Petroleum dynamics in the sea and influence of subseadispersant injection during Deepwater HorizonJonas Grosa,b, Scott A. Socolofskyb,1, Anusha L. Dissanayakeb,c, Inok Junb, Lin Zhaod, Michel C. Boufadeld,Christopher M. Reddye, and J. Samuel Areya,f,1

aSchool of Architecture, Civil and Environmental Engineering, École Polytechnique Fédérale de Lausanne, 1015 Lausanne, Switzerland; bZachry Departmentof Civil Engineering, Texas A&M University, College Station, TX 77843; cDepartment of Marine Sciences, University of Georgia, Athens, GA 30602; dCenterfor Natural Resources Development and Protection, Department of Civil and Environmental Engineering, New Jersey Institute of Technology, Newark, NJ07102; eDepartment of Marine Chemistry and Geochemistry, Woods Hole Oceanographic Institution, Woods Hole, MA 02543; and fDepartment ofEnvironmental Chemistry, Swiss Federal Institute of Aquatic Science and Technology, CH-8600 Dübendorf, Switzerland

Edited by David M. Karl, University of Hawaii, Honolulu, HI, and approved May 16, 2017 (received for review July 28, 2016)

During the Deepwater Horizon disaster, a substantial fraction of the600,000–900,000 tons of released petroleum liquid and natural gasbecame entrapped below the sea surface, but the quantity entrappedand the sequestration mechanisms have remained unclear. We mod-eled the buoyant jet of petroleum liquid droplets, gas bubbles, andentrained seawater, using 279 simulated chemical components, for arepresentative day (June 8, 2010) of the period after the sunkenplatform’s riser pipe was pared at the wellhead (June 4–July 15).The model predicts that 27% of the releasedmass of petroleum fluidsdissolved into the sea during ascent from the pared wellhead(1,505 m depth) to the sea surface, thereby matching observed vola-tile organic compound (VOC) emissions to the atmosphere. Based oncombined results from model simulation and water column measure-ments, 24% of released petroleum fluid mass became channeled intoa stable deep-water intrusion at 900- to 1,300-m depth, as aqueouslydissolved compounds (∼23%) and suspended petroleum liquid micro-droplets (∼0.8%). Dispersant injection at the wellhead decreased themedian initial diameters of simulated petroleum liquid droplets andgas bubbles by 3.2-fold and 3.4-fold, respectively, which increaseddissolution of ascending petroleum fluids by 25%. Faster dissolutionincreased the simulated flows of water-soluble compounds into bi-ologically sparse deep water by 55%, while decreasing the flows ofseveral harmful compounds into biologically rich surface water. Dis-persant injection also decreased the simulated emissions of VOCs tothe atmosphere by 28%, including a 2,000-fold decrease in emissionsof benzene, which lowered health risks for response workers.

Deepwater Horizon | oil spill | petroleum | offshore drilling | dispersant

After the blowout at the Deepwater Horizon drilling platformin the Gulf of Mexico on April 20, 2010, between

600,000 and 900,000 tons of petroleum liquid and natural gaserupted continuously from the damaged Macondo wellheadat >1,500 m water depth for 84 d (1, 2). During this deep-searelease, ∼25% of the released mass of petroleum fluids, or150,000–220,000 tons, was interpreted to have dissolved into thewater column (3). For example, >96% of the released C1–C3hydrocarbons (methane, ethane, and propane), benzene, andtoluene did not surface (3). These water-soluble, volatile com-pounds were largely entrapped in a hydrocarbon-rich water massthat formed a stable intrusion at 900- to 1,300-m depth (2, 4, 5)in the local density stratification (6). The composition of thisdeep-water intrusion (2) suggested that the detected hydrocar-bons were introduced primarily by the selective aqueous disso-lution of released petroleum compounds, while residual “fluidparticles” continued a buoyant ascent toward the sea surface.Here, the term fluid particles refers collectively to petroleumliquid droplets, bubbles composed of natural gas or supercriticalfluid, and immiscible two-phase entities containing petroleumliquid and gas.These events have not been documented for any previous oil

spill. Some efforts (7, 8) to model these phenomena neglectedrelevant physical or chemical processes, and more complete

attempts (9, 10) still do not fully couple the evolving states, compo-sitions, and properties of fluid particles under depth-varying pressureand temperature conditions. However, the pressure-dependent,temperature-dependent, and composition-dependent states andproperties of petroleum fluids influence their behaviors in the deepsea (11), and only indirect observations of these phenomena areavailable (2, 4, 12, 13). Additionally, the influence of subsea dis-persant injection was not tested systematically. Corexit EC9500Adispersant was injected initially at the end of the fallen riser pipeand subsequently at the pared wellhead during 67 d of the release(SI Appendix, Fig. S1), but the impact of this unprecedented in-tervention strategy remains debated.Here, we explain the processes that entrapped water-soluble

petroleum compounds and liquid microdroplets in the deep sea,thereby restraining emissions of volatile compounds to the atmo-sphere during the Deepwater Horizon disaster. A mechanistic nu-merical model is developed to simulate near-field processes withina ∼10-km radius of the wellhead (Fig. 1 and SI Appendix, Fig. S2).Model predictions are validated more broadly than any previouslypublished near-field model of the disaster, based on comparisonswith extensive field observations in the sea and atmosphere (Figs.1–3). Model boundary conditions are parameterized from marine,meteorological, and technical measurements recorded on June 8,

Significance

Environmental risks posed by deep-sea petroleum releases aredifficult to predict and assess. We developed a physical model ofthe buoyant jet of petroleum liquid droplets and gas bubblesgushing into the deep sea, coupled with simulated liquid–gasequilibria and aqueous dissolution kinetics of petroleum com-pounds, for the 2010 Deepwater Horizon disaster. Simulationresults are validated by comparisons with extensive observationdata collected in the sea and atmosphere near the release site.Simulations predict that chemical dispersant, injected at thewellhead to mitigate environmental harm, increased the en-trapment of volatile compounds in the deep sea and therebyimproved air quality at the sea surface. Subsea dispersant in-jection thus lowered human health risks and accelerated re-sponse during the intervention.

Author contributions: J.G., S.A.S., C.M.R., and J.S.A. designed research; J.G., S.A.S., A.L.D.,I.J., L.Z., M.C.B., C.M.R., and J.S.A. performed research; J.G., S.A.S., C.M.R., and J.S.A.analyzed data; and J.G., S.A.S., C.M.R., and J.S.A. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Data deposition: TAMOC model code, input files, and output files are publicly availablethrough the GoMRI Information and Data Cooperative at https://data.gulfresearchinitiative.org (ezid.cdlib.org/id/doi:10.7266/N7DF6P8R).1To whom correspondence may be addressed. Email: arey@alum.mit.edu or ssocolofsky@civil.tamu.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1612518114/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1612518114 PNAS | September 19, 2017 | vol. 114 | no. 38 | 10065–10070

ENVIRONMEN

TAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 2,

202

1

2010, a date chosen based on the availability of observations (Figs.2 and 3, and SI Appendix, Fig. S1) after the damaged Macondowellhead was pared above the blowout preventer (June 3) (1).However, no chemical observations in the sea or atmosphere fromthe disaster are used to calibrate the model. The resulting insightsinto petroleum dynamics in the sea allow understanding of the fac-tors governing environmental and human risks for the period afterthe riser pipe was cut until the release was stopped (June 4–July15, 2010).

Results and DiscussionInitial Size Distributions of Petroleum Liquid Droplets and GasBubbles. According to both thermodynamic calculations (11)and observations (1), the Macondo reservoir fluid occurred asseparated petroleum liquid and natural gas phases at the paredwellhead at 1,505-m depth. Because direct observations of fluidparticle sizes at the source are not available, we use a jet-dropletformation model (VDROP-J) to predict the initial sizes of liquiddroplets and gas bubbles (14). VDROP-J predicts initial stabilizedsize distributions by simulating the breakup and coalescence ofdroplets and bubbles above the wellhead. Subsea dispersant in-jection is assumed to affect the initial size distributions of fluidparticles by reducing the interfacial tension (14) (SI Appendix,Table S1).With subsea dispersant injection, VDROP-J predicts a median

diameter by volume (d50) of 1.3 mm for released liquid dropletsand a d50 of 2.6 mm for released gas bubbles (Fig. 4). Thesevalues are similar to the d50 = 1–2 mm predicted by Spauldinget al. (9, 15) for their (bimodally distributed) droplet sizes, alsoassuming dispersant injection. Our thermodynamic model pre-dicts that natural gas bubbles released at 1,505-m depth are∼190 times more dense than natural gas bubbles at the seasurface, but that their interfacial tension with seawater remainshigh (12 mN·m−1 with dispersant injection) such that their sim-ulated sizes (0.3- to 4.7-mm diameter, according to VDROP-J)are similar to near-surface bubbles. Although these individualbubbles contain ∼190 times more mass than a typical near-surfacebubble, they remain nonetheless buoyant in seawater, having initialrise velocities relative to surrounding water (“slip velocities”) of1.7–18 cm·s−1, which are similar to slip velocities of near-surfacebubbles. Droplets of released petroleum liquid, to the contrary, areslightly less dense at 1,505-m depth (708 kg·m−3) than at the sea

surface (∼820 kg·m−3), due to the presence of light C1–C5compounds in the petroleum liquid phase under such elevatedpressures (SI Appendix, Table S1). However, the densities ofboth droplets and bubbles quickly increase during their earlyascent in the water column, due to rapid dissolution of lightcompounds.To elucidate the coupled effects of transport, dissolution, and

phase transitions for the initially released droplets and bubblespredicted by VRDOP-J, we developed the Texas A&M Oilspill(Outfall) Calculator (16) (TAMOC) (Fig. 1 and SI Appendix, Fig.S2). This modeling suite includes modules to predict the coupled jetand plume dynamics of the buoyant petroleum liquid, gas, andentrained water in the deep sea, and it also simulates the dynamicrise and dissolution of individual droplets and bubbles in the oceanwater column. Petroleum fluid properties and liquid–gas partition-ing are assumed to depend on pressure, temperature, and fluidcompositions according to the Peng–Robinson equation of statewith volume translation and other empirical expressions, based ona model composition of 279 components (11) (SI Appendix, SIText, S-1.7). The Peng–Robinson equation of state with volume

Fig. 1. Conceptual representation of the dynamics of petroleum fluids and affected water masses during the Deepwater Horizon disaster, after the riser pipe wascut (June 4–July 15). Petroleum liquid and gas fluids gush out of the pared wellhead. The resulting vertical buoyant jet of petroleum fluid particles entrains ambientwater, generating a buoyant plume (6). At the predicted maximum height of rise (1,070-m depth for the plume centerline), water in the buoyant plume begins todescend to form an intrusion layer. At the trap height (1,100-m depth), the density of the modeled descending plume matches that of the surrounding ambientseawater, producing a deep-water hydrocarbon-rich intrusion that spans 900- to 1,300-m depth and contains water, dissolved petroleum compounds, and liquiddroplets. Within 10 km downstream from the pared wellhead, most of the mass flow rate of fluid particles, having diameters of ≥140 μm, are predicted to separatefrom the buoyant plume, the descending plume, or from the deep-water intrusion, after which they continue ascent toward the sea surface according to theirindividual slip velocities, while also advected horizontally by ambient currents. Model predictions are validated with chemical and physical observations in theatmosphere and in the deep sea, as explained in the text and in Fig. 3. Airplane graphic courtesy of Pixabay/Clker-Free-Vector-Images.

Fig. 2. Time of arrival of petroleum fluid particle mass at the sea surfaceaccording to predictions by TAMOC with dispersant injection at the paredwellhead (green solid line), predictions of TAMOC without dispersant injectionat the pared wellhead (blue solid line) (SI Appendix, SI Text, S-1.14), and fieldobservation before the riser pipe was cut (vertical pink line). The reportedlyobserved value (3) is the time (∼3 h) to observe a visual reduction in the freshsurface slick at the sea surface near the wellhead during dispersant applicationtests before the riser was cut. We interpret the reported field observation as thetime of arrival of the majority of fluid particle mass at the sea surface withoutsubsea dispersant injection.

10066 | www.pnas.org/cgi/doi/10.1073/pnas.1612518114 Gros et al.

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 2,

202

1

translation is used widely in the petroleum industry to predictthe equilibrium gas–liquid distributions and densities of reser-voir petroleum fluids under varying environmental conditions.Biodegradation processes are assumed negligible in the modelednear-field region of the water column (SI Appendix, SI Text, S-1.17).

Controls on Aqueous Dissolution of Petroleum Fluids in theWater Column.According to these comprehensive model simulations, 27% of themass flow rate of released petroleum fluids became dissolved intothe water column and did not arrive at the sea surface on June 8,assuming subsea dispersant injection (SI Appendix, Table S2). Dis-solved petroleum compounds were channeled predominantly (23%)into the deep-water intrusion (900- to 1,300-m depth), with a smallerpercentage (4%) dissolving in the higher water column (0- to 900-mdepth) (Figs. 1 and 5). These model predictions corroborate a pre-vious estimate of ∼25% of the mass flow rate of released petroleumfluids becoming aqueously dissolved in the water column, based onatmospheric observations (3).What mechanisms entrapped petroleum compounds in deep

water of the Gulf of Mexico after the riser pipe was cut on June3? Our TAMOC simulations indicate that water-soluble com-pounds became partly or wholly dissolved in entrained seawaterduring ascent of a buoyant plume that rose to a maximum cen-terline height of 1,070-m depth before collapsing to form aneutrally buoyant hydrocarbon-rich intrusion that spanneddepths of 900–1,300 m (Figs. 1 and 3). The behavior of the ini-tially ascending plume and the depth of neutral buoyancy for thehydrocarbon-rich intrusion are controlled by the petroleum fluid

release rate, petroleum fluid properties, conditions at the releaselocation, and the ambient stratification and prevailing currentprofile (6).For water-soluble, volatile compounds, which included the C1–C4

hydrocarbons and benzene, toluene, ethylbenzene, and xylenes(BTEX), rapid dissolution was facilitated by the high surface-to-volume ratios of small droplets and bubbles formed above the well-head (Fig. 4). The high pressures (>10 bar) encountered at >100-mdepths enhance the dissolution of C1–C3 hydrocarbons relative tomore shallow depths. For example, thermodynamic calculations in-dicate that C1, C2, and C3 in natural gas bubbles became 140, 33, and11 times more water soluble at the release depth (1,505 m) comparedwith conditions at the sea surface (11). After the rapid removal ofthese C1–C3 compounds by aqueous dissolution, gas bubbles thatformed near the wellhead are predicted to condense into 100% liquiddroplets within their initial 700 m of ascent, illustrating a counterin-tuitive and surprising fluid phase transition (SI Appendix, Fig. S3 andSI Text, S-1.6).The dissolution of water-soluble compounds was further fa-

cilitated by the prolonged immersion of petroleum fluids in thesea before their arrival at the sea surface. For June 8, TAMOCpredicts ascent times of 4.6–10 h for the middle 80% of thesurfacing mass, with small droplets (≤900 μm at the wellhead)taking longer times; 50% of the mass flow rate of residual fluidscontributing to surface slicks arrived at the sea surface within 6 hafter their release at 1,505-m depth, assuming subsea dispersantinjection (Fig. 2). The model predicts that 98% of the mass flowrate contributing to the surface slicks arrived at the sea surface

Fig. 3. Model predictions and reported observationsof chemical composition in the deep-water intrusionand at the sea surface during the Deepwater Horizondisaster, after the riser pipe was cut (June 4–July 15).(A) Abundances of the readily soluble compounds, C2–C3and DOSS, in the deep-water hydrocarbon-rich in-trusion, expressed as the fraction of mass present in theintrusion relative to that released from the wellhead,normalized to C1 (Fi,methane, SI Appendix, SI Text, S-1.9and Eq. S8). Reported field observations of Fi,methane (2,5, 17) (green) agree reasonably with predictions of ourmodel (blue). (B) Abundances of semisoluble andsparingly soluble compounds in the deep-water in-trusion, expressed as the fraction of mass present in theintrusion relative to that released from the wellhead,normalized to benzene (Fi,benzene, SI Appendix, SI Text,S-1.9 and Eqs. S9 and S10). Estimated Fi,benzene valuesfrom field observations (green) exhibit reasonableagreement with our model predictions (blue) for12 semisoluble compounds (toluene to C2-naphtha-lenes, having hypothetical subcooled liquid aqueoussolubilities of 10−4.5 to 10−2 mol·L−1 at deep-sea condi-tions), assuming dispersant injection. Adjusting ourmodel predictions for the presence of liquid micro-droplets (yellow) improves the agreement with fieldobservations for 16 sparingly soluble compounds(phenanthrene to pristane, having hypothetical sub-cooled liquid aqueous solubilities <10−4.5 mol·L−1). Ob-served Fi,benzene values (green) were determined here,based on 4,205 quantified concentration data for29 individual petroleum compounds reportedly mea-sured (40) during June 4–July 15 at 374 sampling loca-tions at 900- to 1,300-m depth, situated at 1- to 10-kmdistance from the wellhead (SI Appendix, SI Text, S-1.9).Error bounds (displayed as whiskers) designate the 95%probability intervals of observed values. (C) Abun-dances of selected volatile organic compounds (VOCs)at the sea surface, assuming dispersant injection,expressed as the fraction of mass arriving at the sea surface relative to that released from the wellhead (SI Appendix, SI Text, Eq. S23). Reported atmospheric ob-servations during May 20–23 and June 8, 10, and 15–27 (3) (green) exhibit approximate agreement with predictions by TAMOC (blue). (D) TAMOC predictions of theabundances of selected VOCs at the sea surface in the absence of dispersant injection, expressed as the fraction of mass arriving at the sea surface relative to thatreleased from the wellhead (SI Appendix, SI Text, Eq. S23). No field observations are available for comparison.

Gros et al. PNAS | September 19, 2017 | vol. 114 | no. 38 | 10067

ENVIRONMEN

TAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 2,

202

1

within a narrow 1.0-km2 oval area that spanned a horizontaldistance of 0.1–2.1 km from the release location (SI Appendix, SIText, S-1.14). These predictions agree with reported observationsthat “essentially all” of the residual petroleum surfaced within a≤2-km2 area, offset laterally by 0.5–1.5 km from the wellhead,based on atmospheric measurements of hydrocarbons on June 8–10(3). The predicted dimensions of the surfacing area are controlledby (i) the variation in horizontal transport distances of variable-size fluid particles having varying ascent speeds in the presence ofambient sea currents and (ii) turbulent horizontal dispersion ofascending fluid particles (SI Appendix, SI Text, S-1.14).

Composition of the Deep-Sea Intrusion: Dissolved Hydrocarbons,Petroleum Liquid Microdroplets, and Dioctyl Sodium Sulfosuccinate.Aqueous dissolution processes rapidly removed readily solublecompounds from the released petroleum liquid and gas and in-troduced these compounds into the deep-water intrusion at 900- to1,300-m depth. For June 8, our model simulations predict that>99%,98%, and 92% of the released C1, C2, and benzene, respectively,became entrapped in the deep-water intrusion at 900- to 1,300-mdepth (SI Appendix, Table S3). We infer that 80% of the injecteddioctyl sodium sulfosuccinate (DOSS), which is a component of thedispersant (Corexit EC9500A) injected at the wellhead, also becameretained in the intrusion (SI Appendix, SI Text, S-1.16). Reportedwater column observations collected in June 2010 confirm that thesecompounds were predominantly retained in the deep sea (2, 5, 17).By comparison, aqueous dissolution led to the partial transfers of

semisoluble petroleum compounds into the intrusion. The “finger-print” of aqueous dissolution can be discerned from the observedabundances of individual petroleum compounds in the intrusion,relative to the abundance of a readily soluble reference component(here, C1 or benzene), divided by the same mass ratio in the reser-voir fluid (SI Appendix, SI Text, Eqs. S8–S10). The result is a “frac-tionation index” that can range in value from 0 (no dissolution) to 1(near-complete dissolution) for each compound. Our model predictsfractionation index values that agree with observations reported at900- to 1,300-m depth within a 1- to 10-km radius of the wellhead(June 4–July 15, 2010) for three readily soluble compounds (C2, C3,and DOSS) and for 12 semisoluble compounds (small hydrocarbonsranging from toluene to C2-naphthalenes; Fig. 3 A and B and SIAppendix, Table S4 and SI Text, S-1.9). We further conclude that allreadily soluble and semisoluble compounds (up to C2-naphthalenes)are predominantly in the aqueously dissolved state in the deep-waterintrusion, based on a detailed equilibrium partitioning analysis ofwater column sample data and also based on simulated dissolutionkinetics for suspended ≤130-μm liquid microdroplets entrapped inthe intrusion (SI Appendix, Table S5 and SI Text, S-1.10). Thesevalidations of the model with field data support our apportionment

of dissolved compounds to the deep-water intrusion for the periodafter the riser pipe was cut (Fig. 5).However, the model systematically underpredicts the slight

but unmistakable presence of 16 sparingly soluble compoundsthat exhibited observed fractionation index values of 0.007–0.021 rel-ative to benzene (phenanthrene through pristane, Fig. 3B), based onavailable water column data from the deep-water intrusion after theriser pipe was cut (June 4–July 15). We interpret this discrepancy as acontribution from stable liquid microdroplets that is underestimatedby our modeling approach. Such microdroplets may have formedthrough tip streaming in the turbulent conditions created by the jet ofpetroleum fluids at the wellhead, further facilitated by dispersantinjection (18). This interpretation is consistent with reported in situholographic camera observations of oil droplets having 30–400-μmdiameter within the deep-water intrusion at distances of 1–9 km fromthe wellhead during June 14–20 (19). According to our simulations, amicrodroplet with an initial diameter of ≤130 μm would experiencesufficiently slow buoyant ascent to remain entrapped in the intrusion(>900-m depth) within a 1- to 10-km range of the wellhead. Based onthe discrepancy between modeled and observed fractionation ofpetroleum compounds in the deep-water intrusion, we estimate that1.2% (or 0.6–2.0%) of the mass flow rate of sparingly soluble andinsoluble material from the wellhead led to microdroplets observed inthe intrusion after the riser was cut (equivalent to 0.8%, or 0.4–1.3%,of the total mass of released petroleum fluids) (Figs. 3B and 5, and SIAppendix, SI Text, S-1.9 and S-1.11). This estimate differs from aprevious estimate of 5–25% of the released insoluble material be-coming entrapped as a liquid phase in the intrusion, which was basedon fractionation data for long-chain alkanes collected in the deepwater column before the riser pipe was cut (May 27–June 2) (3).Based on our results from simulations and water column datacollected after the riser was cut, the composition of the deep-water intrusion is ∼97% aqueously dissolved compounds and

Fig. 4. Binned size distributions of liquid droplets and gas bubbles predictedby the VDROP-J model (14) after 200 m of ascent without dissolution, for June8. (A) Liquid droplets without dispersant; (B) liquid droplets with dispersant;(C) gas bubbles without dispersant; and (D) gas bubbles with dispersant.

Fig. 5. Predicted apportionments at the near-field boundary (∼10 km dis-tance from the wellhead), for petroleum fluids released during the DeepwaterHorizon disaster after the riser pipe was cut (June 4–July 15, 2010). Values arepresented for model simulations (A) with subsea dispersant injection and (B)without subsea dispersant injection, for June 8. *The apportionment to theupper water column does not account for dissolution of petroleum compoundsfrom oil slicks at the sea surface. **The apportionment to liquid petroleummicrodroplets in the deep-water intrusion is based on observations and modelpredictions obtained with dispersant injection; we assumed the same value forthe case without dispersant injection.

10068 | www.pnas.org/cgi/doi/10.1073/pnas.1612518114 Gros et al.

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 2,

202

1

∼3% petroleum liquid microdroplets; again, this result differsfrom the previous estimate of 69% dissolved, 31% liquid, basedon data collected before the riser cut (3). These discrepanciessuggest that paring the riser pipe decreased the flow of liquidpetroleum compounds into the deep-sea intrusion. The outcomesfrom this technical intervention are relevant for interpreting bio-availability (10), ecological impact (20, 21), and biodegradation (5,22–25) of petroleum compounds, all of which depend upon thedroplet sizes, liquid/dissolved distribution, and altered chemicalcomposition of petroleum residues in the deep sea.

Implications for Deposition of Petroleum Residues on the DeepSeafloor. Recent studies suggest that the deep-sea intrusionmay have contributed to the extensive deposition of weatheredpetroleum residues observed on the seafloor within a 3,200-km2

area near the wellhead (24, 26, 27). Although deposition pro-cesses were not included in our simulations, our results confirmthe unambiguous presence of ≤130-μm liquid microdroplets inthe deep-sea intrusion, which would have contributed to the“bathtub ring” previously observed in sediments where the in-trusion impinged on the continental slope at 900- to 1,300-mdepth (26). Our simulations further predict rapid dissolution ofsemisoluble compounds (up to C2-naphthalenes) from suspended≤130-μm microdroplets in the intrusion (SI Appendix, TableS5 and SI Text, S-1.10), which would explain the finding (24, 27)that petroleum residues deposited on the seafloor were severelydepleted of two-ring and three-ring polycyclic aromatic hydrocarbons(PAHs).In addition, previous reports have estimated that 1.9–14.8%

(24, 26) or 0.5–9.1% (28) of the released insoluble oil mass wasdeposited on the seafloor largely within 40 km of the wellhead,based on the observed concentrations of recalcitrant compoundssuch as hopane and n-C38 in seafloor sediments (24, 26). Incontrast, our results reveal that only 0.6–2.0% of the insolublematerial released from the wellhead was channeled into thedeep-sea intrusion, after the riser pipe was pared at the wellhead.Assuming that seafloor residues originated primarily from thedeep-sea intrusion (24, 26), this discrepancy is consistent with thelikely possibility (discussed above) that the mass flow of liquidpetroleum droplets into the deep-sea intrusion was higher beforethe riser pipe was pared on June 3. Alternatively, the seafloorresidues may originate from other sources that supplementedcontributions from the intrusion. Our simulated dissolution ki-netics predict that 10- to 300-μm microdroplets do not becomenegatively buoyant during transport up to 30 km downstreamfrom the wellhead. Therefore, additional processes such as bio-degradation (22, 23), marine oil snow sedimentation, and floccu-lent accumulation (MOSSFA) (29, 30) associated with naturalparticles from riverine inputs and from marine algae, or in situburning (31) are needed to explain the buoyancy loss that wouldlead to fallout deposition. TAMOC simulations of near-field pro-cesses can provide initial conditions for other models of thesephenomena and other processes occurring at the sea surface (32).

Subsea Dispersant Injection Restrains the Arrival of Volatile OrganicCompounds at the Sea Surface. Dispersant injection at the wellheadcan partly influence hazardous exposures of response workers toharmful volatile organic compounds (VOCs) at the sea surface, butthe benefits of this intervention tool remain debated. In modelsimulations, subsea dispersant injection lowered the interfacialtensions of the liquid–seawater and gas–seawater interfaces. Thisproduced smaller fluid particles near the pared wellhead, de-creasing d50 by factors of 3.2 and 3.4 for droplets and bubbles, re-spectively, according to VDROP-J simulations (Fig. 4). Dispersantinjection thus increased the surface-to-volume ratios of dropletsand bubbles by approximately three times, which accelerated dis-solution kinetics. With dispersant injection, ascending fluids arepredicted to arrive at the sea surface within 4.6–10 h (for the middle

80% of the surfacing mass), compared with predicted ascent timesof only 2–4 h without dispersant (Fig. 2). This prediction agreeswith available field observations before the riser pipe was cut(3), which suggest that petroleum liquid surfaced ∼3 h afterrelease without dispersant injection. Dispersant injection there-fore lengthened the available time for dissolution processes to actand also increased the rates of dissolution, both of which in-creased the overall extent of dissolution of petroleum fluids.Due to these mechanisms, the 49 tons of dispersant injected at

the wellhead on June 8 increased the aqueous dissolution ofpetroleum compounds by 25% or 430 tons, based on compari-sons of model simulations with and without dispersant injection(Fig. 5). As a consequence, dispersant injection increased theflow rates of water-soluble compounds into the biologicallysparse deep sea (Figs. 1 and 5 and SI Appendix, Table S3). Ac-cordingly, dispersants lowered the flow rates of these harmfulcompounds into biologically rich surface waters, where pop-ulation densities of sensitive early life-stage biota are muchhigher than in deep water (20).Readily soluble and semisoluble hydrocarbons also represent many

of the VOCs in the reservoir fluid, where VOCs are approximatedhere as petroleum compounds in the C1–C9 boiling point range. Oursimulated atmospheric emissions of VOCs are consistent withreported atmospheric measurements (3) taken duringMay 20–23 andJune 8, 10, and 15–27 above the disaster site, during which dispersantinjection was applied (Fig. 3C). This validation of simulations resultswith field data supports our apportionments of total petroleumcompounds to the sea surface and the water column during the pe-riod after the riser pipe was cut (Fig. 5).Based on model simulations, dispersant injection accelerated

deep-sea dissolution processes that affected many volatile com-pounds, thereby decreasing the mass flow rate of VOCs to theatmosphere by ∼28% (1,000 instead of 1,400 tons of C1–C9 com-pounds emitted on June 8). Dispersant injection also increased thepredicted area where residual fluids reached the sea surface from0.2 to 1.0 km2. This increase in surface area occurred because dis-persant injection decreased the sizes of the ascending droplets, whichled to a broader distribution of ascent speeds (SI Appendix, SI Text,S-1.14), and which displaced the center of mass of petroleum ar-riving at the sea surface from 230-m to 490-m horizontal distancefrom the wellhead. Due to these combined effects of dispersant in-jection, potential inhalation exposures to VOCs were dramaticallydecreased for response personnel at the sea surface (Figs. 3 C and Dand 5). Based on meteorological conditions on June 8, dispersantinjection lowered the average total concentration of VOCs from anestimated ∼400 ppm (without dispersant) to ∼70 ppm (with dis-persant) at 2-m altitude above sea level inside the respective areaswhere 98% of the residual fluids were predicted to arrive at the seasurface. BTEX emissions to the atmosphere were disproportionatelyaffected, due to their high aqueous solubilities (Fig. 3 C and D andSI Appendix, Table S3). For example, dispersant injection decreasedthe predicted mass flow rate of benzene to the atmosphere by2,000 times, lowering the predicted average benzene concentrationsin the atmosphere from ∼1,300 ppb (without dispersant) to ∼0.2 ppb(with dispersant) at 2-m altitude above the surfacing area, duringconditions on June 8 (SI Appendix, SI Text, S-1.15). Dispersant in-jection also led to smaller petroleum droplets reaching the sea sur-face (SI Appendix, Fig. S4).These model predictions corroborate the observation that

VOC concentrations in the atmosphere were decreased at thesea surface above the wellhead during subsea dispersant in-jection and during sea-surface application of dispersant (33).Our predicted atmospheric concentrations for June 8 depend onwind conditions that can vary daily. Nonetheless, our simulationresults rationalize the effectiveness of subsea dispersant injectionas a response strategy for reduced VOC exposure by responsepersonnel, reduced fire hazard, and reduced disruptions to re-sponse work (33). For example, work was stopped on “the small

Gros et al. PNAS | September 19, 2017 | vol. 114 | no. 38 | 10069

ENVIRONMEN

TAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 2,

202

1

city of vessels” tethered to the wellhead during periods whenmeasured atmospheric concentrations of benzene exceeded 5 ppm(33). Therefore, subsea dispersant injection and dispersant ap-plication at the petroleum surfacing location likely favored a moreprompt final containment of the uncontrolled release of fluidsfrom the wellhead.

MethodsVDROP-J. The initial size distributions of droplets and bubbles predicted byVDROP-J are dependent on the size of the wellhead orifice through whichpetroleum fluids were released, and the flow rates, densities, buoyancies,viscosities, and liquid/water and gas/water interfacial tensions of petroleumfluids under local conditions at the wellhead (14) (SI Appendix, Table S1 andSI Text, S-1.1). The initial size distributions predicted by VDROP-J were usedat the petroleum fluids source boundary of TAMOC.

TAMOC. TAMOC includes a bent plumemodel that describes both the buoyantrising plume and the descending plume that forms the intrusion (Fig. 1),including the entrainment of ambient seawater (16, 34). Detailed thermo-dynamic equations of state are applied to simulate oil, gas, and fluid prop-erties under low temperature and extreme pressure (11). Particle fate andtransport follows an approach similar to McGinnis et al. (35), and jet andplume behavior follow an integral model approach (36–38). Our model ex-tends predicted trap heights based on estimates using simplified assumptions

(6), which was an update to the scaling of Asaeda and Imberger (39). Themodel is unique in its ability to combine the detailed thermodynamic be-havior (needed to capture phase equilibria, dissolution, and buoyancy) withthe dynamic equations of motion (plume and intrusion formation). The modelhas been validated to numerous laboratory and field experiments of bubbleplumes (e.g., ref. 38). No model parameters were tuned to improve agree-ment with chemical observations collected in the sea or atmosphere duringthe Deepwater Horizon disaster.

Volatile Organic Compound Concentrations in the Atmosphere. Atmosphericconcentrations of VOCs above the surfacing area were estimated using asuperposition of Gaussian plume models, assuming immediate evaporationof all compounds in the C1–C9 boiling point range from residual fluids as theyreached the sea surface (SI Appendix, SI Text, S-1.12 and S-1.15).

ACKNOWLEDGMENTS. We thank the late Dr. John Hayes for his advice andencouragement and Thomas B. Ryerson, Richard Camilli, David L. Valentine,Deborah French-McCay, Tim Nedwed, and Chuan-Yuan Hsu for discussions.Research was made possible, in part, by grants from the Gulf of MexicoResearch Initiative (GoMRI) to the Center for the Integrated Modeling andAnalysis of the Gulf Ecosystem (C-IMAGE), Dispersion Research on Oil: Physics &Plankton Studies (DROPPS II), Deep Sea to Coast Connectivity in the EasternGulf of Mexico (DEEP-C), and GoMRI request for proposal V (RPV-V). This workis also supported by GoMRI Grant SA 16-30 and National Science FoundationGrants OCE-0960841, RAPID OCE-1043976, CBET-1034112, and EAR-0950600.

1. McNutt MK, et al. (2012) Review of flow rate estimates of the Deepwater Horizon oilspill. Proc Natl Acad Sci USA 109:20260–20267.

2. Reddy CM, et al. (2012) Composition and fate of gas and oil released to the watercolumn during the Deepwater Horizon oil spill. Proc Natl Acad Sci USA 109:20229–20234.

3. Ryerson TB, et al. (2012) Chemical data quantify Deepwater Horizon hydrocarbonflow rate and environmental distribution. Proc Natl Acad Sci USA 109:20246–20253.

4. Camilli R, et al. (2010) Tracking hydrocarbon plume transport and biodegradation atDeepwater Horizon. Science 330:201–204.

5. Valentine DL, et al. (2010) Propane respiration jump-starts microbial response to adeep oil spill. Science 330:208–211.

6. Socolofsky SA, Adams EE, Sherwood CR (2011) Formation dynamics of subsurfacehydrocarbon intrusions following the Deepwater Horizon blowout. Geophys Res Lett38:L09602.

7. Testa JM, Adams EE, North EW, He R (2016) Modeling the influence of deep waterapplication of dispersants on the surface expression of oil: A sensitivity study.J Geophys Res Oceans 121:5995–6008.

8. Lindo-Atichati D, et al. (2016) Simulating the effects of droplet size, high-pressurebiodegradation, and variable flow rate on the subsea evolution of deep plumes fromthe Macondo blowout. Deep Sea Res Part II Top Stud Oceanogr 129:301–310.

9. Spaulding M, Mendelsohn D, Crowley D, Li Z, Bird A (2015) DRAFT technical reportsfor Deepwater Horizon water column injury assessment WC_TR.13: Application ofOILMAP DEEP to the Deepwater Horizon blowout (RPS ASA, South Kingstown, RI).

10. French McCay D, et al. (2015) Technical reports for Deepwater Horizon water columninjury assessment WC_TR.14: Modeling oil fate and exposure concentrations in thedeepwater plume and rising oil resulting from the Deepwater Horizon oil spill (RPSASA, South Kingstown, RI).

11. Gros J, Reddy CM, Nelson RK, Socolofsky SA, Arey JS (2016) Simulating gas−liquid−waterpartitioning and fluid properties of petroleum under pressure: Implications for deep-seablowouts. Environ Sci Technol 50:7397–7408.

12. Camilli R, et al. (2012) Acoustic measurement of the Deepwater Horizon Macondowell flow rate. Proc Natl Acad Sci USA 109:20235–20239.

13. Spier C, Stringfellow WT, Hazen TC, Conrad M (2013) Distribution of hydrocarbonsreleased during the 2010 MC252 oil spill in deep offshore waters. Environ Pollut 173:224–230.

14. Zhao L, et al. (2014) Evolution of droplets in subsea oil and gas blowouts: Develop-ment and validation of the numerical model VDROP-J. Mar Pollut Bull 83:58–69.

15. Li Z, Spaulding M, French McCay D, Crowley D, Payne JR (2017) Development of aunified oil droplet size distribution model with application to surface breaking wavesand subsea blowout releases considering dispersant effects. Mar Pollut Bull 114:247–257.

16. Socolofsky SA, et al. (2015) Texas A&M Oilspill Calculator (TAMOC) modeling suite forsubsea spills. Proceedings of the Thirty-Eighth AMOP Technical Seminar (EnvironmentCanada, Ottawa), pp 153–168.

17. Kujawinski EB, et al. (2011) Fate of dispersants associated with the Deepwater Hori-zon oil spill. Environ Sci Technol 45:1298–1306.

18. Gopalan B, Katz J (2010) Turbulent shearing of crude oil mixed with dispersantsgenerates long microthreads and microdroplets. Phys Rev Lett 104:054501.

19. Li Z, et al. (2015) Technical reports for Deepwater Horizon water column injuryassessment: Oil particle data from the Deepwater Horizon oil spill (RPS ASA, SouthKingstown, RI).

20. French McCay D, Rowe J, Balouskus R, Morandi A, Conor McManus M (2015) Technicalreports for Deepwater Horizon water column injury assessment WC_TR.28: Injuryquantification for planktonic fish and invertebrates in estuarine, shelf and offshorewaters (RPS ASA, South Kingstown, RI).

21. White HK, et al. (2012) Impact of the Deepwater Horizon oil spill on a deep-watercoral community in the Gulf of Mexico. Proc Natl Acad Sci USA 109:20303–20308.

22. Valentine DL, et al. (2012) Dynamic autoinoculation and the microbial ecology of adeep water hydrocarbon irruption. Proc Natl Acad Sci USA 109:20286–20291.

23. Brakstad OG, Nordtug T, Throne-Holst M (2015) Biodegradation of dispersedMacondo oilin seawater at low temperature and different oil droplet sizes. Mar Pollut Bull 93:144–152.

24. Bagby SC, Reddy CM, Aeppli C, Fisher GB, Valentine DL (2017) Persistence and bio-degradation of oil at the ocean floor following Deepwater Horizon. Proc Natl Acad SciUSA 114:E9–E18.

25. Hazen TC, Prince RC, Mahmoudi N (2016) Marine oil biodegradation. Environ SciTechnol 50:2121–2129.

26. Valentine DL, et al. (2014) Fallout plume of submerged oil from Deepwater Horizon.Proc Natl Acad Sci USA 111:15906–15911.

27. Stout SA, Payne JR (2016) Macondo oil in deep-sea sediments. Part 1: Sub-seaweathering of oil deposited on the seafloor. Mar Pollut Bull 111:365–380.

28. Chanton J, et al. (2015) Using natural abundance radiocarbon to trace the flux ofpetrocarbon to the seafloor following the Deepwater Horizon oil spill. Environ SciTechnol 49:847–854.

29. Daly KL, Passow U, Chanton J, Hollander D (2016) Assessing the impacts of oil-associatedmarine snow formation and sedimentation during and after the Deepwater Horizon oilspill. Anthropocene 13:18–33.

30. Passow U (2016) Formation of rapidly-sinking, oil-associated marine snow. Deep SeaRes Part II Top Stud Oceanogr 129:232–240.

31. Stout SA, Payne JR (2016) Chemical composition of floating and sunken in-situ burnresidues from the Deepwater Horizon oil spill. Mar Pollut Bull 108:186–202.

32. Beegle-Krause CJ (2001) General NOAA oil modeling environment (GNOME): A newspill trajectory model. IOSC 2001 Proceedings (Tampa, FL), pp 865–871.

33. US Coast Guard (2011) On scene coordinator report, Deepwater Horizon oil spill (USCoast Guard, Washington, DC).

34. Socolofsky SA (2015) TAMOC (College Station) Available at: https://github.com/socolofs/tamoc. Accessed August 3, 2017.

35. McGinnis DF, Greinert J, Artemov Y, Beaubien SE, Wüest A (2006) Fate of risingmethane bubbles in stratified waters: How much methane reaches the atmosphere?J Geophys Res Oceans 111:C09007.

36. Johansen Ø (2003) Development and verification of deep-water blowout models.MarPollut Bull 47:360–368.

37. Zheng L, Yapa PD, Chen F (2003) Amodel for simulating deepwater oil and gas blowouts.Part I: Theory and model formulation. J Hydraul Res 41:339–351.

38. Socolofsky SA, Bhaumik T, Seol D-G (2008) Double-plume integral models for near-field mixing in multiphase plumes. J Hydraul Eng 134:772–783.

39. Asaeda T, Imberger J (1993) Structure of bubble plumes in linearly stratified envi-ronments. J Fluid Mech 249:35–57.

40. Gulf BP Science Data (2016) Chemistry data associated with water column samplescollected in the Gulf of Mexico from May 2010 through July 2012. Available at: https://data.gulfresearchinitiative.org/data/BP.x750.000:0016#distributionInfo. Accessed August3, 2017.

10070 | www.pnas.org/cgi/doi/10.1073/pnas.1612518114 Gros et al.

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 2,

202

1