Fingerprinting dissolved organic compounds: a potential tool for identifying the surface infiltration environments of meteoric groundwaters Open Access

The chemical composition of groundwater, due to both natural and man-made processes, has long been used for tracing the origins and ages of subsurface waters. Groundwater tracers generally fall into three types: (1) additive tracers/point-source pollutants (Flury and Wai 2003; Abrantes et al. 2018), which disperse rapidly and may be unrecoverable (Filippini et al. 2018), and hence can only be used to trace flow over short distances and timescales; (2) tritium, helium-4 and CFCs, which can be related to specific time points (the bomb pulse tests in the 1950s and the banning of CFCs in the 1980s) to determine the age of modern meteoric groundwaters (Casillas-Trasvina et al. 2022; Okuhata et al. 2022), although tritium concentrations have now decayed to barely detectible levels; and (3) isotope ratios, which can be used to determine the original altitude of groundwater infiltration (e.g. $δ$¹⁸O and $δ$D isotopes) (Prada et al. 2016; Schneeberger et al. 2017; Fackrell et al. 2020), the presence of differing hydrothermal and lithological water sources (e.g. $δ$¹⁸O(SO₄), $δ$³⁴S(SO₄), $δ$³⁷Cl, ³H, $δ$¹⁴C and ⁸⁷Sr) (Pichler 2005; Osman Awaleh et al. 2020), and the mixed origins of old groundwaters based on age (e.g. ³He, ³⁹Ar, ⁸¹Kr and ⁸⁵Kr) (Kralik et al. 2014; Gerber et al. 2017; Avrahamov et al. 2018). Vascular plant biomarkers can be used to determine the origin of dissolved organic matter from near-surface environments (Shen et al. 2015). Most hydrogeological studies use a combination of several tracing techniques to reduce uncertainty in the determination of groundwater origins. Using existing techniques, meteoric waters can be identified, their ages predicted and, where topographical elevation varies, the altitude of infiltration can be estimated. However, for most groundwaters the type of near-surface infiltration environment (e.g. surface-soil type, river bed or lake bed) through which the groundwater infiltrated cannot be reliably determined.

Natural dissolved organic compounds are not routinely used for groundwater tracing, although they are prevalent in all aquatic environments, including groundwater systems. Dissolved organic compounds are input into the groundwater system through the breakdown of solid organic matter in the form of organisms (Shen et al. 2015) and soil/plant litter (Baker et al. 2000), or through the transport of water-soluble organic compounds and pollutants (Khatri and Tyagi 2015) from the surface. Other sources within groundwater include organic matter within the host rock and excretions from subsurface micro-organisms. Through time phyto-, microbial- and chemical- degradation (Chen et al. 2010; Obernosterer and Benner 2004; Zhang et al. 2009) result in a breakdown of larger solid organic compounds into a series of smaller water-soluble compounds. Breakdown of solid organic matter is rapid at the surface where oxygen is readily available to aid in biotic degradation (Keiluweit et al. 2016). Infiltrating meteoric water carries these water-soluble organic compounds into groundwater. Once in the groundwater, decay continues to alter organic compound structure and composition but the rate of decay decreases substantially with increasing depth, due to the increasingly anaerobic conditions (Kortelainen and Karhu 2006). Hence, groundwaters contain a complex array of preserved dissolved organic compounds.

The tracing of groundwater using dissolved organic carbon has seen previous success using a targeted approach (Derrien et al. 2017). Until recent years, the detection, measurement and comparison of the complete organic molecular composition of a water sample, and the relative abundance of individual molecules through a non-targeted approach, has not been readily achievable. However, with the advent of 2D gas chromatography time of flight mass spectrometry (GC × GC-ToF-MS) (Patrushev 2015), detailed organic fingerprinting of water samples is now possible. GC × GC is largely confined to the fields of environmental forensics, where it is used as legal evidence of the relative contributions of individual polluters (McGregor et al. 2012; Amaral et al. 2020), and medical sciences, where minute changes in the organic composition of bodily fluids could provide an early indication of disease (Almstetter et al. 2012).

In this research, we show that differing near-surface infiltration environments (individual soil types and lakes) have distinct dissolved organic signatures and that these signatures can be detected within groundwater samples at depth. We collected samples from a number of surface sites and groundwater samples from multiple boreholes at the Grimsel Test Site, Switzerland. These signatures were then compared visually using principal component analysis (PCA) and placed into statistically significant groups identified through hierarchical cluster analysis (HCA). We show that specific near-surface infiltration environments have differing organic signatures. Further, these organic signatures are distinguishable in groundwater samples at depth and, hence, could be used to indicate the predominant near-surface infiltration environment. Our research demonstrates that organic fingerprinting may prove a useful investigative tool for distinguishing the dominant near-surface infiltration environment(s) through which individual groundwater samples have infiltrated.

The Grimsel Test Site (GTS) is located in the Hasli Valley in the Canton of Bern (Switzerland), and comprises a series of access tunnels and groundwater monitoring boreholes (marked in black in Fig. 1). The entrance tunnel is at the base of the reservoir dam and the GTS site is c. 30 m below the reservoir bed, and between 200 and 500 m below the ground surface (which slopes steeply downwards from south to north). Boreholes and tunnels cut two lithologies: the Central Aar Granite (CAGr) to the north and the Grimsel Granodiorite (GrGr) to the south. The fracture network comprises open (unfilled) and gouge-filled fractures; fracture flow dominates the groundwater system at the GTS (Schneeberger et al. 2016).

At the GTS, potential inflows into the groundwater system are from the infiltration of precipitation (rainfall or snow melt). Meteoric waters infiltrate through surface soils, mountain stream beds and from the reservoir beds (where the reservoir level is above the adjacent groundwater head). We hypothesized that different near-surface water infiltration environments (soils and lake water) at the GTS could give rise to different exposure to potential organic solutes. The surface exposure directly above the GTS varies in slope gradient, orientation and altitude, and largely comprises weathered granite with soil and vegetation-filled fissures. Where soil is present, there are clearly visible spatial variations in soil type. On the east-facing mountain side overlying the GTS there are also several ephemeral streams (Fig. 1a). In addition to soil cover, there are areas of exposed granitoid rock and scree/boulder-covered slopes. Exfoliation fractures, topographical stress fractures and near-vertical tectonic fracture sets cut the surface topography, providing potential surface-water infiltration sites into the groundwater system encompassing the GTS.

Further potential sources of groundwater at the GTS are the surface-water reservoirs. Immediately to the east and south of the GTS, there are two hydropower reservoirs fed by surface runoff and glacial melt (Fig. 1a). These reservoirs are part of a regional pump-storage hydropower network containing multiple reservoirs draining different surface-water catchments. The network of reservoirs is connected by a series of tunnels, pipes and river systems, and reservoir water is regularly pumped both up and down the network, resulting in a well-mixed water body.

Field sampling took place in August 2018. Surface-soil sites for the sampling of soil organic material were severely restricted by the steep topography of the mountainous slopes above the GTS. Four surface-soil sample sites were selected (sites 1–4 in Fig. 1a) that, as far as possible, describe a north–south transect above the GTS and encompass the visibly different soil/sediment types above the GTS. At each site, two samples were collected c. 10 m apart (labelled a and b) to examine the variability of the organic signature at each location. Two aliquots (i and ii) of each soil sample (a and b) from each location (1–4) were subsequently extracted and analysed for their organic signature (see the following ‘Sample preparation’ and ‘Organic analysis’ subsections). Soil samples displayed clear differences based on a visual inspection of the soil. Locations 1 and 2 also contained some visible differences between the duplicate samples taken c. 10 m apart. Figure 2 shows images of the flora and fauna at each location, and circular markers identify the approximate location where each sample was extracted. Soil at site 1 was dark, waterlogged and had very little sand/gravel content. Soil at site 2 was brown, not as waterlogged as site 1 and contained fragments of roots/plant matter. Soil at site 3 consisted mostly of granite particles and was light brown in colour. Site 4, an ephemeral stream bed, mostly contained angular granitoid rock fragments and fine rock flour.

Lake water was sampled (Fig. 1a) at the one location where the water was safely accessible, and where the predominant SW–NE fracture set within the GTS (Fig. 1b) might plausibly intersect the lake. Due to the highly connected nature of the pump-storage hydropower system (which pumps water between the higher and lower reservoirs), any sample is likely to represent an integrated mixture of surface runoff and glacial meltwaters from both upstream and downstream in the hydropower network. Surface soils were collected using clean metal trowels and placed into aluminium foil parcels. Approximately 1 kg of soil was collected per sample, and the samples were double wrapped and stored at 4°C for transport (below ambient soil temperature at the time of sampling) until sample preparation could take place.

Groundwater was sampled from several locations within the GTS (D, E, F, G, H, I and M in Fig. 1b) from four horizontal boreholes labelled B1–B4 (Fig. 1b) and one vertical borehole B5 drilled from the subsurface gallery upwards towards the surface. All boreholes are fitted with isolated packer systems, integrated with water-sampling flow lines. Volumes of groundwater were sampled from seven individually packed intervals within each borehole (Fig. 1b), which allows for the sampling of water that inflows into a specific section of each borehole. Groundwater sampling intervals (labelled alphanumerically) were chosen to sample the different host-rock lithologies and structural geological features, as well as providing spatial coverage across the GTS; locations D–H are in the Central Aar Granite (CAGr), location I sits in the transition zone between the CAGr and the Grimsel Granodiorite (GrGr), and location M lies in the GrGr. Table 1 describes which sample location (D–M) corresponds to which borehole (B1–B5), the altitude of each sample location and the hydraulic head at each sample location. Groundwater sample locations D, E and F are all located within borehole B1. Prior to groundwater sampling, each borehole interval was drained three times to flush out the volume of the borehole sampling interval itself and the sample lines. Draining of the borehole intervals was carried out to remove any water that had been in contact with plastic in the packer system and to ensure that only the formation water was sampled. Groundwater was used to flush the 125 ml sample bottle three times. Samples were collected and sealed under water with PTFE foil-lined caps (US Geological Survey 2006). Groundwater samples were collected in triplicate for GC × GC and duplicated for CFC analysis; however, during shipping from Switzerland to the UK several samples were smashed or spoiled. As a result, only two samples remained from the lake water, I, F, D, E and H, and only one of the samples taken from G and M were preserved for testing. Thus, we are only able to present the groundwater data analysed by GC × GC either in duplicate for most locations, or as a single sample where only one sample survived. Duplicate samples that have been extracted and analysed by GC × GC are indicated in Roman numerals after the sample location in both the groundwater and lake-water samples (i.e. Di and Dii represent two separate water samples taken from sample location D, extracted and analysed by GC × GC).

Physical and chemical parameters (electrical conductivity, pH, redox potential, dissolved oxygen and temperature) were measured in situ during sampling using a flow cell and multiparameter probe (YSI Pro Plus multimeter). Water samples for dissolved ion analysis were filtered using 0.45 µm cellulose acetate filters and stored in 50 ml HDPE centrifuge tubes, with acidified and non-acidified portions stored in a dark fridge and analysed for dissolved ions at the University of Strathclyde within 14 days of sampling, with the exception of alkalinity which was measured on the day of sampling using a HACH digital alkalinity titrator. The methodology used for the sampling and testing of water samples for dissolved inorganic chemistry in this study is described in Stillings et al. (2021).

Water samples for 2D gas chromatography (GC × GC) and chlorofluorocarbon (CFC) analysis were collected in 125 ml Boston rounds (borosilicate glass) with foil cap liners. CFC samples were taken according to IAEA (2006) glass bottle collection method 2, and were analysed for CFC-11 and CFC-12 by the British Geological Survey (BGS). The infiltration date and apparent groundwater age were then calculated based on a piston flow model (IAEA 2006) as previously used in the calculation of tritium ages at the GTS (Keppler 1996; Schneeberger et al. 2017).

Each whole soil sample (c. 1 kg) was freeze dried and homogenized using a pestle and mortar (washed three times with acetone and a further three times with dichloromethane), then extracted with dichloromethane (DCM) : methanol (MeOH) (9 : 1, v : v) using Accelerated Solvent Extractor (ASE) 350 (Dionex) (US EPA Method 3545A (SW-846): US EPA 2017). ASE extraction cells were packed with 2 g of sample and filled with clean sand heated to 550°C for 8 h. Ground and lake-water samples were extracted by separatory funnel liquid–liquid extraction, using the US EPA Method 3510C as a guideline (US EPA 1996). It was not feasible to transport large volumes of groundwater, so extraction was carried out on a reduced 125 ml sample compared to the EPA method but used the same liquid : liquid ratio. Extraction was carried out three times on each water sample, with a solvent mixture of DCM : MeOH (9 : 1, v : v), to recover the extractable dissolved organic signature from each water sample. DCM was used as the main extraction solvent due to its immiscibility with water and ability to dissolve a wide range of organic compounds that can be detected by electron impact mass spectrometry. The extraction resulted in a wide range of detectible compounds from both the solid and aqueous samples. While this extraction is not exhaustive, it was suitable to construct comparable organic fingerprints of the water and soil samples. Due to the dilute nature of the dissolved organics in groundwater, extracted samples were concentrated using a combination of heat and vacuum concentration (Buchi Syncore Analyst, DCM method) to 1.0 ml volume. Where further sample concentration was required, solvent evaporation with a constant stream of pure N₂ was used to reduce the sample to the desired volume.

Two-dimensional gas chromatography time of flight mass spectrometry (GC × GC-ToF-MS) operates in a similar way to standard gas chromatography mass spectrometry systems (GC-MS), except that at the end of the first column the compounds are reinjected onto a second column by use of a thermal modulator. This leads to better analyte separation and greater peak intensity. While, in a standard GC-MS, an individual chromatogram peak may represent several different co-eluting compounds, in GC × GC-ToF-MS the co-eluting compounds are more easily separated and identified by their 2D retention times and mass spectra. The following GC × GC-ToF-MS method was used to analyse soil and water extracts, as adapted from the LECO application note (LECO Corporation USA 2019). Comprehensive signatures of the samples were collected using a LECO (Saint Joseph, MI, USA) time of flight mass spectrometer (Pegasus 4D), with an Agilent 7890A gas chromatography equipped with a LECO thermal modulator. The column set-up was reverse phase, first dimension column DB-17MS (60 m × 0.25 mm i.d. × 0.25 µm: Agilent) polar phase, second dimension column less polar phase Rxi-5Sil MS (1.4 m × 0.25 mm i.d. × 0.25 µm: Restek). Sample injection was splitless using a split/splitless injector set at 260°C, with a helium flow rate of 1.4 ml min⁻¹ for the entirety of the run. The primary oven temperature programme was as follows: initial temperature 50°C, hold for 0.2 min, ramp 3.5°C min⁻¹ to 320°C, hold for 20 min. The secondary oven and thermal modulator had an offset of +10 and +20°C, respectively, from the primary oven. The thermal modulator period was 5 s, and the mass spectrometer transfer line temperature was 300°C with a spectra acquisition rate of 200 spectra s⁻¹. The instrument method was refined through an iterative process changing the temperature ramp and modulation period until a good peak shape and peak separation were found within a standard compound mix containing a semi-volatile standard with 76 different compounds (8270 Standard, Restek) and a 16 compound C10–C40 (even) n-alkane standard (Restek). Repeated injections of the standard compound mix were compared to ensure that the first and second dimension retention times of the compound peaks were consistent across runs. The same standard mix was added as a sample in every subsequent run to ensure that the instrument was working correctly and that the retention times of peaks did not drift between sample batches. This method gave repeatable analyte separation and produced a sufficient number of detectible peaks to build a signature of all the organic compounds contained within the chromatograph.

Processing of 2D gas chromatography data to identify peaks and compounds was carried out using LECO ChromaTOF software. Processing was carried out twice using a low and high signal-to-noise ratio of 50 and 100, respectively. A classification method was applied to remove any compounds related to column bleed or the sample solvent (DCM). The end result for each sample was a 2D chromatograph, and peak table that contained the retention time, intensity, mass spectra and the NIST Library (Linstrom and Mallard 2018) database match for each peak.

The peak tables output from GC × GC analysis can have in excess of 5000 analytes of interest (peaks). Hence, an automated peak table alignment process is required to determine whether peaks with close retention times are genuinely different compounds or whether they are merely misaligned due to a minor shift in retention time between samples. Compound peaks were compared using the statistical compare function within ChromaTOF to align, through pairwise comparison, all the detected peaks within each sample, thus producing an alignment table.

Similarities in the organic fingerprints of each sample were visually identified using PCA (Jolliffe 2002). PCA is a standard technique employed in the analysis of GC × GC data (McGregor et al. 2012). PCA of the alignment table was carried out using R (R Team 2018). PCA determines a set of orthogonal axes, or components (linear combinations of the relative concentrations of the organic compounds), that explain the greatest variance within the data using the fewest components. The underlying similarity between samples can then be elucidated by displaying the samples as coordinates of the first two, most explanatory, principal components. Samples that plot at similar locations will contain similar combinations (or patterns) of the organic compounds. To determine which samples are most similar, hierarchical cluster analysis (HCA) was also performed using R (R Team 2018). HCA determines the shortest distance between each sample. HCA was carried out using the H-clust function within R (R Team 2018). HCA algorithms continue to pair the closest samples, based on their Euclidean distance, until the whole dataset is described within the same cluster. HCA finds a series of clusters that identify similarities between samples. To identify whether clusters were statistically significant, a P-value was calculated using the multiscale bootstrapping ‘pvclust’ function in R and adopting the approximately unbiased approach, as described in Suzuki and Shimodaira (2006).

The groundwater at the GTS is of low conductivity (69–84 µS cm⁻¹) and alkaline (pH 8.83–9.39). Borehole intervals in the south of the GTS have higher dissolved sodium and lower dissolved calcium concentrations than borehole intervals in the north of the GTS (Fig. 3a; see also the data in Supplementary Table S1); these results are consistent with previous findings (Schneeberger et al. 2017; Stillings et al. 2021) that have been shown by Schneeberger et al. (2017) to reflect the change in host-rock lithology from Grimsel Granodiorite (GrGr) in the south of the GTS to Central Aar Granite (CAGr) in the north of the GTS. Groundwater residence time estimates vary between sampling locations in the GTS, which is likely to reflect poor connectivity in the fracture network between surface recharge and sampling locations at depth, giving rise to variably tortuous flow paths (Stillings et al. 2021). The CFC concentrations in samples taken from intervals F, G, H and I, according to IAEA (2006) method 2, and analysed for CFC-11 and CFC-12 by the BGS indicated an apparent groundwater residence time of 57–67 years at this location based on a piston flow model (IAEA 2006). The apparent groundwater age from CFC measurements is consistent with historical and recent tritium measurements. Early tritium measurements from two boreholes (not sampled here) in the south of the GTS (Keppler 1996; Schneeberger et al. 2019) showed an apparent groundwater residence time of between 5 and 36.5 years. More recently, tritium measurements from interval G (see Supplementary Table S1) imply an apparent groundwater age of more than 60 years (Schneeberger et al. 2017). ¹⁴C_DOC dating (Keppler 1996; Schneeberger et al. 2019) shows a similar difference in age estimates, with apparent residence times of 220 ± 180 years for intervals D, E and F in the north, and 13 ± 3 years for intervals (not sampled here) in the south.

In general, the results of the GC × GC analysis show that the samples are highly complex and contain a large number of organic compounds. By way of example, typical GC × GC chromatograms, in the form of 2D contour plots, for soil sample 2b and groundwater sample G are shown in Figure 3b and c. The colour temperature scale denotes high-intensity areas, and each high point represents an individual compound peak. The same compound in each sample will occupy approximately the same retention time in both the first (x-axis) and second dimensions (y-axis), and will plot at the same location on each chromatograph. Similar compounds or groups of compounds elute along predictable trends in the chromatograph. Changes in chain length of the same type of compounds (i.e. carbon number) are reflected by a systematic increase in the retention time. Different groups of compounds have different affinities to the second-dimension column's stationary phase, causing separation by the compound group along the y-axis. So, for example, n-alkanes have a clear peak separation from ketones, alcohols, aromatics, and other branched and unsaturated aliphatic compounds. When comparing the soil sample to the groundwater sample (Fig. 3b, c), there are similar patterns in the elution of specific compounds. However, the relative concentration of the longer chain alkanes and alkenes (labelled in Fig. 3b) is higher in the soil than in the groundwater. The total number of compounds detected in each sample is given in Figure 3d; compound abundance varies significantly, ranging from 826 in groundwater sample E to 5000 in lake-water sample LW.

To identify similarities in the organic signatures of surface and groundwater samples, it is necessary to examine the relative abundance of individual compounds that are common to most samples: that is, to determine whether the ratios (or pattern) of preserved compounds at depth can be compared with the surface-soil and water samples, and, hence, used to indicate the groundwater origin. The statistical compare function within ChromaTOF was used to align the organic compounds, producing a compound alignment table (see Supplementary Table S2). The statistical analysis used 50 organic compounds, which were common to 80% of the samples. To calculate the relative abundance of each compound, the abundance ratios of these 50 aligned compounds within an individual sample was taken (i.e. for each sample the sum of the relative concentrations of all 50 compounds is equal to 1). Figure 4 summarizes the relative abundance of compound classes for the 50 aligned compounds for each sample, where the length of each colour represents the relative fraction of each compound classification. Most repeat samples from the same sample location (labelled i and ii) have similar proportions of each different compound class, and samples of the same type (i.e. groundwater, soil and lake water) are visibly similar (Fig. 4) – with the exceptions of D and E from the groundwater, and 2b from the soil, all three of which have a smaller proportion of acid and alcohol compounds than the other samples. There is a visible variation in the relative abundance compound classes within the groundwater samples F, G, H, I and M that most notably vary in their organic acid content, where the relative abundance ranges from 20 to 50% of the aligned compounds.

Before comparing organic fingerprints between surface samples and groundwater samples, it is first important to understand the variability of aligned compounds in the surface samples to determine if there is a variation in soil organic fingerprint that may be reflected in the groundwater samples. Comparison of all the aligned soil samples using the 50 aligned organic compounds common to all samples as explanatory variables was carried out using PCA (Fig. 5a), and clustering was performed using HCA to identify any significant clusters (Fig. 5b). In both the PCA and the HCA the replicate samples (i and ii) plot in the same location, with the exception of samples 1ai and 1aii that do not plot as closely. Cluster analysis shows that there is more than 95% confidence of two different clusters being present within the soil samples. Locations 1–4 cluster together, while 2b clusters separately from 2a and the other soil sample locations.

To identify whether similarities exist between the groundwater samples, the lake water and the different soil clusters identified in Figure 5b, PCA was carried out using the same 50 aligned organic compounds as explanatory variables, as determined from the peak alignment table (see Supplementary Table S2). The analysis used all samples (soil, lake water and groundwater) in order to determine whether the surface and groundwater sampling sites have clearly distinct organic signatures that form statistically significant clusters based on the aligned compounds between samples. Results from the PCA are shown in Figure 6. Principal component 1 (PC1) explains 33% of the variance and principal component 2 (PC2) explains 19% of the variance, which is not uncommon in datasets such as this with a larger number of variables compared to the number of observations (Ringnér 2008). Most repeat soil and sediment samples (labelled i and ii for direct repeats, and a and b for samples taken in the same geographical area) plot in the same region of the PCA plot (Fig. 6). Soil samples from sites 3 and 4, taken from the lower slopes of the mountain above the GTS, consistently plot in a similar location. Of the two samples taken from location 2, sample 2a plots similarly to the other soils; however, sample 2b, whilst being similar in PC1, is very different in PC2 when compared to all of the other soil samples. This indicates a distinct difference in the organic signature of soils at location 2, which is at the highest elevation of all of the soil sampling sites and was the only location at which roots and plant matter were visually apparent in the soil samples. The lake water (orange) plots in a distinctly different location to any of the soil samples (green), indicating it has a different organic signature that is clearly distinguished within PC1.

Groundwater samples F, G, H, I and M all plot close to the lake-water signature in the PCA (Fig. 6), indicating that they have similar organic fingerprints to each other and could be derived predominantly from infiltration of the lake water. The water level in Lake Raterichsboden lies within the range of groundwater head measurements found throughout the GTS (at the time of sampling the water level in Raterichsboden was higher than the head in interval I but lower than in intervals F, G and M). By comparison, the water level in Lake Grimsel is higher than all head measurements in the GTS. Hence, it may be that the groundwaters in intervals F, G, H, I and M were originally derived from Lake Grimsel, which is immediately upstream of Lake Raterichsboden and hydraulically connected via the pumped-storage hydropower system.

Groundwater samples D and E differ from the other groundwater samples (blue in Fig. 6). All groundwater and lake-water samples contain 44 or more of the 50 aligned compounds (present in 80% of samples), with the exception of Hii, which contains 21 out of the 50 and plots near to Hi; so the differences in PC1 and PC2 for samples D and E cannot be attributed to a smaller number of compounds in these samples. Samples D and E plot with soil sample 2b (Fig. 6), indicating that the groundwaters contain a similar organic signature to soils at higher elevations and are likely to have originated from surface-soil infiltration. Interestingly, groundwater samples D, E and F are all sampled from separate locations at different distances down the borehole; sample locations are separated by a hydraulic packer system with separate flow lines to allow sampling from different distances within the same horizontal borehole. Groundwaters at D and E seem to comprise predominantly water that infiltrated through surface soils, whereas groundwater at interval F, in the same borehole, is likely to have a hydraulic connection to the lakes or is derived from lake water. Their very different organic signatures supports previous research observations that the local fracture network is very poorly connected even between sampling intervals from the same borehole (Stillings et al. 2021).

To identify whether the clusters indicated by the PCA analysis (Fig. 6) are statistically significant, HCA was carried out on the 50 aligned compounds (Fig. 7). Three statistically significant (99% confidence level) clusters are identified and grouped through HCA, indicating that the samples within each cluster have statistically similar organic fingerprints. Within each cluster on the dendrogram (Fig. 7), as expected, the series neighbour (the most similar) in the HCA analysis are replicate samples collected from the same borehole interval or, in the case of the soils, the duplicate extraction of the same homogenized soil sample (i.e. i and ii). In the soils, the next closest pair is generally the ‘b’ sample from a neighbouring location, except for soil samples 1a and 3b. The only sample that does not cluster with its duplicate is sample 1aii, which clusters with the groundwaters, while all the other samples taken from location 1 cluster together within the other soil samples 2a, 3a, 3b, 4a and 4b; we attribute the deviation of sample 1aii to analytical uncertainty probably during the extraction of the organic fingerprint from sample 1aii, as a result this point can be considered as an outlier. For all 12 separate samples taken from the eight groundwater sampling intervals (Figs 6 and 7), the HCA analysis shows it is possible to group and differentiate samples with similar organic fingerprints, which implies that there is likely to be a relationship between samples within the same clusters.

The top 10 compounds with the highest loading magnitudes in PC1 are the compounds that describe the most sample-to-sample variance in the first principal component, presented in Table 1, and, hence, are indicative of the differences between clusters in the organic fingerprints. PC1 is responsible for the separation of the ‘groundwater and lake-water cluster’ from the other two clusters shown in Figure 6. Of the top 10 loading compounds in PC1 some derive from natural sources, for example: decane, 6-4ethyl-2-methyl- has been found in a certain species of plant root extract (Shettima et al. 2013); 1-iodoundecane is commonly found in mammal urine (Achiraman and Archunan 2002) and is also an active compound in some plants (Khammas et al. 2020); and 1-hexene, 4,5-dimethyl- has been detected as an excretory compound from fungi (Simon et al. 2017). Of the top 10 compounds in PC2, which separates the smaller ‘groundwater and soils’ cluster from the other two clusters in Figure 6, two are classified as unknown compounds (unknown compound 147 and 224) as they do not have a higher than 70% match to any compounds within the NIST Library (Linstrom and Mallard 2018). A manual search of unknown 147 and unknown 224 with the library found a 66% match with propanoic acid, 3-hydroxy-2-isopropylidene and a 63% match with 1,1-difluoro-2,2-dimethyl-cyclopropane, respectively. Two other compounds in the top 10 loadings in PC2 are known to derive from natural sources: oxime-, methoxy-phenyl- has been found in plants and in animal mucus (Sallam et al. 2009; Al-Mussawii et al. 2022); 2-undecenal, E- is found in essential oils derived from some plants (Kivcak et al. 2001); and decane has also been found in plants (Cakir et al. 2004) but is also common in petroleum and coal tars (Pan et al. 2012). Thus, decane does not have a specifically discernable natural source. Other compounds (PC1 and PC2 in Table 2) do not have a specific identifiable natural source and could potentially derive from either natural or industrial processes.

The results of the PCA and HCA analyses show that groundwater, lake water and soil waters have distinct, and repeatable (i.e. duplicate samples that fall within the same cluster), dissolved organic signatures, and that these signatures can be potentially be used to determine the predominant influence of near-surface recharge sources on groundwater samples at depth. The PCA and HCA analysis identified three statistically significant clusters, based on their organic signatures. These clusters indicate that at the GTS, most groundwater sampling intervals tap fractures that are hydraulically connected to the lake water. However, sampling intervals D and E, which are in the north of the GTS do not cluster with the lake water, this suggests a second potential infiltration source that could reflect water infiltrating through soils at higher altitudes and which have an organic signature similar to soil sample 2b. The variability in the organic groundwater signatures, particularly from neighbouring sampling intervals within the same borehole (D, E and F), underlines the poorly connected nature of the fracture network. This observation is further supported by the variation in the groundwater age estimates between boreholes at different locations (Keppler 1996; Schneeberger et al. 2019) and previous observations of highly localized perturbations in pH associated with microseismic events during reservoir drainage and maintenance (Stillings et al. 2021).

Previously researchers have successfully discriminated groundwater origins through the use of unique ‘target’ biomarker compounds (Derrien et al. 2017). At the GTS, no such target biomarker compounds were found. Instead, groundwater origins were obtained using an untargeted organic ‘fingerprint’ for each water/soil sample, in which relative concentrations were determined for a large number of common compounds. For organic fingerprinting to be an effective groundwater tracer at other locations, a sufficient number of organic compounds within the surface signatures must be well preserved over time, so as to be identifiable at depth. Organic matter decay rates change most rapidly in the shallow subsurface (to depths of c. 350 m), attributed to the changes in oxidative decomposition (Kortelainen and Karhu 2006). However, active microbial communities have been shown to exist in groundwater systems to depths of up to c. 1 km (e.g. Shimizu et al. 2007; Nyyssönen et al. 2012). These communities will gradually metabolize organic components in the groundwater, leading to progressive degradation of organic parent molecules over time and, hence, along the groundwater flow paths. This degradation of organic parent molecules, into daughter decay products, is likely to explain the large total number of compounds (Fig. 3d) that were found in the lake water and most groundwater samples when compared with the soil samples. At the GTS, where apparent groundwater residence times vary by location, from 5 to 220 years (Keppler 1996; Schneeberger et al. 2017, 2019), the number of detectable organic compounds that remained in 80% of all samples, and were thus usable in the final PCA and HCA analysis, was 50. Our results show that these 50 compounds were sufficient to discriminate between the distinct surface environments, and that these surface organic signatures could still be identified in groundwaters at depth. Evidence from other research fields also suggests that long-term solid and dissolved preservation of organic compounds may not be uncommon; Korkmaz and Gülbay (2007) used specific compounds as indicators of surface deposition environments for petroleum source rocks that are of Jurassic age (Korkmaz and Gülbay 2007), while specific dissolved organic compounds are found preserved in groundwaters of up to 23 kyr in age (Aravena et al. 1995). Further studies, using sites with older and younger groundwaters in differing surface and geological environments, are required to determine the range of geological settings and age of groundwaters for which organic fingerprinting can prove a useful tool for investigating groundwater origins.

It is possible that some of the variability in the PCA analysis in Figure 6 is due to groundwater mixing between surface infiltration water and the lake water, specifically at groundwater sampling location F. Whilst F forms a statistically significant cluster (at >99% confidence level) with the lake water in the HCA analysis, in the PCA it plots between some of the soils and the lake water. To identify whether groundwater mixing could be responsible, future investigations could prepare different proportional mixes of each infiltration source and include these signatures for comparison in the statistical analysis, thereby allowing any potential groundwater mixing to be identified.

The conclusions that can be drawn about the groundwater system at the GTS in this study are limited due to the small total number of samples that it was possible to collect and the restrictions to surface access. Despite the small sample size, the HCA identified three clusters with a 99% confidence level, enabling us to clearly distinguish groundwater sampling locations that are predominantly fed via surface-soil infiltration (D and E) from those that are dominated by lake-water infiltration. In future studies, a larger sample size would reduce the uncertainty when comparing organic fingerprints and might enable clusters to be identified that link individual surface-soil infiltration sites to specific groundwater sampling locations. The use of additional complementary geochemical techniques would also be useful in further constraining the meteoric infiltration locations.

Two-dimensional gas chromatography was used to organically fingerprint surface-soil, lake-water and groundwater samples at the Grimsel Test Site in Switzerland. Three distinct meteoric infiltration types were identified with uniquely different proportions of the same compounds forming their individual organic fingerprints: two types of surface-soil environment and the lake water. These surface infiltration fingerprints were compared to organic signatures found within seven borehole sampling intervals located at a depth of 200–500 m below ground surface, positioned throughout the length of the GTS tunnels. Fifty organic molecules were found to be common to 80% of samples. Using principal component analysis (PCA), the relative abundance of these molecules was used to match the individual borehole samples to their likely surface-water origins. Hierarchical cluster analysis (HCA) was used to identify three statistically significant clusters. These clusters showed that most groundwater sampling intervals were clustered with the lake water and, hence, primarily tap fractures that are hydraulically connected to the lake. Two intervals, however, were clustered with soil taken from the highest altitude sampled on the mountain above the GTS, thus suggesting that these tap fractures connected to surface infiltration water at high altitudes. This research demonstrates that natural organic molecules, and their relative abundance, are sufficiently well preserved in groundwater over timescales of several decades that they can be used to discriminate the near-surface environment(s) through which meteoric groundwater has infiltrated. Organic fingerprinting could be a powerful new tool for an improved understanding of groundwater flow systems, particularly when used in combination with other complementary tracing techniques.

The research forms part of the collaborative Large-Scale Monitoring (LASMO) programme at the Grimsel Test Site. We would like to thank the anonymous reviewers for their suggestions and helpful input.

MS: conceptualization (supporting), data curation (lead), formal analysis (equal), investigation (lead), methodology (lead), validation (equal), visualization (equal), writing – original draft (lead), writing – review & editing (lead); RJL: conceptualization (equal), formal analysis (supporting), funding acquisition (lead), project administration (lead), supervision (equal), visualization (supporting), writing – original draft (supporting), writing – review & editing (equal); ZKS: conceptualization (equal), funding acquisition (equal), investigation (equal), project administration (equal), supervision (equal), visualization (supporting), writing – original draft (supporting), writing – review & editing (equal); RAL: conceptualization (equal), funding acquisition (equal), investigation (equal), methodology (equal), project administration (equal), validation (equal), writing – review & editing (equal); ST: conceptualization (equal), funding acquisition (equal), project administration (equal), writing – review & editing (equal); MK: formal analysis (supporting), methodology (equal), project administration (supporting), resources (equal), software (equal), validation (equal).

This work was funded by the Engineering and Physical Sciences Research Council (grant EP/M506643/1 awarded to R.A. Lord) and the Nuclear Waste Services (grant awarded to R.J. Lunn).

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data that support the figures are available in the Supplementary information dataset ds01.