UNITED STATES ENVIRONMENTAL PROTECTION AGENCY
WASHINGTON D.C. 20460
OFFICE OF THE ADMINISTRATOR
SCIENCE ADVISORY BOARD
April 18, 2017
EPA-CASAC-17-002
The Honorable E. Scott Pruitt
Administrator
U.S. Environmental Protection Agency
1200 Pennsylvania Avenue, N.W.
Washington, D.C. 20460
Subj ect: Consultation on the EPA's Review of the Primary National Ambient Air Quality
Standard for Sulfur Oxides: Risk and Exposure Assessment Planning Document
(External Review Draft - February 2017)
Dear Administrator Pruitt:
EPA's Clean Air Scientific Advisory Committee (CASAC) Sulfur Oxides Panel held a public meeting
on March 21, 2017, to conduct a consultation with EPA staff on the EPA's Review of the Primary
National Ambient Air Quality Standard for Sulfur Oxides: Risk and Exposure Assessment Planning
Document (External Review Draft - February 2017). The Panel generally found the Draft Risk and
Exposure Assessment Planning Document to be a useful roadmap for the development of the Risk and
Exposure Assessment.
The Science Advisory Board Staff Office has developed the consultation as a mechanism to provide
individual expert comments for the EPA's consideration early in the implementation of a project or
action. A consultation is conducted under the normal requirements of the Federal Advisory Committee
Act (FACA), as amended (5 U.S.C., App.), which include advance notice of the public meeting in the
Federal Register.
No consensus report is provided to the EPA because no consensus advice is given. The individual
CASAC Sulfur Oxides Panel members' written comments are provided in Enclosure A.
We thank the EPA for the opportunity to provide advice early in the development of the Risk and
Exposure Assessment and look forward to peer reviewing the completed Risk and Exposure
Assessment.

O
%
$32
^ PRO^°
O

-------
Sincerely,
/s/
Dr. Ana V. Diez Roux, Chair
Clean Air Scientific Advisory Committee
Enclosure

-------
NOTICE
This report has been written as part of the activities of the EPA's Clean Air Scientific Advisory
Committee (CASAC), a federal advisory committee independently chartered to provide extramural
scientific information and advice to the Administrator and other officials of the EPA. The CASAC
provides balanced, expert assessment of scientific matters related to issues and problems facing the
agency. This report has not been reviewed for approval by the agency and, hence, the contents of this
report do not represent the views and policies of the EPA, nor of other agencies within the Executive
Branch of the federal government. In addition, any mention of trade names or commercial products does
not constitute a recommendation for use. The CASAC reports are posted on the EPA website at:
http://www.epa.gov/casac.
1

-------
U.S. Environmental Protection Agency
Clean Air Scientific Advisory Committee
Sulfur Oxides Panel
CHAIR
Dr. Ana V. Diez Roux, Dean, School of Public Health, Drexel University, Philadelphia, PA
CASAC MEMBERS
Dr. Judith Chow, Nazir and Mary Ansari Chair in Entrepreneurialism and Science and Research
Professor, Division of Atmospheric Sciences, Desert Research Institute, Reno, NV
Dr. Jack Harkema, Distinguished University Professor, Department of Pathobiology and Diagnostic
Investigation, College of Veterinary Medicine, Michigan State University, East Lansing, MI
Dr. Donna Kenski, Data Analysis Director, Lake Michigan Air Directors Consortium, Rosemont, IL
Dr. Elizabeth A. (Lianne) Sheppard, Professor of Biostatistics and Professor and Assistant Chair of
Environmental & Occupational Health Sciences, School of Public Health, University of Washington,
Seattle, WA
Dr. Ronald Wyzga, Technical Executive, Air Quality Health and Risk, Electric Power Research
Institute, Palo Alto, CA
CONSULTANTS
Mr. George A. Allen, Senior Scientist, Northeast States for Coordinated Air Use Management
(NESCAUM), Boston, MA
Dr. John R. Balmes, Professor, Department of Medicine, Division of Occupational and Environmental
Medicine, University of California, San Francisco, San Francisco, CA
Dr. James Boylan, Program Manager, Planning & Support Program, Air Protection Branch, Georgia
Department of Natural Resources, Atlanta, GA
Dr. Aaron Cohen, Consulting Scientist, Health Effects Institute, Boston, MA
Dr. Alison C. Cullen, Professor, Daniel J. Evans School of Public Policy and Governance, University
of Washington, Seattle, WA
Dr. Delbert Eatough, Professor of Chemistry, Department of Chemistry and Biochemistry, Brigham
Young University, Provo, UT
11

-------
Dr. H. Christopher Frey, Glenn E. Futrell Distinguished University Professor, Department of Civil,
Construction and Environmental Engineering, College of Engineering, North Carolina State University,
Raleigh, NC
Dr. William C. Griffith,* Associate Director, Department of Environmental and Occupational Health
Sciences, Institute for Risk Analysis & Risk Communication, School of Public Health, University of
Washington, Seattle, WA
Dr. Steven Hanna, President, Hanna Consultants, Kennebunkport, ME
Dr. Daniel Jacob,* Professor, Atmospheric Sciences, School of Engineering and Applied Sciences,
Harvard University, Cambridge, MA
Dr. Farla Kaufman, Epidemiologist, Office of Environmental Health Hazard Assessment,
Reproductive and Cancer Hazards Assessment Section, California EPA, Sacramento, CA
Dr. David Peden, Distinguished Professor of Pediatrics, Medicine & Microbiology/Immunology,
School of Medicine, University of North Carolina at Chapel Hill, Chapel Hill, NC, United States
Dr. Richard Schlesinger,* Associate Dean, Dyson College of Arts and Sciences, Pace University, New
York, NY
Dr. Frank Speizer, Edward Kass Distinguished Professor of Medicine, Channing Division of Network
Medicine, Brigham and Women's Hospital and Harvard Medical School, Boston, MA
Dr. James Ultman, Professor, Chemical Engineering, Bioengineering Program, Pennsylvania State
University, University Park, PA
SCIENCE ADVISORY BOARD STAFF
Mr. Aaron Yeow, Designated Federal Officer, U.S. Environmental Protection Agency, Washington,
DC
* Did not participate in review
111

-------
Enclosure A
Individual Comments by CASAC Sulfur Oxides Panel Members
on the EPA's Review of the Primary National Ambient Air Quality Standardfor Sulfur Oxides:
Risk and Exposure Assessment Planning Document (External Review Draft — February 2017)
Mr. George A. Allen	A-2
Dr. John R. Balmes	A-4
Dr. James Boylan	A-6
Dr. Judith Chow	A-9
Dr. Aaron Cohen	A-12
Dr. Alison C. Cullen	A-13
Dr. Delbert J. Eatough	A-15
Dr. H. Christopher Frey	A-20
Dr. Steven Hanna	A-24
Dr. Jack Harkema	A-28
Dr. Farla Kaufman	A-29
Dr. Donna Kenski	A-31
Dr. Elizabeth A. (Lianne) Sheppard	A-32
Dr. Frank Speizer	A-35
Dr. James S. Ultman	A-37
Dr. Ronald E. Wyzga	A-38
A-l

-------
Mr. George A. Allen
Analytical Approach and Study Area Selection
1.	The overall analytical approach for the Risk and Exposure Assessment (REA) and its appropriateness
for developing spatially and temporally varying 5-minute ambient S02 concentrations, simulating
population-based 5-minute peak exposures, and estimating study area health risk based on controlled
human exposure study data. [Chapter 4]
The overall analytical approach in Figure 4-1 (page 4-2) for this REA is sound. Using a simple linear
(proportional) adjustment to just meet existing [and alternative?] standards is appropriate for S02, since
concentrations of concern are relatively near the sources and on that spatial/temporal scale, S02 is
reasonably conserved, and expected adjustments are small. The choice to use modeled ambient S02
concentrations instead of observed (measured) concentrations provides more detailed local scale spatial
patterns. Modeled hourly S02 concentrations with AERMOD combined with 5-minute variability
information from observations should provide appropriate input for 5-minute exposure modeling
(APEX). Comments on the approach to risk assessment are not my area of expertise.
2.	The criteria identified and approach used to select potential study areas to evaluate for this REA.
[Section 4.1.2, Exposure Domain]
The process of identifying a "short list" of potential study areas is well described and reasonable, based
on monitor[s] in the area having a design value within 10 ppb of the current standard, 5-minute data
from at least one monitor in the study area, and a population of at least 100,000 within 10 km of relevant
monitors. These selection criteria result in the nine areas shown in Figure 4-1 (page 4-7). Some of these
sites have more available data (sites, DV years), resulting in four "very short list" candidates. Modeling
domains would be constrained to within 10 km of relevant emission sources to limit uncertainty in
modeled concentrations. Overall, this approach should result in optimal exposure domains for this REA.
Other Comments
During the meeting it was noted by EPA staff that the REA would use the 2014 National Emissions
Inventory (NEI) instead of 2011 NEI as used in this planning document. This is important given the
large reductions in S02 emissions from EGUs and other sources (such as ultra-low S diesel and heating
oil) over the last several years.
A useful analysis of reported 5-minute S02 concentrations in the context of design values and various
health-relevant 5-minute concentrations is presented in Appendix B of this planning document
(Occurrences of 5-Minute S02 Concentrations of Interest in the Recent Ambient Air Monitoring Data
(2013-2015)). The text of the document only mentions this appendix very briefly (one sentence on page
3-6); it may be useful to bring a summary of the information in Appendix B into chapter 3.
A-2

-------
Table 4-1 lists lead smelting as a source in Marion County IN (Indianapolis); is this correct? It's my
understanding that the last of the domestic lead smelters closed down several years ago. It would be
useful to include a column listing the total 2014 NEI TPY emissions for each of the study areas in this
table.
There is no mention in the planning document of performing exposure risk analysis using potential
alternative standards (concentrations, forms). During the meeting it was noted by EPA staff that this
could be done based on the results of the risk analysis at the current standard. It would be helpful if there
was a brief discussion of this in the planning document.
AERMOD, the EPA regulatory S02 model, will be used in the risk assessment to estimate exposures to
both 1-hour and 5-minute S02. However, AERMOD's performance is evaluated only at 1-hour, and its
performance in estimating distributions of 5-minute peak S02 concentrations is not well characterized.
If this is a potential issue in estimating exposures of concern for the REA, some discussion of this in the
planning document would be useful.
A-3

-------
Dr. John R. Balmes
Analytical Approach and Study Area Selection
1.	The overall analytical approach for the Risk and Exposure Assessment (REA) and its appropriateness
for developing spatially and temporally varying 5-minute ambient S02 concentrations, simulating
population-based 5-minute peak exposures, and estimating study area health risk based on controlled
human exposure study data. [Chapter 4J.
The plan to follow the same conceptual model as used for the 2009 REA seems appropriate.
2.	The criteria identified and approach used to select potential study areas to evaluate for this REA.
[Section 4.1.2]
The planned approach seems reasonable.
Exposure Analysis
1.	The overall approach to be used for the exposure analysis, including the use of the APEX model,
given objectives of the analyses, which include development of 5-minute exposures for input to the risk
assessment, assessment of factors that contribute to the upper percentile population-based 5-minute
exposures. [Section 4.1]
The planned approach to the exposure analysis seems reasonable.
2.	The selected study population groups of interest (adults with asthma, school-aged children with
asthma) for which SO2 exposure estimates are to be developed. [Sections 3.2.1, 4.1.3]
The target study population groups are appropriate based on the review of the literature contained in the
ISA.
Health Risk Assessment
1. The general structure and overall approach that staff plans to use for the risk assessment. [Section
4.2]
The overall approach for the health risk assessment is appropriate given the review of the literature
contained in the ISA.
2. The approaches for using findings from the controlled human exposure studies.
A-4

-------
a.	The health benchmarks identifiedfor this REA. [Sections 3.2.2, 4.2.3]
b.	Plans for developing updated exposure-response functions, including the methodology, and specific
studies to be relied on, for estimating exposure-response relationships for lung function decrements.
[Sections 3.2.2, 4.2.4]
i.	The focus on specific airway responsiveness (sRaw) for this quantitative risk assessment of
short-term exposure-related endpoints.
ii.	The range of exposure concentrations appropriate to include in the dataset for deriving the
exposure-response function.
Given that there are no new controlled human exposure study data, I think that it is reasonable to include
the Linn et al. (1983) and Horstman et al. (1986) data to improve the usefulness of the E-R curves for
lower level exposures. I also like the plan to explore the effects of different forms of the E-R curve, such
as using a curve with the 1000 ppb data removed. Finally, the plan to focus only on SRaw response data
is wise given the paucity of FEV1 data.
A-5

-------
Dr. James Boylan
Ambient Air Concentrations
1. The use of an AERMOD model-based approach to predict hourly concentrations at all receptor
locations within selected study areas. [Sections 3.3.2, 4.1.3.3]
The model-based approach to predict hourly concentrations at all receptor locations within the selected
study areas is appropriate and will better quantify the spatial variation in concentrations compared to
using observations alone. AERMOD is an appropriate model for predicting SO2 concentrations in
ambient air. The approach that is described in the REA includes running AERMOD to obtain 1-hour
SO2 concentrations at all receptors and all hours, then uses the 5-minute SO2 observations to convert the
1-hour AERMOD results into continuous 5-minute results. If AERMOD is performing well, this is a
valid approach.
For the past 2 years, I have been running AERMOD to model SO2 in the state of Georgia to meet the
requirements of EPA's SO2 Data Requirement Rule to inform our SO2 designation recommendations for
the 2010 standard. In my experience, AERMOD does not always perform well and can have significant
over and under predictions depending on site-specific characteristics. At one monitor location in
Georgia, AERMOD over predicted the SO2 concentrations by a factor of 10 (the monitor was half the
standard and the model was 5x the standard).
Page 4-18 of the REA states that "Model performance (e.g., comparison with available monitor data)
can be evaluated using procedures outlined in the EPA Protocol for Determining Best Performing Model
(U.S. EPA, 1992)." A summary of the specific model performance approach that will be implemented
and the model performance "acceptance" criteria needs to be included. The ISA states, "For models
intended for application to compliance assessments (e.g., related to the 1-h daily max SO2 standard), the
model's ability to capture the high end of the concentration distribution is important. Measures such as
robust highest concentration (RHC) (Cox and Tikvart, 1990), and exploratory examinations of quantile-
quantile plots (Chambers et al., 1983) are useful. The RHC represents a smoothed estimate of the top
values in the distribution of hourly concentrations. In contrast, for dispersion modeling in support of
health studies where the model must capture concentrations at specified locations and time periods,
additional measures of bias and scatter are important."
Most published AERMOD model performance evaluations are associated with using the model for
compliance assessments. In these cases, the model's ability to capture the high end of the concentration
distribution is evaluated with Q-Q plots where the highest data point from the model is compared to the
highest data point from the observations even if they occur at different locations, time of day, and season
of the year). In the REA, the model is being used to support health studies where spatial and temporal
accuracy is much more important compared with compliance assessments. The model results need to be
evaluated against observations paired in time and space. The REA needs to discuss "acceptable" model
performance criteria and options for correcting the model results if there are significant biases in the
modeling results. For example, would it be acceptable to over predict the SO2 concentrations by a factor
of 2? How about 5? How about 10? If not, what will be done with these modeling results?
A-6

-------
Also, the overall model performance may look good for the entire year due to compensating errors, but
the daily temporal profiles and/or seasonal profiles might be way off. EPA should consider adjusting the
REA model results up/down to match the SO2 observations. This approach would keep the relative
spatial distributions identified by the model in place, but would adjust the concentration levels to match
the observations. This would minimize the impacts from poor model performance on the ambient SO2
concentrations used in the exposure estimates.
2.	The use of SO2 measurements at ambient air monitors within and near the study areas to estimate
continuous 5-minute concentrations, where appropriate (e.g., filling missing values, for AERMOD
hourly predictions). [Sections 3.3.1, 4.1.3.1, 4.1.3.2]
The comprehensive 2011-2015 1-hour average and 5-minute SO2 data sets will help estimate continuous
5-minute concentrations. On page 4-13, proposed step #3 states "Substitute any hours not having
measurements for any year with a value of zero (0)." However, it might be more appropriate to
substitute missing values with the lower detection limit (2 ppb) or half the LDL (1 ppb), or interpolate
between the values before and after the missing value(s).
Equation 4-1 and Equation 4-3 are simple ways to fill the missing data. The REA proposes "to use a
linear ramp" for Equation 4-2, but this equation is only truly linear for the case where the 5-min max is
equal to 2 x 1-hour average. Equation 4-2 should be updated since the cases where the 5-min max is >2
x 1-hour average produces 5-min values for Ce - Ci that are lower than the cases where the 5-min max is
<2 x 1-hour average. Also, an exponential ramp may be more appropriate than a linear ramp for cases
where the 5-min max is >2 x 1-hour average. In addition, a minimum value should be assigned. Options
for setting an appropriate minimum value could include using 50% or 25% of the 1-hour average SO2
concentrations for the current hour or using the 1-hour average concentration for the hours before and
after the current hour (assuming they are significantly lower than the current hour).
3.	The proportional approach selectedfor adjusting ambient concentrations to simulate air quality that
just meets the existing standard. [Section 4.1.3.4]
In general, the proportional approach selected for adjusting ambient concentrations to simulate air
quality that just meets the existing standard seems to be appropriate. On page 4-20, it is stated "For the
planned REA for the current review, in each study area, F will be calculated by dividing 75 ppb by the
DV and will be used to adjust all SO2 concentrations in a study area by this factor to simulate just
meeting the existing standard." In order to just meet the standard, it would seem appropriate to only
adjust the SO2 concentrations for those receptors with a design value above 75 ppb and leave the
receptors with a design value already below 75 ppb unchanged.
Study Areas and Time Periods (Section 4.1.2.1)
It seems appropriate to model three consecutive years to evaluate variability in exposures across a 3-year
period since this is consistent with the form of the existing standard. Also, modeling domains with
receptors within a 10-km radius of all nearby SO2 emission sources greater than 100 tons/year is
appropriate.
A-7

-------
Four of the nine candidate study areas listed in Table 4-1 were selected for detailed analysis. It would be
helpful to state the specific reasons why each of the other five areas were not selected. Finally, the maps
in Figures 4-2, 4-3, 4-4, and 4-5 are difficult to read and interpret. I suggest showing DVs 0-75 (blue),
76-100 (yellow), and >100 (red). For the large SO2 emission sources, it would be better to represent the
emission sources with different size bubbles that are representative of the size of the SO2 emissions
(small bubble for low emissions, larger bubbles for higher emissions). Also, I would suggest break
points of 100-500, 500-1000, 1000-2000, 2000-5000, >5000 tons.
Also, it is not clear how the four study areas (less than 1.5M people) will be used to calculate the
number and percent of the total population across the country (> 325M people) experiencing 5-minute
SO2 exposures at or above benchmark levels of concern, the number of occurrences of lung function
decrements in the at-risk populations across the country, and the number and percent of the at-risk
populations across the country estimated to experience single or multiple occurrences of those lung
function decrements. This needs to be done since this is a national standard. How will populations
exposed to much higher or much lower ambient SO2 concentrations in other areas of the country be
factored into the analysis?
Other Potential Standard Levels
With so many improvements being implemented to better characterize the 5-minute SO2 concentrations
in ambient air (new 2011-2015 SO2 measurements and enhancements to AERMOD) and the
improvements to the exposure assessment (APEX and CHAD), it would seem appropriate to re-evaluate
other potential standard levels (50 ppb and 100 ppb) besides the current level (75 ppb). If significant
impacts are indicated by the current level of the standard, then a level of 50 ppb should be evaluated. If
minimal impacts are indicated by the current level of the standard, then a level of 100 ppb should be
evaluated.
A-8

-------
Dr. Judith Chow
Analytical Approach and Study Area Selection
2. The criteria identified and approach used to select potential study areas to evaluate for this REA.
[Section 4.1.2]
The four candidate study areas are reasonable as they meet the criteria for air quality data, design values,
and population size. The four areas include different SO2 sources (e.g., steel mill in Cuyahoga County,
OH, lead smelting in Marion County, IN, pulp and paper in Brown County, WI, and a fertilizer plant in
Hillsborough County, FL) in addition to electric generating units (EGUs). However, three of these areas
are close to water bodies (with the exception of inland Marion County, IN). The selection does not cover
all relevant geographical regions (i.e., Midwest, Northeast, South, and West) defined by the National
Health Interview Survey (NHIS, Page 4-26). Complex terrain features (e.g., plume impact on elevated
terrain) are ignored. Sites near the ocean or a lake may experience additional moisture resulting in
enhanced SO2 to sulfate transformation and therefore may not represent typical SO2 exposure.
Although large SO2 sources (>100 tons per year) are shown in Figures 4-2 to 4-5 (Pages 4-8 to 4-11), the
range of emissions from 4,821 to 142,920 tons per year spans more than an order of magnitude; more
refined divisions of large sources would be desirable. It would be helpful to provide the most recent
(e.g., 2013 to 2015) statistical summary of 5-minute hourly and daily 1-hour maximum values, as well
as hourly and daily average values, for the selected areas. With a design value of 65 to 85 ppb in these
four areas, the number of days that 5-minute maximum values exceeded 100, 200, 300, and 400 ppb
benchmark concentrations should be given. Figure B-l of Appendix B shows that nationwide, there are
30 to 90 days with SO2 concentrations exceeding 125 ppb. Perspective should be given with respect to
the exposure-response of these elevated SO2 concentrations.
Detailed national statistics are given in the second draft SOx ISA (U.S. EPA, 2016a) for six focus areas
during the 2013 to 2015 period. Other than four sites in Ohio included in both the Cleveland-Elyria-
Mentor and Cuyahuga County study areas, none of the focus areas (i.e., Gila County, AZ, St. Louis,
MO-IL, Houston-Sugar Land-Baytown, TX, Pittsburgh, PA, and New York-Northern New Jersey-Long
Island, NY-NJ-PA) correspond to the selected REA modeling areas. The EPA may want to consider
including some of the ISA focus areas (especially those like Gila County, AZ) in the West with high
copper smelting emissions (>21,747 tons/year) and elevated (282 ppb) 5-minute hourly maximum SO2
concentrations (see Figure 2-18, Page 2-43 of the second draft SOxISA).
Ambient Air Concentrations
1. The use of an AERMOD model-based approach to predict hourly concentrations at all receptor
locations within selected study areas. [Sections 3.3.2, 4.1.3.3]
It is encouraging that the EPA has made several updates to the AERMOD modeling system and its data
processors such as AERMET, AIRMINUTE, and AERMAP (U.S. EPA, 2016b). With the improvement
A-9

-------
in terrain and meteorological processors, better agreement should be found in hourly SO2 concentrations
between AERMOD model predictions and ambient SO2 measurements. Sensitivity tests need to be
conducted with the new options for adjustment of surface friction velocity under 1-minute low wind
speeds and building downwash to demonstrate the improvements in AERMOD model simulation and
exposure assessment. Performing preliminary model runs for surrounding sources to determine the
spatial scale that best captures concentration gradients seems reasonable. As the current AERMOD
model only predicts hourly SO2 concentrations and does not account for plume looping and short-term
touchdowns, the EPA is encouraged to further update the AERMOD model to estimate shorter
averaging time (e.g., subhourly).
2.	The use of SO2 measurements at ambient air monitors within and near the study areas to estimate
continuous 5-minute concentrations, where appropriate (e.g., filling missing values, for AERMOD
hourly predictions). [Sections 3.3.1, 4.1.3.1, 4.1.3.2]
Although the number of monitors reporting 5-minute concentrations has increased since 2011 (Figure 3-
1, Page 3-6), it is unfortunate that only -40% of monitors in the compliance network report 12
consecutive 5-minute measurements for each hour (Page 3-5). A great deal of effort has been made
(Section 4.1.3.2) to estimate the unreported 5-minute concentrations where only the hourly maximum 5-
minute concentrations are reported.
The approach to estimate the other eleven 5-minute measurements seems reasonable. However, the
progressive decrease in SO2 concentrations in Tables 4-3 and 4-4 (Page 4-16 and 4-17) does not
necessarily reflect the frequency and duration of plume touchdown and downwash mixing, adding
uncertainties to the modeling results.
Federal Reference Method (FRM) instruments for SO2 are capable of producing short-duration averages;
consistent reporting of each 5-minute average by the states is preferred. This would allow for the
examination of consecutive elevated 5-minute SO2 concentrations and a clearer understanding of the
exposure durations and diurnal variations. The EPA is encouraged to acquire past 5-minute
measurements from states (even if it is not fully quality assured) to verify the adequacies of predicting
the other eleven 5-minute SO2 measurements. In the future, the EPA should mandate that states report
each 5-minute average SO2 measurement, as it will provide a database to evaluate a future 5-minute SO2
NAAQS indicator.
3.	The proportional approach selectedfor adjusting ambient concentrations to simulate air quality that
just meets the existing standard.
[Section 4.1.3.4]
As the highest design value (DV) is used to derive a single multiplicative factor (/•') to adjust the
monitored concentrations across the study area, the selection of an appropriate DV is important. The
REA asserts that the adjustment for ambient concentrations used in the exposure assessment is likely to
be small (<10%, Page 4-20), inconsistent with the large variations in DV values (from 78 to 92 ppb in
Marion County, IN and 66 to 93 ppb in Tampa, Hillsborough County, FL over the 2011 to 2013, 2012 to
2014, and 2013 to 2015 periods) shown in Table 4-1 (Page 4-7). The representativeness of DVs needs to
be clarified.
A-10

-------
References
U.S. EPA (2016a). Integrated Science Assessment for Sulfur Oxides—Health Criteria, Second External
Review Draft. EPA/600/R-16/351. U.S. Environmental Protection Agency, Research Triangle Park,
North Carolina 27711.
U.S. EPA (2016b). User's Guide for the AERMOD Terrain Preprocessor (AERMAP). EPA-454/B-16-
012. U.S. Environmental Protection Agency, Research Triangle Park, North Carolina 27711.
A-ll

-------
Dr. Aaron Cohen
General Structure [Section 4.2]
The overall plan for the REA Health Risk Assessment is, for the most part, clearly described. It appears
methodologically sound and consistent with the health evidence as reviewed in the draft ISA.
Approaches for using findings from controlled human exposure studies
The choice of health endpoints appears well-justified (page 4-31, Section 4.2.1).
Page 4-32, para. 1:1 assume that the scaling of ventilation by BSA is the appropriate way to handle
adult-child differences, this approach having been used in other REA, but this is not my area of
expertise.
Health benchmarks [Sections 3.2.2, 4.2.3]
The benchmark levels seem appropriate given the design of the controlled exposure studies.
Plans for E-R functions [Sections 3.2.2, 4.2.4]
The rationale for not including FEVi deserves further discussion. I am not sure I agree, given the ISA
review and though estimates may be less precise than for sRaw FEVi decrements in exposed asthmatics
are adverse. In any case, excluding FEVi seems to contradict what was said in Section 4.2.1.
Section 4.2.4, page 4-34, line 4: there seems to be a word (missing: ".. .people estimated to [have
experienced?] at least one.
Variability, covariability and uncertainty [Section 4.4]
The approaches to addressing variability in the underlying exposure and health evidence and
characterizing uncertainty seem conceptually sound but it is not entirely clear to me what the sources of
variability are that will actually be addressed. Perhaps a small table would help.
I understand that uncertainty will be characterized using sensitivity analysis, but I assume that the risk
estimates will be also presented with uncertainty intervals. This is not discussed but should be, including
which sources of uncertainty such intervals will include, e.g., uncertainty in exposure assignment as well
as in the fitted E-R functions.
A-12

-------
Dr. Alison C. Cullen
The focus of these comments is Chapter 4 - Plan for the Current Health Risk and Exposure Assessment
(Section 4.1 Population Based Exposure Assessment pages 4-1 to 4-30).
This section of Chapter 4 is very informative and lays out the approach EPA proposes for assessing
exposure in a systematic way. Below I outline questions/comments intended to sharpen details and
answer remaining questions.
For the assessment of human exposure to SO2, EPA has selected study areas to represent the US.
Over 100 potential areas fit the selection criteria related to air quality data availability and design
values however only nine satisfied the population size criteria, so clearly population size is a
pivotal criterion. Please say something about the choice of 100,000 as the cut point for
population size. What impact would a different cut point be expected to have on the study area
selection and ultimately the analysis? Also, please expand on the extent to which the final four
study areas are representative of exposure locations of concern regarding SO2 sources and how
these areas reflect on the broader characteristics related to exposure, given the application of the
set of selection criteria.
Further, regarding the selection criteria, either refer to another section of the REA or explain here
why 75% is considered to be complete enough to be the cut point for completeness. Also, clarify
the impact that a different value would have. Finally, a separate point to address/clarify - why are
areas with more complete data prioritized given that a model approach is ultimately used? Some
acknowledgement of the most "relevant" data or the most "valuable" data may be as important as
raw "completeness".
The equations used to estimate missing 5-minute concentrations of SO2 introduce an important
role for the maximum 5-minute concentration and its position relative to the average. Please say
more about the impact of this choice.
In Figure 4-6 given that the one or two data points at the top of the percentile distribution for
daily maximum one hour SO2 are very influential on the fit of the slope of the regression line for
each of the four locations - say more about the applicability of these derived relationships.
Although the exclusion of children under the age of 5 in the at-risk population assessment is
consistent with other REAs for other pollutants as stated in the REA for SO2, and its justification
is outlined, please say something further about the impact of this exclusion on the NAAQS.
Asthma status is obviously important when considering SO2 exposure and risk. Given that the
2014 REA for O3 is referenced for its improved treatment of asthma and geographic level
differences please say a bit more about the approach so that this REA can stand alone (while not
repeating all previous content). Are the four study areas representative of the US national range
in asthma prevalence? Is it possible to add gender breakdowns to Table 4-5?
Clarify on page 4-28 or earlier how interactions between asthma status and exertion level will be
handled, given the limitations of CHAD.
Refer readers to a fuller discussion in this REA (outside of the US EPA 2012b reference which is
a full treatment) of how the population diversity statistic D and the within-person autocorrelation
statistic A are applied (page 4-29).
A-13

-------
Interdependencies are very important as noted in section 4.1.6.5. Personal Attributes, but the
language is vague. How and when will these interdependencies be taken into account (beyond
just "where possible")?
Is asthma status a physical attribute which could be referenced in section 4.1.6.5.2 Physical
Attributes or does it belong elsewhere in the overall section 4.1.6?
A-14

-------
Dr. Delbert J. Eatough
I was specifically asked to focus on Ambient Air Concentrations. However, as I read the document my
areas of concern centered on both the material on Analytic Approach and Study Area Selection (Section
4.1.2) and Ambient Air Concentrations (Section 4.1.3). My comments will address both of those areas. I
will focus on how decisions in the structure of these two sections effect the development of the
"Exposure Modeling (Apex)" purple box and subsequent development of the "Lung Function Exposure-
Response Function Relationship" red box in the Figure 4-1 (page 4-2) Overview of the analysis
approach for the REA. In particular, I will comment how decisions made on the Study Area Selection
(Section 4.1.2) may well lead to the underprediction of response.
Background on Areas of the ISA and REA which contribute to my concerns.
In my written comments on the ISA I pointed out that CAS AC had requested the following of EPA in
the development of the Second Draft ISA:
In the April 15, 2016 letter to Administrator McCarthy we stated,
"The CAS AC finds that the source categories and definitions of major sources are inconsistent
throughout Chapter 2 as well as the entire ISA and recommends that these be consistent. The
chapter should include locations and emissions for point sources (energy-generating units,
integrated steel and iron mills and smelters) near urban centers."
In the 03/10/16 Draft Report we further stated,
"The importance of pollution sources and formation of non-sulfate compounds such as inorganic
particulate S(IV) species, organic S(IV) species (e.g., bis-hydroxy dimethyl sulfone) and organic
S(VI) species (e.g., alkyl sulfates) requires additional discussion. Studies such as Alarie et al.
(1973) and Amdur (1971) demonstrated the relationship between exposure to inorganic S(IV)
compounds and exacerbation of SO2 inhalation responses in animals. These compounds are
potential confounders or moderators of SO2 health effects in epidemiological studies where
copper smelter or integrated steel mill emissions are abundant and the possible influence of these
compounds should be discussed."
And in my Final Comments on Draft IRP I outlined in detail what was known about the above outlined
chemistry and recommended,
"Probably a more fruitful set of data to evaluate the relative importance of aerosol S(IV) species
associated with smelter emissions would involve past epidemiological studies from about two to
three decades ago when smelter emission were much more significant, for example from the TX
smelters in El Paseo (ASARCO Cu smelter, closed in 1999), and Corpus Christi (ASARCO Pb
smelter, closed in 1985), AZ smelters (ASARCO Cu smelter in Hayden, currently operating and
Phelps Dodge Cu smelter in Douglas, closed in 1987), from the Kennecott Cu smelter in Magna,
UT prior to construction of the tall stack, from the Tacoma WA smelter (American Smelting and
A-15

-------
Refining, a Cu smelter specializing in high As ore refining, closed in 1985), or the smelters in
Montana (ASARCO Pb smelter in East Helena, closed in 2001, Anaconda Cu smelter in
Anaconda, closed in 1981) and Idaho (Bunker Hill Pb smelter in Kellogg, closed in 1982). I
know that several epidemiological studies were conducted at these locations, but I am not
familiar with the results of these studies with respect to asthma exacerbation. I recommend that
EPA look at this older data to see if an estimate of the relative potency of S02 and smelter
associated aerosol S(IV) species can be determined. There will not be data on the concentrations
of S(IV) in the aerosols emitted from these sources, so total particulate exposure would need to
be used as a surrogate. The importance of elucidating the effect of these exposures is correctly
alluded to in the ISA on Page 4-12, Line 11."
My comments on the Second Draft ISA point out that these requests were not responded to in the second
draft ISA with the following two consequences:
1.	It would appear from data in the ISA that the highest anthropogenic associated concentrations to
which a population is exposed under current SO2 emissions is dominated by emissions from
smelters and integrated iron and steel mills. Further, with respect to current conditions, high
exposure concentrations resulting from emissions from EGUs is rare. I have asked in my
comments on the ISA for additional information from EPA in the ISA to make this point clearer.
2.	The request to explore the hypothesis that the presence of particulate inorganic S(IV) species in
emissions from smelters and integrated iron and steel mills will result in a greater exacerbation
of asthma in exposed populations will lead to a higher risk than exposure to SO2 alone, such as
you might see in emissions from an EGU was not considered by EPA.
Consequence 1. means that the development of a Risk and Exposure Assessment document which
focuses on EGU emissions will underestimate the highest exposures which will lead to the highest risk.
Consequence 2. Means that if the hypothesis is correct, the risk will be further underestimated by not
focusing on the higher emissions to which populations are exposed from living near a smelter or an
integrated iron or steel mill and which emissions are associated with a higher risk from sulfur oxides
than that due to only SO2.
Section 4.1.2 Exposure Domain
There are two statements in the material in Section 4 which precede 4.1.2 which appear to me to be
contradictory and which directly affect the choice of Exposure Domains. On page 4-1 first paragraph the
REA states "The objective for the REA for this review is to characterize exposure and health risk
associated with S02 from ambient air under conditions just meeting the current primary standard." In
Section 4.1, page 4-3 last paragraph the REA states "Additionally, as part of this analysis, the
population-based statistical distribution of exposures will also be evaluated to identify important
exposure environments and/or influential activities that lead to those estimated as having greatest
potential health risk." I would think the second point is more important than the first if the objective is to
identify the risks which should guide decisions on whether the SO2 standard needs to be revised.
A-16

-------
With these points in mind, let me comment on the four identified potential study areas in Section 4.1.2.
and on the associated figure for that study area.
Fist a general comment on the four figures: Figures, 4-2 through 4-5 are a little confusing. The various
sources shown are given in the Key as squares, but show in the figure as circles.
It would also be very helpful if the sources were specifically identified and not only listed in Table 4-1
as to type and not as to size.
Brown County WI contains Green Bay. It is not one of the sites identified in Figure 2-11 of the second
draft ISA as a site with the 99th percentile of 1-h daily max sulfur dioxide concentrations reported to be
above 75 ppb. The data in Table 4-1 indicate the DV is just at 75 ppb. Impact appears to be from pulp
and paper facilities, mostly near the single monitor in the Study area. It would help if the specific
sources and emissions were given. I assume the EGU is the large source in the central circle. I further
assume that it has a tall stack and will not significantly impact the single monitor in the study area. I
would consider this study area (with only one monitor) to be less valuable than the FL study area.
Cuyahoga County OH contains Cleveland. It is one of the sites identified in Figure 2-11 of the draft ISA
as a site with the 99th percentile of 1-h daily max sulfur dioxide concentrations reported to be above 75
ppb. The DV in Table 4-1 of the REA for this study area is 62. There are four monitors shown in the
study area. The highest concentration for the four is for MONID 390350060. According to Figure 2-13
of the second draft ISA the 99th percentile of 5-minute hourly max sulfur dioxide concentration at that
monitor during 2013-2015 was 61 ppb. Based on the data in Figure 2-13 of the second draft ISA I
believe the emissions source to the right of the four monitors shown in Figure 4-3 is the 2133 tpy
ArcelorMittal integrated iron and steel mill. The smaller emissions source in the middle of the four
sampling stations in Figure 4-3 of the REA is not shown in the ISA. It would be useful to know what
that source is. The advantage of this study area is the presence of four monitors to aid in the APEX
analysis. It is possible that these four monitors are influenced by emissions from an integrated iron and
steel mill. The highest monitoring site in this area given in Figure 2-13 of the ISA is E, which averages
85.7 ppb. Both the monitor and the nearby emissions source are within the domain shown in Figure 4-3,
but they are not shown. The nearby 2745 tpy emission source is not identified, but I have asked for that
to be added to the draft ISA. The large red source in the middle of the upper circle for Figure 4-3 is not
identified, but I am sure it is an EGU (with an adjacent Monitor with a 20 ppb average values not shown
in Figure 4-3) given in Figure 2-13 of the ISA. Distance from Cleveland to ISA E is 30 mi. The monitor
by the power plant is about midway between the two. Why are these two additional monitors (and site)
near Cleveland not included in the study area? It does not seem wisdom to not include the monitor with
the highest SO2 readings and a clear impact from a nearby source in the analysis. If all potential data
were used, this could be a very viable study area. It further has the advantage that it would look at the
probable impact from an integrated steel mill. I would surly also like to know that the emission source
near Painesville is. If it is the Painesville Electric Plant (which does not appear to have a tall stack) it
would be a unique opportunity to include the impact of an EGU in the study area analysis.
Hillsborough County, FL contains Tampa. It is not one of the sites identified in Figure 2-11 of the draft
ISA as a site with the 99th percentile of 1-h daily max sulfur dioxide concentrations reported to be above
75 ppb. The DV in Table 4-1 of the REA for this study area has dropped over the years and is currently
66. Table 4-1 indicates there is only 1 monitor to be included in the analysis for the study area and
sources are Fertilizer Plants and an EGU. However, there are five monitoring sites shown in Figure 4-4.
A-17

-------
Why are the other monitoring sites not being included in the APEX analysis? This would surely increase
the power of the analysis. I assume the red circle (the key looks like it should be a square) centered on
the bottom blue circle is the EGU. I also assume it has a tall stack and will not have a major impact in
the study area. What, specifically is the source vert near to the MONID 120570109 monitoring site.
How probable is it that, even if it has emissions less than 1136 tpy, it will impact the site. Do the data
from the nearby monitor suggest this will be the case? I consider this a reasonable study area if all
monitoring site data are included in the analysis.
Marion County IN contains Indianapolis. It is not one of the sites identified in Figure 2-11 of the draft
ISA as a site with the 99th percentile of 1-h daily max sulfur dioxide concentrations reported to be above
75 ppb. The "lead smelter" listed in Table 4-1 for Marion County IN is actually the RSR-Quemetco
Battery Recycling Facility on the west side of Indianapolis (identified as yellow and directly west of the
monitoring site MONID 180970057) and is not a smelter. There is only one monitor in the study area
and it has a DV 79. There are two other monitoring stations identified but the key indicates they have no
valid data. I would consider this site (with only one monitor) to be less valuable than the FL site.
Your choice of sites is skewed towards lower concentrations and generally avoids emissions from
smelters and integrated steel mills. Because of the general use of tall stacks, EGU emissions will be low
at all sites, with the possible exception of Cuyahoga County if the study area were enlarged as suggested
in my comments. I suggest you at consider dropping both the proposed WI and IN study areas to give
additional areas with multiple monitors and, ideally, increased attention to emissions from integrated
iron and steel mills and smelters.
I suggest you consider including the Detroit, Wayne Co, MI study area. It would be useful to have a map
of that study area like Figures 4-2 through 4-5 to further evaluate that possibility. If the data from the
multiple monitors (6) in this potential study area could all be used in the APEX analysis this would be a
strong point for including Detroit as a study area. In addition, if the possibility of looking at the impact
of emissions from the Zug island steel mill or the Trenton Channel Power Plant located near the steel
mill or the closely located together DTE Belle River Power Plant and St. Clair Power Plant in the
northeast part of Detroit existed this would further indicate it would be an excellent study area. The three
mentioned EGUs do not have tall stacks and they may contribute to more local impacts. The last point
could be determined by examination of the data from the six monitors in Detroit.
I suggest you also consider adding Gila County (Figure 2-18 of the second ISA draft) as a study area. I
recognize that it does not include the population minimum of 100,000 you listed in your criteria for a
study area (current county population, 53,000). However, it does contain four monitoring site, each of
which have 99th percentile 5-minute max concentration above 100 ppb during 2012 - 2015 (Figure 2-18,
second ISA draft). It is the only place where exposure of a population to emissions from smelters can be
modeled. As summarized in my comments on the ISA and my discussion at the start of these comments,
EPA has been charged by CASAC to consider the effect of exposure to particulate inorganic S(IV)
compounds in emissions from smelters as "These compounds are potential confounders or moderators of
SO2 health effects in epidemiological studies where copper smelter or integrated steel mill emissions are
abundant and the possible influence of these compounds should be discussed."
If EPA responds to the request to explore the hypothesis that these compounds are confounders of
exposure to SO2 alone and that the effect of the exposure to both is to increase the exacerbation of
A-18

-------
asthma, then failing to consider smelter emission in the REA analysis will lead to an underestimation of
the Lung Function Exposure - Response Relationship and the Lung Function Risk (Figure 4-1). Of
course, if EPA does not explore the hypotheses and the hypothesis is correct, the health effects of
exposure of asthmatics to SO2 will still be underestimated by EPA as decisions on the future form of the
standard are made.
This will be true no matter how well the modeling exercise described in Section 4.1.3 is conducted.
Section 4.1.3 SO2 Concentrations in Ambient Air
Section 4.1.3.2, page 4-13. Second paragraph. It is not clear to me that the assumption that "where
ambient air measurements are missing likely occur at times where concentrations are relatively low,
thereby yielding slightly lower means and standard deviations when comparing substituted data relative
to the unsubstituted data" is valid. It seems to me that missing data are due to a monitor problem and not
an ambient air concentration condition. What will be the effect if this assumption is not valid and the
missing data are high?
I feel uncomfortable with Equations (4-1), (4-2) and (4-3). It seems to me that an important feature of
the analysis should be the use of multiple monitors and a study area to predict exposure using AERMOD
and APEX. How can 5-minute data from multiple monitors have any comparison value if estimated as
outlined. Are we limited to hourly average modeling for sensible results? I will let modelers in the group
comment on this.
A-19

-------
Dr. H. Christopher Frey
This review focuses on Chapter 4: Plan for the Current Health Risk and Exposure Assessment, on pages
4-1 to 4-30, with a focus on exposure assessment.
The key points from this review are:
•	Please explain why there are no plans for exposure assessment scenarios at levels of air quality
below the current standard.
•	How will the dispersion modeling approach be "supplemented and informed" by air quality
monitoring measurements?
•	Why use a linear ramp (Equation 4-2)?
•	Will (and if so, how) will transport time be accounted for with regard to modeled concentration?
•	Page 4-19 - there seems to be some ambiguity between the concept of a design value versus a
decision to look at annual data. The rationale here is not stated.
•	Figure 4-6: the R2 values do not imply that hourly data are correlated from one year to the next -
this needs to be clearly communicated. The meaning of R2 here, as an indicator of similarity in
the frequency distributions from one year to another, should be communicated carefully.
•	There are some detailed comments on Equations 4-1 to 4-4 and regarding the hour-day-month
specific factors. The latter are not clear.
Page 3-9: the new algorithm for resting metabolic rate should be documented, perhaps in an appendix, in
the 1st draft HREA unless it is documented elsewhere in a report that can be provided to CASAC (e.g.,
the documentation of the expected Spring 2017 release of the APEX update). Similarly, updates to the
algorithm using Ve (if different) should also be documented.
Page 4-3, the decision to evaluate potential risk associated with air quality adjusted to just meeting the
existing S02 standard (and recent unadjusted air quality) seems appropriate if the question is whether
the current standard protects public health with an adequate margin of safety. What about other levels
less than the current standard?
Page 4-6: "we are proposing to use the model-based approach to estimate ambient air concentrations in
each study area, supplemented and informed by available local ambient monitor measurements." Please
explain more clearly how the model-based approach will be "supplemented and informed by"
measurements.
Page 4-6 - it would help to say what cities are included in these counties. More broadly, the choice of
only four study areas needs discussion, as noted by other panelists.
Page 4-13: "Calculate an average distribution of hour-day-month specific factors using measured
values" not clear what is meant by "hour-day-month specific factors." Does this mean factors that are
A-20

-------
specific to each hour for a given day for a given month? If so, the sample size would be very small. Or
does this mean one factor for hour of day, another factor for day of month, and another factor for month
of year? Or other?
Page 4-13: "Instances of where ambient air measurements are missing likely occur at times where
concentrations are relatively low..Explain why this is.
Equation 4-1.1 finally figured this out, but the nomenclature is confusing. Nomenclature such as this
would be more clear:
Ci,h,d,m = estimated concentration for the ith 5 minute interval (I =1,12) in the hth hour for
the dth day in the mth month, [or could just be Ci,h]
Cmax,h,d,m — maximum 5 minute average concentration in the hth hour for the dth day in the mth
month, [or could just be Cmax,h]
Cave,h,d,m — hourly average concentration in the hth hour for the dth day in the mth month, [or
could just be Cmax,h]
i	= index for 5 consecutive non-overlapping 5 minute periods in an hour, from 1 to
12.
Thus, Equation (4-1) would become:
—		 (1*	^maar,h«d,fs
Or possibly:
_ _ (l2XCsgejft) — Cmajg|h
th ~	11
And so on for the other equations.
Page 4-15: Is there some underlying reason to use a linear ramp? This text just describes the linear ramp
but does not offer a reason/rationale for it.
Page 4-16: "Because there is improved representation of 5-minute concentration variability based on the
number of measurements in this data set." This is not clear - improved compared to what and in what
way?
Equation (4-3) is unclear. For example, n is not defined. Could be more clear, perhaps something such
as this:
12 — (n — m)
Where
Ci,h	= concentration for the ith measured 5 minute interval (I =l,m) in the hth hour
Cest,h = estimated 5 minute average concentration in the hth hour, applicable to each of the
n-m 5-minute periods that were not measured.
A-21

-------
Cavc.h = hourly average concentration in the hth hour for the dth day in the mth month, [or
could just be Cmax,h]
m	= number of measured 5 minute periods in hour h (must be 11 or fewer),
n	= number of 5 minute periods in hour h (12).
Page 4-17:
"would" "will"
"perspective" is vague and unclear -> "quantification of'?
Text could address how factors such as complex terrain, built environment, and chemical reactivity and
deposition of SO2 will be addressed. What is known about validation of AERMOD for application to
S02?
Page 4-18:
"can be" -> will be?
"we would need to" -> we will?
Equation 4-4: define the units as applicable.
Page 4-19: mention that the temporal pattern from one 5-min period to another at a monitor will be
delayed compared to the pattern at the emission source because of transport time. For example, for a
wind speed of 2 m/s, an air parcel would be transported 600 meters. Thus, a change in emission rate
would manifest as a change in concentration 5 minutes later at a distance of 600 meters, 10 minutes later
at a distance of 1,200 meters, and so on. At one hour, the air parcel would be 7.2 km from the source,
which would likely still be within the study region. Thus, at a wind speed of 2 m/s, it is likely that the
maximum 5-hour concentration would be misclassified by one hour compared to the timing of the
emissions that lead to such a peak. Or, to put this another way, simulated emissions at hour h would lead
to downwind air concentration at hour h+1 at a distance of 7.2 km to 14.4 km. Does this matter to the
exposure assessment? If so, how much? For sources that emit approximately continuously, exposure
concentration errors would have less error than for sources with more pronounced short-term temporal
variation. Perhaps it can be argued that industrial sources that are the main sources of S02 would be
operating approximately continuously. More information about the load or emissions profile over time
would be helpful.
Section 4.1.3.4 - page 4-19: as noted earlier, please explain the purpose of "air quality adjusted to just
meet the existing primary S02 standard, as well as for recent (unadjusted) air quality." Why not also
look at air quality below the existing standard?
Same paragraph - design values are based on three years of averaging, but it is not clear that this text is
using the term design value correctly. In a five-year period, there would be three design values,
assuming that the three-year average rolls from one year to the next. Thus, there would not be a low
concentration "year" but a low concentration 3-year average. Or explain why annual averages are used
instead of three year averages, and be careful to clarify the choice of design value as a basis for
assessing high or low concentrations in a given year. Please clarify. An example would help.
Page 4-20: Table 4-1 does not show adjustment factors. Need to explain more specifically how an
adjustment factor can be inferred from data given in Table 4-1.
A-22

-------
Figure 4-6: I think this makes sense from the perspective of evaluating whether two data sets are
similarly distributed. However, it should be acknowledged in the text that the data are not actually paired
this way. The data in both years have been sorted and the sorted values have been paired to assess
whether the distributions are similar from one year to another. This does not imply that the data are
correlated. The linear fit and R2 are indications of similarity of the hourly distributions but do not imply
that data in one year are correlated with data in another year. Some readers are likely to misinterpret the
very high R2 values to mean that hourly data in one year are highly correlated to data in the same hour
of a different year, which is of course not valid.
Page 4-23: It would help to either state values of a, P, and k or, if these values (and distributions) have
been developed and reported elsewhere, cite references for the values (and distributions) to be used. Or
otherwise give more insight as the basis of these.
Page 4-28: a comparison of APEX model results (for PM2.5) based on using the Markov-chain
clustering (MCC) algorithm, diversity and autocorrelation approach, and random resample is given by
Che et al. (2014) (Risk Analysis, 34(12):2066-2079).
A-23

-------
Dr. Steven Hanna
Note that my expertise is primarily in atmospheric transport and dispersion modeling and analysis of
observed concentrations, and my comments focus on those areas. I was asked to comment on the areas
related to "Ambient Air Concentrations", and focus on three specific topic areas copied below in bold
italics.
First I have a general request: Please explain why my suggestions on the SOx ISA were not followed here.
These comments regarded better use of the literature and existing methods for estimating peak to mean
concentrations and space and time variability of concentrations (the EPA work on 5 min vs 60 min peak
concentrations appears to have been done without a general review of the topic).
Ambient Air Concentrations
1. The use of an AERMOD model-based approach to predict hourly concentrations at all receptor
locations within selected study areas [Sections 3.3.2, 4.1.3.3]
I agree that, of the existing recommended suite of EPA air quality models, AERMOD is best and that
AERMOD has a basic averaging time of one hour. CMAQ is certainly not applicable. However, line 4 of
p 3-7 suggests that a comprehensive review has been done. If that had happened, the EPA authors would
have discovered that most dispersion models used by other agencies can easily model smaller averaging
times such as 5 minutes. SCIPUFF is a good example, and I thought that SCIPUFF/SCICHEM was on
EPA's list. From a basic science viewpoint (see dispersion texts by Pasquill, Stull, Arya, etc.) the models
for industrial sources can apply to any averaging time. What is needed is parameterization of how
turbulence and turbulence scales vary with averaging time, and this is known and parameterized in models
such as SCIPUFF. AERMOD includes these parameterizations in order to provide required turbulence
inputs. These formulations are "hidden" within the AERMOD software and operational users are not able
to change them.
On 16 March, I confirmed the above facts during phone calls with Jeff Weil, former leader of the
AERMIC group that developed AERMOD, and Akula Venkatram, former member of AERMIC who
developed many of the AERMOD algorithms. I suggest that a subcommittee be formed to advise EPA on
these key scientific facts.
I also asked David Carruthers, ADMS developer (in the UK), and he replied that ADMS has adjustments
to plume width to account for varying averaging times (less than 1 hr). ADMS is the equivalent of
AERMOD and is widely used as a regulatory model in Europe and Asia.
All plume dispersion experts agree that the ratio of 5 min to 1 hr peak concentration is dependent on
stability (i.e., time of day and wind speed and cloudiness) and on nearness to major point sources or
industrial complexes. The largest peak to mean ratios occur near large point sources during meandering
conditions, which are known to have periods of about 5 to 10 minutes.
A-24

-------
The estimation of 5 min values described in section 4.3.3 should also incorporate scientific knowledge
By the way, we feel that the EPA allegation (e.g., line 6 of section 4.1.3.3) that AERMOD is fine for
showing spatial variability but does not show time variability is not a correct generalization and is at odds
with our basic science concepts. Network observations and model simulations verify a large diurnal
variability in SO2 concentrations.
As described on p 4-18, it is a good idea the evaluate the model predictions with the 1 hr and 5 min
observations. However, the 1992 EPA model performance measures have been revised as described in the
2005 AERMOD evaluation report and journal article.
p 4-18, first sentence of second paragraph - As described above, the EPA should at a minimum
acknowledge that this purely-statistical and arbitrary approach could be enhanced by revising AERMOD
to calculate 5 minute concentrations, following existing methods in SCIPUFF, ADMS, and most other
dispersion models used across the globe. AERMOD already has the basic turbulence parameterization
formulas and they can be easily modified to allow for averaging times other than 1 hr. Also, the
preprocessor AERMINUTE can provide one-minute averaged wind inputs.
p 4-19 top half of page - Here too there is a need to use existing basic science concepts. Maybe the basic
science approach does not give significantly different improvements over the arbitrary statistical approach.
But the comparison exercise should be used to demonstrate this.
2. The use of SO2 measurements at ambient air monitors within and near the study areas to estimate 5-
minute concentrations, where appropriate (e.g., filling missing values for AERMOD hourly predictions)
[Sections 3.3.1, 4.1.3.1, 4.1.3.2]
My main comment on this has already been stated in my general request at the beginning. In scientific
studies, it is required that the existing literature first be reviewed. Gifford (1960), Slade (1968) and Turner
(1970) all discuss observations of peak concentration variation with averaging time and propose some
simple formulas. Ralph Larsen (career employee of EPA) spent decades studying this topic and published
many papers, (e.g., see his paper on the log-normal distribution in the 1974 symposium proceedings
referenced below). The literature from 40 to 50 years ago suggests that, on average, peak concentration is
inversely proportional to averaging time (Ta) to the 1/5 power. Thus peak C (5 min)/peak C (1 hr) would
be about 1.6. More recent papers suggest a power of V2 could also be used for certain types of sources,
which gives a ratio of 3.46. In the current report, the EPA should have first reviewed the literature (inside
and outside EPA), then describe where the existing methods are not appropriate, then justify why an
alternate method is being used. Finally, the estimates of the new method should be compared with those of
the old methods.
Section 4.1.3.2 addresses estimating missing values in the air monitoring data. This problem often comes
up in all fields of environmental study. However, in the atmosphere, there is a need to preserve known
correlations in space and time, and these are followed in several EPA analyses that I have reviewed. But I
do not see those principles being followed here. For example, persistence or linear interpolation are
usually assumed if there are only a few missing data. If available observations are indicating a high-
pollution afternoon, then any missing data are likely to be high too. The current EPA report's method
seems to go back to climatological values.
A-25

-------
Similarly, the method on p 4-15 for "filling in" the missing 115 minute values, when only the max 5
minute value in an hour is reported, is non-scientific. At the minimum, use the known probability
distribution function (pdf) of concentration variability (e.g., log-normal is often used). Known correlations
(in time) among 5-minute values could be incorporated too.
3. The proportional approach selectedfor adjusting ambient concentrations to simulate air quality that
just meets the existing standard [Section 4.1.3.4]
I have no comments since I am not sure of the rationale behind what is being done in this section.
Additional comments on the first part of the document (not covered under topic areas discussions)
p 1-2, lines 9-10 - "Advances in modeling tools and techniques and air quality data that have become
available since the last review are also considered" - This statement implies that there has been a
comprehensive review of the field in general; however, the current report suggests that, in my topic area,
only internal EPA documents and work have been considered.
p 1-4, lines 13-15-1 disagree with this statement that "concentrations of SO2 in ambient air do not
exhibit consistently strong temporal variability over daily or seasonal time scales..." Maybe this is true for
samplers in rural areas with no industrial sources nearby, but it does not apply in urban or industrial areas
or within 20 km of large point sources. And I notice later that the regions being considered for further
study are all urban metropolitan areas and include large point sources.
p 1-7 References - All references are to EPA reports instead of to general (possibly outside of EPA)
relevant reports and papers.
p 2-24 References for EPA 2009 REA review section - There are no basic references on atmospheric
processes, dispersion models, or statistical analysis of air quality data provided.
Chapter 3 introduction paragraph (p 3-1) and Conclusions Section 3.5 (p 3-10 and 3-11) - These sections
emphasize that "newly-available information" is being assessed with respect to its potential effects on risk
assessments. The EPA apparently is defining "new information" as that developed within their group (e.g.,
more 5-minute concentration data and revisions to AERMOD and its processors). I would hope that EPA
would also consider relevant information from groups outside of their agency (and outside OAQPS). For
example, the dispersion models of most European countries can model concentrations at a variety of
averaging times, including less than 1 hour. Within the U.S., the DOD's SCIPUFF dispersion model can
also handle any averaging time (following the theoretical derivations in basic turbulence and dispersion
textbooks such as Pasquill, as well as fundamental dispersion formulas by Taylor, Batchelor, and
Richardson). Note that a version of SCIPUFF called SCICHEM is an alternate model available from EPA.
Chapter 3 References include no "non-EPA" studies of statistical analysis of air quality data, and do not
include the various recent updates to the AERMOD dispersion model itself.
A-26

-------
References
EPA, (1974). Proceedings of Symposium on Statistical Aspects of Air Quality data. Chapel Hill NC Nov
1972, EPA-650/4-74-038 Environmental Monitoring Series, 270 pp.
https://nepis.epa.gov/Exe/ZyPURL.cgi?Dockev=2000X6RH.txt.
Gifford, F.A., (1960). Peak-to-average concentration ratios according to a fluctuating plume model. Int. J.
Air Poll. 3, 253-260.
Slade, D.H. (1968). Meteorology and Atomic Energy, U.S. Atomic Energy Commission (No. TID-24190).
U.S. Atomic Energy Commission, pp 109-111, 154-156.
Turner, D.B. (1970). Workbook of Atmospheric Dispersion Estimates (No. 999-AP-26). U.S.
Environmental Protection Agency, pp 37-38.
A-27

-------
Dr. Jack Harkema
Analytical Approach and Study Area Selection
1.	The overall analytical approach for the Risk and Exposure Assessment (REA) and its
appropriateness for developing spatially and temporally varying 5-minute ambient SO2
concentrations, simulating population-based 5-minute peak exposures, and estimating study areas
health risk based on controlled human exposure study data [Chapter 4].
2.	The criteria identified and approach used to select potential study areas to evaluate for this REA
[Section 4.1.2].
Comments and Questions to be considered:
In Chapter 4, the authors should provide more rationale and justification for limiting the main objective
to characterize exposure and health risk associated with S02 from ambient air under conditions just
meeting the current primary standard. Why not include conditions below the primary standard?
On page 4-22, the microenvironments appear to be selected primary for adult asthmatics rather than
asthmatic children (e.g., day schools, preschools, elementary schools). Does the analytical approach
have a built in bias for adult rather than childhood asthma?
Since obesity is more prevalent in children and adults (including those with asthma) then when the
controlled human exposure studies were conducted, how will the proposed risk models take this into
account?
A-28

-------
Dr. Farla Kaufman
Health Risk Assessment
1.	The general structure and overall approach that staff plans to use for the risk assessment. [Section
4.2]
The general structure and overall approach as outlined in section 4.2 of the planning document seem
appropriate and clearly laid out.
2.	The approaches for using findings from the controlled human exposure studies.
a.	The health benchmarks identifiedfor this REA. [Sections 3.2.2, 4.2.3]
Generally, the approaches for using findings from the controlled human exposure studies with regard to
identifying health benchmarks seem appropriate. However, I would add the following caveat.
The REA states there is no evidence to indicate that individuals with severe asthma "would experience
moderate or greater lung function decrements at lower S02 exposure concentrations than individuals
with moderate asthma". However, based on the available data there is not sufficient evidence to the
contrary. As noted in the REA individuals with severe asthma are not generally represented in the
controlled human exposure studies (second draft ISA, p 5-21). The effects of exercise and S02 exposure
in this population are not sufficiently understood, thus have the potential to influence these benchmarks.
Also, noted in the REA (section 4.2.3.2. footnote 55), "that studies utilizing a mouthpiece to deliver
pollutant concentrations cannot be directly compared to studies involving freely breathing subjects, as
nasal absorption of S02 is bypassed during oral breathing..Although the comparison may not be
directly made, the ratio of nasal to oral breathing shifts during exercise where there is more oral
breathing if not all oral breathing. This would seem to be informative to conditions of breathing through
a mouthpiece.
b.	Plans for developing updated exposure-response functions, including the methodology, and
specific studies to be relied on, for estimating exposure-response relationships for lung function
decrements. [Sections 3.2.2, 4.2.4]
i.	The focus on specific airway responsiveness (sRaw) for this quantitative risk
assessment of short-term exposure-related endpoints,
ii.	The range of exposure concentrations appropriate to include in the dataset for
deriving the exposure-response function.
The data seem to support the proposed focus on sRAW for this quantitative risk assessment in terms of
short-term endpoints.
The approach for selecting the range of exposure concentrations to include in the dataset seems
appropriate and well-reasoned.
A-29

-------
The additional analyses, deriving the E-R functions with and without higher exposure concentrations,
seems important to investigate based on the 2009 REA and from the results presented in Figure 4-7 and
Table 4-7.
3. The approach for assessing variability/co-variability and characterizing uncertainty in each part of
the risk assessment and the approach for model sensitivity evaluations. [Section 4.4]
In some instances, more recent survey data is available for factors such as physical attributes. Use of
more recent data could reduce variability and should be used whenever it is available.
Body weight and surface area: Each simulated individual's body mass was assigned using body
mass distributions from the 1994-2004 NHANES data. More recent data is available and show a
change in the body mass distributions. Presumably using more recent data could change the age-
and sex-specific estimated BSA.
Recent data from the National Health Interview Survey show a higher prevalence of asthma in
Blacks especially in Black children (15.4%).
Uncertainties exist in many areas of risk assessment for children when using adjustments factors applied
to data from adults. Many differences between adults and children may influence children's responses
such as lung development throughout childhood, and breathing patterns in children during exercise
(which also may be influenced by other factors, e.g. gender, race, overweight/obese status). These
factors should be considered either quantitatively, where feasible, or qualitatively, where not.
Factors influencing asthma prevalence should also include race and obesity (4.1.6.2.).
This draft REA is based solely on exposure to sulfur dioxide. As known, exposure to mixtures of air
pollutants, is the real world experience and may have greater health impacts than exposure to single
pollutants. Thus, co-pollutants should not only be considered as confounders, but recognized as
potentially modifying factors of the response to S02 exposure, e.g. potentiating or acting synergistically.
Other Comment
Many sections of the document are well written. However, there are sections that could be written in a
clearer manner, as they require much deciphering to determine the information being conveyed (e.g.
3.2.2.).
A-30

-------
Dr. Donna Kenski
Analytical Approach and Study Area Selection
The overall approach seems sound and Chapter 4 starts out with a fairly clear description of the process,
but loses some of that clarity in the sections on adjustments to ambient data, as noted below. I liked the
choices of study areas, although it may be my Midwest bias. The criteria for selecting them were clear
and rational. Terrain wasn't mentioned as a possible complicating factor; should it be?
Ambient Air Concentrations
1.	Use of AERMOD: Based on what was presented in the REA and ISA, AERMOD seems like the
only reasonable choice, and the ISA did a good job documenting its strengths and some
remaining weaknesses. However, the panel discussed the possibility of modifying AERMOD to
model 5-minute data, instead of using the convoluted data adjustment procedures outlined in the
REA. If feasible, that would be a better approach; if not, the data adjustment processes need
much better documentation and explanation.
2.	Use of S02 measurements from ambient monitors: The processes described in Section 4.1.3 on
adjustments to the ambient data, then to the modeled data, were quite confusing. A diagram
might help readers keep the various steps straight. I understand how the 5-minute ambient data
estimation process works but not why it is being done this way. Some additional explanation is
needed. For example, why are two different methods being used to fill in the missing data?
Despite the stated goal of maintaining the features and bounds of the existing monitoring data,
neither of the 2 equations used to estimate 5-minute averages yield realistic-looking data. The
use of two different equations needs to be justified. If the goal is just to fill in with values that
add up to the 1-hour average, then one method of expanding the data should be sufficient.
Explain what advantages are conferred by using multiple methods, and why there is no effort to
generate a realistic distribution of the expanded data.
3.	Proportional approach to adjusting data to just meet the standard: This method is straightforward
and worked well for the last review. I see no reason to change unless issues arise once the actual
analyses are underway.
p.4-13, line 9: transpose is the wrong word. Perhaps you meant transfer or translate?
p. 4-16, last line: delete either
A-31

-------
Dr. Elizabeth A. (Lianne) Sheppard
Exposure Analysis
Charge Question 1:
Overall the approach seems reasonable. We want to think about areas that are representative as a distinct
consideration from those that have data. While without data to do a quantitative assessment this can only
be discussed qualitatively, such a qualitative assessment should be covered in the REA. (See comments
on exposure representativeness below.) The population size cut-off is reasonable based on discussion at
the meeting indicating that low population areas aren't very informative. Consider reporting proportions
in addition to absolute numbers as a way of facilitating broader interpretation. I agree with simplification
of microenvironments.
Charge Question 2\
The selected population groups seem reasonable. Some modifications to the approach, e.g. in sensitivity
analyses, may help foster a generalizable interpretation. For instance, consider using prevalences from
other areas in an effort to make the results more nationally representative. Also consider using different
assumptions for population density and proximity to monitors to make the results more nationally
representative, as well as other environmental justice considerations. I note that the population
characteristics can be separately overlaid on the areas since these are simulation studies. While choosing
areas is limited by monitoring data, since the population information could be overlaid in other ways,
this will facilitate more nationally representative estimates.
General comments in the following bullets:
Comments goal: Think about what analyses (e.g. sensitivity analyses) should be considered to enable the
REA to address the protectiveness of the standard for public health and to ensure analyses are nationally
representative
•	Calibrating and getting 5-minute averages:
o Filling in 5-minute data from the 1-hour average and 5-minute max:
¦	Is it worth doing a more extreme approach where the 2nd 5 minutes is close to the
maximum, as long as the mean is not exceeded, then continue until the remaining 5-
minute values are 0 or small?
¦	Alternatively use an exponential ramp? Is this also more extreme than the current
approach?
o Calibrating: Consider using measurements to adjust AERMOD values, calibrating them to
measurements at nearby monitors.
•	Asthma prevalences: Consider prevalences for other areas (e.g. places with higher prevalences) to
address national impacts? Also obesity distributions from more obese locations? We can also
consider other environmental justice types of considerations.
A-32

-------
•	Approaches to using findings from controlled human exposure studies. (Notes from the meeting
discussion)
o Roughly 6 m children with asthma. Doubling airway resistance will make individual
asthmatic children symptomatic. How we extrapolate to children from the (healthier) adults
with asthma in the controlled exposure studies is an important consideration.
o Is there a threshold? In the population, there will be someone at risk at every exposure, so on
a population level probably no threshold exists.
•	Representativeness of exposure
o We'd like areas to be studied that will facilitate us to extrapolate to the entire US. However,
it seems that the areas to be studied all have sources. Is it worth quantifying the proportion of
the US population within 10 km of sources?
o Monitors: Which are source-oriented, which are population oriented? How relevant is this to
the REA?
o DV areas: How many areas exceed the DV by more than the upper limit used to select areas?
How should this understanding be incorporated in the REA? Or does it belong in the PA?
•	Assessing variability, characterizing uncertainty, and approach for model sensitivity evaluations
o I suggest adding a table of key sensitivity analyses, and assumptions, particularly those that
are expected to be important drivers of the results.
o Given the health effect evidence is from controlled human exposures, I believe impacts of
co-exposures can only be addressed as a source of uncertainty.
Comments on Chapter 4
Overall I thought the chapter was straightforward, easy to read, and covered the important issues. I have
a few suggestions for changes that I hope will improve the document and ultimately the REA.
•	Table 4-1, pdf p 64: I missed clearly understanding the number of monitors in a group and its
misalignment with the monitors shown in the sample maps. Consider whether updates to the text will
help document readers. If helpful, I suggest adding the number of source-oriented monitors in
parentheses to the number of monitors column.
•	Pdf p 70: I suggest (if feasible) the approach to filling in missing monitor data consider using local
data, e.g. from the nearest monitor of the same type. This will be less smooth than the approach
described.
•	Pdf p 70: The discussion of why the 95th percentile is lower once the missing data have been filled in
is reasonable. It would be worth having a bit more documentation of this assertion since the focus of
this work is on modeling exceedances.
•	Pdf p 74, 4-17: It will be interesting to learn what the preliminary model runs show about how to
best capture concentration gradients.
•	P 4-20, pdf p 77: I don't follow how Table 4-1 shows the ambient air concentration adjustments will
be small.
•	Figure 4-6: I suggest ensuring the axes are identical and adding 1:1 lines so the multiple plots on a
page can be interpreted more easily visually.
•	Pdf p 82: The balance of benefit with practical considerations seems appropriate.
•	Pdf p 83: Something is wrong with the wording of the last sentence in the first paragraph.
A-33

-------
Pdf p 86: It would be helpful to provide the reader with some context for the scale or range and
interpretation of typical values for A and D.
P 87 2nd paragraph: Is there something missing before "several"?
P 87: While I agree the person-occurrences statistic is less informative than the count, including it
would provide complementary information and should be considered for inclusion.
Pdf p 93: Given the sensitivity to the E-R function used, I agree with the plan to incorporate
sensitivity analyses to better understand the result of different E-R estimates.
P 98: While there is good experience with the models and results from previous REAs, would it be
worthwhile to list our expectations for the most important sources of variability or uncertainty? Also,
is it worth considering some sensitivity analyses to address important sources of variability and
uncertainty, particularly for quantities that are modeled based on assumptions?
A-34

-------
Dr. Frank Speizer
Chapter 2
This is an excellent summary of what was done for the REA 2009 and provides an excellent road map of
how to proceed with this current REA assessment. Both details of analytical procedures as well as
limitations are discussed. The main issue in redoing this will be the hopefully far increased data base on
5 minutes (and 1 hour) daily averages and max/day.
Chapter 3
I would concur with the staff summary that the issues of change as summarized in chapter justify a
redoing of the risk estimates. The two main issues are the far greater data base on 5 minute so2
measurements and thus the ability to quantitatively reduce uncertainty in the estimates at or around the
current standard (75ppb hourly). There will still be a range of values that will need to be studied. In
addition, I would hope to see in Chapter 4 a discussion of relooking at the form of the standard (5-min
vs. 1-hour) as well as the number of exceedances above 98%ile vs 99%ile.
Page 4-3, Use of APEX model seems justified
Page 4-5-6. The criteria for selection of potential sites seem reasonable down to 9 potential sites. Going
from 9 to 4 seems more arbitrary and raises some concern, particularly as 3 or the 4 sites are contiguous
with large bodies of water. The demography of the places selected also could influence results as the
asthma rates may be particularly high and the proportion of African American will also be high. Need to
discuss: why not use all 9 sites? It would seem to me that once the programs are written one need only
turn the crank. -I think I now understand better why only 4 sites are planned. However, I am still
concerned about the fact that 3 of 4 are near water and thought the suggestion at the meeting by Dr.
Sheppard was a good one.
Page 4-14, Table 4-2: Are not the Maximum values the same by design? It would not appear that a max
could be estimated from missing data. If so, slightly misleading to put in table.
Page 4.16, 4.17 Table 4.3-4.4: For those of us that are naive re the use of formula 4.3 please plan for
someone to demonstrate the calculation that produces the constant values in 4.3 for <2x only vs. in 4.4
for all four columns for C 1... C x.
Page 4.33 1st full paragraph. This paragraph discusses range of calculations as 100-400 ppb. Justification
for lower bound is that group analysis at 200ppb is not significant. However, it is known that there are a
small but significant number of individuals that do respond below 200 and thus under best case scenario
and with an adequate margin of safety should be protected. I suggest we need to quantify how large this
group is to justify not protecting them; and therefore recommend that the analysis be run to 50 ppb.
A-35

-------
Page 4.34 end of first paragraph Description of how individual data will be used to construct APEX
score raises concern about what will be categorized as responsive. For example, in the Linn 1987 paper
there are among the 40 asthmatics (mild=16, severe=24) several who responded to 0.2 with more than a
0.51 drop in FEV. (Although average drop was only about 150 cc.) See above comment as I think this
relates.
Section 4.2: Despite the numerous criticisms and comment above I found this whole section well written
and the logic appropriate. One concern is that as far as I can tell all of the human exposure data that will
be utilized is in adults with and without asthma. However, in section 4.2.4.2 mention is made of
calculations for children and it is not clear what data base is being used for that. I suspect it is
extrapolation of data from adults and if so should be stated.
Section 4.3 Discussion of Variability and Uncertainty. It is not as clear as it might be the relation
between these two concepts. That is, how does variability influence the characterization of uncertainty?
When the calculation crosses the null it DOES NOT mean that the presence of an effect is null. Rather
that, in most cases, there is insufficient data or too much variability for the amount of data there is to be
confident of the central tendency and thus the magnitude of uncertainty will be higher. This will need to
be expanded upon and discussed.
A-36

-------
Dr. James Ultman
Health Risk Assessment
Exposure analysis:
Benchmark levels used in the exposure analysis are apparently based on bronchial constriction
measurements made on adult asthmatics (pg 4-33, para 2). It is erroneous to think of children simply as
small adults. For several reasons, children are probably more vulnerable to SOxthan adults. Very
important are ontogenical factors such as maturity of the immune system and airway anatomy (e.g.,
Foos, et al., 2008, J. Tox. Environ. Health, 71; 149-165). Perhaps an additional lower benchmark level
should be added to the exposure analysis to deal with our uncertainty of a child's response to SO2.
Exposure-Response Relationship:
In developing an exposure-response relationship, bronchial constriction rather than FEVi decrement will
be used as the response variable (pg 4-35, para 2). To support this decision, it would be useful to provide
regressions, such as those in figure 4-7, comparing the two types of measurements. Which of the two
response measurement has lower variability? Which of the two tends to be a more conservative estimate
of a modification in lung function?
The exposure-response relationship is based on laboratory data taken on adults. It would be desirable to
include some discussion of how the relationship might deviate for children. There is a literature on
dosimetry modeling that could indicate how factors such as differences in airway anatomy and route of
breathing come into play (e.g., Ginsberg et. al., 2008, J. Tox. Environ. Health, Part A, 71:166-195).
To evaluate how uncertainties in the exposure-response relationship affect the health risk calculation,
particularly at exposures below 200 ppm where no data is available, the response of a single exposure
bin will be set to zero (pg 4-43, 2nd bullet item). Another approach would be to use the confidence
limits of the regression to examine upper and lower bounds of the errors in the health risk associated
with variability of the data.
A-37

-------
Dr. Ronald Wyzga
Overall, I believe that the plan describes a reasonable approach. Its implementation will involve many
specific decisions and details that will merit further examination and be reviewed by the Panel.
Specific comments:
SO2 emissions and levels have changed dramatically over the recent past. It is important that the REA
use the most recent estimates of SO2 emissions and exposures. For the former, see the comments of Dr.
Chow on the second draft of the ISA.
The combination of a model-based approach supplemented by ambient measurements appears to be the
only way to estimate the needed exposures. I would hope that some effort is expended to evaluate the
implemented approach by possibly applying the method without the use of data from one monitor in an
area where more than one set of monitoring data are available to see how well the chosen method
predicts monitored values.
p. 4-18: It is stated the "Model performance... can be evaluated." I believe it should be evaluated.
p. 4-22:1 believe that exposures for the Indoor MEs will be much lower than the outdoor MEs. For that
reason, I suggest that minimal effort be expended on the indoor MEs. I would support Staff efforts to
take shortcuts in addressing potential exposures for these MEs.
p. 4-42:1 believe that uncertainty analyses and sensitivity analyses should be undertaken to provide as
quantitative as possible estimates of uncertainty associated with the risk assessment outputs. In some
cases, a qualitative approach may be the only possible approach, but I urge that a quantitative approach
be applied to the fullest extent possible.
With respect to the choice of study areas, I believe the suggested approach is clearly articulated and is
reasonable. My only hesitation is that the study areas are is not geographically heterogeneous.
A-38

-------