Modeling Dispersion from Toxic Gas Released after a Train ... · Modeling Dispersion from Toxic Gas...

Modeling Dispersion from Toxic Gas Released after a TrainCollision in Graniteville, SC

Robert L. Buckley, Charles H. Hunter, Robert P. Addis, and Matthew J. ParkerSavannah River National Laboratory, Aiken, SC

ABSTRACTThe Savannah River National Laboratory (SRNL) WeatherInformation and Display System was used to provide me-teorological and atmospheric modeling/consequence as-sessment support to state and local agencies after thecollision of two Norfolk Southern freight trains on themorning of January 6, 2005. This collision resulted in therelease of several toxic chemicals to the environment,including chlorine. The dense and highly toxic cloud ofchlorine gas that formed in the vicinity of the accidentwas responsible for 9 fatalities and caused injuries to morethan 500 others. Transport model results depicting theforecast path of the ongoing release were made availableto emergency managers in the county’s Unified Com-mand Center shortly after SRNL received a request forassistance. Support continued over the ensuing 2 days ofthe active response. The SRNL also provided weatherbriefings and transport/consequence assessment modelresults to responders from the South Carolina Depart-ment of Health and Environmental Control, the Savan-nah River Site (SRS) Emergency Operations Center, De-partment of Energy headquarters, and hazard materialteams dispatched from the SRS. Operational model-gen-erated forecast winds used in consequence assessmentsconducted during the incident were provided at 2-kmhorizontal grid spacing during the accident response.High-resolution Regional Atmospheric Modeling System(RAMS, version 4.3.0) simulation was later performed toexamine potential influences of local topography onplume migration in greater detail. The detailed RAMSsimulation was used to determine meteorology usingmultiple grids with an innermost grid spacing of 125 m.

Results from the two simulations are shown to generallyagree with meteorological observations at the time; con-sequently, local topography did not significantly affectwind in the area. Use of a dense gas dispersion model tosimulate localized plume behavior using the higher-reso-lution winds indicated agreement with fatalities in theimmediate area and visible damage to vegetation.

INTRODUCTIONOn January 6, 2005, at �2:40 a.m., a Norfolk Southernfreight train traveling northbound through the town ofGraniteville, SC, collided with a second freight trainparked on an industrial rail spur, resulting in the ruptureof rail cars transporting liquefied chlorine and other in-dustrial chemicals. The subsequent rapid discharge of �70t of chlorine quickly produced a dense airborne cloud oftoxic gas and aerosols that spread throughout propertyoccupied by a textile mill and into adjacent areas of town,resulting in the deaths of 9 individuals, mainly mill work-ers, and injuries to �500 others.1 Graniteville is located inwest-central South Carolina a few kilometers west ofAiken, SC, and 20 km north of the Department of Ener-gy’s Savannah River Site (SRS).

The threat of a rupture to additional tankers of chlo-rine that were damaged in the accident resulted in closureof businesses and the relocation of �5000 residents for �9days, as crews worked to dispose of the remaining inven-tories. Hundreds of emergency workers from local volun-teer fire departments, Aiken County sheriff and emer-gency management offices, the SRS, and state and federalagencies responded to the event. As part of a mutual aidagreement with Aiken County, the Atmospheric Technol-ogies Group (ATG) of the Savannah River National Labo-ratory (SRNL) at SRS supported decision-makers in thecounty’s Joint Operation Center with hazard conse-quence modeling and meteorological assessments overthe 2-day period of active response.

The SRS is an 800-km2 (310-mi2) nuclear facilityowned by the U.S. Department of Energy and operated bythe Washington Savannah River Company. The SRS islocated in Western South Carolina, �20 km south ofAiken, bordered on the west by the Savannah River (Fig-ure 1). Established in 1950, the SRS produced radionu-clides in support of the nation’s defense and currentlyoperates programs supporting nuclear nonproliferation,disposition of radiological and chemical waste, and envi-ronmental remediation.

The meteorological program developed and oper-ated by SRNL supports SRS operations in the areas of

IMPLICATIONSResponse to emergency events of national significance,such as the train collision described in this paper, requiresa collaborative effort from numerous agencies at all levelsof government. Dispersion modeling of chemical releasesto assess potential downwind hazards is a crucial elementin the implementation of appropriate protective actions.Under often stressful conditions, modeling guidance needsto be easily comprehended and given to decision-makersat the local level in a timely fashion. Experience gainedthrough the response to the Graniteville, SC, train accidentshow that regional modeling and consequence assessmentassets can play an essential role in achieving this goal,providing comprehensive technical support tailored specif-ically for the local decision-maker.

TECHNICAL PAPER ISSN 1047-3289 J. Air & Waste Manage. Assoc. 57:268–278

Copyright 2007 Air & Waste Management Association

268 Journal of the Air & Waste Management Association Volume 57 March 2007

emergency response, compliance with regulatory require-ments for employee and public health and safety, and thedesign and licensing of facilities supporting current andfuture SRS missions. The SRNL also conducts work in theatmospheric sciences for other federal government agen-cies. Components of the meteorological program at SRNLinclude a comprehensive tower-based atmospheric mea-surement program,2 or mesonet, a fully equipped weatherforecast center, which provides access to local and world-wide weather data, and advanced atmospheric modelingcapabilities that provide operational numerical weatherforecasts and atmospheric consequence assessments foremergency response and nonproliferation.

The focus of this paper is on the technical aspects ofhazard consequence modeling conducted by SRNL duringand after the response to the Graniteville accident and thebenefits of leveraging unique technical resources that maybe available in the regional area to support local officialsresponsible for protecting public health and safety duringhazardous materials incidents.

BACKGROUNDThe SRNL’s Weather Information and Display (WIND)System3 provides a comprehensive, automated resourcefor conducting consequence assessment modeling duringemergency response. The WIND System accesses a varietyof real-time sources of local, national, and internationalmeteorological information. Meteorological data col-lected from a local tower network (mesonet) are availablein real-time. A continuous satellite-based feed from a pri-vate sector service is used to obtain regional, national, andinternational data from the National Weather Service(NWS) and other government agencies.

The backbone of the WIND System is a cluster ofUNIX workstations that gather local and regional datafrom the mesonet and satellite feed and archive these datainto a relational database. Meteorological observationsare extracted from the database every 15 min and down-loaded, along with forecast data from operational runs ofa prognostic mesoscale model, to desktop computers. Asuite of consequence assessment models residing locallyon these computers can then be run with a combinationof the local observations and forecasts to generate thereal-time predictions of plume transport and associatedhazards.

In 1996, the SRNL established mutual aid agreementswith five counties surrounding the SRS,4 including AikenCounty, SC. These agreements delineated three areas ofcollaboration through which WIND System resourcescould be leveraged to enhance regional emergency re-sponse programs: (1) assistance from SRNL in establishingmeteorological monitoring stations in critical areas ofneed; (2) custom WIND System software for use by countyofficials in conducting local consequence assessments us-ing observations from the local meteorological mesonet;and (3) providing technical assistance to the county’semergency management team during a real-time responseto events such as the Graniteville accident.

WIND System ComponentsMeteorological Measurements. The mesonet of meteorolog-ical monitoring stations currently incorporated intoWIND System operation is illustrated in Figure 1. Towerslocated adjacent to each of the SRS’s eight major opera-tional areas are instrumented to measure wind speed,wind direction, turbulence, temperature, and moisture atan elevation of 61 m above ground level. A ninth towerlocated near the center of the SRS collects similar data atfour levels up to a height of 61 m. The ATG also operatesinstrumentation on a nearby television tower facility formeasurements of wind, temperature, and moisture at 30,61, and 304 m. Every 15 min, tower observations aretransmitted to the UNIX workstation cluster and ar-chived. In support of the mutual aid agreements, an ad-ditional four monitoring stations were installed in Au-gusta, GA. Measurements of wind and temperature atthese four sites are collected at a single level with heightsranging from 10–60 m above ground. In total, there are14 stations located within a 25-km radius of Granitevillewith variable spacing of 3–15 km between them. Terrainin the general area that includes Graniteville and the SRSconsists of low-rolling, forested hills. Most of the sites inthe mesonet are generally free from any adverse influenceof topography, and measurements are taken at an eleva-tion that is well above the forest canopy. Consequently,wind data from the mesonet are typically expected toaccurately represent conditions throughout the region.One exception may occur during highly stable conditionsat night when terrain-induced drainage flows can occur inshallow valleys (Graniteville is located in such a valley).Measurements from the regional mesonet are supple-mented by Southeastern U.S. surface observations andupper-air soundings from the NWS. There are two NWSsurface stations within 30 km of Graniteville (Augusta’s

Figure 1. Map of the area including the SRS (cross-hatched),Aiken, and Graniteville. Shading represents topographic elevation inmeters above sea level (20-m increments) with lighter shading atlower elevations and darker shading at higher elevations. Urbanareas (and lakes within the SRS) are indicated with the darkestshading. Station locations that are a part of the SRS regional mon-itoring network are indicated by the black circles.

Buckley, Hunter, Addis, and Parker

Volume 57 March 2007 Journal of the Air & Waste Management Association 269

Bush and Daniel Field). However, these stations do notreliably represent the flow features at Graniteville.

Meteorological Forecasts. The ATG has configured a prog-nostic atmospheric model, the Regional AtmosphericModeling System5 (RAMS), to generate routine three-di-mensional forecasts of meteorological conditionsthroughout the Central Savannah River Area (CSRA) ofGeorgia and South Carolina and much of the SoutheastUnited States. Detailed 6-hr forecasts of wind speed, di-rection, turbulence, and other meteorological variablesare generated every 3 hr for a region encompassing theCSRA with an inner horizontal grid spacing of 2 km. Inaddition, regional 36-hr forecasts using RAMS at a 10-kmgrid spacing are generated every 12 hr for an area thatincludes much of Georgia and South Carolina.

For the local simulations used during the train acci-dent, the initial conditions were provided by the NationalCenters for Environmental Prediction (NCEP) Rapid Up-date Cycle (RUC) model.6 The modeling of surface condi-tions requires land-use features such as topography, veg-etation type, and soil type. Variable input soil moistureconditions are also used. Topography data are obtainedfrom the United States Geological Survey (USGS) at vary-ing resolutions, whereas the vegetation data are also USGSproducts based on the Normalized Difference VegetationIndex. The soil type assumed here is sandy clay loam,whereas the variable soil moisture input is obtained fromthe NCEP Eta Data Assimilation System model at �40 kmresolution. The model’s lowest atmospheric level is at20 m above ground. This grid system is designed foremergency response needs at the SRS (centered at33.256 °N, 81.750 °W), located to the south-southeast ofGraniteville by 30 km (Figure 2). Graniteville is located

within the inner, higher-resolution domain.

Atmospheric Transport and Consequence Assessment Models.ATG maintains a suite of transport and dispersion modelsavailable for assessing consequences of hazardous materi-als released to the environment. These models are tailoredto support a broad range of assessment needs during theearly and intermediate phase of response. Two of thesemodels (Puff/Plume and Lagrangian Particle DispersionModel [LPDM]) were used during the Graniteville re-sponse, whereas a more sophisticated non-SRNL model(Hazard Prediction and Assessment Capability [HPAC])was used during the postanalysis. Descriptions of eachmodel follow.

Puff/Plume7 is a segmented trajectory Gaussian dis-persion model that provides decision-makers a rapid andconservative assessment as an initial estimate of potentialdownwind hazards from a chemical or radiological re-lease. Puff or plume release trajectories are constructed for�12 hr of observed and forecast winds with results avail-able in �1 minute. Output can also be exported for usewith geographic information system software.

The LPDM8 provides refined transport and dispersionanalyses on local to regional scales. LPDM uses a three-dimensional wind forecast by RAMS to account for com-plex wind patterns because of the effects of terrain andmesoscale phenomena, such as fronts. Although primar-ily configured to calculate dose and deposition for radio-logical releases, LPDM can be used to simulate the disper-sion of any passive contaminant, such as trace orpollutant gases. The position of all particles is tracked bysolving the Langevin stochastic differential equations forsubgrid-scale turbulent velocities,9 and atmospheric dif-fusion is modeled by a Markov chain process.

The HPAC is a software package developed by theDefense Threat Reduction Agency.10 HPAC is actually asuite of models that allows for various modes of release ofradiological, chemical, and biological agents. HPAC gen-erates interpolated meteorological data fields based onmeteorology and transports the material using a transportand diffusion model (SCIPUFF11,12). SCIPUFF describesdiffusion processes using second-order turbulence closureby relating the dispersion rate to velocity fluctuation sta-tistics, and also provides variance in the concentrationfields, allowing for a measure of uncertainty in the result.

It should be noted that although SRNL did not be-come actively involved in the response until several hoursafter the accident, Puff/Plume and LPDM were still used toassess an ongoing release consisting of mainly passiveoff-gassing of a residual inventory of chlorine remainingin the ruptured tanker. With its more robust handling ofsource term input and its ability to simulate dense gases,HPAC was used in postaccident analyses for detailed sim-ulations of the release during the first 2–3 hr after thecollision when the chlorine cloud likely exhibited densegas behavior.

INCIDENT DESCRIPTIONAs reported by the National Transportation Safety Board,1two trains collided in the town of Graniteville in the earlymorning of January 6, 2005. A stationary train sitting ona siding rail (spur) servicing a textile mill was struck by

Figure 2. Standard local grid configuration used by the SRNL forgenerating mesoscale atmospheric conditions. Also shown arenearby cities including Augusta (Aug) and Columbia (Col), along withinterstate highways (light lines), rivers (dark lines), population cen-ters (shaded areas), and NWS stations.



another train carrying a variety of hazardous chemicals,including liquid chlorine (Cl2). The switching mechanismwas inadvertently set to send trains to the spur, ratherthan to the main track northward and out of town. Theaccident occurred at 2:40 a.m. and resulted in the derail-ment and catastrophic breach to one of several 90-t carscontaining the pressurized liquid Cl2. On contact withthe atmosphere, the Cl2 rapidly vaporized and became anairborne threat to the immediate vicinity. A detailedUSGS map of Graniteville is shown in Figure 3, and aphotograph of the damage at the time of the incident isshown in Figure 4. The figure illustrates the presence ofnumerous residences very near to the site of the collision.

Table 1 provides a synopsis of the range of meteoro-

Figure 3. Detailed USGS map of Graniteville (1:24,000 scale),showing the location of the train collision. The thin lines indicatetopography relief at 10-ft intervals, whereas the black dots representindividual residences. Terrain in the immediate crash area is seen tobe sloped slightly upward from north to south, with steeper sectionsat further distances to the east and west forming a broad, shallowvalley around the accident site. The moving train was travelingnorthward at the time of the collision.

Figure 4. Train collision in Graniteville with details relating to chemicals released and nearby buildings. The picture is oriented such that northis directed to the right (photo courtesy of the Graniteville Volunteer Fire Department).

Table 1. Range of meteorological conditions (min to max) observed fromthe SRNL mesonet (between 3:00 and 6:00 a.m., January 6, 2005).

Variable Values

Wind direction (from) South-southwestWind speed (61 m) 3.2–4.5 m/sec (7–10 mph)Wind speed (surface) 0.9–1.8 m/sec (2–4 mph)Temperature 11–13 oC (52–56 oF)Relative humidity 92–98%Atmospheric stabilitya E and DCloudiness at AGS: Partly cloudy

Notes: Figure 5b indicates winds from the mesonet at 3:00 a.m. aExpressed as aPasquill-Gifford stability (turbulence) class. Classification is based on the standarddeviation of horizontal wind direction fluctuations (��) using a classification schemerecommended by U.S. Environmental Protection Agency.14



logical conditions observed from the WIND System re-gional mesonet at the time of the accident. The south-southwest wind that was observed throughout themesonet and at NWS stations across the region was theresult of clockwise flow around a surface high-pressuresystem centered off the Southeast U.S. coast. A map ofsynoptic conditions the morning of the accident13 isshown in Figure 5a, whereas winds from the regionalmesonet at 3:00 a.m. local time are shown in Figure 5b.The observed conditions given in Table 1 and Figure 5 arebelieved significant to the subsequent behavior of thechlorine cloud.

The uniform nature of observed wind direction andspeed across all of the stations in the mesonet suggested

that synoptic scale pressure gradients generally were likelysufficient to overcome possible microscale flows driven bylocal terrain. Observed wind speeds and turbulence inten-sities were greater than is often observed in winter duringthe early morning hours when very stable conditions arelikely. In fact, measured values of turbulence intensitycorresponded with an atmospheric stability that variedbetween neutral and weakly stable.14 The uniformity ofthe observations and the lack of a strongly stable bound-ary layer provided the SRNL response team confidencethat the regional wind observations were reasonably rep-resentative of conditions in the Graniteville area.

Furthermore, these meteorological conditions wereprobably sufficient to cause some turbulent mixing andsteady erosion of contaminant along the periphery of thedense cloud that formed immediately after the crash andtank rupture. Subsequent passive transport of chlorine gasthen occurred toward slightly higher terrain and less pop-ulated areas to the north-northeast of Graniteville. Tele-vision broadcasts shortly after dawn showed reasonablygood visibility around the crash site, indicating that mostof the initially compact dense cloud had been flushedfrom the shallow valley in which Graniteville is located.The turbulent transport and dispersion of this nighttimecloud likely reduced airborne concentrations of Cl2 overthe longer term (minutes to hours) and quite possiblycould have reduced the number of casualties resultingfrom the accident.

Roughly 1 month after the incident occurred, two ofthe authors surveyed vegetation damage in the area. Veg-etation can be damaged from exposure to chlorine con-centrations as low as 0.1–5 ppm for �2 hr.15 Visible dam-age to vegetation in the area is noted to have covered anarea roughly oriented north-south along the shallow val-ley. The main vegetative damage occurred with pine treesand juniper bushes.16 Photographs such as the onesshown in Figure 6 indicate bleaching of the pine trees(with subsequent death of some trees, such as that shownin Figure 6a). Such visual evidence can be used to infer thespatial extent of the most concentrated portion of theplume. In particular, Figure 6b shows little or no damageto the tops of these pine trees, whereas the lower portionsare adversely affected. This vertical extent of bleachingsuggests that the highest concentrations of Cl2 only ex-tended a few tens of meters into the atmosphere.

Support During IncidentReal-Time Support. Consultations with the Aiken CountyEmergency Management Agency (EMA) began around7:00 a.m. on the morning of January 6. Initially, the mostsignificant challenge for the SRNL response team was thedevelopment of a source term that would provide anadequate basis for assessing protective action measures.Aiken County officials reported that the accident resultedin breaches to railcars containing chlorine, sodium hy-droxide, and cresol. The chlorine was recognized as by farthe most volatile and significant of the three substanceswith respect to potential airborne hazards. Based on re-ports of injuries and possible fatalities occurring duringthe predawn hours, the breach of the chlorine tanker wasbelieved to have resulted in a rapid initial discharge andvaporization of most of the contents, followed by slow,

Figure 5. (a) Synoptic conditions at 7:00 a.m. during the incident.(b) Local mesonet wind observations at 3:00 a.m.



steady off-gassing of a residual inventory in the tank.These assumptions were supported by television reportsfrom the scene, which showed no significant venting of avisible chlorine plume or substantially reduced visibilityat the scene.

Aiken County officials initially implemented an evac-uation order during the predawn for all of the residentswithin a default 1.6-km (1-mi) radius of the incidentscene. Therefore, the primary objective for consequencemodeling at the time of the SRNL activation was to sup-port a reassessment of the initial evacuation order withrespect to the subsequent ongoing release from residualoff gassing. In a situation with only limited incident sceneinformation, the SRNL uses a default source term to allowfor a preliminary atmospheric transport assessment to beconducted. As better data are obtained, refinements to thesource term and subsequent model runs are made. Al-though modeled airborne concentrations are highly un-certain, valuable information on transport of any hazard-ous materials can quickly be assessed. In this case ofsecondary off gassing from a rail car accident, an arbitrarydefault source term of 0.454 kg/sec (60 lb/min) assignedby Puff/Plume was used as a release rate. Although this

release rate is much less than default values that would beapplicable to the initial discharge, it was believed to rep-resent a reasonable value for an off-gassing scenario.

The first Puff/Plume transport model calculation,shown as Figure 7, was posted to an external Web site foraccess by officials at Aiken County’s Unified CommandCenter around 8:00 a.m. Although highly uncertain, re-sults showed that predicted downwind concentrationsgreater than the Emergency Response Planning GuideLevel 2 (ERPG-2) threshold of 3 ppm extended no morethan 1.6 km (1 mi) downwind of the accident site. TheERPG levels for chemicals are developed by the AmericanIndustrial Hygiene Association.17 The ERPG-2 values arethe consensus standard for implementing measures toprotect industrial workers and the public and represent aconcentration below which nearly all individuals couldbe exposed for �1 hr without developing severe or irre-versible health effects.

Discussions continued through the morning withAiken County EMA and with representatives from theSouth Carolina Department of Health and EnvironmentalControl (SC-DHEC) to begin evaluating the accuracy ofthe first model calculation and refine the assumptions onsource term. Unsubstantiated reports by motorists alongInterstate 20, �20 km north of the incident, indicatedthat odors had been noted; however, SC-DHEC reportedno evidence of detectable levels of chlorine in areas down-wind of the immediate incident scene after daylight.These reports provided the SRNL assessment team in-creased confidence that the chosen release rate was rea-sonably conservative with respect to the ongoing residualoff gassing.

By 9:00 a.m., results were available from an LPDMsimulation of downwind transport using RAMS forecasts

Figure 6. Photographs of vegetation damage near the crash site.(a) Pine tree located �30 m south of the crash site. (b) Pine treeslocated �100 m northwest of the crash site. Note browning of thepine trees below the very tops, which remained green in this photo.

Figure 7. Initial Puff/Plume prediction provided to the Aiken CountyEMA and SC-DHEC for the ongoing residual release of chlorine fromthe Graniteville site. Values for the different shadings of dose (ERPGlevels) are indicated on the figure.



of local winds based on initial conditions from the RUCmodel at 1:00 a.m. The forecast period covered by LPDMencompassed 7:00 a.m. to 1:00 p.m. RAMS forecasts forthe Graniteville area showed winds that were consistentwith the regional observations from the mesonet andindicated a persistence of the south-southwest windthroughout the day (Figure 8). NWS and SRNL mesonetobservations of winds during the day verified this RAMSforecast. Animation of LPDM results based on the RAMSforecast showed a corresponding north-northeast trans-port of the residual chlorine release that remained west ofmore densely populated areas near the city of Aiken.

After the initial accident support, the SRNL team con-tinued to provide ongoing support as follows. First, up-dates of Puff/Plume consequence assessment model out-put for the ongoing residual release using observed andforecasted meteorological conditions were posted to theexternal Web site for use by Aiken County and SC-DHECapproximately every 2 hr. Second, Briefings were made forlocal and state officials on consequence modeling resultsand forecasts of meteorological conditions expected forthe upcoming 12–24 hr period. Third, Puff/Plume modelresults were shown to core staff in the SRS EmergencyOperations Center (EOC), U.S. Department of Energy(DOE) headquarters EOC, and the DOE liaison with theDepartment of Homeland Security.

By midday on January 7, responders at the incidentscene began to plan recovery actions for the rupturedtanker, as well as for �1 additional chlorine tanker thathad been damaged during the collision. One dispositionalternative being considered was to physically lift the twodamaged tankers from the wreckage and remove themfrom the scene. Because of concern that such actionscould result in an additional catastrophic rupture, AikenCounty requested SRNL to estimate possible downwind

consequences resulting from the postulated release of theentire contents of the chlorine tanker in question.

Puff/Plume results for this scenario, using meteoro-logical conditions forecasted for that afternoon, showedan area of potentially life-threatening effects (greater thanthe ERPG-3 concentration of 20 ppm) to a distance of 5km from the crash site and potentially irreversible severeeffects (greater than the ERPG-2 concentration of 3 ppm)to a distance of �20 km from the site (Figure 9). Further-more, winds had shifted to a direction that would trans-port the contaminant more toward the city of Aiken andimpact a nearby hospital and other sensitive facilities.SRNL also forecasted stable conditions to occur during thefollowing night, and such conditions would likely haveled to higher airborne concentrations of Cl2 in the imme-diate vicinity over greater periods of time. Model resultsfor this scenario were posted to the external Web site andprovided to Aiken County as a geographic informationsystem layer for display on the Unified Command Centermapping system. These results led to a decision to deferrecovery actions involving movement of the damagedtankers. Eventually, teams were able to reach the tankerand siphon the remaining chlorine inventory onto un-damaged tankers that were brought to the scene.

PostanalysisGiven the complexity of a dense gas release in a valleyresulting from the Graniteville train crash, a much higherresolution simulation was conducted in an attempt tocapture the near-surface wind fields very near the crashsite. The original two-grid system (Figure 2) was modifiedto incorporate a third and fourth nested grid at 500- and125-m horizontal grid spacing, respectively. In addition,the vertical grid spacing was reduced such that the lowest

Figure 8. Forecasted concentration field from RAMS using thestandard 2-grid simulation domain at 12:00 p.m. where differentshading of the plume denotes orders of magnitude change (assum-ing a continuous unit release).

Figure 9. Puff/Plume prediction provided to the Aiken County EMAand SC-DHEC for a potential catastrophic rupture of a second tankeron January 7. Values for the different shadings of dose (ERPGlevels) are indicated on the figure.



vertical level above ground for the two outer grids was�15 m AGL, whereas for the inner two grids it was 7 mAGL (resulting in 14 atmospheric levels �300 m).

RUC data are available at 3-hr intervals. Beginning at1:00 p.m., January 5, 2005, these data were used to createinitial and lateral boundary conditions in RAMS. Fine-resolution topography (100 m) from USGS digital eleva-tion maps was used for topography. Figure 10 shows thetopographic heights for the 500- and 125-m horizontalgrids used in the simulations. The feature of particularinterest is the valley where the accident occurred and thesteeper terrain located just to the east. The northwesternpart of the city of Aiken is located just to the east andsoutheast of this feature.

There were no meteorological observations taken inthe vicinity of the crash site. The closest near-surfacemeasurements were taken at a tower located on the SRScomplex and two NWS sites located at Daniel (DNL) andBush Field (AGS) airports in Augusta, GA. Unfortunately,most likely none of these is totally representative of actualwind conditions in the valley at the time of the incident,

but they are valuable for understanding the synoptic flowof the region. The nearby station locations considered inthe analyses for this paper are given in Table 2 and alsonoted in Figure 2.

As noted previously, observations from the SRNL me-sonet during the incident showed that surface wind

Table 2. Observation information.

ObservationLocation

Latitude(�N)

Longitude(�W)

Elevation(m ASL)

Height(m AGL)

Da

(km)

AGS 33.37 81.97 40 10 26DNL 33.47 82.03 134 10 24SRS climatology 33.25 81.65 90 18 40Graniteville 33.56 81.81 68 10 –b

aApproximate distance from crash site to observation location. bLocation ofcrash site.

Figure 11. Vertical cross-section along the inner grid along thelatitude of the crash site (indicated by the vertical arrow). Plots ofmeteorological variables (vertical velocity [w, m/sec, a], turbulentkinetic energy [TKE, m2/sec2, b], potential temperature [Theta, K, c],and RH [%, d]) are for 3:00 a.m.

Figure 12. Vertical cross-section along the inner grid along thelatitude of the crash site (indicated by the vertical arrow). Plots ofmeteorological variables (vertical velocity [w, m/sec, a], turbulentkinetic energy [TKE, m2/sec2, b], potential temperature [Theta, K, c],and RH [%, d]) are for 11:00 a.m.

Figure 10. Postaccident assessment showing a simulated plume(small diamonds) from the Graniteville crash site (indicated by thelarge dot in the center of the picture) at 3:00 p.m. Topographiccontours, nearby roads (bold lines), population centers, and windbarbs from the 500-m RAMS grid at 7-m AGL are also indicated.



speeds were greater than the near-calm conditions thatfrequently occur during the early morning hours this timeof year. Nonetheless, the detailed simulation indicatedthat surface winds near the crash site were sufficientlylight (�2 m/sec) to allow for minor channeling throughthe valleys of Graniteville at the time of the accident.Wind speeds increased to �2.5 m/sec by sunrise (7:00a.m.) and to 4 m/sec by midafternoon (3:00 p.m.). Thedirection was generally from the south-southwest orsouthwest during the entire period.

Vertical cross-sections of various meteorological vari-ables (vertical velocity [meters per second], turbulent ki-netic energy [square meters per square second], potentialtemperature [kelvin], and relative humidity (RH) [per-cent]) over the lowest 600 m of the atmosphere along awest-to-east orientation intersecting the train collisionlocation are shown in Figure 11 for a time of 3:00 a.m. Thelocation of the train accident is indicated by the arrow.Vertical velocities were very light near the crash site withlow turbulence levels. Potential temperature profiles indi-cated a weakly stratified atmosphere, and humidity levelswere very high (�100%). For the dense gas-behaving

chlorine cloud, sufficiently light wind coupled withslightly downward air motion near the surface wouldsupport gravity-driven flow.

Surface wind speeds increased by sunrise, and thedepth of the boundary layer began to grow (Figure 12). By11:00 a.m., vertical velocity was simulated to be upwardat the crash site, and the boundary layer depth had grownto between 300 and 400 m. Note that large-scale modelsprovide similar boundary layer depth estimates duringthis time. Unfortunately, there were no special soundingsor other specialized equipment (i.e., SODAR [sonic detec-tion and ranging]) taken during the incident with whichto compare simulated midday estimates. The nearest NWSupper-air stations at Atlanta, GA, and Charleston, SC,located �200 km west and southwest, respectively, onlytake soundings twice per day at 7:00 a.m. and 7:00 p.m.Eastern Standard Time. The atmospheric moisture con-tent had also begun to drop because of daytime heatingand convective mixing in the boundary layer. This wouldalso support reports that the cloud had dispersed by thistime.

Figure 13. (a) Comparison of observed wind direction from various locations (isolated markers) and simulated values using RAMS (lines) asa function of local time from the beginning of the incident on January 6, 2005. The bolder line indicates model results from Graniteville, whereno observations of wind direction were taken. (b) Comparison of wind speeds.



Time-series plots of wind speed and direction weregenerated for the period 2:30 a.m. to 7:00 p.m., January 6,and interpolated to the observation locations discussedpreviously. In addition, simulated wind speed and direc-tion as interpolated to the surface level (10 m AGL) inGraniteville at the crash site are also plotted. All of thesimulated values (except Graniteville) were taken frommeteorology generated on grid 2. Values were taken fromthis grid, because the extent of the finer grids did notreach either the SRS or Augusta. Comparisons of winddirection and wind speed are given in Figure 13.

At the time of the collision, all three of the observa-tion sites indicated winds from the south-southwest tosouthwest, although DNL was closer to 225 °, whereas theother stations were closer to 200 °. Directions were gen-erally uniform throughout the period, with slight veeringto 240 ° by late day (5:00 p.m.). Simulated values weregenerally within 20 ° to 40 ° of the observations, althoughearly morning values at SRS indicated a more southerlycomponent. As expected, simulated values at Granitevillelie within values obtained from all of the measurementsites.

Observed wind speeds (Figure 13b) were all �3 m/secat the time of the accident, before peaking at �8 m/sec bylate afternoon. The simulated trends follow along theselines, although values between 2:30 a.m. and 7:00 a.m. atthe airports were �2 m/sec. Simulated wind speeds atGraniteville steadily rose from 1 m/sec at the time of thecrash to a maximum of 4.5 m/sec by 2:00 p.m. beforetailing off later in the afternoon.

HPAC was used to simulate the effects of the densegas very near the release location. An industrial transportaccident was assumed for a tanker containing �20,000 galof Cl2. A major rupture was assumed to occur instanta-neously expelling �62,000 kg of chlorine gas in bothvapor and aerosol phase. This source term is consistentwith the NTSB estimate of a release of �70 t. Dense gascalculations were made within HPAC, and appropriatemedian lethal (LCt) and incapacitating (ICt) concentra-tion levels, as well as ERPG levels, were then determined.Figure 14 illustrates the plume footprint 1 hr after thecrash using the 500-m resolution RAMS winds, with de-tailed road overlays and the approximate location ofdeaths that occurred as a result of the accident. Becauseterrain near the site is actually lower to the southwest ofthe crash site, the relatively light winds coupled withgravity-driven spreading of the dense gas explains whyseveral deaths occurred southwest (upwind) of the wreck,although winds were simulated (and observed) to beblowing weakly from the southwest toward the northeastat the time of the crash.

The simulations appear to be reasonable in one otherway. Also shown in Figure 14 (dark bold line) is an ap-proximate outline of visible damage to vegetation in thearea described earlier. The resulting footprint comparesfavorably with both vegetation damage and the locationsof deaths that occurred as a result of the incident.

CONCLUSIONSThe SRNL’s WIND System, a real-time consequence assess-ment resource for emergency response, performed as de-signed to provide timely support to Aiken County and

state of South Carolina officials responding to the NorfolkSouthern rail accident in Graniteville. Results were usedto reassess the appropriateness of initial protective actionsfor the surrounding community and to plan incident scenerecovery actions. The value of SRNL’s support to local emer-gency managers was noted in subsequent newspaper re-ports.18 Furthermore, experience gained from the Gran-iteville response clearly demonstrates that local/regionalconsequence assessment assets can play a valuable role dur-ing hazardous material incidents of local and even nationalsignificance by providing nearly full-time, customized sup-port targeted directly to local decision-makers.

Detailed numerical simulations of meteorology dur-ing the Graniteville train collision were generated usingthe mesoscale RAMS model nested with four grids ofhorizontal spacing at 8, 2, 0.5, and 0.125 km. The lowestvertical level above ground was 7 m. Simulated fieldscompare well with nearby observations, with transportfor the first day indicating a plume directed betweennorth and northeast.

Interestingly, although the chlorine cloud was moredense than air, the synoptic winds were strong enough atthe time of the accident (3:00 a.m.) to greatly minimize thegravity driven flow into the upwind, shallow valley. In ad-dition, turbulent mixing was sufficient to cause significanttransport of the plume over slightly higher terrain to thenorth and northeast. Use of a dense gas model to simulatelocalized effects indicates agreement with fatalities in theimmediate area and visible damage to vegetation.

REFERENCES1. National Transportation Safety Board. Collision of Norfolk Southern

Freight Train 192 with Standing Norfolk Southern Local Train P22 with

Figure 14. Plume showing transport of chlorine vapor as simulatedby HPAC using RAMS meteorology (shown as wind barbs) at 3:40a.m. Isopleths indicate varying LCt and ICt levels, as well as ERPGlevels. The thick straight lines indicate railroad tracks, the thin linesindicate roads, and the large dots show the location of deaths thatoccurred from the accident. In addition, the thickest line denotes thevisible extent of vegetation damage �1 month after the incident.(Note that dose levels for isopleth levels correspond to 83,000,19,000, 3000, 1400, 870, 130.5, 43.5, and 21.75 mg min/m3).



Subsequent Hazardous Materials Released at Graniteville, South Carolina,January 6, 2005: Railroad Accident Report, 2005; available at http://www.ntsb.gov/publictn/2005/RAR0504.pdf (accessed 2006).

2. Parker, M.J.; Addis, R.P. Meteorological Monitoring Program at theSavannah River Site; WSRC-TR-93-0106; Westinghouse SavannahRiver Company: Aiken, SC, 1993.

3. Hunter, C.H. The Weather INformation and Display System; Publication05A00978-01; Westinghouse Savannah River Company: Aiken, SC,2005.

4. Hunter, C.H.; Parker, M.J.; Snyder, G.L. Meteorological Partnershipsfor the Savannah River Site/Central Savannah River Area. In Proceed-ings of the 7th Topical Meeting on Emergency Preparedness and Response;Santa Fe, NM, September 14–17, 1999; American Nuclear Society:LaGrange Park, IL, 1999.

5. Pielke, R.A.; Cotton, W.R.; Walko, R.L.; Tremback, C.J.; Lyons, W.A.;Grasso, L.D.; Nicholls, M.E.; Moran, M.D.; Wesley, D.A.; Lee, T.J.;Copeland, J.H. A Comprehensive Meteorological Modeling System—RAMS; Meteor. Atmos. Phys. 1992, 49, 69-91.

6. Benjamin, S.G.; Grell, G.; Brown, J.M.; Smirnova, T.G.; Bleck, R;.Mesoscale Weather Prediction with the RUC Hybrid Isentropic—Ter-rain-Following Coordinate Model; Mon. Wea. Rev. 2004, 132, 473-494.

7. Garrett, A.J.; Murphy C., Jr. A Puff-Plume Atmospheric Deposition Modelfor Use at SRP in Emergency Response Situations; DP-1595; DuPont deNemours and Company, Savannah River Laboratory: Aiken, SC, 1981.

8. Uliasz, M. The Atmospheric Mesoscale Dispersion Modeling System;J. Appl. Meteor. 1993, 32, 139-149.

9. Gifford, F.A. Horizontal Diffusion in the Atmosphere: a Lagrangian-Dynamical Theory; Atmos. Environ. 1982, 16, 505-512.

10. Defense Threat Reduction Agency. The HPAC User’s Guide. HazardPrediction and Assessment Capability, Version 4.0; Defense Threat Re-duction Agency: San Diego, CA, 2001.

11. Sykes, R.I.; Parker, S.F.; Henn, D.S.; Lewellen, W.S. Numerical Simula-tion of ANATEX Tracer Data Using a Turbulence Closure Model forLong-Range Dispersion; J. Appl. Meteor. 1993, 32, 929-947.

12. Sykes, R.I.; Parker, S.F.; Henn, D.S.; Cerasoli, C.P.; Santos, L.P. PC-SCIPUFF Version 1.2PD, Technical Documentation; ARAP Rep. 718;

Titan Research and Technology Division, Titan Corp.: Princeton, NJ,1998; available at http://www.titan.com/products-services/336/download_scipuff.html (accessed 2006).

13. National Oceanic and Atmospheric Administration. Daily WeatherMaps. 2005; available at http://www.hpc.ncep.noaa.gov/daily_wxmap/index_20050106.html (accessed 2006).

14. Meteorological Monitoring Guidance for Regulatory Modeling Applications;EPA-454/R-99-005; U.S. Environmental Protection Agency: ResearchTriangle Park, NC, 2000.

15. Sikora, E.J.; Chappelka, A.H. Air Pollution Damage to Plants. ACESPublications, ANR-0913: Alabama Cooperative Extension System,1996; available at http://www.aces.edu/pubs/docs/A/ANR-0913 (accessed2006).

16. Holmes, S.B. Chlorine Gas Takes Affect on Graniteville Plant Life. TheAiken Standard, February 12, 2005.

17. The AIHA 2005 Handbook. Emergency Response Planning Guidelines andWorkplace Environmental Exposure Level Guides. American IndustrialHygiene Association: Fairfax, VA, 2005.

18. Nesbitt, J. Lab Aided Emergency Effort. The Augusta Chronicle, January26, 2005.

About the AuthorsAll of the authors are with the Savannah River National Lab-oratory. Address correspondence to Robert L. Buckley, Sa-vannah River National Laboratory, Savannah River Site, 773A-A1008, Aiken, SC 29808; phone: �1-803-725-1926; fax: �1-803-725-4233; e-mail: [email protected].



Date post:	30-Apr-2020
Category:	Documents
Upload:	others
View:	14 times
Download:	0 times

Modeling Dispersion from Toxic Gas Released after a Train ... · Modeling Dispersion from Toxic Gas...

Documents