(m)
UNITED STATES ENVIRONMENTAL PROTECTION AGENCY
ATMOSPHERIC RESEARCH AND EXPOSURE ASSESSMENT LABORATORY
RESEARCH TRIANGLE PARK
NORTH CAROLINA 277II
September 19, 1990
Memorandum
SUBJECT:
FROM:
Status Report of Research on Population Exposure Methodology for
Mobile Source Pollutants
Gary J. Foley	A^5fetSLX>	
Director, Atmospheric Research and
Exposure Assessment Laboratory (MD-75)
TO:
Richard D. Wilson
Office of Mobile Sources (ANR-455)
I am writing in response to your request for a status report on AREAL's
FY89-90 mobile source population exposure research. Attachment 1 summarizes
the research that we are conducting with resources provided to support Issue D
of the Air and Radiation Research Committee (i.e., A101 D21). Attachment 2
presents a recent example of the research AREAL also conducts to support
short-term, high-priority requests for assistance from your office.
As you know, the available resources in A101 D21 (about 5 FTEs and $150K
annually) are not adequate to address the broad range of mobile source
population exposure research issues of concern to your office. However, AREAL
frequently is able to increase our productivity in this area by conducting
research with extramural funds from other sources that is directly relevant to
mobile source population exposure scientific issues. I have included
descriptions of the related research programs in Attachment 1 also.
With the emphasis placed on mobile source pollutants in the draft Clean
Air Act Amendment legislation, we hope that population exposure research will
receive increased priority during FY92-93 and, ultimately, increased funding.
In the interim, I have provided information in Attachment 1 about those
research activities which will be continued or initiated during FY91.
Attachments
cc: Rick A. Linthurst, RD-680
John D. Bachmann, MD-11
Charles Gray, Ann Arbor
Mark Joyce, ANR-443
Phil Lorang, Ann Arbor
Dale Pahl, MD-56
Gary Evans, MD-56

-------
ATTACHMENT 1
STATUS REPORT ON POPULATION EXPOSURE METHODOLOGY
FOR
MOBILE SOURCE POLLUTANTS
ATMOSPHERIC RESEARCH AND EXPOSURE ASSESSMENT LABORATORY
OFFICE OF RESEARCH AND DEVELOPMENT
U.S. ENVIRONMENTAL PROTECTION AGENCY
RESEARCH TRIANGLE PARK, NORTH CAROLINA
SEPTEMBER 1990

-------
STATUS REPORT ON POPULATION EXPOSURE METHODOLOGY
FOR
MOBILE SOURCE POLLUTANTS
ATMOSPHERIC RESEARCH AND EXPOSURE ASSESSMENT LABORATORY
OFFICE OF RESEARCH AND DEVELOPMENT
U.S. ENVIRONMENTAL PROTECTION AGENCY
RESEARCH TRIANGLE PARK, NORTH CAROLINA
SEPTEMBER 1990

-------
INTRODUCTION
This status report covers the major activities conducted
during 1989 and 1990 by AREAL/RTP under Issue D (and others) of
the Air Research Committee as they relate to population exposure
methodology for mobile source pollutants. The .report is
essentially an update of a December 1988 report on work done in
support of Issue D. In addition to exposure modeling activities,
descriptions are included of monitoring and methods development
programs judged to be relevant to the issue. Finally, information
is provided on research activities which are scheduled to be
continued or initiated next year.
1989/90 ACTIVITIES
EXPOSURE MODELING
The following summaries contain information on current
activities in exposure modeling. The research discussed in these
summaries addresses the issue of serial correlation of exposures.
Previous modeling approaches have estimated exposures in a
deterministic or statistical fashion. Both of which have ignored
the effects of serial correlation. The authors recognized that
this tends to underestimate maximum exposures and that the effect
of serial correlation in exposure estimates cannot be ignored.
Specifically, model enhancements have been made to NEM (National
Exposure Model) that relate exposures this hour with ambient
concentrations that existed the previous hour. Previous versions
of NEM used simple linear regression to relate exposure to
simultaneous ambient concentrations. The proposed version of the
model includes a correlation coefficient to relate concentrations
within microenvironments to ambient concentrations this hour and
the previous hour. Model development has continued with SHAPE
(Simulation of Human Activity and Pollutant Exposure) to
statistically characterize the effect of serial correlation in
exposure estimates. The third summary describes a preliminary
analysis that attempts to quantify the magnitude of the effect of
serial correlation.
1. Modification to NEM Model
The NEM modeling methodology was initially developed in 1979
for estimating population exposure to various pollutants. This
methodology has evolved over the years to its present form which
attempts to account for the variation of exposure throughout urban
populations. NEM divides a defined human population into specific
groups, simulates their movements through environmental settings
of varying air quality, and provides estimates of exposure in these
settings. The model achieves this through the use of realistic
activity patterns based on actual diary notes.
2

-------
A critical component of the NEM methodology is the model used
to predict exposure from ambient data. Previous versions of NEM
used simple linear regression to relate exposure to simultaneous
ambient concentration within each microenvironment. However, the
correlation coefficient between these variables was found to be
very low. New modeling research summarized in the report
referenced below attempts to improve the predictive capability of
the model by accounting for the influence of previous ambient
levels and the dependence of prior exposure levels. This research
represents the first attempt to account for these factors in the
major microenvironments where people spend most of their time.
Personal exposure data collected in the Denver Personal Monitoring
Study and ambient fixed site monitoring data in Denver were used
to establish the relationships between these two variables.
Modeling results from this analysis suggest that the use of
the concurrent and previous hourly ambient concentration provides
a slight improvement in the estimation of exposure levels. The
percent variation explained with this model was approximately 10-
15% greater than earlier values, depending upon the
microenvironment. A large percentage of variation remains
unexplained, 50-75% for most environmental settings. The component
of exposure not explained by ambient level is incorporated into the
residual term. By accounting for the observed dependence of the
residuals, the model is conceptually closer to the structure of the
observed data and the freguency distribution of predicted CO
exposures is much closer to the distribution of the actual
exposures. This result is particularly important to OAQPS for the
assessment of the number of people exposed to levels exceeding
alternative ambient air quality standards.
Research Products - The Incorporation of Serial Correlation into
a Version of NEM Applicable to Carbon Monoxide by T. Johnson et.
al., PEI.
EPA Contact - Tom McCurdy (Mail Drop: 56 FTS: 629-5658).
2. Refinements of Averaging Time Exposure Models
Additional investigation has been performed on the
incorporation of auto correlation into averaging time models. When
profiles of activity patterns are used to generate time series of
simulated exposure, one typically samples from exposure
distributions which are microenvironment-specific to each activity.
If the simulation time step is short, then independent sampling at
each time step, ignoring autocorrelation, will result in aggregates
with too little variability from one simulation to another.
Autocorrelation can often be modeled with one or two extra
parameters and then used in the simulation. Furthermore, one may
substantially reduce computation by generating a single averaged
exposure for each activity segment whose distribution depends in
3

-------
a simple way on the activity duration and the modeled
autocorrelation. The process has been illustrated using the El
Camino Real commuting exposure study data of Ott, Switzer, and
Willits. These results will appear soon in a published report
(Toxicology and Industrial Health).
The microenvironmental approach to total human exposure
modeling requires the development of submodels that realistically
describe the concentrations experienced in the microenvironments
that the population occupies. These submodels should be based on,
experimental data from pollutant exposures found in the'
microenvironments, reflecting both the appropriate expected values
and the appropriate variability of concentrations. For some
applications, the nature of the time series also may be important.
This report deals primarily with the variability of the time series
of concentrations.
A two-step method of analyzing the variability was applied to
one-minute average CO exposures measured on 88 trips on El Camino
Real over a one-year period in California. The two-step method
decomposed the variability in two parts: between-trip variability
and within-trip variability. To deal with the between-trip
variability, the one-minute average CO concentrations on each trip
were normalized by dividing them by the overall mean concentration
for each trip. This step appeared justified by the observed
proportionality between the trip standard deviations and trip
means. The similarity of the frequency distributions of the
resulting normalized CO concentrations for the different trips was
exploited to obtain pooled variogram estimates needed for the
averaging time model.
When different averaging times were considered, the same
approximate lognormal distributional form emerged, with variability
expressible via the averaging time model. A fitted parametric
averaging time model, based on an exponential variogram, gave
estimates of variability very close to the empirically calculated
variability for a wide range of averaging times.
The resulting parametric averaging time model can be used in
microenvironmental probability models to simulate time-averaged
exposures for microenvironmental segments (trips) when the segment
average is selected from an appropriately described distribution.
Research Products - Averaging Time Modeling of Exposure Simulation
with Application to the El Camino Vehicle Data by W.R. Ott, P.
Switzer, and N. Willits, January 1990.
EPA Contact - Wayne Ott (Stanford University 415/367-8158).
4

-------
3. A Sensitivity Analysis on the Effect of Serial Correlation on
Exposure Estimates
Statistical methods of estimating concentration values for use
in human exposure estimates have become increasingly more popular
because of the complexities in correlating the temporal and spatial
concentration variations within microenvironments with the location
of people. The number of variables and their associated
uncertainty make deterministic models difficult to use. SHAPE and
NEM which include statistical methods for characterization of the
concentration values for different microenvironments, are examples
of human exposure models that are a combination of deterministic
and statistical methodologies. In SHAPE, concentrations are
estimated by sampling from a concentration distribution for a given
microenvironment in a Monte-Carlo fashion. In the simulations
conducted thus far, the SHAPE model has ignored serial correlation
effects. The authors recognized that this tends to underestimate
maximum exposures since the concentration in a given
microenvironment is related to source terms and atmospheric
processes. This method tends to underestimate for another reason.
If an individual is in a "dirty" microenvironment, for example a
commute that exposes him to particularly high CO levels, he tends
to be exposed to that level every commute. Monte-Carlo simulations
conducted thus far have made no provision for this effect, and
therefore tend to underestimate the highest exposures and
overestimate the lowest exposures. The purpose of this sensitivity
study is to quantify the factors affecting serial correlation in
indoor microenvironments. Further, we investigate in a very
preliminary way use of personal exposure monitoring data to infer
the value of variables needed to estimate indoor concentrations
such as the rates of air exchange, pollutant removal, and pollutant
generation.
Indoor concentration values were simulated by using a mass
consistent box model driven by a 48-hour S02 data set collected
during the RAPS study. One thousand 8-hour averages were generated
for three different scenarios. The Monte-carlo simulations
underestimated the highest exposures by about 20% and overestimated
the lower concentrations by about 30%. The difference between
using the actual air exchange rate each hour and an average air
exchange rate for each microenvironment during the 48-hour period
was slight. This suggest that a model like NEM could be updated
to include the affects of serial correlation since a mass
consistent box model could be included rather easily.
We conclude that the use of personal exposure monitoring
data to derive rate constants may be useful for order of magnitude
estimates. More study is needed to ascertain whether the positive
correlations and relative magnitudes of the time-averaged cross-
5

-------
products terms would behave as seen in our analysis if real-world
data were used. Our finding suggest that the use of PEM data would
underestimate air exchange rates, underestimate emission rates, and
overestimate removal rates where only one, or at the most, two
processes are in effect.
EPA contact - Bill Petersen (Mail Drop:80 FTS: 629-1376).
EXPOSURE MONITORING
Several monitoring programs have been conducted by AREAL/RTP
during the last two years which have shed light on the
concentration ranges for mobile source pollutants which are
typically encountered in the urban environment. The research goals
established for these programs generally included field evaluation
of methods, as well as collecting a database for interpretative
purposes. The first activity describes the results of monitoring
in-vehicle concentrations for a number of mobile source pollutants
under a variety of driving conditions. A key finding of this study
was that in-vehicle levels of VOCs can be predicted by a simple
linear regression model based on the fixed-site measurements
obtained from the vicinity of the traffic routes. The next program
discussed describes a two-year database collected at a 10-site
pilot network for urban VOC measurements (including formaldehyde)
known as the TAMS network. The third summary describes a joint
EPA/State program to monitor many of the same compounds at sites
selected and maintained by State agency personnel.
1. Exposure to VOCs. CO. 03. and N02 Under Different Conditions
Personal exposure to volatile organic compounds (VOCs) has
become of increasing concern as various regulatory options are
being considered. Data from several studies including the
California TEAM Study suggest that exposure to VOCs such as
benzene, m-jD-xy lene, and ethylbenzene is associated with automotive
vehicle use.
The in-vehicle concentrations of 24 gasoline-related volatile
organic compounds (VOCs) and three criteria air pollutants, ozone,
carbon monoxide, and nitrogen dioxide, were measured in the summer
of 1988, in Raleigh, North Carolina. Two, four-door sedans of
different ages were used to evaluate in-vehicle concentrations of
these compounds under different ventilation conditions on different
roads with different traffic patterns. Isopentane was the most
abundant aliphatic hydrocarbon (median 1-h concentration was 52.6
jig/m ) and toluene was the most abundant aromatic VOC (median 1-
h concentration was 43.1 nq/w) measured inside the vehicles. The
median concentrations of CO, N02, and 03 were 11 ppm, 81 ppb, and
11 ppb, respectively. No differences were found in the in-vehicle
concentrations of 24 VOCs, CO, N02, and 03 between the two
experimental cars. In-vehicle VOC and CO concentrations were
highest for the urban roadway, second highest for the interstate
6

-------
highway, and lowest for the rural road. The median concentration
ratio of urban/interstate/rural for each VOC was about 10/6/1. No
differences in in-vehicle VOC concentrations were found between
morning and afternoon rush hour driving, but higher in-vehicle
ozone and N02 concentrations were found during afternoon driving.
The ventilation condition that gave the highest in-vehicle VOC
levels was the one in which the windows were closed, the vent was
on and the fan was on. The ventilation condition that gave the
lowest VOC levels was the one in which the air conditioning was
used. The in-vehicle/car exterior concentration ratio for VOCs,
CO, and no2 was slightly higher than 1. The VOC concentration
measured by a pedestrian on the sidewalk was lower than the in-
vehicle measurements but higher than the fixed-site measurements
on urban roadways 50 m from streets. All 24 VOC concentrations
were positively correlated with each other. The VOC measurements
were positively correlated with the CO measurement and negatively
correlated with the ozone measurement. For the data in this study,
in-vehicle VOC levels can be predicted by a simple linear
regression model based on the fixed-site measurements obtained from
the vicinity of the traffic routes.
Research Products - Driver's Exposure to Volatile Organic
Compounds, CO, Ozone, N02 under Different Driving Conditions by C.
Chan et. al.
EPA Contact - Bill Nelson (Mail Drop:56 FTS: 629-3184).
2. Toxics Air Monitoring System (TAMS)
During FY-88 and FY-89, ambient air samples were routinely
collected in stainless steel canisters at 10 urban sites as part
of the TAMS monitoring network. The sites were selected to
represent typical industrial and commercial areas in Boston,
Chicago, Houston, and the Seattle/Tacoma metropolitan region.
Flow-controlled air samples were drawn over a 24-hr period
beginning at 6:00 a.m. every 12th day. These whole-air samples
were analyzed by GC/MSD for selected VOCs including benzene,
toluene and other aromatic compounds. The MSD was operated in the
selective ion monitoring (SIM) mode in order to optimize resolution
for the compounds of interest. Concentrations were reported
quantitatively if they exceeded 0.10 ppbv.	In addition,
cartridges coated with 2,4-DNPH were collected concurrently and
analyzed by HPLC for formaldehyde.
Quality assurance procedures were incorporated into the
program in order to provide measures of the accuracy and precision
of the methods used. The QA data revealed that the original DQOs
established for TAMS were being achieved in terms of data
completeness (>85%), precision (<20%), and accuracy (<20%) .
Ambient VOC concentrations ranged from below detection (<0.10 ppbv)
to 20.0 ppbv. Compound ratios and correlations indicated that
motor vehicle emissions accounted for most compounds with mean
7

-------
concentrations greater than 1.0 ppbv. An intensive study was
conducted at one Houston TAMS site from August 15 - 25, 1989 during
which sequential 3-hr samples were collected and analyzed to
provide information on diurnal VOC concentration patterns. Routine
sampling continued through August 20 and all samples were analyzed
by September 30, 1989.
Research Products - Final Report on the Operations and Findings of
TAMS by G. F. Evans, June 1990.
Database available on EPA's Aerometric Information Retrieval System
(AIRS).
Design and Implementation Plan for the Restructured TAMS by G. F.
Evans and J. D. Pleil, September 1990.
EPA Contact - G. F. Evans (Mail Drop: 56 FTS: 629-3124).
3. Urban Air Toxics Monitoring Program (UATMP1
The UATMP is a joint undertaking by U.S. EPA and several
participating State environmental agencies to determine ambient air
toxics concentrations and assess the risk they may pose to human
health. The network is aimed at three major groups of pollutants
including: the volatile organic compounds (VOCs), airborne
particulates (trace metals and benzo[a]pyrene), and aldehydes (as
a separate class of organics). Each is monitored in a different
way - the VOCs using evacuated canisters, the particulates using
a high volume sampler to collect samples on glass fiber filters,
and the aldehydes using cartridges containing a 2,4-DNPH substrate.
In all cases, field sampling is conducted by State agency
personnel and samples are sent to a contract laboratory (Radian
Corporation) for analysis.
A report has been prepared to present data summaries and other
analyses of the fiscal year 1988 aldehyde data from the UATMP
network. The aldehydes quantified in this report are acetaldehyde,
formaldehyde, and acrolein. Some aspects of the data presented
include seasonal patterns, weekday versus weekend patterns,
correlations among the pollutants, and comparison of UATMP data
with data from other monitoring programs. It is not possible to
productively compare results from one city with another due to
siting differences. The VOC and particulate data collected in the
UATMP program will be reported on in a future report.
Research Products - UATMP Results of Aldehyde Monitoring for FY88
EPA Contact - Robert Faoro (Mail Drop:J.4 FTS: 629-5459).
8

-------
METHODS DEVELOPMENT
The Atmospheric Research and Exposure Assessment Laboratory
(AREAL/RTP) is responsible for the assessment of environmental
monitoring technology and systems for air and for supplying
technical support to other groups in the agency including the
Office of Air and Radiation (OAR), the Office of Toxic Substances
(OTS), and the Office of Mobile Sources (OMS). In this regard, the
determination of exposure to formaldehyde vapor is an area of
continuing importance. The continuous real-time monitors
described in the first summary provide capability for measurements
over the range of concentrations expected in indoor and outdoor
air, and can thus help define the extent of personal exposure to
formaldehyde. The passive sampling device discussed in the second
summary offers a relatively inexpensive, portable, and accurate
method for determining integrated exposures to a wide variety of
mobile source pollutants including VOCs, ozone, and
formaldehyde.
1.	Real-Time Monitors for Gaseous Formaldehyde
Two new methods for real-time measurement of gaseous
formaldehyde have been developed, one based on a direct
spectroscopic approach and the other on continuous scrubbing of
gaseous formaldehyde into aqueous solution for subsequent analysis.
Both monitors are portable, but the latter method does require
occasional replenishment of simple aqueous reagent solutions. This
method, however, provides a detection limit of 0.20 ppbv which is
about two orders of magnitude below that currently achievable with
the direct approach.
The aqueous scrubbing/analysis method was field-tested by
continuous operation over a 10-day period during which outdoor and
indoor air samples were taken for alternating half-hour periods.
Concentrations observed in the field study ranged from 0.2 to 7.0
ppbv and from 10 to >50 ppbv in outdoor and indoor air,
respectively. Also, a comparison of real-time (aqueous
scrubbing/analysis) with integrated (DNPH/HPLC) measurements showed
close agreement even at the 1.0 ppbv concentration level.
Research Products - Final Report on the Development of Real-Time
Monitors for Gaseous Formaldehyde by T. J. Kelly and R. H. Barnes,
September 1989.
EPA Contact - William McClenny (Mail Drop: 44 FTS: 629-3158).
2.	Thermally Desorbable Passive Sampling Devices (PSDs)
A passive sampler was developed for short-term, low-level
ambient air monitoring applications. This small, stainless steel
device is simply designed and inexpensive. It has a high
equivalent sampling rate, is reusable and rechargeable, and is
9

-------
designed for thermal desorption. Laboratory and field tests with
Tenax GC as the sorbent have shown that the device compares very
favorably with active (pump-based) samplers and has much better
sensitivity than commercial passive samplers which utilize
activated charcoal.
Performance was examined under controlled test chamber
atmospheres and in actual outdoor and indoor situations. Sampling
rates were calculated for several volatile organic compounds. An
extensive evaluation of the effects of air velocity on performance
also was undertaken. Also, a modification has been made by
replacing the granular sorbent with filter paper treated with
appropriate reagents to trap nitrogen dioxide or formaldehyde.
Effective sampling rates for these compounds have also been
computed.
Research Products - Modifications of a High-Efficiency Passive
Sampler to Determine Nitrogen Dioxide or Formaldehyde in Air, J.D.
Mulik et. al., January 1989.
EPA Contact - James Mulik (Mail Drop: 44 FTS: 629-3067).
FUTURE ACTIVITIES
Plans within AREAL for next year include the development
of the design of a hybrid model, that will attempt to strike a
balance between deterministic and statistical models. Current
design criteria for the hybrid model call for the capability to use
empirical data values, and simulated values of concentration within
the microenvironments. Wayne Ott will continue his research at
Stanford on the effect of serial correlation on exposure estimates.
Both of these projects will be primarily in-house efforts because
of little or no extramural funding. Therefore both will proceed
slowly. At the current funding level the development of the hybrid
exposure model is scheduled to be completed in FY-97. Scheduled
work with NEM next year include the incorporation of a mass balance
approach for the simulation of microenvironmental concentration
values.
The driving habits of individuals and the general population
lead to the observation that Americans spend many hours during a
week in their cars for the purposes of commuting to and from work,
pleasure, and family support activities. Within the Denver and
Washington CO exposure studies there were examples of individuals
who had significant CO exposure while driving to work. For many
individuals studied the total amount of time spent in their
automobile was equivalent to or exceeded the time spent outdoors
during the day. Based upon typical U.S. adult time-activity
patterns, similar results could be expected for the general
10

-------
population. Considering the preceding, it is important to examine
the intensity of exposures that result from the complex mixture of
chemicals than can enter or be emitted from the vehicle cabin. The
major automobile source would be the exhaust from an internal
combustion engine that used gasoline, or methanol fuel.
The emissions from the different types of fuels used could be
quite different, therefore, key compounds such as benzene and
toluene from the gasoline powered automobiles and formaldehyde from
methanol powered automobiles will be selected for analysis.
Further, as shown by compounds such as CO, there will be
penetration of emissions into the cabin. Depending upon the
driving conditions, emissions entering the cabin could be a
combination of emissions from other automobiles on a road or
highway and the vehicle being driven, or primarily the emissions
from the vehicle being driven.
An investigation is now in progress with the following two
primary objectives:
1.	To characterize the types and intensity of human exposures to
compounds than can penetrate into or are emitted within the
cabin of an automobile for different types of driving
conditions.
2.	To compare and contrast the intensity and types of exposures
that can occur for methanol powered automobiles with those
associated with the standard non-leaded gasoline powered
vehicles.
An individual can spend many hours each day in an automobile,
thus it is possible to have exposures to single or multiple
compounds which are emitted within or penetrate into the vehicle
cabin. Therefore, it is valuable to determine if exposures to
specific chemicals or chemical classes are significant for time
scales that might produce health outcomes. Further, if methanol
powered or other alternative fuel powered automobiles are
introduced to the U.S. automobile fleet, it is important to
evaluate potential exposures relative to exposures associated with
unleaded gasoline powered automobiles.
11

-------