Environ Fluid Mech
DOI 10.1007/s10652-006-9001-8
O R I G I NA L A RT I C L E
Multiscale plume transport from the collapse
of 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 levels of airborne contaminants in New York City and nearby areas on September 11,
2001 through December, 2001. This catastrophic event revealed the vulnerability of
the urban environment, and the inability of many existing air monitoring systems to
operate efficiently in a crisis. The contaminants released circulated within the street
canyons, but were also lifted above the urban canopy and transported over large distances, reflecting the fact that pollutant transport affects multiple scales, from single
buildings through city blocks to mesoscales. In this study, ground-and space-based
observations were combined with numerical weather forecast fields to initialize finescale numerical simulations. The effort is aimed at reconstructing pollutant dispersion
from the WTC in New York City to surrounding areas, to provide means for eventually evaluating its effect on population and environment. Atmospheric dynamics were
calculated with the multi-grid Regional Atmospheric Modeling System (RAMS), covering scales from 250 m to 300 km and contaminant transport was studied using the
Hybrid Particle and Concentration Transport (HYPACT) model that accepts RAMS
meteorological output. The RAMS/HYPACT results were tested against PM2.5 observations 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, USA
e-mail: [email protected]
N. Lahoti · P. J. Lioy · P. G. Georgopoulos
Department of Environmental and Occupational Medicine, UMDNJ—R.W. Johnson Medical
School, Piscataway, NJ 08854, USA
N. Lahoti · P. J. Lioy · P. G. Georgopoulos
Environmental & Occupational Health Sciences Institute, UMDNJ—R.W. Johnson Medical
School & Rutgers University, Piscataway, NJ 08854, USA
D. J. Diner · R. Kahn
Jet 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 well
as plume directionality. By comparing calculated and observed concentrations, the
effective magnitude of the aerosol source was estimated. The simulated pollutant
distributions are being used to characterize levels of human exposure and associated
environmental health impacts.
Keywords Aerosol plume · Particulate matter · Transport · Urban pollution ·
Regional Atmospheric Modeling System · Hybrid Particle and Concentration
Transport Model · Multi-angle Imaging SpectroRadiometer · World Trade Center ·
9/11 · Terrorist attack
1 Introduction
This study considers the transport of airborne contaminants (mostly particulate matter [PM] or aerosols) produced by the collapse of the world trade center (WTC) in
New York City (NYC) on September 11, 2001, and by the subsequent burning of
the remaining materials. The massive release of aerosols and gases on September
11 affected numerous residents and commuters in the surrounding New York/New
Jersey (NY/NJ) area. The continued threat of terrorist attacks on major cities raises
a new issue: developing a better understanding of the ambient exposures and associated 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 dispersion
in NYC and surrounding areas using available ground- and space-based observations
and numerical modeling to better characterize its environmental and health effects.
The north and south WTC buildings were set on fire by terrorist attacks at 0846
EDT and 0903 EDT, respectively, on September 11, 2001. The collapse of the WTC
South Tower at 0959 EDT followed by the crash of the North Tower at 1029 EDT
instantaneously produced vast amounts of coarse and fine airborne particles that
spread upward and into the streets of southern Manhattan. This initially produced
an intensive but relatively short-term particulate mass and gaseous release into the
urban atmosphere. Materials were deposited on roofs, streets and other flat surfaces
and were re-suspended later by the wind, contributing to the overall airborne contamination levels from September 11 through September 13. The remains of the
WTC complex, covering a 16-acre area known as ground zero, burned with varying
degrees of intensity until September 14, occasionally reaching temperatures exceeding 1, 000◦ C. After September 14 the fire began to diminish due to rain. The fire at
ground zero produced a continuous source of hazardous gases and aerosols for an
extended period of time, which were dispersed in NYC and the surrounding areas.
A detailed spatial and temporal evaluation of the airborne contaminant distribution is needed to fully understand the environmental and health impacts of the
WTC’s collapse. However, the existing ground-based observation networks (both for
meteorological characteristics and particulate matter) are fairly sparse for this purpose, even in the NY/NJ metropolitan area. Many monitoring stations in the vicinity
of the WTC did not operate properly, as they were completely plugged by large
amounts of dust immediately after the collapse, or they were unavailable because
of the short-term nature of the initial releases combined with the loss of electricity.
Quantitative, satellite-based measurements were limited in temporal coverage. As a
result, many important characteristics of the dispersed pollution field could not be
easily determined to help understand the details of associated human exposures. In
Environ Fluid Mech
the current study, numerical modeling of micrometeorological fields and PM transport was combined with available observations to reconstruct plume behavior as
realistically 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-recognized
health issue [4]. This region is one of the most densely populated in the US. Among
the sources regularly contributing to atmospheric aerosol loading are local industries
and utility generation, motor vehicle emissions, residential cooking and heating, dust
raised from disturbed soils, and marine aerosol production over the coastal waters.
In addition, long-range transport of emissions from the industrialized Midwest and
Baltimore–Washington areas contribute to the overall pollution level [5–12]. However, the unexpected nature of the catastrophic event on September 11, 2001 prevented the collection of adequate amounts of quantitative information necessary to
establish risk.
Transport and deposition of atmospheric tracers is highly dependent on local circulation, turbulent mixing in the boundary layer, terrain, and precipitation. Using
meteorological fields with a coarse spatial resolution often causes uncertainties in calculations of contaminant distribution. Unfortunately, fine-scale meteorological fields
are not available from observations, and operational forecast models provide meteorological fields with spatial sampling that is not sufficient for high-resolution transport calculations. In this study, the Regional Atmospheric Modeling System (RAMS)
(http://www.atmet.com/html/docs/documentation.shtml) was employed to downscale
the analysis fields from the Eta Weather Prediction Model [13] conducted with spatial
resolution of approximately 32 km. The RAMS databases of land elevation, vegetation
cover, and sea surface temperature were improved to account for fine-scale effects of
the land-surface boundary conditions and sea surface temperature. The downscaled
meteorological fields were used in the Hybrid Particle and Concentration Transport model (HYPACT) (http://www.atmet.com/html/docs/documentation.shtml) for
the fine grid transport and deposition calculations. The HYPACT uses RAMS meteorological output for calculating aerosol transport from localized sources combining
Lagrangian and Eulerian approaches. RAMS is a comprehensive mesoscale meteorological model, which is not fully capable of simulating the flow within the city’s street
canyons. However, it can accurately calculate the flow above the buildings, linking it
to larger-scale meteorological structures. To account for the effects of buildings on
the flow in the boundary layer, the surface roughness over Manhattan was increased
up 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 conducted in three nested domains (Fig. 1). The largest domain has a regional scale of
300 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 resolutions. The collapse of the WTC towers and the fire at ground zero were not explicitly
described to define emissions of aerosols and gases. More detailed computational fluid
dynamics (CFD) simulations need to be conducted to calculate those processes and to
obtain characteristics of air flow in the street canyons [3, 14]. However, CFD simulations require realistic lateral and upper boundary conditions that can be obtained only
from fine-scale meteorological calculations like those conducted using RAMS. This
study relied on available observational data to evaluate the time-dependent height of
the convective cell generated by the fire at ground zero, and quantified the magnitude
of the aerosol emissions source from the comparison of the simulated and available
observed 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 show
the location of the ASOS and buoy stations. Land elevation is shown with black contours. Land
cover 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 AVHRR
multi-channel observations, is shown by red contours
Environ Fluid Mech
PM distribution patterns over NYC, NJ, and NY were tested against available observations that included satellite retrievals, surface meteorological observations, and
available PM2.5 measurements. It was found that the simulations compare favorably
with observations and allow effective reconstruction of the plume evolution.
This article is organized as follows: Sect. 2, describes the modeling approach and
the simulation setup; Sect. 3 briefly describes meteorological conditions, discusses
results, and provides evaluation of the sensitivity of the results to the parameters and
initialization. Results are summarized in Sect. 4.
2 Methodology
To conduct multiscale atmospheric transport calculations, micrometeorological fields
need to be calculated with sufficient accuracy and spatial resolution. To estimate
time-varying human exposures it is also necessary to simulate the distribution of airborne contaminants with a spatial resolution of, at minimum, a city block. The routine
Eta model forecast, that provides the best available meteorological fields, resides at
the National Centers for Environmental Prediction (NCEP) and has a spatial resolution of about 32 km. Therefore, dynamically downscaling the Eta fields using RAMS
and additional available observations was required. The micrometeorological fields
obtained 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 bulk
cloud microphysics, and surface interaction parameterizations [15, 16]. The governing equations are approximated using the hybrid implicit-in-the-vertical time-split
difference scheme of Tripoli and Cotton [17]. RAMS predicts the 3-D fields of three
velocity components, temperature, water vapor mixing ratio, pressure, sub-grid-scale
turbulent kinetic energy, and several types of cloud hydrometeors including cloud
water, ice, graupel, and snow.
The horizontal grid uses a rotated polar-stereographic projection. In the vertical
direction, RAMS employs a sigma-Z terrain-following coordinate system [18]. Grid
nesting is used in RAMS to provide high-spatial resolution in selected areas, while
covering a large domain at lower resolution. Therefore, effects of large-scale circulation patterns can be transferred to an internal fine resolution region. A nested grid
occupies a region within the computational domain of its coarser parent grid. For
the external domain, lateral boundary conditions are applied by exponential relaxing
(nudging) the calculated fields toward the flow obtained from the forecast model in
the grid-belt along the lateral boundaries [19]. The relaxation coefficient follows a
parabolic function of the distance from the boundary and is constant in height. For
the internal domains the two-way interactions between nested grids are calculated
following Clark and Farley [20].
Various parameterization modules were available for most physical processes,
including radiation, turbulence, and land/atmosphere interaction. As vertical and horizontal resolutions are relatively different in this study, vertical turbulent eddy mixing
was parameterized using the 2.5 level scheme of Mellor and Yamada [21, 22] based on
a prognostic equation for turbulent kinetic energy. Horizontal turbulent mixing was
Environ Fluid Mech
calculated using turbulent diffusion coefficients calculated from the tensor of deformation [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 parameterizations are standard features of RAMS [27]. A modified version of the Kain and
Fritsch convective scheme [28, 29] was recently implemented in RAMS [30]. However, for cloud-resolving calculations, as in this study, RAMS does not require any
convective parameterization.
The cloud microphysics scheme is based on Tripoli and Cotton [17, 31] and Cotton
et al. [32]. This scheme consists of a set of conservation equations for water vapor
and six hydrometeor types: cloud droplets, raindrops, pristine ice, snow, graupel, and
aggregates. Their tendencies are affected by advection, turbulence, and microphysical
transformations in size distribution and from one class to another.
Calculation of land-atmosphere interaction is based on the Land EcosystemAtmosphere Feedback (LEAF-2) model [33], with 12 soil textural classes and 18
vegetation types. LEAF-2 predicts soil temperature and water content, snow cover,
vegetation, and canopy air as well as turbulent and radiative exchanges between these
components. LEAF-2 uses a mosaic approach where the grid cells are subdivided into
smaller portions or “patches” corresponding to different surface characteristics occurring in the area covered by the grid cell.
RAMS has been tested in numerous applications for atmospheric chemistry and
air pollution, including recent studies of the sulfur cycle and acid deposition in East
Asia [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 for
climate downscaling over the US [30, 37].
The triple-nested domain used in the present simulations, centered at the coordinates 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 projection
axes at 40.783◦ N and 73.967◦ W. The parent domain is necessary to accommodate
mesoscale structures to be downscaled to two smaller nested domains centered at the
same central point, covering areas of 54 × 54 and 10.5 × 10.5 km, respectively. Going
forward, the three domains will be referred to as grids 1, 2, and 3. The spatial resolutions of the three grids arre 4 × 4, 1 × 1, and 0.25 × 0.25 km and the number of grid
points 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 at
the top of the domain, at an altitude of 16 km.
The original RAMS land elevation and vegetation cover data sets are of 1 km
resolution, which is sufficient to calculate mesoscale circulation but is not adequate
for the needs of the present study. To conduct very fine-resolution simulations in
the metropolitan area it was necessary to improve the model databases. First the
high-resolution National Land Cover Dataset (NLCD) and National Elevation Data
(NED) from the United States Geological Survey (USGS) were adopted. NED, a
raster product, is available on the Internet at http://edcwww.cr.usgs.gov/doc/edchome/ndcdb/ndcdb.html. NED has a resolution of 1 arc-second or about 30 m for the
US. A visual basic software application was developed to read pixel values and
convert them to a digital data file. Land elevation is shown as black contours in
Fig. 1.
NLCD is a multi-layer and multi-source database that contains a 30 m
resolution, 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 digital
data 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 produced. In Fig. 1 vegetation classes are distinguished by color.
The original RAMS Sea Surface Temperature (SST) is based on the climatologically
averaged monthly mean 1 × 1◦ resolution data set [38]. However, fine-scale pollutant
transport can be affected by sea breezes initiated by the actual land/sea temperature
contrast; therefore, the simulations in this study used real-time SST, providing better
spatial and temporal resolution for the NY metropolitan area. For this purpose 1 km
resolution, multi-channel, Advanced Very High Resolution Radiometer (AVHRR)
satellite retrievals [39] were acquired from the Marine Remote Sensing Laboratory
of the Rutgers University Institute for Marine and Coastal Sciences. This data set had
to 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 September 11–14, 2001 is shown as red contours over a blue background corresponding
to the “ocean–lake–river–stream” surface classification group.
The terrain in Fig. 1 is fairly flat, not exceeding 400 m in elevation. The fine-scale
features were degraded on the parent grid in Fig. 1; nevertheless the 200 m highnarrow Palisades Cliff on the west side of the Hudson River, northwest of NYC, is
well captured and the coastline is well approximated. The dominant land cover type
is urban, with small intrusions of grassland, marsh, and trees. The AVHRR SST for
September 11–14, 2001 shows warm areas of 297–298◦ K related to the Gulf stream
path. The colder waters of 296◦ K and below are transported southward from the Labrador Sea, along Long Island and the NJ coast. The ocean temperature in the Gulf
Stream region is fairly patchy, but becomes smoother near the coast.
The AVHRR SSTs were tested with buoy observations available from the National
Data Buoy Center (http://www.ndbc.noaa.gov/to_station.shtml). The two stations
closest 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 composited
SST, sampled at 40.46◦ N, 73.5◦ W along the NY coast and shown as a solid curve without marks, with the Ambrose Light (closed circles) and Long Island (open circles)
buoy stations. The AVHRR SST, sampled between the stations, compares favorably
with the station observations, catching all SST changes during this period. The buoy
hourly output shows more high-frequency variations, but the three-day average AVHRR composites show fairly accurately that SST decreases from September 11 to
September 17, and then stabilizes at about 294.5◦ K, and at the end of September
drops again to 291◦ K. The SST change during the first week following the WTC collapse is most important, as it drove breeze circulation during the period when the
emissions were most intensive and the plume was especially dense.
The meteorology calculations depended on the initial and boundary conditions
that were developed using the objective analysis package within RAMS. These objectively analyzed fields are calculated from the three-hourly Eta model operational
analysis [40, 41]. The Eta Model data were provided by the National Center for
Atmospheric Research (NCAR) in gridded binary (GRIB) format, on a horizontal grid with a spatial resolution of 32 km. They included surface pressure, surface
elevation, and 3-D fields of pressure, temperature, water vapor mixing ratio, and horizontal wind components at 26 pressure levels, for the entire US. RAMS is able to
combine 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 retrieval
at 40.46◦ N, 73.5◦ W (solid line) and hourly observations from the long island (40.25◦ N, 73.17◦ W) and
Ambrose Light (40.46◦ N, 73.83◦ W) buoy stations near the NJ coast (open and filled circles, respectively)
fields could be enhanced by surface station data from NCEP and Automated Surface
Observation Stations (ASOS) available from the National Climate Data Center
(NCDC) (http://www4.ncdc.noaa.gov/cgi-win/wwcgi.dll?wwdi∼ASOSPhotos). The
five ASOS stations located closest to the WTC were used in the present study, and
are 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 laboratory on Long Island. Therefore, upper air observations could not be used in the
analysis.
The objectively analyzed 3-hourly fields helped constrain the flow near the boundaries of the grid 1 domain using relaxation type boundary conditions [19] with an
efolding relaxation time of 30 min. at the 5-grid-cell boundary belt. In addition, to
keep the flow close to the observations during the entire simulation period, horizontal
velocity, potential temperature, and Exner function = (p/p0 )R/Cp [42] were nudged
in the interior of the domain, with a much greater relaxation time time of 12 h to allow
small-scale high-frequency disturbances to develop. (In this formula, p and p0 are air
pressure at given locations and base state pressure at the ground, respectively, R is the
gas 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 radiative 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 stations
shown in Fig. 1. Comparison with observations revealed that these settings produced
results superior to others tested in the course of this study. The ASOS observations
accounted for observed fine structure of the flow in the vicinity of the WTC that was
lost in the 32 km resolution Eta analysis. The inclusion of 6-hourly station data from
NCEP did not produce an improvement because they were too sparse to affect local
circulation structures. In addition, they caused inhomogeneity in the driving fields
because the NCEP data are not available at each 3-h time step. Below, the results are
presented along with a sensitivity analysis discussing the dependence of the results on
model 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 included
cement, cellulose, glass fibers, asbestos, lead, and polycyclic aromatic hydrocarbons
(PAHs). The WTC debris deposited in the Hudson River and then transported downstream left a distinct signature on NY Harbor sediments, affecting the sedimentary
records of Ca, S, Sr, Cu, and Zn. In this study all types of aerosols and gases associated with WTC emissions were treated as tracers, and their transport calculated
off-line using HYPACT model, Version 1.2. HYPACT model calculates temporal and
spatial distributions of atmospheric pollutants using 3-D, time-dependent wind and
turbulence fields. It can account for multiple sources and various weather regimes,
including complex terrain flows, land/sea breezes, or circulation in urban areas. Species
can 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 wind
and potential temperature fields simulated in RAMS. The turbulence characteristics
are calculated diagnostically from available meteorological information using the turbulent closure of Mellor and Yamada [21, 22]. Particle interaction with the surface is
parameterized following Boughton et al. [42]. Above 100 m, the probability of particle
deposition for the timescales of interest is negligible. If the particle falls below this
height, the probability that the particle is deposited is computed in HYPACT from
the transition probability density given by Monin [44].
HYPACT simulations were conducted for the entire period of RAMS simulations
from September 11 to October 8, 2001. The HYPACT uses a 30-s. time step, interpolating RAMS 30-min. output at each time step. HYPACT output was archived every
30-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 in
both WTC towers whereby pollutants were released into the atmosphere at an altitude
of about 1500 m. The collapse of the main two structures produced a very fine-scale
intensive low-level jet that mixed pulverized construction materials vertically in a
column 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. These
fine-scale processes are the subject of on-going CFD studies, at a spatial resolution of
a 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, the
particles 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]. The
calculations in the present study did not account for those secondary sources. This
probably resulted in an underestimation of the overall airborne contaminant level
in the simulations. The fire on the site of the WTC that developed after the collapse of
the buildings produced a continuous source of aerosols that was most intense during
the first 3 days, well exceeding the background level. As the fire receded, the effective
altitude of the source and the emissions release rate gradually decreased.
On separate days fires were present at different locations within the 16-acre ground
zero area. When the heat released from the fire was high, it initiated intense convection
that mixed combustion products in the vertical column. The altitude of the convective
mixing depends on the magnitude of thermal heat flux from the fire, as well as atmospheric conditions. The situation is even more complex for the initial dust emission
caused by the collapse of the WTC main structures.
Therefore, for the purposes of the calculations in the present study, it was assumed
that aerosols were released from the entire area of ground zero. Because of numerous uncertainties, direct simulation of the convection caused by the fire was not
performed; instead, an approximation of the time evolution of the altitude of the
convective column was made using photographs taken from the ground and satellite
observations.
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 WTC
taken from Newark, NJ on September 11–17, 2001. The photos show that on September 11 from 0856 to 0950 EDT, after the attack on the buildings by aircraft but before
the collapse of the buildings, the plume rose above the urban canopy to the height of
about 1,000–1,500 m. At noon on September 11 the plume reached its highest altitude
of about 1,800 m. On the next day, at 0500 EDT (and probably during the night), the
altitude of the plume was below 400 m, reaching 1,500 m at 1200 EDT. However, in
the late afternoon on September 12 the altitude of the plume decreased to 400 m. On
September 13–17 the plume was mostly confined to the 200–400 m layer, sporadically
rising 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 to
decrease and stabilized above the urban canopy at about 150–200 m.
In addition to surface-based photographic observations, the WTC plume was observed from space. Figure 3 shows imagery and height retrievals derived from Multiangle Imaging SpectroRadiometer (MISR) observations. The MISR flies in sun-synchronous, polar orbit aboard NASA’s Terra spacecraft, and measures upwelling radiance from Earth in four spectral bands centered at 446, 558, 672, and 866 nm, at each
of nine fixed viewing angles spread out along the flight path from 70.5◦ forward to
70.5◦ aft [48]. It is a push-broom imager, providing nearly pole-to-pole coverage of
a 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) on
September 12, 2001. The upper panel depicts MISR 70◦ forward image of natural color reflectance for
Terra orbit 9,237, prominently showing the smoke plume, and indicating four patches, for which stereo-height histograms were derived. The lower panel shows histograms of height generated using the
60 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 approximately the size of the histogram bins. They demonstrate that the plume height is roughly 1250 m near
the WTC. Points in the histograms at 2.5 km altitude and higher are mostly cumulus clouds within the
patches, 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 12
channels, 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 Spacebased Multi-angle Imaging, submitted manuscript). The WTC and other mid-latitude
sites are viewed 1–2 times per week (see http://www-misr.jpl.nasa.gov for more details
about MISR).
The MISR contributes to knowledge of the global aerosol budget, providing tight
constraints on aerosol optical depth from well-calibrated spectral radiances measured at precisely known air-mass factors ranging from one to three. The multi-angle
observations also sample a wide range of scattering angles (about 50–160◦ at midlatitudes), offering additional constraints on particle shape, size distribution, and
single-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 features
in the multi-angle images, such as near fire, dust, or volcanic aerosol source regions, a
stereo-matching technique automatically retrieves plume-top height [50]. The height
retrieval is performed both with and without MISR-derived wind correction. For the
WTC case on September 12, the wind correction is very small, because the plume is
oriented nearly normal to the plane of the multi-angle views.
Retrieval of plume-top and cloud-top heights make use of the stereoscopic nature
of MISR data, and employs rapid pattern matching algorithms [51] to determine
the geometric parallax (horizontal displacement) of cloud and plume features due to
Environ Fluid Mech
their altitude above the surface. Photogrammetric calculations using accurate camera geometric models transform the derived parallaxes into cloud-top heights. Using
the nadir and near-nadir cameras, as is done in generating MISR’s operational stereo
product, the quantized accuracy of the resulting height field is ∼ 560 m. However, thin
plumes do not produce sufficient image contrast at these angles for the pattern matching algorithms to work. This was addressed with special processing that used more
oblique angle images, which enhance the plume appearance relative to the surface
background. Using the 60 and 70◦ pair of angles, for example, results in a quantized
height 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 indicating four patches, for which stereo-height histograms were derived. The image was
acquired at 1603 UTC (1203 EDT) at 275 m pixel resolution in the red band and at
1.1 km in the green and blue; the red band data were used to sharpen the image as
a whole to 275 m effective resolution. The near-vertical rise of the buoyant plume
directly above the WTC, viewed obliquely by the MISR 70◦ forward-looking camera
as Terra flew southward, is revealed at high resolution, projected on the scene as an
apparent south-trending column of smoke.
The lower panel in Fig. 3 shows histograms of height, with 250 m bin size and
1.1 km horizontal (pixel) resolution, for the four patches indicated in the upper panel
of 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 greater
sensitivity to thin hazes than the standard MISR stereo height product. The results
show that the height of the plume top is roughly at 1,250 m near the WTC, and it
spreads upward slightly, downwind. The points in the histograms at 2.5 km and higher
are mostly cumulus clouds within the patches, although the points at 2 km, especially
those in Patch 3, are probably part of the smoke plume. Thus, inference of vertical
mixing downwind rests on the Patch 3 data. Thinning downwind is indicated by both
the small number of pixels, for which MISR stereo heights could retrieve in Patch
4, 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 the
ground-based photos taken from Newark on September 12, that show the top of the
plume 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 column
volume with the 250 × 250 m base centered at the WTC, which roughly corresponds
to the entire area of ground zero. The effective time-varying altitude of the volume
source was chosen to be 1500 m during the first 52 h (i.e., until 1400 EDT on September 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 simulations
on October 8. One-hundred Lagrangian particles per second were emitted randomly
and statistically uniformly in the volume of the source, with a unit total mass-release
rate of 1 kg/s. Below, the magnitude of the source is estimated by comparing observed
and calculated concentrations.
3 Results
The weather on September 11–13, 2001 was clear, dry, with afternoon temperatures
reaching 80◦ F. A cold front approached the NY metropolitan area on September
14 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 an
average of 20◦ F to the lower 60s, 5–8◦ F below the climatological average. The most
intense precipitation fell on September 14, with total amounts reaching 1.9 in. On
September 17 a high-pressure system held across the region, bringing sunny skies
until September 20. A weak front passed the region producing scattered showers
of only 0.83 and 0.36 in/day on September 20 and 21, respectively. From September
22 to 24 the temperature rose to the low 70s, exceeding the climatology average by
3–7◦ 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◦ F
on 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 1
were fairly weak, reaching only 0.3, 0.41, 0.04, 0.36, and 0.1 in/day, respectively. During the first week of October, a high-pressure system brought sunny weather and a
strong diurnal temperature cycle; the daily maximum exceeded 80◦ F. A cold front
approached from the north on October 6, reducing temperatures to the lower 50s and
bringing weak scattered showers of 0.12 in/day. In general, the entire period was very
dry. There were no significant precipitation events that could wash out aerosols from
the atmosphere or off surfaces except for the showers that occurred on September 14.
That precipitation was an important factor affecting aerosol lifetime, decreasing fire
intensity and reducing aerosol emissions at ground zero. Further, it would wash away
most of the outdoor re-suspendable dust.
Through the course of this study, the sensitivity of the results was tested with respect
to several alternative model settings including coarse and fine resolution RAMS databases (Land Elevation, Land Cover, and SST), different nudging strength, and time
step. The micrometeorological and tracer fields obtained with the model settings discussed in Sect. 2.1 appear to be more consistent with observations than calculations
conducted with the different model configurations; it was assumed that these fields
were superior and were therefore used in all simulations. Discussed in this section
is the sensitivity of the simulated plume to variations in the driving meteorological
fields, as a result of implementing the ASOS and NCEP data, with the model settings
fixed at those from Sect. 2.1. More specifically, a comparison is made of results from
runs 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 plume
dispersion with respect to the release rate of Lagrangian particles was evaluated.
Then a comparison of the simulated wind fields with the observations from ASOS
stations was conducted, and the simulated concentrations were tested against PM2.5
observations from the roofs of three Public Schools in NYC where data are available
every hour. At the end of this section the plume transport on September 12, which
was characterized by a very rapid change in plume direction, is discussed and the
simulated plume compared with with the satellite images.
3.1 Plume calculations using different meteorological fields
Figure 4 depicts relative plume concentrations, shown as percentages, normalized
to the maximum concentration at the WTC location in the lowest model layer of
Grid 3 (10 m altitude), calculated using different meteorological fields. Specifically,
transport simulations were conducted using the RAMS fields calculated with initial
and 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-h
period from 0800 to 1600 EDT on September 11, 2001, (b) Same as (a) but simulated with Eta + ASOS
initialization, (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 second
using Eta initialization. This particle release rate was 10 times higher in magnitude
than that in the employed routine simulations and it effectively improved the plume
discrete spatial approximation during the entire run, compared to the run with a release rate of 100 particles per second (Fig. 4d). Concentrations were averaged for the
initial 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 with
Eta+ASOS and Eta+ASOS+NCEP fields, the direction of highest concentration rotated a little bit more to the east than in simulations with Eta fields only (Fig. 4a). The
plume calculated with the 1,000-particle-per-second release rate shows more spatial
dispersion, but in general the results are close to the simulations with the 100-particleper-second release rate. From this analysis, which shows weak dependence on driving
meteorological fields and spatial approximation of plume, it can be concluded that the
RAMS initialization using Eta and ASOS data and the 100-particle-per-second release
rate allow sufficiently accurate calculation of the concentration field for the initial post
event period. The root mean square error between the concentration field obtained in
calculations with different initial conditions and particle release rate shown in Fig. 4
did not exceed 10%. It is also shown that for the first 8 h following the attack the
plume affected mostly lower Manhattan and northwest Brooklyn. In Brooklyn the
concentrations 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 simulations. Therefore, in order to test the RAMS simulations, calculations initialized
with the Eta fields only were compared with (in this case, independent) ASOS wind
observations (see Fig. 5). The comparisons were conducted at five locations: Teterboro Airport (TEB), Newark Airport (EWR), NY Central Park, LaGuardia Airport
(LGA), and JFK Airport (JFK). The stations are shown in Fig. 1 and their exact
geographic coordinates are reported in Fig. 5. The comparison was conducted for
about 4 days and data are presented in universal time (UTC = EDT + 5 = EST + 4).
The simulated winds generally compare well with observations, both in magnitude
and direction for all the locations. The wind pattern at Central Park is most complex
because it is affected by local circulation in central Manhattan.
The surface wind speed did not exceed approximately 5 m/s at all station locations (Fig. 5). There were several distinct wind regimes. The wind was predominantly
northwesterly from September 11 to midday on September 12. Then the wind blew
predominantly from the south on September 13. On September 14 the wind was fairly
variable both in space and time, blowing predominantly southwest in the morning.
The fast-moving cold front on September 14 brought north and then northeast winds
almost simultaneously at all stations.
A vertical solid line is drawn at 1600 UTC (1200 EDT) on September 12, when
the wind direction changed dramatically and it was especially difficult to compare the
plume position with observations. This issue is discussed further in Sect. 3.4. Winds
calculated using Eta + ASOS data and Eta + ASOS + NCEP data (not shown) are
not substantially different from those using only Eta driving fields. This might be
expected from the analysis presented in Sect. 3.1. This confirms that Eta 3-D input is
mostly important for the meteorology simulations. It also shows that RAMS with the
selected setting configuration is capable of downscaling the wind field to the very fine
resolution needed for transport calculations.
3.3 Concentrations
Plume evolution was calculated using HYPACT, driven by meteorological fields from
RAMS 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 locations in the vicinity of the WTC shown in Fig. 1. Vertical solid line shows 1200 EDT, when the plume
was observed by Landsat and MISR blowing in the southwest direction. The horizontal axis shows
universal time (UTC)
Environ Fluid Mech
To evaluate the transport simulations, PM2.5 concentration data that were routinely
collected on the roofs of public school buildings in NYC were used. These instruments
were not compromised by the initial dust and smoke release caused by the collapse
and continued to collect reliable data during the entire period of interest, beginning on
September 11, 2001. For comparison three school locations were selected where observations were reported every hour: Public School (PS)-64 at 600 E 6th Street in Lower
Manhattan.; PS-199 at 3920 48th Avenue in Queens; and PS-274 at 800 Bushwick
Avenue in Brooklyn. These schools (except PS-274) were at a sufficient distance from
the southeast sector of the major impact immediately after the collapse of the WTC
buildings 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 closest
to the WTC, and received the (relatively) highest level of PM. PS-199 is located in
the same northeast sector from the WTC at a distance of about 7–8 km. Therefore it
is possible that the plume reached these two locations at about the same time. PS-274
is south of PS 199, in the east-southeast sector from the WTC where plume characteristics 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 these
distances the plume was well organized and did not experience fluctuations related to
near-source processes.
Figure 6 depicts observed and simulated concentrations for the three school locations between 1200 UTC (0800 EDT) on September 11 and 2400 UTC (2000 EDT)
September 14. In the simulations, concentrations were sampled from the fourth model
layer at an altitude of 89 m. The exact geographic coordinates of the schools are given
in Fig. 6. Using the simulation and observation data, the concentrations were normalized to the maximum value of the peaks found at each location. Two key questions
were considered in analyzing the data: (a) Do the model outputs reproduce the
observed timing of the concentration spikes? and (b) How large should the aerosol
release rate be in simulations in order to reproduce the observed aerosol concentrations in the plume? The second question is of great scientific and practical interest
because the actual aerosol release rate was never measured. The WTC aerosol emission source varied in magnitude and spatial distribution and released aerosols and
gases at varying altitudes. Moreover, there were multiple additional sources, such
as transportation, industrial activities, long-distance transport from remote sources,
and local re-suspension of deposited aerosol particles, not accounted for in the simulations that produced background concentrations. Therefore an attempt was made
to estimate the aerosol source intensity using concentrations in the peaks that were
significantly above background PM levels. The effect of aerosols on the flow was not
considered, therefore transport is linear and concentrations are linearly proportional
to the magnitude of the source. This assumption could fail if aerosol optical depth was
high 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 well
estimated, showing almost perfect timing for the observed concentration peaks. For
September 11, the simulations do not show any signal at the three locations. The
increase of observed concentrations seen at PS-274, which is closest to the main
direction of the plume transport on this day, was not that high. The series of peaks
on September 12–14 is fairly similar at PS-64 and PS-199, and has a shape different
from that at PS-274. For PS-64 and PS-199 a comparison was made of simulated and
observed 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)-64
in 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 be
about 0.2 and 0.01 kg/s, respectively, to reproduce the observed concentrations. All
locations gave similar estimates within a factor of 2.
These effective emission rates are representative of smoke produced by the fire at
ground zero. It must be emphasized that, using the above approach, only the peak
aerosol production is estimated; the source could have yielded different emission
rates at other times. Recycled aerosol deposits could be another complication. For
example, 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 dust
because more material was deposited in this sector on September 11. In addition, the
effective plume altitude, as well as the aerosol release rate, are fairly variable and are
only roughly approximated in the simulations. The aerosol surface concentrations are
also sensitive to the turbulent structure of the boundary layer, since, they develop as
a result of horizontal transport and vertical mixing from the core of the plume that
could 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 concentration spikes at different locations, providing confidence in the simulated plume
directionality. Here, analyses were conducted for what is probably the most complex
period of plume evolution in the afternoon of September 12. The simulations were
tested against visual Landsat imagery (Fig. 7), which provides a higher-resolution view
than the MISR imagery in Fig. 3. Figure 7d shows the observed plume blowing toward
the southwest at 1530 UTC (1130 EDT) half an hour before the image in Fig. 3 was
taken. The plume is viewed in the nadir only by Landsat, so all layers containing
aerosols are superposed. The plume spreads from the WTC along the Hudson River
toward the Bayonne Peninsula and Newark Bay. It turns out that it is very difficult to
reproduce this plume position in simulations because of rapid wind direction changes
at this time. The Landsat and MISR images catch the extreme southwest position of
the 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, concentrations 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 chosen
in 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 MISR
image (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 the
surface (see Fig. 5), with a significant vertical sheer. Therefore, the position of
the plume observed from space is very sensitive to the altitude of the upper part
of the plume seen in nadir view from space. Because of the strong dependence of
wind on time and altitude, this case provides an ideal consistency test between simulations and satellite observations. The MISR estimates the altitude of the top of the
plume during this period to be about 1,250 m with an uncertainty of about 250 m. The
simulations show that the portions of the plume below 750 m and above 1,050 m do
not rotate as far clockwise as in Fig. 7d, and move counterclockwise rapidly soon after
1600 UTC (1200 EDT). This suggests that the core of the smoke layer was between
750 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 at
the Central Park and the LaGuardia Airport stations are there short, sporadic periods
of easterly wind. At LaGuardia Airport, the model captures those easterly winds very
well. 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 and
evolution. Consistent with the MISR plume-top altitude retrievals, the simulations
show that most of the aerosol mass was likely transported within the 750–1,050 m
layer. This demonstrates important internal consistency between fine-scale atmospheric dynamics, transport calculations, and satellite observations. Nevertheless, this
analysis also indicates that comparisons between simulations and observations should
be conducted with reasonable caution, especially for periods of strong changes in wind
magnitude and direction.
Environ Fluid Mech
Fig. 7 Simulated with the Eta+ASOS initialization, vertically averaged from 600 to 1,000 m, and
normalized tracer concentrations on September 12, 2001 at (a) 1600 UTC (1200 EDT), (b) 1630
UTC (1230 EDT), (c) 1700 UTC (1300 EDT), and (d) Landsat image at 1600 UTC (1200 EDT) on
September 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 using
the 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 concentrations for September 11–15 in the lower layer of the model at the altitude of 10 m. The contour
intervals 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 to
this sector. It is not surprising that a weak PM2.5 concentration spike was detected
only 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 dispersion
was observed on September 12 between 0800–1600 EDT, corresponding to the rapid
wind direction variations shown in Fig. 5. The plume rotated fairly rapidly, producing
small PM2.5 spikes at all three school locations.
At 1600–2400 EDT on September 12, the plume first affected the north-to-northeast 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 25th
Street and 1st Avenue approximately 1.0 km northeast of the WTC. Although, it was
collected from a long term sampler, the only time it was affected by the WTC plume
was 1600–2400 EDT on September 12 as is seen in Fig. 8. It was assumed that the
instrument was affected by the plume for 1/2 h and then clogged and stopped sampling. Based upon the lack of prior contact with the plume, it was estimated that
the ambient concentration of fine particles detected during that half-hour period was
≈ 500 to 600 µg/m3 (L.C. Chen, personal communication). However, the modeling
results in Fig. 8 suggest that the plume was over the area for a few hours. Thus, the
estimate by Chen could be accurate, if the sampler clogged in the first 1/2 hour; the
values could be lower depending upon how long the sampler remained operational
over the 3-hour period.
On the morning of September 13, the wind moved the plume to the east, dispersing
PM in a wide southeast–northeast sector. This transport produces significant PM2.5
spikes at PS-64, PS-199 and PS-274; the signal arrived a couple of hours earlier at
PS-274 than at the other two locations.
Later on September 13, and on the morning of September 14, the plume blew
in the east, north and south directions, producing concentration peaks at all three
schools between about 0600–1000 UTC (0200–0600 EDT). At the end of the day
on September 14, the wind again blew to the east and southeast, producing detectable signals at PS-199 and PS-274. On the morning of September 15, as shown in
Fig. 8, the plume blew to the south. Figures for the rest of this period are not shown
because the magnitude of the source decreased significantly after the September
14 rain.
The period documented shows that the simulated plume was distributed fairly consistently with the ground-based PM2.5 observations that were made a few kilometers
from the WTC site. The predominant transport during this entire period was to the
south-southeast (about 75% probability). There were two episodes of northward aerosol transport, affecting upper Manhattan on September 12 and 13, three periods
when the plume blew to the east and northeast, affecting Queens and the Bronx, and
one 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 analysis conducted using the CALMET-CALPUFF modeling system in Gilliam et al. [3].
Both CALMET–CALPUFF and RAMS/HYPACT employed a constant altitude aerosol source of 50 m for the entire simulation period. However Gilliam et al. presented
calculations performed with coarser horizontal resolution (500×500 m) and employed
a different turbulent closure which caused effectively more rapid dispersion of the
plume. 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 of
the present study.
4 Summary and Conclusions
This study demonstrates that the combination of numerical modeling and groundand space-based observations allow reconstruction of the aerosol plume from the
WTC site for the entire period of interest. Micrometeorological and tracer transport
modeling provides a framework for utilizing available observations. This increases the
credibility of the simulations, and allows rough estimation of the effective emission
rates of PM observed in the urban atmosphere during this event.
The results of this work can be summarized as follows:
(1)
(2)
(3)
(4)
(5)
High-quality, fine-scale micrometeorological fields produced as part of this study
were consistent with the observations for the entire period of interest following the September 11, 2001 event. The PM transport compares favorably with
ground-based observations and satellite images. Simulations give reliable information about the spatial distribution of pollutants and the timing of the maximum near-surface concentrations at different locations.
On September 12, the southeast transport was very episodic, making comparisons with observations difficult. However, the model accurately reproduces the
plume directionality on this day. Simulated PM concentrations at the surface, and
plume directionality, are sensitive to the altitude of the convective column developed by the fire at ground zero. MISR estimated the altitude of the plume on
September 12 to be about 1,000–1,250 m at 1600 UTC (1200 EDT). The MISR
retrievals allow better calibration of the surface estimates of plume altitude.
The vertical structure of the fine-scale wind field also appears to be consistent
with the MISR altitude estimates. This example shows an important role that
quantitative satellite retrievals can play in monitoring fine-scale processes during
disasters, when ground-based observation networks are suppressed or destroyed
by catastrophic conditions.
The simulated fine-scale meteorological fields from mesoscale model simulations account for the larger-scale meteorological structures and could be used as
boundary conditions for CFD calculations.
During the first 3 days, when the magnitude of aerosol source was largest, aerosols were transported predominantly to the south-southeast, affecting lower
Manhattan and Brooklyn. However, over Brooklyn the aerosol concentrations
were at least an order of magnitude lower than would be expected in the vicinity
of the WTC.
It was found that the effective peak fine PM release rates from the fire at ground
zero have to be in the range of 10–200 g/s to be consistent with PM2.5 concentrations observed in Manhattan, Brooklyn, and Queens during the 3 days following
the collapse of the WTC. However, this estimate implicitly included contributions from secondary sources (such as re-suspension of particles) that were not
accounted for in the simulations. Presumably, their contribution was relatively
small in the 5 km vicinity of the source during the considered period when the
primary 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 the
collapse of the WTC.
The present study estimated from observations rather than simulating directly the aerosol emission processes that resulted from the collapse of the main WTC structures
and the fire at ground zero. These processes define the amount of material released
and the way in which it was initially distributed vertically. At fine spatial scales these
processes are controlled by meteorological conditions, such as the vertical temperature gradient, wind shear, and boundary layer turbulence; also they are dependant on
the detailed hydrodynamic flow within street canyons surrounding ground zero.
To reduce uncertainties in future studies, it would be interesting to simulate emissions and transport interactively, and to account realistically for aerosol source variability as a function of the micrometeorological environment. Having interactive
aerosol release modeling capability would also allow one to calculate aerosol microphysics explicitly and estimate its effect on aerosol transport and deposition patterns.
Acknowledgements This work was sponsored by USEPA grant CR827033. Additional support was
provided by a supplement to the NIEHS EOHSI center grant P30 ES05022. We thank Praveen Amar
of NESCAUM for providing plume photographs; Jennifer Bosch of the Rutgers University Institute
of Marine and Coastal Sciences for providing AVHRR SST retrievals; the developers of RAMS
and 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 Stenchikov
was partially supported by NJDEP grant SR04-048. The research of David Diner and Ralph Kahn
is supported, in part, by the MISR project at JPL, under contract with NASA. Ralph Kahn is also
supported by the NASA Climate and Radiation Research & Analysis program, under H. Maring. We
thank 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, Zhong
MH, Cohen MD, Prophete C, Yang I, Stiles R, Chee G, Johnson W, Porcja R, Alimokhtari S, Hale
RC, Weschler C, Chen LC (2002) Characterization of the dust/smoke aerosol that settled east of
the world trade center (WTC) in Lower Manhattan after the collapse of the WTC 11 September
2001. Environ Health Perspect 110:703–714
2. Landrigan PJ, Lioy PJ, Thurston G, Berkowitz G, Chen LC, Chillrud SN, Gavett SH, Georgopoulos PG, Geyh AS, Levin S, Perera F, Rappaport SM, Small C (2004) Health and environmental
consequences 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 on
human 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 and
Radiation
5. Bornstein RD, Johnson DS (1977) Urban rural wind velocity differences. Atmos Environ
11:597–604
6. Bornstein RD, Thunis P, Schayes G (1994) Observation and simulation of urban-topography barrier 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 York
pp 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 of
sulfate 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 in
particle 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, Blumenthal
DL, Anderson JA (1998) Pollutant transport during a regional O-3 episode in the mid-Atlantic
states. J Air Waste Manage Assoc 48:786–797
12. Taubman BF, Marufu LT, Piety CA, Doddridge BG, Stehr JW, Dickerson RR (2004) Airborne
characterization of the chemical, optical, and meteorological properties, and origins of a combined
ozone-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 dynamics. J Atmos Sci 35:1070–1096
14. Rehm RG, Pitts WM, Baum HR, Evans DD, Prasad K, McGrattan KB, Forney GP (2003) Initial
Model 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 Safety
Science, Boston, MA, pp 25–40
15. Cotton WR, Pielke RA, Walko RL, Liston GE, Tremback CJ, Jiang H, McAnelly RL, Harrington
JY, Nicholls ME, Carrio GG, McFadden JP (2003) RAMS 2001: Current status and future directions. 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 modeling
system—RAMS. Meteorol Atmos Phys 49:69–91
17. Tripoli GJ, Cotton WR (1982) The Colorado state university three-dimensional cloud/mesoscale
model—1982. part I: general theoretical framework and sensitivity experiments. J de Recherches
Atmos 16:185–220
18. Gal-Chen T, Somerville RCJ (1975) On the use of a coordinate transformation for the solution
of 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 Roy
Meteorol Soc 102:405–418
20. Clark TL, Farley RD (1984) Severe downslope windstorm calculations in two and three spatial
dimensions using anelastic interactive grid nesting: a possible mechanism for gustiness. J Atmos
Sci 41:329–350
21. Mellor GL, Yamada T (1974) A hierarchy of turbulence closure models for planetary boundary
layers. J Atmos Sci 31:1791–1806
22. Mellor GL, Yamada T (1982) Development of a turbulence closure model for geophysical fluid
problems. Rev Geophys Space Phys 20:851–875
23. Smagorinsky J (1963) General circulation experiments with the primitive equations. Part I: the
basic experiment. Mon Weather Rev 91:99–164
24. Harrington JY (1997) The effects of radiative and microphysical processes on simulated warm
and 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 convection
on large-scale flow. J Atmos Sci 31:1232–1240
26. Fritsch JM, Chappell CF (1980) Numerical prediction of convectively driven mesoscale pressure
systems. part I: convective parameterization. J Atmos Sci 37:1722–1733
27. Tremback CJ (1990) Numerical simulation of a mesoscale convective complex: Model development and numerical results. Ph.D. Dissertation, Department of Atmospheric Science, Colorado
State University
28. Kain JS, Fritsch JM (1990) A one-dimensional entraining/detraining plume model and its application in convective parameterization. J Atmos Sci 47:2784–2802
29. Kain JS, Fritsch JM (1993) Convective parameterization for mesoscale models: The Kain-Fritsch scheme. In: Emanuel KA, Raymond DJ, (eds) The representation of cumulus convection in
numerical models.American Meteorological Society, Boston, MA, pp 165–170
30. Miguez-Macho G, Stenchikov GL, Robock A (2005) Regional climate simulations over North
America: 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 the
observed 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 threedimensional cloud/mesoscale model 1982—part II: an ice phase parameterization. J de Recherches
Atmos 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 for
environmental 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-range
transport 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 emissions on tropospheric ozone formation over Greece. Phys Chem Earth Part C-Solar-Terrestial
Planet Sci 24:507–513
36. Fast JD, Doran JC, Shaw WJ, Coulter RL, Martin TJ (2000) The evolution of the boundary layer
and 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 of
domain 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 and
satellite SST analysis for climate. J Clim 15:1609–1625
39. Bernstein RL (1982) Sea surface temperature estimation using the NOAA 6 satellite advanced
very 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 an
Appalachian 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 environmental prediction. Weather Forecasting 11:391–413
42. Boughton BA, Delaurentis JM, Dunn WE (1987) A stochastic model of particle dispersion in the
atmosphere. Bound-Lay Meteorol 40:147–163
43. Oktay SD, Brabander DJ, Smith JP, Kada J, Bullen T, Olsen CR (2003) WTC Geochemical
fingerprint 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. Adv
Geophys 6:435–436
45. Huber A, Freeman M, Spencer R, Bell B, Kuehlert K, Schwartz W (2005) Applications of CFD
simulations of pollutant transport and dispersion within ambient urban building environments:
including homeland security. In: Annual Conference of the Air & Waste Management Association, Minneapolis, MN, June 21–24, 2005
46. Lioy PJ, Georgopoulos PG (2006) The anatomy of the exposures that occurred around the world
trade 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 and
human exposures that occurred post 9–11 (Chap 2). In: Gaffney JS, Marley NA (eds) Urban
Aerosols and Their Impacts: Lessons Learned from the World Trade Center Tragedy. American
Chemical Society, Washington, DC
48. Diner DJ, Beckert JC, Reilly TH, Bruegge CJ, Conel JE, Kahn RA, Martonchik JV, Ackerman
TP, 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 experiment
overview. IEEE Trans Geosci Remote Sens 36:1072–1087
49. Kahn R, Banerjee P, McDonald D (2001) Sensitivity of multiangle imaging to natural mixtures of
aerosols over ocean. J Geophys Res-Atmos 106:18219–18238
50. Moroney C, Davies R, Muller JP (2002) Operational retrieval of cloud-top heights using MISR
data. IEEE Trans Geosci Remote Sens 40:1532–1540
51. Muller JP, Mandanayake A, Moroney C, Davies R, Diner DJ, Paradise S (2002) MISR stereoscopic
image matchers: techniques and results. IEEE Trans Geosci Remote Sens 40:1547–1559