United States Science Advisory EPA-SAB-EC-00-015
Environmental Board (1400A) August 2000
Protection Agency Washington DC iviviv.epa.gov/sab
&EPA REVIEW OF DRAFT AIR
TOXICS MONITORING
STRATEGY CONCEPT
PAPER
REVIEW BY THE AIR TOXICS
MONITORING SUBCOMMITTEE OF
THE SCIENCE ADVISORY BOARD
-------
August 18, 2000
EPA-SAB-EC-00-015
Honorable Carol M. Browner
Administrator
U.S. Environmental Protection Agency
1200 Pennsylvania Ave.
Washington, D.C. 20460
RE: Review of Draft Air Toxics Monitoring Strategy Concept Paper
Dear Ms. Browner:
On March 29-30, 2000, the Science Advisory Board's (SAB's) Air Toxics Monitoring
Subcommittee of the SAB Executive Committee reviewed the February 29, 2000 drafts of the Air
Toxics Monitoring Strategy Concept Paper and the Protocol for Model-to-Monitor Comparisons. The
Office of Air Quality, Planning, and Standards (OAQPS) prepared both documents as part of the
National Air Toxics Assessment (NATA). The accompanying SAB report responds to charge
questions concerning the air toxic monitoring objectives and principles, the phased strategy for the
design of a national air toxics network, and the Model-to-Monitor evaluation protocol.
In briefest terms, the Agency is taking a sound, scientific approach with the available resources.
OAQPS has appropriately decided to address a limited number of objectives. They are approaching
these objectives in a logical, informed and step-wise fashion. The Subcommittee expects that
systematic planning will continue to be done as strategies are phased in to allow optimal use of available
resources. It is crucially important not to spread the available resources so thinly that nothing can be
done well. Therefore, the Subcommittee's additional suggestions for valuable work should be
considered in the event that additional resources become available. These suggestions are included in
this report.
In summary, the Subcommittee finds that the concept paper presents a reasonable phased
strategy to design a national air toxics network. The Agency has identified the most important uses for
ambient air toxics data and the Subcommittee offered suggestions for other types of air toxics
monitoring data or data uses. As first steps in the design of the national network, the Subcommittee
endorses the Agency's data analysis of existing air toxics monitoring data, coupled with focused pilot
studies for a core set of hazardous air pollutants (HAPs) in a limited number of areas. Overall, the
Subcommittee supports the goals of the Model-to-Monitor evaluation protocol and has provided a
number of specific comments for applications of the protocol.
-------
The Subcommittee wishes to highlight certain findings concerning the existing documents:
a) The Subcommittee commends the Agency for the quality of both the Air Toxics
Monitoring Concept Paper and the Model-to-Monitor strategy. A great deal of careful
thought and evaluation has gone into this process. Both reports are terse, clear, and
well written.
b) Although the Agency has identified the most important uses for ambient air toxics data,
the Subcommittee identified additional types of air toxics monitoring data and data uses.
c) The concept paper presents a reasonable phased strategy to design a national air toxics
network.
d) With only minor modifications, the Agency's strategy for neighborhood-scale sampling
over a 24-hour period is appropriate.
e) The Agency's data analysis of existing air toxics monitoring data, coupled with focused
pilot studies for a core set of hazardous air pollutants (HAPs) in a limited number of
areas, are appropriate first steps in the design of the national network.
The Subcommittee encourages the use of tools, such as the data quality objective
process, to improve the relevance and reliability of the exposure information required
f>
by the Agency.
g) Overall, the Subcommittee supports the goals of the Model-to-Monitor evaluation
protocol and provides a number of specific comments for improving the application of
the protocol.
The Subcommittee also commented on the possibilities for an expanded study because there
are a number of areas in which the monitoring activities and data evaluation could be improved or
expanded, if more resources are made available.
a) A well-defined and consistent scientific framework addressing both modeling and
measurement problems for hazardous air pollutants would be valuable.
b) It would be useful to understand how well monitoring and modeling define source-to-
concentration relationships for certain types of pollutants.
c) It is not clear how outdoor ambient air measurements are to be related to population
exposure, which occurs mostly indoors. Consideration should be given to parallel pilot
studies on indoor air monitoring, using techniques having similar detection levels as
-------
those in the proposed outdoor study, in order to establish indoor/outdoor relationships
that would permit better estimation of indoor exposures and the apportionment of the
outdoor contributions to the exposures.
d) If multimedia pollutants were included in both the monitoring and modeling framework,
additional exposure routes could be considered. This consideration is important
because several classes of HAPs are multimedia pollutants; for example, metals and
semi-volatile organic compounds that are transferred through food chains, and for
which the inhalation route plays a relatively minor role.
In the future, NATA and the monitoring data it provides will be a resource of enormous
scientific value. Understanding air toxics in the environment is an important area where much has been
achieved and yet much remains to be done. Because currently the Agency is rich in models and poor in
data, data collection efforts such as this should be given high priority. Therefore, the Subcommittee
hopes that the Congress and the Agency will continue to provide the resources to OAQPS to expand
this program and that the Agency continues to collaborate with state and local agencies.
The data from this proposed study are likely to be used for additional analyses, which, if done
well, can be very valuable. As an example, environmental justice programs will be interested in air
quality monitoring at the neighborhood scale. With respect to facility siting and human exposure,
disadvantaged communities often focus on air quality because air is the medium in which contaminants
are most mobile. An earlier SAB report on secondary use of data (EPA 1999) and the Agency's new
Office of Environmental Information may provide useful guidance for the appropriate secondary use of
data. Other potential secondary uses are identified in the attached report.
The Agency is now launching a serious effort to measure hazardous air
pollutants in ambient air. The Agency and the SAB anticipate that the resulting
information on concentrations will improve the scientific basis for
understanding exposure to these chemicals and the resulting risk. However,
in addition to air concentration data, estimates of exposure and risk also depend upon assumptions and
models. The SAB reminds the Agency that future advances in modeling and risk assessment methods
may permit further improvements in its analyses by incorporating more realistic and credible unit risk
factors as well as the new air concentration data.
We appreciate the opportunity to provide advice on this effort at this early stage. The Agency
staff was open, collegial, and cognizant of capabilities and limitations of the concept paper. They were
also accepting of and responsive to the Subcommittee's suggestions. The SAB is open to reviewing
-------
elements of this monitoring strategy at appropriate stages of its development. We look forward to the
response of the Assistant Administrator for Air and Radiation.
Sincerely,
/s/
Dr. Morton Lippmann, Interim Chair
Science Advisory Board
/s/
Dr. Thomas McKone, Chair
Air Toxics Monitoring Subcommittee
Science Advisory Board
-------
NOTICE
This report has been written as part of the activities of the Science Advisory Board, a public
advisory group providing extramural scientific information and advice to the Administrator and other
officials of the Environmental Protection Agency. The Board is structured to provide balanced, expert
assessment of scientific matters related to problems facing the Agency. This report has not been
reviewed for approval by the Agency and, hence, the contents of this report do not necessarily
represent the views and policies of the Environmental Protection Agency, nor of other agencies in the
Executive Branch of the Federal government, nor does mention of trade names or commercial products
constitute a recommendation for use.
Distribution and Availability: This Science Advisory Board report is provided to the EPA
Administrator, senior Agency management, appropriate program staff, interested members of the
-------
public, and is posted on the SAB website (www.epa.gov/sab). Information on its availability is also
provided in the SAB's monthly newsletter (Happenings at the Science Advisory Board). Additional
copies and further information are available from the SAB Staff.
ABSTRACT
On March 29-30, 2000, the Science Advisory Board's (SAB's) Air Toxics Monitoring
Subcommittee of the SAB Executive Committee conducted a peer review of the February 29, 2000
drafts of the Air Toxics Monitoring Strategy Concept Paper and the Protocol for Model-to-Monitor
Comparisons. Both documents are part of the National Air Toxics Assessment (NAT A) and were
prepared by the Office of Air Quality, Planning, and Standards (OAQPS).
The Subcommittee commended the Agency for their effort in developing both the Air Toxics
Monitoring Concept Paper and the Monitoring-to-Models strategy. The concept paper presents a
reasonable phased strategy to design a national air toxics network. The Subcommittee endorses the
Agency's data analysis of existing air toxics monitoring data, coupled with focused pilot studies for a
core set of hazardous air pollutants (HAPs) in a limited number of areas. The Agency has identified the
most important uses for ambient air toxics data and the Subcommittee offered suggestions for other
types of air toxics monitoring data and/or data uses.
NATA and the monitoring data it provides will be a resource of enormous scientific value. The
Subcommittee identified a number of areas in which the monitoring activities and data evaluation could
be improved or expanded, particularly if more resources are made available.
Keywords: hazardous air pollutants, air toxics, monitoring, NATA.
-------
TABLE OF CONTENTS
1. EXECUTIVE SUMMARY 1
2. INTRODUCTION 2
2.1 Background 2
2.2 Charge 4
2.3 SAB Review Process 5
3. RESPONSES TO THE CHARGE 7
3.1 Summary of the Subcommittee Responses 8
3.1.1 Question 1: Does the air toxics monitoring concept paper describe appropriate
air toxic monitoring objectives and principles, particularly ones that will permit
the collection
of monitoring data to support the initial National Air
Toxics Assessment activities? 8
3.1.1.1 Question 1 (a): Does the Subcommittee believe
that the EPA has identified the most important uses
for ambient air toxics data? Are there other types of
air toxics monitoring data or data uses that should
also be identified for near-term or for future air toxics
monitoring activities? 9
3.1.1.2 Question l(b): Does the Subcommittee believe that
neighborhood-scale monitoring is appropriate for evaluating ASPEN air
quality predictions and later for developing
long-term ambient air quality trends? Are there other
appropriate monitoring scales, perhaps for other data uses,
that the Subcommittee would suggest? 10
3.1.1.3 Question l(c): Does the Subcommittee believe that a basic
24-hour sample taken at a frequency sufficient to fulfill the objectives of
the program is adequate to provide the model
reality check and supply data for the characterization of
ambient hazardous air pollutant (HAPs) concentrations? In particular,
what are the Subcommittee's thoughts on the use
of 24-hour samples collected once in 12 days for model
evaluation and at a more frequent (say, l-in-6 or l-in-3 day) schedule
in the future for trends assessment at permanently
located monitoring sites? 11
3.1.2 Question 2: Does the air toxics monitoring concept paper present a reasonable
phased strategy to design a national air toxics network? 11
3.1.2.1 Question 2(a): Does the Subcommittee believe that data
iii
-------
analyses of existing air toxics monitoring data, coupled with focused
pilot studies for a core set of HAPs in a limited number
of areas are appropriate first steps in the design of the
national network. What additional or alternative approaches
are suggested? 11
3.1.2.2 Question 2(b): Given that the State and local agencies
have been measuring ambient air toxics with the Toxic Organic (TO)
and Inorganic (IO) methods as described in the paper,
are these methods appropriate for the continued routine
monitoring of the target Urban Air Toxics Strategy compounds
in a national monitoring network? If not, are there alternative methods
that the Subcommittee would recommend? 12
3.1.3 Question 3 12
3.1.3.1 Question 3(a): Do the data analysis approaches provide
enough information to allow appropriate interpretations of
model results to support the development of model
improvements in the future and to assist with the design of the national
monitoring network? 12
3.1.3.2 Question 3(b): Are there some HAPs for which these
approaches appear inadequate? If so, can the Sub-committee suggest
alternative approaches for these? 13
3.1.3.3 Question 3(c): As noted in the paper, annual-average concentrations
and comparisons to modeled estimates can be uncertain when a large
percentage of the measurements are
below the method detection limit (MDL). To estimate annual-average
concentrations from monitoring data, EPA generally substitutes half the
MDL. Does the Subcommittee suggest any alternative statistical
approaches? 14
3.2 Detailed Responses to Question 1 14
3.2.1 Question l(a) 16
3.2.1.1 Current Primary Uses for Air Toxics Data 16
3.2.1.2 Potential Secondary Uses of Air Toxics Data 17
3.2.1.3 Other near-term and future uses for data 18
3.2.2 Question l(b): Does the Subcommittee believe that neighborhood-scale
monitoring is appropriate for evaluating ASPEN air quality predictions and later
for developing long-term ambient air quality trends? Are there other
appropriate monitoring scales, perhaps for other data uses, that the
Subcommittee would suggest? 19
3.2.2.1 The Problem With Rural Census Tracts 20
3.2.2.2 Co-Location With Other Monitors 20
IV
-------
3.2.3 Question l(c): Does the Subcommittee believe that a basic 24-hour sample
taken at a frequency sufficient to fulfill the objectives of the program is adequate
to provide the model reality check and supply
data for the characterization of ambient hazardous air pollutant (HAPs)
concentrations? In particular, what are the Subcommittee's thoughts
on the use of 24-hour samples collected once in 12 days for model evaluation
and at a more frequent (say, l-in-6 or l-in-3 day) schedule in the future for
trends assessment at permanently located monitoring
sites? 20
3.2.3.1 Time Resolution of Samples 21
3.2.3.2 Seasonal Variation and Annual Average 22
3.3 Detailed Responses to Question 2: Does the air toxics monitoring concept
paper present a reasonable phased strategy to design a national air toxics network? . 22
3.3.1 Question 2(a): Does the Subcommittee believe that data analyses of existing air
toxics monitoring data, coupled with focused pilot studies
for a core set of HAPs in a limited number of areas are appropriate
first steps in the design of the national network. What additional or alternative
approaches are suggested? 22
3.3.1.1 Data Gaps and Pilot Studies 23
3.3.1.2 Mobile and Stationary Sources 23
3.3.1.3 Personal Monitors 23
3.3.1.4 Conveying Uncertainty 23
3.3.2 Question 2(b): Given that the State and local agencies have been measuring
ambient air toxics with the Toxic Organic (TO) and
Inorganic (IO) methods, as described in the paper, are these methods
appropriate for the continued routine monitoring of the target Urban
Air Toxics Strategy compounds in a national monitoring network?
If not, are there alternative methods which the Subcommittee would
recommend? 24
3.3.2.1 Matching the Method to the Need 24
3.3.2.2 Selecting the Appropriate Method 25
3.4 Detailed Responses to Question 3 on Model-to-Monitor Comparison 27
3.4.1 Question 3(a): Do the data analysis approaches provide enough information to
allow appropriate interpretations of model results to support the development
of model improvements in the future and to assist with the design
of the national monitoring network? 27
3.4.1.1 Proposed Data Analysis Methods 27
3.4.1.2 Use of Stratification 28
3.4.1.3 Model Diagnostics and Model Reliability 29
3.4.1.4 Alternate Methods for Comparing Models to
Measurements 30
-------
3.4.2 Question 3(b): Are there some HAPs for which these approaches
appear inadequate? If so, can the Subcommittee suggest alternative approaches
for these? 30
3.4.3 Question 3(c): As noted in the paper, annual-average concentrations
and comparisons to modeled estimates can be uncertain when a large
percentage of the measurements are below the method detection limit (MDL).
To estimate annual-average concentrations from monitoring
data, EPA generally substitutes half the MDL. Does the Subcommittee suggest
any alternative statistical approaches? 31
3.4.3.1 The Importance of Dealing With Measurements Below
theMDL 31
3.4.3.2 Cautions 32
3.4.3.3 Alternate Statistical Approaches for Data Below the MDL 33
GLOSSARY G-l
REFERENCES R-l
APPENDIX A: Exposure Measurement Issues A-1
APPENDIX B: Summary of Elements of the EPA Quality System and an
Introduction to the Data Quality Objectives Process B-l
APPENDIX C: Rosters B-3
VI
-------
1. EXECUTIVE SUMMARY
On March 29-30, 2000, the Science Advisory Board's (SAB's) Air Toxics Monitoring
Subcommittee of the SAB Executive Committee reviewed the February 29, 2000 drafts of the Air
Toxics Monitoring Strategy Concept Paper and the Protocol for Model-to-Monitor Comparisons. The
Office of Air Quality, Planning, and Standards (OAQPS) prepared both documents in support of the
National Air Toxics Assessment (NATA) program.
The Subcommittee commends the Agency for the documents. A great deal of careful thought
and evaluation has gone into this process with the result that both reports are terse, clear, and well-
written.
The Agency has identified the most important uses for ambient air toxics data, and the concept
paper presents a reasonable phased strategy to design a national air toxics network. The Subcommittee
endorses the Agency's data analysis of existing air toxics monitoring data and focused pilot monitoring
studies for a core set of hazardous air pollutants (HAPs) in a limited number of areas as first steps in the
design of the national network. With minor changes, the Subcommittee supports the goals of the
Model-to-Monitor evaluation protocol and the Agency's strategy for neighborhood-scale sampling over
24-hour periods.
The NATA and the monitoring data it provides will be a resource of enormous scientific value
for understanding air toxics in the environment. This is an important area where much has been
achieved and yet much remains to be done. Because the Agency is currently rich in models and poor in
data, data collection efforts such as this are very important. Therefore, the Subcommittee identified a
number of areas where the monitoring activities and data evaluation could be improved or expanded,
particularly if more resources are made available.
The Subcommittee expressed concern about the initial exclusion of multimedia pollutants from
both the monitoring and modeling framework and the related factor that only inhalation is considered as
an exposure route. Even though multimedia pollutants are not the first priority of an air-monitoring
program, several important classes of HAPs are multimedia pollutants. Metals and semi-volatile
organic compounds, for example, are transferred through food chains; for these, inhalation is not the
primary direct route of exposure. By excluding such compounds from the first phase of the program,
there is some chance they could be excluded over the long-term because absence of information could
be interpreted as the absence of a problem. Thus, the Subcommittee recommends a process that
provides a continuing incentive for including multimedia and multi-pathway concentration and exposure
data.
The Subcommittee also recommends the development and use of tools, such as the Data
Quality Objective Process, to link sampling strategies to the relevance and reliability of the exposure
information required by the Agency.
-------
The Subcommittee hopes that the Congress and the Agency will continue to provide the
resources to OAQPS to expand this program. It is also important to continue the Agency's
collaborations with state and local agencies.
-------
2. INTRODUCTION
2.1 Background
The Clean Air Act (CAA) regulates 188 hazardous air pollutants (HAPs), also called "air
toxics", because they have been associated with a wide variety of adverse health effects. These air
toxics are emitted from multiple sources and result in population and ecosystem exposure. Typically,
people experience exposures to multiple HAPs from many sources. Exposures of concern result not
only from the inhalation of these HAPs, but also, for some HAPs, from multi-pathway exposures to air
emissions. For example, air emissions of mercury are deposited in water affecting fish and those people
who are exposed to mercury through their consumption of contaminated fish.
One of EPA's goals is to reduce air toxics emissions by 75% from 1993 levels. When tools are
available to assess the residual risk, EPA plans to modify that goal to focus on reducing risks associated
with exposure to air toxics. EPA's long-term goal is to eliminate unacceptable risks of cancer and other
significant health problems resulting from exposures to air toxics emissions and to substantially reduce
or eliminate adverse effects on our ecological resources.
To meet these goals, EPA has developed an Air Toxics Program (ATP) to characterize,
prioritize, and equitably address the impacts of HAPs on the public health and the environment. The
ATP seeks to address air toxics problems through a strategic combination of agencies, activities and
authorities, including regulatory approaches and voluntary partnerships. It includes four elements:
a) Source-specific standards and sector-based standards, including Section 112
standards; i.e., Maximum Achievable Control Technology (MACT), Generally
Achievable Control Technology (GACT), residual risk standards, and Section 129
standards.
b) National, regional, and community-based initiatives to focus on multi-media and
cumulative risks, such as the Integrated Urban Air Toxics Strategy, Great Waters,
Mercury initiatives, Persistent Bioaccumulative Toxics (PBT) and Total Maximum Daily
Load (TMDL) initiatives, and Clean Air Partnerships.
c) National Air Toxics Assessment (NATA) that will help EPA identify areas of concern,
characterize risks, and track progress. These activities include expanded air toxics
monitoring, improving and periodically updating emissions inventories, national-and
local-scale air quality and exposure modeling, and continued research on effects and
assessment tools, leading to improved characterizations of air toxics risk and reductions
in risk resulting from ongoing and future implementation of air toxics emissions control
standards and initiatives.
-------
d) Education and outreach.
The ATP depends on quantifying the impacts of air toxics emissions on public health and the
environment. Therefore, EPA has initiated a National Air Toxics Assessment (NATA) to provide the
best technical information regarding air toxics emissions, ambient concentrations, and health impacts to
support the development of sound policies in the ATP. These activities include:
a) measurement of air toxics emission rates from individual pollution sources;
b) compilation of comprehensive air toxics emission inventories for local, State, and
national domains;
c) measurement of ambient concentrations of air toxics at monitoring sites throughout the
nation(urban and rural);
d) analyses of patterns and trends in ambient air toxics measurements;
e) estimation of ambient and multimedia air toxics concentrations from emission
inventories, using dispersion and deposition modeling;
f) estimation of human and environmental exposures to air toxics;
g) assessment of risks due to air toxics; and
h) ongoing research in the above areas to improve assessments over time.
Emissions data, ambient concentration measurements, modeled estimates, and health and
environmental impact information are all needed to fully assess air toxics impacts and to characterize
risk. Specifically, emissions data are needed to quantify the sources of air toxics and aid in the
development of control strategies. Ambient monitoring data can be used to evaluate the atmospheric
dispersion and deposition that describe the fate and transport of air toxics in the atmosphere. Because
ambient measurements cannot be made everywhere, modeled estimates are used to extrapolate to
locations without monitors. A combination of reliable modeling systems and a well-designed ambient
network is thought to be the best approach for estimating ambient concentrations and population and
ecosystem exposure across the nation.
Exposure assessments and health effects information integrate all of these data into an
understanding of the implications of air toxics impacts and to characterize air toxics risks. Ambient
measurements provided by routine monitoring programs, together with personal exposure
measurements obtained from ongoing research studies, are important to evaluate these air quality and
exposure models.
-------
The Air Toxics Monitoring Strategy Concept Paper and related documents were drafted and
submitted for review to provide a logical and scientifically strong basis to meet these data needs.
2.2 Charge
The focus of the present SAB review was to evaluate the adequacy of the February 29, 2000,
drafts of the Air Toxics Monitoring Strategy Concept Paper and the Protocol for Model-to-Monitor
Comparisons. The former describes a phased approach towards meeting the monitoring objectives of
the Air Toxics Program. Both documents are part of the National Air Toxics Assessment and were
prepared by the Office of Air Quality, Planning, and Standards. The Agency asked the SAB to focus
on three specific questions.
1. Does the air toxics monitoring concept paper describe appropriate air
toxic monitoring objectives and principles, particularly ones that will permit the
collection of monitoring data to support the initial National Air Toxics
Assessment activities? Specifically,
a) Does the Subcommittee believe that [OAQPS] identified the most
important uses for ambient air toxics data? Are there other types
of air toxics monitoring data or data uses that should also be
identified for near-term or for future air toxics monitoring
activities?
b) Does the Subcommittee believe that neighborhood-scale
monitoring is appropriate for evaluating ASPEN air quality
predictions and later for developing long-term ambient air quality
trends? Are there other appropriate monitoring scales, perhaps
for other data uses, that the Subcommittee would suggest?
c) Does the Subcommittee believe that a basic 24-hour sample taken
at a frequency sufficient to fulfill the objectives of the program is
adequate to provide the model reality check and supply data for
the characterization of ambient hazardous air pollutant (HAPs)
concentrations? In particular, what are the Subcommittee's
thoughts on the use of 24-hour samples collected once in 12 days
for model evaluation and at a more frequent (say, l-in-6 or l-in-3
day) schedule in the future for trends assessment at permanently
located monitoring sites?
2. Does the air toxics monitoring concept paper present a reasonable phased
strategy to design a national air toxics network?
-------
a) Does the Subcommittee believe that data analyses of existing air
toxics monitoring data, coupled with focused pilot studies for a
core set of HAPs in a limited number of areas are appropriate first
steps in the design of the national network. What additional or
alternative approaches are suggested?
b) Given that the State and local agencies have been measuring
ambient air toxics with the Toxic Organic (TO) and Inorganic (IO)
methods as described in the paper, are these methods appropriate
for the continued routine monitoring of the target Urban Air
Toxics Strategy compounds in a national monitoring network? If
not, are there alternative methods which the Subcommittee would
recommend?
3. In addition to your comments on the overall monitoring strategy, we seek
your advice on the monitor-to-model evaluation protocol.
a) Do the data analysis approaches provide enough information to allow
appropriate interpretations of model results to support the development of
model improvements in the future and to assist with the design of the
national monitoring network?
b) Are there some HAPsfor which these approaches appear
inadequate? If so, can the Subcommittee suggest alternative
approaches for these?
c) As noted in the paper, annual-average concentrations and
comparisons to modeled estimates can be uncertain when a large
percentage of the measurements are below the method detection
limit (MDL). To estimate annual-average concentrations from
monitoring data, EPA generally substitutes one-half the MDL.
Does the Subcommittee suggest any alternative statistical
approaches?
2.3 SAB Review Process
Members of the Subcommittee were recruited from a variety of SAB Standing Committees and
consultants to form a body of reviewers well-acquainted with the work of the SAB's Clean Air
Scientific Advisory Committee, the Drinking Water Committee, the Environmental Engineering
Committee, the Environmental Modeling Subcommittee, the Integrated Human Exposure Committee,
and the Research Strategies Advisory Committee. Various Subcommittee members also served on the
-------
key relevant National Research Council reviews. The purpose of this formation was to provide
continuity and consistency in advice on this important topic.
The Subcommittee met in public session on March 29-30, 2000, in Washington, DC. This
report is based upon written comments prepared before and during the meeting by Subcommittee
members and subsequently edited by the Subcommittee and approved by mail May 30, 2000. The
report was then transmitted to the SAB's Executive Committee, for approval at a public meeting June
30, 2000.
-------
3. RESPONSES TO THE CHARGE
Controlling the exposure of human populations and ecosystems to environmental contaminants
depends upon understanding the links among multiple pollutant sources, exposure pathways, and
adverse effects. This is a large and difficult problem. Both researchers and regulators are aware that
air toxics data needs are real and pressing. Although much has been learned, much remains to be done.
Currently, the Agency is rich in models and poor in data. Therefore, NATA and the monitoring data it
provides will be a resource of enormous scientific value.
Given the size and complexity of the problem, and the paucity of the data, it is not surprising
that the Agency documents contain an ambitious list of goals for the proposed monitoring system and
that the Subcommittee has identified even more. The challenge is in selecting from among these
important needs, a set of goals that are achievable within the resources available.
In this chapter, the Subcommittee, based on members' expertise and experience, provides
guidance on which goals to pursue in which order. The Subcommittee did not make a detailed
consideration of resources. The use of the Agency's Data Quality Objective Process (See Appendix
B.) or other systematic planning process can help managers select among the many worthy and
competing goals because it links the decision(s) to be made, the data and certainty required, and the
project-specific blueprint for obtaining data appropriate for decision-making. Such planning processes
provide a reality check. If the goal cannot be achieved within the resources available, then another
(achievable) goal may be preferable.
In summary, the Subcommittee finds that the concept paper presents a reasonable phased
strategy to design a national air toxics network. The Agency has identified the most important uses for
ambient air toxics data and the Subcommittee offered suggestions for other types of air toxics
monitoring data or data uses. As first steps in the design of the national network, the Subcommittee
endorses the Agency's data analysis of existing air toxics monitoring data, coupled with focused pilot
studies for a core set of hazardous air pollutants (HAPs) in a limited number of areas. Overall, the
Subcommittee supports the goals of the Model-to-Monitor evaluation protocol and has provided a
number of specific comments for applications of the protocol.
The Subcommittee identified a number of areas in which the data evaluation could be improved
or expanded, if more resources were to be made available; these areas include indoor monitoring and
multimedia studies. The proposed program does not address actual population exposures, which
mostly occur indoors. Current personal exposure monitoring methods do not have the sensitivity of
those used in the proposed network. A pilot program to measure indoor levels with similar methods
should be considered along with methods to determine the relationship of outdoor concentrations to
indoor levels.
-------
For many pollutants, transport through air is governed by deposition to and re-emission from
soil, water and vegetation. For the purposes of this report, these are called multi-media pollutants.
EPA's proposed approach excludes multimedia pollutants from both the monitoring and modeling
framework, and only inhalation is considered as an exposure route. Although multimedia pollutants are
not the first priority of an air-monitoring program, several important classes of HAPs are multimedia
pollutants; for example metals and semi-volatile organic compounds that are transferred through food
chains. The Subcommittee is not concerned about this priority, but fears that by excluding such
compounds from the first phase of the program, there is some chance they could be excluded over the
long-term.
Because data collection efforts such as this should be given high priority, the Subcommittee
hopes that the Congress and the Agency will continue to provide the resources to OAQPS to expand
this program.
The following section begins with a summary of the Subcommittee's response and
recommendations for each of the questions. More detailed discussions to support and expand on the
recommendations follow the summaries.
3.1 Summary of the Subcommittee Responses
3.1.1 Question 1: Does the air toxics monitoring concept paper describe appropriate
air toxic monitoring objectives and principles, particularly ones that will permit
the collection of monitoring data to support the initial National Air Toxics
Assessment activities?
The Subcommittee found that the monitoring objectives and principles described in the concept
paper are appropriate and will permit the collection of the data necessary to support the initial NATA
activities. However, the Subcommittee recommends that to better meet their specific aims, the EPA
should develop a detailed, step-wise process that can clarify how they will achieve the primary
objectives of the program. In this process, careful consideration should be given to the following issues:
a) Prioritize activities and milestones needed to achieve the primary objectives of NATA,
following the Government Performance and Results Act (GPRA) model.
b) Develop a set of criteria for judging the adequacy of the network in terms of the
objectives.
c) Develop data quality criteria for each of the activities.
d) Establish acceptable levels of data uncertainties (measurement errors) and model
estimate uncertainties for both extant data and the new data to be collected.
-------
e) Develop a process for including potential modification(s) of the plan in order to
address unforeseen events or new findings.
f) Involve those EPA regional offices as well as local/state authorities that will be
expected to contribute to NATA activities as much as possible, from the initial stages of
plan development.
g) Include discussion of guidelines for siting monitors in rural areas in the revised Concept
Paper.
3.1.1.1 Question l(a): Does the Subcommittee believe that the EPA has identified the
most important uses for ambient air toxics data? Are there other types of air toxics
monitoring data or data uses that should also be identified for near-term or for future
air toxics monitoring activities?
The Subcommittee believes that the document identifies the most important uses of the ambient
air toxics data as they relate to the main objectives of NATA. As with any other data collection effort,
there will be unforeseen, future uses of these data. The Subcommittee identified the following near-term
and future uses of these data:
a) Identification of unpermitted and unreported emissions.
b) Input to permitting and siting decisions for new facilities.
c) "Reality check" on actual emission reductions vs. model-derived estimates.
d) Support and evaluation of the impact of EPA and local/state programs.
e) Evaluation tool for other models that include or require air toxics information, including
multi-media and cross-media exposure models.
However, the Subcommittee notes that some of these potential uses could contradict Agency
policy as specified in 40 CFR Part 51 of April 21, 2000, section 10.2.2 and thus could be
inappropriate uses of current source-receptor modeling techniques. Because of the foreseen and
unforeseen uses of these data, the Subcommittee recommends that:
a) input be obtained from other Agency programs that deal with multiple uses of data,
including the Quality System (See Appendix B for description) and the efforts on
secondary uses of data.
10
-------
b) EPA develop a careful description of the data collected as well as a description of the
capabilities and limitations of the data.
c) Utilize the Pilot Study to gain needed information on appropriate sampling frequency
and time-resolution (for example, day/night) sampling.
d) EPA consider using multimedia monitoring for persistent air toxics because such
monitoring would be useful in evaluating models that include or require air toxics
information.
e) EPA collaborate with other on-going exposure and health surveys as a means on
increasing the potential uses of the data.
3.1.1.2 Question l(b): Does the Subcommittee believe that neighborhood-scale
monitoring is appropriate for evaluating ASPEN air quality predictions and later for
developing long-term ambient air quality trends? Are there other appropriate
monitoring scales, perhaps for other data uses, that the Subcommittee would suggest?
The Subcommittee found that the neighborhood-scale monitoring approach is generally
appropriate for the evaluation of ASPEN estimates and long-term air quality trends. However, because
of the national scope of the program and the diversity of sites, this scale may not be applicable in all
cases and may need to be modified. The Subcommittee recommends that:
a) Neighborhood-scale sampling remain the main focus of the strategy. However, over the
longer term, the monitoring scale used for toxic air pollutants should be guided by both
the objectives of the monitoring program and the characteristics of the pollutants being
monitored
b) The Agency should consider a micro-scale-type emission site in the Pilot study to
assess variability within a neighborhood scale. However, because of the increase of
sampling required by such an effort and the likely additional demand on resources; this
situation may be more efficiently addressed through modeling in the main phase of the
program.
c) Assuring that modern laboratory equipment (in some cases less than five years old) be
used for analysis of samples.
d) Ecological resource exposures should also be considered, in addition to the human
exposure focus.
11
-------
e) Co-locating monitors at existing monitoring locations used for other purposes should be
evaluated carefully in the context of the objectives of NATA.
12
-------
3.1.1.3 Question l(c): Does the Subcommittee believe that a basic 24-hour sample
taken at a frequency sufficient to fulfill the objectives of the program is adequate to
provide the model reality check and supply data for the characterization of ambient
hazardous air pollutant (HAPs) concentrations? In particular, what are the
Subcommittee's thoughts on the use of 24-hour samples collected once in 12 days for
model evaluation and at a more frequent (say, l-in-6 or l-in-3 day) schedule in the
future for trends assessment at permanently located monitoring sites?
The Subcommittee believes that the one in twelve day, 24-hour sampling frame, while
reasonable in principle, may not be consistent with the objectives of NAT A for all compounds and
foreseen uses of the data. As with the issue of spatial location of sites, the development of data quality
criteria should assist in this process. The Subcommittee recommends that EPA:
a) Develop a multi-tier sampling frame after careful consideration of the objectives of
NAT A, the data quality criteria, the nature of the air toxic (or class of compound) and
the temporal variability of the emission sources, and the specific uses of the data.
b) Consider also future uses of the data in developing the temporal sampling time frame.
3.1.2 Question 2: Does the air toxics monitoring concept paper present a reasonable
phased strategy to design a national air toxics network?
3.1.2.1 Question 2(a): Does the Subcommittee believe that data analyses of existing
air toxics monitoring data, coupled with focused pilot studies for a core set of HAPs in
a limited number of areas are appropriate first steps in the design of the national
network. What additional or alternative approaches are suggested?
The Subcommittee determined that the proposed phased strategy for NAT A is reasonable as
an initial approach to developing the network. However, the Subcommittee suggested alternate
approaches for evaluating the information derived from the program. The Subcommittee noted that
there should be careful definition of the "decision unit" to which the samples are applied. In addition,
the subcommittee observed that there is a significant leap between ambient air concentration
measurements and estimation of exposures, which requires also indoor concentration data and
information on time-activity patterns. Since these data will not be collected, the Subcommittee
recommends that EPA consider the need for developing penetration factors in parallel research
programs. (Penetration factors help establish the relationship between indoor and outdoor air quality.)
13
-------
3.1.2.2 Question 2(b): Given that the State and local agencies have been measuring
ambient air toxics with the Toxic Organic (TO) and Inorganic (IO) methods as
described in the paper, are these methods appropriate for the continued routine
monitoring of the target Urban Air Toxics Strategy compounds in a national monitoring
network? If not, are there alternative methods that the Subcommittee would
recommend?
The Subcommittee agrees with EPA that the TO and IO methods are generally appropriate for
use in the national monitoring network. However, the Subcommittee noted some limitations to these
methods. Some of these limitations are compound-specific, and others relate to the resources required
to meet the objectives and stated sampling goals. The Subcommittee recommends that the Agency
consider:
a) Evaluating each method in light of the data quality criteria and the objectives of each
activity and stage of the program withing the framework of the available resources, and
use less expensive alternatives when appropriate.
b) Considering alternatives to TO or IO methods for specific compounds for which these
methods are known to be inadequate( e.g., DNPH-based methods for acrolein or
SUMMA canisters for polar compounds).
c) Developing a clear set of criteria that will guide the selection of alternate methods in the
future as sampling/analytical methodology evolves.
3.1.3 Question 3. The Monitor-to-Model Evaluation Protocol
3.1.3.1 Question 3(a): Do the data analysis approaches provide enough information to
allow appropriate interpretations of model results to support the development of model
improvements in the future and to assist with the design of the national monitoring
network?
The Subcommittee generally agrees with the proposed data analysis approach as an initial step.
However, the Subcommittee notes that any effort at model validation will be limited by a number of
factors, including: lack of multimedia measurements for relevant air toxics; the limitation of the analytical
methods for measuring certain HAPS; and the well-recognized limitations of emission inventories. The
Subcommittee is very concerned about the exclusion of models for multi-media pollutants that move
and accumulate in soil, sediment, water, and the food chain, and the consequences of this exclusion for
achieving the objectives of the program. The Subcommittee recommends that the Agency:
a) Develop specific performance assessment criteria for the comparison of model
estimates and measurements.
14
-------
b) Perform Monte Carlo simulation of model outputs as a means of parametrizing the
model estimates for future comparison with measurements.
c) Prioritize the data analysis approaches from the least-sophisticated but more easily
understood and conveyed (e.g., rank correlation) to the more sophisticated but less
easy to explain (e.g., tests of medians and quartiles).
d) Further develop the stratification approach to data analysis to include source categories,
type of source, type of terrain and meteorological variables.
e) Develop a set of metrics to evaluate in detail the potential causes for
disagreement between model estimates and measurements.
f) Consider that persistent HAPS may have different dispersive characteristics
g) Start to develop support for inclusion of multi-media monitoring and modeling efforts.
h) Consider use of additional statistical analysis methods such as multivariate regression
and test of medians and quartiles.
3.1.3.2 Question 3(b): Are there some HAPs for which these approaches appear
inadequate? If so, can the Subcommittee suggest alternative approaches for these?
The Subcommittee noted that there are HAPS for which the Model-to-Monitor comparison
approaches will be inadequate. The Subcommittee expressed particular concern that multimedia
pollutants are excluded from both the monitoring and modeling framework. In addition only inhalation
is considered as an exposure route. The Subcommittee recognizes that currently there may not be
sufficient resources to include multimedia pollutants in the first phase of the monitoring efforts.
Nevertheless, understanding the behavior and effects of multimedia HAPs will eventually require a
multimedia monitoring strategy and multimedia exposure models. This should be addressed in
subsequent phases of the program.
In addition to the concern expressed about the inadequacy of the approach for multimedia
pollutants, the Subcommittee noted that the Model-to-Monitor approach could have inadequacies for
the following HAPs categories for the reasons noted:
a) Those for which problems exist with current sampling and analytical methodology (e.g.,
acrolein, acrylonitrile).
15
-------
b) Those for which adequate detection levels are attainable but for which current analytical
methods are limited to those HAPS that have ambient concentration significantly above
the exposure-level of concern.?
c) HAPS having uncertain, or poorly established, emission inventories.
d) Because multi-media pollutants can be locally cycled, they will be observed by monitors
but their real concentrations will not be captured in local emissions inventories.
3.1.3.3 Question 3(c): As noted in the paper, annual-average concentrations and
comparisons to modeled estimates can be uncertain when a large percentage of the
measurements are below the method detection limit (MDL). To estimate annual-
average concentrations from monitoring data, EPA generally substitutes half the
MDL. Does the Subcommittee suggest any alternative statistical approaches?
The Subcommittee believes that substitution by one-half the MDL, while a fairly robust
approach, may not be appropriate for all air toxics and all situations. The Subcommittee recommends
that the Agency:
a) Develop a set of criteria for what would constitute an acceptable fraction of below
detection concentrations as part of the data quality objectives, sampling and analytical
methods, and spatial and temporal sampling considerations.
b) If possible, incorporate the need for obtaining larger proportion of above- detection
values as a criterion for the sampling/analysis approach.
c) Select the MDL substitution method that best fits the amount and distribution of the
data, the fraction of values that are below the MDL, and the specific objective for
which the data will be used.
The Subcommittee recommended that laboratories report all data with the associated
uncertainties rather than an MDL, because the MDL is a variable in and of itself. This way, end users
of the data can decide for themselves the appropriate uncertainties for utilizing the data. For data sets
that already contain MDL values, the Subcommittee makes no specific recommendation, but offers
several approaches for statistical treatment of MDLs.
16
-------
3.2 Detailed Responses to Question 1: Does the air toxics monitoring concept paper
describe appropriate air toxic monitoring objectives and principles, particularly ones
that will permit the collection of monitoring data to support the initial National Air
Toxics Assessment activities?
The objectives of the National Air Toxics Assessment (NAT A) program are to "help EPA
identify areas of concern, characterize human health and ecosystem risks and track progress." The
activities of the NATA program are reiterated here for the purpose of discussion. The specific aims are:
a) measurement of air toxics emission rates from individual pollution sources;
b) compilation of comprehensive air toxics emission inventories for local, state, and
national domains;
c) measurement of ambient concentrations of air toxics at monitoring sites throughout the
nation;
d) analysis of patterns and trends in ambient air toxics measurements;
e) estimation of ambient and multimedia air toxics concentrations from emission inventories
using dispersion and deposition modeling;
f) the estimation of human and ecological resource exposures to air toxics; and
g) the assessment of human and ecological risks due to air toxics.
Specific aims c-g should be considered as tactical approaches for achieving aims a and b, the
overall primary objective of NATA.
Among the activities that derive from these specific aims, the concept paper indicates that
ambient monitoring data are needed to assess the air toxics inventory, air toxics modeling and trends in
HAP concentrations. The Subcommittee recommends that these uses of the monitoring network be
prioritized to provide a sound rationale for the study design. Establishing a well-designed ambient
network(s) for estimating ambient concentrations could also assist in determining exposures across the
nation when combined with appropriate models. However, from what was presented at the meeting, it
is not clear how the EPA will know:
a) Whether the ambient network is well designed?
b) What criteria would be used to assure that the network is adequate?
17
-------
c) How will the benefits of this program be assessed in terms of contributions to Agency
programs beyond those established under the Clean Air Act?
The Agency is now launching a serious effort to measure hazardous air
pollutants in ambient air. The Agency and the SAB anticipate that the resulting
information on concentrations will improve the scientific basis for
understanding exposure to these chemicals and the resulting risk. However,
in addition to air concentration data, estimates of exposure and risk also depend upon assumptions and
models. The SAB reminds the Agency that future advances in modeling and risk assessment methods
may permit further improvements in its analyses by incorporating more realistic and credible unit risk
factors as well as the new air concentration data. These questions could be better answered if priorities
for the network are clearly defined.
The draft Air Toxics Monitoring Concept Paper identified a number of important scientific
issues. However, a well thought-out, step-wise process that addresses these issues and focuses on
achieving the primary monitoring objectives should be developed. The EPA could begin this process
by defining as a goal the acceptable levels of uncertainties and then the acceptable data quality that are
needed for achieving the goal. These uncertainties should be communicated in all data analysis activities
or assessments based on these data.
If monitoring data are to be used to estimate exposures, measurement error is of significant
concern because it can be a potential source of bias. Exposure measurement error can occur when
using ambient air measurements to estimate human and ecological exposures. Because of these errors,
data provided by centrally located monitors rather than exposures measured on individuals could affect
the relative risk estimates in ways that are difficult to predict. The inevitability of such errors causes
uncertainty about the true magnitude of the estimated effects of individual air toxics on health. For a
more complete discussion of exposure measurement and its relationship to sources, see APPENDIX A:
Exposure Measurement Issues.
The proposed air toxics monitoring program is based on the assumption that state and local
agencies will make the most of the measurements, and that EPA will provide some funding to the States
for this. Although state and local agencies have the expertise to carry out field sampling programs, with
adequate guidance from EPA, other tasks, such as personal monitoring, methods development, and
other research-oriented objectives, are not typically within the ability and authority of the State and
local agencies.
18
-------
3.2.1 Question l(a): Does the Subcommittee believe that OAQPS has identified the
most important uses for ambient air toxics data? Are there other types of air toxics
monitoring data or data uses that should also be identified for near-term or for future
air toxics monitoring activities?
To address this question the Subcommittee considered current, near-term and future uses of
air-toxics data. These uses were also evaluated as being primary, that is directly related to the specific
aims of NAT A, and secondary, of use to other EPA and state programs.
3.2.1.1 Current Primary Uses for Air Toxics Data
EPA has identified many important uses of the data, including: model parameterization, model
evaluation, trends analysis for GPRA, background measurements, source characterization, national air
toxics assessments, and residual risk program assessments.
Because the National Ambient Monitoring Network will be the primary source of air toxics
data for many uses, the data should be collected in a manner appropriate for multiple uses. The
Agency's Quality System and the SAB's reports on secondary uses of data may be helpful in
determining how this may best be done.
Some additional uses of the data will naturally evolve from the 1-2 year pilot program. From a
regulatory perspective, data gathered during this pilot phase will:
a) characterize ambient concentrations and deposition in representative monitoring areas,
b) provide a "reality check" to dispersion and deposition models, and
c) decide on the appropriate quantity and quality of measurements in a national monitoring
network.
The Subcommittee agrees that multiple sites operating over at least a one year period in several
different regions of the country will be needed to adequately characterize a given monitoring area and
provide a minimal "reality check" on current models.
The Subcommittee agrees that EPA should focus on the "Urban HAP List" developed as part
of the Integrated Urban Air Toxics Strategy. These chemicals are of relatively high priority both for
health risk and because of frequency of detection. EPA may also wish to consider groups of substances
that serve as "fingerprints" for specific source emissions because this will make the initial data sets
amenable to alternative source-receptor modeling as an additional "reality check" on the dispersion and
deposition models.
19
-------
Although the Subcommittee members expressed a range of views on the selection of HAPs to
be measured during the pilot phase, there was consensus that the target list should reflect the following
issues:
a) practicality of measurement,
b) relative toxicity at ambient concentrations of the compound, and
c) inclusion of more compounds as the monitoring program matures.
Some thought the initial measurement phase should include as many compounds as is practical, given
method limitations and probabilities of detection, and that this list should include at least the 18 core
compounds identified in the concept paper, with possible addition of other multi-media pollutants that
include an air pathway. Others advocated that the initial measurement phase focus on fewer
compounds, selecting those whose behavior is most consistent with the assumptions of the dispersion
and deposition models to be tested.
3.2.1.2 Potential Secondary Uses of Air Toxics Data
Given the scarcity of air toxics monitoring data, EPA should prepare for the use of the publicly
available data from this project by others for additional analyses. The Subcommittee notes that some of
these uses could contradict Agency policy for the use of monitoring data as specified in 40 CFR Part
51 of April 21, 2000 section 10.2.2. Furthermore, there is the potential for inappropriate uses of
presently available dispersion models, most of which cannot confidently distinguish contributions from
individual sources to ambient monitors. By providing sustained attention to the Agency's quality system,
available modeling techniques and documentation, EPA can provide a scientific basis for determining
whether or not the data can be used appropriately to support secondary uses. The Subcommittee
expects EPA will restrict itself to appropriate uses and hopes that others will do the same. However,
some individuals and organizations may be less sensitive to the nuances of appropriate secondary uses
of data. Others may use data inappropriately to reach conclusions that are not scientifically defensible.
If so, then EPA will find that the reasoning and documentation required by the Agency's quality system
provides a credible basis for disagreement with the faulty analyses. The following are areas where the
Subcommittee anticipates secondary uses of the data from this study:
a) Identifying non-permitted emissions, not found on current inventories, such as fugitive
air emissions, air emissions from commercial underground injection sites, etc.;
b) Determining the potential impact of siting new facilities using chemicals monitored by the
project;
20
-------
c) Determining if reported large emission reductions have actually occurred, or if reported
reductions were due to changes in calculation method;
d) Supporting and assessing the impacts of other EPA programs, such as OSWER's
voluntary reduction program for persistent bioaccumulative toxics (PBTs);
e) Supporting the fish advisory program's efforts to determine whether there are
advisories for certain chemicals in areas where air concentrations indicate that there
may be problems; and
f) Informing analyses by community groups, the private sector and local governments for
decisions about environmental monitoring and management activities.
The Subcommittee reminds the Agency of an earlier SAB report related to the secondary use
of data collected by the Agency, Secondary Data Use Subcommittee Review ofCEIS's Draft
"Data Suitability Review of Major EPA Databases " (EPA-S AB-EC-99-010). In that report, the
SAB found that the 1998 draft CEIS document was appropriate for evaluating the general suitability of
databases for a range of secondary uses. The Center for Environmental Information and Statistics
(CEIS) as been merged into EPA's new Office of Environmental Information. Both the SAB report
and the Agency's Office of Environmental Information may be of use in planning for and influencing the
secondary uses of data from this study.
3.2.1.3 Other near-term and future uses for data
In planning the monitoring program, EPA should make use of existing data on air toxics, but be
aware that data quality criteria be assigned based on measurement quality and the time frame. In the
initial pilot phase of the monitoring activities and as an aid in the selection of sites nationwide, data from
EPA-funded and non-EPA funded research studies could also be used even if it was not collected
according to EPA-approved monitoring methods.
If resources are available, monitoring of multiple media for chemical contaminants (e.g, PCBs,
mercury, dioxin) can strengthen the exposure models that are the basis for risk assessments required by
the Clean Air Act (e.g., residual risk assessments). These media should include: soil, surface water,
sediment, fish and plant foliage.
Data from this study can contribute to the assessment of models other than ASPEN, including
TKDVI.FaTE and hazardous waste combustion exposure models. These data could also be of use to
those conducting personal exposure studies. EPA collaboration with organizations planning large (NIH
or NCHS type) health surveys could be of benefit to both organizations.
21
-------
3.2.2 Question l(b): Does the Subcommittee believe that neighborhood-scale
monitoring is appropriate for evaluating ASPEN air quality predictions and later for
developing long-term ambient air quality trends? Are there other appropriate
monitoring scales, perhaps for other data uses, that the Subcommittee would suggest?
The Subcommittee agrees that neighborhood scale monitoring is the correct choice for
measurements in urban areas during the pilot stage of this program. Neighborhood-scale monitoring is
meant to be representative of a 0.5 to 4 km horizontal scale. This is appropriate for comparison with
ASPEN results for urban census tracts because the tracts average 2.3 km2 in area, according to the
Cumulative Exposure Project study. Neighborhood-scale monitoring is also a good starting point for
tracking long-term ambient air quality trends in urban areas. However, over the longer term, the
monitoring scale used for toxic air pollutants should be guided by both the objectives of the monitoring
program and the characteristics of the pollutants being monitored, e.g. their reaction or removal rates
and the distribution of sources that determine the spatial variability in their concentrations. Different
scales may be used depending on the compound class and purpose.
In many cases, temporal variability in concentration will be more important that spatial
variability (for example, a background site or a city with few sources), and location could be specified
at the urban rather than neighborhood scale. However, for some comparisons, the neighborhood scale
could be too large because of the impact of significant sources in very-close-in areas. In these cases,
source-receptor characteristics will dominate siting decisions. In these situations, EPA may wish to take
a multi-tier approach to site selection, based on careful evaluation of potential site conditions and the
multiple uses of the data.
Conducting limited monitoring to assess spatial variations within a census tract is not essential
during the pilot stage of the program. Experience from the recent MATES-II study in southern
California has shown that the use of monitors to detect areas with localized higher concentrations of
toxic air pollutants is more difficult than usually perceived. Distances in the order of 200 meters or so
can have concentration gradients of up to two orders of magnitude. Thus an array of monitors may be
needed to depict influences from point sources. Therefore, the Subcommittee recommends avoiding
siphoning off resources for monitors and placing the initial focus of the national monitoring program
clearly on the neighborhood scale.
3.2.2.1 The Problem With Rural Census Tracts
In contrast to urban census tracts, the average area of the rural census tracts considered in the
Cumulative Exposure Project was 242 km2. In the future, background monitors representative of
larger areas may be needed for comparison with model results for rural census tracts. Because
OAQPS plans to include "rural" monitors in the pilot stage, the revised concept paper should discuss
guidelines for siting them (e.g., at NSF Long-term Ecological Research Sites).
22
-------
Neither the monitoring scales considered in the concept paper nor the census tract divisions
used in ASPEN modeling are appropriate for assessing ecological exposures. Therefore, in the future,
the program may wish to consider establishing some monitors for ecological exposures, at which time
siting criteria that are appropriate for this purpose should be developed.
3.2.2.2 Co-Location With Other Monitors
EPA should be cautious about recommending the co-location of air-toxics monitors with
existing PAMs sites because these locations may not be optimal for assessing air toxics exposures.
3.2.3 Question l(c): Does the Subcommittee believe that a basic 24-hour sample
taken at a frequency sufficient to fulfill the objectives of the program is
adequate to provide the model reality check and supply data for the
characterization of ambient hazardous air pollutant (HAPs) concentrations? In
particular, what are the Subcommittee's thoughts on the use of 24-hour samples
collected once in 12 days for model evaluation and at a more frequent (say, 1-
in-6 or l-in-3 day) schedule in the future for trends assessment at permanently
located monitoring sites?
The Subcommittee believes that the 24-hour sampling frame, while reasonable in principle, may
not always be consistent with the objectives of NAT A for all compounds and foreseen uses of the data.
In some cases a longer sampling period may be more useful. For example, longer duration
samples may help elucidate the relationship of long-term exposures to chronic health effects. Therefore,
sampling for 24-hour periods may not be appropriate if the purpose is to compare the mean of the
measurements with a model estimate of average concentration for a longer period, such as one year.
Assuming that there are no sampling duration-related artifacts, and that the monitoring approach
permits, a longer sampling time (e.g., 1 week) is beneficial. The longer sampling time would result in a
smaller fraction of the data being below the MDL, lower cost, and the quasi-continuous samples will
provide even better comparison data.
Sometimes a shorter sampling period will be more appropriate. For example shorter duration
samples may help elucidate the relationship of brief exposures to acute health effects. If the
measurements will be compared with a model estimate of the maximum 24-hour concentration over the
period of a year then shorter-term sampling is a better approach. Sampling for 24-hour periods may
also be appropriate if future trend analyses address concentrations at the upper end of the distribution.
The highest concentrations, which characterize the upper tail, occur infrequently. As a result many
measurements are needed to characterize the tail.
Because chemistry affects how compounds behave in the environment, what effects they cause,
and how they can best be measured, sampling duration could be considered compound-specific, or
23
-------
perhaps compound class-specific. For example, for sorbent-based methods, but not canister sampling,
sampling over longer time periods will reduce the detection levels because a larger quantity of sample is
collected. To address this issue, the Agency must consider the specific uses of the data as foreseen
now, brainstorm to determine potential future uses, and then adopt sampling times that satisfy the
requirements of all, within the available resources.
EPA's Guidance for the Data Quality Objectives Process (QA/G-4) should be consulted as
a guide to planning for projects where the objective of the study is to collect environmental data in
support of an Agency program. This is important where the results of the study will be used to make a
specific decision. Data Quality Objectives (see Appendix B) developed with the above considerations
can help guide decisions regarding:
a) the appropriateness of taking 24-hr samples,
b) the appropriate frequency to use for sample collection (every 3, 6, 12, etc. days),
c) the number of monitoring stations needed to produce sufficient data to reduce
uncertainties to an acceptable level, and
d) the spatial distribution of monitoring stations needed with respect to the population and
environment. In the absence of a rigorous approach, it is likely that the modeled results
will have large unacceptable and ill-defined uncertainties.
3.2.3.1 Time Resolution of Samples
The precision/bias desired for the models and for risk estimates should also be a factor in
setting the frequency of sample collection. Too long a cycle may miss episodic emissions and very
short sampling cycles will require more resources.
Consider the case of a power plant as an example. During the summer, electricity demand can
vary as much as 40 percent from day to day, especially during heat waves. To meet that power
demand, electric power companies increase generation at their plants, many of which burn coal or fuel
oil, and these plants emit VOCs in addition to nitrogen and sulfur oxides, mercury, and other metals.
VOC emissions are a function of the amount of fuel consumed, which increases during increased
demand. Unfortunately, although nitrogen oxide emissions also increase, these emissions are more a
factor of the burn temperature, air flow for combustion, and duration of burn, and so are not a good
proxy for VOC emissions (most utility-owned power plants continuously monitor their nitrogen oxide
emissions). Therefore, with short-term heat waves, the one-in-twelve monitoring proposed will likely
not be frequent enough to capture the short-term large increases in VOC emissions. According to
EPA's AIRS database, electric power plants burning coal and oil are among the top producers of
24
-------
VOCs nationally, and will likely constitute the major source of VOCs in some neighborhoods or rural
areas. (National Environmental Trust, May 1997)
3.2.3.2 Seasonal Variation and Annual Average
Currently available data from sixth day sampling cycles could be used to determine the effect on
precision of less frequent sampling. This could be done by examining the impact on the autocorrelations
for each HAP. For HAPs that vary by season a sampling frequency greater than 7 samples per season
(1 in 12 days) may be needed in order to precisely describe variability and to obtain an accurate annual
average.
Diurnal variation and resulting health effects are also issues for certain HAPS. Where these
issues are important, EPA could consider selecting a number of sites (e.g. 1-3 sites) for 12-hr day/night
samples. Although the Subcommittee does not favor co-location of monitors (see response to question
l(b) above), the Subcommittee suggests that PAMS data should be analyzed for diurnal information
regarding the toxics compounds measured in that program. In some situations, sampling done on a 1 in
12 basis may preserve resources and allow a larger number of sites to be monitored. More frequent
sampling at a selected number of sites (e.g., 1-3) should have 1 in 6 sampling; and perhaps one site
should have 1 in 3 sampling to help elucidate the potential importance of variations. Future long-term
monitoring should be based on results of the pilot study.
3.3 Detailed Responses to Question 2: Does the air toxics monitoring concept paper
present a reasonable phased strategy to design a national air toxics network?
3.3.1 Question 2(a): Does the Subcommittee believe that data analyses of existing
air toxics monitoring data, coupled with focused pilot studies for a core set of HAPs in
a limited number of areas are appropriate first steps in the design of the national
network. What additional or alternative approaches are suggested?
In general, the "data analyses of existing air toxics monitoring data, coupled with focused pilot
studies for a core set of HAPS in a limited number of areas are appropriate first steps in the design of
the national network." However, these analyses should not be limited to the generation of statistical
summaries. They must also focus on how the dimensions of existing measurements in space and time
relate to the dimensions of the decision units of interest.
Any statistical analysis must not only supply a data summarization, but also supply a description
of how these summary statistics are related to the entity about which inferences are desired. This entity
is the "Decision Unit." It is critical to clearly define the spatial and temporal boundaries of this entity
before sampling and/or ascribing any meaning to pre-existing sampling data. This is step 4 of the DQO
process. (See Appendix B for a description of the DQO process.)
25
-------
Conversely, when employing existing monitors, a clear description of the spatial and temporal
boundaries of the entity (air volume) actually monitored is essential. If these boundaries do not coincide
with those of the Decision Unit, then the data generated by the monitor are of little value in making
inferences regarding the Decision Unit.
3.3.1.1 Data Gaps and Pilot Studies
Analyses of the existing air toxics data will be quite helpful in identifying data gaps and planning
for focused pilot studies. EPA and the Subcommittee have already identified a few data gaps that may
merit pilot studies. For example, a pilot study that would address vertical variability may be helpful to
determine if monitors are adequately representing exposure critical for ecological studies as well as
human health risk assessment.
3.3.1.2 Mobile and Stationary Sources
The air toxics monitoring concept paper presents a reasonably phased strategy for designing a
national air toxics network and the proposed pilot study is a good step toward designing this. In
choosing sampling sites for this network, EPA should consider selecting sites that will be representative
of emissions from mobile sources, sites that will be representative of emissions from stationary sources,
and sites where there are contaminants from both mobile and stationary sources.
3.3.1.3 Personal Monitors
Personal monitors seldom have the sensitivity that is currently achieved by ambient air sampling.
In addition, personal monitors are not likely to be used on a large scale. However, characterizing
exposure requires both indoor and outdoor concentration levels. Therefore, EPA should consider
developing outdoor/indoor penetration factors for selected air toxics in parallel research programs,
perhaps those associated with PM2.5 studies.
3.3.1.4 Conveying Uncertainty
It is very important to convey the level of uncertainty. This includes uncertainty in the models
and the risk assessment, as well as sampling/analysis. In this way, reasonable expectations of precision
can be communicated to the public.
3.3.2 Question 2(b): Given that the State and local agencies have been measuring
ambient air toxics with the Toxic Organic (TO) and Inorganic (IO) methods, as
described in the paper, are these methods appropriate for the continued routine
26
-------
monitoring of the target Urban Air Toxics Strategy compounds in a national monitoring
network? If not, are there alternative methods which the Subcommittee would
recommend?
The technical criteria must be compatible with the data quality objectives of the study and
EPA's Guidance for the Data Quality Objectives Process (QA/G-4) is a useful approach to
working these issues out.
The methods proposed for the network are the currently accepted "gold standard" procedures
tested and approved by the EPA. For example, if the purpose of the study is to understand exposures
near the current MDL, these methods are probably necessary. This does not mean that using these
methods is always the best choice; other uses of the data may have objectives more compatible with
other methods. To make an analogy to medical testing, screening tests with relatively high false
positives are often used in situations where large numbers of measurements need to be made; then the
more expensive confirmatory tests are performed where the screening tests suggested there might be a
problem.
The DQO process enables the Agency to think through its objectives and the costs of meeting
them in a systematic way. Studies that involve collection and analysis of large numbers of samples often
need to balance factors such as the number of samples that can be obtained at any given site, and the
number of sampling locations and test methods. Because the development of a sampling network
involves a series of decisions, the balance among these considerations may change from decision to
decision. Sometimes resources are inadequate to meet the original goals. If so, decisions must be
made about altering the goals, the approach or the resources.
3.3.2.1 Matching the Method to the Need
In this section, the Subcommittee is addressing the question, "which are the best methods at
each stage of the project." The idea is to suit the test methods to the budget, the intended use of the
data at each stage and how it may impact the next stage of the project.
Site Selection - Where the site selection is the issue, a screening sampling method may be the
most appropriate approach because it is more important that the sampling locations are identified than
that the individual measurements are exact. The most rigorous methods could be used to analyze
samples from a few selected sites to provide a measure of reliability for the screening tools.
Accuracy - Once the sites are selected, then the accuracy of the measurement may become
the driver for the sampling and analytical approach. If so, then the most rigorous method is the more
desirable approach.
27
-------
Once the sites to be included in the network are selected, then "gold standard" methods could
be applied routinely. In some cases, improvements could be made to the TO-sampling methods. For
example, silica-lined Summa canisters may provide for more stability of polar species and there is
evidence that they are suitable for target HAPS such as ethylene oxide and acrylonitrile.
While the current TO methods are generally appropriate, the utility of silica lined canisters
should be explored in order to extend the TO-14,15 approach to polar compounds (ethylene oxide,
acrolein, acrylonitrile) that are difficult to sample with current stainless steel canisters. Frequent
performance audit checks of laboratories doing TO- methods should be performed to insure that these
laboratories are meeting accuracy, precision and detection level criteria.
Methods and Costs Influence the Number of Sampling Sites - Based on these
considerations, the Subcommittee recommends that the Agency reconsider the type of possible
sampling methods. The Subcommittee notes two examples. One, if the Agency compares the cost of
integrated vs. real time monitoring methods and finds real time methods to be satisfactory as screening
tools and less expensive overall than integrated sampling, then they may be a more suitable choice for
the initial phase of selecting sampling locations. Two, passive sampling may be appropriate for some
HAPS over longer sampling periods.
3.3.2.2 Selecting the Appropriate Method
The Subcommittee suggests EPA use the following key criteria to assess the appropriateness of
methods for use in monitoring the target Urban Air Toxics. These criteria include:
a) the ease of sample collection by minimally trained individuals,
b) the ease of sample transport and storage,
c) the stability of the pollutants in the collected state,
d) the detection limits of the method (i.e., the combination of sample size collected and the
sensitivity of the instrumental measurement technique),
e) the method precision and bias,
f) general ruggedness of the method,
g) the existence of other extant data collected with the same methods (comparability
issues), and
h) cost of analysis per sample.
28
-------
Advantages of Current TO- Methods - The TO-4A, -13 A, -14A and -15 methods for
collecting pesticides/PCBs, BaP/PAHs, and VOCs (non-polar and oxygenates) are well established
procedures. Most state and local agencies have acclimated to using these procedures and are
experienced with these methods. A considerable body of data now exists with these methods and the
precision and bias are well understood. It is also likely that after the DQOs are defined as described
above, the precision and bias will be acceptable for the objectives of NATA. Sample transport and
storage and the stability of the pollutants in the collected state will also likely meet the criteria for this
program. The use of GC/MS methods for measuring organics is highly recommended because the
instrumental method yields data with the fewest false positive results, i.e., misidentification is very low.
However, the Subcommittee strongly recommends that modern GC/MS instruments be employed,
because their sensitivity is as much as 50-fold higher than those 5 or more years old. This will help
reduce the percentage of non-measurable values for each of the HAPs, a concern in the use of
modeling techniques for estimating exposure. GC/MS methods employing full scanning approach
should be used to permit the very valuable opportunity of retrospectively examining data for other or
new HAPs in the future. While a major goal of NATA is the ability to perform a trends assessment for
the target list of HAPs, full scan GC/MS acquisition of data will also permit identification of emerging
pollution problems.
Disadvantages of Current TO- Methods - The preparation of collection materials and the
collection procedures prescribed in these methods are labor intensive. The sample work-up for the
pesticides/PCBs and BaP/PAHs is also labor intensive. Taking into account the entire effort to
implement TO-4A, -9A and -13A methods, it is not surprising that the costs are high per sample.
Because the VOCs are measured directly from the collection device, i.e., no work-up is needed, the
cost is somewhat less. Nevertheless, the cost for analyzing thousands of samples collected throughout
the monitoring network will be very expensive.
Comments on Other Methods - TO-11A for the collection and measurement of
formaldehyde and other aldehydes is also a well established method. The method employs high
performance liquid chromatography with UV detection. The drawback of the method is that it is limited
to a target list of aldehydes. Because modern HPLC/MS technology provides sensitive measurements,
the Subcommittee recommends that HPLC/MS be used for measuring the dinitrophenylhydrazine
derivatives of aldehydes. This approach will permit the opportunity to identify emerging trends in new
pollutants that would be missed with the current method of detection.
IO-3 is used for collecting fine and total suspended particulate matter on filters and for the
analysis of metals. Similar to its counterpart TO methods, this method is a well defined and used
method. Alternatives to X-ray fluorescence measurement of metals do exist; however, many of them
require sample digestion (destruction).
General Comments - The Agency should also encourage the use of state-of-the-art
instrumentation with each of the analytical methods used in NATA. For example, modern mass
29
-------
spectrometers have improved sensitivity such that a factor of over 50 can be achieved compared to
instruments manufactured in the mid-1990s. This added sensitivity will permit the measurement of
HAPs at lower levels than in past years and yield fewer non-detects. Also, high throughput systems will
be available in a few years that will permit the simultaneous analysis of 3 to 6 samples, thus lowering the
analysis costs significantly. The Agency should encourage analytical laboratories to stay abreast with
these developments for use in NATA.
3.4 Detailed Responses to Question 3 on Model-to-Monitor Comparison
3.4.1 Question 3(a): Do the data analysis approaches provide enough information to
allow appropriate interpretations of model results to support the development of model
improvements in the future and to assist with the design of the national monitoring
network?
The data analysis approaches provide a good starting point to evaluate the performance of the
ASPEN model and to facilitate improvements to the monitoring network. The data analysis approaches
provide the opportunity to calibrate the ASPEN or other models with monitoring data. Once
calibrated, ASPEN could be applied with greater confidence in situations where the monitoring data are
limited. Similarly, calibration will allow EPA to use the ASPEN (or other models) to design monitoring
networks.
To define performance objectives for models used to predict air toxic levels in time and space,
EPA must first establish acceptable uncertainties. There are two broad categories of uncertainties:
those that are associated with the model's ability to describe the system of interest and those that are
associated with the input parameters to the model. In general, model uncertainties are larger and more
difficult to assess than input parameter uncertainties. Both should be evaluated to the extent possible in
order to gain confidence in a model's performance for use in this program. This will require
understanding which input parameters are the most influential in determining the outcome or, if not
known, performing sensitivity testing of the models. The potential sources of uncertainties in ASPEN
have been identified. However, the sensitivity of input parameters leading to the magnitude of
uncertainties is unclear. Once this knowledge is in-hand, effort should be devoted to obtaining high
quality data for the more significant input parameters consistent with achieving the desired level of
confidence. The results from modeling efforts should display confidence intervals.
3.4.1.1 Proposed Data Analysis Methods
Assessment Tools - The development of performance assessment methods is key to
evaluating the model relative to the monitoring data. In addition to straight-scatter plots and box
plots, EPA used three performance assessment tools in the report and at the meeting: probability plots,
Spearman's rank correlation test of measured and modeled concentrations, and a point-to-range test of
monitored concentrations with the model estimates for the county in which the monitor is located. The
30
-------
first two tools are for classification analysis, whereas the latter makes more explicit use of the exact
concentrations. The probability plot is based on the transformation of model measurement pairs into a
hit (1) or miss (0) score with all location/measurement pairs plotted on a logistic curve. The rank
correlation compares the ranks of the monitored averages with the ranks of the model estimates, to see
if the model and monitors rank the sites similarly. The point-to-range test compares the point
observation to the range of modeled concentrations for the selected county.
Priorities - All three methods have clear capabilities and limitations. The Subcommittee found
the rank correlation approach easier to carry out and to communicate to decision makers and thus
suggested it should be given highest priority as an evaluation tool. The Subcommittee suggested that the
rank scatter-plot would be a useful adjunct for assessing the rank correlation. Scatter plots of absolute
concentrations are also useful; these could be done with model estimates for neighboring census tracts
as well as for the exact monitor location.
Second priority should be given to the point-to-range test. The point-to-range test is useful
because it is an "on-the-ground" test, and the Subcommittee recommends its use with one caveat.
Because of the variation in census tract size, the observation of a miss in one range may not be the same
as another. This might cause confusion for those trying to interpret the results. It may be more useful to
compare the monitoring average to the range of model estimates for several neighboring census tracts
than for the range for the county that includes the monitor, given the variability in county size and
monitor location within a county. The monitor may be located near the boundary of a census tract
and/or county.
The use of probability plots offers the potential for more statistical sophistication, but had the
disadvantages that the method is difficult to explain and that it is difficult to interpret the results. This is a
significant disadvantage for information that must be communicated to a broad range of decision
makers.
3.4.1.2 Use of Stratification
Stratification to evaluate monitoring data is very useful and should be strongly encouraged. The
stratification of samples is a reasonable way to discern causes of model and monitoring mismatches,
with the caveat (which EPA recognizes) that stratification "slices a thin pie [of samples] even thinner."
The proposed stratification variables are reasonable, these are:
a) urban vs nonurban,
b) geographic/climatological region
c) pollutant level, and
-------
d) source-oriented monitors vs others.
However, the Subcommittee suggests that additional factors be considered in developing the
stratification, such as:
a) wind speed,
b) terrain,
c) season, and
d) source categories (such as point, mobile, area, etc).
The terrain would be difficult to represent as a single variable for a census tract; although a
digital elevation model may be helpful, it is also a great deal of work. Subcommittee members vary on
their views of which stratification variables suggested by EPA are the most important. Some favor
pollutant level (above the MDL) and geographic region, while others think geographic region as it
relates to meteorological factors could be informative, and stratifying by pollutant concentrations could
give significant hints about whether the model or the measurements are in error. Errors in the emissions
inventory may overwhelm any differences in model output due to the presence of terrain features or
other stratification variables, but this assumption would need to be tested.
The stratification provides the opportunity to "bin" (i.e., organize) observations in a way that
allows the monitoring and model results to be linked to similar attributes. For example, by evaluating
monitor/model concentration ratios among all sites with low (or high) median wind speeds, scientists
can look for any systematic problems in the model performance under the low-wind conditions.
3.4.1.3 Model Diagnostics and Model Reliability
Although the stratified model comparisons will give a flavor for how well the models work
relative to monitored concentrations, these comparisons will not provide detailed diagnostics on model
performance. Thus, the comparisons cannot be expected to provide enough diagnostic information to
determine how the ASPEN or other EPA models may be corrected or improved. The Subcommittee
suggested ways to address this issue.
First, EPA should identify the purpose for the model and the expected reliability of that
performance. The conceptual plan for the monitor-to-model evaluation does not yet include
quantitative metrics or criteria to identify the differences in measured and modeled average
concentrations that would be a cause for concern. Such a metric could be used to indicate the need for
a more detailed evaluation of the model structure. To start examining problems that might arise from
meteorological inputs or treatment of dispersion in the model, it will be necessary to investigate model-
32
-------
monitor comparisons at finer time scales than the proposed annual average comparisons. One way to
address this is to work with a subset of chemicals that can be linked to the same source, such as
benzene, toluene, and ethylbenzene from fuels and see how they agree in the monitoring, model and
source data. One could look at comparisons between monitored and modeled ratios of pairs of
contaminants at any time scale - finer time resolution isn't necessary. If ratios don't match, one might
suspect the emissions inventory, since meteorology should affect pairs of pollutants from similar sources
in similar ways and thus cancel out of the ratio.
At concentrations at or below the MDL, the ASPEN model may not be easily evaluated when
there is a large percentage of non-detects in the monitoring data. This is unfortunate, given that
detection levels for some chemicals, such as some VOCs, are several orders of magnitude higher than
the critical health values.
Measured concentrations and modeled estimates may be expected to be different if the
pollutant is a persistent and is distributed into other media. In this case monitors may reflect long-term
emissions from soil or vegetation that are not predicted by ASPEN. The nine HAPs proposed for the
Model-to-Monitor comparison should include a persistent, bioaccumulative organic pollutant such as a
3-4 ring PAH or PCBs unless it is demonstrated that the dispersive characteristics of these chemicals
are similar to at least one of the nine PAHs. Because transport in the environment depends upon
partitioning, chemicals will have similar dispersive characteristics when they partition in similar
proportions into organic and lipid components.
3.4.1.4 Alternate Methods for Comparing Models to Measurements
The Subcommittee offered suggestions for alternate methods for evaluating the congruency of
predictions and observations.
a) Multivariate regression can be applied similarly to both the monitoring data and the
model predictions. That is, premises can be tested for dependence of observed or
predicted concentrations on variations in source, climate conditions, terrain factors,
source category, etc.
b) Gilbert and Simpson (1990) have proposed a method for evaluating soil concentrations,
that has been adopted by the USEPA (1994) and the Nuclear Regulatory Commission
(1995). This approach involves a test of medians and a quantile test on extremes.
Both methods require a "statistical" distribution of observed annual average
concentrations and model predictions. These entities result from temporal and spatial
series with perhaps the superposition of measurement error. Therefore, it may be
difficult to define what we have when we say we have congruency within some
geographical area.
33
-------
c) The Canadian NTRI data are more detailed than NTI data with respect to contaminant
emissions, such as PERC, PAHs, and dioxin, from some sources, and these could
facilitate the identification of gaps in the NTI.
3.4.2 Question 3(b): Are there some HAPs for which these approaches appear
inadequate? If so, can the Subcommittee suggest alternative approaches for these?
Problems exist for HAPS in the following categories:
a) Those for which problems exist with current sampling and analytical methodology
(acrolein, acrylonitrile). These may be resolved by considering alternative techniques
such as substitution of silica-lined canisters for stainless steel canisters in TO-14, TO-
15.
b) Those for which detection levels are attainable but current analytical methods produce
measurements that are significantly above the 10"6 risk based concentration. This is
particularly true for the class of volatile organic compounds. Because there is not any
likelihood that the method detection limits will be significantly reduced in the next few
years, the issue of dealing with "non-detects" becomes critical here. Some form of
sensitivity analysis should be carried out on the effects of "non-detects" on the annual
average concentrations derived from monitoring data.
c) HAPS having uncertain, or poorly established, emission inventories. Values should be
assigned to designate the quality of the emission inventories for specific HAPS and their
contribution to the uncertainty of modeled concentrations.
Problems may exist for compounds that can be adsorbed on soil, biota, or water surface and
then locally cycled and re-emitted in such a manner that they are not included in local emissions
inventories. This categories would include semi-volatiles, organic species such as polychlorinated
byphenyls and polycyclic organic matter.
The Subcommittee remains concerned that multimedia pollutants are excluded from both the
monitoring and modeling framework. In addition, only inhalation is considered as an exposure route.
This excludes several classes of HAPs - semi-volatile compounds that are transferred through food
chains. The Subcommittee recognizes that there are not sufficient resources to include multimedia
pollutants in the first phase of the monitoring efforts. However, the Subcommittee is concerned that,
because they are left out of the first phase, they will not be considered in the future and absence of
information could be interpreted as the absence of a problem. Providing adequate attention to multi-
media HAPs will require a multimedia monitoring strategy and a multimedia exposure model.
34
-------
3.4.3 Question 3(c): As noted in the paper, annual-average concentrations and
comparisons to modeled estimates can be uncertain when a large percentage of the
measurements are below the method detection limit (MDL). To estimate annual-
average concentrations from monitoring data, EPA generally substitutes half the
MDL. Does the Subcommittee suggest any alternative statistical approaches?
The Subcommittee observed that the substitution by half the MDL, which is a fairly robust
approach and one commonly used, may not be appropriate for all air toxics and all situations. In
reviewing this issue the Subcommittee noted that dealing with data below the MDL is an important issue
and must be confronted. However, the subcommittee offers cautions about interpreting the MDL and
how this impacts methods for dealing with monitoring data below the MDL.
3.4.3.1 The Importance of Dealing With Measurements Below the MDL
At and below the MDL, concentrations cannot be measured reliably and this makes certain
analyses difficult. The proportion of non-detects and how they are treated in the analysis can change
the estimate of the averages. Where toxic potency factors are high, even small changes in the estimate
of the averages can result in substantial variability (and additional uncertainty) in the calculated risk.
Where the proportion of non-detects is large, the ability to compare modeled versus measured
concentrations is impaired because the measures of central tendency and the variability depend upon
the estimates of distribution parameters. As a result, these estimates become more and more uncertain
as the fraction of the concentration data that is below the MDL increases.
3.4.3.2 Cautions
The appropriate interpretation and use of the MDL is not a simple matter.
a) Because there are differences in the way MDLs are determined and reported, the
method of determining the MDL for a particular measurement must be known and
understood.
b) The MDL is a variable itself and methods for determining MDLs can vary from
laboratory to laboratory.
For example, an MDL value could have been calculated as 3X the instrumental limit of
detection (IDL) or been based on blank variability for a sorbent-based method. It
could also mean that a particular peak or ion in a chromatogram was not seen at all, or
that a peak was observed but the concentration was lower than the mean blank, or that
the value was below a certain percentile of the distribution of possible blank values. In
each case, the value could be treated differently because their reliability is different, and
we may want to substitute them using different approaches.
35
-------
c) The MDL depends not only on the specific sampling and analysis methods but also on
the sample duration, and a number of other factors that may be related to the specific
operator of the site and laboratory.
d) The MDL can vary among laboratories even if they follow the same written procedures.
Furthermore, there are biases among laboratories.
e) The MDL can vary within a single laboratory because more than one analyst may be
involved or for other reasons.
f) The MDL is not a specific and constant value, even for a single laboratory consistently
using a standard sampling/analytical methodology. The MDL may vary from day to day
and from sample to sample.
Although it would be ideal if each reported measurement had an associated MDL value, it
would be extremely cumbersome and expensive to determine it.
The consequence of all these issues is, that before undertaking significant analyses, EPA should
understand what each reported MDL value actually means, how it was calculated, and that the
replacement approach should include only like-estimates of the MDL. Otherwise there will be
additional and inappropriately added variance to the data set, which will be reflected in the estimation of
means and variances.
3.4.3.3 Alternate Statistical Approaches for Data Below the MDL
There are a number of approaches available to obtain robust estimates of the parameters of the
distribution of concentrations when a significant fraction of the observations is below the MDL. The
Subcommittee suggested two types of approaches for addressing this issue: approaches that apply at
the front end, that is at the time the laboratory analyses are being conducted, and approaches that apply
at the back end, that is once the data base has been assembled.
Front-End Approaches - There are a number of approaches that may offer improved
estimates of the parameters of the distribution-one at the front end, at the time the laboratory analyses
are being conducted, and the other at the back end, once the data base has been assembled. Some
examples are provided below.
a) Sampling and/or analysis methods may be modified to increase the sensitivity of the
measurement, so that the percentage of the concentration data that falls below the MDL
is very small. Such modifications can be done at the point of sampling, chemical
analysis, and/or data reporting.
36
-------
b) Sample duration can be lengthened, so that more analyte or volatilization is collected.
There are both practical limitations to this approach and potential technical problems
due to sampling artifacts (e.g., decomposition of more reactive analytes during
prolonged sampling times).
c) The method of "standard additions" can be used at the point of analysis. In this method,
a known amount of analyte (standard) is added to a sample matrix. The idea is that the
standard's response, as measured by the instrument, will increase (adjusted for any
dilution and matrix effects) proportionally to the concentration of the analyte in the
sample. Typically this is done at three different concentration levels, usually consistent
with points on a calibration curve. If the slope of the standard additions curve parallels
that of the calibration curve, then the net difference between the curves at the y-
intercept is the concentration of the analyte of interest. In essence, the added amount
produces a measurable value above the MDL, and the known amount is subtracted,
giving the remaining value of the analyte. As long as the resulting value is greater than
the uncertainty in the calibration curve, a reasonable estimate for the original non-
detectable concentration can be made. The drawback to this method is that it is very
labor intensive, requiring at least two, and preferably three additional runs for each
sample with non-detectable concentrations. This approach is very difficult to apply to
canister sampling. Where there are a significant number of non-detectable
concentrations, this process can substantially increase laboratory analytical costs.
d) Another front-end approach involves data reporting. If both the confidence about the
MDL estimate and the value are reported, then this uncertainty can be incorporated in
any comparison with a corresponding model estimate of concentration.
e) Just because an analytical result is below the MDL does not mean that the laboratory
has not been able to measure a value, but rather that the measurement has less reliability
than others that are above the MDL. When background concentrations are not an issue
(i.e., if the method is sorbent-based and there is a background of the analyte, then the
sample MDL is determined differently) MDL's are determined by running a low-level
standard many times (e.g., 20 runs), and determining the variability in terms of its
standard deviation about the measured mean. Three times the standard deviation (3-
sigma) is added to the y-intercept of the calibration curve to determine the MDL. This
results in a 99% confidence level that the data points above the MDL are quantifiable.
At 2-sigma, the confidence is about 95%; at 1-sigma, about 65%. Thus, values
measured below the MDL could be reported along with its corresponding sigma value.
f) Measurements reported as below the 1-sigma level would be considered to be in the
noise, and not reportable. Those using the data could then determine what confidence
interval is appropriate, and select values based on the desired level. Several
37
-------
Subcommittee members stated that it is more useful to have laboratories report all data
with associated uncertainties than to have the laboratories censor the data. This
approach is much less labor intensive than the "standard additions" approach, in that
additional laboratory analyses are not necessary.
Back-End Approaches - Once a measurement result has been reported as below the MDL,
there is a question of to how to handle this result in subsequent mathematical and statistical calculations.
Gilbert and Kinnison (1981) reviewed some of the more popular methods of dealing with this problem.
Other methods have been proposed (see Schmoyer et al. 1996).
The choice of back-end method depends upon what assumptions are consistent with the form
of the underlying statistical distribution of the measurement data. For example, it is not appropriate to
assume that a set of measurement results at any sampling location should fit a statistical distribution
when these observations may be more likely to reflect a time- or spatial series. Thus, that which
appears as a useful statistical model from an observed data distribution, e.g. normal, lognormal, Weibull
models, may in reality result from a non-random process. This frequently compromises the statistical
justification behind a method for dealing with observations reported as below the MDL.
The convention of using half the MDL replacement is simple and based in an acceptable
rationale. If there is a set of measurements with unknown concentrations, but it is known these values
are between true 0 and an upper limit, and it is further assumed that the distribution of those values
between true 0 and the upper limit is normal and the MDL values in the data set are an unbiased sample
of that distribution, then the best estimator of the mean of the distribution is the mean of the MDL
estimates, which approximates half the MDL. This approach is fairly robust in many cases, but it
becomes less so as the proportion of values below the MDL increases and/or the distribution of the
data deviates strongly from normalcy. Then, the method chosen (i.e., the one that will provide the most
stable estimate of the mean) depends on what we know about the distribution of the data, the
percentage of values below the MDL, and the eventual uses of the data. It is also important to consider
if the additional computational effort is cost effective.
The key questions are:
a) How much does the selected substitution affect the measure of central tendency and
variance?
b) Is the effect important given a pre-selected criterion for a good fit between the mean of
the measurement and the modeled estimate of concentration?
As an example of the second question, the effect of the replacement method on the estimated
distribution parameters may not be important if the criterion for good match is one order of magnitude.
However it could be very important if the criterion for good fit is a factor of 2.
38
-------
When comparing sets of data pair wise for the purposes of determining if they are similar or not
making use of values below the MDL becomes even more complex, particularly when the total number
of pair values is low, and the proportion of data below the MDL is high. Examples of these types of
pair comparison are (1) the same compound measured by two different methods or (2) measured
concentrations compared with model estimates. Example (2) is closest to the comparison of the
parameters of the distribution of measurements to a single model estimate of average concentration
presumably without variability. In this case, the answer can be succinctly described as the replacement
method that provides the most stable estimate of the measure of central tendency of the distribution of
measurements.
There are a number of other approaches available to make use of measurements below the
MDL, including replacement by MDL divided by the square root of 2, Monte Carlo methods, random
selection below the MDL (if the distribution of the data is known), and maximum likelihood methods.
Discussion of these various methods and the situations where they might apply can be found in a
number of sources. Some user-friendly description with direct application to environmental data can be
found, for example, in Gilbert, (1987).
A particularly attractive alternative approach is to utilize known information to estimate the
unknown information. Given the percentage of values below the MDL, the mean and standard
deviation for the values above the MDL, the underlying distribution (e.g., lognormal), the mean and
standard deviation for the entire distribution could be estimated. Or, one could use this technique to
create a simple look-up table that estimates the below-MDL values based on the percentage of non-
detects, for example:
a) If 50% of data are below the MDL, the average of those values is 50% of the MDL;
b) if 25% of the data are below the MDL, the average of those values is 75% of the
MDL;
c) if 90% of the data are below the MDL, the average of those values is 10% of the
MDL; etc.
However, these methods can only be applied if the shape of the distribution is known. If, for
example, the mode of the distribution is higher than the MDL value, then other approaches may be
needed, probably a maximum likelihood approach (see, for example, the Cohen method in Gilbert,
(1987, pg 182-183).
Finally, for purposes of Model-to-Monitor comparisons, it may be more useful to set low-level
model predictions to the monitored value reported as below the MDL. This might well minimize the
impact of the choice of convention on dealing with measurements reported as below the MDL on
comparisons between model predictions and monitor data.
39
-------
40
-------
GLOSSARY
Air toxics
ATP
CAA
DQO
EPA
GACT
GCMS
GPRA
HAPs
HPLC/MS
IDL
IO
MACT
MDL
NATA
NSF
NTI
OAQPS
OSWER
PAHs
PBT
SAB
TMDL
TO
TRIM
TREVLFaTE
VOC
188 hazardous air pollutants regulated under the Clean Air Act
Air Toxics Program
Clean Air Act
Data Quality Objectives
Environmental Protection Agency
Generally Achievable Control Technology
Gas chromatography/mass spectrometry
Government Performance and Results Act
Hazardous air pollutants
High pressure liquid chromatography/mass spectrometry
Instrumental limit of detection
Inorganic method
Maximum Achievable Control Technology
Method detection limit
National Air Toxics Assessment
National Science Foundation
National Toxics Inventory
Office of Air Quality, Planning, and Standards
Office of Solid Waste and Emergency Response
Polynuclear aromatic hydrocarbons
Persistent Bioaccumulative Toxics
Science Advisory Board
Total Maximum Daily Load
Toxic Organic method
Total Risk Integrated Methodology
Multi-media fate within the TRIM package
Volatile Organic Compounds
G-l
-------
REFERENCES
Gilbert, R.O., and R. R. Kinnison, 1981, "Statistical Methods for Estimating the Mean and Variance
from Radionuclide Data Sets Containing Negative, Unreported or Less-Than Values," Health
Phvsics Vol. 40, pp. 377-390.
Gilbert, Richard O. (1987) Estimating the Mean and Variance from Censored Data Sets, In: Statistical
Methods for Environmental Pollution Monitoring, Van Nostrand Reinhold Company, New
York, NY., pp 177-185
Gilbert, R. O., and J. C. Simpson, 1990, "Statistical sampling and Analysis Issues and Needs for
Testing Attainment of Background-Based Cleanup Standards at Superfund Sites," In
Proceedings of The Workshop on Superfund Hazardous Waste: Statistical Issues in
Characterizing a Site: Protocols. Tools, and Research Needs. Arlington VA, February 21-22.
National Environmental Trust, "Up in Smoke: Industry Lobbyists' Big Lie about Barbecues and the
New Clean Air Act Standards", (Table 1), May 1997
Schmoyer, R. L., J. J. Beauchamp, C. C. Brandt and F. O. Hoffman, Jr, 1996, "Difficulties with the
lognormal model in mean estimation and testing." Environmental and Ecological Statistics. Vol.
3, No. 1, pp. 81-97.
USEPA, 1994, Methods for Evaluating the Attainment of Cleanup Standards. Volume 3: Reference-
Based Standards For Soils and Solid Media. Washington D.C., EPA 230-R-94-004.
U.S. Nuclear Regulatory Commission, 1995, A Nonparametric Statistical Methodology for the Design
and Analysis of Final Status Decommissioning Surveys. NUREG
R-l
-------
APPENDIX A: Exposure Measurement Issues
Exposure measurement error not only includes any errors resulting from the measurement
instrument, but also considers the error in assigning an individual's exposure based on instruments some
distance away (spatial) from each individual in the study population. Measurement errors may consist
of "classical" error, which causes bias in measures of association in most situations, and "Berkson"
error, which causes little or no bias. Berkson error occurs when the expectation of the measured value
is not the true value but near the average of the true values. There are three components of exposure
measurement error:
1. the individual exposure and average of personal exposures,
2. the average personal exposure and true ambient levels, and
3. the measured ambient level and true ambient level.
"True" exposures cannot easily be measured. The major Berkson error component is the
difference between an individual's actual exposure to a particular pollutant and the average individual
exposures of everyone in a geographical area of interest. The average individual exposures will not be
known in NATA, but instead the ambient levels will be measured by one or a few monitors in an area.
The difference between the monitor measurements and the average personal exposure is the remaining
error and is more of the "classical" error that is likely to introduce bias in the risk estimate.
This remaining error can be further decomposed into: the difference between the average
personal exposure and the true ambient level, and the difference between the true ambient level and the
measured ambient level. The difference between the true and measured ambient level probably would
not introduce bias if the average measurement from available monitors is an unbiased estimate of the
true, spatially averaged ambient level. This leaves the difference between average personal and ambient
levels as the most likely cause of bias. In NATA the average personal exposure will be modeled from
the measured ambient level data. An approach that can be used to correct for such biases is to use
regression calibration which uses data on both the error-prone daily neighborhood and fixed-site
ambient level measurements and personal exposure measurements for some persons on the same days.
Such data can be used to calibrate, that is, adjust, the ambient exposure measures by estimating from a
regression model the change in average personal exposures corresponding to a unit change in ambient
levels. Once the calibration factor is known, the estimated change in risk per unit change in ambient
levels can be corrected so that they apply to changes in personal exposures. Regression calibrations
can be obtained from TEAM, PTEAM, NHEXAS, EXPOLIS and THEES data where personal and
ambient levels were measured for a number of the HAPS.
A-l
-------
APPENDIX B: Summary of Elements of the EPA Quality System and an
Introduction to the Data Quality Objectives Process
The Agency's quality policy is consistent with ANSI/ASQC E-4 and is defined in EPA Order
5360.1 CHG 1 (1998), the Quality Manual and the organizational components designed for policy
implementation as described by the Agency's Quality System (EPA QA/G-0). The quality system
provides the framework for planning, implementing, and assessing work performed by the organization
for carrying out required quality assurance and quality control.
EPA has a comprehensive system of tools for managing its data collection and use activities to
assure data quality. The management tools used in the organizational level of the EPA Quality
System include Quality Management Plans and Management System Reviews. The technical tools
used in the project level of the EPA Quality System include the Data Quality Objectives Process,
Quality Assurance Project Plans,
Standard Operating Procedures, Technical Assessments, and Data Quality Assessment.
At the management level, the Quality System requires that organizations prepare Quality
Management Plan (QMP). The QMP provides an overview of responsibilities and lines of authority
with regards to quality issues within an organization. Therefore, not only does ETV have a QMP, but
the verification partners and subcontractors are required to develop and implement their own QMPs.
The ETV program calls these documents Quality and Management Plans.
Organizations with QMPs review their own performance and develop Quality Assurance
Annual Report and Work Plans (QAARWP) that provide information on the previous yearis
QA/QC activities and those planned for the current year. The QAARWP functions as an important
management tool at the organizational level as well as at the Agency-wide level when QAARWP
supplied information is compiled across organizations.
At longer multi-year intervals EPA conducts periodic Management System Reviews for
organizations. An MSR consists of a site visit; a draft report that details findings and recommended
corrective actions, consideration of the reviewed organization's formal response to the draft report and
the authoring of a final report.
At the project level, the data life cycle of planning, implementation and assessment becomes
important. The data life cycle begins with systematic planning. EPA recommends that this required
planning be conducted using the Data Quality Objectives (DOO) Process. The DQO process is a
strategic planning approach based on the scientific method that is used to prepare for a data collection
activity. It provides a systematic procedure for defining the criteria that a data collection design should
satisfy, including when to collect samples, where to collect samples, the tolerable level of decision
errors for the study, and how many samples to collect.
B-l
-------
EPA has prepared Guidance for the Data Quality Objectives Process (QA/G-4). This
guidance document applies to projects where the objective of the study is to collect environmental data
in support of an Agency program, and, the results of the study will be used to make a specific decision.
DQOs are qualitative and quantitative statements that clarify study objective(s), define the most
appropriate type of data to collect, determine the most appropriate conditions from which to collect the
data, and specify tolerable limits on the decision errors which will be used as the basis for establishing
the quantity and quality of data needed to support the decision. The QA/G-4 provides guidance on
using a systematic planning process to develop DQOs; it is based on a graded approach.
Briefly, the seven steps in the DQO process are:
1. State the problem
2. Identify the decision
3. Identify the inputs to the decision
4. Define the study boundaries
5. Develop a decision rule
6. Specify tolerable limits on decision errors
7. Optimize the design
The Quality Assurance Project Plan (OAPP) is the principal output of the DQO process
and is the project-specific blueprint for obtaining data appropriate for decision-making. The QAPP
translates the DQOs into performance specifications and QA/QC procedures for the data collectors.
QAPPs provide a second level of assurance that the test will be performed in a matter to generated
objective and useful information of known quality.
The final step in the data life cycle is the Data Quality Assessment (DOA) which determines
whether the acquired data meet the assumptions and objectives of the systematic planning process that
resulted in their collection. In other words, the DQA determines whether the data are usable because
they are of the quantity and quality required to support Agency decisions.
B-2
-------
APPENDIX C: ROSTERS
Air Toxics Monitoring Strategy Subcommittee
Executive Committee
-------
U.S. ENVIRONMENTAL PROTECTION AGENCY
SCIENCE ADVISORY BOARD
AD HOC SUBCOMMITTEE OF THE EXECUTIVE COMMITTEE:
AIR TOXICS MONITORING STRATEGY SUBCOMMITTEE
FYOO
CHAIRMAN*
Dr. Thomas E. McKone, Staff Scientist Lawrence Berkeley Laboratory and Adjunct Professor,
School of Public Health, University of California, Berkeley, CA
MEMBERS OF SAB COMMITTEES*
Dr. Maria Morandi, Assistant Professor, University of Texas Health Science Center at Houston,
School of Public Health, Houston, TX
CONSULTANTS TO THE SAB*
Dr. Rebecca Ann Efroymson, Research Staff Member, Environmental Sciences Division, Oak Ridge
National Laboratory, Oak Ridge, TN
Dr. Timothy Larson, Professor, Dept. of Civil and Environmental Engineering, University of
Washington, Seattle, WA
Dr. Karl R. Loos, Environmental Advisor, Equilon Enterprises LLC, Houston, TX
Dr. Jana Milford, Associate Professor, Department of Mechanical Engineering, University of
Colorado, Boulder, CO
Dr. Thomas Natan, Research Director, National Environmental Trust, Washington, DC
Dr. Edo Pellizari, Research Triangle Institute, Research Triangle Park, NC
Mr. Doug Splitstone, Principal, Splitstone & Assoc., Marysville, PA
Mr. Mel Zeldin, Director, Monitoring and Analysis Division, South Coast Air Quality Management
District, Diamond Bar, CA
SCIENCE ADVISORY BOARD STAFF
Kathleen White Conway, Designated Federal Officer, Science Advisory Board (1400A),
Environmental Protection Agency, 1200 Pennsylvania Avenue, NW, Washington, DC 20460
C-2
-------
Mary L. Winston, Management Assistant, Science Advisory Board (1400A), Environmental
Protection Agency, 1200 Pennsylvania Avenue, NW, Washington, DC
C-3
-------
* Members of this SAB Subcommittee consist of
a. SAB Members: Experts appointed by the Administrator to two-year terms to serve on one
of the 10 SAB Standing Committees.
b. SAB Consultants: Experts appointed by the SAB Staff Director to a one-year renewable
term to address a particular issue.
C-4
-------
U.S. ENVIRONMENTAL PROTECTION AGENCY
SCIENCE ADVISORY BOARD
EXECUTIVE COMMITTEE
FY2000
INTERIM CHAIR
Dr. Morton Lippmann, Professor, Nelson Institute of Environmental Medicine, New York University
School of Medicine, Tuxedo, NY
MEMBERS
Dr. Henry A. Anderson, Chief Medical Officer, Wisconsin Division of Public Health, Madison, WI
Dr. Richard J. Bull, MoBull Consulting, Kennewick, WA
Dr. Maureen L. Cropper, Principal Economist, DECRG, The World Bank,
Washington, DC
Dr. Kenneth W. Cummins, Senior Advisory Scientist, California Cooperative Fishery Research Unit
and Adjunct Professor, Fisheries Department, Humboldt State University, Arcata, CA
Dr. Linda Greer, Senior Scientist, Natural Resources Defense Council,
Washington, DC
Dr. Hilary I. Inyang, University Professor and Director, Center for Environmental Engineering,
Science and Technology (CEEST), University of Massachusetts Lowell, Lowell, MA
Dr. Janet A. Johnson, Senior Radiation Scientist, Shepherd Miller, Inc.,
Fort Collins, CO
Dr. Roger E. Kasperson, University Professor and Director, The George Perkins Marsh Institute,
Clark University, Worcester, MA
Dr. Joe L. Mauderly, Director & Senior Scientist, Lovelace Respiratory Research Institute,
Albuquerque, NM
Dr. M. Granger Morgan, Head, Department of Engineering & Public Policy, Carnegie Mellon
University, Pittsburgh, PA
Dr. William Randall Seeker, Senior Vice President, General Electric Energy and Environmental
Research Corp., Irvine, CA
C-5
-------
Dr. William H. Smith, Professor of Forest Biology, Yale University, New Haven, CT
Dr. Robert N. Stavins, Albert Pratt Professor of Business and Government, Faculty Chair,
Environment and Natural Resources Program, John F. Kennedy School of Government,
Harvard University, Cambridge, MA
Dr. Mark J. Utell, Professor of Medicine and Environmental Medicine, University of Rochester
Medical Center, Rochester, NY
Dr. Terry F. Young, Senior Consulting Scientist, Environmental Defense Fund, Oakland, CA
LIAISON FOR CHILDRENS HEALTH PROTECTION ADVISORY COMMITTEE
Mr. J. Thomas Carrato, Assistant General Counsel, Regulatory Affairs, Monsanto Company, St.
Louis, MO
LIAISON FOR SCIENCE ADVISORY PANEL
Dr. Ronald Kendall, Director & Professor, The Institute of Environmental & Human Health, Texas
Tech University/Texas Tech University Health Sciences Center, Lubbock, TX
LIAISON FOR ORD BOARD OF SCIENTIFIC COUNSELORS
Dr. Costel D. Denson, Professor of Chemical Engineering, University of Delaware, Newark, DE
SCIENCE ADVISORY BOARD STAFF
Dr. Donald G. Barnes, Staff Director/Designated Federal Officer, Environmental Protection Agency,
Science Advisory Board (1400A),1200 Pennsylvania Avenue, NW, Ariel Rios North Lobby,
Room 6450, Washington, DC 20460
Ms. Priscilla Y. Tillery-Gadson, Program Specialist, Environmental Protection Agency, Science
Advisory Board (1400A), 1200 Pennsylvania Avenue, NW, Ariel Rios North Lobby, Room
6450, Washington, DC 20460
Ms. Betty B. Fortune, Office Assistant, Environmental Protection Agency, Science Advisory Board
(1400A), 1200 Pennsylvania Avenue, NW, Ariel Rios North Lobby, Room 6450,
Washington, DC 20460
C-6
------- |