EPA/600/R-18/281 j September 2018
www.epa.gov/research
United States
Environmental Protection
Agency
Assessment of SHEDS model
for air sample placement based
on population exposure
estimates following a
anthracis outdoor release
Office of Research and Development
National Exposure Research Laboratory
-------
EPA/600/R-18/281
September 2018
Assessment of SHEDS model for air sample placement
based on population exposure estimates following a
Bacillus anthracis outdoor release
by
Janet Burke, PhD
Computational Exposure Division
National Exposure Research Laboratory
Office of Research and Development
Durham, NC 27711
Computational Exposure Division
National Exposure Research Laboratory
Durham, NC 27711
-------
Disclaimer
This document has been reviewed by the U.S. Environmental Protection Agency, Office of Research
and Development, and approved for publication.
-------
Abstract
As part of EPA/ORD's Homeland Security research program, there is a need to improve strategies for
emergency response following a wide-area release of a biological agent. Modeling tools that simulate
the dispersion of biological agents for a wide-area release may be used to prioritize air sample placement
based on estimated concentrations. However, dispersion models don't account for people's behaviors
and activities which could result in some populations having greater risk of exposure. Human exposure
models that account for variability in population demographics, human activity patterns, and the factors
influencing infiltration of outdoor air indoors have previously been developed, and could be used to
better guide decontamination efforts by incorporating potential risk of exposure in air sampling
strategies.
To explore this, a case study application was performed to assess the utility of the Stochastic Human
Exposure and Dose Simulation for Particulate Matter (SHEDS-PM) model that provides estimates of
human exposures by simulating representative individuals for a specific geographic location. The
individuals time series of exposure and dose are estimated using human activity pattern data matched to
each individual and the concentrations for each location they spend time in such as outdoors, indoors,
and in vehicles. Results from the case study highlighted the impact that demographics and other factors
such as day of week have on model estimates of exposure and dose due to their influence on activity
patterns, as well as the importance of accounting for population mobility. Key advantages of the
SHEDS model and its output identified through this case study are summarized, as well as current
limitations and options for addressing them.
-------
Table of Contents
Disclaimer ii
Abstract iii
Table of Figures v
Tables vi
Acronyms and Abbreviations vii
Acknowledgments viii
1. Introduction 1
2. Methods 2
2.1. Overview of SHEDS-PM 2
2.1. Initial test runs of the SHEDS-PM model 3
2.2. SHEDS-PM refinements 5
2.3. QUIC model application 5
2.3.1. Development of Bacillus anthracis release scenario 6
2.3.2. Model inputs 6
2.4. SHEDS-PM model application 6
3. Results 10
3.1. QUIC modeled concentrations 10
3.2. SHEDS-PM modeled exposure and dose 11
4. Discussion 22
4.1. Advantages of SHEDS model 23
4.2. Limitations of SHEDS model 24
4.3. Other modeling approaches 26
5. Conclusions 27
References 28
iv
-------
Table of Figures
Figure 1. EPA SHEDS-PM 3.7 user interface showing screens for selection of inputs and
analysis of model outputs 3
Figure 2. QUIC model domain for Los Angeles simulation showing concentration plume for Yp
release scenario provided by LANL 4
Figure 3. Meteorological data (wind direction, wind speed) from Los Angeles Intl Airport (LAX)
for 2017 by hour of the day 7
Figure 4. Location of hypothetical releases of biological agents used in QUIC model application
scenarios (Ba release at Pershing Square and Yp release near Skid Row) 8
Figure 5. Satellite view of Los Angeles with census tract boundaries (blue) and QUIC model
domain (yellow) 9
Figure 6. QUIC model results for Pershing Square Ba release scenario April 22, 2017 11
Figure 7. Hourly Ba concentrations (|jg/m3) for census tracts with concentrations above zero.
12
Figure 8. Table of population, housing and employment data for census tracts within the QUIC
model domain for Los Angeles, CA with map showing census tract locations 13
Figure 9. Distribution of daily mean exposure (|jg/m3) and deposited dose (|jg) across all
SHEDS-PM simulated individuals based on QUIC modeled concentrations for Ba release
at Pershing Square on Saturday April 22, 2017 14
Figure 10. Distribution of time spent in different locations (minutes) across all simulated
individuals for (a) weekend (Saturday, no commuting) and (b) weekday with commuting
included 15
Figure 11. Variability in residential air exchange rates (hr1) for SHEDS-PM simulation 16
Figure 12. Distribution of daily mean exposure (|jg/m3) (upper) and deposited dose (|jg)
(lower) for all SHEDS-PM simulated individuals (a), and for individuals in the three census
tracts surrounding the Ba release at Pershing Square (b, c, d) on Saturday April 22, 2017.
17
Figure 13. Census tract mean and 99th percentile exposures (|jg/m3) (upper) and deposited
dose (|jg) (lower) predicted by SHEDS-PM based on QUIC modeled concentrations for Ba
release at Pershing Square (star) on Saturday April 22, 2017 18
Figure 14. Distribution of daily mean exposure (|jg/m3) and deposited dose (|jg) across all
SHEDS-PM simulated individuals based on QUIC modeled concentrations for Ba release
at Pershing Square on a weekday with commuting included, overall (left) and by different
locations (right) 19
Figure 15. Distribution of daily mean exposure (|jg/m3) and deposited dose (|jg) by different
locations for employed (right) and not employed (left) simulated individuals based on QUIC
modeled concentrations for Ba release at Pershing Square on a weekday with commuting
included 20
Figure 16. Census tract mean and 99th percentile exposures (|jg/m3) (upper) and deposited
dose (|jg) (lower) predicted by SHEDS-PM based on QUIC modeled concentrations for Ba
release at Pershing Square (star) on a weekday with commuting 21
Figure 17. Census tract mean and 99th percentile exposures (|jg/m3) (upper) and deposited
dose (|jg) (lower) predicted by SHEDS-PM based on QUIC modeled concentrations for Ba
release at Pershing Square (star) on a weekday with commuting 22
Figure 18. Map of US Census unit boundaries for Los Angeles with tracts (dark green), block
groups (light green) and blocks (gray) which could be implemented into SHEDS for
simulation at finer spatial resolutions 25
V
-------
Tables
Table 1. Wind direction and wind speed inputs used in application of QUIC model for Pershing
Square Ba release scenario April 22, 2017 8
Table 2. Input parameters for SHEDS-PM microenvironment concentration equations 9
vi
-------
Acronyms and Abbreviations
Ba
Bacillus anthracis
CHAD
Consolidated Human Activity Database
GUI
Graphical user interface
LANL
Los Alamos National Laboratory
MMD
Mass median diameter
PM
Particulate matter
QUIC
Quick Urban and Industrial Complex
SHEDS
Stochastic Human Exposure and Dose Simulation
Yp
Yersinia pestis
vii
-------
Acknowledgments
We acknowledge the contribution of Mike Brown from Los Alamos National Laboratory, for helpful
discussions and guidance on the QUIC model, and for providing the model code and input files for the
Los Angeles, CA modeling scenario used for this report.
We also acknowledge the contributions of JoEllen Brandmeyer and Sarav Arunachalam from University
of North Carolina Institute for the Environment for their support under contract no. EP-D-12-044 in
creating input files, identifying required changes to the SHEDS-PM model, and updating the model
code.
Acknowledgements are also extended to EPA reviewers who provided helpful comments on the draft
report: Tim Boe, David Heist, Leroy Mickelsen, and Michael Pirhalla.
viii
-------
1. Introduction
As part of EPA/ORD's Homeland Security research program, there is a need to improve strategies for
emergency response following a wide-area release of a biological agent. Estimating the potential risk of
human exposures to biological agents for this type of scenario could provide useful information to
support these strategies, given that human behavior and activities can impact the level of exposures. For
example, people typically spend much of their time indoors (>80%), and the amount of outdoor air that
gets indoors depends on several factors including building characteristics, meteorological conditions,
and occupant behaviors. Population mobility within an area such as for work and school commuting,
will also impact exposures if levels at those locations differ from the home location. Human exposure
models that account for variability in population demographics, human activity patterns, and the factors
influencing infiltration indoors have previously been developed to predict exposures to air pollutants for
exposure and risk assessments. This type of modeling may also be useful for informing emergency
response and planning for wide-area releases of airborne biological agents.
To explore this, research was initiated to assess the utility of a specific model, the Stochastic Human
Exposure and Dose Simulation for Particulate Matter (SHEDS-PM), for estimating population
distributions of exposure and dose for biological agents such as Bacillus anthracis following an outdoor
release within an urban area.
The SHEDS-PM model is a physically-based probabilistic model for estimating population exposures
and dose for particulate matter (PM) air pollution that was previously developed by EPA/ORD's
National Exposure Research Laboratory (NERL). The SHEDS-PM model estimates population
distributions of exposure and dose by simulating the time series of exposure for individuals that
demographically represent a population of interest. US Census demographic data are used to randomly
select individuals from the population, and human activity pattern data from EPA's Consolidated
Human Activity Pattern Database (CHAD) are randomly assigned to each simulated individual to
account for the time people spend in different locations (i.e. indoors, outdoors, in vehicles). Time
varying exposures are calculated for each simulated individual based on PM concentrations and
exposure factors provided as input to the model. The model algorithms also calculate inhaled PM dose
by estimated breathing rates that vary with physical activity over the time series of exposures, as well as
deposited PM dose based on particle-size specific deposition to three regions of the lung. Statistical
methods for incorporating both variability and uncertainty in the model inputs are used to obtain the
predicted distribution of exposure and dose that characterizes the variability across the population, and
the uncertainty associated with those predicted distributions.
SHEDS-PM has previously been applied for estimating population distributions of exposures to PM2.5,
particles less than 2.5 |im in diameter (Berrocal et al., 2011; Cao and Frey, 2011; Isakov et al., 2009).
However, the model is not specific to PM2.5 as the model algorithms have the flexibility to simulate
particles of different sizes and density. The SHEDS-PM model can therefore be applied to airborne
biological agents, and the output can provide the range in exposure and dose across a population
(variability), as well as the likelihood of exposures above a certain level. Additional relevant features of
the SHEDS-PM model include:
• Simulates individuals representing the population of a specific geographic location in the U.S.
• Particle size distribution can be specified as an input
• Uses a microenvironmental approach to estimate concentrations for each location an individual
spends time in (indoors/outdoors/in-vehicles)
1
-------
• Calculates inhalation exposure and dose (inhaled dose, deposited dose) for airborne particles
• Accounts for movement of commuters away from home location
• Has user interface for selecting model inputs and visualizing results
In addition, this approach has previously been used as the human exposure modeling component of a
source-to-dose-to-effect modeling framework for emergency events that was applied to assess the
impact of hypothetical releases of Bacillus anthracis spores (Isukapalli et al., 2008).
Therefore, the overall goal of this research was to assess the utility of the SHEDS-PM model for
supporting EPA's emergency response research needs for a wide-area release of a biological agent.
Program partners had identified air sample placement as a high-priority research need, and the SHEDS
model could provide population exposure estimates to inform air sample placement. The approach used
for the assessment focused on development of a demonstration case study application of the model. The
case study approach identified requirements for application of the model to outdoor releases of
biological agents, including input data for the model and model code refinements needed.
This report summarizes the research conducted to date for the three main aspects of this assessment:
(a) perform test runs of the SHEDS-PM model using outdoor concentrations of a biological agent in
order to identify and prioritize required changes to the model code, (b) obtain air concentrations from an
outdoor release scenario for a biological agent to use as input for application of the model, and (c) apply
the model using the air concentrations from the outdoor release scenario to demonstrate the potential
utility of the model output.
2. Methods
2.1. Overview of SHEDS-PM
SHEDS-PM was developed using MATLAB® software (The Mathworks Inc., Natick, MA) and
compiled as a stand-alone executable that runs on Windows operating systems without any specific
software required. The model has a graphical user interface (GUI) for selecting input data files,
specifying model run parameters, and analysis of model results (Figure 1). To apply SHEDS-PM, the
user must provide an input file of outdoor concentrations for the population of interest. The input
concentration data should ideally be at the census tract level in terms of spatial resolution but can be
from multiple monitoring locations or a single domain-wide concentration, and the temporal resolution
can be hourly concentration values or 24-hour average concentrations. The particle size distribution for
the input concentration data must also be specified, and includes options for defining mass median
diameter (MMD) and standard deviation for up to two size modes. Other required inputs such as
population demographic data from the US Census and human activity pattern data are included as
databases with the model.
Options for a model run scenario are also selected through the GUI, and include the range of dates and
specific census tracts for the run, as well as the number of individuals or percent of population to
simulate for each census tract. Human activity diaries are matched to the simulated individuals by age,
gender and day type (weekday, Saturday, Sunday) with optional diary matching by employment status
and housing type. Additional required inputs include parameters of equations used to estimate
concentrations in various locations that people spend time based on the outdoor concentrations supplied
as input to the model. Different options for these equations can be selected for each location (e.g.
indoors at home, school or store) depending on the amount of data available for the indoor/outdoor
2
-------
Stochastic Human Exposure and Dose Simulation
Particulate Matter (SHEDS-PM)
Edit / View Model Run Inputs
Analyze Results
Outputs
Inputs
Output Optiot
Model Run Inputs
Individual Time
Series
Population
Distributions
Day Number
Concentration File
(mins): |i3
Microenvironment
Pie Chart
Map by Census Tract
Census T ract Selection
Export Output to Excel
Figure l. EPA SHEDS-PM 3.7 user interface showing screens for selection of inputs and analysis of
model outputs.
concentration relationship. The options include a simple scaling factor, a linear regression equation
(with slope and intercept parameters), and a mass balance equation that requires several parameters for
calculating indoor concentrations such as penetration, deposition, and air exchange rates.
2.1. Initial test runs of the SHEDS-PM model
Initial test runs were conducted with EPA SHEDS-PM 3.7, the latest version of the stand-alone
executable (ver. 12/9/2011), to identify modifications to the model code needed for application to
biological agents. Input databases provided in EPA SHEDS-PM 3 .7 were used for these test runs, and
included census tract population, employment, housing and commuting data from the 2000 U.S. Census,
as well as human activity data from CHAD (version 2010 update). Default values for PM2.5 particle size
distribution and microenvironmental equation parameters were used (Burke and Vedantham, 2009), as
these initial test runs were exploratory and focused on the functionality of the model.
Input concentrations for a hypothetical release of a biological agent within an urban area were needed to
conduct the model test runs. Options for obtaining concentration data were investigated, and included a
search of published measurement studies and model applications. Criteria used to assess available data
included relevance to the objectives of this case study application (i.e. data for an outdoor release of
Bacillus anthracis or similar biological agent), and appropriateness for use as input concentrations to the
model (i.e. includes data for multiple census tracts within an urban area and over multiple time periods).
3
-------
The QUIC (Quick Urban & Industrial Complex) model (http://www. 1 anl.gov/projects/quicf) was
identified as an appropriate source of input concentration data from the literature review based on these
criteria. The QUIC dispersion modeling system was developed at Los Alamos National Laboratory
(LANL) to resolve the three-dimensional flow and plume transport and dispersion of airborne
contaminants around buildings at neighborhood scales with fast runtimes (Brown, 2014). The developer
of the model at LANL agreed to provide modeled concentration data from a previous application of
QUIC with permission from the application sponsor. However, LANL's prior applications of the QUIC
model for an outdoor release of Bacillus anthracis (Ba) were classified, and could not be provided for
this project. LANL was able to provide QUIC model output from a simulated release of Yersiniapestis
(.Yp) at a single location in Los Angeles, CA since these results were in the public domain and could
serve as example concentration output from QUIC for a biological agent in the exploratory initial testing
(Inglesby et al., 2000; Inglesby et al, 2002).
LANL provided gridded hourly Yp concentrations (in g/m3) across downtown Los Angeles over a 4-
hour period from the QUIC model (Figure 2). The release scenario was 100 g of Yp released as a line
source along a roadway at 10 am on 05/08/2015, and a particle size of 5 microns MMD was used for Yp
in this simulation. The downtown Los Angeles domain was 3.5 km by 3.5 km in size with a grid
resolution of 20 m by 20 m, a higher resolution than a typical census tract. The gridded output was
modified to the census tract format required for input to SHEDS-PM using ArcGIS (Esri, Redlands,
CA). Spatial Analyst tools in ArcGIS were used to calculate the mean of all grid points within the
borders of each census tract, and mean concentration converted to units of |ig/m3. The resulting hourly
mean census tract concentrations were used to conduct initial test runs of SHEDS-PM.
Often*
Untimrrty
of Southern
CsUfaflM
Figure 2. QUIC model domain for Los Angeles simulation showing concentration plume for Yp release
scenario provided by LANL
4
-------
2.2.SHEDS-PM refinements
The goal of the initial test runs of the EPA SHEDS-PM 3.7 executable was to identify code
modifications required for application of the model to an outdoor release of a biological agent.
However, the Windows computing environment had improved substantially since the model code was
last changed. Consequently, the SHEDS-PM source code had to be updated first to run within
MATLAB on the Windows 10 (64-bit) operating system before any modifications could be made.
A major update to the SHEDS-PM source code was a structural change to the input databases. Open
Database Connectivity (ODBC) between MATLAB version R2016b and Microsoft Access 2016 was
implemented for the databases used as input to the model. Also, primary keys and other indices were
added to the Access input databases to increase data throughput. MATLAB code for connecting to the
Access databases was also streamlined using object-oriented classes in MATLAB along with member
functions that are called. Additionally, some MATLAB functions were no longer supported and had to
be replaced with current functionality, and minor changes in syntax for many MATLAB functions
throughout the source code required updating.
In addition to updates required by the current software platform, parts of the code that were identified as
hardcoded rather than defined by the inputs needed to be modified. One example is in the code for
particle size distribution which is an important input that differs for biological agents. This code was
revised to make all variables used in the particle size distribution calculation be defined by the user input
specified through the GUI.
The open source repository GitHub was used to manage code changes and archive source code. As
incremental changes to the SHEDS-PM source code were made, branches off the main trunk were
created to provide inline documentation of all changes. Unit testing was performed to evaluate if the
incremental updates worked as expected. All source code and database changes have been committed to
a Git repository.
2.3.QUIC model application
Initial test runs of the SHEDS-PM model using QUIC input concentration data for an outdoor release in
downtown Los Angeles demonstrated that a combined application of the models could accomplish the
goals for the case study. However, the input concentration data provided by LANL from QUIC were
not ideal for assessment of the SHEDS model for several reasons. First, the location of the bioagent
release for the QUIC simulation was near 'Skid Row', a homeless area covering several city blocks to
the east of downtown Los Angeles. As a result, the area with highest concentrations was not a typical
residential or commercial area, and the most highly exposed population had limited demographic and
housing variability. Second, the bioagent simulated was Yp, which differs from Ba in physical
properties that influence concentrations. The 5-micron MMD and decay rate of 10% per minute used
for the Yp simulation resulted in near zero concentrations after just 30 minutes. The MMD for Ba is
likely smaller and the decay rate is considered negligible, 1% per minute (Stuart and Wilkening, 2005),
so concentrations of Ba would be higher over the same simulated time period. Also, the predominant
wind direction for Los Angeles is generally from the west, whereas winds from the east were used for
the Yp simulation.
To create input concentrations specifically for the SHEDS assessment, LANL provided EPA/ORD with
a license for the QUIC model along with the set of inputs used for the Los Angeles Yp application. A
5
-------
hypothetical Ba release scenario for Los Angeles was developed for the QUIC model simulation with
the goal of producing higher concentrations in more populated areas over a longer time.
2.3.1. Development of Bacillus anthracis release scenario
Meteorological conditions conducive to higher concentrations were identified using data from the Los
Angeles Intl Airport (LAX). Five-minute wind direction and wind speed data for 2017 (obtained from
https://mesonet.agron.iastate.edu/ASOS/) showed both a diurnal and seasonal trend of alternating
between land and sea breeze conditions. The diurnal trend for all of 2017 is shown in Figure 3.
Overnight and in the morning, winds tended to be from the east at lower wind speeds (land breeze),
which transitions after sunrise to winds from the west with higher winds speeds (sea breeze). Sea breeze
conditions dominate throughout the daytime and continue until late evening with a transition back to
land breeze after sunset. Changing wind direction and wind speed during these transitions between land
and sea breeze conditions should result in a release being dispersed across a larger area compared to the
more consistent winds midday, for example, and were further investigated for this scenario. The time of
day for the transition between wind patterns changed with season, with April and May 2017 having the
transition from land to sea breeze occur around 9 am. Several dates in April and May were identified as
having similar transitions between land and sea breeze conditions that could be used for this scenario.
Pershing Square was selected for the release because of its central location within the QUIC domain and
proximity to the downtown area with tall buildings (Figure 4). Public events are often held in Pershing
Square, such as lunchtime concerts and Friday night movies. A large public event occurred at Pershing
Square on Saturday April 22, 2017 starting at 9 AM. Conditions on this date had the potential for higher
concentrations over a broad area with calm winds at 10 am changing to winds from the southeast and
then from the west with higher wind speeds by 11 am, and so was selected for the QUIC model scenario.
This outdoor release scenario was used to examine exposures for the population residing at their home
location on a Saturday morning (not those attending the event in Pershing Square).
2.3.2. Model inputs
QUIC version 6.26 (02/03/2017) was applied for a release of Ba (100 g) as a line source along the road
adjacent to the east side of the park for approximately 100 m over 5 minutes beginning at 10 am on
April 22, 2017. Particle size distribution was specified as 3.5 |im MMD and geometric standard
deviation of 1.05, to represent the relatively uniform size distribution of Ba spores assumed for
weapons-grade quality anthrax (Nicogossian et al., 2011). Wind speed and wind direction inputs for the
2-hour simulation are shown in Table 1. QUIC output concentrations (in g/m3) were saved every 15
minutes, and output to text files.
2.4.SHEDS-PM model application
EPA SHEDS-PM 3.7a, the updated version of the source code running within MATLAB R2016b
(Windows 10), was applied with Ba input concentrations produced by the QUIC model for the Ba
release scenario described above. Figure 5 shows the 27 census tracts for Los Angeles that were at least
partially within the QUIC model domain. To create census tract concentration input files for the
SHEDS-PM application, the Spatial Analyst tools in ArcGIS (Esri, Redlands, CA) were used to
calculate the mean of all grid points from QUIC within the borders of each census tract for each 15-
minute output file from QUIC. The 15-minute census tract concentrations were averaged over each hour
to provide hourly averaged input concentrations for each census tract and converted to units of |ig/m3.
The particle size distribution was set to the same as specified for the QUIC simulation: 3.5 |im MMD
and geometric standard deviation of 1.05.
6
-------
LAX Wind Direction 2017
¦ N ¦ NNE BENE IE ¦ ESE BSSE BS BSSW BWSW BW B WNW B NNW B N00 BVRB
Hour of Day
<- Land Breeze -> <- Sea Breeze ->
LAX Wind Speed 2017
B<3 B 3 - 5 B 5 - 7 17-9 B9-11 Bll-13 B13-15 B 15 +
Hour of Day
Figure 3. Meteorological data (wind direction, wind speed) from Los Angeles Intl Airport (LAX) for
2017 by hour of the day illustrating pattern of nighttime land breeze (easterly winds with lower wind
speeds) and daytime sea breeze (westerly winds with higher wind speeds), with transition periods noted
by red dash boxes. Wind direction divided into 12 categories of 30 degrees each, with additional
categories for no wind (N00) and variable wind (VRB). Wind speed units=miles per hour.
Input databases for the 2000 U.S. Census, as well as human activity data from CHAD (version 2010
update) currently available in the SHEDS-PM model were used. The number of individuals simulated
per tract was 100, a minimum representative sample for characterizing variability in population
demographics for a census tract. Time-activity diaries were matched to each simulated individual based
on the default criteria of age, gender and day of week. Microenvironments specified for the simulation
included outdoors, indoors at home, school, office, store or other indoor locations, as well as in vehicles.
Default values for microenvironmental equation parameters available for PM2.5 were used (Table 2),
since the particle size distribution for the Ba input concentrations was similar (3.5 |im). Daily inhalation
exposure (|ig/m3), inhaled dose (|ig), and deposited dose (|ig) for each simulated individual were output
from the model.
7
-------
So Release
f
£
Ti
Figure 4. Location of hypothetical releases of biological agents used in QUIC m odel application
scenarios (Ba release at Pershing Square and Yp release near Skid Row) (left); and 3D view of buildings
within Los Angeles domain (color indicates building height) (right).
Table 1. Wind direction and wind speed inputs used in application of QUIC model for Pershing Square
Ba release scenario April 22, 2017.
Time
Wind Direction
(degrees)
Wind Speed
(m s1)
10:00
30
0.5
10:15
170
2.1
10:30
160
2.1
10:45
270
4.1
11:00
270
4.1
11:15
280
4.1
11:30
270
4.1
11:45
280
5.1
12:00
270
5.7
8
-------
Figure 5. Satellite view of Los Angeles with census tract boundaries (blue) and QUIC model domain
(yellow).
Table 2. Input parameters for SHEDS-PM microenvironment concentration equations.
Microenvironment
Equation Type
Parameter
Parameter Value*
Outdoors
All Outdoors
Scaling factor
1.0
Penetration
/V(0.97, 0.02)
Deposition
N(0.3, 0.095)
Air exchange rate:
- Spring
/og/V(0.449, 2.226)
Home
Mass balance
Volume:
- Single family detached
logN(411.6, 1.649)
Indoors
- Single family attached
- Apartment
- Other
/og/V(327.0, 1.649)
logN(219.2, 1.553)
/og/V(208.5, 1.433)
Office
Scaling factor
0.3
School
Scaling factor
0.6
Store
Scaling factor
0.75
Other Indoor
Scaling factor
0.5
In vehicle
All Vehicle
Scaling factor
0.8
*N(x,y) = Normal distribution with mean x and standard deviation y; !ogN(x,y) = Lognormal distribution with geometric mean x and
geometric standard deviation y.
9
-------
To compare estimates of Ba exposure and dose when work commuting is considered, an additional
SHEDS-PM model run was conducted with the commuting algorithm applied. This algorithm uses data
from the US Census on where the population from each tract commutes to for work. If a simulated
individual is employed, and the activity diary assigned to that individual includes the work activity, then
the individual is randomly assigned to a work tract. The concentration for the work tract is used to
calculate exposure and dose for that individual during the work activity instead of the home tract
concentration. However, the work tract could be the home tract depending on the commuting
proportions for each tract. All the same SHEDS-PM model run specifications were used, along with the
same input concentration data from the QUIC simulation except that the date in the input concentration
file was changed to a weekday (April 24, 2017) and diary matching by employment status required for
simulation of commuting.
Since Ba concentrations for only 27 census tracts were able to be obtained from the QUIC model
domain for this scenario, the commuting algorithm in SHEDS-PM was only able to move individuals
between this limited set of census tracts. An additional SHEDS-PM simulation was performed to
address that individuals living outside the QUIC domain commute to the area of Los Angeles for work.
This third simulation was performed using a total of 326 census tracts with Ba concentrations of zero
added to the input file for all the other census tracts and the same SHEDS-PM model run specifications
as above.
3. Results
3.1. QUIC modeled concentrations
Application of QUIC for the Ba release scenario described above provided a time series of input
concentrations needed for the assessment of SHEDS-PM. The QUIC modeled concentrations are
displayed in Figure 6, which shows the progression of the concentration plume at 15-minute intervals
from the simulation of a Ba release near Pershing Square on April 22, 2017. The plume remains
centered over the release location initially (10:15 am), then is gradually dispersed across the downtown
area with high-rise buildings (10:30 am). The plume continues to expand over more residential areas to
the north (10:45 am) until the wind direction shifts to from the west (11:00 am). The higher wind speeds
with winds from the west, further disperses the concentration plume to the east over additional areas
north and east of Pershing Square (11:15 am). This hypothetical Ba release scenario resulted in higher
concentrations in more populated areas over a longer time period for the SHEDS-PM case study when
compared to the previous Yp release scenario.
As described above, hourly mean Ba concentrations for each of the 27 census tracts within the QUIC
modeling domain were calculated from the gridded concentration output from QUIC to create the input
concentration file for SHEDS-PM. Figure 7 shows the variation in Ba concentrations in |ig/m3 by hour
of the day for 17 of the census tracts that had above zero concentrations for at least one hour (i.e. 10
census tracts had zero concentration for all hours). Census tract mean concentrations were zero until the
release at 10 am. Census tracts with red and orange bars had the highest mean concentrations over the
first hour of the simulation (hour 10), and lower concentrations for the second hour (hour 11), with
concentrations dropping to very low or to zero by the third hour (hour 12). Census tracts with yellow
bars had mean concentrations in the middle of the range, with some tracts increasing in concentration
from the first hour to the second hour. Census tracts with blue bars had lower mean concentrations and
typically only above zero for one hour (11 am). Figure 8 (right) displays the location of the census tracts
within the QUIC model domain (yellow) and relative to the release.
10
-------
Ill
Figure 6. QUIC model results for Pershing Square Ba release scenario April 22, 2017.
The table in Figure 8 shows that total population also varied among the census tracts. Several census
tracts with higher concentrations also had larger populations (approx. 4000 people or more for 207300,
207500, 208000, 208300), while other tracts with higher concentrations had fewer people (207710,
209200). The type of housing also varied by census tract, with similar numbers of single family homes
and apartments for a few tracts, whereas housing for most tracts was dominated by apartments. Most
census tracts had similar levels of employed versus unemployed people, except for 206020. Tract
206020 had unusual demographics (high total population with none employed and few housing units)
due to it containing a large prison and railyard, so it was excluded from the SHEDS-PM simulation.
3.2.SHEDS-PM modeled exposure and dose
The SHEDS-PM model was applied using the hourly mean Ba concentrations for April 22, 2017 from
the QUIC model, and the results are displayed in Figures 9 through 13. Figure 9 shows the distribution
of the daily mean exposure (|ig/m3) and deposited dose (jig) across all simulated individuals, overall and
by the different locations they spent time in. In all cases, the median value is zero due to the hours prior
to and following the release where concentrations were zero for individuals in the tracts impacted by the
11
-------
Hour
l.E+00
l.E-01
m
l.E-02
E
—
l.E-03
00
3
l.E-04
c
o
l.E-05
5
to
l.E-06
4->
c
l.E-07
d)
u
l.E-08
c
o
l.E-09
u
l.E-10
^%^>ecb% ° ° Census Tract
Figure 7. Hourly Ba concentrations (|ig/m3) for census tracts with concentrations above zero.
release, as well as the individuals for tracts not impacted by the plume having concentrations of zero for
the entire 24 hours. Daily exposure and dose were much higher for the home location compared to any
other location due to individuals spending most of their time at home as shown in Figure 10(a). Since
April 22, 2017 was a Saturday, and this simulation matched activity patterns by day of week and did not
include any mobility between census tracts such as for commuting, the exposure concentrations for each
location differed only by the fraction of the outdoor Ba concentration assigned to that location. For
example, when a simulated individual spent time outdoors the exposure concentration would be
equivalent to the input Ba concentration for that hour, whereas for time spent in an office building the
exposure concentration would be calculated as 30% of the input Ba concentration for that hour (see
Table 2).
For the home location, one of the important drivers of indoor concentrations is the air exchange rate
which represents the relative volume of air indoors that is exchanged with outdoor air per hour. Many
factors influence the air exchange rate in homes such as the size and age of the home, as well as
meteorological conditions (temperature, wind speed). Distributions of air exchange rates for different
regions based on housing stock and seasons are used as input to the SHEDS-PM model. Figure 11
displays the variability in residential air exchange rates for this simulation, overall and between census
tracts. While most homes were assigned air exchange rates of 0.4 - 0.6 hr"1, others had air exchange
rates of 1.0 hr"1 or higher, which would contribute to greater infiltration of outdoor Ba indoors for these
homes. Variation between census tracts was also evident, with mean values ranging from 0.52 - 0.76
hr"1, indicating that exposures while indoors at home can also vary due to factors such as air exchange
rates.
12
-------
Census
Tract
Total
Population
Housing
Employment
Total
Single
Family/
Duplex
Total
Apartments
Total
Employed
Total
Unemployed
197500
5263
758
996
2264
1619
197600
2984
467
477
1182
1049
206020
10852
15
18 10868
206030
955
60
250
306 315
206040
3445
322
952
1296 1197
206050
2488
126
489
1038[ 1032
206200
3477
45
289
765
2547
206300
4995
72
466
1062
3576
207100
5753
332
1239
1632
3212
207300
3739
66
235
847
2788
207400
1237
7
0
0
1094
207500
4098
71
344
1880
2079
207710
1229
15
270
287
921
207900
1993
58
138
835
858
208000
42.53
384
731
1211
1534
208300
6893
398
1390
2342
2510
209101
6800
161
1141
2376
2428
209102
4677
54
1007
1756
1785
209200
1467
5
399
563
477
209300
3100
75
909
1189
1244
209520
2772
58
768
1168
948
209820
2708
194
798
919
960
210010
3607
129
715
1187
1403
224010
2529
53
354
886
1095
224200
3067
234
700
1004
1202
224320
3293
198
847
1133
1161
226000
4232
495
514
1337
1897
197!
9761
207100
206020
209102720921
207710
:073I
226000
I I Census tracts
QUIC model domain
EH Pershing Square
Figure 8. Table of population, housing and employment data for census tracts within the QUIC model
domain for Los Angeles, CA with map showing census tract locations (bold tracts had Ba
concentrations above zero following release).
13
-------
10-3 Total Exposure
Total Exp.
10
x 10"'
Oj
CD
O
Cl
X
\c
0.3
0.25
0.15
o
Cl
0.05
Total Deposited Dose
Total Dep. Dose
0.15
0.1
O)
-------
=> 800
1200
1000
(/)
£
1 800
1
600
400
200
0
,0^"
M'
-------
Histogram - Air Exchange Rate
Figure 11. Variability in residential air exchange rates (hr"1) for SHEDS-PM simulation. Census tract
mean air exchange rate for 100 simulated individuals in each tract (left) and frequency distribution of air
exchange rate for all simulated individuals (n=2500) (right).
Figure 12 shows that the distributions of the daily mean exposure (|ig/m3) and deposited dose (|ig) for
the three census tracts surrounding the Ba release at Pershing Square were different than for all census
tracts combined. For these three census tracts (207300, 207710 and 207500), all simulated individuals
had daily mean exposures and deposited dose values above zero. Tract 207500 had exposure and
deposited dose values similar to the range for all tracts combined, while the other two tracts (207300 and
207710) had exposures 2 to 4 times higher, and deposited dose values as much as an order of magnitude
higher. These differences in exposures and deposited dose between census tracts are also shown in the
maps of the mean and 99th percentiles for each tract in Figure 13. Only the three census tracts
surrounding the Ba release at Pershing Square had mean exposure and deposited dose above zero, and
individuals living in census tract 207300 had higher exposure and dose.
The potential impact of commuting on exposure and dose for this release scenario was also investigated.
As described above, the SHEDS-PM model was applied using the same concentration input file of
modeled Ba concentrations for the census tracts within the QUIC domain, only the date was modified to
a weekday (April 24, 2017) and the commuting option selected for the model run. Although the input
concentrations were the same, the activity patterns assigned to the simulated individuals were different
since they were selected from weekday diaries, and concentrations from census tracts other than the
home tract may have been assigned for individuals when at work. Results for this SHEDS-PM
simulation that included commuting are displayed in Figures 14 through 17.
16
-------
(a) All Tracts
3 Total Exposure
:
(b) Tract 207300
(c) Tract 207710
T^O.04
Total Exposure
~r~
0.035
0.03
E 0.025
O)
~ 0.02
dj
w
§. 0.015
HI
0.01
I
I
0.005
_L
0
Total Exp.
(d) Tract 207500
0.04
Total Exposure
0.035
0.03
0.025
N
O)
•7* 0.02
ZJ
H
8. 0.015
X
LU
0.01
0.005
_L
0
Total Exp.
K "io"3 Total Exposure
Total Exp.
0.3
Total Deposited Dose
0.25
0.2
"35
r0-15
tn
I
O
—1—1
0.1
0.05
0
Total Dep. Dose
Total Deposited Dose
Total Dep. Dose
Total Deposited Dose
Total Dep. Dose
Total Deposited Dose
Total Dep. I
Figure 12. Distribution of daily mean exposure (|ig/m3) (upper) and deposited dose (|ig) (lower) for all
SHEDS-PM simulated individuals (a), and for individuals in the three census tracts surrounding the Ba
release at Pershing Square (b, c, d) on Saturday April 22, 2017. Orange star indicates different range for
axis.
17
-------
Total Exposure- Mean Value Total Exposure - 99th percentile value
-118.268
Total Deposited Dose - 99th percentile value
Total Deposited Dose- Mean Value
-118.251
-118.268
^ I I I
0 0.33094 0.66188 0.99282 1.32376 1.6547 1.98564 2.31658 2.64752
Figure 13. Census tract mean and 99th percentile exposures (,ug/m3) (upper) and deposited dose (jig)
(lower) predicted by SHEDS-PM based on QUIC modeled concentrations for Ba release at Pershing
Square (star) on Saturday April 22, 2017.
Figure 14 shows that the overall distribution of exposure was slightly lower, while deposited dose was
somewhat higher overall, compared to the simulation for a Saturday without commuting (Figure 9).
Higher deposited dose with similar exposure levels could be due to the weekday activity diaries having
greater physical activity levels (higher breathing rates resulting in greater dose) than the Saturday
diaries. Although exposure and dose were also highest for the home location for this weekday simulation
with commuting, the time spent in other locations was significantly higher as shown in Figure 10(b).
This is further highlighted in Figure 15 which compares exposure and dose for individuals that were
employed (31% of the population) to the rest of the population. Because employed individuals could be
assigned to census tracts different than their home tract while working, exposure and dose were highest
for the office location for employed individuals. For individuals not employed (children, elderly and
non-working adults), the home location dominated their exposure and dose.
18
-------
x 10
3 Total Exposure
Total Exp.
10
x 10'*
*0^
,0s .V*
•V
,4**
0X
„<*9
,,.C*
Vs
AO"
¦\
-------
Not employed
Employed
h -V"
^ ^ ^ X° <**
^ ^ ^ .AX**'
,d^ ^
0
h»
0s ,4»' .«•' .0s' ,4»V ,tt» ..4.1s
.C^ ^°°
+
°0.05
+
0
+
+
^ ,0<
^ ^
O0^6 Oot,C o<£e
-------
Results for the third SHEDS-PM simulation that included census tracts from outside the QUIC model
domain to examine the impact of worker commuting across the broader Los Angeles area are shown in
Figure 17. These maps show that for the more than 300 census tracts with Ba concentrations of zero for
all hours, some individuals who commuted spent enough time in the tracts impacted by the Ba release to
result in elevated daily exposure and dose.
Total Exposure- Mean Value
Total Exposure - 99th percentile value
0.00486746 0.00973493 0.0146024 0.0194699 0.0243373 0.0292048 0.0340722 0.038939
Total Deposited Dose- Mean Value
Total Deposited Dose - 99th percentile value
34,046 —
0.242771 0.485543 0.728314 0.971086 1.21386
Figure 16. Census tract mean and 99th percentile exposures (|ig/m3) (upper) and deposited dose (ug)
(lower) predicted by SHEDS-PM based on QUIC modeled concentrations for Ba release at Pershing
Square (star) on a weekday with commuting.
21
-------
Total Exposure - 99th percentile value
Total Exposure- Mean Value
Total Deposited Dose- Mean Value Total Deposited Dose - 99th percentile value
34.035
0 0.263239 0.526478 0.789718 1.05296 1.3162 1.57944 1.84267 2.10591
Figure 17. Census tract mean and 99th percentile exposures (|ig/m3) (upper) and deposited dose (jig)
(lower) predicted by SHEDS-PM based on QUIC modeled concentrations for Ba release at Pershing
Square (star) on a weekday with commuting.
4. Discussion
EPA's SHEDS-PM model provides estimates of human exposures by simulating representative
individuals for a specific geographic location. The individuals time series of exposure and dose are
estimated using human activity pattern data matched to each individual by factors that influence activity
patterns (e.g., demographics and day of week), and the air concentrations for each location they spend
time in (e.g., outdoors, indoors, in vehicle). This assessment of the SHEDS model was initiated because
these features of the model had potential to provide useful information on human exposures following
an outdoor release of a biological agent for emergency response and planning, such as for guiding air
sample placement.
22
-------
Overall, the results of this initial assessment demonstrate the potential utility of the SHEDS model for
this purpose. However, the SHEDS model in its current form is likely too complex for application
during the response immediately following a wide-area release of a biological agent. The SHEDS
modeling approach may be more appropriate for emergency response planning when another model
such as QUIC is also applied, and various scenarios are investigated, such as for future high-profile
security events or locations. Alternatively, the combination of the underlying data sources included in
the SHEDS model (i.e. population demographics, housing types, and commuting patterns from US
Census data along with human activity pattern data) may provide the most important information for
prioritizing decontamination to reduce potential risk of exposure, and could be incorporated in other
tools or approaches for guiding air sample placement in emergency response.
Key advantages of the SHEDS model and its output identified through this case study are summarized
below, as well as current limitations and options for addressing them. In addition, the exposure and dose
modeling capabilities within the QUIC model are discussed.
4.1. Advantages of SHEDS model
The main advantages of the SHEDS modeling approach are that the inputs are referenced to a specific
geographic location, that the relationships between population demographics and human activity
patterns are accounted for, and that the mobility of the population in space and time for worker
commuting can be included.
Population simulated for a geographic location. Use of census data in the SHEDS model provides the
reference to the geographic location for the inputs to each simulation. Currently, census tracts are the
spatial unit of the input data for population demographics and housing type data. A simulation
population is generated that demographically represents the population of each census tract in terms of
age and gender, as well as for employment status and types of housing units if selected. When the
concentration data used as input to the model have similar spatial resolution, the model provides the
ability to link concentration variability with population variability for that geographic location. This
aspect of the model provides refined estimates of human exposures that could be important for
application to biological agents for emergency response planning given the potential for large spatial
gradients in concentrations following an outdoor release and the need to identify the population at risk
and locate samplers to appropriately assess their risk.
Population demographics, housing factors and activity patterns. The case study results highlighted the
impact that demographics and other factors such as day of week have on model estimates of exposure
and dose due to their influence on activity patterns. The census tracts near the release location differed
in concentrations as well as population characteristics, housing types and employment status. The age
and gender proportions differed by census tract and the activity patterns assigned contributed to the
variation in exposure and dose in addition to the concentration differences. Age and gender differences
in time spent outdoors, indoors and in vehicles can influence the risk of exposure, and age and gender
specific factors are also used in the calculation of dose.
Comparison of the SHEDS simulations for a weekday versus a weekend day showed that overall the
time spent at home contributed the most to exposure and dose for both, although the amount was lower
for the weekday simulation. More time was spent in locations other than the home for the weekday
simulation, with median time spent in offices and schools increasing to approximately 4 and 6 hours
respectively. Although the same concentrations were used as input for both simulations, the estimates of
23
-------
exposure and dose were not the same due to the time spent in locations other than home and the different
indoor-outdoor factors for those locations. However, since time spent at home typically contributed the
most to exposure and dose, an important feature of the SHEDS model is the ability to specify different
types of housing and the factors that contribute to residential indoor exposure estimates such as air
exchange rates.
These relationships between demographics, activity patterns and housing factors were incorporated in
the SHEDS model design for air pollutant exposure assessment, but may be equally relevant for
estimating exposures and dose following an outdoor release of a biological agent.
Population mobility for worker commuting. Mobility of the population in space and time such as for
worker commuting may also be a critical component of exposure assessment for biological agents. In
this case study, accounting for worker commuting patterns affected the SHEDS results in both directions
depending on location and population. Some census tracts far from the release and outside the QUIC
model domain had high 99th percentile exposures and deposited dose despite no direct impact from the
release on those census tracts. The population simulated for these tracts had individuals that commuted
to census tracts near the release location which had higher concentrations than their home census tract,
thus elevating their exposure and dose. On the other hand, some individuals whose home census tract
was near the release location commuted to census tracts with lower or zero concentrations (not impacted
by plume) which lowered their exposure and dose. Using the worker commuting capability in the
SHEDS model, this case study highlighted the importance of accounting for population mobility due to
the impact that may have on exposures to biological agents following an outdoor release.
4.2. Limitations of SHEDS model
Since SHEDS-PM was developed for application to particulate matter air pollution and to estimate the
overall variability in exposure and dose across a population, the demonstration case study also identified
limitations for application of the model to outdoor releases of biological agents. These limitations
include the relevant spatial and temporal scales, options for map display of outputs, the concentration
units used, population mobility factors simulated, and lung deposition calculations for bioaerosols.
Spatial and temporal scale. Clearly, one major limitation for application of the SHEDS model for
emergency response is the scale, with the current model limited to census tracts for the spatial scale and
hourly concentrations inputs for the temporal scale. Simulation of releases of biological agents (or other
airborne releases relevant to emergency response) require higher spatial and temporal resolution due to
the potential for short duration extreme concentrations over a localized area that may be the most
relevant exposures to capture.
However, the structure of the SHEDS model code allows for more flexibility in scale if the input data
for different scales could be developed and a consistent set of inputs provided for an application. For
example, a database of US Census demographic, housing, and commuting data could be compiled with
data for all spatial levels of census units available, including census tracts, block groups, and blocks.
Figure 18 displays the boundaries of US Census tracts, block groups, and blocks for the Los Angeles
domain for comparison. The appropriate spatial scale of input data could then be selected by the user
for the application. Modification of the model code to support this flexibility in spatial scale would
require relatively straightforward changes to the code. Development of the database from census data
would also be relatively straightforward, but would also require routine updating as new US Census data
are released.
24
-------
For temporal resolution, the SHEDS model currently has code that is flexible in the input data, from
hourly to 24-hour data. Human activity pattern data in the CHAD database is also available at a finer
temporal resolution, with 1 hour being the maximum length for an activity record. Modification of the
model code to support time resolution less than 1 hour would also be relatively straightforward.
Map display of model outputs. One of the key features of the SHEDS model that could be directly
useful for guiding air sample placement based on human exposure estimates for biological agents is the
capability to display model outputs on a map of the census tracts included in the simulation. However,
the options currently available for mapping of outputs are limited to the display of distribution statistics
for exposure, intake dose, and deposited dose, as well as the input concentrations. Mapping of the
model results could be modified to include more relevant information for emergency response such as
population demographics, housing characteristics, indoor concentrations or infiltration. An additional
map feature that could display the impact of commuting on exposures could also make the output more
useful. Model code for the mapping feature would also have to be modified to support the different
census unit options if the model was made m ore flexible in spatial resolution as described above.
msm
r II —
[manitiouIaveh^ !
% w/i
T- ''V -VI '
¦¦ f ih,
j
k ,4"v
i // 3%// fr*-*
) I Ijj \i r t .7/
3 : ^ "yTTH ST I 1
VIOLET ST
V K}
SACH^MFNTO S T
¦PORTER,,STi
fc'O BLVD
-11 »26FiWtSif|
Figure 18. Map of US Census unit boundaries for Los Angeles with tracts (dark green), block groups
(light green) and blocks (gray) which could be implemented into SHEDS for simulation at finer spatial
resolutions. [Source: https://tigerweb.geo.census.gov/tigerwebmain/TIGERweb apps.html ]
25
-------
Concentration units. Currently, the SHEDS model assumes input concentrations in units of |ig/m3. For
biological agents such as Ba the units may need to be modifiable to allow for the metric of interest for
infection or detection to be simulated (e.g. the number of spores). Options for addressing the units issue
could include revising the code to allow complete flexibility in the model for user defined units so the
model would perform exposure and dose calculations and provide output in the units specified, or
continue to require input concentrations in mass units but allow conversion of the model output to
different units based on a conversion factor (default or user provided).
Population mobility factors. The SHEDS model currently captures population mobility within the
model domain related to worker commuting. However, other types of mobility may be important for
some segments of the population such as children commuting to school or for activities such as
shopping, especially for suburban areas with large retail locations. Incorporating these aspects of
population mobility within the SHEDS model has been constrained by the lack of a single source for
available data (e.g., US Census) and differences between urban areas such as for school commuting
(Xue et al., 2009). The availability of population mobility data may improve as approaches for utilizing
different data sources advances. For example, Landscan USA provides high-resolution population
distribution data over space and time (Bhaduri et al, 2007), but the data are limited to static daytime and
nighttime population counts that do not include demographic data (e.g. age, gender). Agent-based
modeling approaches are also being used to create synthetic populations that accounting for movement
in space and time (Pires et al, 2018). This type of data could be of value for guiding air sample
placement as an indicator of potential exposure by providing estimates of the location of the population
in space and time with the demographic characteristics needed for human exposure estimates (Aubrecht
et al, 2013).
Lung deposition for bioaerosols. The SHEDS model includes the standard ICRP model for calculating
the deposition of particles to three regions of the lung (Burke and Vedantham, 2009). The ICRP model
assumes particles are spherical shaped, but bioaerosols tend to be non-spherical or elongated in shape.
Guha et al. (2014) modified the ICRP model to address this limitation and the Matlab code is publicly
available for download. The SHEDS model could be updated with this new ICRP model code to
account for the specific nature of bioaerosol deposition to the lung.
4.3. Other modeling approaches
The QUIC model was applied to generate Ba input concentrations for this assessment of the SHEDS
model. However, QUIC also has capabilities for estimating population exposures and dose. For
example, using the building data required as input to the model, infiltration can be calculated based on
typical parameters for each type of building (e.g. office building) or specified for each individual
building, to estimate indoor concentrations. The QUIC model also has a population exposure calculator
that uses either daytime or nighttime population counts provided with the model to assign counts to the
buildings, outdoors, or in vehicles, and uses a fixed 'protection' factor for indoors and vehicles. The
QUIC model approach to estimating exposures takes advantage of the individual building data required
as input to the model to allocate the population to different locations spatially and the concentrations for
those locations. In contrast, the focus of the SHEDS model is on simulating individuals in the
population using population demographics and housing type proportions that vary spatially. The
individuals are assigned to different locations (indoors, outdoors, in vehicles) over time with different
concentrations, but these are not fixed locations such as the individual buildings defined for QUIC
simulations.
Inhalation of aerosols can also be simulated by QUIC based on a single breathing rate and fixed lung
deposition fractions. This differs from the SHEDS modeling approach that incorporates variability in
26
-------
breathing rates by age and gender as well as for different activities (e.g., sleeping vs. exercise), and
particle-size specific deposition to the lung. However, the QUIC model goes beyond inhalation and
dose to estimate response using lethal concentration thresholds provided as input. Estimates of the total
affected population, the affected population by location (indoor, outdoor and vehicle), and the total area
affected, can be generated and output from the QUIC model.
Additional features of the QUIC model were identified that could also be further explored for improving
estimates of exposure and dose for biological agents following an outdoor release. One example is the
resuspension feature which uses the deposition output from a release simulation as input to a second
model run for incorporating resuspension.
Finally, although there would be an advantage to having one modeling tool that integrates simulation of
dispersion, concentrations, exposures and dose, as well as response, it is important to also consider other
sources of concentration data that could be used as input to SHEDS and whether further development of
SHEDS for biological agents should be oriented toward application with other data or models. One
possibility is application of SHEDS together with screening level models used for emergency response
such as HP AC (DTRA, 2015) which do not have the extensive input data requirements of the QUIC
model.
5. Conclusions
Current research supports the need for models for emergency preparedness that not only include
dispersion modeling but also estimate doses based on exposure duration, breathing rate, lung volume,
and particle size distribution, as well as dose-response models to estimate infection probabilities (Van
Leuken et al, 2016). The SHEDS model provides many features that could be useful when applied to
wide-area releases of biological agents to address this research need. The SHEDS model could be
refined to provide output that informs air sample placement based on population exposure estimates to
help prioritize areas for decontamination when applied with other models such as QUIC. Other potential
uses of the SHEDS model could include estimating residual risk for populations outside the primary
contaminated zone following an event, estimating the risk of exposure to resuspended biological agents
that remain after the release, or reconstruction of exposures in a post-event analysis. Future research
should focus on maximizing the utility the exposure modeling approach either through modification of
the current SHEDS model or enhancing the exposure and dose estimation methods within other
modeling tools.
27
-------
References
Aubrecht, C., Ozceylan, D., Steinnocher, K., Freire, S. (2013) Multi-level geospatial modeling of human
exposure patterns and vulnerability indicators. Nat Hazards. 68: 147
Berrocal, V. J., Gelfand, A. E., Holland, D. M., Burke, J. and Miranda, M. L. (2011) On the use of a
PM2.5 exposure simulator to explain birth weight. Environmetrics, 22: 553-571.
Bhaduri, B., Bright, E., Coleman, P., Urban, M.L. (2007) LandScan USA: a high-resolution geospatial
and temporal modeling approach for population distribution and dynamics. GeoJournal 69: 103-117.
Brown, M. (2014) QUIC: A Fast, High-Resolution 3D Building-Aware Urban Transport and Dispersion
Modeling System. EM Magazine April issue, http://awma.org/publications/em-magazine/archive
Burke, J. and R. Vedantham (2009) User Guide Stochastic Human Exposure and Dose Simulation for
Particulate Matter (SHEDS-PM) Version 3.5.
Cao, Y., Frey, H. C. (2011). Geographic differences in inter-individual variability of human exposure to
fine particulate matter. Atmospheric Environment, 45(32), 5684-5691.
DTRA (2015) Hazard Prediction and Assessment Capability (HPAC) Version 6.4 available from
Defense Threat Reduction Agency (DTRA), Ft. Belvoir, VA.
Guha, S., Hariharan, P., Myers, M. R. (2014) Enhancement of ICRP's Lung Deposition Model for
Pathogenic Bioaerosols. Aerosol Science and Technology, 48:12, 1226-1235.
Inglesby T., Dennis D., Henderson D., Bartlett, J., Ascher, M., Eitzen, E., Fine, A., Friedlander, A.,
Hauer, J., Koerner, J., Layton, M., McDade, J., Osterholm, M., O'Toole, T., Parker, G., Perl, T., Russell,
P., Schoch-Spana, M., Tonat, K. for the Working Group on Civilian Biodefense (2000) Plague as a
Biological Weapon: Medical and Public Health Management. JAMA. 283(17): 2281-2290.
Inglesby, T., O'Toole, T., Henderson, D., Bartlett, J., Ascher, M., Eitzen, E., Friedlander, A.,
Gerberding, J., Hauer, J., Hughes, J., McDade, J., Osterholm, M., Parker, G., Perl, T., Russell, P., Tonat,
K. for the Working Group on Civilian Biodefense (2002) Anthrax as a Biological Weapon, 2002
Updated Recommendations for Management. JAMA. 288(17): 2236-2252.
Isakov, V., Touma, J., Burke, J., Lobdell, D., Palma, T., Rosenbaum, A., Ozkaynak, H. (2009)
Combining regional- and local-scale air quality models with exposure models for use in environmental
health studies. Journal of the Air & Waste Management Association, 59: 461-472.
Isukapalli, S. S., P. J. Lioy, and P. G. Georgopoulos (2008) Mechanistic modeling of emergency events:
Assessing the impact of hypothetical releases of anthrax. Risk Analysis, 28(3): 723-740.
Nicogossian, A., Schintler, L., Boybeyi, Z. (2011) Modeling Urban Atmospheric Anthrax Spores
Dispersion: Assessment of Health Impacts and Policy Implications. World Medical & Health Policy,
3(3): Article 5.
Pires, B., Korkmaz, G., Ensor, K., Higdon, D., Keller, S., Lewis, B., Schroeder, A. (2018) Estimating
individualized exposure impacts from ambient ozone levels: A synthetic information approach.
Environmental Modelling & Software, 103:146-57.
28
-------
Stuart, A.L. and D.A. Wilkening (2005) Degradation of Biological Weapons Agents in the Environment:
Implications for Terrorism Response. Environ. Sci. Technol. 39: 2736-2743.
Van Leuken, J.P.G., Swart, A.N., Havelaar, A.H., Van Pul, A., Van der Hoek, W., Heederik, D. (2016)
Atmospheric dispersion modelling of bioaerosols that are pathogenic to humans and livestock - A
review to inform risk assessment studies. Microbial Risk Analysis. 1: 19-39.
Xue, J., McCurdy, T., Burke, J., Bhaduri, B., Liu, C., Nutaro, J., Patterson, L. (2010) Analyses of school
commuting data for exposure modeling purposes. Journal of Exposure Analysis and Environmental
Epidemiology, 20: 69-78.
29
-------
*>EPA
United States
Environmental Protection
Agency
PRESORTED
STANDARD POSTAGE
& FEES PAID EPA
PERMIT NO. G-35
Office of Research and
Development (8101R)
Washington, DC 20460
Offal Business
Penalty for Private Use
$300
------- |