Life Cycle Greenhouse Gas Emissions and Freshwater Consumption Associated with Bakken Tight

Life cycle greenhouse gas emissions and freshwater PNAS PLUS consumption associated with Bakken tight oil

Ian J. Laurenzia,1, Joule A. Bergersonb, and Kavan Motazedib

aCorporate Strategic Research, ExxonMobil Research and Engineering, Annandale, NJ 08801; and bDepartment of Chemical and Petroleum Engineering, University of Calgary, Calgary, AB Canada T2N 1N4

Edited by M. Granger Morgan, Carnegie Mellon University, Pittsburgh, PA, and approved September 23, 2016 (received for review June 6, 2016) In recent years, hydraulic fracturing and horizontal drilling have fining, transportation of crude and refined products, and the final been applied to extract crude oil from tight reservoirs, including the use of refined products. To estimate these impacts in a compre- Bakken formation. There is growing interest in understanding the hensive way (i.e., cradle to grave) we use life cycle assessment greenhouse gas (GHG) emissions associated with the development (LCA) (8). When conducted in accordance with International of tight oil. We conducted a life cycle assessment of Bakken crude Organization for Standardization (ISO) guidelines, LCA yields using data from operations throughout the supply chain, including estimates of the environmental impacts of products in terms of drilling and completion, refining, and use of refined products. If their function or use, which in turn allows environmental compar- associated gas is gathered throughout the Bakken well life cycle, ison with alternative products. then the well to wheel GHG emissions are estimated to be 89 g In this study, we assessed the cumulative greenhouse gas (GHG) – CO2eq/MJ (80% CI, 87 94) of Bakken-derived gasoline and 90 g emissions and freshwater consumption associated with fuels refined – CO2eq/MJ (80% CI, 88 94) of diesel. If associated gas is flared for the from Bakken crude, using functional units of 1 MJ [low heating first 12 mo of production, then life cycle GHG emissions increase by 5% value (LHV) basis] of two of the refined products (gasoline and on average. Regardless of the level of flaring, the Bakken life cycle diesel). To this end, we constructed a life cycle inventory for the GHG emissions are comparable to those of other crudes refined in the Bakken upstream activities using well pad data from XTO Energy United States because flaring GHG emissions are largely offset at the (an ExxonMobil affiliate). This inventory includes data for drilling, refinery due to the physical properties of this tight oil. We also flowback, operations associated with the production of gas, water SCIENCE assessed the life cycle freshwater consumptions of Bakken-derived and crude, corrosion and scale inhibition, and flaring of both as- SUSTAINABILITY – gasoline and diesel to be 1.14 (80% CI, 0.67 2.15) and 1.22 barrel/barrel sociated gas and tank vapors. Our LCA also uses production data – (80% CI, 0.71 2.29), respectively, 13% of which is associated with from XTO Energy and more than 60 other Bakken operators in hydraulic fracturing. North Dakota. It also includes an inventory for the refining of Bakken crude, constructed from the open source Petroleum Re- life cycle assessment | unconventional resources | petroleum | finery Life Cycle Inventory Model (PRELIM) to estimate refinery hydraulic fracturing | flaring GHG emissions (9, 10), as well as ExxonMobil data for refinery freshwater consumption. We provide these inventories among he Bakken tight oil play extends over the Williston basin, in- other supporting data in SI Appendix, sections 2–4. Tcluding areas of Montana and Saskatchewan but primarily lo- The life of a Bakken well begins with drilling, which is typically cated within North Dakota. In 2000, ∼2,000 barrels (bbl) per day of powered by diesel-fueled equipment. Bakken wells descend about Bakken crude (1) were produced on average. After the introduction 10,000 ft before turning horizontal and extend about 10,000 ft of horizontal drilling and hydraulic fracturing to the region, pro- laterally into the Middle Bakken or Three Forks formation. As the duction increased rapidly: production in 2010 was 200,000 barrels well is drilled, multiple layers of cement and steel casings are per day (of 5,475,000 barrels per day of US production), and inserted to prevent loss of hydrocarbons from the completed well exceeded 1.2 million barrels per day (of 9.4 million barrels per day of and prevent contamination of aquifers near the surface. Afterward, US production) throughout most of 2015 (2, 3). This rapid growth of production from the Bakken and other tight Significance oil plays including the Eagle Ford and Niobrara has changed the global energy landscape. For example, imports of Nigerian crudes The growth of production from tight oil plays such as the Bakken to the United States—constituting more than 10% of US crude and Eagle Ford has prompted public interest in understanding imports in 2010 (4) —fell by more than 90% between 2010 and the greenhouse gas (GHG) emissions and freshwater consump- 2015 as a direct consequence of tight oil development (5, 6). Or- tion associated with these resources, specifically with regard to ganization of the Petroleum Exporting Countries (OPEC) crude hydraulic fracturing and flaring. Therefore, we conducted a imports to the United States fell by 41% over the same time in- comprehensive life cycle assessment of Bakken crude, using terval (4). thousands of data from XTO Energy and other Bakken opera- As production from tight oil resources has increased, there has tors, establishing robust estimates of the footprint of current been growing interest in the environmental impacts, particularly in production practices. We conclude that flaring and hydraulic association with hydraulic fracturing and flaring of associated gas, fracturing have a small impact on the life cycle (well to wheel) i.e., gas produced along with oil. In 2011, the US Energy Information GHG emissions associated with Bakken and that these GHG “ Administration (EIA) reported that over one-third of natural gas emissions are comparable to those of other crudes. produced in North Dakota is flared or otherwise not marketed” (7). In practice, there are two sources of flaring in tight oil plays: flaring Author contributions: I.J.L. and J.A.B. designed research; I.J.L. and K.M. performed re- due to the absence of a gas pipeline connection (lack of hookup) search; I.J.L. and K.M. analyzed data; and I.J.L. and J.A.B. wrote the paper. and flaring due to gas pipeline capacity constraints, i.e., a gas Conflict of interest statement: I.J.L. is utilized by ExxonMobil Research & Engineering. pipeline connection to the well exists, but a portion of the gas is This article is a PNAS Direct Submission. flared due to insufficient capacity of the pipeline. Freely available online through the PNAS open access option. Environmental impacts associated with drilling and production— 1To whom correspondence should be addressed. Email: ian.j.laurenzi@exxonmobil.com. — including flaring are a subset of the impacts associated with any This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. crude over its life cycle. Other impacts may be associated with re- 1073/pnas.1607475113/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1607475113 PNAS Early Edition | 1of9 Downloaded by guest on October 4, 2021 the well is completed, which includes hydraulic fracturing (11) and crude slate in 2012 on a per barrel basis (14), but the sources of the insertion of tubing through which the oil is produced. Well emissions differ. The emissions associated with the production of completion also includes flowback, whereby gas, crude oil, and Bakken crude arise from a variety of factors. Flaring associated water (including water injected during fracturing) are produced with pipeline capacity constraints yields 15.2 kg CO2eq/bbl (SI from the well until the flow of gas and liquids is steady. These Appendix,TableS25). Other direct GHG emissions arise from the produced fluids flow to a temporary three-phase separator yielding use of associated gas as fuel for the heater treater, flaring of volatile crude oil, water, and associated gas. Stabilization of the crude oil in organic compounds that separate from crude tanks at the pad, the the separator is accomplished via heating and is fueled by the as- flare pilot (i.e., a small, continuous flame that will ignite hydro- sociated gas that leaves the separator. Crude oil and water are sent carbons when needed), and flaring of associated gas during the to temporary storage tanks, and the associated gas not used as fuel flowback phase of well completion. There are also GHG emissions for the separator is typically flared. Refracturing has not been associated with the generation of electricity used by the pumping necessary for Bakken wells to date, although individual operators unit (pump jack). These GHGs are not emitted at the pad; most have experimented with it on a case-by-case basis. are emitted at centralized offsite fossil-fueled power plants. GHG Once the flow of oil, gas, and water from the well becomes emissions associated with operation of the pumping unit constitute sufficiently steady, the well is connected to a permanent separator the second largest source of upstream GHG emissions: The life known as a heater treater with permanent storage tanks, and the cycle GHG emissions associated with grid power used in this LCA production phase of the well begins. Associated gas not used at the were 691 kg CO2eq/MWh (Methods and SI Appendix,Fig.S11). well pad (e.g., as fuel for phase separation at the heater treater) is Estimates for upstream freshwater consumption are reported in typically routed to a gas gathering system (pipeline). Liquid Fig. 2. Hydraulic fracturing and batch treatments of corrosion and products are intermittently transported by truck to their destina- scale inhibition are the largest direct consumers of freshwater over tions. Produced water is transported to class II salt water disposal the life cycle of the well. However, grid electricity used by the (SWD) sites (12). Crude oil is transported to a refinery via pipeline pumping unit consumes water as well: large-scale power plants or by rail car. There, the crude is converted into gasoline, diesel, consume freshwater due to closed-loop cooling and other opera- and other refined products, which are transported via various tions associated with extraction of power plant fuel (e.g., coal modes (pipeline, rail, truck, and waterway) to service stations mining) (15). Freshwater consumption associated with Bakken around the United States and are used to power vehicles. crude is very time dependent: hydraulic fracturing occurs at the We divide the life cycle of Bakken crude into the following beginning of the well life cycle, whereas batch treatments of cor- phases: drilling and completion, production, crude transportation, rosion and scale inhibitor (which require freshwater diluent; SI refining, transportation of refined products, and vehicle operation. Appendix, section 4.2.5.2), presently amount to around 200 barrels We denote the first two of these as the upstream, and the first five per month and decrease with decreasing production of the well as well to tank. We consider several Bakken operation scenarios over its life. Freshwater consumption associated with grid electricity and refining configurations in this LCA: in our base case, a pipeline use also decreases with decreasing production, due to decreased forthesaleofassociatedgasispresentforthecompletelifeofa power consumed by the pumping unit. well, with the exception of the flowback phase of well completion. The upstream phases of the Bakken life cycle account for 40% However, a fraction of the gas produced will still be flared due to of the freshwater consumption from the well to the refinery—about pipeline capacity constraints. Crude is transported by either rail or athirdofthelifecyclefreshwater consumption. We note that these pipeline (two options) to a medium conversion refinery featuring a fluidized catalytic cracker (FCC) and no residual oil conversion (coking)—representative of many refineries that process Bakken as a feedstock. We also considered a number of scenarios to assess the sensitivity of our results to operating conditions. Among these, we considered the impacts of flaring associated gas for 3, 6, or 12 mo before connecting a well to a gas gathering pipeline. We evaluated the effect of alternative refinery configurations and alternative methods of allocating GHG emissions and freshwater consumption to upstream and refinery coproducts. We also made use of our datasets to quantify the statistical distributions and ranges of environmental impacts. Collectively, the results of our LCA for these cases facilitate (i) comparisons with previously reported estimates of life cycle GHG emissions for gasoline and diesel sourced from other crudes and (ii) evaluation of the relative impacts of specific life cycle stages and operations from “well to wheel.” Results Upstream. As of September 2015, 92% of the Bakken wells in North Dakota were connected to a gas gathering system (SI Appendix, section 1.1.1). Therefore, we selected our base case scenario to be a well that has a gas pipeline connection throughout production, with the exception of the flowback phase of well completion. Despite the gas pipeline connection, some gas will be flared due to pipeline capacity constraints [17% on average (13); SI Appendix,section Fig. 1. Upstream GHG emissions expressed in IPCC AR5 GWPs (100 y). Total 4.2.8]. The total upstream GHG emissions (amortized over the GHG emissions over the well life cycle are amortized over the expected ultimate recovery (EUR) of the well. Relative emissions of CH and CO asso- total volume of crude produced from the well over its lifetime) 4 2 SI Appendix ciated with flaring reflect a 98% flare efficiency (41). XTO, calculated from corresponding to this case are illustrated in Fig. 1 ( , XTO data; ND, calculated from data available from the North Dakota De- Table S25). partment of Mineral Resources (42). GHG emissions associated with (19) The GHG emissions associated with the Bakken upstream (44 kg “other sources” are reported in the SI Appendix, Table S25. Emissions as- CO2eq/bbl crude) are equivalent to those reported for the US sociated with flaring are highlighted.

2of9 | www.pnas.org/cgi/doi/10.1073/pnas.1607475113 Laurenzi et al. Downloaded by guest on October 4, 2021 Bakken occur when refined products are used as transportation PNAS PLUS fuels (tank to wheel); GHG emissions from well to tank (WTT) constitute 18% of the life cycle GHG emissions. GHG emissions associated with the upstream phases (drilling, completion, and production) constitute 8.9% of the life cycle GHG emissions; 3.1% of life cycle GHG emissions arise from flaring due to chal- lenges or constraints on existing gathering systems. Only 0.6% of life cycle GHG emissions are due to operations associated with hydraulic fracturing and flowback flaring, including proppant and additive manufacture and transport. Absolute values of these and other GHG emissions are reported in SI Appendix,TablesS25 and S26. In Fig. 1, we showed that the upstream GHG emissions for Bakken are 44 kg CO2eq/bbl. As we show in Fig. 3, refining of Bakken in a medium conversion refinery featuring fluidized catalytic cracking and no residual oil conversion results in 24.7 kg CO2eq/bbl (5.9 g CO2eq/MJ gasoline) of direct GHG emissions and 2.5 kg CO2eq/bbl (0.60 g CO2eq/MJ gasoline) associated with purchased electricity from the US grid (Methods), assuming no power is cogenerated at the refinery. This estimate is consistent with the reported GHG emissions from the Tesoro Mandan refinery (649,000 ton CO2eq in 2013), which processed about 60,000 barrels Fig. 2. Upstream freshwater consumption for base case LCA of Bakken crude. per day of Bakken crude in 2013 (16). Total freshwater consumption over the well life cycle is amortized over the expected ultimate recovery (EUR) of the well. Freshwater is directly consumed The physical properties of Bakken crude substantially reduce the by hydraulic fracturing and (batch) corrosion and scale inhibition. By contrast, GHG emissions associated with its refining relative to heavy (i.e., the consumption of electricity by the pumping unit indirectly consumes fresh low API/high density) and “sour” (i.e., high sulfur content) crudes. SCIENCE water: this fresh water is consumed at large scale power plants (coal, natural Bakken crude has a low sulfur content (0.097% by mass; SI Ap- gas, nuclear) not located at the well pad. In our analysis, we have not included pendix,TableS15); hence, the hydrocarbon streams generated via SUSTAINABILITY freshwater consumption associated with hydropower. XTO, estimate is made refining require less hydrotreating. Moreover, catalytic reforming using data from XTO Energy. Thirty-nine percent of the upstream freshwater of the relatively large naphtha fraction of Bakken and other tight consumption is associated with hydraulic fracturing. crudes yields more than enough hydrogen to meet the demand from the hydrotreaters. Hence, the carbon-intensive process of estimates include the freshwater consumption associated with all steam methane reforming (SMR) is not likely to be needed to contributors to grid electricity except hydropower (Methods). produce refined products from this feedstock. Many refineries currently processing crude produced in the Bakken formation do Well to Wheel. The life cycle [well to wheel (WTW)] GHG emis- not have a SMR, including Tesoro’s Mandan refinery in North sions associated with Bakken-derived gasoline are reported in Fig. 3 Dakota and both refineries (Delta Trainer and Philadelphia En- using the common functional unit of “megajoule (LHV) of gaso- ergy Solutions) in the Philadelphia metropolitan area (17). Other line.” More than 80% of the GHG emissions associated with tight oil crudes such as Eagle Ford and Niobrara may have similar

Fig. 3. Life cycle GHG emissions associated with gasoline refined from Bakken crude, base case (pipeline connection throughout production, medium FCC). GHG reported in terms of IPCC AR5 GWPs (100-y time horizon). Note: a functional unit of megajoule (LHV basis) of gasoline is used here, whereas a functional unit of bbl crude is used in Figs. 1 and 2. Life cycle GHG emissions for diesel refined from Bakken crude are reported in SI Appendix,Fig.S33. Tank to wheel GHG emissions are adopted from GREET 2013 (35).

Laurenzi et al. PNAS Early Edition | 3of9 Downloaded by guest on October 4, 2021 refinery GHG emissions as a consequence of their similar prop- the LCA: in the absence of a gas pipeline connection, all GHG erties (SI Appendix,section5.6). emissions resulting from flaring of associated gas are allocated to A breakdown of the life cycle (well to wheel) freshwater con- the crude, and less gas is sold as a coproduct over the well life cycle. sumption associated with Bakken-derived gasoline is illustrated in Therefore, the fraction of upstream GHG emissions allocated to Fig. 4. The largest freshwater consumer is the refinery (∼0.65 bbl associated gas will decrease. For instance, GHG emissions associ- consumed/bbl gasoline, or 0.13 bbl consumed/MJ gasoline), with ated with heater treater fuel use are 5.14 kg CO2eq/bbl crude for a other operations including hydraulic fracturing and grid electricity well connected to a gas pipeline through its life cycle. However, if used by the pumping unit consuming smaller quantities. the net associated gas (i.e., gas not used as fuel at the pad) is flared Similar results were obtained for the life cycle impacts associated for the first 12 mo of the well lifetime, GHG emissions associated with diesel manufactured from Bakken (SI Appendix, Figs. S33 and with the heater treater rise to 5.41 kg CO2eq/bbl. S34). For the base case (gas hookup throughout production), life Flaring of associated gas due to the absence of a gathering cycle GHG emissions were 90 g CO2eq/MJ LHV (WTT: 15 g pipeline connection has a smaller impact on the life cycle GHG CO2eq/MJ LHV), and life cycle freshwater consumption was 1.22 bbl emissions than the WTT GHG emissions because most of the freshwater/bbl diesel. life cycle GHG emissions are associated with the combustion of finished fuels (e.g., gasoline). The life cycle GHG emissions for Flaring of Associated Gas. As we have discussed, our base case all four flaring scenarios are less than or equal to the US EPA Bakken well is connected to a gas gathering pipeline after the well Renewable Fuel Standard 2 (RFS2) baseline of 94 g CO2eq/MJ is completed. Current regulations in North Dakota permit opera- LHV (18). tors to flare associated gas from the first well drilled at a well pad for up to a year without a gas pipeline, i.e., all gas that is not used at Characterization of Uncertainty and Variability. The results thus the pad is flared until the well is connected to a gathering system. discussed quantify life cycle impacts of Bakken crude calculated However, subsequent wells are subject to North Dakota flaring using averages of datasets (e.g., the average flowback gas volume) order limits. We illustrate the dependency of the well to tank and or average values of random variables (e.g., the volume of hydro- life cycle GHG emissions on flaring due to lack of gas pipeline carbon vapors that evolve from crude in a tank) of wells in the connection in Fig. 5. Bakken. In practice, a range of life cycle impacts may result due to If gas is flared for the first 6 mo of production, the increase in variability (actual differences in operations or conditions by life life cycle GHG emissions will amount to 2.8 g CO2eq/MJ gasoline cycle phase, actual differences in refinery configuration, etc.). (3.2%); an additional 6 mo of flaring results in an additional 1.7 g Moreover, parameters used in the LCA have intrinsic uncertainty CO2eq/MJ gasoline (1.8%). A similar trend is observed for diesel. that cannot be reduced without the acquisition of additional The impact on the time interval is a consequence of the nonlinear measurements. To quantify the ranges of life cycle impacts due to decline of production of Bakken wells: as time proceeds, less gas is uncertainty and variability, we used Monte Carlo (MC) simulation produced. Decline in production also induces a secondary effect in (Methods). In all simulations, we used a medium conversion refinery with an FCC and no gas oil hydrocracking, which is typical of the type of refinery processing Bakken today. Moreover, the types of refinery expansions currently being considered by industry are represented by this PRELIM configuration, i.e., increasing distil- lation and FCC capacity (19). Results of our Monte Carlo simulations for life cycle GHG emissions associated with Bakken-derived gasoline are illustrated in Fig. 6; results for crude and diesel are reported in SI Appendix,Figs. S43 and S47, respectively. GHG emissions associated with the Bakken upstream range from 32 to 71 kg CO2eq/bbl (80% CI); life cycle GHG emissions associated with Bakken gasoline range from 87 to 94 g CO2eq/MJ LHV, and those of Bakken diesel range from 88 to 94 g CO2eq/MJ LHV. The range of life cycle GHG emissions is largely driven by the variability in the expected ultimate recoveries (EURs) of Bakken wells. The life cycle GHG emissions are not as sensitive to any other individual factor (SI Appendix,Fig.S38). For example, variability due to EUR is much larger than the effect of flaring of associated gas due to absence of a pipeline connection (SI Appendix,Fig.S38). For this reason, we conclude that the method by which EUR is calculated from production time series is critical to the accuracy of the results of an LCA of tight oil (Methods). In SI Appendix, Figs. S46 and S49, we report the results of our MC simulations for the upstream and life cycle freshwater consumptions, respectively. Freshwater consumption associated with the Bakken upstream ranges from 0.25 to 0.96 bbl/bbl crude (80% CI); life cycle freshwater consumption associated with Bakken gasoline ranges from 87 to 94 bbl/bbl gasoline, and the freshwater Fig. 4. Life cycle freshwater consumption associated with Bakken-derived consumption for Bakken diesel ranges from 88 to 94 bbl/bbl diesel. gasoline, base case (pipeline connection throughout production, medium FCC). Unlike GHG emissions, the range of upstream freshwater con- Freshwater consumption associated with hydropower is excluded. Refinery sumption is driven by a combination of the ranges of the freshwater electricity use (MWh/bbl gasoline) was estimated via the PRELIM model, and consumption associated with hydraulic fracturing, scale and corro- freshwater consumption associated with grid electricity (bbl/MWh) was cal- sion inhibition, and the ultimate recovery of oil from the well. Life culated using the data in the US EIA electricity data file and estimates of generator freshwater consumption from NETL and NREL. Thirteen percent of cycle impacts are also defined by several factors: both those asso- the life cycle freshwater consumption is associated with hydraulic fracturing. ciated with the upstream as well as EUR, and refinery freshwater EM, average ExxonMobil refinery freshwater consumption. 1 bbl gasoline = consumption. The results of sensitivity analyses for both upstream 5,144 MJ (LHV basis). and life cycle results are reported in SI Appendix,Figs.S40–S42.

4of9 | www.pnas.org/cgi/doi/10.1073/pnas.1607475113 Laurenzi et al. Downloaded by guest on October 4, 2021 PNAS PLUS SCIENCE Fig. 5. Effect of flaring of associated gas on the life cycle GHG emissions for gasoline refined from Bakken in a medium-conversion refinery with an FCC. GHG emissions reported using IPCC AR5 GWPs (100-y time horizon). SUSTAINABILITY

Comparison with Other Studies. Generally speaking, detailed com- terms of 100-y GWPs reported in the Fifth Assessment Report parisons cannot be made among the findings of alternative LCAs of the Intergovernmental Panel on Climate Change (IPCC of similar products due to different methods of modeling, types of AR5). Generally speaking, our estimate of life cycle GHG input data used, system boundaries, etc. However, it is informative emissions associated with Bakken-derived gasoline are consis- to compare among alternative studies. In SI Appendix,Fig.S50,we tent with estimates of life cycle GHG emissions associated with compare our findings to other frequently cited studies of life cycle gasoline derived from other sources (92–96 g CO2eq/MJ). One impacts of petroleum fuels (18, 20–24). We only compared our exception is the estimate of the recently published WTW study results to studies that explicitly reported CO2,CH4 and N2O of the Joint Research Centre-EUCAR-CONCAWE (JEC) emissions, thereby allowing for the expression of all study results in collaboration (23). The JEC reports an estimate of 87 g CO2eq/MJ, which is the lowest estimate of well to wheel GHG emissions among these studies. Differences among alternative LCAs of refined products may be attributed to different modeling assumptions, particularly regarding refining, crude production, transportation, and the impacts associated with electricity. Moreover, they have included different crudes. For instance, the NETL Petroleum Baseline and GHGe- nius (20) explicitly include oil sands among the crudes used to manufacture the average megajoule of gasoline. Discussion In this work, we discussed the impact of current Bakken operations, but changes in operations are already in progress. For instance, many operators delay production until a gathering pipeline is connected to the equipment at the pad: a practice preceding North Dakota Industrial Commission (Flaring) Order 24665, which required the capturing of 74% of associated gas by the end of 2014, 77% of associated gas by March 2016, and greater limits being phased in thereafter (25). Furthermore, new developments in separations technology require less fuel than current heater treaters, further lowering GHG emissions at the pad. Increasing the capacity of gathering systems to eliminate production-related Fig. 6. Distributions of the life cycle GHG emissions from Bakken-derived flaring would decrease life cycle GHG emissions from 89 to 87 g gasoline, calculated via MC simulation for all four associated gas pipeline CO eq/MJ (2%). cases; the first (zero months until pipeline connection) is characteristic of 2 most of the wells in the Bakken shale play (Methods). Blue lines, 10th and During the preparation of this manuscript, Brandt et al. pub- 90th percentiles; gray lines, results from LCAs using expectation values of lished an LCA of Bakken crude at the website of Argonne National distributed variables; green lines, EPA RFS2 baseline. Results are reported in Labs (26). This study, like ours, used data from the North Dakota terms of IPCC AR5 GWPs (100-y time horizon). Department of Mineral resources. However, the data inventory did

Laurenzi et al. PNAS Early Edition | 5of9 Downloaded by guest on October 4, 2021 not include flowback gas volumes or Bakken-specific refinery GHG Recently published work has called attention to fugitive emis- estimates. Due to the minimal impact of flowback flaring, their sions in the Bakken shale play (33–35). Some of these studies used estimate of GHG emissions from well to refinery gate (i.e., all satellite or airplane flux measurements, i.e., top-down approaches. operations preceding the refinery) was 8.8 kg CO2eq/MJ gasoline, Such approaches measure or estimate emissions that fall outside similar to ours (Fig. 3). However, Brandt et al. overestimated re- of the oil system boundary, such as fugitive emissions associated finery emissions, resulting in an overestimate of the life cycle GHG with gathering pipelines, compressors, and gas treatment and emissions associated with gasoline refined from Bakken crude. processing plants. To construct a bottom-up inventory of GHG Both studies report similar emissions to the US Environmental emissions for the entire Bakken shale play, one would need to Protection Agency (EPA) Petroleum Baseline. quantify these impacts and integrate them with the impacts There are considerably fewer investigations of freshwater con- assessed in this study. These impacts would be different from sumption to which we may compare our own, as tight oil is an those associated with other “rich” gases (e.g., those of the western emerging energy resource. Recently, Scanlon et al. (27) estimated Marcellus) due to differences in the operating pressures of the the freshwater consumption associated with hydraulic fracturing of gathering systems as well as gas composition. These differences, in Bakken crude from the Central Basin to be 3.08 million gallons turn, necessitate different technologies for separating natural gas (gal)/well, which is consistent with our data (mean: 3.06 million liquids from the associated gas, which will change the emissions gal/well, 80% CI: 2.5–3.6 million gal/well; SI Appendix,Fig.S17). profile for the gas in the Bakken (36). Although we have not in- They also estimated volumes ranging from 0.82 to 2.01 million gal/ vestigated these impacts in this study, they could be investigated in well for other regions of the Bakken tight oil play. However, they future work. adoptedanEURof8.2× 106 gal of oil equivalent per well (195 One recent study has, however, revealed specific equipment kbbl/well), which is about half our estimate of the average Bakken sources of fugitive emissions via helicopter-based infrared camera well EUR. This results in a large difference between our esti- surveys of Bakken pads (35). That study reported 13.8% of Bakken mate of freshwater consumption per barrel of crude produced over pads yield fugitive emissions and that 93% of those emissions were the well lifetime and that of Scanlon et al. As we discussed pre- associated with tank vapors. As SI Appendix,Fig.S20shows, tank viously, this illustrates the importance of the EUR in LCA of vapors have an average methane composition of 20% (volume). tight oil. The other hydrocarbon components are not greenhouse gases, but King and Webber (28) and the researchers at Argonne National do yield two to seven times more CO2 than methane on combustion Laboratory (24) were among the first to assess the life cycle as a consequence of stoichiometry. Therefore, the GHG emissions freshwater consumption associated with products refined from associated with venting tank vapors are slightly lower than those for crude oil. Our findings for the Bakken life cycle substantially dis- flaring them (SI Appendix,section4.2.3). If one were to model the agree with these studies due to (i) differences in upstream opera- disposition of tank vapors as being vented (e.g., via a leaky tank tions and (ii) differences in life cycle inventory data. Previous hatch) rather than flared, then the upstream GHG emissions de- studies associated upstream water consumption with “water crease from 44 (Fig. 1) to 43 kg CO2eq/bbl, and the life cycle GHG flooding,” steam injection, and/or CO2 injection for enhanced oil emissions decrease from 89.1 to 89.0 g CO2eq/MJ gasoline. Al- recovery (29); however, none of these operations are used for tight though fugitive emissions do not impact the carbon intensity of oil wells in the Bakken or elsewhere (e.g., the Eagle Ford). Hence, Bakken crude fuels, we believe experimental investigations of tank Argonne’s estimate of 26.48 gal of freshwater are consumed for vapor and other crude-specific emissions akin to the investigations every MMBtu of crude produced on-shore, which is a factor of 10 of Allen et al. for natural gas systems (37–39) are timely. larger than our estimate of upstream freshwater consumption, is not pertinent to Bakken crude. Furthermore, Argonne estimates Conclusions refinery freshwater consumption at “1.53 gal of water is consumed Our data-driven results show that there is a significant overlap foreachgallonofcrude” (29), which is based on earlier studies between the ranges of life cycle (well to wheel) GHG emissions (30–32). However, as our data show (SI Appendix,Fig.S25), associated with Bakken tight oil and the recent mixture of crudes ExxonMobil refineries consume less than half of this volume of refined in the United States. Moreover, upstream flaring has a freshwater on average. relatively small effect on the life cycle GHG emissions. Indeed, To our knowledge, our LCA is the first to assess the GHG our comprehensive analysis shows that the life cycle GHG emis- emissions and freshwater consumption associated with corrosion sions from Bakken-derived gasoline and diesel are lower than the and scale inhibition. In some cases, corrosion and scale inhibitors US EPA’s petroleum GHG emission baselines for the Renewable are added continuously, usually due to high salt concentrations in Fuel Standard (18), regardless of the duration of flaring. If asso- the produced water. In such cases, freshwater is not injected along ciated gas is flared for the first 12 mo of production (370 mo over with the inhibitors. Our LCA featured a representative distribution the average well life), life cycle GHG emissions associated with the of wells using both continuous and batch treatments for corrosion refined products will increase by 5% on average (4.5 g CO2eq/MJ and scale inhibition (SI Appendix, section 4.2.5.2). gasoline, 4.2 g CO2eq/MJ diesel), still remaining on par with the Ultimately, GREET 2014 reports 71 gal freshwater consumed/ EPA baseline. MMBtu (LHV) of gasoline (about 8 bbl/bbl gasoline, and 29 gal As our findings illustrate, the largest GHG emissions sources of freshwater consumed/MMBtu (LHV) of diesel (about 3.7 bbl/bbl the Bakken upstream are (i) flaring of associated gas during pro- diesel). By contrast, King and Webber, using similar assumptions duction, (ii) prime movers associated with the pumping unit, regarding conventional production practices, arrived at a life cycle (iii) fuel consumption associated with heater treaters, and (iv)flaring water consumption of 1.4–2.9 bbl/bbl (28). Although our estimates of tank vapors. However, conventional crudes extracted from on- of life cycle freshwater consumption for Bakken are lower than shore reservoirs often feature additional operations over their life both previous estimates for conventional crudes, we do not believe cycles, including those associated with secondary and tertiary re- there is sufficient evidence to claim that life cycle freshwater con- covery. Sound comparison of the GHG emissions associated with sumption for tight oil is less than that of “conventional” crudes conventional and tight crudes requires explicit consideration of the produced on-shore. Estimates of freshwater consumption for en- time evolution of field operations over their life cycles. We consid- hanced oil recovery (EOR) used by previous studies appear to be ered the time evolution of Bakken production in this LCA, but >20 y old and are possibly nonrepresentative for current operating the time evolution of conventional crude production has not been practices (32). Therefore, the apparent difference between Bakken sufficiently captured in previous studies. Therefore, we believe new and other crudes produced on-shore may be due to old data or research addressing the freshwater consumption associated with modeling artifacts. conventional on-shore production, including EOR, is timely and will

6of9 | www.pnas.org/cgi/doi/10.1073/pnas.1607475113 Laurenzi et al. Downloaded by guest on October 4, 2021 help quantify differences, if any, between the freshwater consump- crude characteristics. Light end products are used as refinery fuel gas, repre- PNAS PLUS tion associated with tight and conventional crudes. senting an upper bound for GHG emissions and a lower bound for resource utilization (e.g., LPG is not part of the product slate, nor are chemical feedstocks Methods such as ethylene). In plant-level allocation, refinery impacts are allocated to refined products in The goal of this investigation was to estimate the life cycle or cradle to grave proportion to their fractions of the energy (or mass) of total refined products. GHG emissions and freshwater consumption associated with gasoline and diesel Process-level allocation, by contrast, requires explicit accounting of material refined from Bakken crude. The estimates are intended to (i) facilitate com- through each unit of the refinery. The PRELIM model used for the estimation of parisons with previously reported estimates of life cycle GHG emissions for refinery impacts accounts for the flows of hydrocarbons through individual gasoline and diesel (i.e., sourced from other crudes), (ii) evaluate the impacts refinery process units. Therefore, we were able to investigate the effect of both associated with specific operations (e.g., flaring), and (iii) explicitly investigate allocation methods. In our base case, we use process-level allocation. the effects of modeling choices on the results (e.g., allocation of impacts at Environmental impacts for all operations up to and including the refinery the refinery). are allocated to the aforementioned refined products via energy content. We used a functional unit of megajoule of fuel combusted (LHV basis) for both gasoline and diesel, adopting a common metric for liquid fuels. However, Sensitivity and Scenario Analysis. There are several sources of uncertainty and we note that comparison of the life cycle GHG emissions associated with diesel variability in the well to tank emissions of Bakken crudes that must be taken into and gasoline necessitates consideration of the engine efficiencies and drive account to inform decisions about the implications of GHG emissions. These cycles associated with vehicles that use these fuels, i.e., a functional unit of miles sources include variability in performance across wells (e.g., estimated ultimate of transport (for light duty vehicles) or ton-miles of transport (for heavy-duty recovery of oil) and variability in operating conditions (e.g., emissions intensity vehicles). Therefore, results for diesel and gasoline reported in this work should of electricity consumed). We conducted sensitivity analyses to explore the effect not be directly compared with each other. of variability and uncertainty on our estimates of life cycle GHG emissions. SI Appendix, Tables S30 and S31 summarize the variables that are included in the System Definition. The tight oil system boundary included operations such as uncertainty analyses. drilling, hydraulic fracturing, separation of crude, produced water and asso- We evaluated the sensitivity of our results to flaring associated with lack of ciated gas at the pad, flaring, transportation of crude, refining, transportation infrastructure as follows: we considered scenarios in which associated gas is of refined products, and end use (e.g., combustion as fuel for vehicles). We also flared due to lack of pipeline connection for the first 3, 6, and 12 mo of include the life cycle GHG emissions and freshwater consumption associated production. In these scenarios, the impacts associated with flaring are allocated with the generation of electricity used in these processes, cementing and casing entirely to oil. As previously discussed, our base case scenario features a pipeline of wells, proppant and gel frac additives used for hydraulic fracturing, and scale connection for gas from the beginning of production, i.e., gas is flared only

and corrosion inhibitor used during the production phase of the well life cycle. during completion and as a consequence of upsets associated with infra- SCIENCE Additional details including an illustration of the system boundary are provided structural constraints. We conducted assessments for a maximum of 12 mo of SUSTAINABILITY in SI Appendix, section 1 and Fig. S1. We note that equipment and operations flaring because the US Bureau of Land Management limits flaring to 1 y for “on- associated with the associated gas gathering system (e.g., gas compression and shore federal and Indian oil and gas leases” (45) and North Dakota Industrial dehydration) are not included in the system boundary, following the ap- Commission (Commission) order no. 2466549, which limited flaring to 15% of proaches of the US EPA (40, 41) and prior LCAs of petroleum-derived gasoline produced gas in 2015. From our production data, we estimate that Bakken (20). The magnitude of fugitive emissions associated with the crude oil path- wells will produce 110 million scf (MMscf) in the first year and 505 MMscf over way are tested in the sensitivity analysis. the well lifetime. Thus, we consider flaring of gas for the first 12 mo to be an In September 2015, the North Dakota Department of Mineral Resources oil appropriate upper limit for our sensitivity analysis. and gas database (42) reported that 92% of ∼10,000 Bakken wells were con- We considered the sensitivity of our results to the refinery configuration: the nected to gas gathering systems (SI Appendix, section 1.1.1). The production use of plant-wide or process-level allocation at the refinery. A sensitivity analysis from these pools exceeded 95% of the oil production from North Dakota. evaluating the effect of plant-level and process-level allocation is reported in SI Therefore, we defined our base case Bakken well configuration to be a well Appendix, Figs. S38 and S39 and Tables S26 and S27. We also investigated the featuring a pipeline connection for gas from the beginning of production, i.e., effect of mass- vs. energy-based allocation in our sensitivity analysis (SI Appendix, gas is flared only during completion and as a consequence of upsets associated Figs. S38 and S39). with infrastructural constraints. We then explored the impact of pipeline capacity constraints in the sensitivity analysis. MC Simulation. To quantify the ranges of life cycle impacts due to uncertainty and variability, we used MC simulation, adopting the procedure described by Impact Assessment. We assessed freshwater consumption and GHG (CO2,CH4, Laurenzi and Jersey (36). Each of the variables listed in SI Appendix,TableS30 and N2O) emissions in this study. GHG emissions were quantified in terms of was selected randomly from its distribution or data set in each MC trial. CO2 equivalents, using 100-y GWPs from the Fifth Assessment Report of the Through testing, we determined that 5,000 MC trials were sufficient to char- IPCC (43) (AR5), in accordance with the recommendations of the Kyoto pro- acterize the distributions of GHG emissions and freshwater consumption as- tocol (44). The GWPs for methane and N2O (100-y, AR5) are 30 g CO2eq/kg CH4 sociated with the life cycle of Bakken crude, as well as individual life cycle and 265 kg CO2eq/kg N2O, respectively. Methane emissions primarily arise as a phases. Emissions associated with combustion of gasoline and diesel were consequence of uncombusted associated gas and fugitives associated with considered constant in these simulations; the coefficient of variation for these heater treaters and gas meters. emissions has been reported to be 2% (46). Direct refinery GHG emissions per barrel of crude were also considered to be constant. Coproduct Accounting. In addition to oil, Bakken wells produce water and (associated) gas. Associated gas is used as a fuel for equipment at the well pad, Characteristics of Bakken Hydrocarbons. Bakken is a light, sweet crude (API > 38, sold as a coproduct, or flared when no infrastructure exists or is insufficient. sulfur < 0.5%wt). The hydrocarbon-rich associated gas has a methane content Therefore, environmental impacts associated with drilling, completion, and of 52.8% by volume. Additional characteristics are reported in SI Appendix, production (including flaring) are allocated to both oil and gas. Table S15. Based on the raw gas composition, we estimate GHG emissions of

Upstream GHG emissions and freshwater consumption are allocated to oil 0.0886 kg CO2/scf of combusted associated gas and 0.3064 kg CO2eq/scf of and associated gas that is delivered to a gathering pipeline. GHG emissions vented or fugitive associated gas. associated with the flaring of associated gas due to the absence of a pipeline connection are completely allocated to crude because there is no coproduct in Electricity and Fuel Impacts. Fuels and electricity are used in many processes this case. However, GHG emissions associated with flaring caused by capacity throughout the life cycle of Bakken crude. Impacts associated with US electricity constraints are allocated to both associated gas and oil. In the base case, 78.4% were estimated from National Energy Technology Laboratory (NETL) and of the upstream GHG emissions and freshwater consumption are allocated to Argonne LCAs of power generation (24, 47) and the US grid mix of 2013. crude oil, corresponding to the relative amounts of energy of crude oil and Impacts associated with the use of non–Bakken-derived diesel (e.g., for dril- associated gas sold. We explore the sensitivity of this allocation in SI Appendix, ling) were adopted from the NETL Petroleum Baseline study (20) (SI Appendix, section 5.3.1 and 5.3.2. Table S9) and King and Webber (28). Additional details associated with elec- Downstream GHG emissions are allocated to gasoline, jet fuel, ultra-low sulfur tricity and fuel impacts are provided in SI Appendix,section2.

diesel, heating fuel oil, bunker fuel/asphalt, petcoke, and surplus H2 generated via catalytic reforming of naphtha (if any), subject to the limitations of the Transportation Impacts. Refined products are transported by pipeline, rail, PRELIM model. The product slate is defined by the refinery configuration and truck, and waterway. We followed the method used by NETL (20) to assess

Laurenzi et al. PNAS Early Edition | 7of9 Downloaded by guest on October 4, 2021 the life cycle impacts of each mode of transport using updated data of 98%, following the convention of the US EPA (41). In our sensitivity analysis, sources (48–50). Electricity consumption associated with pipelines for both we varied the flare destruction efficiency from 90% to 100% to investigate its crude and refined products was modeled using data reported by Hooker effect on the upstream and life cycle GHG emissions (SI Appendix, section 5.3.1). (51) for seven pipeline operators in 1981, which should serve as a con- We model the gas that is not combusted (2%) as vented. servative upper bound for electricity use. Additional details are provided in SI Appendix, section 2.2. Refining Impacts. We used the PRELIM to estimate the GHG emissions and product slates associated with the refining of Bakken crude (9, 10). PRELIM is a Production Impacts. For about 3 mo after well completion, oil, gas, and water mass and energy based process unit-level tool that permits estimation of en- flow due to natural pressure differences between the reservoir and surface. ergy use and GHG emissions associated with processing of specific crude oils Subsequently, a pumping unit (pump jack) will be used for artificial lift. Some of within a range of configurations in a refinery. In PRELIM, a material balance the gas is used as fuel for the heater treater (and in some cases, an engine that connects the different process units along with flow splitters that together drives the pumping unit); the rest is sold or flared. Water and oil flow from the simulate different refinery configurations, whereas flows are split using ratios heater treater to separate tanks. Hydrocarbon vapors entrained in the crude that have been obtained from discussions with (non-ExxonMobil) refinery and water evolve from solution and flow through piping to the flare. In this experts (SI Appendix, section 4.4.1). LCA, we have not considered vapor recovery from the tanks (i.e., repressuri- Two Bakken crude oil assays were analyzed, to investigate the sensitivity of zation and injection of tank vapors into the gathering system). the results to the particular assay. The model was run for both assays in all seven The EUR of wells is a key factor in the environmental assessment of tight oil. refinery configurations: hydroskimming, medium conversion [with FCC, with In this study, we calculated EURs for 3513 hydraulically fractured Bakken wells gas-oil hydrocracking, and with FCC and gas oil hydrocracking (GO-HC)], and (horizontally) drilled since 2010, using production data from the IHS Enerdeq deep conversion with FCC, with GO-HC, and with FCC and GO-HC. These sets of service (52). The wells were drilled and completed by all Bakken operators. We runs were conducted with both mass- and energy-based allocation to refined regressed EURs from production data using the technique of Ilk (53), as it is products to investigate the impact of allocation method on the results. well suited for modeling of unconventional gas and oil production. Well lifetimes were estimated concurrently using the same dataset. We then ex- Mass or energy allocation methods were used to allocate the GHG cluded wells for which water production and oil production data were in- emissions to all final products [i.e., gasoline, jet fuel, diesel, fuel oil, bunker complete. Production figures (kbbl crude, kbbl produced water, EUR, and well C, coke, sulfur, and naphtha catalytic reformer (NCR) hydrogen, if a surplus lifetime) for the resulting 2264 wells were ultimately used. The average EUR is of hydrogen existed], subject to the refinery configuration. For instance, if a 372 kbbl, and average lifetime is 370 mo. Additional production data are surplus of hydrogen was produced for a particular run, then emissions provided in SI Appendix, Table S16 and Fig. S12. would be allocated to it. In all PRELIM runs, results were reported based on We also used production data from 57 wells drilled and completed by XTO the higher heating values (HHV) of final products, and we considered only Energy (an ExxonMobil affiliate) in 2013. These data included gas oil ratios, straight run naphtha to be fed to the NCR. Additional information is lengthsofwellcasings,linersandtubing,cementuse,proppantuse,dieseluse provided in SI Appendix. for hydraulic fracturing and drilling, and flowback gas volumes. Key data and Some refinery process units use water as a heat transfer medium for cooling statistics for these wells are reported in SI Appendix, Table S17. or heating, whereas others use water directly in refinery processes. Although Additional data and figures of pertinence to the assessment of impacts refineries reuse steam, process water and cooling water, some water is lost due associated with casings (54), cement, corrosion and scale inhibitors, hydraulic to evaporation (e.g., in cooling water towers), blowdown, or other processes. fracturing additives, artificial lift (operation of the pumping unit), fugitives This water is consumptively lost, and is replenished with fresh water. PRELIM (40), tank recirculation pumps (55), tank vapors (56), class II disposal of pro- does not currently account for freshwater consumption. Therefore, we mod- duced water, flaring due to infrastructural constraints (46), and crude trans- eled freshwater consumption for refining using data from 32 ExxonMobil re- portation modes (13) and distances are provided in SI Appendix,section4.1. fineries in 2013 (SI Appendix). The average refinery freshwater consumption is XTO Energy uses the same service providers used by other oil and gas com- about 0.7 bbl/bbl crude refined; the median is about 0.5 bbl/bbl crude refined. panies for drilling, hydraulic fracturing, and other upstream operations that We allocated freshwater to refined products at a plant-wide level. directly generate GHG emissions. Therefore, it is highly likely that the analysis we conducted is representative of the industry generally, although we have Fuel Combustion. GHG emissions associated with the combustion of fuels re- not conducted a formal survey of service providers to compare practices for fined from Bakken crude (tank to wheel) are adopted from GREET 2013 (24) (SI XTO wells with practices for other wells. XTO produced about 6.5% of Bakken Appendix, Table S10). production in 2015. Fugitive emissions associated with heater treaters (which use mechanical ACKNOWLEDGMENTS. We thank Matthew Briney, Derick Lucas, Nate dump valves in lieu of pneumatic level controllers) and gas meters (which Kurczewski, Robby Denton, Martin Wipf, Gib Jersey, Paul Krishna, delineate the boundary between the gas and oil system boundaries) were Michael O’Connor, Jessica Abella, Dharik Mallapragada, and Victor modeled using emission factors from the EPA GHG Inventory (40). DePaola for providing subject matter expertise and technical data used GHG emissions associated with flowback flaring depend on the flare efficiency, in this study. This investigation was funded by ExxonMobil Research i.e., the fraction of gas that is combusted. In this analysis, we use a flare efficiency & Engineering.

1. Energy Information Administration (EIA) (2011) Bakken formation oil and gas drilling 9. Abella JP, Bergerson JA (2012) Model to investigate energy and greenhouse gas activity mirrors development in the Barnett. Available at www.eia.gov/todayinenergy/ emissions implications of refining petroleum: Impacts of crude quality and refinery detail.cfm?id=3750. Accessed March 6, 2014. configuration. Environ Sci Technol 46(24):13037–13047. 2. Energy Information Administration (EIA) (2015) Bakken Region Drilling Productivity 10. Abella JP, Motazedi K, Bergerson JA (2015) Petroleum refinery life cycle inventory Report November 2015. Available at www.eia.gov/petroleum/drilling/pdf/bakken.pdf/. model (PRELIM). Available at ucalgary.ca/lcaost/prelim. Accessed March 6, 2015. Accessed November 23, 2014. 11. Energy API, Hydraulic fracturing: Safe oil and natural gas extraction. Available at www.api. 3. Energy Information Administration (EIA) (2016) Crude oil production. Available at org/oil-and-natural-gas/wells-to-consumer/exploration-and-production/hydraulic-fracturing/ www.eia.gov/dnav/pet/pet_crd_crpdn_adc_mbblpd_a.htm. Accessed July 27, 2016. fracking-safe-oil-gas-extraction. Accessed January 27, 2016. 4. Energy Information Administration (EIA) (2016) U.S. imports by country of origin. Avail- 12. US Environmental Protection Agency, Class II oil and gas related injection wells. Avail- able at https://www.eia.gov/dnav/pet/pet_move_impcus_a2_nus_epc0_im0_mbblpd_a.htm. able at https://www.epa.gov/uic/class-ii-oil-and-gas-related-injection-wells. Accessed Oc- Accessed July 27, 2016. tober 26, 2016. 5. Energy Information Administration (EIA) (2014) U.S. imports of crude oil from Africa 13. North Dakota Pipeline Authority, Monthly update. Available at https://northdakotapipe- fell by more than 90 percent between 2010 and early 2014. Available at www.eia.gov/ lines.com/directors-cut/. Accessed February 8, 2015. petroleum/weekly/archive/2014/140521/twipprint.html. Accessed March 19, 2014. 14. Forrest J, Dereniwski C, Birn K (2014) Comparing GHG intensity of the oil sands and 6. US Energy Information Administration (2013) Recent decline in Gulf Coast crude oil the average US crude oil (IHS Cera). Available at https://www.ihs.com/products/energy- imports mainly affects lighter grades. This Week in Petroleum October 30. Available at industry-oil-sands-dialogue.html. Accessed October 26, 2016. https://www.eia.gov/petroleum/weekly/archive/2013/131030/twipprint.html. Accessed 15. Skone TJ, Adder JM (2012) Power Systems Life Cycle Analysis Tool (Power LCAT) (US October 26, 2016. DOE National Energy Technology Laboratory, Washington, DC). 7. Energy Information Administration (EIA) (2011) Over one-third of natural gas pro- 16. US Environmental Protection Agency (2015) GHG reporting program data sets. duced in North Dakota is flared or otherwise not marketed. Available at www.eia. Available at https://www.epa.gov/ghgreporting/ghg-reporting-program-data-sets. gov/todayinenergy/detail.cfm?id=4030. Accessed March 19, 2014. Accessed October 26, 2016. 8. International Organisation for Standardisation (ISO) (2006) Environmental Management— 17. US Environmental Protection Agency, Refinery capacity report. Available at www.eia. Life Cycle Assessment—Principles and Framework (ISO, Geneva). gov/petroleum/refinerycapacity/. Accessed January 27, 2016.

8of9 | www.pnas.org/cgi/doi/10.1073/pnas.1607475113 Laurenzi et al. Downloaded by guest on October 4, 2021 18. US Environmental Protection Agency (2010) Renewable fuel standard program (RFS2) 37. Allen DT, et al. (2013) Measurements of methane emissions at natural gas production PNAS PLUS regulatory impact analysis, EPA-420-R-10-006. Available at https://www.epa.gov/otaq/ sites in the United States. Proc Natl Acad Sci USA 110(44):17768–17773. renewablefuels/420r10006.pdf. Accessed April 8, 2014. 38. Allen DT (2014) Methane emissions from natural gas production and use: reconciling 19. Energy Information Administration (EIA) (2015) Technical options for processing addi- bottom-up and top-down measurements. Curr Opin Chem Eng 5:78–83. tional light tight oil volumes within the United States. Available at www.eia.gov/analysis/ 39. Allen DT, et al. (2015) Methane emissions from process equipment at natural gas pro- studies/petroleum/lto/pdf/lightightoil.pdf. Accessed January 27, 2016. duction sites in the United States: liquid unloadings. Environ Sci Technol 49(1):641–648. 20. Gerdes K, Skone TJ (2008) Development of Baseline Data and Analysis of Life Cycle 40. US Environmental Protection Agency (2014) Inventory of U.S. Greenhouse Gas Emissions Greenhouse Gas Emissions of Petroleum-Based Fuels (US DOE National Energy Tech- and Sinks: 1990-2012 (US Environmental Protection Agency, Washington, DC). nology Laboratory, Washington, DC). 41. US Environmental Protection Agency/Gas Research Institute (1996) Methane Emissions 21. O’Connor D (2014) GHGenius 4.03a–A model for life cycle assessment of trans- from the Natural Gas Industry, Volume 6: Vented and Combustion Sources,eds portation fuels. Available at www.ghgenius.ca/. Accessed June 16, 2014. Harrison M, et al. (Radian International LLC for National Risk Management Research 22. California Air Resources Board (CARB), CA-GREET 2.0. Available at https://www.arb.ca. Laboratory, Research Triangle Park, NC). gov/fuels/lcfs/ca-greet/ca-greet.htm. Accessed November 14, 2014. 42. North Dakota Industrial Commission, Department of Mineral Resources, Oil and Gas 23. Joint Research Centre/European Oil Company Organization for Environment, Health, Division. Available at https://www.dmr.nd.gov/oilgas/. Accessed February 2014. and Safety/European Council for Automotive Research and Development (2014) Well- 43. IPCC (2013) Climate Change 2013: The Physical Science Basis. Contribution of Working to-Wheels Analysis of Future Automotive Fuels and Powertrains in the European Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Context. Available at iet.jrc.ec.europa.eu/about-jec/downloads. Accessed October 26, Change (Cambridge Univ Press, Cambridge, UK). 2016. 44. United Nations Framework Convention on Climate Change (1997) The Kyoto Protocol 24. Argonne National Laboratory (2014) The Greenhouse Gases, Regulated Emissions, to the United Nations Convention on Climate Change. Available at unfccc.int/cop5/ and Energy Use in Transportation (GREET) Fuel Cycle Model (Transportation Tech- resource/docs/cop3/07a01.pdf. Accessed April 1, 2015. nology R&D Center, US Dept Energy, Argonne, IL). 45. US Department of the Interior, Notice to lessees and operators of onshore federal and 25. North Dakota Industrial Commission (2015) Commission order no 24665. Available at indian oil and gas leases (NTL-4A), section IV. Available at www.blm.gov/wy/st/en/ https://www.dmr.nd.gov/oilgas/GuidancePolicyNorthDakotaIndustrialCommission- programs/energy/Oil_and_Gas/docs/ntl_4a.html. Accessed October 26, 2016. order24665.pdf. Accessed March 6, 2016. 46. Venkatesh A, Jaramillo P, Griffin WM, Matthews HS (2011) Uncertainty analysis of life 26. Brandt AR, et al. (2015) Energy intensity and greenhouse gas emissions from crude oil cycle greenhouse gas emissions from petroleum-based fuels and impacts on low production in the Bakken formation: Input data and analysis methods. Available at carbon fuel policies. Environ Sci Technol 45(1):125–131. https://greet.es.anl.gov/publication-bakken-oil. Accessed July 27, 2016. 47. Drennen TE, Andruski J (2012) Power Systems Life Cycle Analysis Tool (Power LCAT) 27. Scanlon BR, Reedy RC, Nicot JP (2014) Comparison of water use for hydraulic frac- (US DOE National Energy Technology Laboratory, Washington, DC). turing for unconventional oil and gas versus conventional oil. Environ Sci Technol 48. US Energy Information Administration (2011) Voluntary reporting of greenhouse 48(20):12386–12393. gases program (technical assistance): Fuel emission factors. Available at www.eia.gov/ 28. King CW, Webber ME (2008) The water intensity of the plugged-in automotive oiaf/1605/coefficients.html. Accessed March 6, 2014. economy. Environ Sci Technol 42(12):4305–4311. 49. Davis SC, Diegel SW, Boundy RG, Moore S (2014) 2013 Vehicle Technologies Market 29. Wu M, Chiu Y (2011) Consumptive Water Use in the Production of Ethanol and Report (US DOE Oak Ridge National Laboratories, Oak Ridge, TN). Petroleum Gasoline—2011 Update (Argonne National Laboratories, Argonne, IL). 50. Davis SC, Diegel SW, Boundy RG (2012) Transportation Energy Data Book, 31 Ed (Oak SCIENCE

30. Ellis M, Dillich S, Margolis N (2001) Industrial Water Use and Its Energy Implications Ridge National Laboratories, Oak Ridge, TN). SUSTAINABILITY (Energetics Incorporated for the US Department of Energy, Office of Energy Efficiency 51. Hooker JN (1981) Oil Pipeline Energy Consumption and Efficiency (Oak Ridge Na- and Renewable Energy, Office of Industrial Technologies, Washington, DC). tional Laboratories, Oak Ridge, TN). 31. Arena BJ, Buchan MS (2006) Water and the refinery—An introduction to growing 52. IHS Markit, US well and production data. Available at https://www.ihs.com/industry/ issues impacting refinery water use. Proceedings of the American Institute of energy-services.html. Accessed June 13, 2014. Chemical Engineers (AIChE) Chicago Symposium. Available at https://web.archive.org/ 53. Ilk D, Perego AD, Rushing JA, Blasingame TA (2008) Integrating multiple production web/20080703203517/http://www.aiche-chicago.org/symposium06/abstract.htm. analysis techniques to assess tight gas sand reserves: Defining a new paradigm for Accessed October 27, 2016. industry best practices. Proceedings of the CIPC/SPE Gas Technology Symposium 2008 32. Gleick P (1994) Water and energy. Annu Rev Energy Environ 19:267–299. Joint Conference (Society of Petroleum Engineers, Richardson, TX). 33. Kort EA, et al. (2016) Fugitive emissions from the Bakken shale illustrate role of shale 54. Norgate TE, Lovel RR (2004) Water Use in Metal Production: A Life Cycle Perspective production in global ethane shift. Geophys Res Lett 43:4617–4623. (CSIRO Minerals, Clayton South, VIC, Australia). 34. Schneising O, et al. (2014) Remote sensing of fugitive methane emissions from oil and 55. US Environmental Protection Agency (2008) Compilation of Air Pollutant Emission gas production in North American tight geologic formations. Earths Futur 2:548–558. Factors, Chapter 3: Stationary Internal Combustion Sources (Research Triangle Park, NC). 35. Lyon DR, et al. (2016) Aerial surveys of elevated hydrocarbon emissions from oil and 56. North Dakota Department of Health, Division of Air Quality (2011) Bakken pool oil and gas gas production sites. Environ Sci Technol 50(9):4877–4886. production facilities: Air pollution control permitting & compliance guidance. Available at 36. Laurenzi IJ, Jersey GR (2013) Life cycle greenhouse gas emissions and freshwater https://www.ndhealth.gov/AQ/Policy/20110502Oil%20%20Gas%20Permitting%20Guidance. consumption of Marcellus shale gas. Environ Sci Technol 47(9):4896–4903. pdf. Accessed October 26, 2016.

Laurenzi et al. PNAS Early Edition | 9of9 Downloaded by guest on October 4, 2021