Fern Gully Lagoon, the δ¹⁸O and δ²H of those waters fell close to the LELs defined by surface waters. At Swallow Lagoon and Brown Lake, the LEL intersects the MWL near to the mean of all rainfall samples; however, at Welsby Lagoon and Fern Gully Lagoon, the MWL-LEL intersects fall closer to the minimum δ¹⁸O and δ²H values associated with winter rainfall (Figure 5). As a consequence of the clearly defined LELs, δ¹⁸O_L consistently correlates with the deuterium excess of wetland water at all locations, with R ² values ranging between 0.85–0.95 depending on site.

Monthly change in lake depth and equivalent change in δ¹⁸O_L was calculated to assess the relationship between hydrological and stable isotope change, whilst excluding the effect of changes which occurred either prior to the monitoring period, or during gaps in the record. Monthly change in lake depth at Swallow Lagoon and Brown Lake exhibited significant correlations with the equivalent monthly change in δ¹⁸O_L at both sites, with an R ² of 0.63 and slope of –5.76 for Swallow Lagoon and an R ² of 0.83 and a slope of –4.37 for Brown Lake (Figures 6A,B) reflecting the greater range of variability at Swallow Lagoon. At Welsby Lagoon and Fern Gully Lagoon change in depth and change in δ¹⁸O_L exhibit weaker, yet statistically significant (p < 0.05) relationships (Figures 6C,D), albeit subject to several possible outliers which appear to have resulted in lower regression slopes than the majority of data imply. No clear reason for outlier rejection was identified, but it is possible that these values reflect short term (i.e., sub-monthly) rainfall events with distinctly different δ¹⁸O which, in turn, compensate for changes in δ¹⁸O_L associated with lake filling or evaporation.

Water level variability at Brown Lake and Swallow Lagoon was modelled for December 2011 to February 2020 (Figure 7). To check for the possible influence of regional groundwater influence, regional groundwater head height was compared to lake surface and lake floor levels, and throughout both records lake levels were perched well above groundwater, despite groundwater levels exhibiting a similar, albeit lagged pattern of variability (Figure 7). In order to simulate observed water depth variability through this period, it was necessary to vary two key model parameters: rainfall catchment area and daily outseepage loss rate. For Brown Lake, the catchment area was set at 2x lake area from December 2011 to February 2013, which was a period of high precipitation, and at 1.5x from February 2013 to January 2014. Catchment area was made constant at 1x for the rest of the model run (Figure 7A). Outseepage at Brown Lake was set at 4.20 mm/day from December 2011 to January 2014, and 2.75 mm/day from January 2014 to October 2014. For a brief window of time in 2015, outseepage at Brown Lake was set at 2.0 mm/day, whereas for August 2016 to February 2020, outseepage was set at 0.2 mm/day (Figure 7A). The RMSE of observed vs. modelled lake water depth at Brown Lake varied between the five calibration periods within a range of 0.025–0.085 m.

For Swallow Lagoon, it was necessary to vary the calibration parameters more frequently in order to accurately simulate wetland water depths. Between December 2011 and February 2013, catchment area was set initially at 1x lake area until January 2012, then at 2x between January 2012 to November 2012, before returning to 1x from November 2012 to March 2013, and again for April 2014 to February 2015 (Figure 7B). Between February 2015 to March 2016, catchment area for Swallow Lagoon was set to 1.5x. From August 2016 to February 2012 catchment area was set to 0.5x lake area (Figure 7B). Outseepage rate at Swallow Lagoon was varied between 0.5 and 4 mm/day, with the highest rate of outseepage required to simulate water depth between April and October 2014, and the lowest outseepage rate applied between February 2018 to February 2020. The RMSE of observed vs. modelled lake water depth at Swallow Lagoon varied between the eight calibration periods within a range of 0.028–0.086.

Overall, it is apparent that during periods of higher wetland water depth, which are also periods of higher effective precipitation, it is necessary to set parameters for higher runoff efficiency (catchment size) and higher wetland outseepage. This need to vary the model parameters meant that it is impossible to independently validate the model. Furthermore, model projections outside of the calibration period are likely to be highly uncertain. The model does provide utility, however, for understanding the key fluxes into, and out, of the wetlands during the monitoring period, and also to examine how hypothetical drought conditions might affect the wetlands over a number of years. In this respect, under the scenario where the driest year on record is repeated for 22 years, neither Brown Lake nor Swallow Lagoon was simulated to dry completely (Figure 8). Both wetlands lost water on an annual cycle, yet approached depth minima after 4 years of this climate regime for Swallow Lagoon and after 8 years for Brown Lake. This pattern of declining water level reached asymptote at around 1 m deep for Swallow Lagoon and 3.75 m deep for Brown Lake, with very minor additional annual losses.

Water depths and δ¹⁸O at Fern Gully Lagoon and Welsby Lagoon exhibited markedly different patterns to the two deeper study sites, due to them both drying completely over the study period (Figures 3, 4). Both palustrine wetlands exhibited very similar patterns in depth and δ¹⁸O, and these differences in the range of variability are likely explained by differences in wetland and catchment surface area. During the years 2015–2016 inclusive, water depths fluctuated markedly at both sites, with a general trend of decline punctuated by a period of increase between May and August 2016. These decreases and increases in water depth coincided with corresponding increase and decrease in δ¹⁸O_L (Figure 4). Water levels dropped to that of the sediment-water interface after 2016, whereupon the hydrology of both palustrine wetlands became characterised by flashy periods of filling and subsequent evaporation. Isotope data coverage through this period is more sparse, but in general, those events were characterised by declines in δ¹⁸O_L during filling, followed by subsequent isotopic enrichment. The full range of isotopic enrichment could not be quantified due to the short timespan of the events—which would have required daily scale monitoring. Interestingly, although wetland water depths increased marginally from early 2018 onwards, the δ¹⁸O of surface waters also increased at both Fern Gully and Welsby Lagoon. In the absence of a comparable change in δ¹⁸O_P, this pattern can be explained by the general persistence of water in both lagoons which, after filling during high rainfall in early 2018, underwent progressive evaporation throughout the year. Transpiration by emergent or littoral vegetation is not a viable explanation for the observed variability in δ¹⁸O_GW since plants are non-discriminatory with respect to the δ¹⁸O of water they take up (Allison et al., 1984; Ehleringer and Dawson, 1992).

Comparison between surface and groundwater levels at Welsby Lagoon and Fern Gully Lagoon suggest a complex hydrological interplay in these shallow palustrine wetlands. At both sites, groundwater levels exhibited patterns which closely resembled those of surface water depths, suggesting that the processes which drive recharge and loss are tied (Figure 3). At both sites, periods of surface water level increase were matched by immediate rises in local groundwater levels, suggesting that the recharge of these subsurface aquifers occurred rapidly, alongside rises in surface water level (Figure 3). This, in turn, implies that the subsurface aquifers are relatively small in volume, and also laterally highly permeable—consistent with their sandy substrate. The monitored groundwater levels at both sites were consistently lower than those of the surface water by between 0.2 and 0.5 m (Figure 3), supporting previous observations that the monitored bores are located down-stream of groundwater flow through the wetland (Draper, 2017).

The groundwater δ¹⁸O data for Welsby and Fern Gully Lagoons highlight further complexity and heterogeneity within these systems. Overall, δ¹⁸O_GW values at both lagoons fell within a similar range to δ¹⁸O_L (Figure 4). During periods of higher surface water level, δ¹⁸O_GW values adhered to local evaporation lines and were enriched relative to rainfall, further indicating that the monitored groundwaters were fed in part by outseepage from evaporated lagoon water (Figures 4, 5). However, during periods of low or absent surface water, the δ¹⁸O_GW at both lagoons exhibited relatively constant values within the range of the δ¹⁸O of precipitation (Figures 4, 5) and the regional groundwater and perched aquifer δ¹⁸O values (Hofmann et al., 2020). At Welsby Lagoon, two neighbouring groundwater bores exhibited contrasting patterns of change. δ¹⁸O values from bore #1 (∼13 m closer to the wetland) were consistently higher by 2–4‰ than equivalent δ¹⁸O_P, suggesting evaporation prior to infiltration. In addition, the bore #1 δ¹⁸O_GW values exhibit a pattern similar to a lagged δ¹⁸O_L as might be expected for wetland outflow (Figure 4). By contrast, the δ¹⁸O_GW in bore #2 was consistently 1–3‰ lower than at bore #1, except for a brief period in early 2017 (Figure 4). Although lower than in bore #1, the δ¹⁸O_GW at bore #2 were still high, relative to rainfall, suggesting that the groundwater at this location received a smaller fraction of wetland outseepage, despite its proximity to the lagoon. Following a large amount of low δ¹⁸O_P precipitation in June 2016 and April 2017, δ¹⁸O_GW at bore #2 decreased to ∼ –6‰, indicative of direct recharge by rainfall. Subsequently, relatively constant δ¹⁸O_GW at bore #2 at ∼ –6‰ from late 2017 to early 2018 implies that this aquifer became disconnected from wetland outseepage from there on (Figure 4). Throughout this period the water levels at bore #2 continued to fluctuate in tandem with both bore #1 groundwater and surface water (Figure 3), indicating that between mid-2017 and early 2019 the aquifer underwent recharge from rainfall and loss by outseepage without mixing with surface water and without any notable effects of evaporative loss. A further point of interest is that the groundwater head heights at Welsby bore #2 were virtually identical to those at bore #1, which did exhibit δ¹⁸O_GW patterns that suggest continued connectivity with the surface hydrology. This implies lateral heterogeneity in the shallow aquifers beneath Welsby Lagoon, as well as lateral variability in the extent to which those aquifers are connected to the surface water.

Similar groundwater patterns emerged at Fern Gully Lagoon. When the lagoon had standing water during late 2015 and early 2016, δ¹⁸O_GW values from bore #1 (closer to the wetland) were similar to those of surface δ¹⁸O_L (Figure 4). Following the large amount of low δ¹⁸O_P rainfall in mid-2016, shallow groundwater at both bores at Fern Gully Lagoon dropped to ∼ –5‰. Subsequently, despite a marked increase in the δ¹⁸O_L of surface waters over the following year, only a gradual pattern of increase was observed in δ¹⁸O_GW, suggesting only a small fraction of recharge from wetland outseepage, presumably due to the low volume of water in the wetland. Nevertheless, as at Welsby Lagoon, the groundwater levels at Fern Gully Lagoon exhibited abrupt periods of rise and fall in concert with surface water levels, indicating that through this period their recharge was dominated by direct infiltration by rainwater.

From these observations, it is reasonable to conclude that the shallow groundwater systems at these lagoons are only intermittently connected with the surface waters at their respective wetlands. During the monitored period, these groundwaters received a combination of direct infiltration and outseepage from the wetlands, seemingly subject to wetland surface water depth and volume. It appears that during periods of low water level, the groundwater aquifers received very little outseepage from the wetland, yet still received recharge by direct infiltration of rainfall.

Despite the relatively high rainfall and short water residence times of wetlands on North Stradbroke Island, the lake water balance and δ¹⁸O_L of both Swallow Lagoon and Brown Lake was strongly influenced by evaporation. Both lakes underwent a marked decline in lake level between mid-2015 and early 2017, by ∼2 m at Swallow Lagoon and ∼1.5 m at Brown Lake (Figures 3, 4). Groundwater levels were only monitored at Brown Lake, and there—unlike at the palustrine wetlands—groundwater levels were consistently level with or higher than surface water levels, indicating that local groundwater probably drained into the lake throughout the monitoring period. Over the same period, the δ¹⁸O_L of both lakes increased, from ∼0 to +6‰ at Swallow Lagoon and from ∼+1 to +5‰ at Brown Lake (Figure 4). The δ¹⁸O_L data, as well as the similarity between lakes, again illustrates the dominant role of evaporation in lake level decline, rather than loss of water via subsurface drainage, which should have no effect on δ¹⁸O_L. Transpiration by littoral or emergent aquatic plants is another possible pathway for water loss, however in this case the uptake of water by plant roots should not discriminate oxygen isotopes (Allison et al., 1984; Ehleringer and Dawson, 1992), therefore not explaining the observed patterns in δ¹⁸O_L. Changes in vegetation may, however, affect evaporation rates via shading and sheltering, both in the catchment and on the lake water surface. Changes in δ¹⁸O_L cannot be explained by changes in δ¹⁸O_P, despite the similar pattern exhibited by δ¹⁸O_P and δ¹⁸O_L. During the major period of lake level decline and δ¹⁸O_L increase, δ¹⁸O_P remained relatively constant at ∼–2‰, except for 2 months with high δ¹⁸O_P > +1‰ (January 2016 and November 2016), both of which were characterised by low precipitation (<100 mm), as well as a month with high precipitation amount (>200 mm) and markedly low δ¹⁸O_P value of –7‰ in May 2016 (Figure 4). Whilst the month of high precipitation clearly led to a decline in δ¹⁸O_L at both sites, overall δ¹⁸O_P and δ¹⁸O_L do not correlate and in particular the trend of increasing δ¹⁸O_L is not matched by a concurrent increase in δ¹⁸O_P.

From January 2017 through to December 2019, δ¹⁸O_L at Swallow Lagoon and Brown Lake declined by 2‰ and 1‰, respectively, exhibiting a narrower range of variability through the year 2018 than observed during previous years (Figure 4). This decline in δ¹⁸O occurred during a period of continued lake level decline at Swallow Lagoon, and a period of low, yet stable, lake levels at Brown Lake. Thus, at first glance these patterns appear contrary to the expected negative relationship between δ¹⁸O and lake level, where lake water evaporation drives an increase in δ¹⁸O. One explanation for this pattern could be a decline in the δ¹⁸O of mean precipitation, however the continued adherence to the LEL (Figure 5) does not support this hypothesis. The mean annual δ¹⁸O of rainfall (un-weighted) declined by ∼1‰ during this period, however throughout the record, the effect of high amount, low δ¹⁸O_P events on the total δ¹⁸O of the two lakes was relatively minor and short lived—for example the high monthly rainfall event in June 2017 (248 mm, of which 110 mm fell on 14/6/2016) was associated with a decrease in δ¹⁸O_P by ∼ –4‰, yet only a ∼–2‰ decrease in δ¹⁸O_L at Swallow Lagoon which persisted for just 2 months (Figure 4). Rather than a significant change in the source or δ¹⁸O of moisture reaching the lagoons between 2017 and 2019, it is more likely that the decline in δ¹⁸O in both lakes over this period reflects the decrease in wetland surface area resulting from lake shallowing, which in turn caused a decrease in total evaporation. It would also be expected that the residence time of water in both lakes would also have decreased due to the decreased volume. At Brown Lake, which has a larger lake surface and catchment area (Table 1), a period of greater hydrological stability appears to have developed between 2017 and 2019 suggesting that for this period at least, water inputs vs. evaporation were near equilibrium. This balance was interrupted by a period of high evaporation and low precipitation in early 2019 which resulted in further lake level decrease and an increase in δ¹⁸O_L at Brown Lake (Figure 4). By contrast, the continual decline in water depth at Swallow Lagoon suggests that, with a much smaller lake surface and catchment area, the flux of evaporative losses continued to exceed those from inflowing rainfall during the same period. This difference in hydrological stability between the lagoons also explains the larger seasonal range in δ¹⁸O at Swallow Lagoon, compared to Brown Lake (Figure 4).

To make informed assessment of the future risks to wetland hydrology and ecology it is necessary to be able to adequately model historical variability in these systems. In this study, it was possible to model 10 years of hydrological change at Swallow Lagoon and Brown Lake using a reasonably simple hydrological mass balance model (Lobegeiger et al., 2015). However, it was only possible to accurately simulate lake depth change at these wetlands by varying two key parameters through time: catchment runoff and throughflow (termed catchment area in the model; basically short term transfer of rainfall into the lake) and wetland outseepage rate (Figure 7). The need to vary the calibration parameters through time further highlights the significant complexity regarding the response of these wetlands and their surrounding catchments to climate variability. For the lacustrine wetlands, higher runoff (catchment area) parameterisation was required for periods of higher effective moisture and higher lake level, between 2012 and 2013 at both lakes, and also for 2015 at Swallow Lagoon. This requirement to parameterise larger runoff for these periods suggests that changes occurred in the seasonality or intensity of precipitation, or in catchment permeability (Potter et al., 2005). Lower permeability during times of higher effective moisture may be a consequence of saturated moisture content in the soils, or a higher organic concentration, particularly of leaf litter, due to higher terrestrial productivity. Therefore, major uncertainties remain regarding the way that water flows into these lakes during periods of higher precipitation—namely, the fraction of water that enters via runoff versus that which flows into the lakes via shallow groundwater. Improving our understanding of these processes is crucial in order to better manage landscapes to aid lake water recharge under future climate scenarios.

In addition to altering catchment run off behaviour, it was also necessary to vary the outseepage rate through time. For Brown Lake, it was clear that a higher outseepage rate was necessary at times of higher water depth (Figure 7). For Swallow Lagoon, a similar pattern emerged, albeit more complex. In general, however, it is apparent that periods of higher lake water depth are associated with a higher discharge of non-evaporative water losses, of which outseepage to groundwater is the most likely pathway. An increase in groundwater discharge during periods of higher lake level could be explained by either higher water pressures during deeper lake phases, or by the overtopping of impermeable layers at higher lake depths. An additional factor may be the uptake of water by catchment vegetation during periods of higher precipitation, which would also not result in changes to δ¹⁸O_L. In general, it is apparent that understanding the role of soil and lake sediment permeability, infiltration, outseepage and vegetation is crucial to understanding the trajectory of wetlands on North Stradbroke Island, and in similar environments, in response to future climate change.

Given the complexities related to lake hydrological responses to climate, and the consequent need to adjust model parameterisation through time, model projections of lake depth under future scenarios are currently uncertain. Nevertheless, it is feasible to use the model—as calibrated for the 2016–2019 shallow lake phase—to explore the potential resilience of the lake hydrology to an ongoing period of drought. A systematic analysis of total annual rainfall and total annual evaporation since 1890 at Dunwich Post Office (BoM 2020) indicated that the year 2019 was the driest year since records began (SI Figure 2). To explore the effect of simply continuing with the climate of 2019 over the coming decades, the calibrated model was run for both Swallow Lagoon and Brown Lake, based on a hypothetical repetition of 2019 climate until 2038 (Figure 8). Interestingly, neither lake dried up completely under this scenario: seasonal water depths at Swallow Lagoon remained consistent with the seasonal cycle of 2019–2020, whereas Brown Lake levels continued to decline but reached asymptote after 2024 (Figure 8). Given that the models assume no groundwater input and relatively minor catchment runoff, this stabilization of lake depths supports the interpretation that reducing water surface area during lake level decline causes a decrease in evaporative water loss, thus establishing an equilibrium with inputs derived from direct precipitation and runoff. This natural process of self-regulation is linked to the bathymetry of the lakes. Swallow Lagoon has a roughly conical bathymetry while Brown Lake has a number of deeper sub-basins which would retain water even with marked water level declines (Supplementary Figure S1). In addition, the model experiments suggest that a long-term decline in water depth in these lakes requires a long-term decrease in the precipitation to evaporation ratio, whereas a low but fixed precipitation to evaporation ratio might not dry the lakes beyond a certain level. Consequently, if precipitation were to decline over several years at or beyond 2019 levels, then it is feasible that Swallow Lagoon, in particular, might dry out entirely. Similarly, further reduction in the runoff and throughflow efficiency, as we have inferred for the last decade (Figure 7), could also accelerate the demise of these lakes. In order to make more reliable projections of future hydrological change, however, it is necessary to develop, calibrate and validate more sophisticated hydrological models which combine responsive lake water thermodynamics, surface-groundwater interaction and stable isotope geochemistry [e.g. (Hostetler and Bartlein, 1990; Jones and Imbers, 2010; Steinman et al., 2010; Ankor and Tyler, 2019). Such modelling exercises should be a priority in order to better understand the resilience of these important wetlands under future climate scenarios. Given that changes in water level of the magnitude observed in our study are likely to result in fundamental alterations to water quality (for example the extent of lake bottom anoxia), a greater understanding of the water quality and biological response to these alterations is also needed.

The wetlands of North Stradbroke Island offer significant value as palaeoclimate archives, being some of the oldest extant water bodies in Australia, and providing some of the most complete sedimentary records spanning the last ∼100,000 years (Tibby et al., 2017; Cadd et al., 2018; Kemp et al., 2020; Lewis et al., 2020). Our observations of the contemporary isotope hydrology of North Stradbroke Island wetlands provides a useful basis for the interpretation of palaeoclimate records from the sediments of these wetlands, and similar inter-dunal lake systems in the humid subtropics in Australia and worldwide. Importantly, the isotope hydrology of all the monitored wetlands between 2015 and 2019 was dominated by the balance between precipitation vs. evaporation, despite the relatively high rainfall in south-eastern Queensland and the porous geology, which promotes rapid recharge and short water residence times. The effects of evaporation on water isotopes were largely consistent between wetlands, irrespective of the surface area or depth of the water body, suggesting that similar processes can be used to explain isotope hydrological change throughout the Quaternary history of the wetlands. This is particularly relevant for Welsby and Fern Gully Lagoons, which are two of Australia’s oldest extant wetlands with sediments that have accumulated over tens to hundreds of thousands of years (Tibby et al., 2017). At Welsby Lagoon, due to progressive basin infilling by sediment, the wetland evolved from a deeper lake prior to 40,000 years ago, into a shallow palustrine wetland similar to the contemporary system (Cadd et al., 2018). The hydrology of ancient Welsby Lagoon is more likely to have been analogous to the contemporary Brown Lake, however our results indicate that the mode of interpretation of δ¹⁸O_L from the sediments of Welsby Lagoon can be reasonably consistent throughout.

Although the evaporation related patterns of change were similar between the four monitored wetlands, the magnitude of variability in δ¹⁸O_L varied significantly between sites due to differences in the surface area and volume of the wetland. Of the two lacustrine wetlands, Swallow Lagoon exhibited a larger range in δ¹⁸O_L, which can be explained by the smaller lake volume, lake surface and catchment area, compared to Brown Lake, meaning that the total influx of water was less and thus the effects of evaporation on δ¹⁸O were proportionately greater. The isotope hydrology of the two palustrine wetlands, Welsby Lagoon and Fern Gully Lagoon, was characterised by periods of recharge followed by subsequent evaporation and isotopic enrichment. In these cases, differences in the magnitude of δ¹⁸O_L change can be explained by differences in surface area, and thus the total evaporation rate.

In δ¹⁸O vs. δD space, each of the four wetlands exhibit local evaporation lines with similar slopes (Figure 5). The intersect of the LEL with the regional meteoric water line (which is defined by the δ¹⁸O and δD of precipitation), can be used to estimate the weighted isotopic composition of water recharging the wetlands. Interestingly, for Swallow Lagoon and Brown Lake, these recharge estimates fall close to the mean δ¹⁸O_P and δD_P for the monitored period, whereas at Welsby Lagoon and Fern Gully Lagoon, the estimated isotopic composition of recharge waters are closer to the lower, winter δ¹⁸O_P and δD_P values (Figure 5). One explanation for this is that Welsby and Fern Gully Lagoons have smaller water volumes and water residence times, thus their δ¹⁸O_L values are weighted towards recent precipitation events, which predominantly occurred in winter over the monitored period. By contrast, with longer water residence times in both the lakes and surrounding groundwater, Brown Lake and Swallow Lagoon receive input waters with an annually integrated δ¹⁸O. That these waters include a higher proportion of summer rainfall might imply that the catchment groundwaters are preferentially recharged by intense, summer rainfall. Alternatively, it is possible that the lake and ground water δ¹⁸O at the two lakes retain a signal from higher δ¹⁸O_P rainfall events which occurred prior to the commencement of monitoring. Irrespective of the causes, interpretation of ancient δ¹⁸O variability from wetland sediments must consider the possibility of variable seasonal bias depending on the prevailing annual distribution of rainfall, and depth and hydrology of the wetland at the time, whereby shallow palustrine wetlands and their sediments may record a more seasonally-biased signal of lake water isotopes compared to deeper lakes.

Similarly, the monitoring data provide some insights into the way that plant and animal tissue might accrue oxygen and isotope hydrological signatures, which are then preserved in the sediments. All of the wetlands host emergent aquatic plants, with roots that accumulate water from the top 1–2 m depth beneath the sediment surface, and this is particularly the case in the palustrine wetlands. The δ¹⁸O of plant remains in these wetlands, which are the major contributor of organic matter to the lake sediments, can be recorded in the sediments within cellulose, lipid biomarkers or the isotope composition of grazing invertebrates. It is therefore possible that such higher plant derived organic matter has a δ¹⁸O signal derived in part from sediment interstitial water or groundwater, contrasting with algal or bacterial organic matter, which should more faithfully record the signal of evaporated surface water. The δ¹⁸O of shallow groundwaters is fed by a combination of recharge direct from precipitation as well as outseepage from wetland surface water. Therefore, plant tissue δ¹⁸O may likely reflect a composite signal from these sources. Ongoing monitoring of plant matter in these environments aims to provide further insights into this relationship.

Differences in the sensitivity of wetland isotope hydrology to seasonal and inter-annual climate variability (Figure 6) raises the enticing potential to combine palaeo- δ¹⁸O data from multiple sites within close proximity to solve for regional hydrological fluxes, namely precipitation amount vs. evaporation. This potential is particularly promising for perennial lake systems such as Swallow Lagoon and Brown Lake. However, such an approach also comes with significant caveats, most notably that basin morphology also varies as a function of lake water depth, which in turn affects lake recharge, evaporation rate and outseepage. In addition, as indicated by our modelling (Figure 7), changes in wetland outseepage and catchment runoff can also vary as a function of climate change. Therefore, to use between-lake differences in δ¹⁸O to quantify past hydrological change requires the development and application of more detailed isotope-hydrological models which incorporate variable lake bathymetry, informed by geophysical profiles of the basin’s sedimentary fill, as well as accurate estimates of sediment accumulation through time. The development of such data and models for inter-dunal wetlands represents an important objective for future research.