J4.4	LINKING AIR TOXICS CONCENTRATION FROM CMAQ TO THE HAPEM5
EXPOSURE MODEL AT NEIGHBORHOOD SCALES FOR THE PHILADELPHIA AREA
Jason Ching*1, Thomas Pierce*1, Ted Palma2, William Hutzell3,
Ruen Tang4, Alan Cimorelli5, and Jerold Herwehe6
Atmospheric Sciences Modeling Division, ARL, NOAA, RTP, NC EPA/600/A-04/085
2Office of Air Quality Planning and Standards, USEPA, RTP, NC
3National Exposure Research Laboratory, USEPA, RTP, NC
4 Computer Sciences Corporation, RTP, NC
5Air Programs Division, Region III, USEPA, Philadelphia, PA
""Atmospheric Turbulence and Diffusion Division, ARL, NOAA, Oak Ridge, TN
1. INTRODUCTION
Historically, Gaussian plume models have
provided estimates of ambient concentrations of air
pollutants for input to human exposure models.
However, most Gaussian-based modeling systems
do not account for complex chemical reactions and
struggle to account for background concentrations.
The USEPA is developing the capability to link air
toxics (AT) concentrations from an advanced
photochemical grid model to the Hazardous Air
Pollutant Exposure Model (HAPEM). The basis
for AT modeling is the Community Multi-scale Air
Quality (CMAQ) modeling system (Byun and
Ching, 1999), a "one-atmosphere" chemical
transport model. Because the AT model must
simulate the spatial distribution of "toxic hot spots"
across an urban area, the CMAQ system needs to
account for specific toxic compounds at a fine-scale
grid resolution. Recently, the HAPEM4
(www.epa.gov/ttn/atw/nata/modelexp.html) model
was extended to consider concentration variability
(HAPEM5). We are proposing a neighborhood-
scale modeling paradigm that will couple AT
concentration estimates from CMAQ at relatively
fine grid resolutions to estimates of within-grid
variability obtained from ambient concentration
distribution functions (CDFs) developed for each
grid cell. Under this paradigm, the CMAQ system
will be linked to HAPEM5, which has the
capability to incorporate information on the
statistical variability of the ambient air pollutant
concentrations. For the pilot study described in this
paper, information provided to HAPEM5 will
include the mean, median, and the 90th percentile of
*On assignment to the National Exposure Research
Laboratory, U.S. Environmental Protection
Agency. Corresponding author's email address:
ching.j ason@epa. gov
the concentration distribution. Because HAPEM5
typically requires at least one year of pollutant
concentration values, CMAQ must provide a
simulation over at least a one year period.
The CMAQ modeling system has been
configured with a modified version of the Carbon
Bond IV chemical mechanism that explicitly treats
a number of gas-phase air toxic compounds. The
system, known as CMAQ-AT, has been run for an
annual period in a nested mode at 36, 12, and 4 km
grid mesh resolutions using the 1999 National
Emission Inventory (www.epa.gov/ttn/chief/net/)
and meteorological outputs from 2001 simulations
with the Penn State/NCAR Mesoscale Meteor-
ological Model (MM5) (box.mmm.ucar.edu/mm5/).
The 36 km grid mesh encompasses the continental
United States, while the 12 and 4 km grid meshes
encompass Philadelphia and Delaware.
In prior investigations of Philadelphia, Ching
et al. (2004) employed a 1.3 km nest to produce
finely-resolved concentration fields of photo-
chemical pollutants. Because our pilot study
requires annual simulations, the grid resolution has
been initially restricted to a 4 km grid mesh to
explore the feasibility of linking CMAQ-AT with
human exposure models. Eventually, we plan to
explore the use of finer-grid meshes and the use of
the Industrial Source Complex (ISC) model with a
high-resolution receptor network capable of
providing within-grid concentration distributions of
slow reacting species.
This paper highlights the following interim
results: (a) outputs of CMAQ-AT for several toxic
air pollutants (formaldehyde, acetaldehyde,
acrolein, 1-3 butadiene, and benzene); (b) com-
parisons of model outputs to observations from an
available monitoring site; and, (c) linkages of
annual concentrations from CMAQ-AT to
HAPEM-5.

-------
2. RESULTS
The CMAQ-AT results are presented here in a
manner consistent with the input requirements for
HAPEM5. Specifically, HAPEM5 requires that
concentration values be grouped into eight 3-hourly
annualized diurnal time blocks. For example, the
first time interval represents the first three hours of
each day (00-03 LT) averaged over the year, the
second interval represents the next three hours (03-
06 LT) averaged over the year and so on.
HAPEM5 attempts to account for concentration
variability by accepting as inputs the mean, median
and 90th percentile values of each diurnal time
interval. Thus, some of the CMAQ-AT results
shown below will illustrate the temporal variability
captured by the mean, median and 90th percentile
values. In addition, we show results that illustrate
the contributions from primary and secondary
sources for formaldehyde, acetaldehyde and
acrolein. We will focus on results extracted from
two 4-km grid cells over central Philadelphia.
Although HAPEM5 is designed to perform
assessments on a census tract basis, we will assume
as a first step that the concentration in each census
tract can be associated with the grid cell overlaying
the centroid of a census tract.
2.1 Results of CMAQ annual simulations at 4
km grid size
In Figure 1, formaldehyde concentrations from
a single 4-km grid cell are grouped to show the
annual time series for the eight 3-hourly time
diurnal time intervals. The modeled results are
taken from layer 1 of grid cell (26,47). Layer 1 is
approximately 38 m deep. Figure 1 shows similar
seasonal patterns for all diurnal time periods,
although the differences appear slightly more
pronounced for the nighttime periods.
Annual daily time series of formaldehyde
for eight diurnal time intervals
(CMAQ-AT @ 4 km, central Philadelphia, 2001 meteorology)
20
18
16
14
12
10


















i





tr
It
Utr
iir
X
1


il i

bin
¦ ¦¦rim
r . w
i 'i
:, "i
hil:
hrnn* mm i
T[|M
|
a
£
00-03
03-06
06-09
09-12
12-15
15-15
18-21
21-24
Figure 1. Daily concentrations of formaldehyde for 2001 as simulated in layer 1 of CMAQ-AT over
central Philadelphia (26,47). Each of the eight time intervals contains three-hour averages for each day of
the year; 00-03 corresponds to midnight to 3 a.m.

-------
1
Oh
£
•8
o
Ph
20
15
10
CMAQ-AT simulation for central Philadelphia
:
|—
	 Pntnort/
-
-
	r i ill Jell v
	Secondary
i
i 111!
i i


m
: ".T'.OTWI
50	100 150 200 250
Day of year (2001)
300
350
Figure 2a. Annual variability in primary and secondary formaldehyde as modeled by CMAQ-AT for a 4
km grid cell (26,47) over central Philadelphia. The data are grouped into three-hour averages from the
2001 simulation.
CMAQ-AT simulation for central Philadelphia
13
O
o
o
•fi
&
o
20
15
10
50
100	150 200 250
Day of Year (2001)
300
350
Figure 2b. Ratio of secondary formaldehyde to primary formaldehyde using the data shown in Figure 2a.
The second set of figures explores the
differences between primary and secondary
formaldehyde. In these figures, the concentrations
are plotted as an annual time series, unlike the time
interval values shown in Figure 1. CMAQ-AT
allows formaldehyde to be tracked separately as a
primary and a secondary species. Primary
formaldehyde can be linked to the direct emission
of formaldehyde, while secondary formaldehyde
results from photochemical reactions, especially the
reaction of isoprene with the hydroxyl radical.
Figure 2a shows the annual time series of the
primary and secondary formaldehyde species for a
central Philadelphia grid cell. The primary species

-------
varies less on a seasonal basis than the secondary
species. Further, the contribution of the primary
species to total formaldehyde is typically larger
than that of the secondary species during the colder
months, but the opposite is true during the wanner
months. Figure 2b shows the ratio of the secondary
to primary contributions for the same time period
and grid cell. Figure 2b confirms that the
secondary contribution is smaller than the primary
contribution during the colder months, but the
secondary contribution eventually greatly exceeds
the primary contribution during the hotter summer
months. This behavior is attributed to the increase
in photochemical activity during warm, sunny
periods.
12
CMAQ-AT simulation for central Philadelphia
January
August
4	5	6
Day of Month (2001)
8*
&
o

4.5
4.0
3.5
3.0
2.5
2.0
1.5
1.0
0.5
0.0
CMAQ-AT simulation for central Philadelphia
50	100	150 200 250
Day of year (2001)
300
350
Figure 3. Fonnaldehyde concentrations from CMAQ-AT at a 4 km central Philadelphia grid cell (26,47)
for a week in January and in August. Simulated concentrations are three-hour averages.
Figure 4. Annual time series of benzene simulated by CMAQ-AT for a 4 km grid cell (26,47) over central
Philadelphia. Concentrations are taken from layer 1 and grouped into 3-hour averages.

-------
Figure 3 compares the variation in modeled
formaldehyde (in 3-hour averages) for a week in
January 2001 and a week in August 2001. Diurnal
variations appear relatively small for both months,
especially for August. This suggests that the
temporal variability for modeled formaldehyde may
be due primarily to seasonal changes upon which
are superimposed finer temporal variations due to
synoptic events.
Results for a slower reacting compound,
benzene, are shown in Figure 4. The annual time
series for benzene is similar to that seen for the
primary contribution of formaldehyde species, with
peak values occurring during the colder months.
This is attributed to the trapping of the pollutants
during periods of lower mixing heights.
Annual CMAQ-AT simulation for central Philadelphia
^ 3.0
I 2.5
3
-------
The next set of results provides relevant
summary statistics of the annual simulations as
needed for running HAPEM5. Results are shown
from a 4 km grid cell over central Philadelphia,
although these results are taken from a grid cell
(27,46) just SE of the grid cell shown in Figures 1-
4.
Figure 5 compares the primary and secondary
contributions of formaldehyde annualized for the
eight diurnal time periods. The secondary
contribution to total formaldehyde is greater than
the primary contribution for six of the eight time
intervals. The two contributions are about equally
divided during the morning (06-09 LT) and evening
(18-20 LT) commuter traffic periods.
Figure 6 shows, on an annual basis for most
diurnal time periods, that the secondary
contribution is greater than the primary contribution
for formaldehyde and acetaldehyde. For modeled
values of acrolein, most of the contribution is
primary. This suggests that reactive toxic com-
pounds should be modeled explicitly since the role
of atmospheric chemistry varies by chemical
species.
A statistic that has direct relevance to
HAPEM5 is the 90th percentile of the annual
concentration. Figure 7 compares that 90th
percentile value to the mean value for three toxic
pollutants. Ratios are shown for the eight
annualized 3-hour time intervals needed for input to
HAPEM5. In general, the ratio of the 90th
percentile to the mean is about a factor of two for
formaldehyde, aceteldehyde and acrolein.
Differences in the ratios are surprisingly small
between the three pollutants. Diurnally, the ratio
shows some variability with higher values for
fonnaldeyde during the early morning (03-06) and
for acrolein during the late afternoon (15-18).
Annual CM A Q-AT simulation for central Philadelphia
3.0
2.5
2.0
1.5
1.0
0.5
0.0
¦ FORM
ACET ~ ACRO
In
i\
00-03 03-06 06-09 09-12 12-15 15-18 18-21 21-24
Local time
Figure 7. Comparison of 90th percentile versus mean values of formaldehyde (FORM), acetaldehyde
(ACET), and acrolein (ACRO) as computed for each diurnal time period (3-hour averages) for grid cell
(27,46).
2.2 Comparison of monitoring and model
outputs
Although detailed field observations for the
Philadelphia modeling domain are lacking, air toxic
concentration measurements are available from a
single monitor in Camden, New Jersey that is part
of EPA's Urban Air Toxic Monitoring Program
(UATMP) (ERG, 2002). Observations taken as
part of the UATMP were made over a 24 h period
every 6 to 12 days. The Camden site is located just
east of Philadelphia in a semi-industrial area. The
primary emission sources are located mainly to the
west and to the north of the Camden site. Further
details on the site and the sampling protocol are
available from ERG (2002).
Concentrations from this monitoring site for a
few toxic compounds were compared to matching
modeled concentrations simulated with the 4-km
and 36-km versions of CMAQ-AT for 2001.
Concentrations from layer 1 of the grid cell
overlaying the Camden site were extracted from the
two CMAQ-AT simulations. Comparisons for 1,3-
butadiene, formaldehyde, acetaldehyde, and
benzene are shown in Table 1. In general, the
modeled mean values compared reasonably well
against the observed values, and the means,
standard deviations, and correlations from the 4 km
-------
version of CMAQ-AT compared more closely with
the observed values than the modeled values
extracted from the 36 km version of CMAQ-AT.
While this limited comparison is encouraging, to
properly assess and evaluate model performance
will require observational data from more than one
monitor location.
Table 1. Comparison of air toxic concentrations measured at the Camden, NJ, site to CMAQ-AT (layer 1).
All samples, except benzene, are 24-hour averages; benzene is a 1-hour average.
Compound
n
Mean (ug/m3)
Obs 4 km 36 km
Std. deviation (ug/m3)
Obs 4 km 36 km
Correlation
4 km 36 km
1,3-Butadiene
Formaldehyde
Acetaldehyde
Benzene
28
44
44
1328
0.33 0.18 0.12
3.68 2.91 2.25
2.09 2.49 1.92
1.11 1.02 0.77
0.34 0.12 0.08
3.21 2.13 1.52
1.42 1.20 0.78
1.06 0.71 0.40
0.07 0.09
0.42 0.38
0.45 0.44
0.48 0.41
2.3 HAPEM5 Results
2.3.1 Model Linkage
An air quality dispersion model, such as the
CMAQ model, estimates an ambient concentration
for a given time period at a given location. If a
human stayed at a fixed location for a specified
time period (in this case, one year), then the
ambient concentration predicted by an air quality
dispersion model would equal the "apparent"
exposure, or the concentration available for the
human to breathe. In the real world, however,
people generally move from location to location
(e.g., from home to work, or home to school).
Also, most people do not spend their entire day
outdoors; a majority of time is spent in indoor
locations (e.g., the home, workplace, school, or
vehicle). Studies have shown that air quality
concentrations in indoor environments can be quite
different than those in the outdoor environment.
Because of these factors, a human exposure model
is generally employed to consider these factors and
predict the "apparent" inhalation exposure.
The HAPEM5 is a screening-level exposure
model designed to predict the "apparent" inhalation
exposure for the general population, or a specific
sub-population, over spatial scales ranging from
urban environment to nationwide.
HAPEM5 uses the general approach of
tracking representatives of specified demographic
groups as they move among indoor and outdoor
microenvironments and among geographic
locations. The estimated pollutant concentrations in
each microenvironment visited are combined into a
time-weighted average concentration, which is
assigned to members of the demographic group.
HAPEM5 uses four primary sources of
information: population data from the US Census,
population activity data from human diary data,
microenvironmental data, contained within the
model and air quality data that is provided for the
study region.
As human activity data generally exhibits a
diurnal pattern the air quality data provided to
HAPEM5 must capture the expected diurnal
pattern. The hourly CMAQ results were utilized to
build a diurnal temporal pattern by averaging the
model results over 3-hour blocks (i.e., midnight -
3am, 3am-6am...) for the entire year. To help
exhibit the range in the diurnal pattern, similar
patterns were built utilizing the median and 90th
percentile ambient levels. Such temporal
information is useful for risk assessment in both
bounding the range of potential exposure levels and
in helping to define the annual variability in
exposure.
As ambient predictions from CMAQ are
calculated on a regularly spaced grid, the HAPEM5
model is designed to perform its exposure
assessments on a census tract. Thus, for input to
HAPEM5, the CMAQ air quality estimates were
converted to a census tract resolution by selecting
the grid concentration overlaying each census
centroid. In an urban area, such as the Philadelphia
study area, even with a 4 km grid, this generally
results in multiple census tracts residing a single
grid cell. The study area for HAPEM5 included the
381 census tracts in the Philadelphia area.
Ambient input files for HAPEM5 have been
built from the CMAQ-AT diurnal distributions at
the census tract level. For comparison in this paper,
three CMAQ-AT diurnal distributions (mean,
median, and 90th percentile) were examined for two
pollutants: a reactive pollutant (formaldehyde) and
a relatively non- reactive pollutant (benzene).
-------
2.3.2 HAPEM5 Model Results
A total of six HAPEM5 simulations were made
using the three diurnal distributions and two
pollutants (benzene and formaldehyde). The
average exposure for the Philadelphia area
estimated from these simulations, along with the
draft 1999 NATA HAPEM5 results which uses a
Gaussian plume model (USEPA, 2004), is
presented in Table 2. Future analyses will examine
the spatial variability of these exposure estimates
across the individual census tracts in Philadelphia.
As a benchmark, the CMAQ results are
compared to the HAPEM5 model results from the
1999 National Air Toxic Assessment (NATA). The
CMAQ-AT's estimates for benzene are about half
of that predicted by NATA and about 85% of the
NATA estimated formaldehyde exposure values.
Table 2. Average Philadelphia exposure levels (ug/m3) computed withHAPEM5.

CMAQ-AT 4 km

NATA

Mean DD Median
90% DD
Mean
Compound
DD

DD

Benzene
1.26 0.96
2.63
2.23
Formaldehyde
2.15 1.63
4.60
2.57
DD= diurnal distributions pattern
An examination of the temporal variation of
the diurnal patterns shows that exposure levels are
almost doubled as compared to using average
annual patterns. The 90th percentile estimate
provides an upper estimate that is comparable to
day-to-day variations in ambient levels.
Toxicologists believe that capturing this variation is
important in characterizing both acute as well as
chronic risk (USEPA, 2001).
3. SUMMARY
This paper has shown that a sophisticated
chemical grid model can be used to provide the air
toxic concentration fields needed to drive an
exposure model. A comparison of the model
results with a limited set of observations suggests
that the model performance is reasonable. For this
pilot study, air toxic concentrations generated by
the CMAQ-AT model for a 4 km grid mesh
overlaying Philadelphia were successfully
formatted for direct input to the human exposure
model HAPEM5.
Disclaimer: This paper has been reviewed in
accordance with United States Environmental
Protection Agency's peer and administrative review
policies and approved for presentation and
publication.
4. REFERENCES
Byun, D. and J. Ching, 1999: Science Algorithms
of the EPA Models-3 Community Multiscale Air
Quality (CMAQ) Modeling System, EPA/600/R-
99/030, U.S. Environmental Protection Agency,
Research Triangle Park, NC.
Ching, J., S. Dupont, J. Herwehe, T. Otte, A.
Lacser, D. Byun, and R. Tang, 2004: Air quality
modeling at coarse-to-fine scales in urban areas, In
Proceedings of the Sixth AMS Conference on
Atmospheric Chemistry: Air Quality in Megacities,
American Meteorological Society, Seattle,
Washington, January 11-15, 2004.
ERG, 2002: 2001 Urban Air Toxics Monitoring
Program (UATMP), EPA/R-02-010, Environmental
Protection Agency, Research Triangle Park, NC.
www.epa.gov/ttn/amtic/airtxfil.html
USEPA, 2001: Evaluating the National-Scale Air
Toxics Assessment 1996 Data - An Advisory by the
EPA Science Advisory Board , U.S. Environmental
Protection Agency Science Advisory
Environmental Board (1400A), Washington,
DC, EPA-SAB-EC-ADV-02-001, December 2001.
USEPA, 2004: The ASPEN Model, Office of Air
Quality Planning and Standards, U.S.
Environmental Protection Agency, Research
Triangle Park, NC. [Available from
www. epa. uov/ttn/atw/nata/aspcn. htmll 1
-------