The Bakken Air Quality Study (BAQS) was conducted to assess the mix of pollutants impacting national parks and Class I areas in the Bakken region. Although elevated pollutant levels can occur anytime of the year, measurements were focused primarily on winter months. The first BAQS study period was in 2013, with measurements from 15 February to 6 April. The study was conducted at five field sites: FOUS, KNRI, MELA, THRO-N and THRO-S. THRO-N served as the core sampling site. At the core site, high time resolution measurements were made of NOx, CO, Total Reactive Nitrogen, O3, SO2, black carbon and aerosol light scattering. More extensive data were also obtained at lower time resolution (6 h–1 week) of organic and inorganic composition of particles and gases. The other four sites were not as heavily instrumented. FOUS, MELA and KNRI had 48 h integrated samples (6 days a week) of inorganic gas and particulate composition, real time ozone measurements, automated precipitation samplers and Radiello passive samplers, which measured weekly integrated concentrations of SO2, NO2, NH3 and O3. A nephelometer was also deployed at KNRI. Because THRO-S is heavily instrumented through state and federal monitoring programs, only passive samplers were deployed at this site. Meteorological data were available at all sites. In addition to the sampling sites, 2 days of measurements of methane and VOCs were made using a mobile laboratory. A detailed list of measurements from the first study period is given in Table 1.

The second study period ran from 23 November 2013 through 28 March 2014, encompassing the largest increasing trends in sulfate and nitrate as determined from IMPROVE observations (Hand et al., 2012a). During the second study, measurements were limited to three sites, with increased emphasis on higher time resolution data collection. THRO-N remained the core site, while FOUS and MELA served as satellite sites. At THRO-N and FOUS, additional measurements of gas and particle concentrations and compositions were made, including VOC measurements (Table 2). The VOC data provide markers for many of the potential air pollutant sources in this region. Mobile measurements were also conducted (see Sect. 2.2).

These study periods correspond to months when temperatures are typically below freezing, and where minimum winter temperatures can fall below -30 ∘C. Based on meteorological data collected during 2002–2013 at Watford City (near THRO-N), MELA, and LOST, the predominant wind direction in the study region was southwesterly and the second most common direction was northwesterly, though airmasses can arrive from all directions. There are spatial, diurnal, and seasonal fluctuations around this predominant pattern. Seasonally, air masses from the northwest are most common during fall and winter; this was generally observed during BAQS. Transport from easterly directions is most likely during spring and summer. Average wind speeds in winter were in the range of 3–5 m s-1 at all of the study sites, with Watford City (near THRO-N) having the slowest and LOST the highest mean speeds.

Many of the measurements listed in Tables 1 and 2 will be described in detail in forthcoming publications, and so are not discussed further. Here we provide a brief description of measurements from the Results and Discussion section.

For real time measurements of NOx and SO2 during the first study period, sampling was from a common inlet ∼ 3 m above ground level. The sampling line was 0.64 cm OD Teflon tubing. For SO2 from the real time measurements, a calibration was performed prior to the study. For NOx, calibrations were conducted daily using certified, traceable standards provided by Airgas (Prenni et al., 2014). Every calibration included zero air and a span concentration, with calibration gases introduced at the sample inlets.

NOx measurements were made using a chemiluminescence instrument (Teledyne 201E). The technique alternately measures NO directly and measures NOx by first converting NO2 to NO using a molybdenum converter. NO is reacted with ozone forming NO2 in an excited state which emits radiation while decaying to the ground state.

Real time SO2 measurements were made during the first study period using a Thermo Scientific SO2 Analyzer (Model 43C), which uses pulsed fluorescence and has a detection limit of 1 ppbv (60 s averaging). During the second study period, and during both studies at the satellite sites, SO2 concentrations were also derived from University Research Glassware (URG) samplers.

Twenty-four hour samples were collected using URG annular denuder/filter-pack samplers from 8 a.m. to 8 a.m. local time at THRO-N during both study periods. During the first study period, 48 h samples were also collected at FOUS, MELA and KNRI, covering 6 days per week. During the second study, 24 h samples were collected at FOUS, and weekly samples were collected at MELA. Extracted samples were analyzed for inorganic gas and particulate species using ion chromatography (IC). Sample collection and analysis procedures were similar to those described elsewhere (Benedict et al., 2013).

Real time black carbon (BC) data were collected using a multi-wavelength aethalometer at the core site (Magee Scientific AE-31). We follow the recommendation of Petzold et al. (2013) in designating aethalometer measurements as BC and measurements from the IMPROVE program as EC. The sample is collected on quartz fiber filter tape and absorption is measured at seven wavelengths from 370–950 nm. For this study, a PM2.5 inlet was used and BC mass was determined as the mean of the masses measured from all wavelengths; no further corrections were implemented. Aethalometer data were logged as 5 min averages. The instrument was factory calibrated prior to the first study period and has a sensitivity of < 0.1 µg m-3.

The THRO-N site also had an IMPROVE particle monitor that collected 24 h samples. Samples were collected daily, on the same schedule as the URG samplers (8 a.m. to 8 a.m.). Modules A, C, and D were used during the study. Modules A and C collect fine particles (PM2.5), while module D collects both fine and coarse particles (PM10). Module A is equipped with a Teflon^® filter that is analyzed for PM2.5 gravimetric fine mass, elemental concentration, and light absorption. Module C utilizes a quartz fiber filter that is analyzed by thermal optical reflectance (TOR) for organic carbon and EC. Module D utilizes a Teflon filter to determine PM10 aerosol mass concentrations gravimetrically. Module A was used during the first study, and Modules A and C were used during the second study period. Module D was used for a limited time during the second measurement period.

For VOC measurements, whole air samples were collected at THRO-N, FOUS, and MELA, as well as at various locations throughout the Bakken region as part of the mobile measurements. During the second study period, samples were collected into evacuated 2 L passivated, stainless steel canisters. A total of 40 individual VOCs were quantified from the canister samples using a five-channel, three gas chromatograph (GC) analytical system which employed three flame ionization detectors (FIDs), one electron capture detector (ECD) and one mass spectrometer (MS). The gases analyzed included C2–C10 nonmethane hydrocarbons (NMHCs), C1–C2 halocarbons, C1–C5 alkyl nitrates and reduced sulfur compounds. The analytical system and methodology are similar to those used in previous studies (Russo et al., 2010a, b; Sive, 1998; Swarthout et al., 2013, 2015; Zhou et al., 2010). Multiple whole air standards were used during sample analysis (analyzed every 10 samples). The measurement precision, represented by the relative standard deviation of the peak areas for each compound in the whole air standards, was 1–8 % for the NMHCs, 3–10 % for the halocarbons, 3–8 % for the alkyl nitrates and 3–5 % for the sulfur compounds. For the second study period, a canister sample was collected twice per day at THRO-N, four times per week at FOUS, and once per week at MELA. For approximately 1 month of the study (19 December 2013–31 January 2014), canisters were collected only once per week at FOUS.

Meteorological data were collected with a Climatronics All-In-One Weather Sensor (Part Number 102780), co-located with the gas measurements.

Before presenting measurements from these studies, we first examine monitoring data during the two time periods of the intensive field campaigns to determine if the measurement periods were typical for the region. To this end, we use long-term air monitoring data from the region (EPA AirData: https://ofmext.epa.gov/AQDMRS/aqdmrs.html). In Fig. 2, box plots of mean daily values of SO2 and NO2 concentrations at THRO-N are presented for all available data since 2000. Data are also shown separately for each month of the BAQS study periods. As shown in Fig. 2, there was significant variability for SO2 and NO2 during the study periods; e.g. daily average concentrations of NO2 ranged from near zero to 7 ppbv. March and December 2013 showed elevated concentrations of both species at THRO-N, with median values for both months falling above the 75th percentile, and December 2013 having the highest median concentrations for each of these species during the two intensive study periods. Using the Wilcoxon Rank Sum test, we determined that March and December 2013 were the only 2 months during the studies in which NO2 and SO2 were both significantly greater than the historical data (p < 0.05). To better understand the cause for the elevated concentrations, hourly ensemble back trajectories with a maximum length of 5 days were generated using version 4.9 of the Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model (Draxler and Hess, 1998), as shown in Fig. 3 for NO2 data collected during the two study periods. Gridded meteorological data from the 12 km North American Mesoscale Model (NAM12, http://www.emc.ncep.noaa.gov/NAM/.php) (Janjić, 2003) were used as input. During the study period, back trajectory analysis showed that the periods with highest concentrations (top 5 %) for SO2 and NO2 corresponded to trajectories that were shorter (slower speeds) and were more likely to be impacted by closer sources. In contrast, the lower concentration days had higher wind speeds and winds were preferentially from the west.

Although NO2 and SO2 were significantly higher at THRO-N in March and December 2013, EC concentrations from the IMPROVE network (http://views.cira.colostate.edu/fed/; Malm et al., 1994) at THRO-S were not elevated relative to historical data (Fig. S2). EC typically peaks in summer, when wildfires influence much of the west. Further, THRO-S is at the southern end of the oil and gas fields and winds at THRO-S are primarily out of the northwest and south, so that THRO-S may be less influenced by oil and gas emissions. Comparing EC concentrations across the region during the study period, we observed an increase in EC in going from THRO-S northward to THRO-N and LOST (discussed further below). Thus aerosol concentrations at THRO-S may be driven more by regional reductions in particulate matter (Fig. S1), while sites farther north appear to be impacted by local sources. Considering both THRO-N and THRO-S measurements, we find no instances from the two study periods when the median values for all three species fell outside of the interquartile range, indicating that regionally the study periods were not anomalous relative to past years.

Across the United States, emissions from power plants have decreased dramatically in recent decades as the result of legislatively mandated controls, leading to broad improvements in air quality (Hand et al., 2014; Rieder et al., 2013; Sickles II and Shadwick, 2015). In the region surrounding the Bakken, annual power plant SO2 emissions have decreased four-fold over the past 20 years and NOx emissions have been cut in half (http://ampd.epa.gov/ampd/). At the same time, the number of producing oil and gas wells in the ND Bakken region increased by nearly a factor of 50 (https://www.dmr.nd.gov/oilgas/) from January 2005 to January 2015. To better understand the impact of these changing emissions, we again use long-term monitoring data.

Figure 4 shows annually averaged SO2 concentrations collected from monitoring sites at six locations in western ND (EPA AirData; Fig. 1). THRO-N, THRO-S, and LOST all fall within the area with oil and gas activities, although, as noted above, THRO-S may be influenced more by regional trends than local sources. Dunn falls on the outskirts of the Bakken region. Beulah and Hannover both lie to the east of most of the activity, near KNRI and several coal-fired power plants (see Fig. 1), which represent major sources of SO2 in the region. As shown in Fig. 4, SO2 concentrations are declining throughout the region, particularly at sites closer to the power plants, consistent with observations in the eastern United States (Sickles II and Shadwick, 2015; Hand et al., 2012b) and decreasing SO2 emissions from power plants across the United States. These reductions were determined to be significant at all of the sites in Fig. 4 except LOST and THRO-N using the Theil-Sen method (Sen, 1968; Theil, 1950) for trend analysis (p < 0.001; monthly averaged values). Trend analysis throughout the paper was conducted using the Open Air package in R (Carslaw and Ropkins, 2012; Carslaw, 2014).

Despite these reductions, power plants still represent a large source of SO2 in the region, exceeding that from oil and gas development (http://ampd.epa.gov/ampd/; Grant et al., 2014). The influence of SO2 emissions from regional power plants was observed during BAQS on multiple occasions. In Fig. 5, SO2 concentrations from the URG samplers at four sites are shown for the first study period. Data are presented based on the time resolution at which they were collected (24 or 48 h samples). Data collected using the real time SO2 instrument at THRO-N were compared to the URG data, and showed reasonable agreement averaging over the same time periods (not shown; R2 = 0.87; real time instrument produced higher values, slope = 1.19). Apparent in the figure are the higher concentrations observed at KNRI, which is located east of the Bakken and very near several power plants (Fig. 1). We focus on two high SO2 events at KNRI during the first measurement campaign: 20–22 February 2013 (8 a.m. to 8 a.m.) and 27–29 March 2013 (8 a.m. to 8 a.m.). On 20–22 February, 48 h average concentrations at KNRI were ∼ 8 ppbv, the highest concentrations observed during the study. During this event, none of the other sites had elevated SO2. Comparing back-trajectories for each of the sites, we see that the air masses which impacted KNRI during this 2-day period passed directly over several coal-fired power plants (Fig. 6a), while the air masses reaching the other sites had very little influence from these same plants. Further, the air masses which reached THRO-N during this time period spent minimal time in the Bakken region, and NOx and BC concentrations at THRO-N were relatively low during this event (Fig. 7).

During the episode on 27–29 March slower moving air masses with changing wind directions impacted THRO-N, as well as FOUS and MELA, and the airmasses spent more time over the Bakken region (Fig. 6b), yielding considerably higher concentrations of NOx and BC at the core site (Fig. 7). However, the air masses which reached THRO-N had only a minor influence from emissions sources east of the Bakken, and SO2 concentrations were again low. While THRO-N, MELA and FOUS were minimally impacted by power plants, KNRI was impacted by several plants on these dates, and KNRI had elevated SO2 concentrations at this time (Fig. 5). These observations are consistent with regional power plants largely influencing SO2 concentrations, and emissions sources from within the Bakken, likely tied to the many sources associated with oil and gas activities, leading to the observed increases in NOx and BC.

Long-term monitoring data for NO2 (EPA AirData) are consistent with these observations and provide an interesting contrast to SO2. Like SO2, NO2 concentrations have decreased in Hannover and Beulah (Fig. 8), east of the Bakken region (p < 0.001; Theil-Sen method, monthly averaged values), likely driven by decreasing NOx emissions from power plants. In contrast, THRO-N and Dunn, within and at the outskirts of the oil and gas production region, show no significant trends when considering the entire time period shown. However, when limiting the data to the past 10 years (2005–2014), when oil and gas activities intensified, significant increasing trends in NO2 are observed at both THRO-N and Dunn (p < 0.001; Theil-Sen method, monthly averaged values). Finally, LOST shows a significant (p < 0.001; Theil-Sen method, monthly averaged values) trend of increasing NO2 throughout the time period shown. These changes are consistent with increasing NOx emissions from oil and gas activities, which more than doubled in the Williston Basin from 2009 to 2011 and are expected to continue to increase (Grant et al., 2014). Unlike SO2, NOx emissions from oil and gas are similar in magnitude to those from regional power plants (Grant et al., 2014; http://ampd.epa.gov/ampd/). Trends of increasing NO2 have also been observed in the Marcellus Shale region (Carlton et al., 2014) and the Canadian Oil Sands (McLinden et al., 2012), with increases corresponding to increased activities related to oil and gas extraction.

To further explore the influence of oil and gas emissions, we consider monitoring data at LOST. Focusing on NO2 and segregating measurements by local wind direction, we find significant (p < 0.01; Theil-Sen method, monthly averaged values) trends of increasing NO2 only when winds are out of the W, SW and S (Fig. 9), areas with major oil and gas development. EC concentrations from IMPROVE, which have many of the same sources as NO2, showed an identical pattern. In contrast, the only significant trend for SO2 at LOST is for winds out of the south, where concentrations have decreased, likely from decreasing power plant emissions.

To better establish a connection between measured pollutants at the study sites and regional oil and gas activities, we consider data from the second study period, 23 November 2013–28 March 2014. These measurements included the use of canisters to collect air samples for analysis of key tracer species. Of particular interest for the Bakken are the light alkanes, which serve as markers for oil and gas activity (e.g. Gilman et al., 2013; Petron et al., 2012; Swarthout et al., 2013, 2015). Figure 10 summarizes measurements of ethane, propane, n-butane, and n-pentane throughout the campaign at all three sites. The mean ethane and propane mixing ratios from all three sites were 16 and 15 ppbv, respectively, with maximum values approaching 100 ppbv for ethane and 150 ppbv for propane. The i-butane levels ranged from 0.1–22 ppbv, and n-butane peaked over 60 ppbv. The i-pentane and n-pentane had comparable mixing ratios with mean concentrations of ∼ 1.2 ppbv (range < 0.1–17 ppbv). These concentrations are significantly higher than typically observed in remote regions (Russo et al., 2010b) and are comparable to levels observed in urban areas known to be influenced by petrochemical industry emissions, as has been observed in other oil and gas basins (Swarthout et al., 2013, 2015). Despite variability in absolute concentrations throughout the study, these data provide evidence that emissions related to oil and gas activities were observed at THRO-S, FOUS, and MELA during the second study period.

To better characterize the extent of this impact, we focus on pentane measurements from all of the sites. Recent studies have used the ratio of pentane isomers to identify air masses that are influenced by oil and gas emissions. Although this ratio varies by basin, a ratio of i-pentane to n-pentane which falls at or below one is generally indicative of oil and gas emissions (Swarthout et al., 2013, 2015; Gilman et al., 2013), whereas higher ratios correspond to background conditions, largely resulting from automobile emissions and fuel evaporation (e.g. Russo et al., 2010b). The i-pentane to n-pentane ratios for all sites for the entire sampling period are shown in Fig. 11; the slope is 0.77. Although there is scatter in the data, particularly at the lowest concentrations; only two out of 287 samples at THRO-N, FOUS and MELA had i-pentane to n-pentane ratios that were consistent with background air; all other samples indicated oil and gas influence. These data not only confirm that oil and gas emissions are impacting the region, but also that this influence was present at nearly all times during the second study period.

Mobile measurements collected throughout the Bakken region support these data. Background concentrations for CH4 observed in the Bakken region for the sampling period of 10–16 December 2013 were 2.2 ± 0.4 ppmv, above expected background levels of < 2 ppmv for a remote location (Farrell et al., 2013; Wofsy et al., 2011), with peak measured concentrations reaching 16.1 ppmv (1 min average). BC concentrations also were elevated for a remote region, with average concentrations of 900 ± 100 ng m-3. To better demonstrate the direct impact of oil and gas activities on these species in the region, two mobile sampling periods from the Bakken region are shown in Fig. 12. One set of measurements was located on the Indian Hill oil field, where there was an active flare at the time of the measurement. The other set of measurements was located on the Painted Woods oil field, with no active flare. Figure 12a shows the data collected near the Indian Hill location (active flare) where an increase in CH4 concentrations corresponded to high concentrations of BC. During the flaring, maximum BC concentrations near the site were approximately 4 times higher than the regional BC concentrations. The data collected from the Painted Woods oil field, with no active flare, are shown in Fig. 12b. Without flaring, these measurements show elevated CH4 concentrations (∼ 7 times above the regional background average), with no corresponding increase in the BC. These areas thus provide sources of VOCs, and, when flaring is present, BC.

Using the light alkanes as markers for local oil and gas activities, we compared alkane concentrations to measurements of NOx, SO2 and BC throughout the second study period. Timelines of all of these species are shown in Fig. 13 for THRO-N. In the figure, we use ethane as a marker for oil and gas emissions, but all of the light alkanes showed similar results. NOx, SO2 and BC concentrations are daily average values; in contrast, ethane data are calculated as the average of two grab samples per day: one collected in the morning (typically 8 a.m.), and one collected in the afternoon (typically 4 p.m.). Concentrations of NOx and BC were correlated with ethane (correlation coefficients, r = 0.75 for NOx and r = 0.70 for BC) throughout the study period. Although these measurements do not identify which emissions source drives the elevated concentrations for NOx and BC, the data suggest that VOCs, NOx and BC likely have collocated sources. SO2 was not as strongly correlated with ethane (r = 0.42). The lower correlation is presumably because SO2 comes largely from different sources, as discussed above.

Measurements collected at THRO-N as part of the first intensive study (February–April 2013) also showed that NOx and EC were correlated (Fig. 7; r = 0.81 for 1 h data) with elevated concentrations observed throughout the campaign. Although we did not make routine measurements of VOCs during the first study period, we expect that these measurements were likely impacted by similar sources as observed in the second study period; i.e. oil and gas related emissions. Hourly NOx and EC concentrations reached 10 ppbv and 1.3 µg m-3, respectively, during the first study period; the maximum observed hourly SO2 concentration was just over 9 ppbv. As discussed above, higher concentrations for all species typically were observed during periods of low wind speeds and changing wind directions. Such conditions allow pollutants to accumulate in the region, particularly those with local sources.

The data shown thus far suggest that emissions from oil and gas activities are impacting air quality in the region, raising ambient concentrations of VOCs, NOx, and EC. Next, we use VOC measurements to estimate the amount of photochemical processing within the air masses that reached THRO-N during the study. To this end, we use the alkyl nitrate to parent hydrocarbon ratios (R-ONO2 / R-H) to estimate air mass age for measurements at THRO-N (e.g. Bertman et al., 1995; Russo et al., 2010a; Simpson et al., 2003; Swarthout et al., 2013). The modeled ratios of 2-pentyl nitrate to n-pentane versus 2-butyl nitrate to n-butane are presented in Fig. 14 (solid line) along with the measurements of these species. For the model, a diurnally averaged OH concentration of 5 × 105 molec cm-3 was assumed. The measured ratios fall on the modeled line, suggesting that the photochemical sources of the alkyl nitrates are reasonably well represented in the model. Results indicate a photochemical processing time of < 2 days throughout the majority of the campaign. These results are similar to wintertime VOC measurements in the Denver-Julesburg basin in NE Colorado (Swarthout et al., 2013). For a windspeed of 1.44 m s-1, the median windspeed observed at THRO-N during the second study, a processing time of 2 days corresponds to a transport distance of ∼ 250 km. Because the absolute air mass age determined from the estimates are OH radical concentration dependent, these estimates are subject to uncertainty. However, the results show that processes are occurring on relatively short timescales and are associated with fresh emissions, rather than aged air masses, and so point to emissions within the Bakken, rather than long-range transport from other oil and gas basins. NOx and BC concentrations also are shown on the alkyl nitrate evolution plots in Fig. 14a and b, respectively. These data show that the highest levels of NOx and BC occur in air masses with short processing times (< 12 h), consistent with the data presented thus far, and further implicating local sources for NOx and BC; this is particularly relevant for BC, which has a longer atmospheric lifetime. A similar plot is shown for SO2 in Fig. S3.