A combined-method approach to trace submarine groundwater discharge from a coastal karst aquifer in Ireland

The groundwater catchment of Bell Harbour (Fig. 1)—as previously adopted by McCormack et al. (2017); Schuler et al. (2018)—is located in the north-eastern part of the Burren Limestone Plateau in the west of the Republic of Ireland. The Burren, including the upland catchment of Bell Harbour, is a temperate glaciokarst landscape, which has been subject to repeated glaciation during the Pleistocene, showing features typical of glaciation such as ice-plucked crags, scoured rock surfaces, limestone pavements and erratic boulders (Simms 2014). The mean annual air temperature is 13.6 °C and the annual rainfall was 1,386 and 1,560 mm for the hydrological years 2017 and 2018, respectively, as measured at the weather station C1 (Fig. 1). Multiple valleys intersect the study area that range in elevation between sea level in the north and up to 340 m above sea level (masl) along the escarpments. Along the escarpments the bare outcrop shows high degrees of karstification.

The entire area is underlain by Lower Carboniferous well-bedded and pure limestones ranging between the Tubber formation (early Viséan), the Lower Burren and Upper Burren formation (mid-Viséan, Asbian), and the Slievenaglasha Formation (late Viséan, Brigantian). The presence of irregular limestone surfaces is interpreted as palaeokarst (Pracht et al. 2015). The total thickness of limestone reaches at least 510 m. The strata dip generally to the south, ranging between 2 and 3°.

The deformation of Carboniferous rocks in Ireland is largely attributed to the Variscan orogeny and the associated north/northwest-oriented compression (Graham 2009). The Variscan contractional deformation caused the formation of calcite veins of a few microns to 0.5 m width, which are laterally and vertically consistent across bedding discontinuities (Gillespie et al. 2001).

A suite of veins with more exotic mineralisation, including fluorite, quartz and occasional sulphides, occurs in a ~10-km wide north–south zone centred near Carran (see Fig. 1). These generally wider veins range from 0.05 to 1 m, and are part of a structure which extends 20 km across all stratigraphic sequences of the Burren (J. Walsh et al., University College Dublin, unpublished paper, 2019). The veins are visible in outcrop, and have been observed at more than 100 m depth, for example along the vertically oriented cave Poll Gonzo (Bunce 2010). The horizontal persistency of veins along strikes spans over 7 km, interconnecting caves (Mac Sharry 2006). Another example is the recently discovered sinkhole Deelin Pot, which is a 18 m deep and 5-m-wide excavated shaft oriented along north–south oriented calcite veins containing also silica veining, previously filled with glacial debris (Bunce and Drew 2017). Furthermore, slickenside veins were recorded at several depths in a 490-m deep borehole drilled by the Geological Survey Ireland (GSI) at location BH1 (Fig. 1).

The veins may be dissolved out or still intact. Where intact, the contact between limestone host rock and the (calcite) vein material may act as an inception horizon (Lowe and Waters 2014) and pathway of preferential dissolution of the limestone rock. The presence of sulphide minerals infill provides a source of sulphur for the formation of sulphuric acids to increase the rate of dissolution. Therefore, veins must be considered important for regional groundwater flow, as their lateral and vertical extent is believed to have largely contributed to north-south groundwater flow paths (Moore and Walsh 2013) and (J. Walsh et al., University College Dublin, unpublished paper, 2019). The joints pattern is very prominent, and it is known to extend several kilometres off-shore, at least to 80 m below sea level (mbsl; Kozich and Sautter 2018).

Offshore sinkholes are known to exist within Bell Harbour Bay. The use of marine electrical resistivity tomography (ERT) confirmed that these sinkholes are partly hydraulically activated, hence, acting as submarine springs. These sinkholes are generally filled with sediments, yet, there are single examples that indicate the absence of a sediment infill (O’Connell et al. 2018). The geology and structure of the telogenetic karst Burren Plateau extends several kilometres off-shore west of the Burren (Gillespie and Sautter 2018), and therefore sinkholes may accompany these structures, yet, to the knowledge of the authors, no detailed studies have yet confirmed this.

Approximately 60% of the Burren Plateau drains directly to the sea, and significant quantities of groundwater discharge occurs along the shore via intertidal springs (Drew 2018). The intertidal springs that are relevant to this study were previously traced successfully (Table 1) and are illustrated in Fig. 1. All tracer tests were carried out qualitatively; hence, the average flow velocities are approximated and range between 21 and 460 m/h. For the Bell Harbour catchment, only one successful tracer test was carried out over a distance of 2.2 km.

Known locations of SGD are either in conjunction with intertidal springs along the western shore of the Burren Plateau or limited to Bell Harbour and Kinvara Bay (Drew 2003; O’Connell et al. 2018) with estimated discharge rates ranging between 0 and 4.3 m³/s, and 5–16 m³/s respectively (McCormack et al. 2014; Schuler et al. 2018). On a larger scale, significant areas of sea-surface-temperature anomalies interpreted as SGD were detected on the western side of the Burren Plateau linked to structural geology (Wilson and Rocha 2012). Due to the scale of these temperature anomalies, the rate of SGD may exceed the abovementioned numbers.

Within the Bell Harbour catchment, the groundwater flow is a conduit-dominated updip from south to north (McCormack et al. 2017). Discharge of the study area occurs via SiGD, including the intertidal spring Pouldoody, into Bell Harbour Bay. A shallow north–south conduit of ~2 m diameter is argued to link shallow flow within the catchment and springs in Bell Harbour Bay (McCormack et al. 2017; Fig. 1). The SiGD regime fluctuates seasonally corresponding to the overall piezometric state of the aquifer driving an overflow mechanism. Previous tracer tests suggest that groundwater flow velocities may reach up to 460 m/h (Table 1). Vertical groundwater flow was measured up to 176 mbsl in BH1 reaching up to 97 m/h. This deep flow was hypothesised to drain the coastal aquifer as SGD into Galway Bay as an additional outlet to the SiGD into Bell Harbour Bay (Schuler et al. 2018), with one goal of this study to provide confirmation by artificial tracer tests.

The high degree of karstification limits surface-water features to short reaches of ephemeral streams (Drew 1990). South of the study area and the Burren Plateau, the Fergus River drains large areas from the southern part of the Burren. Recharge in the study area originates from rainfall, percolating rapidly towards the phreatic zone based on the response of borehole hydrographs (Schuler et al. 2018). Recharge is assumed to get concentrated along preferential flow paths, as for example within Poll Gonzo, where an underground stream rapidly falls 40 m until reaching the permanent local water table at ~31 masl (Bunce 2010). Two tracer studies were conducted, in 2011 and 2015, using 5 kg of uranine and 4 kg of rhodamine WT, respectively, to trace flow paths from Poll Gonzo, but without success. In addition to the tracer tests illustrated in Fig. 1, 11 tracer tests have been carried out in the uplands of the Burren Plateau. Four of these tests investigated a hydraulic connection to the shore, yet, none proved confirmatory (Bunce and Drew 2017). Relatively fast, concentrated and mainly deep groundwater flow may have bypassed any tracer from shallow sampling locations. Furthermore, it is assumed that large dissolution features such as Deelin Pot must be well connected to such deep conduit network.

Within the study area, different sampling sites were set up to record hydroclimatic data. A full weather station at C1 (38 masl, Fig. 1) records rainfall in 15-min intervals using a CEL tipping-bucket rain gauge (Casella, Bedford, UK) attached to a Rainlogger Model 3002 (Solinst Canada Ltd., Ontario, Canada) data logger, wind speed in 15-min intervals using an ultrasonic wind sensor (Gill Instruments Ltd) attached to a CR800 data logger (Campbell Scientific Ltd., Shepshed, UK), and atmospheric pressure in 1-h intervals using an INW PT2X sensor (Seametrics, Kent, WA, USA). The water depth of the Fergus River at FR (~23 masl, Fig. 1) was monitored at 1-h intervals using a Mini-Diver502 (SchlumbergerWater Services, British Colombia, Canada). A stage-discharge curve was established through 13 spot discharge measurements using an OTT acoustic-digital-current meter (OTT Hydromet GmbH, Kempten, Germany). Temperature, level and electrical conductivity (temperature compensated) of intertidal groundwater discharge at Pouldoody spring at SiGD1 (−0.1 masl, Fig. 1) were measured in 1-h intervals using a Solinst LTC sensor (Solinst Canada Ltd., Canada). The LTC sampler at SiGD was fixed on a concrete platform placed into the near-shore sediments 1-2 m away and ~1 m below the spring outlet. A second Solinst LTC was used to monitor temperature, level and electrical conductivity (temperature compensated) in the sea onboard of a vessel. Bathymetry data used in this study comprise a 5- and 10-m grid provided by the GSI. Land topography is based on the Shuttle Radar Topography Mission (SRTM; USGS 2014).

Two types of tracers were used for this study: (1) floating biodegradable wood chips, and (2) fluorescent dyes in the form of the conservative tracer uranine (Acid Yellow 73) and the nonconservative tracer rhodamine WT (Acid Red; Field et al. 1995; Leibundgut et al. 2009). Sampling of fluorescent dyes follows the principle of excitation of a water sample and measuring the ensuing emission intensity. The fluorescence of a dye tracer is pH-dependent, and not considered to be a factor in this environment where the pH is approximately 7. Solubility is 300 g/L for uranine compared to 3-20 g/L for rhodamine WT (Leibundgut et al. 2009). Excitation and emission maxima for uranine are 491 and 516 nm respectively, and 561 and 586 nm for rhodamine WT.

The sorption behaviour of tracers is particularly important with regard to the quantitative analysis. In general, the higher the solubility in water, the lower the sorption capacity of the substance. Rhodamine WT shows a strong sorption behaviour, while uranine has a very low sorption tendency (Field 2002) which applies to the given environment.

Given the uncertainty of SGD locations in the sea, the suitability of floating wood chips as a natural and biodegradable tracer was tested. The reasoning for using wood chips was to mitigate the problem of spatial integration (dispersion) related to solute tracers: there is no detection limit as the wood chips are measured by visual observation. Hence, a single detected floating wood chip may be sufficient to confirm the success of the tracer test. The ideal solid tracer would have a density between freshwater (1 kg/m³) and seawater (~1.035 kg/m³) in order to be neutrally buoyant within the freshwater underground river and conduits of the aquifer before being discharged into the sea where it would float up to the sea surface and thereby be visually detectable.

The buoyancy of five different types of wood was tested in a stirring freshwater tank over the period of 10 days, ranging in density from 660 to 960 kg/m³ (TRADA 2018). The chips tested were of the size 4 × 40 × 50 mm. The best performance in terms of neutral buoyancy in freshwater was achieved by Keruing (Dipterocarpus spp.). When salinity was increased, the buoyancy increased. A total of 250-L of Keruing chips were manually produced, soaked in freshwater and packed in 10 × 30 L caving bags. The concentration of wood chips >2 mm grain size was counted as >1,000 parts per 50 ml—ergo, the total amount of chips (>2 mm) for the volume of 250 l was estimated at 5.71 million.

The geophysical analysis of the submarine sinkholes considered as potential SGD locations, show both the presence and absence of a sediment infill (O’Connell et al. 2018). Such sediments might play a role in blocking or attenuating the solid wood, which needs to be considered in the interpretation of results.

The two fluorescent dyes and wood chips were injected at three different sites on 14 April 2018: two injection sites located within the cave Poll Gonzo with the entrance 7.5 km south of Bell Harbour Bay and the openly accessible sinkhole Deelin Pot 5.9 km south of Bell Harbour Bay.

Access to Poll Gonzo required a team of six cavers to abseil three pitches of ~20 m each on fixed ropes. The uranine was dissolved in-situ within a 100-L inflatable swimming pool and released into the underground stream upstream of a waterfall at ~70 masl and 56 m below ground level (Table 2, site 1a).

The 250 L of wood chips were released at the bottom of the cave at ~31 masl at the local water table, where three experienced speleologists abseiled down another pitch with difficult access (Table 2, site 1b). Rhodamine WT was injected into the bottom of Deelin Pot 1.67 km north-northwest of Poll Gonzo, where a 1 m³ carbon-fibre tank with a release valve at its bottom was installed to mix the dye in-situ (Table 2, site 2). The total volume of water used for dilution and flushing was 2.7 m³.

Sampling for the tracers at potential outlets was conducted at terrestrial locations, as well as offshore along transects to account for the numerous potential SGD locations in Galway Bay. The terrestrial outlets were sampled using a fluorometer and/or activated charcoal kept in place for the period of observation. Detection of fluorescent tracers was done qualitatively using permeable nylon bags of activated charcoal pellets (technical grade, AppliChem Panreac). Each bag was filled with ~20-g pellets.

For off-shore sampling, GGUN-FL30 field fluorometers (Albillia Co., Switzerland) were used to (semi-) quantitatively analyse the fluorescence. GGUN-FL field fluorometers can measure three distinct dyes in parallel as well as turbidity at a sampling interval ≥10 s. Readings in mV are converted into ppb using a linear calibration to three tracer standards (1, 10 and 100 ppb) for each tracer made using local sea water and dye from the batch injected; hence, background fluorescence in the sea is broadly accounted for. The minimum detection limit for uranine in clear water is 0.02 ppb, and ≤0.2 ppb for rhodamine WT (Schnegg 2002). Visual monitoring for wood chips was conducted offshore from onboard of the vessel.

Stationary sampling was conducted at 10 different sites (Table 3; Fig. 3). A field fluorometer was installed at the Fergus River to assess the unlikely scenario of southwards transport. A second field fluorometer was installed at Pouldoody spring, along with a conductivity-temperature-depth (CTD) sensor to record the tidal fluctuation, temperature and conductivity as indicators for the occurrence of SiGD at Pouldoody spring.

Qualitative sampling methods using activated charcoal bags varied by site. Within the open borehole (BH1), four bags were tied on a 200-m-long bottom-weighted rope hung in the borehole to sample depths ranging from 15 to −170 masl. The six known intertidal coastal springs were monitored with individual bags tied to concrete slabs at the intertidal outlets: Ballyvaughan West and East, Corranroo West, Toberbreen and Kinvara West and East. Furthermore, two charcoal bags were attached to a floating buoy anchored in the seabed in the centre of the outlet of Bell Harbour Bay (Fig. 1), sampling the top sea water column. The high tidal range largely empties Bell Harbour Bay during ebb tide, such that fluorescence may have been detected at that site. The charcoal bags were collected on 19 April and the replacement set collected on 25 April.

Mobile sampling of sea surface fluorescence and conductivity over transects was conducted over 21 h on four successive days from a vessel in the sea targeting the areas of temperature anomalies previously identified by remote sensing. Sampling focused on the central and western area, as the temperature anomalies were considered to be more explicit than those of the east, which may be due to freshwater dilutions from Kinvara Bay/surface water. Based on estimated travel times (Table 1), sampling started 36 h after injection and lasted for 4 days. The fluorometer was tied on ropes pulled by the vessel, while also cabled to the data logger and a computer on board. The ropes were weighted to make sure the fluorometer did not float, sampling between 0.03 and 5.77 m (average 1.01 m) below the sea surface. In addition, the speed of the vessel was very slow at 3–5 km/h to reduce perturbation and erroneous readings due to air bubbles from the propeller. A CTD sensor was attached close to the fluorometer to sample for depth, temperature, and conductivity as indicator for SGD. A hand-held global positioning system (GPS; GPSmap 60CSx, Garmin, KS, USA) was synchronised with the fluorometer and CTD sensor, recording the geographic position of the vessel in 3-s intervals.

Charcoal bags were collected on 19 and 25 April 2018, 5 and 11 days after the injections, respectively. All sample bags were individually sealed and kept protected from sunlight in a closed container. The charcoal samples were analysed within days following collection. The Smart Solution was used for rhodamine WT consisting of 1-propanol, deionized water and 30% NH₄OH in the ratio 5:3:2 (Smart and Simpson 2002). For uranine, the eluant was a mixture of 70% isopropyl alcohol, 30% high purity deionized water and 10 g/L sodium hydroxide. The analysis was carried out using a Cary Eclipse Fluorescence Spectrophotometer (Agilent Technologies, Santa Clara, CA, USA) based in the environmental laboratory of the GSI in Dublin, Ireland. The samples were analysed in sequence; hence, the elution time of a charcoal sampler ranged from at least one, to a maximum of 2 h.

In contrast to the rather clean continuous signal achievable with terrestrial stationary monitoring, sampling in the sea is impacted by perturbation through currents, drag of the boat, and air bubbles causing potentially erroneous readings. Hence, potentially erroneous readings were filtered following data acquisition. The filter compares the ratio of observed mV signals of given tracers with the ratio of mV signals of the same tracers as established during the calibration procedure using the values of the corresponding photodetectors which were L1 and L2 for the two tracers rhodamine and uranine, only the ratios of L1 and L2 were recorded in the raw files.

First, the ratio of measured mV signals for a tracer T_n of interest is established using the measured signals of L1 during calibration L1_cal [mV] and L2 during calibration L2_cal [mV]. The calibration ratio r₀ is then given by Eq. (1),with the associated signal for pure water W [mV] and the tracer with n = 1 for uranine and n = 2 for rhodamine.

The result is a characteristic ratio for each of the three tracers that can be monitored, corrected by subtracting the background signal of water. The actual observed ratio between photodetector 1 L1_obs [mV] and photodetector 2 L2_obs [mV] at time step t is established to yield r_t following Eq. (2)again, corrected for the background value of water.

Ideally, \( {r}_t={r}_{0,{T}_n} \) but this is never exactly the case. Since L2 is measured 100 ms after L1, the measurement is rejected if a large deviation of r_t from \( {r}_{0,{T}_n} \) is observed, indicating the occurrence of air bubbles during the measurement of L1 or L2. Here, an observation was considered reliable if its r_t ranges between the lower and upper limit of the ratios of the two tracers considered, which are r₁ = 0.035 for uranine and r₂ = 2.597 for rhodamine. In addition, a threshold deviation of ±20% was defined as acceptable so that \( 0.8\times {r}_{0,{T}_1}<{r}_{t,{T}_n}<1.2\times {r}_{0,{T}_2} \).

Because of its nature, the method is called ‘quotient method’. This approach was considered to be very conservative in order to reliably define the validity of a single tracer reading following a transparent filtering sequence.

None of the emission spectra of the eluant from charcoal samples collected on 19 and 25 April show emission peaks of 516 nm for uranine (Fig. 4a,b), nor an emission peak of 586 nm for rhodamine (Fig. 4c,d) samples. All charcoal samples are interpreted to be free of any of the two dyes.

The time series fluorometry (Fig. 5a) shows the results obtained from Pouldoody spring, along with water level fluctuations measured at the monitoring site. The obvious twice-diurnal water-level fluctuation is a function of the tidal cycles. Further, the fluctuation of conductivity indicates active discharge of the spring with notable minimum values of up to 2.9 mS/cm during low water levels starting on 17 April. Groundwater discharge is at maximum when the observed EC is at its minimum as a result of the low water level (ebb tide). At such times, the recorded concentrations of rhodamine and uranine peak <0.3 ppb and slightly above 1 ppb, respectively. The impact of the tidal fluctuation on the readings of both dyes is believed to be the result of changing chemistry and composition of the water at the site, probably dissolved organic matter (DOM). As a consequence of these low concentrations, the results are interpreted as negative.

Figure 5b shows the result of the field fluorometer installed at the Fergus River to the south of the tracer injection points. The river flow exhibits a clear response to the two preceding rainfall events of 11.6 mm (15 April 3:00 to 16 April 1:00) and 14 mm (16 April 14:00 to 17 April 4:00). As a result, discharge increases from 1.6 to 12.8 m³/s along with concurrent increases in turbidity. Uranine and rhodamine show both a correlation with turbidity peaking at 2 and 3 ppb respectively. Correlation between fluorescent dyes and turbidity within this range is common (Schnegg 2002). In addition, measured fluorescence exhibits a negative correlation with the temperature in the stream; hence, altogether, the observed dye concentrations are interpreted as a mixture of natural background and/or anthropogenic sources.

Figure 6a,b shows the concentrations of uranine and rhodamine for the filtered data in semi-log scale along with measured conductivity in the sea, depth of sampling and hourly rainfall. Conductivity records are missing for day 1 (16 April). Monitoring was limited to the eastern transect on day 1 due to very harsh weather conditions and a red weather warning for Galway Bay issued by the Irish meteorological service MetEireann. During days 2–3, sampling focussed to the centre and the western transect, for the reasons previously discussed, and that more elevated tracer readings were noticed in real-time in the centre and the west on day 2.

The raw data show 4,998 readings with maximum values for rhodamine and uranine reaching 208.6, 457.8 ppb respectively (Fig. 6b). The quotient method filtered 12.7% potentially erroneous readings, leaving 4,365 readings or 87.3% to be considered (Fig. 6b). As a result, the maximum concentrations of rhodamine and uranine considered reliable are 29.3 and 50.7 ppb respectively.

Both maximum dye concentrations were recorded on sampling day 2 (17 April) which correlate with the lowest conductivity <5 mS/cm values also recorded on day 2. Periods of low conductivity and high tracer readings are highlighted in pink rectangular dashed boxes (day 2: 15:05–15:20; day 4: 11:15–11:28). Importantly, the fluorometer and CTD sensor were very shallow during these two periods (0.18, 0.23 m). The average sampling depths on days 1, 2 and 3 are 0.81, 1.20 and 0.96 m, respectively (average of 1.01 m). Low conductivity levels may be linked to the two rainfall events between 15 April and 17 April with a total of 22.6 mm. It is hypothesised that the rapid recharge increased SGD with the resultant lowering of conductivity in the sea at these locations. In comparison, the conductivity on the 18 and 19 of April, during which there was no rainfall, are much more homogeneous and relatively constant. It is assumed that without rainfall, the rate of SGD is lower. No correlation between conductivity and tracer concentration occurred on day 3 with the conductivity readings remaining high at 50 mS/cm, with exceptions.

The fact that high tracer readings correlate in time and space with low EC readings suggest consistency of the method. Overall, it was assumed that a large amount of dilution weakens and spatially integrates the signal, and therefore a clear correlation cannot be expected. In order to evaluate the spatial pattern of recorded values, the georeferenced and filtered ppb readings for rhodamine and uranine were mapped, along with monitored conductivity of the sea water and the total track recorded by the GPS (Fig. 7). Both dyes are plotted for values >2 ppb in order to not overload the map with low (and presumably not meaningful) concentrations.

In general, tracer was recovered in a relatively large geographical area, mostly in three clusters: cluster 1 in the west, cluster 2 in the centre and cluster 3 in the east. In all clusters, both dyes were recovered. Altogether, a spatial correlation between dye readings and conductivity occurred. Cluster 1 contains rhodamine and uranine readings of days 2–4. This cluster contains the maximum observations for rhodamine and uranine reaching 29.3 and 50.7 ppb respectively, associated with low conductivity readings down to 3.4 mS/cm on day 2 (Fig. 6, *1) corresponding to travel times of ~160 and 165 m/h for rhodamine and uranine. Also, Cluster 2 relates mainly to the highly correlated low conductivity and high tracer readings observed on Day 2 (Fig. 6, *1) with associated travel times of ~>181 and 194 m/h for rhodamine and uranine. These two areas are considered as SGD ‘hot spots’.

The longest distance between injection sites and sampling locations occurred in cluster 3, where uranine was sampled 15.8 km away from Poll Gonzo and rhodamine 14. 6 km away from Deelin Pot, which correspond to travel times of ~331 and 339 m/h. No wood chips were observed during offshore monitoring. During the first 3 days of sampling, the sea was very rough due to strong winds with maximum average hourly wind speeds of 8–11 m/s. These conditions made the work onboard difficult and monitoring of wood chips on the sea surface challenging. On day 4, the sea became very calm, but no wood chips were seen or sampled; therefore, the result of the tracer test using wood chips is considered as inconclusive/negative.