The contaminated site is located in Upper Austria in proximity of the river Danube. Between 1896 and 1972, a railroad sleeper impregnation plant, encompassing approximately 11 hectar, was in operation. Sleepers, poles, and other wooden materials were impregnated using distilled coal tar, zinc chloride, and other heavy metal salts in mobile (up to approximately 1900) and fixed tank and steam pressure impregnation (in the following) facilities. Documents from the early operation period indicate that spent creosote was disposed of in groundwater wells, i.e., without passage through the vadose zone. Cooling water was drained close to the impregnation facilities, possibly entailing increased contaminant mobility. Local surface contamination arose from losses during manipulation at the site of impregnation and creosote storage facilities and storage of recently treated material. Following site closure in the 1970s, the area is continuously used for industrial purposes. Contaminated soil was partly disposed of, and a large part of the surface is now sealed and poorly accessible for investigations. At the time of risk recognition in the 1990s, the creosote had formed a DNAPL at the site and in the groundwater downstream region following the aquitard relief approximately 15 below surface, spreading over an area of approximately 30.000 m², and with a thickness of up to 100 cm. Maximum length and width are about 450 and 40 m, totalling a volume of 2 million liters of creosote. The groundwater flow velocity is approximately 0.8–1 m/d, at a table inclination of 1.0–1.5 permille, and a permeability of the aquifer sandy gravel of 9 × 10⁻⁴−9 × 10⁻³ m/s. A detailed description is available elsewhere (UBA, 2013). Our recent publication provides information about aqueous contamination, iron mineral availability, bacterial, and archaeal microbiome, aquifer stratification and describes natural attenuation processes (Scherr et al., 2016).

Groundwater samples were collected from deep, i.e., DNAPL-containing, and shallow wells above the DNAPL and in up- and downstream regions, and analyzed for their content of coal tar derived constituents. The DNAPL sample, approximately 2 L, was collected from a central downstream deep well using a disposable plastic pump that was carefully immersed approximately 20 cm below the DNAPL/water interface, and stored in a glass bottle at 4°C in the dark until further analytical and experimental treatment.

Equilibration experiments using site DNAPL and groundwater samples to obtain c_{i, sat, lab} (Table 1) were conducted in the laboratory at 20°C in the dark as described earlier (Scherr et al., 2016). Briefly, approximately each 10 g of DNAPL were weighed into triplicate 2 L Pyrex bottles which were carefully filled with 1 L of site groundwater. Following gentle mixing on an orbital shaker (10 rpm) for 24 h, avoiding DNAPL dispersion, the bottles were left to rest for 1 week. Aqueous samples from different levels were withdrawn using a volumetric pipette in regular intervals to determine the point of equilibrium.

Extraction of aqueous and DNAPL samples was conducted following ultrasound-assisted (1 h) dissolution in heptane and dichloromethane (Scherr et al., 2016). Light and dense solvent and corresponding aqueous residues were separated in separation funnels, and gently evaporated if required. Analysis of extracted water, groundwater, and tar DNAPL samples using gas chromatography coupled to flame ionization detection (GC-FID) was perfomed in a modification of ISO DIN 16703 (Scherr K. et al., 2007). Unidimensional gas chromatographic (GC-MS) and comprehensive two-dimensional gas chromatography coupled to mass spectrometry (GCxGC-MS) analysis was conducted as described earlier (Hasinger et al., 2012; Vasilieva et al., 2012a,b). GCxGC-MS analysis is efficiently applied for the qualitative analysis of complex hydrocarbon mixtures (Rowland et al., 2012; Wilde et al., 2015). Compound identification must be carefully performed via library searches, requiring differentiation from those not included in the library, which is supported by the use of authentic standards similar to unidimensional GC. Equivalent carbon numbers on GC-FID were identified via comparison with authentic standards of normal alkanes purchased from Sigma. The presented data are averages of triplicate extractions. For quantified compounds, the following abbreviations are used: NAP Naphthalene, PHE Phenanthrene, FLA Fluoranthene, CAN Acenaphthene, FLU Fluorene, DBF Dibenzofuran, DBT Dibenzothiophene, CAR Carbazol, PYR Pyrene, 1NAP 1-Methylnaphthalene, 2NAP 2-Methylnaphthalene, BiP Biphenyl, ANT Anthracene, BaA Benzo(a)anthracene, CHR Chrysene, 26DN 2,6-Dimethylnaphthalene, BT Benzothiophene, IDA Indan, IDE Inden, 2ANT 2-Methylanthracene, ACY Acenaphtylene, BhQ Benzo(h)quinolin, BbF Benzo(b)fluoranthene, BaP Benzo(a)pyrene, 1PNA 1-Phenylnaphthalene, PHO Phenanthridinone, BkF Benzo(k)fluoranthene, ACR Acridine, and PHI Phenanthridine.

The original DNAPL composition on the present site is not known, and has likely undergone substantial changes over extended time of plant operation and the following period of weathering in the aquifer. Moreover, technical advances in coal gasification led to different tar qualities used over time, which vary especially in the content of acids and heterocyclic compounds (Adam, 1932; Birak and Miller, 2009). The DNAPL sample of concern was sampled from a deep groundwater well approximately 250 m in groundwater flow direction from the downstream border of the original site. The DNAPLs's low specific gravity of 1.01, but also low dynamic viscosity (around 15 mPa·s) (UBA, 2013) are unusual even for lighter and distilled tars (Brown et al., 2006; Birak and Miller, 2009). Earlier investigations (UBA, 2013) have noted a decline of viscosity and specific gravity along the direction of migration along the aquitard relief, with 40 mPa·s and a specific gravity of 1.1 in the center of the site, suggesting an impact of transport in the composition of migratory DNAPL. Subsurface chromatography effects have been observed for high molecular weight hydrocarbons in crude oil reservoirs (Bastow et al., 2007), and a similar depletion may have occurred during DNAPL migration presently.

Figures 1-3 show the results of qualitative and quantitative DNAPL hydrocarbon analysis via GC-FID, GCxGC-MS and a synopsis of GC-MS and GC-FID analysis, respectively. Despite the long exposure time—40–140 years, under consideration of site operation time—hydrophilic compounds such as Indene and Indane (aqueous solubility >100 mg/L) and NAP are abundant inside the DNAPL body, indicating little impact of weathering. Other studies have found groundwater well DNAPL samples to be less weathered than those from the vadose zone, also with NAP remaining the dominant compound (Brown et al., 2006; Birak and Miller, 2009). In contrast, DNAPL solidification, adjoined by depletion of low molecular weight compounds has been observed in simulated aquifers within only a few years (Liu et al., 2009). The investigated samples are from a depth of approximately 20 cm below the groundwater/DNAPL interface. Based on the identified compounds, 37% of DNAPL molecules contain at least one aromatic ring (Figure 3, corresponding to 40% by weight), and homocyclic molecules dominate over heterocyclic ones. The residual, poorly resolved mixture in the GC-FID chromatogram (Figure 1) is referred to in the following as uncharacterized hydrocarbons (UHC), and represents a substantial coal tar mass fraction (Figure 3), although of unknown composition. The coal tar visibly dissolved without residue in dichloromethane or n-heptane upon sonication, with a recovery between 94.7 and 95.3%, respectively, eluting between n-C₁₀ and n-C₄₀ on GC-FID. Solubilisation in saturated solvents reflects the low abundance of asphaltenes (Kharrat et al., 2007), and may be attributed to the migratory character of the sampled DNAPL and the distillation process. BTEX and phenols were not detectable (Scherr et al., 2016), and inorganic contaminants, typically Arsenic, Chromium, Lead, and Cyanides associated to coal tar, were not within the focus of the present study.

Coal tars and creosotes consist of up to 10.000 of individual constituents with the molecular weight of pitch range constituents exceeding 10,000 Da (Millan et al., 2005; Herod et al., 2007), rendering a comprehensive compositional assessment difficult, especially for high molecular weight tars. This requires the use of indirect methods to assess DNAPL average molecular weight and functional group chemistry. The traditional method for the determination of the average molecular weight, vapor pressure osmometry, has been criticized for its dependence on sample properties (Brown et al., 2005), and was not available in the period of data compilation, prompting for an alternative method of assessment. Previously (Vasilieva et al., 2012a,b; Scherr et al., 2016), we used high-resolution two dimensional gas chromatography coupled to quadrupole (GCxGC-MS) and time-of-flight mass spectrometry (GCxGC-TOF-MS) for the identification of coal tar oil constituents. Figure 2 presents an annotated GCxGC-MS chromatogram of a DNAPL sample. Beside those identified based on comparison of mass spectra and retention behavior with authentic standards on GC-MS (see Figure 3), a variety of additional DNAPL constituents were identified based on their mass spectral library match; thus, most isomers are only tentatively assigned. The presently used GC-FID system is set up for the quantification up to 563 Da, or ECN-40 (n-tetracontane) and qualification compounds of a boiling point of up to 550°C (Scherr K. E. et al., 2007; Erlacher et al., 2013), encompassing typical creosote distillation ranges. Beyond 6-ring PAH, neither GC-FID nor GCxGC-MS analysis of fully dissolved coal tar gave profound signals. Up to 90% of coal tar mass is known to be aromatic (Morgan et al., 2008). Mid-molecular weight compounds, predominantly methylated homocyclic compounds were identified on GCxGC-MS (Figure 2), and are also visible as unresolved mixture in the corresponding GC-FID chromatogram (Figure 1), with an equivalent carbon number (ECN) below 30. The unresolved mixture's centroid is located around ECN-15, indicative of the average carbon number of compounds comprising the UHC. Together with the aromatic, alkylated nature of the uncalibrated compounds, methylanthracene (C₁₅H₁₅) was chosen as a model compound for the stoichiometric conversion of UHC range hydrocarbons (Figure 3). Trimethyl- and ethylmethylbenzenes add to to the light fraction, but were not quantified, as indicated in Figure 2. Consequently, the coal tar DNAPL has an average molecular weight (MW) of approximately 180 g/mole, representing an untypically light DNAPL (Brown et al., 2006), moreover in the light the weathering period, where light components are expected to be preferentially depleted. This may also be indicative of migratory effects, where higher molecular weight compounds may be separated in subsurface chromatographic effects during the movement through the aquifer. In addition, high molecular weight compounds such as asphaltenes, if present originally, may have been held in solution by saturated constituents, and were precipitated from the unstable DNAPL following alkane dissolution during weathering. Precipitation is ideally to be expected when the mole fraction of a compound exceeds its solid/liquid fugacity ratio (Peters et al., 1997), which is nearly the case for anthracene in the present DNAPL (Table 1).

Following Raoult's law, ideality in hydrophobic mixtures, with an activity coefficient γ_i = 1 can be assumed for all constitutents i provided their similarity in size, structure or shape and functional group chemistry (Peters et al., 1999; Eberhardt and Grathwohl, 2002; Klenk and Grathwohl, 2002). However, if molecular interaction or arrangement deviates substantially from that of a pure, unmixed liquid, excess enthalpy or entropy become nozero, respectively; if this is the case for hydrophobic mixtures, is not known a priori, owing to the complexity of possible coal tar dense or light NAPLs, and their degree of post-release alteration, i.e., weathering (Brown et al., 2006; Birak and Miller, 2009). Compositional changes during DNAPL weathering, with a continuous depletion of hydrophilic compounds, further contribute to the dynamic character of phase composition, relative abundance constituent properties, and thus, individual activity. Activity coefficients exceeding unity suggest self-association tendencies, and those below 1 to have molecular properties resulting in excess solvation tendencies. Structural decomposition using the group contribution method for UNIFAC (Universal Quasichemical Functional Group Activity Coefficient (Fredenslund et al., 1977) modeling indicate the most abundant structures, including methylanthracene as model UHC compound, in the present DNAPL to be ACH (aromatic carbon with one hydrogen atom attached, 69%), AC (aromatic carbon, 29%) and, to a lesser extent, ACCH₃ (methylated aromatic carbon, 0.5%), similar to comparable light coal tars (Peters et al., 1999). Heteroaromatic group fractions (C₄S, C₅H₃N, and C-O) each contribute two orders of magnitude less.

Peters et al. (1999) provided estimations of γ_i based on the UNIFAC group contribution method for 43 tar oil components, which ranged between 0.14 and 1.13 for quinoline and ethylbenzene, respectively, in the two light DNAPLs of comparable molecular composition (average MW around 250 g/mol), suggesting excess solvation tendencies. Based on the similarity of MW, functional group distribution and molecular composition, we adopted the activity coefficients γ_i provided there. Values for γ_i averaged (n = 2), for the presently most abundant compounds, to 1.00, 0.99, and 0.98 for FLA, PHE, and FLU, respectively, and 0.93 and 0.94 for NAP and ACN, while the two heavier DNAPLs had slightly higher activity coefficients (e.g., γ_PHE = 1.02, γ_NAP = 0.99; Table 1). More condensed PAH such as BaA and heavier congeners, showed γ_i only slightly exceeding unity. Generally, alkylated PAH tend to have γ_i > 1, and N-heterocyclic and hydrophilic compounds γ_i < 1.

Table 1 shows calculated aqueous equilibrium concentrations (c_{i, sat, calc}) for some abundant DNAPL constituents, incorporating subcooled liquid solubilities and activity coefficients in extension of data presented earlier (Scherr et al., 2016), and comparison with experimental (c_{i, sat, lab}) and groundwater concentration data in proximity of the DNAPL surface (c_{i, w, field}). Equilibrium concentrations are significantly higher than those presented earlier (Scherr et al., 2016) when using a simplified version of Raoult's law, owing predominantly to the consideration of fugacity ratios than to the consideration of activity coefficients, deviating largely unsubstantially from unity (Table 1). They are in good agreement with those obtained from our laboratory equilibrium experiments (c_{i, sat, lab}), with the exception of Naphthalene, which is overpredicted by a factor of 2.6.

Taking qualitative groundwater concentration data into account (Table 1), Figure 4 presents a GCxGC-MS profile of a low-flow groundwater sample taken approximately 20 cm above the groundwater/DNAPL interface in a well, i.e., above the open DNAPL surface. The profile is substantially different from that of the DNAPL and presents a limited range of hydrophilic compounds, encompassing a variety of dimethylnaphthalenes, which are underrepresented in the DNAPL. Previous calculations (Scherr et al., 2016) indicated the thickness of the theoretical boundary layer in the aquifer above the interface, where equilibrium concentrations (i.e., c_{i, sat}, Table 1) are expected to prevail, to be at least 80 cm for the case of NAP, and even greater over a free DNAPL surface inside the well. There, any depletive, anaerobic natural attenuation processes are expected to occurr at a lower rate than DNAPL dissolution (Eberhardt and Grathwohl, 2002; Lee, 2004; Scherr et al., 2016). Thus, the groundwater sample can be assumed to be in equilibrium with the DNAPL. However, while laboratory data, using a freshly sampled DNAPL, agree well with equilibrium assumptions, significantly lower concentrations were recovered throughout the field tests (additional groundwater data presented in Scherr et al., 2016), indicating interfacial alterations during weathering.

At the DNAPL-groundwater interface, a variety of effects are expected to occur. Alshafie and Ghoshal (2004) reported the formation of viscous, semi-rigid films, providing significant resistance to solute diffusion into the water phase, resulting in a reduced mass transfer from the DNAPL to the aqueous phase. There, depletion of hydrophobic compounds, bulk DNAPL diffusion effects and changed wetting properties were ruled out as driving factors. After 1 year of weathering, Nelson et al. (1996) found no enrichment, depletion or polymerization at the interface, but the formation of a semigelatinous film formed by weakly bonded water in an emulsified interface. However, the formation of a skin could not be verified due to the in situ character of the study. Based on field PAH concentrations, a hypothetical representation of the interfacial composition or selectivity of any film or skin for PAH solubilization from the bulk DNAPL, f_{interface, calc} can be calculated based on Raoult's law (Table 1, Figure 5—compare to bulk DNAPL, Figure 3). Using this approach, NAP and ACN are overproportionally depleted, or selectively kept from release, possibly due to emulsification, with the mid-molecular weight PHE becoming the dominant compound.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.