Rare earth element geochemistry of outcrop and core samples from the Marcellus Shale

Sample acquisition

Fresh exposures from Marcellus Shale outcrops (N = 11) were collected as part of research at the Department of Energy, National Energy Technology Laboratory (NETL) in the Industrial Carbon Management Initiative (ICMI) between May 2010 and September 2011. Four outcrops were sampled from northern, surface exposures in New York State (NY) while the remaining outcrop samples originated from southern, surface exposures in West Virginia (WV) and Pennsylvania (PA). Outcrop samples were primarily from the Union Springs or Oatka Creek members of the Marcellus Shale. Samples (N = 6) at six depth intervals (between 7,780 and 7,920 ft below ground surface) from a single core were provided by an industrial partner, under terms of confidentiality, operating in Greene County, PA. Locations of core and outcrop samples can be seen in Figure 1. Details of the outcrop samples, including lithologic and stratigraphic descriptions, are given in Additional file 1: Table S1.

While the samples studied here were not exclusively from gas-bearing members of the Marcellus, if the REE are to be used as tracers it is important to have thorough characterization of the REE in over- and underlying strata. Induced fractures (and therefore fluids) often propagate “out-of-zone” and, at times, hundreds of meters vertically above and below the perforation midpoint [22]. Therefore the variety of strata (within the Marcellus) studied here are generally of interest for naturally occurring tracer applications.

Rare earth element abundance, correlations, and profiles

Concentrations of the study analytes are summarized in Figure 2, with sample-wise results presented in Table 1. In general, the REE varied over three orders of magnitude with 95% of all measurements in the range between 0.324 and 75.2 ppm. As expected, the REE exhibited a “zig-zag” pattern of abundance, consistent with the Oddo-Harkins effect [23]. Several samples were enriched—relative to world black shales [24]—in Pr, Dy, Ho, and Er, however most samples fell within the typical range for black shales compiled by Ketris and Yudovich [24].

Sample name	Sc	Y	La	Ce	Pr	Nd	Sm	Eu	Gd	Tb	Dy	Ho	Er	Tm	Yb	Lu
Bedford, PA	7.37	22.5	15.3	30.8	4.14	16.2	3.74	0.81	4.50	0.65	3.67	0.86	2.53	0.34	2.3	0.32
Canoga, NY (OCM)a	13.0	33.2	26.4	44.2	6.54	26.1	5.99	1.31	6.49	0.97	5.02	1.21	3.48	0.52	3.22	0.47
Canoga, NY (OCM)a	12.3	33.6	26.5	42.5	6.44	26.1	5.71	1.22	6.96	0.93	4.83	1.20	3.47	0.48	3.04	0.48
1-DGLS	6.51	29.5	19.0	34.4	4.89	20.0	4.88	1.07	5.16	0.82	4.77	1.20	3.51	0.52	3.52	0.54
Petersburg, WV (N)	12.6	27.0	32.2	65.4	7.90	31.8	6.23	1.27	5.89	0.83	4.47	1.06	2.90	0.42	2.73	0.42
C-7789	19.1	18.3	28.7	61.2	7.14	26.6	4.99	1.13	4.46	0.66	3.34	0.81	2.30	0.35	2.33	0.36
Whip Gap, WV	5.42	11.3	12.7	21.1	2.75	9.65	1.88	0.41	1.83	0.28	1.66	0.41	1.16	0.19	1.25	0.17
Burlington, WV (1)	21.2	27.8	44.7	97.0	11.0	42.2	7.83	1.60	6.38	0.91	4.95	1.11	3.39	0.51	3.12	0.46
Canoga, NY (USM)	17.2	32.6	38.4	75.6	9.17	33.4	6.55	1.48	6.62	0.97	5.08	1.20	3.51	0.49	3.15	0.44
C-7838	22.7	30.5	38.6	75.9	9.08	34.8	6.82	1.46	6.25	0.93	4.93	1.20	3.60	0.52	3.32	0.50
Petersburg, WV (W)	23.0	30.6	45.5	96.5	11.2	40.3	7.86	1.62	6.62	1.00	5.23	1.26	3.53	0.53	3.51	0.52
C-7907	10.8	15.4	37.1	53.0	5.90	19.3	3.21	0.64	2.90	0.40	2.34	0.58	1.81	0.27	1.81	0.29
C-7801b	22.7	29.2	39.6	81.6	10.0	37.5	7.95	1.53	7.01	1.04	5.25	1.24	3.69	0.54	3.39	0.49
C-7801b	22.5	31.3	39.5	81.9	9.88	38.4	8.14	1.54	7.05	1.04	5.17	1.22	3.54	0.54	3.47	0.50
Le Roy, NY	20.2	36.6	35.4	73.4	9.26	36.2	7.99	1.77	7.81	1.19	6.13	1.45	4.23	0.62	4.35	0.61
Marcellus, NY	20.9	33.6	42.5	88.7	10.3	39.3	7.80	1.64	7.19	1.05	5.51	1.33	3.91	0.53	3.49	0.54
Burlington, WV (2)	4.37	24.0	18.2	32.5	4.28	18.4	3.92	1.18	4.70	0.64	3.31	0.78	2.08	0.29	1.68	0.27
C-7813	23.1	27.3	40.4	79.5	9.52	34.8	6.82	1.33	5.51	0.83	4.4	1.10	3.37	0.48	3.12	0.46

All results are mass fractions, reported in parts per million (mg/kg) with three significant-digit precision. Average analytical uncertainty was 3.4% from 5 sweeps during analysis.

^{a, b}Method duplicates.

The REE were also highly, positively correlated in these samples. The median interelement correlation (Spearman’s \(\rho\)) was 0.81 while 95% of all correlations fell between 0.47 and 0.98. The minimum observed correlation (0.33) was between La and Y. In general, REE tended to correlate most strongly with the nearest elements, with Sc correlating better with the LREE and Y with the HREE (Additional file 1: Figure S3). Overall, the high correlations exhibited in these samples were consistent with correlations determined in aqueous media [25], which was expected given the ubiquitous occurrence and coherent chemical properties of the REE.

Rare earth element concentrations were statistically compared between core and outcrop samples and between outcrop localities to determine if presumed weathering of outcrops or regional variations might yield systematically different REE concentrations. Despite the REE concentrations in the outcrop samples appearing to be more variable than in the core samples (e.g. Eu in Figure 2), no statistically significant differences in element variability were determined between the outcrop and core samples (Ansari-Bradley test for difference in scale parameter; \(P \approx 1\) for all elements following Bonferroni–Holm corrections for multiple comparisons). Similarly, no statistically significant differences were found in the central tendencies of any of the REE between the two sample types (Wilcoxon rank-sum test for location shift; \(P \approx 1\) for all elements, corrected for multiple comparisons). Analogous, parametric tests (Bartlett test for homogeneity of variance and a t test) were performed, also indicating no significant differences (Additional file 1, Section: “Outcrop-core statistical comparison”).

Testing of reduced dimension variables, such as the total REE content, similarly exhibited no differences between sample types. This could indicate that surface weathering processes did not appreciably alter the REE composition. Alternatively, the small sample size leads to aggregation of the samples as “outcrops” since insufficient samples were available to compare among members of the Marcellus (e.g. Union Springs vs. Oatka Creek). This could lead to false negative test results as inter-strata variability could obscure variability due to weathering.

Application of the PERMANOVA test further confirmed the lack of difference between the two sample types in bulk REE content (\(P > 0.5\) from 10,000 permutations). While the apparent differences in dispersion or variance between the types may not be detectable given the small sample size, the similarity of medians corresponds with the findings of Chermak and Schreiber [18], where numerous, non-REE analytes agreed between core samples from different geographies within the Marcellus.

Similar results (i.e. no statistically significant differences) were obtained for uni- and multivariate comparisons between northern and southern outcrop samples. This indicates that inter-regional variability of the bulk REE composition of the shale may be less significant than intra-regional variability (i.e. at the stratigraphic or mineralogical scale). However, the current dataset is insufficient to make meaningful, statistical comparisons between stratigraphic groups.

REE profiles of these samples were variable, with enrichments of all REE weight classes—LREE, MREE, and HREE—observed in PAAS-normalized patterns (Figure 3a, b). However, most samples exhibited LREE depletion (that is they had MREE/LREE and HREE/LREE ratios >1) with MREE enrichments predominating (Figure 3b). Similarly, some samples exhibited negative Ce anomalies (Ce* < 1), but most samples had Ce and Eu anomalies near 1 (anomalies not pictured) fitting with an anoxic to sulfidic, sedimentary environments such as those proposed for the Marcellus Shale [16, 20, 26]. No statistically significant differences were observed in REE patterns as either sample type (core vs. outcrop) or sample locality (North vs. South). Taken together, these results imply that variability in the REE profiles of the Marcellus Shale is dominated at the mineral scale.

Crystalline mineralogy determined by XRD

The results of semi-quantitative XRD analyses are presented in Figure 4. The predominant crystalline mineral phases identified in these samples were quartz (classified as a major phase in 14 of 15 samples analyzed and as minor in 1 of 15), illite (10/15 major, 3/15 minor, 1/15 trace, and 1/15 non-detect), pyrite (2/15 major, 7/15 minor, 1/15 trace, and 5/15 non-detect), and calcite (6/15 major, 2/15 minor, and 7/15 non-detect). These results agree with the compilation of Chermak and Schreiber [18], who found other Marcellus samples to be predominantly phyllo- and tecto-silicates, while other gas shales (such as the Antrim and Eagle Ford) were more carbonaceous.

Comparisons among diffraction spectra and hierarchical cluster analysis of these spectra for all samples are found in Figure 5. The cluster analysis shows some potential differences between core and outcrop samples as four of six core samples cluster strongly (along with one outcrop). However, the PERMANOVA test indicates no statistically significant differences between the XRD patterns of either cores or outcrops (P > 0.1 from 10,000 permutations). If the two disparate core samples, “1-DGLS” and “C-7907” are removed from the analysis, a slightly significant PERMANOVA result is achieved (P < 0.05 from 10,000 permutations). The primary mineralogical difference between these two core samples and the other cores is the inferred presence of a major calcite phase (Figure 4). These samples (“1-DGLS” and “C-7907”) also exhibited the greatest REE profile fractionation: in Figure 3a, b these samples are the green triangle and the blue square, respectively, which exhibit significant profile fractionation.

Similarly, the results of cluster analysis provide little confidence in discernable, mineralogical differences between the regionalized outcrop samples. PERMANOVA testing confirms this observation, with no significant differences as a function of location (P > 0.5 from 10,000 permutations). However, the apparent lack of regionality (with respect to these mineralogical and elemental analyses) may be an artifact of sample size (as other geochemical parameters are known to be highly, regionally variable in the Marcellus play [5, 7]) or may arise from the pooling of samples from unique strata.

Relationships between REE profiles and mineralogy

The Mantel test was used to test for correlation between intersample distances calculated as a function of REE abundance and XRD spectra correlations. A moderate, positive correlation was observed (Spearman’s \(\rho\) = 0.53, P < 0.001), indicating that differences in the crystalline mineralogy of the samples is a significant control on REE profile variability. This hypothesis was further explored by applying Wilcoxon tests to both the degree of fractionation metric and the total REE content using the semi-quantitative XRD results for each mineral as the predictor variable.

In this analysis, the presence of major illite or calcite phases was shown to have significant, contrasting effects on the REE abundance and fractionation (Figure 6). Total REE abundance showed a strong positive correlation with illite-enriched samples (P < 0.005). The Hodges–Lehmann estimator (HL) indicates that samples with a major illite phase had approximately 98 ppm more total REE (95% CI: 39–158 ppm) than samples without a major illite phase. Additionally, samples with major illite phases were between 14 and 112% less fractionated than samples without a major illite phase (HL 95% CI; P < 0.01). The latter finding seems intuitive as the degree of fractionation is calculated relative to a composite of clayey shales (PAAS). This result also indicated that the bulk of the REE concentration is likely found in the illite (or other clay) phases of the samples or in trace phases correlated with the clays. The mechanism of REE occurrence (i.e. sorbed or structurally incorporated) in these phases is not immediately elucidated, as the separate mineralogical fractions of the shale were not directly analyzed. Abanda and Hannigan [27] found that approximately 70% of the total REE content was likely associated with the silicate/clay fraction of Utica shale samples.

Conversely, samples with more calcite were between 54 and 400% more fractionated (HL 95% CI; P < 0.005), with 6–120 ppm less total REE than samples without a major calcite phase (HL 95% CI; P < 0.05), corroborating the conclusions drawn regarding differences in core samples, where dissimilar samples had a significant calcite fraction (Figures 3, 4). LREE-depletion has been observed in carbonate fractions of shales, potentially being excluded from the crystal lattice while MREE and HREE, with more similar ionic radii to Ca, are coprecipitated [27].

These postulates are supported by analyzing correlations between the major elements of the shale (i.e. Al, Ca, Fe, K, Mg, Na, and Si; reported for the outcrop samples studied here by Dilmore et al. [28]) and the total REE content as well as the degree of REE-profile fractionation (Additional file 1: Figures S4, S5). Namely, strong positive correlations were observed between total REE content and Al, Fe, K, Mg, and Na. This supports the hypothesis of total REE correlating with clay phases. Given the general abundance of these elements in all geologic media, substantial conclusions cannot be drawn on this data alone. However, Ver Straeten et al. [17] utilized related multivariate statistics to infer mineral inputs into the Devonian Appalachian Basin. Similar to Condie [29], no correlation was observed between total REE and P, indicating that minor phosphate minerals, which can be strong REE accumulators [30, 31], did not contribute significantly to the REE content of these samples.

The implications of these hypothesized mineral associations can be related to the potential for these shales to release REE during hydraulic fracturing. Since the REE may be structurally bound within the clays (as opposed to sorbed at surface sites) [32], it is possible that produced water REE profiles will not resemble those of the bulk shale. Yan et al. [33] found the REE to reside predominantly in the fine-grained fraction of a glacial till, clayey aquitard, but associated evenly between seven mineral fractions (elucidated through sequential leaching); the REE profiles of the adsorbed and exchangeable cations fraction, which were MREE- to HREE-enriched accounting for 9–10% of the total REE in those samples [33], most closely resembled the majority of profiles observed here. Conversely, the more readily soluble fractions (such as the carbonates, which often produce LREE-depletion [27]) may be responsible for REE profiles observed in produced waters, which could be used for source identification in the event of brine intrusion or waste spillage. More study is necessary to determine the release mechanisms of the REE under conditions relevant to hydraulic fracturing and solid waste disposal.

Materials

For sample fusion, lithium metaborate (LiBO₂) was acquired from Acros Organics (99% purity; Lot # A0317552). Trace-metal grade nitric acid (HNO₃) was used for fusion dissolution and as the background solvent for ICP-MS analysis (BDH ARISTAR^® Plus, VWR; assay 69 wt%; Lot # 1113050). Single element standard solutions (~1,000 µg/L) of the REE and all elements necessary for internal and external standardization were obtained from inorganic ventures. All acid dilutions were performed on a gravimetric basis using ultrapure water (ASTM Type I, 18.2 MΩ/cm), prepared using a Barnstead NANOpure^® water purification system.

Rare earth element abundance analysis

Aliquots (~100 mg) of finely powdered sample were fused with ~1 g LiBO₂ in graphite crucibles at 1,000°C for 30 min to yield a homogenous, molten fusion. Samples were quickly removed from the furnace and poured into pre-weighed 125 mL HDPE bottles partially filled with 5% HNO₃. After all fusions were dispensed, the total volume of digestate was brought up to ~100 mL with 5% HNO₃ and weighed. Bottles were then placed into an ultrasonic bath for 2 h to break apart any remaining particulate matter.

Mineralogical analysis

Mineralogy of the shale samples was investigated by synchrotron-based X-ray diffraction (XRD). XRD measurements were made on beamline 11-3 at the Stanford Synchrotron Radiation Lightsource (SSRL) using powdered samples. Incident X-rays (\(\lambda =\) 0.9744 Å, 12,735 eV) were focused using a bent cube root I-beam Si (311) monochromator. A MAR345 area detector positioned 120 mm downstream of the sample was used to collect diffraction scans with a dwell time of 90 s. The collected images were integrated and converted into degrees 2θ using area diffraction machine (open source) software. The diffraction patterns were background subtracted and peak matched using Xpert Highscore Plus using a reference library obtained from the Crystallography Open Database, which was converted to synchrotron energy.

Based on previous analyses of Marcellus Shale samples, reported in Chermak and Schreiber [18], and examination of the patterns reported here, diffraction data were qualitatively partitioned to seven potential minerals (COD code in parentheses): quartz (1011097), calcite (9007867), dolomite (1200014), pyrite (5000115), illite (9013723), chlorite (9000158), and montmorillonite (9002779). The relative volume fraction of each mineral within this model assemblage was estimated for every sample by evaluation of several parameters including total diffraction peak intensity, goodness of peak fits, and contribution to the overall fitting. Additional mineral phases did not constitute significant fractions of the crystalline mineralogy and were not included. A list of the specific reference spectra used is given in the Additional file 1: Table S4.

Data analysis