Direct Trace Element Determination in Oil and Gas Produced Waters with Inductively Coupled Plasma‐Optical Emission Spectrometry: Advantages of High‐Salinity Tolerance

Waters co‐produced during petroleum extraction are the largest waste streams from oil and gas development. Reuse or disposal of these waters is difficult due to their high salinities and the sheer volumes generated. Produced waters (PWs) may also contain valuable mineral commodities. While an understanding of produced water trace element composition is required for evaluating the associated resource and waste potential of these materials, measuring trace elements in brines is challenging due to the dilution requirements of typical methods. Alternatively, inductively coupled plasma‐optical emission spectrometry (ICP‐OES) has shown promise as being capable of direct measurements of trace elements within PWs with minimal dilution. Here, we evaluate direct ICP‐OES trace element quantification in PWs for seventeen trace elements (As, Al, Ba, Be, Cd, Cr, Co, Cu, Hg, Mo, Ni, Pb, Rb, Sb, U, V and Zn) within fifteen PWs from five U.S. continuous reservoirs. The total analytical uncertainties associated with the trace element levels determined using ICP‐OES were estimated to be better than ± 30% (2s) except for Rb, which could not be determined due to ionisation interferences. The ICP‐OES results are compared with trace element levels determined using inductively coupled plasma‐mass spectrometry from the same samples. Our results demonstrate the potential for direct analysis of high‐salinity waters using ICP‐OES with minimal dilution and provide trace element concentrations in waters from several important U.S. petroleum‐generating reservoirs where available data are sparse.

determination in seawater (Winge et al. 1977, Sturgeon et al. 1980, Nickson et al. 1999, Dehouck et al. 2016, they generally have not been tested for the complex matrices typical of PWs, which can have salinities that greatly exceed those of seawater (~35 g l -1 ). This lack of comparability between tested methods and these sample types indicates that the quality of existing produced water data is not well known, and recent work on PWs from the Appalachian Basin by Tasker et al. (2019) has shown that trace element determinations from these sample types may often be over ± 20% different from the most probable values. This analytical uncertainty is further exacerbated by a lack of reference materials for brines, which would help validate quantitation accuracy. Given the recent and rapid development of hydrocarbon production from low-permeability reservoirs, and the concomitant production of large volumes of associated PWs, there is a need for tested, reliable methods for the determination of the trace element composition of these fluids.
The most common approach to quantify trace elements in PWs has been to dilute samples up to and exceeding a factor of 10000, creating a sample matrix similar to surface waters, followed by analysis via inductively coupled plasmamass spectrometry (ICP-MS; Hayes 2009, Dresel and Rose 2010, Lauer et al. 2016, Peterman et al. 2017, Nicot et al. 2018, Tasker et al. 2019. Such an approach is required due the deleterious effect of high concentrations of dissolved solids on the mass spectrometer ion beam, and dilution additionally removes many of the analytical problems associated with trace element quantification in PWs (e.g., matrix interferences). However, the large dilution factors needed for high-salinity sample analysis results in poorer detection limits relative to lower salinity waters. Other approaches include pre-concentration of elements and separation from the sample matrix via ion exchange and column chemistry (Kamata et al. 1987, Jackson and Mahmood 1994, Nickson et al. 1999, Silva et al. 2015 or cloud point extraction (Bezerra et al. 2007), but such methods are not practical for more than several elements at a time.
An alternative approach has been suggested for the determination of trace elements in seawater involving analysis at full or near full salinity by inductively coupled plasmaoptical emission spectrometry (ICP-OES) with the use of an argon humidifier and high-salinity nebuliser (Chausseau andLebouil 2011, Spanos 2014). Although ICP-OES has significantly lower instrumental sensitivity and detection limits than ICP-MS, the robust design of ICP-OES can accept highsalinity samples up to and exceeding the salinity of seawater without dilution. In this approach, the lower resolution of the ICP-OES is balanced by the minimal sample dilution required. While there are several reports on the analysis of PWs or PWaffected surface waters using ICP-OES, these studies have generally focused on lower salinity samples [≤ 1 g l -1 total dissolved solids (TDS)] (Pancras et al. 2015) or have analysed for only a limited number of trace elements (Bezerra et al. 2007, Penha et al. 2015, Souza et al. 2017). However, a published ICP-OES method applied to PW samples with salinities greater than seawater (> 35 g l -1 ) for a broad suite of trace elements is lacking. Thus, this work represents the first effort to present a tested method for the quantification of trace elements in high-salinity PWs from a broad suite of U.S. oil and gas reservoirs by ICP-OES.
Here, we discuss findings from a comparative investigation to detect and quantify concentrations of seventeen trace elements (As, Al, Ba, Be, Cd, Co, Cr, Cu, Hg, Mo, Ni, Pb, Rb, Sb, U, V and Zn) in PW samples from five continuous (low permeability) U.S. petroleum reservoirs by ICP-OES utilising an argon humidifier and a high-salinity nebuliser (to prevent clogging of the nebuliser during analytical runs). These results are compared against ICP-MS trace element analyses of the same samples demonstrating the validity of an ICP-OES approach as a rigorous analytical method for the quantification of trace elements in high-salinity PWs. Additionally, we present information on the trace element composition of PWs from several important U.S. low-permeability petroleum reservoirs for which (aside from the Marcellus Formation) there are limited data on trace element concentrations in the literature.

Experimental Samples
Three PW samples from each of the Eagle Ford (Gulf Coast Basin), Marcellus (Appalachian Basin), Middle Bakken/Three Forks (Williston Basin), Wolfcamp/Cline (Permian Basin) and Utica/Point Pleasant (Appalachian Basin) reservoirs that spanned the available sample salinity range (i.e., one low, one medium and one high-salinity sample each) were selected for analysis from the U.S. Geological Survey (USGS) PW sample catalogue (Table 1; Figure 1). The fifteen selected samples were all collected during previous USGS field campaigns, filtered to 0.45 µm and acidified in the field to a pH of < 2 with ultra-pure HNO 3 (Fisher Scientific, https://www.fishersci.com), and stored at 4°C in high-density polyethylene bottles, as described previously (e.g., Engle et al. 2016). All samples were analysed within three years of collection. TDS were determined for each sample following U.S. Environmental Protection Agency (EPA) Method 160.1 (1999) using gravimetric analysis following evaporation. The samples span a TDS range of 17-370 g l -1 ; this range is inclusive of nearly all PWs from the United States (Blondes et al. 2018). While there are several studies reporting PW trace element concentrations for the Marcellus (Hayes 2009, Pancras et al. 2015, Ziemkiewicz and He 2015, Tasker et al. 2019, there are only a few reported PW trace element concentrations for the Bakken/Three Forks (Lauer et al. 2016, Peterman et al. 2017 and Eagle Ford (Maguire-Boyle and Barron 2014, Nicot et al. 2018 and no publicly available data for PWs from the Wolfcamp/ Cline and Utica/Point Pleasant reservoirs. Inductively coupled plasma-optical emission spectrometry ICP-OES analyses were performed in the Brine Research Instrumental and Experimental (BRInE) laboratory (https:// usgs.gov/centers/eersc/science/oil-and-gas-waters-project) at the USGS National Center in Reston, Virginia. All analyses were conducted on a Horiba Ultima Expert ICP-OES, which has a vertical torch with a radial plasma view geometry and is equipped with a high TDS sample introduction system consisting of a Meinhard type K nebuliser, a cyclonic spray chamber and an argon humidifier. Ultra-high purity (UHP) grade argon (Ar) was used as the plasma, sheath and nebulisation gas and was used as received. UHP grade nitrogen (N 2 ) was used to purge the optical path and detector compartment. The ICP-OES detector gain and voltage were optimised for detection of a 1000 μg l -1 standard solution (Table 2). The atomic emission lines used for the trace element determinations by ICP-OES are given in Table 3 and were primarily taken from EPA Method 200.7 (1994a). Different emission lines were used for the quantification of Ba, Be and Cd than those listed in U.S. EPA Method 200.7 (1994a) due to the superior performance (i.e., higher precision) observed here when using the alternative lines; no suggested emission lines are given for Rb and U in Method 200.7. During an experimental run, each emission line was scanned with nine points with~0.5 s acquisition per point and the maximum intensity of the peak centre was recorded. Each emission line was measured in triplicate, and the standard deviations for each recorded emission line intensity were typically < 2%. The emission line intensities were baseline corrected using the mean intensity of five laboratory  blanks (50 g l -1 Na-Ca-Cl solution in 2% HNO 3 ; discussed below) measured at that wavelength.
Trace element concentrations were determined using the mean maximum intensity for each element emission line following baseline correction and a linear regression fit to a log-log standard calibration curve. The log-log calibration method was used to prevent violating the assumption of homoscedasticity (homogeneity of variance) in normal least squares regression and was selected over the more common weighted least squares regression as the weights are difficult to estimate when using only a few (~5) calibration points (Ryan 2009). When prediction is the primary objective of regression analysis, homoscedasticity is perhaps more critical than other regression quality metrics because it ensures that precision of estimation is equal across the calibration range (Reimann et al. 2008). In this study, the ICP-OES calibration curves typically spanned the 10-5000 μg l -1 concentration range. Standard solutions containing all seventeen trace elements were prepared to the desired concentration from single element certified 1000000 μg l -1 stock solutions in 2% HNO 3 purchased from High Purity Standards Inc. (Charleston, South Carolina).
Samples were minimally pre-diluted (Table 1) to a target TDS of 50 g l -1 prior to ICP-OES analysis such that salinity was relatively constant between samples and could be matched to the calibration materials. The samples were not digested prior to analysis to remove dissolved organic carbon, as has been suggested for PW analysis by ICP-OES by several previous researchers (e.g., Pancras et al. 2015, Penha et al. 2015. Nearly, all natural PWs with a salinity > 10 g l -1 are dominated by the ions Na, Ca and Cl (Kharaka and Hanor 2014). As salinity of PWs increases, the average Na/Ca ratio tends to decrease (Figure 2; Blondes et al. 2018). For instance, at 100 g l -1 salinity the mean molar Na/Ca ratio is approximately 10, but at 300 g l -1 , it is typically closer to 4. As the molar ratio of elements within the PW samples is unaffected by dilution, samples diluted down to 50 g l -1 for ICP-OES analysis will have a different cation matrix depending on their original salinity. To account for these shifts, the standard solutions used to generate calibration curves were tested in Na-Ca-Cl amended solutions using Na/Ca molar ratios of both 4 and 10. The Na-Ca-Cl solutions used were prepared from certified high purity salts (ACS grade, > 99.0% metals basis or better).
It is important to note that the salts used to generate the Na-Ca-Cl background can contain trace metal impurities, up to 5 mg kg -1 , primarily as Ba and Pb. To check the influence of metal impurities from the Na-Ca-Cl background on our results, we compared calibration curves generated in Na-Ca-Cl backgrounds prepared from solid salts with curves generated in the presence of a custom, certified 99.999% pure Na-Ca-Cl blend (Na/Ca = 10) purchased from High Purity Standards Inc. (Charleston, South Carolina) and used without further purification. Here, no significant differences were observed between calibrations carried out in the different Na-Ca-Cl backgrounds, and as such all further measurements were done with backgrounds prepared from solid salts.

Inductively coupled plasma-mass spectrometry
All ICP-MS analyses were carried out by an accredited commercial laboratory (ISO 17025). This laboratory required sample TDS levels to be less than 0.05% by mass (~0.5 g l -1 ) necessitating sample dilution from~409 to 10009. Samples were sent to the laboratory both as aliquots that had been previously diluted to a target TDS of 0.4 g l -1 (dilution factors provided in Table 1) and as minimally diluted aliquots (typically 1-39). Samples were sent in two batches to control for any biasing factors from the sample dilution step between the two laboratories. Pre-diluted PWs were sent in triplicate, while the minimally dilute PW samples were sent in duplicate due to sample volume limitations. Three additional Eagle Ford PWs and one Wolfcamp/Cline PW were included in the minimally dilute sample set. The specific dilution factors used by the commercial laboratory to bring the minimally diluted samples into the ICP-MS analytical range are unavailable; however, it is reasonable to assume that these were the minimum dilutions required to achieve the required 0.5 g l -1 TDS concentration. Regardless, as samples analysed via ICP-MS were heavily diluted (either by USGS staff or by the commercial laboratory), outside of isobaric interferences (which would be identified through matrix spike response), variation in Na/Ca ratios in the samples is unlikely to affect results and thus did not warrant matrix-matching of the reference materials.

Quality control
To determine performance of both the ICP-OES and ICP-MS analyses, several quality control (QC) samples were included with the sample batches. These included blanks, spike-amended blanks (blank spikes) and aliquots from five of the produced water samples spiked with known masses of target elements (matrix spikes). Typical blank and matrix spike concentrations are provided in Table S1. Furthermore, although no reference materials for trace elements in PWs or brines are available to our knowledge, appropriate reference surrogates were chosen based on the salinity of the post-diluted samples that were analysed using both the ICP-MS and ICP-OES. In the case of ICP-MS analyses where the target salinity was relatively low (0.4 g l -1 ), surface water trace USGS reference samples (herein referred to as USGS T reference materials, USGS T-193 and USGS T-225, respectively) were selected (https://bqs.usgs.gov/srs/). For the ICP-OES analyses, a simulated seawater reference material (High Purity Standards Inc., Charleston, South Carolina) was used as the surrogate reference sample as it has a similar salinity (~35 g l -1 ) and major ion composition to that of many PWs (see Table S2). These reference surrogates were used as received without further purification.
A variety of metrics were used to quantify sample quality from the various results. Analytical precision for the ICP-MS results was estimated from the relative standard deviation (RSD) from triplicate analyses. ICP-OES precision was determined by propagating the standard deviation of the triplicate intensity measurements for each sample at each analytical wavelength (Table 3) with the precision of the linear regression fits to the calibration curve data. The ICP-OES measurement precision determined in this way was typically better than ± 5% at the 2s level of uncertainty. Additionally, the RSD from duplicate ICP-OES measurements provides another measure of analytical precision. Analytical accuracy for both methods was estimated from percent recovery from matrix spiked samples as the use of matrix spikes informs both analytical accuracy and provides information on potential matrix interferences. In our opinion, the use of matrix spikes is the best measure of total uncertainty for PW trace element determinations using both ICP-OES and ICP-MS because no certified reference materials exist, to our knowledge, for these materials. Here, the total uncertainties for the ICP-OES measurements are estimated to be ± 30% (2s) based on matrix spike recoveries for all trace element analytes, except for Rb as discussed below.

ICP-OES quantification of trace elements in produced waters
Trace element determinations from PWs may contain artefacts caused by the variation in the Na/Ca ratio between samples with different initial salinities. To evaluate the presence of any such matrix effects, the nominal relative per cent differences (RPDs) were calculated between the ICP-OES response at each analytical wavelength (Table 3) from two standard solutions (containing seventeen trace elements at 1000 μg l -1 ) in the presence of a 50 g l -1 Na-Ca-Cl amended background with a Na/Ca molar ratio of four and ten, respectively. The RPDs observed for the ICP-OES response from all analytical wavelengths used for the seventeen trace element determinations between the two solutions with varying Na/Ca molar ratios were generally within ± 10%, which is less than the total uncertainty of the ICP-OES measurement. This indicates that the Na/Ca molar ratio of the produced water samples is not a critical parameter to match when making calibration standard solutions for PWs with varying TDS levels. For this reason, all trace element determinations in this study were done using calibration standard solutions in the presence of a 50 g l -1 Na-Ca-Cl background with a Na/Ca molar ratio of 10 as this ratio is representative for a large number of U.S. PWs (Blondes et al. 2018).
Estimated method signal detection limits (MDLs) for each analytical wavelength used in the trace element quantifications were determined according to Equation 1: where y MDL is the instrument signal detection limit; s MDL is the standard deviation from replicate measurements (n = 7) of a standard solution containing all seventeen trace elements at a concentration of 50 μg l -1 in the presence of a 50 g l -1 Na-Ca-Cl background solution with a Na/Ca molar ratio of ten; and 3.14 corresponds to the value of Student's t-test for six degrees of freedom at the 98% confidence interval (Harris 2003). The element-specific limits of quantitation (LOQs) were determined for each analytical wavelength by multiplying s MDL by a factor of 10. The MDLs and LOQs for each analytical wavelength used in this study are given in Table 3. The MDLs and LOQs determined here are similar to other reported ICP-OES detection limits for trace elements within PWs (Pancras et al. 2015, Penha et al. 2015 and only slightly larger (~29) than those typically reported for trace elements in low-salinity waters determined by ICP-OES (U.S. Environmental Protection Agency 1994b). Differences from U.S. EPA Method 200.7 (1994a) can be attributed to signal interferences from the high Na-Ca-Cl background (~50 g l -1 ) that were amended to the reference materials in order to match the salinity of the produced water samples.
The trace element MDLs in the presence of a 50 g l -1 Na-Ca-Cl background were generally within the 1-30 μg l -1 range and the corresponding respective LOQs encompassed the 6-80 μg l -1 concentration regime. For uranium, the MDL determined was notably higher, 100 μg l -1 . This is attributed to well-known interferences from major elements (e.g., Ca, Mg, etc.) present in the PWs (Kamata et al. 1987, Santos et al. 2010). Determination of Rb by ICP-OES in the presence of a high-salinity background containing large amounts of Na is greatly influenced by ionisation effects (Luecke 1991, Olesik 1991, Morishige and Kimura 2008. These deleterious effects prevent the quantification of Rb in the PWs examined by ICP-OES. Furthermore, the MDLs determined here are low enough to quantify As, Ba, Be, Cd, Cr and Cu in PWs at the U.S. EPA national primary drinking water standard level (U.S. Environmental Protection Agency 2009). For elements Hg, Pb, Sb and U, the ICP-OES MDLs are 59, 29, 1.7 9 and 3.3 9 times larger, respectively, than the US EPA regulated maximum contaminant level (U.S. Environmental Protection Agency 2009). There are no U.S. EPA primary drinking water reference materials for Al, Co, Mo, Ni, V or Zn.
The principal measure of ICP-OES analyses quality was the stability of the calibration curves as assessed through the use of blank spikes analysed in the concurrent experimental run as the samples. Typical blank spike recoveries for all trace elements, excluding Rb, were within the 85-115% range indicating the calibration curves were stable (U.S. Environmental Protection Agency 1994b). Other QC samples that were used to assess the performance of the ICP-OES method included a surrogate reference material, that is, a simulated seawater with a TDS level of 35 g l -1 , as no certified reference materials are available, to our knowledge, for high-salinity PWs and matrix spikes. Complete ICP-OE QC results for all PWs are provided in Tables S3-S7. The certified seawater reference material used here only had concentrations greater than the determined ICP-OES MDLs (Table 3) for Al, As, Ba and Cu (see Table S2). Recoveries of Al and Cu from the seawater reference material were typically within ± 15% of the certified concentrations, well within the estimated uncertainty for the ICP-OES method used here, while the recoveries for As and Ba were notably worse, typically ± 50% of the certified concentrations. The source for the poor recoveries for As and Ba from the simulated seawater reference material is unclear.
Matrix spike recoveries for every trace element were within the range 70-130% and were often much better, within 90-110%. Given the lack of reference materials currently available for high-salinity PWs, matrix spikes are key to assessing the accuracy of the measured trace element concentrations by ICP-OES and they additionally reveal the influence that the matrix may have for the concentration determinations. Here, the matrix spike recoveries indicate that the PW matrix up to the 50 g l -1 TDS level does not significantly affect the determination of any trace element by ICP-OES using the method applied in this study, similar to observations made by Penha et al. (2015). While other authors have shown that digestion of PWs can be an important step prior to ICP-OES analysis (e.g., Pancras et al. 2015, Penha et al. 2015, the typical matrix spike recovery achieved here suggests that sample digestion was not critical for the direct detection of trace elements using ICP-OES for the PWs analysed in this study, even for samples with dissolved organic carbon levels approaching~500 mg l -1 . Results from the ICP-OES analyses of the fifteen produced water samples are provided in Table 4. For the trace elements undetected in the samples, the concentrations reported correspond to the MDL of that analyte multiplied by the dilution factor used to bring the sample to an approximate TDS of 50 g l -1 . This value represents the upper-limit concentration for the respective trace element within the individual PWs. Elements detected in the PWs above the MDL but below the LOQ have been flagged as semiquantitative. Perhaps unsurprisingly, given differences in their depositional environment and regional geologies, the PW trace element concentrations determined with ICP-OES vary widely between the five reservoirs investigated. While Be was the only element undetected in all PWs tested, Cr, Hg and V were either undetected or below the LOQ for the majority of samples. Ba was the only element detected in every sample and exhibited the highest concentrations detected. For PWs from the Eagle Ford, Marcellus and Utica/Point Pleasant reservoirs, the Ba concentrations represent a lower-bound estimate as the measured values exceeded the highest calibrator used (5000 μg l -1 ).
Other elements detected in a wide sub-set of samples included Cu, Mo, Pb, Sb, U and Zn. The detection of Ba, Pb, Sb and U at concentrations that exceed the EPA primary drinking water standards (U.S. Environmental Protection Agency 2009) is of special note. Additionally, there are two observations in the data given in Table 4 that should be highlighted: (a) the high levels of Zn (≥ 20 mg l -1 ) detected in all PWs from the Middle Bakken/Three Forks reservoir; and (b) the relative absence of any trace elements in PWs from the Wolfcamp/Cline reservoir. While the controls on base metal concentrations in PWs are not well understood, enrichment of these elements (up to and exceeding 100 mg l -1 ) typically occurs in very saline brines (Kharaka et al. 1987). This observation tends to agree with our results; the samples from the Middle Bakken/Three Forks reservoirs had the highest TDS levels of all samples in the study. For the Wolfcamp/Cline PWs analysed here, only marginal levels of Ba, Mo and Zn were detected. No literature values, to our knowledge, are available for trace elements within PWs from the Wolfcamp/Cline reservoir, although Engle et al. (2016) reported upper-limit Ba values of < 20 mg l -1 for Wolfcamp brines, which prohibits any comparison with our results. However, the limited trace elements observed for these samples here signify that PWs from the Wolfcamp/Cline may exhibit a relatively simple geochemical fingerprint compared with other PWs that may be indicative of their origin. For instance, Engle et al. (2016) argued that although Wolfcamp/Cline PWs originated as evaporated palaeo-seawater, there is evidence that the ions diffused into the shale at some point early in its hydrogeological evolution. In this case, many of the trace elements may have been transport-limited, owing to low concentration gradients and diffusivity constants.
The trace element concentrations determined with ICP-OES in this study are comparable to previously reported PW trace element concentrations from the Bakken/Three Forks (Lauer et al. 2016, Peterman et al. 2017, the Eagle Ford (Nicot et al. 2018), and the Marcellus (Hayes 2009, Pancras et al. 2015, Ziemkiewicz and He 2015, Tasker et al. 2019. With the exception of Pancras et al. (2015), the previously reported PW trace element levels were determined using ICP-MS, although the values presented in Tasker et al. (2019) represent most probable values from an intermethod comparison where both ICP-OES and ICP-MS approaches were used. The similarity between the trace element levels determined here using ICP-OES from solutions at a 50 g l -1 TDS level with these previously published PW trace element concentrations determined using ICP-MS suggests that the two techniques may be performing comparably in the quantification of trace elements in PWs.

Comparison with ICP-MS
For both sample sets (i.e., the pre-diluted to~0.4 g l -1 PWs and the minimally diluted PW aliquots) analysed via ICP-MS, calibration curves created by the commercial laboratory were not provided for further analysis. Similarly, the isotopes used in the determination of each element were not provided, even when specifically requested. Regardless, reasonable guesses about the masses used can be made based on the relative abundance of isotopes, known isobaric interferences (May and Wiedmeyer 1994), and published methods for the quantification of trace element in water samples by ICP-MS (U.S. Environmental Protection Agency 1994a).
Despite attempting to maximise the likelihood of trace element detection in the samples by pre-diluting near the maximum acceptable salinity limit required by the commercial laboratory, a majority of the data (~75%) was reported below the provided MDLs (Table S8). For the ten samples submitted that were closer to their original salinities (i.e., the minimally diluted aliquots), this was also the case, although higher proportions of samples were quantified for Al, Cu and U. The similarities in the number of non-detections between the pre-diluted PW sample set versus the minimally diluted sample set submitted to the commercial laboratory indicate that the dilution procedure used to bring the samples into a TDS range analysable with ICP-MS had a minimal impact (if any) on the results presented here. Given the number of nondetections within the ICP-MS results, these data are reported in aggregate (Table 5; i.e., results for individual reservoirs are not discussed). Only Ba, Pb, Rb and Zn exhibited detectable concentrations in more than 50% of the samples for both samples sets; however, Al, Ni and Cu were also detected for a number of PW samples. This is likely a result of large dilution factors (Table 1), despite relatively low MDLs of ICP-MS (U.S. Environmental Protection Agency 1994a). Other authors have been successful in producing trace element data in PW samples from some of the same reservoirs in this study (Peterman et al. 2010, Ziemkiewicz and He 2015, Akob et al. 2016, Lauer et al. 2016, Peterman et al. 2017, Nicot et al. 2018, Tasker et al. 2019, but only rarely is this achieved by commercial laboratories. Although the exact reason for the low proportion of non-qualified data is unknown, research laboratories may be more willing to run samples at higher salinity and/or can invest time to determine sample set-specific detection limits which may be lower than the multiple-application MDLs often generated by commercial laboratories. Additionally, sample heterogeneity or associated dissolved organic matter within the PWs may have deleteriously affected the ICP-MS analyses. Precision for the ICP-MS analysis of the PW samples, as estimated from RSDs from triplicate analysis of each sample, showed fairly high uncertainty in some samples for Al (up to 152%), Ba (up to 76%), Ni (up to 63%), Pb (up to 139%) and Zn (up to 100%), although the median RSDs were more reasonable (typically < 30%).
Results from the analysis of USGS T-194 and T-225 reference materials by ICP-MS show that most elements reported values with 10% of the most probable value (MPV), or barring that, within two F-pseudo sigma of the MPV (see Table S9). Only results for Be and Zn fell outside of these bounds for both reference materials. These results suggest that while there were some deviations in accurate quantification of trace elements in the PWs, there were no major systematic problems with measurement accuracy. Comparing the ICP-MS results for the reference materials with the ICP-MS observations for the PWs (Table 5) further supports the finding that the large dilution factors needed to bring the samples within the ICP-MS method TDS range may inhibit detection for trace elements within PWs.
Additional insight into ICP-MS analysis accuracy and issues of interferences from the sample matrix can be examined through response to matrix spikes. Recoveries for 15 matrix spikes are summarised in Table S10. Determination of matrix spike recovery is difficult to assess for both sample sets (i.e., minimally diluted and pre-diluted) as the majority of the un-spiked samples exhibited concentrations below detection limits for most elements (Tables 5 and S8). This signifies that the un-spiked concentrations are not known and cannot be accurately estimated. We have adopted the conservative assumption that, for cases where the trace element analyte is below the MDL, the un-spiked concentrations are assumed to be equal to the upper-limit concentrations provided by the commercial laboratory, most likely equal to the dilution factor used multiplied by the MDL. This approach results in potential negative biases for the matrix spike recoveries. Fourteen elements (Al, As, Ba, Be, Cd, Co, Cu, Mo, Ni, Rb, Sb, U, V and Zn) in the two sample sets produced spike recoveries between 80 and 120%, suggesting acceptable accuracy and minimal impact from matrix interferences for the majority of analytes. Of the remaining analytes, Pb had matrix recoveries consistently > 100%, while both Cr and Hg had matrix recoveries consistently < 100%. Potential isobaric interferences exist for Pb (May and Wiedmeyer 1994); 206 Pb + has the same m/z as 190 Pt 16 O + , 207 Pb + has the same m/z as 191 Ir 16 O + and 208 Pb + has the same m/z as 192 Pt 16 O + . However, the abundance of Pt and Ir within PWs is expected to be low (<< 1 μg l -1 ; Li 1991) and, as such, the source of the anomalous high matrix spike recoveries for Pb is unclear. Finally, the consistently low recoveries for Cr and Hg in the matrix spikes suggest that the sample matrix may have been degrading sample accuracy or other problems with the ICP-MS analyses were encountered.

Conclusions
This work details efforts to develop an ICP-OES based method to measure a suite of seventeen trace elements (As, Al, Ba, Be, Cd, Co, Cr, Cu, Hg, Mo, Ni, Pb, Rb, Sb, U, V and Zn) within PWs originating from five different oil and gas producing continuous reservoirs (see Table 1; Figure 1). Furthermore, the ICP-OES results have been compared with the trace element concentrations determined with ICP-MS by a commercial laboratory within the same PW samples. The majority of the trace elements under study were detectable and quantified directly within the fifteen PW samples following minimal (if any) dilution using the ICP-OES method, with the exception of Rb. Rubidium quantification with ICP-OES in the presence of high-salinity matrices suffers from ionisation effects similar to observations for potassium. The detection limits determined for the ICP-OES method in the presence of a 50 g l -1 Na-Ca-Cl background are adequate to detect all trace elements under study at concentrations generally ≥ 30 μg l -1 , with the exception of Rb and U (Table 3). In contrast, the ICP-MS analyses were unable to detect most trace elements within the PW samples (Table 5), which is fairly surprising given the ICP-MS method's high sensitivity (low detection limit). The elements detected in the PWs with ICP-MS included Ba, Pb, Rb and Zn; these represent the most prevalent elements within the PWs sampled.
The precision and accuracy of the two methods were assessed by using and comparing the RSDs of replicate sample measurements along with reference materials and matrix spikes, respectively. If the precision uncertainties for the two methods are compared for the three elements that both methods were able to detect (Ba, Pb and Zn), the median precision estimates are typically < ± 50%, and in many cases < ± 10%, for both methods. However, for ICP-MS the range of precision estimates observed for these elements is large, up to ± 100%, whereas the precision estimate ranges determined with ICP-OES are typically less. The reasons for the higher precision uncertainties found when using ICP-MS remain unclear but may include sample heterogeneity, instrument drift or poor precision during sample dilution and handling.
Recovery of the relevant trace elements for two surrogate reference materials by ICP-MS was typically within ± 15% of the most probable concentration values indicating that this degree of uncertainty may represent the accuracy for the ICP-MS approach. Further testing of the ICP-MS method through the use of matrix spikes showed that the degree of uncertainty associated with accuracy may be actually > ± 15% as median recoveries for one third of all trace elements were outside of this range. As there are very few commercially available trace element reference materials with TDS levels approaching the 50 g l -1 level (only one to our knowledge: certified reference seawater which is still only 35 g l -1 ), the accuracy estimates for the ICP-OES method were entirely determined through the use of matrix spikes. For the produced water samples with verified calibration curves, the matrix spike recoveries were observed to be within the 85-115% range. This indicates that an estimated accuracy of ± 15% at the 1s level of uncertainty is appropriate for trace element determination in these produced water samples with the ICP-OES approach.
Taken together, the results presented here indicate that ICP-OES analysis using a high-salinity nebuliser and an argon humidifier is an appropriate approach for determining trace element concentrations within high-salinity PW samples. In this context, the ICP-OES method outperforms the more common ICP-MS approach in terms of trace elements detected directly within PWs spanning the~17-370 g l -1 TDS range with similar levels of experimental precision and accuracy. Of important note is that the ICP-MS results presented here are based on analyses carried out by a commercial laboratory, which may not routinely maximise brine-specific analytical performance that is typical of a research laboratory. Our laboratory is continuing efforts to refine the performance of the ICP-OES approach detailed here, and while this methodology performs well for a majority of PW samples, there are occasional failures (e.g., As, Cu, Pb and Sb for the Marcellus PWs). The underlying reasons behind these failures have yet to be identified and are the focus of our future efforts in this area. Finally, this study provides trace element concentrations for PWs from several important U.S. petroleum-generating continuous reservoirs where available data are sparse (e.g., the Middle Bakken/ Three Forks, Marcellus and Eagle Ford) or completely lacking (e.g., the Wolfcamp/Cline and Utica/Point Pleasant). These data should be useful to future researchers evaluating the environmental impacts and resource potential of wastewaters from these petroleum reservoirs.

Supporting information
The following supporting information may be found in the online version of this article: Table S1. Concentrations of the blank and matrix spike QA/QC samples used in the ICP-OES) and ICP-MS analyses. Table S2. Reported concentrations for surrogate reference materials used in this study. Table S3. Aggregate ICP-OES measurement results for produced water samples MBTF-1-3. Table S4. Aggregate ICP-OES measurement results for produced water samples EF-1-3. Table S5. Aggregate ICP-OES measurement results for produced water samples M-1-3. Table S6. Aggregate ICP-OES measurement results for produced water samples WC-1-3. Table S7. Aggregate ICP-OES measurement results for produced water samples UPP-1-3. Table S8. Trace element concentrations for produced water samples determined using ICP-MS. Table S9. ICP-MS results for USGS T-193 and T-225 reference materials. Table S10. Summary of ICP-MS results for matrix spikes from both pre-diluted and minimally diluted sample sets.