![ORCID Icon]({"type":"image","src":"/templates/jsp/_style2/_tandf/images/orcid.svg"} "ORCID Icon"){kind=small}
http://orcid.org/0000-0001-8493-9522
ABSTRACT
ABSTRACT
In this study, the authors apply two different dispersion models to evaluate flux chamber measurements of emissions of 58 organic compounds, including C2–C11 hydrocarbons and methanol, ethanol, and isopropanol from oil- and gas-produced water ponds in the Uintah Basin. Field measurement campaigns using the flux chamber technique were performed at a limited number of produced water ponds in the basin throughout 2013–2016. Inverse-modeling results showed significantly higher emissions than were measured by the flux chamber. Discrepancies between the two methods vary across hydrocarbon compounds and are largest in alcohols due to their physical chemistries. This finding, in combination with findings in a related study using the WATER9 wastewater emission model, suggests that the flux chamber technique may underestimate organic compound emissions, especially alcohols, due to its limited coverage of the pond area and alteration of environmental conditions, especially wind speed. Comparisons of inverse-model estimations with flux chamber measurements varied significantly with the complexity of pond facilities and geometries. Both model results and flux chamber measurements suggest significant contributions from produced water ponds to total organic compound emission from oil and gas productions in the basin.
Implications: This research is a component of an extensive study that showed significant amount of hydrocarbon emissions from produced water ponds in the Uintah Basin, Utah. Such findings have important meanings to air quality management agencies in developing control strategies for air pollution in oil and gas fields, especially for the Uintah Basin in which ozone pollutions frequently occurred in winter seasons.
Introduction
Organic compound emissions from produced water ponds are poorly characterized, and they are not included or fully accounted for in oil and gas emission inventories (AECOM Environment and Sonoma Technology, Inc. 2014; U.S. Environmental Protection Agency [EPA] 2017) Information gaps include unquantified amounts and composition of gases emitted from these facilities. Produced water ponds are a source of organic compounds that, along with nitrogen oxides (NOx), are major precursors to ozone formation in the near-surface layer of the atmosphere. Field measurement campaigns using a flux chamber technique have been performed to measure organic compound emissions from produced water ponds in the Uintah Basin of eastern Utah. Lyman et al. (2018) discuss these measurements in detail, and Mansfield et al. (2018) develop mass-transfer coefficients and perform other computational analyses of this data set. Steady-state flux chamber systems (also known as open-chamber or active-chamber systems as opposed to non-steady-state or passive-chamber systems) provide accurate measurements at the point of sampling, but they cover a limited area and are prone to altering environmental conditions (e.g., temperature, pressure, wind). Flux chamber systems are commonly known for masking the effect of external wind, which plays a critical role in determining hydrocarbon emission rates (Lee et al. 2004; Schwarzenbach, Gschwend, and Imboden 2005). For example, early studies (Liss 1973, 1975) discussed the effect the flux chamber on altering the atmospheric and water turbulence that drive the transfer coefficients of gases between air-water interfaces. This could be explained with the popular two-film model by which chemical compounds transfer between liquid and atmosphere through thin liquid-and-gas films at the liquid-gas interface (Bianchi and Varney 1997; Sadek et al. 1996). Wind tunnel experiments demonstrated that turbulence increases with wind speed and enhances the overall transfer coefficient from liquid phase to gas phase (Lee et al. 2004; Parker et al. 2010). Cole et al. (2007) reported that a flux chamber at 0.5 turnovers/min air exchange rate yielded 25% lower emission rates than the open air surface. Effects of wind speed on increasing emission rates is more pronounced for highly soluble compounds (Jiang and Kaye 1996). Values of the Henry constant of several representative hydrocarbon compounds classified into groups that are sensitive to air-phase turbulence, liquid-phase turbulence, or both air- and liquid-phase turbulences are given in Hudson and Ayoko (2008). The size of the flux chamber could also affect emission rates, as a smaller chamber height increases the wind speed gradient near the measuring surface, thus enhancing the emission rate (Frechen et al. 2004; Hudson and Ayoko 2009). Several other artifacts of the flux chambers reported in various flux chamber studies are discussed by Davidson et al. (2002): Firstly, both non-steady-state and steady-state chambers decrease diffusive concentration gradients at the air-soil interface, leading to underestimation of fluxes. Secondly, pressure differentials between the inside and outside of chambers could cause over- or underestimation of fluxes if venting is not properly designed, although the non-steady-state chamber is more susceptible to this issue than the steady-state one. The duration of each sampling event also must be carefully chosen, as too short of a duration yields inaccurate measurement, whereas an extended duration leads to saturation inside the chamber, introducing biases to the measurements by static or closed chambers (Khalil, Rasmussen, and Shearer 1998). Such potential biases in the flux measurement methods present the necessity for evaluating flux chamber measurements with alternative flux estimation methods.
To evaluate our flux chamber measurements, we applied an inverse dispersion modeling technique with evacuated canister sampling. Inverse dispersion modeling techniques have been widely used in various studies covering different spatial and temporal scales, source types, and tracers and have employed traditional dispersion models (Hensen et al. 2009; O’Shaughnessy and Altmaier 2011) or backward Lagrangian stochastic models (Aylor and Flesch 2001; Flesch et al. 2009, 2007). Dispersion modeling methods work well with constant wind conditions and simple terrain. Also, receptor sites must be carefully chosen so that the impact of the source of interest on the receptor can be isolated from impacts of surrounding sources. Because of its limitations, this method is not well suited for a comprehensive characterization of many different pond sites in many different conditions, but it is well suited to provide a check on chamber measurements.
Methodology
Field measurement design
Measurement methods for meteorological parameters and evacuated canister measurements of organic compounds are provided by Lyman et al. (2018). In summary, 163 flux chamber measurements were conducted at eight facilities in fall, winter, spring, and summer during 2015–2016 in the Uintah Basin, among which concurrent canister measurements for inverse modeling were performed for one landfarm and three produced water pond facilities. We placed a set of evacuated canisters along the borders of the produced water ponds and landfarm. We either mounted the canisters on tripods or connected them with 0.6-cm-diameter stainless tubes so that air intakes were at 1 m above the ground. Air intakes consisted of a stainless steel frit followed by a stainless critical orifice that restricted flow for a 1-hr fill time. We analyzed the canisters for 58 organic compounds, including C2–C11 hydrocarbons and methanol, ethanol, and isopropanol. Analytical methods are described by Lyman et al. (2018). For inverse modeling with AERMOD (American Meteorological Society/U.S. Environmental Protection Agency Regulatory Model), the concentrations were converted from volume mixing ratio (ppb) to mass concentration (μg m−3) based on the measured temperature and pressure. The number of canisters for each deployment varied depending on the ponds’ area and geometry and number of collocated ponds. For each canister sampling, we placed a canister upwind of the ponds to characterize the background organic compound concentrations. However, in many cases, we found that the background canister may have been contaminated with the pond’s plume as wind direction changed unexpectedly. We examined such circumstances carefully when performing inverse-modeling analyses.
We measured meteorological parameters, including temperature, relative humidity, wind speed and direction, barometric pressure, and total incoming solar radiation, at 6 m above ground level (AGL). Additionally, we measured the temperature gradients between 1 and 6 m above ground to characterize atmospheric stability. All meteorological parameters were measured within a few meters of the ponds and for at least 24 hr before the canister sampling time. For some measurement campaigns, we measured wind speed on top of the chamber, about 30 cm above the water level, for model sensitivity analyses.
Inverse dispersion modeling
Inverse dispersion modeling applies an initial and arbitrary emission rate to estimate pollutant concentrations at predefined receptors (evacuated canisters, in our case) and adjusts the emission rate until the estimated pollutant concentrations approximate the measured concentrations at the receptors. The derived emission rates are then compared with flux chamber measurements, and differences are analyzed. In this study, we used the EPA regulatory AERMOD model (Cimorelli et al., 2004) and the steady-state heavy gas dispersion model, steady state heavy gas dispersion model (HEGADAS-S) (Witlox 1994), a component of the atmospheric dispersion models developed by Shell Research, Ltd., for inverse dispersion modeling.
We processed geophysical parameters (surface roughness, albedo, Bowen ratio, elevations) for AERMOD’s meteorological module (AERMET) by AERMET’s surface characteristic processor (AERSURFACE) version 13016. We processed on-site meteorological measurements with AERMET version 15181. Although the pollutant ID can be specified in AERMOD, this has no effect on calculations made in the model. As such, we simulated each of the 58 measured compounds in the same manner, except for simulations of dry and wet deposition. However, we found little to no difference due to dry and wet deposition among AERMOD-simulated concentrations of different compounds. We attribute this to the fact that the canisters were very close to the ponds, whereas differences in concentrations due to deposition would only be observable further downwind.
AERMOD’s standard approach allows options to calculate pollutant concentrations for various averaging times (i.e., annually, monthly, daily, hourly), and 1-hr averaged concentration is its standard calculation. Accordingly, we processed measured meteorological parameters in AERMET to produce 1-hr inputs for AERMOD, and we collected evacuated canister samples over intervals of 1 hr. However, we often found that wind direction changed significantly during the sampling period, especially during wintertime and under weak wind conditions. Under such conditions, the single-vector 1-hr wind field poorly represented the actual wind field during the sampling period. To better represent pollutant dispersion, we divided the 1-hr period into four 15-min subperiods, processed the meteorological inputs, and performed “1-hr” AERMOD simulations for each 15-min subperiod. We then averaged the four 15-min AERMOD outputs to obtain the 1-hr modeled concentrations for evaluation with canister samples.
AERMOD allowed pond-like sources to be simulated as area, volume, or open-pit sources. We examined scenarios in which the produced water ponds were modeled as different source types. We found no single source type that worked best to characterize all produced water ponds. We used the area or open-pit source type, depending on the characteristics of the pond.
The hazard assessment software package for modelling the release dispersion of hydrogen fluoride and ideal gases (HGSYSTEM) is developed at Shell. HEGADAS is the heavy gas dispersion from area sources module in HGSYSTEM. It includes steady-state, HEGADAS-S, and time-dependent, HEGADAS-T (Witlox 1994), versions. We used HEGADAS-S for this study. We derived meteorological parameters required for HEGADAS-S from AERMET and on-site measurements.
Organic compounds can be simulated by HEGADAS-S as individual compounds or as mixtures. The default database in HEGADAS-S only covers 15 out of 58 compounds measured in this study. We extended the database with parameters for the remaining compounds (e.g., critical temperature, critical pressure, specific heat of vapor, etc.) based on literature reviews. As such, each compound was simulated separately in HEGADAS-S. Note that no chemical reaction is considered in HEGADAS-S.
Although AERMOD allows flexibility in defining the pond’s geometry, HEGADAS-S can only characterize a pond as a simple two-dimensional rectangular area source. Wind direction is not an input in HEGADAS-S, and the model assumes that wind direction is normal to the source width. This led to difficulties in characterizing produced water ponds in this study, as many ponds are in irregular shapes and wind came from various angles relative to the ponds. To overcome this issue, we discretized the pond into a series of rectangular ponds that are normal to the wind direction and that conserve the pond geometry and total area. Additionally, berms enclosing ponds are not considered in HEGADAS-S. Berms cause turbulence inside and over the ponds, which does not approximate the assumption of steady and streamlined flow in HEGADAS-S. Further details and special treatments on these issues are discussed in Case Study.
The most significant differences between AERMOD and HEGADAS-S are in the dispersion algorithms. AERMOD assumes a Gaussian distribution of the plume concentrations in both horizontal and vertical directions under stable conditions, and a bi-Gaussian vertical distribution in the convective boundary layer (Cimorelli et al. 2004). Vertical structure of the planetary boundary layer is examined extensively in AERMOD based on measured meteorological parameters, and thermal- and mechanical-driven turbulences are the driving factors of plume transport and dispersion. Meanwhile, as a model developed for dispersion of heavy gases, HEGADAS-S assumes the formation of a dense plume after the moment the gas is released to the atmosphere. An effective half-width plume is determined, along which concentrations are uniform to the concentration at the plume’s center line. Vertical and cross-wind dispersion coefficients are determined as Gaussian plumes at the flanks of the dense plume. At the immediate downwind distance, the half-width plume conforms to the pond’s physical half-width and the Gaussian term is zero. Furthermore, heat transfer from water vapor and the surface, as well as thermodynamic characteristics of the gas/mixture, is considered in determining the volume, density, and temperature of the plume. HEGADAS-S relies on a user-specified Pasquill-Gifford stability scale, friction velocity, and Monin-Obukhov length to determine the vertical wind profile from which plume entrancement and cross-wind dispersion coefficients are determined (Witlox 1994).
Results and discussion
Case study
We chose one measurement campaign among all the campaigns conducted in the Uintah Basin during 2015–2016 to discuss in detail here. We chose this campaign because the wind direction was consistent throughout the measurement period and because the layout of the facility we sampled was relatively simple, with one active pond and two skim ponds. At other facilities, multiple active ponds interfered with canister measurements, making the results of inverse modeling more difficult to interpret and resulting in high uncertainty. We discuss the results of studies at other facilities and times in a subsequent section.
Pond description
The produced water pond (Figure 1) in this measurement campaign was roughly 0.3 ha at the water surface, which was 2 m below the top of surrounding berms (surface level). At the facility, produced water is received from trucks, which discharge it into several oil separation tanks. After the tanks, the water is deposited into two skim ponds with surface area of less than 30 m2. From the skim ponds, water is discharged at about 30 cm above the active pond’s berm before running off along the berm’s slope into the pond.
Published online:
11 May 2018Figure 1. Produced water pond layout. The gray-shaded polygon depicts the water surface of the pond; the dot line encloses the pond’s berms; the oval and rectangle represent oil-skim tanks and ponds, respectively; dash lines indicate the discrete subponds normal to wind direction; S0–S9 indicate locations of the canister samplings; FL01–FL04 are locations of flux chamber measurements; the star indicates the location of meteorological measurements. The wind rose shows wind speed and direction throughout the measurement period. Actual locations of S0, S8, and S9 are further than they are presented in this figure, and we slightly modified the pond facility layout to not reveal its identity.
![]({"type":"image","src":"/na101/home/literatum/publisher/tandf/journals/content/uawm20/2018/uawm20.v068.i07/10962247.2018.1426654/20180620/images/medium/uawm_a_1426654_f0001_oc.jpg"})
Figure 1. Produced water pond layout. The gray-shaded polygon depicts the water surface of the pond; the dot line encloses the pond’s berms; the oval and rectangle represent oil-skim tanks and ponds, respectively; dash lines indicate the discrete subponds normal to wind direction; S0–S9 indicate locations of the canister samplings; FL01–FL04 are locations of flux chamber measurements; the star indicates the location of meteorological measurements. The wind rose shows wind speed and direction throughout the measurement period. Actual locations of S0, S8, and S9 are further than they are presented in this figure, and we slightly modified the pond facility layout to not reveal its identity.
Measurement setup
During the 1-hr measurement period, wind direction was relatively consistent at 33°, with an average wind speed of 2.4 m sec−1 at 6 m height above the surface (Figure 1). Nine canisters were deployed in the configuration shown in Figure 1. Figure 2 shows organic compound concentrations grouped by the number of carbon atoms (expressed here as C2–C10 excluding alcohols) measured at sample locations S0–S9. As expected, organic compound concentrations were lowest at S0, as this canister was designated for characterizing background concentrations. S5 and S6 were immediately upwind of the pond, and their concentrations were similar to S0. Except for C2 and C3 compounds, all organic compound concentrations were highest at S1 (located downwind of the skim pond), followed by S2 and S3. The fact that organic compound concentrations were higher at S2 and S3 than at S4 is an interesting finding, since S4 was downwind of the entire pond and could be expected to have been exposed to emissions over a greater area of the pond, leading to higher concentrations (discussed further below). S9 had higher organic compound concentrations than many of the other sampling locations even though it was distant from and not directly downwind of the produced water pond but was instead more directly downwind of liquid storage tanks at the facility. C2 and C3 compounds made up a higher percentage of total organics measured at S9, whereas emissions from produced water tend to be dominated by C6 and heavier compounds (Lyman et al. 2018), providing more evidence that these locations were influenced by storage tanks, not produced water emissions. We excluded S9 from inverse modeling.
Published online:
11 May 2018Figure 2. Concentration distributions of organic compounds, as classified by (left) number of carbon atoms, excluding alcohols, and (right) by carbon-bond type, measured by canisters S0–S9.
![]({"type":"image","src":"/na101/home/literatum/publisher/tandf/journals/content/uawm20/2018/uawm20.v068.i07/10962247.2018.1426654/20180620/images/medium/uawm_a_1426654_f0002_b.gif"})
Figure 2. Concentration distributions of organic compounds, as classified by (left) number of carbon atoms, excluding alcohols, and (right) by carbon-bond type, measured by canisters S0–S9.
We collected five flux chamber measurements on the active pond on the same day that we collected evacuated canister samples, but one measurement had technical issue and was excluded. Figure 3 presents emission rates of C2–C10 compounds measured by the flux chamber at those locations. Flux chamber measurements were fairly consistent across locations over the pond and were dominated by emissions of C6–C9 compounds. The speciation of emissions from the flux chamber measurements corresponded better with speciation in the S1–S3 canisters than speciation in other canisters, perhaps indicating a stronger influence from produced water at those sampling locations.
Published online:
11 May 2018Figure 3. Emission rates measured by the flux chamber. Emission rates are presented in mmol m−2 hr−1 for relative comparison with the concentrations presented in Figure 2.
![]({"type":"image","src":"/na101/home/literatum/publisher/tandf/journals/content/uawm20/2018/uawm20.v068.i07/10962247.2018.1426654/20180620/images/medium/uawm_a_1426654_f0003_b.gif"})
Figure 3. Emission rates measured by the flux chamber. Emission rates are presented in mmol m−2 hr−1 for relative comparison with the concentrations presented in Figure 2.
Inverse-modeling results with HEGADAS-S
To best characterize the pond’s geometry and wind direction, we represented this pond in HEGADAS-S as 15 subponds, as shown in Figure 1. We took the total estimated concentration imposed by all the subponds at each receptor location and compared that with the corresponding canister measurement to deduce the pond’s emission rate. For each of the 58 compounds, we started the HEGADAS-S model with an arbitrary emission rate of 10−5 kg sec−1 m−2 for each of the 15 subponds and then scaled the emission rate based on the ratio of estimated/observed concentrations across applicable canisters. For this case study, S2, S3, and S4 were downwind of the pond and could be used to determine the main pond’s emissions. Canisters S5, S6, and S7 had organic compound concentrations within the range of the background canister (S0) and were treated as such. S1 and S8 were not only influenced by emissions from the active pond but also from the skim ponds. As discussed above, S9 was excluded for this case study.
Estimated emission rates for the active pond are presented in Table 1. Table 1 also shows flux chamber measurements before and after correction with wind speed (more discussion is below). Figure 4 shows the reconstructed hydrocarbon concentrations estimated by HEGADAS-S using estimated emission rates. Distributions of organic compound concentrations noticeably differed from canister measurements both in speciation and magnitude. Specifically, estimated organic compound concentrations were highest at S4, followed by S3 and S2, whereas canister measurements show highest organic compound concentrations at S3, followed by S2 and S4. HEGADAS-S underestimated total organic compound concentrations at all sampling locations. At S3, for example, estimated organic compound concentrations were about 50 ppb lower than measured values, after taking into account background concentrations (calculated as the average of S0, S5, and S6). Note that estimated concentrations at S1 are much lower than measured concentrations, since influences from the skim pond were not accounted for.
Table 1. Organic compound emission rates (mg m−2 hr−1) estimated by HEGADAS-S (HEGA) and AERMOD (AERM), and flux chamber (FLUX) grouped by carbon-bond types under various scenarios.
Published online:
11 May 2018Figure 4. Reconstructed organic compound concentrations at canister locations as estimated by HEGADAS-S with adjusted emission rates that were determined using all subponds (left column) and using only upwind partial ponds, as determined by fluid dynamics modeling (right column).
![]({"type":"image","src":"/na101/home/literatum/publisher/tandf/journals/content/uawm20/2018/uawm20.v068.i07/10962247.2018.1426654/20180620/images/medium/uawm_a_1426654_f0004_b.gif"})
Figure 4. Reconstructed organic compound concentrations at canister locations as estimated by HEGADAS-S with adjusted emission rates that were determined using all subponds (left column) and using only upwind partial ponds, as determined by fluid dynamics modeling (right column).
We hypothesize that this discrepancy was mainly due to the wake effect of the berms surrounding the pond. Since the water level was 2 m below the top of the berm (and the ground level), wind measurements at 6 m AGL were not able to represent wind fields at near-water level (in both magnitude and direction). Thus, using 6 m AGL wind measurements for the HEGADAS-S simulation can be expected to result in discrepancies between estimated and observed values. To test this hypothesis, we utilized a computational fluid dynamics model, the Open source Field Operation And Manipulation (OpenFOAM) model (Weller et al. 1998), to simulate wind flow at the near-water level (see Supplemental Information for more details). One of the findings is that emissions from the lower end of the pond do not directly reach receptors in that area but instead exit through the upper end of the pond. We performed another HEGADAS simulation with the assumption that all of the pond’s emissions exit through the upwind subponds, and we scaled the results by the ratio of total pond area to the upwind area. As shown in Figure 4, the reconstructed organic compound concentrations approximate the canister-measured values better under this scenario. Interestingly, HEGADAS estimated all organic compound concentrations lower at S4 than at S2 and S3, except for alcohols. We attribute this behavior to the light molecular weight and high emission rate of alcohols, especially methanol, both of which resulted in wider spreading of the dense center plume, as estimated by HEGADAS-S in comparison with heavier compounds with lower emission rates.
The flux chamber–measured organic compound emission rates were lower than those estimated by HEGADAS-S across all compound groups in both the full- and partial-pond scenarios (Figure 5). HEGADAS-S estimated higher emissions in the full-pond scenario than in the partial-pond scenario. In the partial-pond scenario, HEGADAS-S emissions of C6–C8 compounds were within a factor of 3 compared with flux chamber measurements. The largest discrepancies were observed for alcohols, especially methanol, for which HEGADAS-S estimated as much as 6 times higher emissions than flux chamber measurements. Nevertheless, correlations between estimated and measured emission rates were high (>0.96) in all scenarios.
Published online:
11 May 2018Figure 5. Comparisons of organic compound emissions as estimated by HEGADAS-S in full-pond (black) and partial-pond (gray) scenarios with flux chamber measurements (wind-corrected).
![]({"type":"image","src":"/na101/home/literatum/publisher/tandf/journals/content/uawm20/2018/uawm20.v068.i07/10962247.2018.1426654/20180620/images/medium/uawm_a_1426654_f0005_b.gif"})
Figure 5. Comparisons of organic compound emissions as estimated by HEGADAS-S in full-pond (black) and partial-pond (gray) scenarios with flux chamber measurements (wind-corrected).
Differences in emission estimation in full-pond and partial-pond scenarios illustrate the critical role of wind fields in inverse-model studies. Another finding from the OpenFOAM model analyses is that wind speed at near-water level is much lower than at 6 m AGL. Obviously, wind speed measurements at 6 m AGL are not representative of wind speed at the near-water level, which determines the rate of volatilization of organics from the water. In fact, whereas wind speed at 6 m AGL during this case study averaged 2.4 m sec−1, wind speed measured at about 0.5 m above the water level at the flux chamber averaged only 1.5 m sec−1. To test the influence of near-water-level versus 6 m AGL wind speed, we performed a HEGADAS-S simulation with the near-water-level wind speed, rather than the 6 m AGL wind speed. In comparison with 6 m AGL wind speed simulations, the lower wind speed near water level led to a denser plume at immediate downwind locations (i.e., higher concentrations at canister locations), which led to a lower estimated emission strength (Table 1). The estimated emission strength under this scenario, however, was still higher than measured values.
Inverse-modeling results with AERMOD
We modeled emissions from the produced water pond in AERMOD as an open pit and as an area source. When modeling as an open pit, AERMOD applies an empirical formula to calculate an effective area source as a fractional size relative to the pit opening based on the pit’s dimensions, its depth, and wind direction (EPA 1995). This treatment comes from findings in wind tunnel studies (Perry, Thompson, and Petersen 1994) that emissions within the pit are emitted primarily from the upwind area of the pit instead of universally over the entire pit. Once the effective area is determined, AERMOD simulates the pit as a typical area source with emission rates scaled by the ratio of the actual area over the effective area. This treatment is similar to our treatment to HEGADAS-S model based on findings from the OpenFOAM model.
Starting with the same initial emission rates as HEGADAS-S (10−5 kg sec−1 m−2), AERMOD estimated lower concentrations at the same canister locations. As shown in Table 1, emission rates estimated by AERMOD were much higher than HEGADAS-S estimations and flux chamber measurements in both open-pit and area source simulations. The largest discrepancies were for alcohols, where AERMOD estimates were as much as 20 times more than flux chamber values. In comparison with HEGADAS-S, the reconstructed organic compound concentrations obtained from AERMOD estimated emission rates were not as good at replicating canister measurements (Figure 6). Correlations (R2) between AERMOD’s estimated and measured concentrations were similar for all individual organic compounds (0.4–0.65 for most compounds). For HEGADAS-S, the corresponding R2 were highest for C4–C8 compounds (0.68–0.79) and lowest for C2–C3 compounds (less than 0.1). Note that C4–C8 compounds made up most of the flux chamber–measured total emission rate (excluding alcohols).
Published online:
11 May 2018Figure 6. Reconstructed organic concentrations at defined canisters as simulated by AERMOD with adjusted emission rates and simulating the produced water pond as open pit (left) and area source (right).
![]({"type":"image","src":"/na101/home/literatum/publisher/tandf/journals/content/uawm20/2018/uawm20.v068.i07/10962247.2018.1426654/20180620/images/medium/uawm_a_1426654_f0006_b.gif"})
Figure 6. Reconstructed organic concentrations at defined canisters as simulated by AERMOD with adjusted emission rates and simulating the produced water pond as open pit (left) and area source (right).
The finding that alcohols had the largest discrepancies between flux chamber measurements and inverse models (20 times in AERMOD and 3–4 times in HEGEDAS-S) is due to the fact that their emissions are most affected by the shielding effect of the flux chamber. Alcohols are highly soluble, and their emission rates depend heavily on turbulence and concentration gradient in the gas phase (Hudson and Ayoko 2008; Parker et al. 2010), which are altered by the presence of the flux chamber. For comparison, Henry constants of aromatic compounds depend primarily on turbulence in the water (Hudson and Ayoko 2008); thus, emissions of aromatics are less impacted by the flux chamber. In other words, the flux chamber likely underestimates the emissions of alcohols more than any other of the measured compounds (Mansfield et al. 2018). As illustrated in Table 1, emissions of aromatics had fewer discrepancies between the flux chamber and inverse models. Figures 5 and 7 show that discrepancies between measured and estimated emissions were smallest for less soluble compounds (C6+) and largest for more soluble compounds (C2–C5, and alcohols).
Published online:
11 May 2018Figure 7. Comparisons of organic compound emissions as estimated by AERMOD in the area (black) and open-pit (gray) source approximations with flux chamber measurements (wind-corrected).
![]({"type":"image","src":"/na101/home/literatum/publisher/tandf/journals/content/uawm20/2018/uawm20.v068.i07/10962247.2018.1426654/20180620/images/medium/uawm_a_1426654_f0007_b.gif"})
Figure 7. Comparisons of organic compound emissions as estimated by AERMOD in the area (black) and open-pit (gray) source approximations with flux chamber measurements (wind-corrected).
Differences between HEGADAS-S and AERMOD emission estimations were due to differences in how the two models treat the plume after it is released into the atmosphere, as discussed in Methodology.
The most noticeable difference in the two approaches is in the vertical diffusion term: HEGADAS-S estimates a much denser but shallower plume than AERMOD. For example, at a same canister location, raising the receptor height from 1 to 3 m above ground level resulted in a 70% concentration reduction in HEGADAS-S but only a 20% reduction in AERMOD. The denser plume in HEGADAS-S resulted in lower emission estimations in comparison with AERMOD.
Modeling results from other measurement campaigns
Comparisons of organic compound emissions estimated by HEGADAS-S and AERMOD with flux chamber measurements in other measurement campaigns are presented in Table 2. Values presented in this Table are the lowest estimations obtained in all examined scenarios in AERMOD and HEGADAS-S. We consider the lowest emission estimations, as the inverse models tend to overestimate actual emissions (more discussions below). The April 2015 measurement campaign was conducted at a landfarm facility. Since the landfarm was at ground level, rather than in a recessed pit, we simulated the landfarm in AERMOD as an area source, not an open pit. Interestingly, at this facility, we found that the estimations of AERMOD and HEGADAS-S agreed well, which we attribute to the absence of complex topography. We found many challenges in applying the inverse-modeling technique in the May 2015 campaigns, which were conducted at a produced water pond facility with multiple ponds. The presence of multiple, adjacent ponds, in combination with strong wind variations, made the canister measurements unreliable for characterizing emissions on the ponds where flux chamber measurements were conducted. Comparisons of flux chamber measurements with AERMOD estimations in the May 2015 campaigns were totally different from other measurement campaigns, and the results are not shown in Table 2. Two measurement campaigns were conducted at the same facility in July 2016.
Table 2. Comparison of emission rates as measured by the flux chamber with and without corrected wind (FLUX and FLUX-C, respectively) and as estimated by AERMOD (AERM) and HEGADAS-S (HEGA) inverse modeling (mg m−2 hr−1).
Overall, organic compound emission estimations from AERMOD and HEGADAS-S were consistently higher than flux chamber measurements, including flux chamber measurements with a wind speed correction applied. Higher emission rates as estimated by inverse modeling than those derived from direct measurements are common findings in inverse-modeling studies. Martin, Doshi, and Moore (2006) attributed inadequate simulation of plume buoyancy and underestimation of concentrations at elevated receptors by the Industrial Source Complex, Short Term model (ISCST3; an earlier version of AERMOD) as the reason for anomalously high estimation of ammonia emissions from a swine finishing facility in Iowa. Marchant et al. (2011) showed that AERMOD estimated emissions were as much as 3 times higher than estimations by Light Detection and Ranging (LIDAR) measurements of particulate emissions from a dairy facility. A comparison of estimated particulate matter emissions from fall tillage in California showed 2–4 times higher emissions from AERMOD estimates than from LIDAR measurements (Moore et al. 2013). Most of these studies attributed high biases in model results to discrepancies in the dispersion models. This is likely true, but we expect that our flux chamber measurements were biased low, exacerbating the difference between measured and modeled results. Mansfield et al. (2018) showed that organic compound emissions estimated by the WATER9 wastewater treatment model (EPA 1994, 2001) are about 1 order of magnitude higher than flux chamber measurements. Flux chambers mask the effect of external wind speed (Mansfield et al. 2018; Parker et al. 2013): The chamber used in this work covered an area of about 0.13 m2, and the mixing fan in the chamber created an effective wind speed of less than 0.5 m sec−1 (Lyman et al. 2018), which was lower than the measured wind speed near the water surface. Since air-liquid partitioning of most compounds depends on air turbulence, higher wind speeds lead to higher emissions, and we expect that the flux chamber measurement is lower than the actual emissions. Mansfield et al. (2018) developed wind speed correction factors for the flux chamber measurements based on WATER9’s formulas of mass-transfer coefficients in water and gas phases. More specifically, correction factor CW is the ratio of mass-transfer coefficient in water phase determined with wind speed measured at reference height (e.g., 10 m) to that coefficient determined with effective wind speed of 0.5 m sec−1 inside the flux chamber. Gas-phase correction factor CA is defined analogously. CA is applied to flux chamber measurements of alcohols, whereas CW is applied to measurements of alkenes and aromatics due to their soluble characteristics as discussed above. As illustrated in Table 1 and Table 2, the wind-corrected measurements agree with HEGADAS-S and AERMOD estimations better than the uncorrected ones.
Mansfield et al. (2018) estimates that the total nonmethane organic emission from all produced water pond facilities in the Uintah Basin, derived from flux chamber measurements, is 7494 ton yr−1 (with upper and lower 95% confidence limits of 3466 and 18,552 ton yr−1, respectively). Such amount accounts for 7% of total nonmethane organic compound emission from all oil- and gas-related sources in the basin on average over four emission inventories (Mansfield et al. 2018). Using HEGADAS-S estimations, which are the lowest of the two models and average 2.5 times higher than wind-corrected flux chamber measurements overall, NMHC emissions from produced water ponds as estimated by the inverse model would be approximately 17% of the total emission in the basin. Although HEGADAS-S may overestimate actual emissions, this work shows that produced water ponds are a large contributor of ozone-forming organic compounds to the Uintah Basin atmosphere.
| Compound group | FLUX | FLUX (wind-corrected) | HEGA Full-pond, 6 m AGL wind speed | HEGA Partial-pond, 6 m AGL wind speed | HEGA Partial-pond, near-water-level wind speed | AERM Open pit | AERM Area source |
|---|---|---|---|---|---|---|---|
| Alkanes | 34.0 | 37.9 | 185.5 | 149.6 | 111.8 | 738.0 | 525.1 |
| Alkenes | 0.0 | 0.0 | 7.7 | 6.2 | 4.9 | 21.8 | 16.0 |
| Aromatic | 60.4 | 67.1 | 177.0 | 142.9 | 113.4 | 690.3 | 490.6 |
| Alcohols | 201.5 | 659.6 | 1075.1 | 866.3 | 691.2 | 3691.3 | 2658.2 |
| Measurement campaign | Means of estimation | Alkanesa | Alkenesa | Aromaticsa | Alcoholsa | Total NMHCsa |
|---|---|---|---|---|---|---|
| April 2015 (Landfarm) | FLUX | 287.5 | 4.6 | 83.8 | 2.4 | 378.3 |
| FLUX-C | The developed wind correction factors are not applicable for air-soil exchange interface | |||||
| AERM | 2891.0 | 0.4 | 1068.8 | 29.6 | 3990 | |
| HEGA | 2690.4 | 0.8 | 1027.1 | 11.1 | 3729 | |
| April 2016 (Pond) | FLUX | 34.0 | 0.0 | 60.4 | 201.5 | 296 |
| FLUX-C | 37.9 | 0.0 | 67.1 | 659.6 | 765 | |
| AERM-b | 714.8 | 1.3 | 628.1 | 312.8 | 1657 | |
| HEGA-b | Was not performed due to calm and strongly variate wind | |||||
| AERM-b | 525.1 | 16.0 | 490.6 | 2658.2 | 3690 | |
| HEGA-b | 149.6 | 6.2 | 142.9 | 866.3 | 1165 | |
| AERM-b | 659.7 | 26.8 | 748.6 | 1729.3 | 3164 | |
| HEGA-b | 495.1 | 15.1 | 388.4 | 1148.4 | 2047 | |
| Jul2016 (Pond) | FLUX | 24.4 | 0.0 | 67.6 | 108.8 | 201 |
| FLUX-C | 46.0 | 0.1 | 123.3 | 545.6 | 715 | |
| AERM-b | 261.7 | 4.2 | 265.8 | 1327.0 | 1859 | |
| HEGA-b | 114.9 | 1.5 | 143.3 | 686.3 | 946 | |
| AERM-b | 417.4 | 3.9 | 1063.7 | 1935.8 | 3421 | |
| HEGA-b | 160.8 | 0.8 | 424.0 | 728.8 | 1314 | |
| Geometric means of ratios | HEGA/FLUX-C | 5.3 | 2.7 | 3.6 | 1.8 | 2.5 |
| AERM/FLUX-C | 11.5 | 5.2 | 7.5 | 2.8 | 4.2 | |
| HEGA/AERM | 0.5 | 0.5 | 0.5 | 0.4 | 0.5 | |
Notes. aFor flux chamber, the presented values are averages of all measurements made in same day and at same pond/landfarm of the evacuated canister sampling; for AERM and HEGA, the presented values are lowest estimations obtained in all examined simulation scenarios.
bThe number indicates the order of measurement campaigns performed at same facility.
Supplemental Information
Download PDF (412 KB)Related Research Data
Acknowledgment
The authors are grateful to the produced water disposal companies that allowed them to access the facilities sampled in this study. The authors also appreciate valuable comments from anonymous reviewers that have contributed to the scientific strength of this study.
Supplemental data
Supplemental data for this paper can be accessed on the publisher’s website.