Environ Fluid MechDOI 10.1007/s10652-006-9001-8

O R I G I NA L A RT I C L E

Multiscale plume transport from the collapseof the World Trade Center on september 11, 2001

Georgiy Stenchikov · Nilesh Lahoti ·David J. Diner · Ralph Kahn · Paul J. Lioy ·Panos G. Georgopoulos

Received: 23 November 2005 / Accepted: 16 May 2006© Springer Science+Business Media B.V. 2006

Abstract The collapse of the world trade center (WTC) produced enhanced lev-els of airborne contaminants in New York City and nearby areas on September 11,2001 through December, 2001. This catastrophic event revealed the vulnerability ofthe urban environment, and the inability of many existing air monitoring systems tooperate efficiently in a crisis. The contaminants released circulated within the streetcanyons, but were also lifted above the urban canopy and transported over large dis-tances, reflecting the fact that pollutant transport affects multiple scales, from singlebuildings through city blocks to mesoscales. In this study, ground-and space-basedobservations were combined with numerical weather forecast fields to initialize fine-scale numerical simulations. The effort is aimed at reconstructing pollutant dispersionfrom the WTC in New York City to surrounding areas, to provide means for eventu-ally evaluating its effect on population and environment. Atmospheric dynamics werecalculated with the multi-grid Regional Atmospheric Modeling System (RAMS), cov-ering scales from 250 m to 300 km and contaminant transport was studied using theHybrid Particle and Concentration Transport (HYPACT) model that accepts RAMSmeteorological output. The RAMS/HYPACT results were tested against PM2.5 obser-vations from the roofs of public schools in New York City (NYC), Landsat images,and Multi-angle Imaging SpectroRadiometer (MISR) retrievals. Calculations accu-

G. Stenchikov (B)Department of Environmental Sciences, Rutgers University, New Brunswick, NJ 08901, USAe-mail: gera@envsci.rutgers.edu

N. Lahoti · P. J. Lioy · P. G. GeorgopoulosDepartment of Environmental and Occupational Medicine, UMDNJ—R.W. Johnson MedicalSchool, Piscataway, NJ 08854, USA

N. Lahoti · P. J. Lioy · P. G. GeorgopoulosEnvironmental & Occupational Health Sciences Institute, UMDNJ—R.W. Johnson MedicalSchool & Rutgers University, Piscataway, NJ 08854, USA

D. J. Diner · R. KahnJet Propulsion Laboratory, California Institute of Technology, Pasadena, CA 91109, USA

Environ Fluid Mech

rately reproduced locations and timing of PM2.5 peak aerosol concentrations, as wellas plume directionality. By comparing calculated and observed concentrations, theeffective magnitude of the aerosol source was estimated. The simulated pollutantdistributions are being used to characterize levels of human exposure and associatedenvironmental health impacts.

Keywords Aerosol plume · Particulate matter · Transport · Urban pollution ·Regional Atmospheric Modeling System · Hybrid Particle and ConcentrationTransport Model · Multi-angle Imaging SpectroRadiometer · World Trade Center ·9/11 · Terrorist attack

1 Introduction

This study considers the transport of airborne contaminants (mostly particulate mat-ter [PM] or aerosols) produced by the collapse of the world trade center (WTC) inNew York City (NYC) on September 11, 2001, and by the subsequent burning ofthe remaining materials. The massive release of aerosols and gases on September11 affected numerous residents and commuters in the surrounding New York/NewJersey (NY/NJ) area. The continued threat of terrorist attacks on major cities raisesa new issue: developing a better understanding of the ambient exposures and asso-ciated health effects caused by a massive pollutant release in highly populated areas[1–3]. The overall objective of this study is to reconstruct the WTC plume dispersionin NYC and surrounding areas using available ground- and space-based observationsand numerical modeling to better characterize its environmental and health effects.

The north and south WTC buildings were set on fire by terrorist attacks at 0846EDT and 0903 EDT, respectively, on September 11, 2001. The collapse of the WTCSouth Tower at 0959 EDT followed by the crash of the North Tower at 1029 EDTinstantaneously produced vast amounts of coarse and fine airborne particles thatspread upward and into the streets of southern Manhattan. This initially producedan intensive but relatively short-term particulate mass and gaseous release into theurban atmosphere. Materials were deposited on roofs, streets and other flat surfacesand were re-suspended later by the wind, contributing to the overall airborne con-tamination levels from September 11 through September 13. The remains of theWTC complex, covering a 16-acre area known as ground zero, burned with varyingdegrees of intensity until September 14, occasionally reaching temperatures exceed-ing 1, 000◦C. After September 14 the fire began to diminish due to rain. The fire atground zero produced a continuous source of hazardous gases and aerosols for anextended period of time, which were dispersed in NYC and the surrounding areas.

A detailed spatial and temporal evaluation of the airborne contaminant distri-bution is needed to fully understand the environmental and health impacts of theWTC’s collapse. However, the existing ground-based observation networks (both formeteorological characteristics and particulate matter) are fairly sparse for this pur-pose, even in the NY/NJ metropolitan area. Many monitoring stations in the vicinityof the WTC did not operate properly, as they were completely plugged by largeamounts of dust immediately after the collapse, or they were unavailable becauseof the short-term nature of the initial releases combined with the loss of electricity.Quantitative, satellite-based measurements were limited in temporal coverage. As aresult, many important characteristics of the dispersed pollution field could not beeasily determined to help understand the details of associated human exposures. In

Environ Fluid Mech

the current study, numerical modeling of micrometeorological fields and PM trans-port was combined with available observations to reconstruct plume behavior asrealistically as possible, and to quantitatively estimate the rate of pollutant release.

In NYC and the nearby areas, air pollution is a long-standing and well-recognizedhealth issue [4]. This region is one of the most densely populated in the US. Amongthe sources regularly contributing to atmospheric aerosol loading are local industriesand utility generation, motor vehicle emissions, residential cooking and heating, dustraised from disturbed soils, and marine aerosol production over the coastal waters.In addition, long-range transport of emissions from the industrialized Midwest andBaltimore–Washington areas contribute to the overall pollution level [5–12]. How-ever, the unexpected nature of the catastrophic event on September 11, 2001 pre-vented the collection of adequate amounts of quantitative information necessary toestablish risk.

Transport and deposition of atmospheric tracers is highly dependent on local cir-culation, turbulent mixing in the boundary layer, terrain, and precipitation. Usingmeteorological fields with a coarse spatial resolution often causes uncertainties in cal-culations of contaminant distribution. Unfortunately, fine-scale meteorological fieldsare not available from observations, and operational forecast models provide mete-orological fields with spatial sampling that is not sufficient for high-resolution trans-port calculations. In this study, the Regional Atmospheric Modeling System (RAMS)(http://www.atmet.com/html/docs/documentation.shtml) was employed to downscalethe analysis fields from the Eta Weather Prediction Model [13] conducted with spatialresolution of approximately 32 km. The RAMS databases of land elevation, vegetationcover, and sea surface temperature were improved to account for fine-scale effects ofthe land-surface boundary conditions and sea surface temperature. The downscaledmeteorological fields were used in the Hybrid Particle and Concentration Trans-port model (HYPACT) (http://www.atmet.com/html/docs/documentation.shtml) forthe fine grid transport and deposition calculations. The HYPACT uses RAMS mete-orological output for calculating aerosol transport from localized sources combiningLagrangian and Eulerian approaches. RAMS is a comprehensive mesoscale meteoro-logical model, which is not fully capable of simulating the flow within the city’s streetcanyons. However, it can accurately calculate the flow above the buildings, linking itto larger-scale meteorological structures. To account for the effects of buildings onthe flow in the boundary layer, the surface roughness over Manhattan was increasedup to 1 m, which is a typical magnitude for metropolitan urban areas [6].

To account for the multi-scale structure of the transport, calculations were con-ducted in three nested domains (Fig. 1). The largest domain has a regional scale of300 km, covering NYC and nearby areas of NY/NJ with the grid spacing of 4,000 m.The internal domains allowed calculation of the flow at 1,000 and 250 m2 spatial reso-lutions. The collapse of the WTC towers and the fire at ground zero were not explicitlydescribed to define emissions of aerosols and gases. More detailed computational fluiddynamics (CFD) simulations need to be conducted to calculate those processes and toobtain characteristics of air flow in the street canyons [3, 14]. However, CFD simula-tions require realistic lateral and upper boundary conditions that can be obtained onlyfrom fine-scale meteorological calculations like those conducted using RAMS. Thisstudy relied on available observational data to evaluate the time-dependent height ofthe convective cell generated by the fire at ground zero, and quantified the magnitudeof the aerosol emissions source from the comparison of the simulated and availableobserved concentrations at a number of distant locations (>3 km). The calculated

Environ Fluid Mech

Fig. 1 The model domains used in the simulations, referenced as grids 1, 2, and 3. Symbols showthe location of the ASOS and buoy stations. Land elevation is shown with black contours. Landcover classes from the USGS National Land Cover Dataset are distinguished by color over the land.Three-day average sea surface temperature (K) for September 13–15, 2001, retrieved from AVHRRmulti-channel observations, is shown by red contours

Environ Fluid Mech

PM distribution patterns over NYC, NJ, and NY were tested against available obser-vations that included satellite retrievals, surface meteorological observations, andavailable PM2.5 measurements. It was found that the simulations compare favorablywith observations and allow effective reconstruction of the plume evolution.

This article is organized as follows: Sect. 2, describes the modeling approach andthe simulation setup; Sect. 3 briefly describes meteorological conditions, discussesresults, and provides evaluation of the sensitivity of the results to the parameters andinitialization. Results are summarized in Sect. 4.

2 Methodology

To conduct multiscale atmospheric transport calculations, micrometeorological fieldsneed to be calculated with sufficient accuracy and spatial resolution. To estimatetime-varying human exposures it is also necessary to simulate the distribution of air-borne contaminants with a spatial resolution of, at minimum, a city block. The routineEta model forecast, that provides the best available meteorological fields, resides atthe National Centers for Environmental Prediction (NCEP) and has a spatial resolu-tion of about 32 km. Therefore, dynamically downscaling the Eta fields using RAMSand additional available observations was required. The micrometeorological fieldsobtained this way were then input to HYPACT for off-line transport calculations.

2.1 Calculation of meteorological fields

RAMS Version 4.3 was employed in the analyses to calculate meteorological fields.RAMS is a compressible, non-hydrostatic, regional model with well-developed bulkcloud microphysics, and surface interaction parameterizations [15, 16]. The govern-ing equations are approximated using the hybrid implicit-in-the-vertical time-splitdifference scheme of Tripoli and Cotton [17]. RAMS predicts the 3-D fields of threevelocity components, temperature, water vapor mixing ratio, pressure, sub-grid-scaleturbulent kinetic energy, and several types of cloud hydrometeors including cloudwater, ice, graupel, and snow.

The horizontal grid uses a rotated polar-stereographic projection. In the verticaldirection, RAMS employs a sigma-Z terrain-following coordinate system [18]. Gridnesting is used in RAMS to provide high-spatial resolution in selected areas, whilecovering a large domain at lower resolution. Therefore, effects of large-scale circula-tion patterns can be transferred to an internal fine resolution region. A nested gridoccupies a region within the computational domain of its coarser parent grid. Forthe external domain, lateral boundary conditions are applied by exponential relaxing(nudging) the calculated fields toward the flow obtained from the forecast model inthe grid-belt along the lateral boundaries [19]. The relaxation coefficient follows aparabolic function of the distance from the boundary and is constant in height. Forthe internal domains the two-way interactions between nested grids are calculatedfollowing Clark and Farley [20].

Various parameterization modules were available for most physical processes,including radiation, turbulence, and land/atmosphere interaction. As vertical and hor-izontal resolutions are relatively different in this study, vertical turbulent eddy mixingwas parameterized using the 2.5 level scheme of Mellor and Yamada [21, 22] based ona prognostic equation for turbulent kinetic energy. Horizontal turbulent mixing was

Environ Fluid Mech

calculated using turbulent diffusion coefficients calculated from the tensor of defor-mation [23]. The Two-Stream Delta-Eddington radiative schemes of Harrington [24]was used for radiative transport.

Modified versions of Kuo [25] and Fritsch and Chappell [26] convective parame-terizations are standard features of RAMS [27]. A modified version of the Kain andFritsch convective scheme [28, 29] was recently implemented in RAMS [30]. How-ever, for cloud-resolving calculations, as in this study, RAMS does not require anyconvective parameterization.

The cloud microphysics scheme is based on Tripoli and Cotton [17, 31] and Cottonet al. [32]. This scheme consists of a set of conservation equations for water vaporand six hydrometeor types: cloud droplets, raindrops, pristine ice, snow, graupel, andaggregates. Their tendencies are affected by advection, turbulence, and microphysicaltransformations in size distribution and from one class to another.

Calculation of land-atmosphere interaction is based on the Land Ecosystem-Atmosphere Feedback (LEAF-2) model [33], with 12 soil textural classes and 18vegetation types. LEAF-2 predicts soil temperature and water content, snow cover,vegetation, and canopy air as well as turbulent and radiative exchanges between thesecomponents. LEAF-2 uses a mosaic approach where the grid cells are subdivided intosmaller portions or “patches” corresponding to different surface characteristics occur-ring in the area covered by the grid cell.

RAMS has been tested in numerous applications for atmospheric chemistry andair pollution, including recent studies of the sulfur cycle and acid deposition in EastAsia [34], calculations of the chemical production of tropospheric ozone over Greece[35] and in the area of Phoenix, Arizona [36]. RAMS has also been recently used forclimate downscaling over the US [30, 37].

The triple-nested domain used in the present simulations, centered at the coordi-nates of the WTC (74.03◦W, 40712◦N), is shown in Fig. 1. The largest (or “parent”)domain covers a 300×300 km area in a polar stereographic projection with projectionaxes at 40.783◦N and 73.967◦W. The parent domain is necessary to accommodatemesoscale structures to be downscaled to two smaller nested domains centered at thesame central point, covering areas of 54 × 54 and 10.5 × 10.5 km, respectively. Goingforward, the three domains will be referred to as grids 1, 2, and 3. The spatial resolu-tions of the three grids arre 4 × 4, 1 × 1, and 0.25 × 0.25 km and the number of gridpoints is 75 × 75, 54 × 54, and 42 × 42, respectively. The vertical grid is non-uniform,containing 39 levels starting from a 20 m-thick surface layer, and reaching 1700 m atthe top of the domain, at an altitude of 16 km.

The original RAMS land elevation and vegetation cover data sets are of 1 kmresolution, which is sufficient to calculate mesoscale circulation but is not adequatefor the needs of the present study. To conduct very fine-resolution simulations inthe metropolitan area it was necessary to improve the model databases. First thehigh-resolution National Land Cover Dataset (NLCD) and National Elevation Data(NED) from the United States Geological Survey (USGS) were adopted. NED, araster product, is available on the Internet at http://edcwww.cr.usgs.gov/doc/edchom-e/ndcdb/ndcdb.html. NED has a resolution of 1 arc-second or about 30 m for theUS. A visual basic software application was developed to read pixel values andconvert them to a digital data file. Land elevation is shown as black contours inFig. 1.

NLCD is a multi-layer and multi-source database that contains a 30 mresolution, 21-class land classification for the territory of the US, in the form of visual

Environ Fluid Mech

images. A visual basic script was used to read pixel values and to produce a digitaldata file. The NLCD classes were then converted to Olson type classes (http://edcdaac.usgs.gov/glcc/globdoc1_2.html) and a LEAF2 database for RAMS was pro-duced. In Fig. 1 vegetation classes are distinguished by color.

The original RAMS Sea Surface Temperature (SST) is based on the climatologicallyaveraged monthly mean 1 × 1◦ resolution data set [38]. However, fine-scale pollutanttransport can be affected by sea breezes initiated by the actual land/sea temperaturecontrast; therefore, the simulations in this study used real-time SST, providing betterspatial and temporal resolution for the NY metropolitan area. For this purpose 1 kmresolution, multi-channel, Advanced Very High Resolution Radiometer (AVHRR)satellite retrievals [39] were acquired from the Marine Remote Sensing Laboratoryof the Rutgers University Institute for Marine and Coastal Sciences. This data set hadto be processed to remove the effect of clouds seen in the instantaneous retrievals,in order to produce 3-day SST composites. In Fig. 1 the composite SST field for Sep-tember 11–14, 2001 is shown as red contours over a blue background correspondingto the “ocean–lake–river–stream” surface classification group.

The terrain in Fig. 1 is fairly flat, not exceeding 400 m in elevation. The fine-scalefeatures were degraded on the parent grid in Fig. 1; nevertheless the 200 m high-narrow Palisades Cliff on the west side of the Hudson River, northwest of NYC, iswell captured and the coastline is well approximated. The dominant land cover typeis urban, with small intrusions of grassland, marsh, and trees. The AVHRR SST forSeptember 11–14, 2001 shows warm areas of 297–298◦K related to the Gulf streampath. The colder waters of 296◦K and below are transported southward from the Lab-rador Sea, along Long Island and the NJ coast. The ocean temperature in the GulfStream region is fairly patchy, but becomes smoother near the coast.

The AVHRR SSTs were tested with buoy observations available from the NationalData Buoy Center (http://www.ndbc.noaa.gov/to_station.shtml). The two stationsclosest to NY/NJ coast were chosen for comparison: ambrose light (station ID ALSN6,located at 40.46◦N, 73.83◦W); and Long Island (station ID 44025, located at 40.25◦N,73.17◦W). These are also marked in Fig. 1. Figure 2 compares AVHRR compositedSST, sampled at 40.46◦N, 73.5◦W along the NY coast and shown as a solid curve with-out marks, with the Ambrose Light (closed circles) and Long Island (open circles)buoy stations. The AVHRR SST, sampled between the stations, compares favorablywith the station observations, catching all SST changes during this period. The buoyhourly output shows more high-frequency variations, but the three-day average AV-HRR composites show fairly accurately that SST decreases from September 11 toSeptember 17, and then stabilizes at about 294.5◦K, and at the end of Septemberdrops again to 291◦K. The SST change during the first week following the WTC col-lapse is most important, as it drove breeze circulation during the period when theemissions were most intensive and the plume was especially dense.

The meteorology calculations depended on the initial and boundary conditionsthat were developed using the objective analysis package within RAMS. These objec-tively analyzed fields are calculated from the three-hourly Eta model operationalanalysis [40, 41]. The Eta Model data were provided by the National Center forAtmospheric Research (NCAR) in gridded binary (GRIB) format, on a horizon-tal grid with a spatial resolution of 32 km. They included surface pressure, surfaceelevation, and 3-D fields of pressure, temperature, water vapor mixing ratio, and hor-izontal wind components at 26 pressure levels, for the entire US. RAMS is able tocombine and blend several input data sets in the data analysis. For example, the Eta

Environ Fluid Mech

Fig. 2 Three-day composited sea surface temperature (K) at the NJ coast from the AVHRR retrievalat 40.46◦N, 73.5◦W (solid line) and hourly observations from the long island (40.25◦N, 73.17◦W) andAmbrose Light (40.46◦N, 73.83◦W) buoy stations near the NJ coast (open and filled circles, respec-tively)

fields could be enhanced by surface station data from NCEP and Automated SurfaceObservation Stations (ASOS) available from the National Climate Data Center(NCDC) (http://www4.ncdc.noaa.gov/cgi-win/wwcgi.dll?wwdi∼ASOSPhotos). Thefive ASOS stations located closest to the WTC were used in the present study, andare shown in Fig. 1. Unfortunately, upper air observations in the NY area are sparse;for example, the closest rawinsonde soundings are taken at Brookhaven national lab-oratory on Long Island. Therefore, upper air observations could not be used in theanalysis.

The objectively analyzed 3-hourly fields helped constrain the flow near the bound-aries of the grid 1 domain using relaxation type boundary conditions [19] with anefolding relaxation time of 30 min. at the 5-grid-cell boundary belt. In addition, tokeep the flow close to the observations during the entire simulation period, horizontalvelocity, potential temperature, and Exner function � = (p/p0)

R/Cp [42] were nudgedin the interior of the domain, with a much greater relaxation time time of 12 h to allowsmall-scale high-frequency disturbances to develop. (In this formula, p and p0 are airpressure at given locations and base state pressure at the ground, respectively, R is thegas constant for dry air, and Cp is specific thermal capacity of air at constant pressure.)

The majority of RAMS simulations in this study were conducted using the radi-ative scheme of Harrington [24], the turbulent closure of Mellor and Yamada [22],

Environ Fluid Mech

and driving fields calculated using Eta fields and 3-hourly data from ASOS stationsshown in Fig. 1. Comparison with observations revealed that these settings producedresults superior to others tested in the course of this study. The ASOS observationsaccounted for observed fine structure of the flow in the vicinity of the WTC that waslost in the 32 km resolution Eta analysis. The inclusion of 6-hourly station data fromNCEP did not produce an improvement because they were too sparse to affect localcirculation structures. In addition, they caused inhomogeneity in the driving fieldsbecause the NCEP data are not available at each 3-h time step. Below, the results arepresented along with a sensitivity analysis discussing the dependence of the results onmodel parameters and driving field variations.

RAMS integrations were conducted for 4 weeks, from September 11 to October 8,2001, with a time step of 12 s. The meteorological fields were saved every 30-min.

2.2 Calculation of pollutant transport

The chemical analysis of sampled aerosol particles that had settled to the ground[1] and in NY Harbor sediments [43] show that the initial WTC emissions includedcement, cellulose, glass fibers, asbestos, lead, and polycyclic aromatic hydrocarbons(PAHs). The WTC debris deposited in the Hudson River and then transported down-stream left a distinct signature on NY Harbor sediments, affecting the sedimentaryrecords of Ca, S, Sr, Cu, and Zn. In this study all types of aerosols and gases asso-ciated with WTC emissions were treated as tracers, and their transport calculatedoff-line using HYPACT model, Version 1.2. HYPACT model calculates temporal andspatial distributions of atmospheric pollutants using 3-D, time-dependent wind andturbulence fields. It can account for multiple sources and various weather regimes,including complex terrain flows, land/sea breezes, or circulation in urban areas. Speciescan include gases and a spectrum of aerosol sizes. Source geometry can include point,line, area, and volume sources of various orientations. HYPACT is driven by windand potential temperature fields simulated in RAMS. The turbulence characteristicsare calculated diagnostically from available meteorological information using the tur-bulent closure of Mellor and Yamada [21, 22]. Particle interaction with the surface isparameterized following Boughton et al. [42]. Above 100 m, the probability of particledeposition for the timescales of interest is negligible. If the particle falls below thisheight, the probability that the particle is deposited is computed in HYPACT fromthe transition probability density given by Monin [44].

HYPACT simulations were conducted for the entire period of RAMS simulationsfrom September 11 to October 8, 2001. The HYPACT uses a 30-s. time step, interpo-lating RAMS 30-min. output at each time step. HYPACT output was archived every30-min.

2.3 Primary and secondary particulate matter sources

The HYPACT transport calculations were driven by the RAMS meteorological fields,and aerosol or gaseous pollutant sources. A source is characterized by the position,surface area, altitude, and rate of pollutant release. The terrorist attack caused fires inboth WTC towers whereby pollutants were released into the atmosphere at an altitudeof about 1500 m. The collapse of the main two structures produced a very fine-scaleintensive low-level jet that mixed pulverized construction materials vertically in acolumn at least 500 m high, and pushed pollutants into the nearby streets. However,

Environ Fluid Mech

the exact shape, altitude, and magnitude of the emission sources are not known. Thesefine-scale processes are the subject of on-going CFD studies, at a spatial resolution ofa few meters [14, 45].

A significant amount of PM from this initial release was deposited on the roofs,streets, and other man-made and natural surfaces. Later, re-suspended by wind, theparticles were released and contributed as secondary sources to the overall pollution,until they were cleaned up or washed out by rain on September 14 [46, 47]. Thecalculations in the present study did not account for those secondary sources. Thisprobably resulted in an underestimation of the overall airborne contaminant levelin the simulations. The fire on the site of the WTC that developed after the collapse ofthe buildings produced a continuous source of aerosols that was most intense duringthe first 3 days, well exceeding the background level. As the fire receded, the effectivealtitude of the source and the emissions release rate gradually decreased.

On separate days fires were present at different locations within the 16-acre groundzero area. When the heat released from the fire was high, it initiated intense convectionthat mixed combustion products in the vertical column. The altitude of the convectivemixing depends on the magnitude of thermal heat flux from the fire, as well as atmo-spheric conditions. The situation is even more complex for the initial dust emissioncaused by the collapse of the WTC main structures.

Therefore, for the purposes of the calculations in the present study, it was assumedthat aerosols were released from the entire area of ground zero. Because of numer-ous uncertainties, direct simulation of the convection caused by the fire was notperformed; instead, an approximation of the time evolution of the altitude of theconvective column was made using photographs taken from the ground and satelliteobservations.

2.4 Plume altitude observations

The North Eastern states for Coordinated Air Use and Management (NESCAUM)organization provided a series of photographs of the plume rising from the WTCtaken from Newark, NJ on September 11–17, 2001. The photos show that on Septem-ber 11 from 0856 to 0950 EDT, after the attack on the buildings by aircraft but beforethe collapse of the buildings, the plume rose above the urban canopy to the height ofabout 1,000–1,500 m. At noon on September 11 the plume reached its highest altitudeof about 1,800 m. On the next day, at 0500 EDT (and probably during the night), thealtitude of the plume was below 400 m, reaching 1,500 m at 1200 EDT. However, inthe late afternoon on September 12 the altitude of the plume decreased to 400 m. OnSeptember 13–17 the plume was mostly confined to the 200–400 m layer, sporadicallyrising to 800 m in the middle of the day when solar radiation heated the aerosol layer,increasing its buoyancy. After September 17 the altitude of the plume continued todecrease and stabilized above the urban canopy at about 150–200 m.

In addition to surface-based photographic observations, the WTC plume was ob-served from space. Figure 3 shows imagery and height retrievals derived from Multi-angle Imaging SpectroRadiometer (MISR) observations. The MISR flies in sun-syn-chronous, polar orbit aboard NASA’s Terra spacecraft, and measures upwelling radi-ance from Earth in four spectral bands centered at 446, 558, 672, and 866 nm, at eachof nine fixed viewing angles spread out along the flight path from 70.5◦ forward to70.5◦ aft [48]. It is a push-broom imager, providing nearly pole-to-pole coverage ofa 400 km wide swath on the day side of each orbit. MISR’s highest spatial sampling

Environ Fluid Mech

Fig. 3 The MISR stereo height analysis of the WTC smoke plume at 1603 UTC (1203 EDT) onSeptember 12, 2001. The upper panel depicts MISR 70◦ forward image of natural color reflectance forTerra orbit 9,237, prominently showing the smoke plume, and indicating four patches, for which ste-reo-height histograms were derived. The lower panel shows histograms of height generated using the60 and 70◦ forward MISR views, with 250 m vertical bin size and 1.1 km horizontal (pixel) resolution,for the four patches indicated in the upper panel. The stereo product vertical resolution is approxi-mately the size of the histogram bins. They demonstrate that the plume height is roughly 1250 m nearthe WTC. Points in the histograms at 2.5 km altitude and higher are mostly cumulus clouds within thepatches, whereas the points at 2 km in Patch 3 are probably part of the smoke plume.

is 275 m at all angles, and global data are routinely acquired at full resolution in 12channels, 1.1 km resolution in the others (see e.g. Kahn et al. [49], Moroney et al.[50], Muller et al. [51], and Kahn et al., Aerosol Source Characterization from Space-based Multi-angle Imaging, submitted manuscript). The WTC and other mid-latitudesites are viewed 1–2 times per week (see http://www-misr.jpl.nasa.gov for more detailsabout MISR).

The MISR contributes to knowledge of the global aerosol budget, providing tightconstraints on aerosol optical depth from well-calibrated spectral radiances mea-sured at precisely known air-mass factors ranging from one to three. The multi-angleobservations also sample a wide range of scattering angles (about 50–160◦ at mid-latitudes), offering additional constraints on particle shape, size distribution, andsingle-scattering albedo, particularly over dark, uniform surfaces such as the ocean(e.g., Kahn et al. [49]). In situations where a plume has discernable contrast featuresin the multi-angle images, such as near fire, dust, or volcanic aerosol source regions, astereo-matching technique automatically retrieves plume-top height [50]. The heightretrieval is performed both with and without MISR-derived wind correction. For theWTC case on September 12, the wind correction is very small, because the plume isoriented nearly normal to the plane of the multi-angle views.

Retrieval of plume-top and cloud-top heights make use of the stereoscopic natureof MISR data, and employs rapid pattern matching algorithms [51] to determinethe geometric parallax (horizontal displacement) of cloud and plume features due to

Environ Fluid Mech

their altitude above the surface. Photogrammetric calculations using accurate cam-era geometric models transform the derived parallaxes into cloud-top heights. Usingthe nadir and near-nadir cameras, as is done in generating MISR’s operational stereoproduct, the quantized accuracy of the resulting height field is ∼ 560 m. However, thinplumes do not produce sufficient image contrast at these angles for the pattern match-ing algorithms to work. This was addressed with special processing that used moreoblique angle images, which enhance the plume appearance relative to the surfacebackground. Using the 60 and 70◦ pair of angles, for example, results in a quantizedheight resolution of ∼ 250 m.

The upper panel in Fig. 3 depicts a natural color (RGB) image from MISR’s 70◦forward view for Terra orbit 9,237, showing the smoke plume prominently and indi-cating four patches, for which stereo-height histograms were derived. The image wasacquired at 1603 UTC (1203 EDT) at 275 m pixel resolution in the red band and at1.1 km in the green and blue; the red band data were used to sharpen the image asa whole to 275 m effective resolution. The near-vertical rise of the buoyant plumedirectly above the WTC, viewed obliquely by the MISR 70◦ forward-looking cameraas Terra flew southward, is revealed at high resolution, projected on the scene as anapparent south-trending column of smoke.

The lower panel in Fig. 3 shows histograms of height, with 250 m bin size and1.1 km horizontal (pixel) resolution, for the four patches indicated in the upper panelof Fig. 3. As noted above, the vertical resolution of the height retrieval is about 250 m,as these heights were generated using the 60 and 70◦ forward views, providing greatersensitivity to thin hazes than the standard MISR stereo height product. The resultsshow that the height of the plume top is roughly at 1,250 m near the WTC, and itspreads upward slightly, downwind. The points in the histograms at 2.5 km and higherare mostly cumulus clouds within the patches, although the points at 2 km, especiallythose in Patch 3, are probably part of the smoke plume. Thus, inference of verticalmixing downwind rests on the Patch 3 data. Thinning downwind is indicated by boththe small number of pixels, for which MISR stereo heights could retrieve in Patch4, and the widening and increased transparency of the plume itself in the image.The estimates of plume altitude derived from MISR are in good agreement with theground-based photos taken from Newark on September 12, that show the top of theplume rising vertically to an altitude about 1,000–1,500 m in the middle of the day.

Therefore, in the simulations aerosols were released in the atmospheric columnvolume with the 250 × 250 m base centered at the WTC, which roughly correspondsto the entire area of ground zero. The effective time-varying altitude of the volumesource was chosen to be 1500 m during the first 52 h (i.e., until 1400 EDT on Septem-ber 12). After 52, 72, and 96 h the altitude of the effective source decreased to 500,300, and 150 m respectively. The 150 m source was kept until the end of simulationson October 8. One-hundred Lagrangian particles per second were emitted randomlyand statistically uniformly in the volume of the source, with a unit total mass-releaserate of 1 kg/s. Below, the magnitude of the source is estimated by comparing observedand calculated concentrations.

3 Results

The weather on September 11–13, 2001 was clear, dry, with afternoon temperaturesreaching 80◦F. A cold front approached the NY metropolitan area on September14 from the north and passed by on September 15, bringing brisk northerly winds

Environ Fluid Mech

and scattered showers. The temperature dropped during September 14–16 by anaverage of 20◦F to the lower 60s, 5–8◦F below the climatological average. The mostintense precipitation fell on September 14, with total amounts reaching 1.9 in. OnSeptember 17 a high-pressure system held across the region, bringing sunny skiesuntil September 20. A weak front passed the region producing scattered showersof only 0.83 and 0.36 in/day on September 20 and 21, respectively. From September22 to 24 the temperature rose to the low 70s, exceeding the climatology average by3–7◦F. On September 25 a cold front approached and stabilized in the area for 4 days.The temperature decreased to 65◦F on September 29 and dropped further to 50◦Fon September 30 and October 1, which is 4–10◦F below the climatological average.The accompanying scattered showers on September 24, 25, 29, 30, and October 1were fairly weak, reaching only 0.3, 0.41, 0.04, 0.36, and 0.1 in/day, respectively. Dur-ing the first week of October, a high-pressure system brought sunny weather and astrong diurnal temperature cycle; the daily maximum exceeded 80◦F. A cold frontapproached from the north on October 6, reducing temperatures to the lower 50s andbringing weak scattered showers of 0.12 in/day. In general, the entire period was verydry. There were no significant precipitation events that could wash out aerosols fromthe atmosphere or off surfaces except for the showers that occurred on September 14.That precipitation was an important factor affecting aerosol lifetime, decreasing fireintensity and reducing aerosol emissions at ground zero. Further, it would wash awaymost of the outdoor re-suspendable dust.

Through the course of this study, the sensitivity of the results was tested with respectto several alternative model settings including coarse and fine resolution RAMS da-tabases (Land Elevation, Land Cover, and SST), different nudging strength, and timestep. The micrometeorological and tracer fields obtained with the model settings dis-cussed in Sect. 2.1 appear to be more consistent with observations than calculationsconducted with the different model configurations; it was assumed that these fieldswere superior and were therefore used in all simulations. Discussed in this sectionis the sensitivity of the simulated plume to variations in the driving meteorologicalfields, as a result of implementing the ASOS and NCEP data, with the model settingsfixed at those from Sect. 2.1. More specifically, a comparison is made of results fromruns when the initial and boundary conditions were constructed using only Eta data,Eta and ASOS data, and Eta, ASOS, and NCEP data, which is referred to as Eta,Eta + ASOS, and Eta + ASOS + NCEP, respectively. Also the sensitivity of the plumedispersion with respect to the release rate of Lagrangian particles was evaluated.Then a comparison of the simulated wind fields with the observations from ASOSstations was conducted, and the simulated concentrations were tested against PM2.5observations from the roofs of three Public Schools in NYC where data are availableevery hour. At the end of this section the plume transport on September 12, whichwas characterized by a very rapid change in plume direction, is discussed and thesimulated plume compared with with the satellite images.

3.1 Plume calculations using different meteorological fields

Figure 4 depicts relative plume concentrations, shown as percentages, normalizedto the maximum concentration at the WTC location in the lowest model layer ofGrid 3 (10 m altitude), calculated using different meteorological fields. Specifically,transport simulations were conducted using the RAMS fields calculated with initialand boundary conditions prepared using Eta fields (Fig. 4a), Eta and ASOS surface

Environ Fluid Mech

Fig. 4 (a) Simulated with Eta initialization, low-level tracer concentrations averaged for the 8-hperiod from 0800 to 1600 EDT on September 11, 2001, (b) Same as (a) but simulated with Eta + ASOSinitialization, (c) Same as (a) but simulated with Eta + ASOS + NCEP run initialization, (d) Same as(a) but simulated with Eta run initialization and 1,000 particle per second release rate

measurements (Fig. 4b), and Eta, ASOS, and NCEP surface station data (Fig. 4c).Simulations were also conducted releasing 1,000 Lagrangian particles per secondusing Eta initialization. This particle release rate was 10 times higher in magnitudethan that in the employed routine simulations and it effectively improved the plumediscrete spatial approximation during the entire run, compared to the run with a rel-ease rate of 100 particles per second (Fig. 4d). Concentrations were averaged for theinitial 8 h post collapse. During this period, aerosol from the WTC was transported

Environ Fluid Mech

to the south-southeast. All simulations produced similar results. In simulations withEta+ASOS and Eta+ASOS+NCEP fields, the direction of highest concentration rot-ated a little bit more to the east than in simulations with Eta fields only (Fig. 4a). Theplume calculated with the 1,000-particle-per-second release rate shows more spatialdispersion, but in general the results are close to the simulations with the 100-particle-per-second release rate. From this analysis, which shows weak dependence on drivingmeteorological fields and spatial approximation of plume, it can be concluded that theRAMS initialization using Eta and ASOS data and the 100-particle-per-second releaserate allow sufficiently accurate calculation of the concentration field for the initial postevent period. The root mean square error between the concentration field obtained incalculations with different initial conditions and particle release rate shown in Fig. 4did not exceed 10%. It is also shown that for the first 8 h following the attack theplume affected mostly lower Manhattan and northwest Brooklyn. In Brooklyn theconcentrations were less than 10% of the peak value seen over Manhattan.

3.2 Winds

ASOS provided the best wind observations available to test the accuracy of the sim-ulations. Therefore, in order to test the RAMS simulations, calculations initializedwith the Eta fields only were compared with (in this case, independent) ASOS windobservations (see Fig. 5). The comparisons were conducted at five locations: Teter-boro Airport (TEB), Newark Airport (EWR), NY Central Park, LaGuardia Airport(LGA), and JFK Airport (JFK). The stations are shown in Fig. 1 and their exactgeographic coordinates are reported in Fig. 5. The comparison was conducted forabout 4 days and data are presented in universal time (UTC = EDT + 5 = EST + 4).The simulated winds generally compare well with observations, both in magnitudeand direction for all the locations. The wind pattern at Central Park is most complexbecause it is affected by local circulation in central Manhattan.

The surface wind speed did not exceed approximately 5 m/s at all station loca-tions (Fig. 5). There were several distinct wind regimes. The wind was predominantlynorthwesterly from September 11 to midday on September 12. Then the wind blewpredominantly from the south on September 13. On September 14 the wind was fairlyvariable both in space and time, blowing predominantly southwest in the morning.The fast-moving cold front on September 14 brought north and then northeast windsalmost simultaneously at all stations.

A vertical solid line is drawn at 1600 UTC (1200 EDT) on September 12, whenthe wind direction changed dramatically and it was especially difficult to compare theplume position with observations. This issue is discussed further in Sect. 3.4. Windscalculated using Eta + ASOS data and Eta + ASOS + NCEP data (not shown) arenot substantially different from those using only Eta driving fields. This might beexpected from the analysis presented in Sect. 3.1. This confirms that Eta 3-D input ismostly important for the meteorology simulations. It also shows that RAMS with theselected setting configuration is capable of downscaling the wind field to the very fineresolution needed for transport calculations.

3.3 Concentrations

Plume evolution was calculated using HYPACT, driven by meteorological fields fromRAMS simulations, and initialized with Eta and ASOS data, as discussed in Sect. 2.2.

Environ Fluid Mech

Fig. 5 Simulated (with Eta initialization) and observed surface wind vectors for ASOS station loca-tions in the vicinity of the WTC shown in Fig. 1. Vertical solid line shows 1200 EDT, when the plumewas observed by Landsat and MISR blowing in the southwest direction. The horizontal axis showsuniversal time (UTC)

Environ Fluid Mech

To evaluate the transport simulations, PM2.5 concentration data that were routinelycollected on the roofs of public school buildings in NYC were used. These instrumentswere not compromised by the initial dust and smoke release caused by the collapseand continued to collect reliable data during the entire period of interest, beginning onSeptember 11, 2001. For comparison three school locations were selected where obser-vations were reported every hour: Public School (PS)-64 at 600 E 6th Street in LowerManhattan.; PS-199 at 3920 48th Avenue in Queens; and PS-274 at 800 BushwickAvenue in Brooklyn. These schools (except PS-274) were at a sufficient distance fromthe southeast sector of the major impact immediately after the collapse of the WTCbuildings so their sensors were not damaged by the initially intense PM release.

PS-64, located about 3 km northeast of the WTC, was the observation site closestto the WTC, and received the (relatively) highest level of PM. PS-199 is located inthe same northeast sector from the WTC at a distance of about 7–8 km. Therefore itis possible that the plume reached these two locations at about the same time. PS-274is south of PS 199, in the east-southeast sector from the WTC where plume charac-teristics and timing might be different from the above two school locations. However,PS-274 is separated from the WTC by about the same distance as PS-199. At thesedistances the plume was well organized and did not experience fluctuations related tonear-source processes.

Figure 6 depicts observed and simulated concentrations for the three school loca-tions between 1200 UTC (0800 EDT) on September 11 and 2400 UTC (2000 EDT)September 14. In the simulations, concentrations were sampled from the fourth modellayer at an altitude of 89 m. The exact geographic coordinates of the schools are givenin Fig. 6. Using the simulation and observation data, the concentrations were normal-ized to the maximum value of the peaks found at each location. Two key questionswere considered in analyzing the data: (a) Do the model outputs reproduce theobserved timing of the concentration spikes? and (b) How large should the aerosolrelease rate be in simulations in order to reproduce the observed aerosol concentra-tions in the plume? The second question is of great scientific and practical interestbecause the actual aerosol release rate was never measured. The WTC aerosol emis-sion source varied in magnitude and spatial distribution and released aerosols andgases at varying altitudes. Moreover, there were multiple additional sources, suchas transportation, industrial activities, long-distance transport from remote sources,and local re-suspension of deposited aerosol particles, not accounted for in the sim-ulations that produced background concentrations. Therefore an attempt was madeto estimate the aerosol source intensity using concentrations in the peaks that weresignificantly above background PM levels. The effect of aerosols on the flow was notconsidered, therefore transport is linear and concentrations are linearly proportionalto the magnitude of the source. This assumption could fail if aerosol optical depth washigh and its radiative heating/cooling effect on the hydrodynamic flow was significant.However, this was not the case for the WTC plume.

It must be mentioned that for all three locations major plume impact was wellestimated, showing almost perfect timing for the observed concentration peaks. ForSeptember 11, the simulations do not show any signal at the three locations. Theincrease of observed concentrations seen at PS-274, which is closest to the maindirection of the plume transport on this day, was not that high. The series of peakson September 12–14 is fairly similar at PS-64 and PS-199, and has a shape differentfrom that at PS-274. For PS-64 and PS-199 a comparison was made of simulated andobserved peak concentrations to avoid differences in the timing of the corresponding

Environ Fluid Mech

Fig. 6 Simulated with the Eta+ASOS initialization, normalized tracer concentrations at 89 m altitude(solid line) and normalized observed PM2.5 concentrations (open circles) at, (a) Public School (PS)-64in lower Manhattan, sampled at grid 3 resolution, (b) PS-199 in Queens, sampled at grid 2 resolution,(c) PS-274 in Brooklyn, sampled at grid 2 resolution. The horizontal axis shows universal time (UTC)

plume passings. The effective PM release rates on September 13 and 14 need to beabout 0.2 and 0.01 kg/s, respectively, to reproduce the observed concentrations. Alllocations gave similar estimates within a factor of 2.

These effective emission rates are representative of smoke produced by the fire atground zero. It must be emphasized that, using the above approach, only the peakaerosol production is estimated; the source could have yielded different emissionrates at other times. Recycled aerosol deposits could be another complication. Forexample, at the PS-274 site, higher concentrations observed on September 13 and 14,but not reproduced by the simulations, might be caused by resuspension of WTC dustbecause more material was deposited in this sector on September 11. In addition, theeffective plume altitude, as well as the aerosol release rate, are fairly variable and areonly roughly approximated in the simulations. The aerosol surface concentrations arealso sensitive to the turbulent structure of the boundary layer, since, they develop asa result of horizontal transport and vertical mixing from the core of the plume thatcould be (e.g., in case of high-elevated aerosol source for a fire) as high as 800–1,000 m.

Environ Fluid Mech

3.4 Transport directionality

In Sect. 3.3 it was shown that the model accurately calculates the timing of con-centration spikes at different locations, providing confidence in the simulated plumedirectionality. Here, analyses were conducted for what is probably the most complexperiod of plume evolution in the afternoon of September 12. The simulations weretested against visual Landsat imagery (Fig. 7), which provides a higher-resolution viewthan the MISR imagery in Fig. 3. Figure 7d shows the observed plume blowing towardthe southwest at 1530 UTC (1130 EDT) half an hour before the image in Fig. 3 wastaken. The plume is viewed in the nadir only by Landsat, so all layers containingaerosols are superposed. The plume spreads from the WTC along the Hudson Rivertoward the Bayonne Peninsula and Newark Bay. It turns out that it is very difficult toreproduce this plume position in simulations because of rapid wind direction changesat this time. The Landsat and MISR images catch the extreme southwest position ofthe plume, just after it rotated clockwise and was about to rotate back.

Figure 7a presents the position of the simulated plume at 1530 UTC (1130 EDT),when it almost reaches its extreme southwest position. (In these figures, concentra-tions are vertically averaged from 750 to 1,050 m, to mimic nadir satellite imagery,and are then normalized to the maximum value.) The range of altitudes was chosenin accordance with MISR estimates of the plume altitude (see histograms in Fig. 3).Figures 7b and 7c show the subsequent plume positions with a 30-min. time step.The simulated plume compares favorably with the Landsat image (Fig. 7d) and MISRimage (Fig. 3), and it reproduces almost perfectly the directionality of plume transport,(Fig. 7a, b and c).

As mentioned above, the wind on September 12 was variable, especially near thesurface (see Fig. 5), with a significant vertical sheer. Therefore, the position ofthe plume observed from space is very sensitive to the altitude of the upper partof the plume seen in nadir view from space. Because of the strong dependence ofwind on time and altitude, this case provides an ideal consistency test between simu-lations and satellite observations. The MISR estimates the altitude of the top of theplume during this period to be about 1,250 m with an uncertainty of about 250 m. Thesimulations show that the portions of the plume below 750 m and above 1,050 m donot rotate as far clockwise as in Fig. 7d, and move counterclockwise rapidly soon after1600 UTC (1200 EDT). This suggests that the core of the smoke layer was between750 and 1,050 m, which is within the error bars of the MISR estimates.

The ASOS wind observations in Fig. 5 confirm the rapid, near-surface wind changes.All stations show northeast wind before 1530–1600 UTC (1130–1200 EDT). Only atthe Central Park and the LaGuardia Airport stations are there short, sporadic periodsof easterly wind. At LaGuardia Airport, the model captures those easterly winds verywell. All stations then show a sudden change in wind direction, at about 1600 UTC(1200 EDT).

The simulations produce an accurate approximation of plume directionality andevolution. Consistent with the MISR plume-top altitude retrievals, the simulationsshow that most of the aerosol mass was likely transported within the 750–1,050 mlayer. This demonstrates important internal consistency between fine-scale atmo-spheric dynamics, transport calculations, and satellite observations. Nevertheless, thisanalysis also indicates that comparisons between simulations and observations shouldbe conducted with reasonable caution, especially for periods of strong changes in windmagnitude and direction.

Environ Fluid Mech

Fig. 7 Simulated with the Eta+ASOS initialization, vertically averaged from 600 to 1,000 m, andnormalized tracer concentrations on September 12, 2001 at (a) 1600 UTC (1200 EDT), (b) 1630UTC (1230 EDT), (c) 1700 UTC (1300 EDT), and (d) Landsat image at 1600 UTC (1200 EDT) onSeptember 12 showing the plume blowing southwest

3.5 Spatial-temporal distribution of aerosol concentration near the surface

Figure 8 shows evolution of the 8-h average plume concentration for the 96 h period,beginning at 0800 EDT on September 11, normalized as in Figs. 4 and 7 and usingthe same contour intervals and color scheme. In Fig. 8 local EDT times are used.

Environ Fluid Mech

Fig. 8 Simulated with the Eta + ASOS initialization, 8-h average normalized low-level tracer con-centrations for September 11–15 in the lower layer of the model at the altitude of 10 m. The contourintervals and the color scheme are the same as in Fig. 7

Environ Fluid Mech

During the first 24 h, the plume moved south-southeast, and was nearly confined tothis sector. It is not surprising that a weak PM2.5 concentration spike was detectedonly at PS-274, located in Brooklyn. The area of high-relative concentrations > 50%of the maximum value during this period spreads only about 3–5 km from the WTC,barely reaching the Brooklyn coast at Buttermilk Channel. More aerosol dispersionwas observed on September 12 between 0800–1600 EDT, corresponding to the rapidwind direction variations shown in Fig. 5. The plume rotated fairly rapidly, producingsmall PM2.5 spikes at all three school locations.

At 1600–2400 EDT on September 12, the plume first affected the north-to-north-east sector producing high concentrations at distances of 1–1.5 km from the WTC.It is interesting that one PM2.5 sample was taken by NY University (NYU) at 25thStreet and 1st Avenue approximately 1.0 km northeast of the WTC. Although, it wascollected from a long term sampler, the only time it was affected by the WTC plumewas 1600–2400 EDT on September 12 as is seen in Fig. 8. It was assumed that theinstrument was affected by the plume for 1/2 h and then clogged and stopped sam-pling. Based upon the lack of prior contact with the plume, it was estimated thatthe ambient concentration of fine particles detected during that half-hour period was≈ 500 to 600 µg/m3 (L.C. Chen, personal communication). However, the modelingresults in Fig. 8 suggest that the plume was over the area for a few hours. Thus, theestimate by Chen could be accurate, if the sampler clogged in the first 1/2 hour; thevalues could be lower depending upon how long the sampler remained operationalover the 3-hour period.

On the morning of September 13, the wind moved the plume to the east, dispersingPM in a wide southeast–northeast sector. This transport produces significant PM2.5spikes at PS-64, PS-199 and PS-274; the signal arrived a couple of hours earlier atPS-274 than at the other two locations.

Later on September 13, and on the morning of September 14, the plume blewin the east, north and south directions, producing concentration peaks at all threeschools between about 0600–1000 UTC (0200–0600 EDT). At the end of the dayon September 14, the wind again blew to the east and southeast, producing detect-able signals at PS-199 and PS-274. On the morning of September 15, as shown inFig. 8, the plume blew to the south. Figures for the rest of this period are not shownbecause the magnitude of the source decreased significantly after the September14 rain.

The period documented shows that the simulated plume was distributed fairly con-sistently with the ground-based PM2.5 observations that were made a few kilometersfrom the WTC site. The predominant transport during this entire period was to thesouth-southeast (about 75% probability). There were two episodes of northward aer-osol transport, affecting upper Manhattan on September 12 and 13, three periodswhen the plume blew to the east and northeast, affecting Queens and the Bronx, andone period on September 12 when the plume blew to NJ (see Fig. 8).

The results discussed in this section are consistent with the plume dispersion anal-ysis conducted using the CALMET-CALPUFF modeling system in Gilliam et al. [3].Both CALMET–CALPUFF and RAMS/HYPACT employed a constant altitude aer-osol source of 50 m for the entire simulation period. However Gilliam et al. presentedcalculations performed with coarser horizontal resolution (500×500 m) and employeda different turbulent closure which caused effectively more rapid dispersion of theplume. For example relative concentrations over Brooklyn obtained in Gilliam et al.

Environ Fluid Mech

were about one order of magnitude lower than the RAMS/HYPACT simulations ofthe present study.

4 Summary and Conclusions

This study demonstrates that the combination of numerical modeling and ground-and space-based observations allow reconstruction of the aerosol plume from theWTC site for the entire period of interest. Micrometeorological and tracer transportmodeling provides a framework for utilizing available observations. This increases thecredibility of the simulations, and allows rough estimation of the effective emissionrates of PM observed in the urban atmosphere during this event.

The results of this work can be summarized as follows:

(1) High-quality, fine-scale micrometeorological fields produced as part of this studywere consistent with the observations for the entire period of interest follow-ing the September 11, 2001 event. The PM transport compares favorably withground-based observations and satellite images. Simulations give reliable infor-mation about the spatial distribution of pollutants and the timing of the maxi-mum near-surface concentrations at different locations.

(2) On September 12, the southeast transport was very episodic, making compari-sons with observations difficult. However, the model accurately reproduces theplume directionality on this day. Simulated PM concentrations at the surface, andplume directionality, are sensitive to the altitude of the convective column devel-oped by the fire at ground zero. MISR estimated the altitude of the plume onSeptember 12 to be about 1,000–1,250 m at 1600 UTC (1200 EDT). The MISRretrievals allow better calibration of the surface estimates of plume altitude.The vertical structure of the fine-scale wind field also appears to be consistentwith the MISR altitude estimates. This example shows an important role thatquantitative satellite retrievals can play in monitoring fine-scale processes duringdisasters, when ground-based observation networks are suppressed or destroyedby catastrophic conditions.

(3) The simulated fine-scale meteorological fields from mesoscale model simula-tions account for the larger-scale meteorological structures and could be used asboundary conditions for CFD calculations.

(4) During the first 3 days, when the magnitude of aerosol source was largest, aero-sols were transported predominantly to the south-southeast, affecting lowerManhattan and Brooklyn. However, over Brooklyn the aerosol concentrationswere at least an order of magnitude lower than would be expected in the vicinityof the WTC.

(5) It was found that the effective peak fine PM release rates from the fire at groundzero have to be in the range of 10–200 g/s to be consistent with PM2.5 concentra-tions observed in Manhattan, Brooklyn, and Queens during the 3 days followingthe collapse of the WTC. However, this estimate implicitly included contribu-tions from secondary sources (such as re-suspension of particles) that were notaccounted for in the simulations. Presumably, their contribution was relativelysmall in the 5 km vicinity of the source during the considered period when theprimary aerosol source from the fire at ground zero was relatively strong. The

Environ Fluid Mech

September 14 rain reduced the fire, and removed aerosols deposited after thecollapse of the WTC.

The present study estimated from observations rather than simulating directly the aer-osol emission processes that resulted from the collapse of the main WTC structuresand the fire at ground zero. These processes define the amount of material releasedand the way in which it was initially distributed vertically. At fine spatial scales theseprocesses are controlled by meteorological conditions, such as the vertical tempera-ture gradient, wind shear, and boundary layer turbulence; also they are dependant onthe detailed hydrodynamic flow within street canyons surrounding ground zero.

To reduce uncertainties in future studies, it would be interesting to simulate emis-sions and transport interactively, and to account realistically for aerosol source var-iability as a function of the micrometeorological environment. Having interactiveaerosol release modeling capability would also allow one to calculate aerosol micro-physics explicitly and estimate its effect on aerosol transport and deposition patterns.

Acknowledgements This work was sponsored by USEPA grant CR827033. Additional support wasprovided by a supplement to the NIEHS EOHSI center grant P30 ES05022. We thank Praveen Amarof NESCAUM for providing plume photographs; Jennifer Bosch of the Rutgers University Instituteof Marine and Coastal Sciences for providing AVHRR SST retrievals; the developers of RAMSand HYPACT, Bob Walko and Craig Tremback, for consulting on RAMS/HYPACT modifications;and Linda Everett of EOHSI for help with editing and manuscript preparation. Georgiy Stenchikovwas partially supported by NJDEP grant SR04-048. The research of David Diner and Ralph Kahnis supported, in part, by the MISR project at JPL, under contract with NASA. Ralph Kahn is alsosupported by the NASA Climate and Radiation Research & Analysis program, under H. Maring. Wethank Catherine Moroney of JPL for the special stereo processing of the MISR WTC data.

References

1. Lioy PJ, Weisel CP, Millette JR, Eisenreich S, Vallero D, Offenberg J, Buckley B, Turpin B, ZhongMH, Cohen MD, Prophete C, Yang I, Stiles R, Chee G, Johnson W, Porcja R, Alimokhtari S, HaleRC, Weschler C, Chen LC (2002) Characterization of the dust/smoke aerosol that settled east ofthe world trade center (WTC) in Lower Manhattan after the collapse of the WTC 11 September2001. Environ Health Perspect 110:703–714

2. Landrigan PJ, Lioy PJ, Thurston G, Berkowitz G, Chen LC, Chillrud SN, Gavett SH, Georgopo-ulos PG, Geyh AS, Levin S, Perera F, Rappaport SM, Small C (2004) Health and environmentalconsequences of the WTC disaster. Environ Health Perspect 112:731–739

3. Huber A, Georgopoulos P, Gilliam R, Stenchikov G, Wang S-W, Kelly B, Feingersh H (2004)Modeling air pollution from the collapse of the WTC and assessing the potential impacts onhuman exposures. Environ Manage. February 2004:35–40

4. USEPA (2003) EPA Acid Rain Program 2002 Progress Report EPA-430-R-03-011. Washington,D. C: US Environmental Protection Agency. Clean Air Markets Division, Office of Air andRadiation

5. Bornstein RD, Johnson DS (1977) Urban rural wind velocity differences. Atmos Environ11:597–604

6. Bornstein RD, Thunis P, Schayes G (1994) Observation and simulation of urban-topography bar-rier effects on boundary layer structure using the three-dimensional TVM/URBMET model. In:(Gryning, S-E, Millán, MM, eds) Air pollution and its Application X. Plenum Press, New Yorkpp 101–108

7. Dickerson RR, Kondragunta S, Stenchikov G, Civerolo KL, Doddridge BG, Holben BN (1997)The impact of aerosols on solar ultraviolet radiation and photochemical smog. Science 278:827–830

8. Malm WC, Schichtel BA, Ames RB, Gebhart KA (2002) A 10-year spatial and temporal trend ofsulfate across the United States. J Geophys Res 107:4627

Environ Fluid Mech

9. Malm WC, Sisler JF, Huffman D, Eldred RA, Cahill TA (1994) Spatial and seasonal trends inparticle concentration and optical extinction in the United States. J Geophys Res-Atmos 99:1347–1370

10. Marufu LT, Taubman BF, Bloomer B, Piety CA, Doddridge BG, Stehr JW, Dickerson RR (2004)The 2003 North American electrical blackout: An accidental experiment in atmospheric chemistry.Geophys Res Lett 31:L13106

11. Ryan WF, Doddridge BG, Dickerson RR, Morales RM, Hallock KA, Roberts PT, BlumenthalDL, Anderson JA (1998) Pollutant transport during a regional O-3 episode in the mid-Atlanticstates. J Air Waste Manage Assoc 48:786–797

12. Taubman BF, Marufu LT, Piety CA, Doddridge BG, Stehr JW, Dickerson RR (2004) Airbornecharacterization of the chemical, optical, and meteorological properties, and origins of a combinedozone-haze episode over the eastern United States. J Atmos Sci 61:1781–1793

13. Klemp JB, Wilhelmson RB (1978) The simulation of three-dimensional convective storm dynam-ics. J Atmos Sci 35:1070–1096

14. Rehm RG, Pitts WM, Baum HR, Evans DD, Prasad K, McGrattan KB, Forney GP (2003) InitialModel for Fires in the world trade center Towers. In: Evans DD (ed) Fire Safety Science—Proceedings of the Seventh International Symposium. International Association for Fire SafetyScience, Boston, MA, pp 25–40

15. Cotton WR, Pielke RA, Walko RL, Liston GE, Tremback CJ, Jiang H, McAnelly RL, HarringtonJY, Nicholls ME, Carrio GG, McFadden JP (2003) RAMS 2001: Current status and future direc-tions. Meteorol Atmos Phys 82:5–29

16. Pielke RA, Cotton WR, Walko RL, Tremback CJ, Lyons WA, Crasso LD, Nicholls ME,Moran MD, Wesley DA, Lee TJ, Copeland JH (1992) A comprehensive meteorological modelingsystem—RAMS. Meteorol Atmos Phys 49:69–91

17. Tripoli GJ, Cotton WR (1982) The Colorado state university three-dimensional cloud/mesoscalemodel—1982. part I: general theoretical framework and sensitivity experiments. J de RecherchesAtmos 16:185–220

18. Gal-Chen T, Somerville RCJ (1975) On the use of a coordinate transformation for the solutionof the Navier-Stokes equations. J Comput Phys 17:209–228

19. Davies HC (1976) A lateral boundary formulation for multi-level prediction models. Q J RoyMeteorol Soc 102:405–418

20. Clark TL, Farley RD (1984) Severe downslope windstorm calculations in two and three spatialdimensions using anelastic interactive grid nesting: a possible mechanism for gustiness. J AtmosSci 41:329–350

21. Mellor GL, Yamada T (1974) A hierarchy of turbulence closure models for planetary boundarylayers. J Atmos Sci 31:1791–1806

22. Mellor GL, Yamada T (1982) Development of a turbulence closure model for geophysical fluidproblems. Rev Geophys Space Phys 20:851–875

23. Smagorinsky J (1963) General circulation experiments with the primitive equations. Part I: thebasic experiment. Mon Weather Rev 91:99–164

24. Harrington JY (1997) The effects of radiative and microphysical processes on simulated warmand transition season Arctic stratus. Ph.D. Dissertation, Department of Atmospheric Science,Colorado State University

25. Kuo HL (1974) Further studies of the parameterization of the influence of cumulus convectionon large-scale flow. J Atmos Sci 31:1232–1240

26. Fritsch JM, Chappell CF (1980) Numerical prediction of convectively driven mesoscale pressuresystems. part I: convective parameterization. J Atmos Sci 37:1722–1733

27. Tremback CJ (1990) Numerical simulation of a mesoscale convective complex: Model develop-ment and numerical results. Ph.D. Dissertation, Department of Atmospheric Science, ColoradoState University

28. Kain JS, Fritsch JM (1990) A one-dimensional entraining/detraining plume model and its appli-cation in convective parameterization. J Atmos Sci 47:2784–2802

29. Kain JS, Fritsch JM (1993) Convective parameterization for mesoscale models: The Kain-Frit-sch scheme. In: Emanuel KA, Raymond DJ, (eds) The representation of cumulus convection innumerical models.American Meteorological Society, Boston, MA, pp 165–170

30. Miguez-Macho G, Stenchikov GL, Robock A (2005) Regional climate simulations over NorthAmerica: interaction of local processes with improved large-scale flow. J Clim 18:1227–1246

31. Tripoli GJ, Cotton WR (1980) A numerical investigation of several factors contributing to theobserved variable intensity of deep convection over south Florida. J Appl Meteorol 19:1037–1063

32. Cotton WR, Stephens MA, Nehrkorn T, Tripoli GJ (1982) The Colorado State University three-dimensional cloud/mesoscale model 1982—part II: an ice phase parameterization. J de RecherchesAtmos 16:295–320

Environ Fluid Mech

33. Walko RL, Band LE, Baron J, Kittel TGF, Lammers R, Lee TJ, Ojima D, Pielke RA, Taylor C,Tague C, Tremback CJ, Vidale PL (2000) Coupled atmosphere-biophysics-hydrology models forenvironmental modeling. J Appl Meteorol 39:931–944

34. Carmichael GR, Calori G, Hayami H, Uno I, Cho SY, Engardt M, Kim SB, Ichikawa Y, Ikeda Y,Woo JH, Ueda H, Amann M (2002) The MICS-Asia study: model intercomparison of long-rangetransport and sulfur deposition in east Asia. Atmos Environ 36:175–199

35. Varinou M, Kallos G, Tsiligiridis G, Sistla G (1999) The role of anthropogenic and biogenic emis-sions on tropospheric ozone formation over Greece. Phys Chem Earth Part C-Solar-TerrestialPlanet Sci 24:507–513

36. Fast JD, Doran JC, Shaw WJ, Coulter RL, Martin TJ (2000) The evolution of the boundary layerand its effect on air chemistry in the Phoenix area. J Geophys Res-Atmos 105:22833–22848

37. Miguez-Macho G, Stenchikov GL, Robock A (2004) Spectral nudging to eliminate the effects ofdomain position and geometry in regional climate model simulations. J Geophys Res-Atmos 109:D13104

38. Reynolds RW, Rayner NA, Smith TM, Stokes DC, Wang W (2002) An improved in situ andsatellite SST analysis for climate. J Clim 15:1609–1625

39. Bernstein RL (1982) Sea surface temperature estimation using the NOAA 6 satellite advancedvery high resolution radiometer. J Geophys Res 87:9455–9465

40. Mesinger F, Janjic ZI, Nickovic S, Gavrilov D, Deaven DG (1988) The step-mountain coordinate:model description and performance for cases of Alpine lee cyclogenesis and for a case of anAppalachian redevelopment. Mon Weather Rev 116:1493–1518

41. Rogers E, Black TL, Deaven DG, DiMego GJ, Zhao QY, Baldwin M, Junker NW, Lin Y (1996)Changes to the operational “early” eta analysis/forecast system at the national centers for envi-ronmental prediction. Weather Forecasting 11:391–413

42. Boughton BA, Delaurentis JM, Dunn WE (1987) A stochastic model of particle dispersion in theatmosphere. Bound-Lay Meteorol 40:147–163

43. Oktay SD, Brabander DJ, Smith JP, Kada J, Bullen T, Olsen CR (2003) WTC Geochemicalfingerprint recorded in New York harbor sediments. EOS Trans Am Geophys Union 84:21

44. Monin AS (1959) On the boundary condition on the earth surface for diffusing pollution. AdvGeophys 6:435–436

45. Huber A, Freeman M, Spencer R, Bell B, Kuehlert K, Schwartz W (2005) Applications of CFDsimulations of pollutant transport and dispersion within ambient urban building environments:including homeland security. In: Annual Conference of the Air & Waste Management Associa-tion, Minneapolis, MN, June 21–24, 2005

46. Lioy PJ, Georgopoulos PG (2006) The anatomy of the exposures that occurred around the worldtrade center site: 9–11 and beyond. New York Academy of Sciences In Press

47. Lioy PJ, Weisel C, Georgopoulos PG (2005) An overview of the environmental conditions andhuman exposures that occurred post 9–11 (Chap 2). In: Gaffney JS, Marley NA (eds) UrbanAerosols and Their Impacts: Lessons Learned from the World Trade Center Tragedy. AmericanChemical Society, Washington, DC

48. Diner DJ, Beckert JC, Reilly TH, Bruegge CJ, Conel JE, Kahn RA, Martonchik JV, AckermanTP, Davies R, Gerstl SAW, Gordon HR, Muller JP, Myneni RB, Sellers PJ, Pinty B, Verstraete MM(1998) Multi-angle imaging spectroradiometer (MISR)—instrument description and experimentoverview. IEEE Trans Geosci Remote Sens 36:1072–1087

49. Kahn R, Banerjee P, McDonald D (2001) Sensitivity of multiangle imaging to natural mixtures ofaerosols over ocean. J Geophys Res-Atmos 106:18219–18238

50. Moroney C, Davies R, Muller JP (2002) Operational retrieval of cloud-top heights using MISRdata. IEEE Trans Geosci Remote Sens 40:1532–1540

51. Muller JP, Mandanayake A, Moroney C, Davies R, Diner DJ, Paradise S (2002) MISR stereoscopicimage matchers: techniques and results. IEEE Trans Geosci Remote Sens 40:1547–1559