Spatial Variability in Pollutants:
Implications for Exposure Assessment

Dr. Larry T. Cupitt

National Exposure Research Laboratory
US EPA, Research Triangle Park, NC

ABSTRACT

Measurement studies of air pollutants in populated areas have demonstrated that
ambient air concentrations across the community can range from being relatively
uniform to being highly variable in space and time. People, too, are variable - moving
across the community throughout the day and participating in various activities that
affect their actual exposures. The resulting exposure profiles are a function of
temporally- and spatially-varying concentrations and activities. The National Exposure
Research Laboratory and others have undertaken a number of studies to assess that
variability in time and space for a variety of pollutants, source-related emissions, and
human activities.

Statistical analyses and modeling have been used to assess the variability in
exposure metrics and to relate those metrics to outcomes in complex systems. The
impact of that variability on the selection of exposure metric and exposure
classification approach (from simple metrics and statistical associations; to statistical
interpolation that fuses observational data and modeling results, to cohort estimates
of varying complexity and sophistication, to state-of-the-science probabilistic human
exposure and dose models, to personal exposure measurement studies) have been
explored. Outcome data bases (e.g., environmental or public health data) are also
examined; stratifying or matching the outcome data with an appropriate exposure
metric is often limited by the content, sparseness, or other restrictions on the outcome
data sets.

The efforts to evaluate the value of improved exposure metrics on the ability to relate
those metrics with outcomes in complex systems have met with varying degrees of
success. This work describes the results of recent efforts, mostly involving air
pollutants, to improve the sophistication in the exposure estimate and classification in
order to improve the quality of associations between exposure and outcomes at the
vend of the complex systems, both human and environmental

Pollutant Variability

Spatial Variability
Paired Koui y Samples as i fu ction of Separation Distance

• •
	•









Median Commu

: ¦ ¦



.11 Miles







i .

Distance (miles)



Relative change in concentration of ozone,
benzene, and PERC plotted as a function of
distance between sampling sites. Trend
lines illustrate that ozone is consistent and
well correlated across the urban area.

Benzene is more variable, but relatively
consistent across the urban area. PERC is
highly variable and even negatively correlated
between sites. Contemporaneous 1-h samples
collected in Atlanta, GA. Assumes isotropic
gradients.

Ratio and variability of
concentrations measured at an
ambient monitoring site and other
locations dominated by specific
source types. Particulate sulfate is a

stable component of regional fine
PM: Benzene is associated with local
sources across the urban area and is
more variable. 24-h samples
collected in Detroit, Ml. (Source:
Ron Williams, NERL, EPA)

Humans Move through the Environment

Nighttime.

' '.-V .' > '

Nighttime vs. Daytime Population Densities calculated for Houston,
Texas by LandScan population distribution model. (90 m grid cell. See
vwwv.oml/gov/gist)

Delay due to Congestion
Annual Average - 86 US Cities

/I

, *



s\











-

























'





H 1













The Average US Driver
Spends 55 minutes per day behind the wheel
Drives 29 miles a day

Only 15% of trips are for commuting - median
commute distance of 11 miles
45 % of daily trips are taken for shopping and
errands

27 % of daily trips are social and recreational

Better Exposure Estimates
Spatial & Temporal Resolution

Integrating monitoring and modeling:

Observations for ground truth, models for spatial profiles and temporal patterns
Modeled

Combining regional scale fused data with local scale models:

Regional CMAQ	Local,ASPEN	Combined

Apply human exposure models, improving exposure factors systematically to account |
for movement across areas and exposure events and processes:

1

INDOORS

OUTDOORS

TRANSIT

itlif



.A









Exposure is the
time-weighted sum
all exposures from
the different
microenvironments
in which a person
spends time.

Use census data & LandScan data go improve travel profiles for work, school, shopping,
recreational. Move population into / out of polluted areas throughout day
Ambient Concentrations

Relating Public Health Data & Environmental Data

' Much of the available public health data represent events like hospital and Emergent
Department (ED) visits We and our collaborators have tended to use ED visits for
illnesses like asthma, cardiovascular events, or pulmonary illness. Often the data
include a count of the events, by diagnoses or illness, on a particular date, at a
particular hospital or ED. The location of facility is known, but information about the
patient (e.g., their residential address) is often protected for privacy reasons or may be
aggregated to the zip code or county level Additional information that would be useful
for estimating or stratifying exposures, like smoking habits, commuting patterns,
pesticide usage habits, or occupation, are often not available.

Daily counts at each ED represent the small fraction of a large population who become
sick enough to visit an ED on a particular day. The health data are, therefore, naturally
integrated over a large population and represent the health outcomes of people who
live across a fairly wide area (US population of 291 million people & 7569 hospitals =
38,500 per hospital: at US urban average of -5500 people per square mile, this
represents an area of 7 sq mi., or a square - A/a km per side: a census tracts usually
have 2,500 - 8,000 residents). If one is to correlate the ED counts with environmental
data (e.g., exposures to a particular air pollutant), one would expect a valid relationship
only for pollutant exposures that affect the population exposure across the area. Many
"non-ambient" exposures are un-correlated with ambient exposures and lead to a
consistent distribution of exposures for a population (even across cities) from day-to-
day: as such, those "non-ambienf exposures would not be expected to correlate with
the spatially-distributed public health data in a temporal analysis In addition, the public
health data represent daily (or sometimes longer) totals, integrating responses for at
least 24 hr, and are confounded by a potential lag between exposure and onset, and
between onset and the very "human" decision to go to the ED

Results are still preliminary, but improved concentration data seems to improve
correlations with public health data. Higher resolution on the spatially-resolved

vmay be of limited benefit, however

Dve m
3d data m

References

National Household Travel Survey, Bureau of Travel Statistics
2005 Annual Urban Mobility Report, Texas Transportation Institute
Ott etal., JAWMA, (2000), 1390-1406.

Wilson and Brauer, JESEE, (2006), 264-274.


-------