PA454/R-94-035
&EPA
United States
Environmental Protection
Agency
Office of Air Quality
Planning and Standards
Research Triangle Park NC 27711
EPA-454/R-94-035
December 1994
Air
Clean Air Act ENVIRONMENTAL
Ozone Design Value Study: PS?N
Final Report DALLAS, TEXAS
A Report to Congress
-------
-------
-------
PA454/R-94-035
United States
Environmental Protection
Agency
Office of Air Quality
Planning and Standards
Research Triangle Park NC 27711
EPA-454/R-94-035
December 1994
Air
© EPA Clean Air Act
ENVIRONMENTAL
Ozone Design Value Study: AGENCYN
Final Report
DALLAS, TEXAS
A Report to Congress
-------
The Clean Air Act Ozone Design Value Study
Final Report
'•p
f\\
Office of Air Quality Planning and Standards
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
-------
DISCLAIMER
This report has been reviewed by the Office of Air Quality Planning and Standards, U.S.
Environmental Protection Agency, and has been approved for publication. Mention of trade
names or commercial products is not intended to constitute endorsement or recommendation for
use.
-------
Table of Contents
1 EXECUTIVE SUMMARY 1-1
INTRODUCTION 1-1
National Ambient Air Quality Standard for Ozone 1-2
Ozone Design Values 1-2
Regulatory History of Design Values 1-3
EPA Design Value Methodology 1-3
THE OZONE DESIGN VALUE STUDY 1-4
Spatial Representativeness 1-6
Temporal Representativeness 1-6
ALTERNATIVE OZONE DESIGN VALUE ESTIMATION METHODS 1-8
USE OF TIME-SERIES MODELS 1-10
PUBLIC OUTREACH EFFORTS 1-11
OTHER CONSIDERATIONS 1-12
Adjusting for Transported Ozone Levels 1-12
Adjusting for Meteorological Variability 1-13
PEER AND PUBLIC REVIEW 1-14
MAJOR FINDINGS 1-17
CONCLUSIONS 1-19
REFERENCES 1-21
2 INTRODUCTION 2-1
OZONE STANDARDS AND DESIGN VALUES 2-1
National Ambient Air Quality Standard for Ozone 2-1
Implementation of the Ozone NAAQS: Estimating the Expected
Exceedance Rate 2-2
Ozone Design Values 2-3
Estimating Design Values 2-4
THE OZONE DESIGN VALUE STUDY 2-7
Spatial Representativeness 2-9
Temporal Representativeness 2-10
Reasonable Indicator Criteria 2-10
Other Design Value Issues 2-10
STRUCTURE OF THE REPORT 2-11
REFERENCES 2-12
3 BACKGROUND 3-1
CHEMISTRY OF OZONE FORMATION 3-2
DISTRIBUTIONAL CHARACTERISTICS OF TROPOSPHERIC OZONE .... 3-6
Monitoring Ambient Ozone Concentrations 3-6
Spatial Distribution 3-6
Temporal Distribution 3-9
in
-------
Seasonal Patterns 3-9
Diurnal Patterns 3-17
Year-to-Year Variation 3-17
Other Technical Issues 3-20
SUMMARY 3-21
REFERENCES 3-21
4 OZONE DESIGN VALUE METHODOLOGIES 4-1
OZONE GUIDELINE DESIGN VALUE METHODS 4-1
EPA Table Look-up Method 4-1
Use of Statistical Distributions 4-2
Fitting Methods 4-4
Lognormal Distributions 4-5
Extreme Value Theory Approximations 4-6
Tail Exponential and Related Approaches 4-7
Conditional Probability Approach 4-8
OTHER DESIGN VALUE APPROACHES 4-9
California Method for Computing Recurrence Rate Values 4-9
EPA Tabular Method Extended to Multiple Years 4-10
Control Strategy Design Values 4-10
SUMMARY 4-11
REFERENCES 4-12
5 EVALUATION APPROACHES FOR OZONE DESIGN VALUE METHODS . . 5-1
INTRODUCTION 5-1
OZONE AIR QUALITY DATABASE 5-1
Spatial Distribution of Ozone Monitors 5-2
Trends in Ozone Monitoring Coverage 5-3
TIME-SERIES MODELING APPROACH 5-3
Time-Series Literature Review 5-3
Concentration Time-Series Models 5-6
Exceedance Time-Series Models 5-7
Computation of the Distribution of the K\h Highest Value 5-8
Time-Series Model Development 5-9
Basic Form of the Daily Maximum 1-hour Ozone Time-Series
Model 5-11
Model Evaluation 5-18
SUMMARY 5-18
REFERENCES 5-20
6 COMPARISONS AMONG ALTERNATIVE DESIGN VALUE METHODS 6-1
ALTERNATIVE DESIGN VALUE METHODS 6-1
Breiman Tail Exponential Method 6-1
California Air Resources Board Method 6-2
Distribution Fitting Method 6-2
Percentile Method 6-3
IV
-------
COMPARISONS USING AMBIENT OZONE DATA 6-3
Selected Metropolitan Area Results, 1989-91 6-3
Comparisons For All Ozone Sites, 1987-89 6-10
Comparisons Among Multi-year Tabular Method Design Value
Estimates 6-11
Multi-year Design Value Estimates in Selected Cities 6-17
Atlanta 6-18
Chicago 6-18
New York City 6-18
Summary of Findings from the Air Quality Design Value Comparisons . 6-21
TIME-SERIES MODELING SIMULATION RESULTS 6-22
Methodology 6-22
Comparison of Alternative Design Value Estimation Methods 6-24
Comparison of Biases of 3-Year Design Values 6-26
Comparison of Standard Deviations of 3-Year Design Values ... 6-28
Comparisons of Tail-Exponential Multi-Year Methods 6-30
Effect of Averaging Period on Alternative Design Value Methods 6-32
Surrogate Design Values 6-36
Summary of Conclusions from the Time-Series Modeling Study 6-38
REFERENCES 6-42
7 THE ROLE OF METEOROLOGY IN OZONE FORMATION 7-1
METEOROLOGICAL INFLUENCES ON OZONE CONCENTRATIONS .... 7-1
Insolation 7-2
Ventilation 7-2
Transport 7-4
Indirect Measures 7-4
ADJUSTING OZONE CONCENTRATIONS TO ACCOUNT FOR
METEOROLOGICAL INFLUENCES 7-7
Regression Methods 7-8
Classification Methods 7-10
Methods for Defining Categories 7-10
Adjustment Procedures Based on Meteorological Categories .... 7-12
Other Adjustment Methods 7-15
SUMMARY AND CONCLUSIONS 7-18
REFERENCES 7-28
8 ASSESSING THE IMPACT OF TRANSPORTED OZONE AND PRECURSORS . . 8-1
CHRONOLOGY OF A MULTI-REGIONAL OZONE EPISODE, JUNE 17-
20, 1987 8-1
ADJUSTING OZONE DESIGN VALUES FOR TRANSPORT 8-7
SUMMARY 8-11
REFERENCES 8-11
9 DETECTING TRENDS IN OZONE DESIGN VALUES 9-1
STATISTICAL APPROACHES TO TREND ANALYSES 9-2
-------
Linear Model Approaches 9-4
Nonparametric Methods 9-6
Extreme Value Theory Approaches 9-8
TIME-SERIES MODELS 9-10
OZONE TRENDS ADJUSTED FOR METEOROLOGY 9-11
SUMMARY AND CONCLUSIONS 9-11
REFERENCES 9-13
10 ALTERNATIVE AIR QUALITY INDICATORS 10-1
OZONE NAAQS RELATED INDICATORS 10-1
ALTERNATIVE AIR QUALITY INDICATORS 10-3
TEMPORAL VARIABILITY IN OZONE INDICATORS 10-10
SPATIAL VARIATIONS IN OZONE CONCENTRATIONS 10-16
Analysis of Monitoring Data 10-16
Spatial Indicators of Air Quality 10-17
SUMMARY 10-21
REFERENCES 10-24
11 PUBLIC OUTREACH EFFORTS 11-1
OVERVIEW OF PUBLIC MEETING COMMENTS 11-1
Increase Robustness of the Design Value 11-3
Adjust for Annual Meteorological Differences 11-3
Compliance Test for Attainment 11-4
Measurement Interference and Bias 11-4
Size and Location of Ozone Monitoring Network 11-5
Transport from Nearby Areas 11-5
Refine Meteorological Models 11-5
SUMMARY OF PUBLIC COMMENTS ON DRAFT REPORT 11-6
Distinction Between Statistical Population Parameters and
Estimated Parameters 11-7
Increase Stability of the Design Value Estimator 11-8
Attainment Test 11-10
Adjust Design Values for Annual Meteorological Fluctuations 11-11
Influence of Transport from Nearby Areas 11-12
Influence of Spatial Variability in Ozone Concentrations 11-13
Response to Previous Public Comments 11-13
SUMMARY OF PEER REVIEW COMMENTS ON DRAFT REPORT 11-13
REFERENCES 11-17
12 SUMMARY OF FINDINGS 12-1
MAJOR FINDINGS 12-2
CONCLUSIONS 12-5
REFERENCES 12-6
VI
-------
List of Tables
1-1. Ozone classifications specified in the 1990 Clean Air Act Amendments 1-1
1-2. Ozone design value rank based on number of years of data 1-4
1-3. Impact on ozone area classifications of varying the number of years when
estimating the ozone design value using the EPA tabular method 1-7
1-4. Average and maximum design values as estimated for the period 1989-91 by each
method 1-10
2-1. Ozone design value rank 2-5
2-2. Hypothetical example of ozone design value determination (case with two O3 sites
in an area, each year at least 75 percent complete) 2-6
2-3. Ozone classifications specified in the 1990 Clean Air Act Amendments 2-7
3-1. Ozone monitoring season by state 3-18
4-1. Ozone Design Value Rank Based on Number of Years of Data 4-2
5-1. Comparison of simulated and ambient ozone design values for sites in New York
and Houston, 1988 5-19
6-1. Average and maximum design values as estimated for the period 1989-91 by each
method 6-5
6-2. Comparison of Number of Areas with Multi-year Tabular Method Ozone Design
Values Equal to or Greater than 0.125 ppm 6-13
6-3. Ozone Nonattainment Area Classifications 6-15
6-4. Impact on Ozone Area Classifications of Varying the Number of Years when
Estimating the Ozone Design Value Using the EPA Tabular Method 6-16
6-5. Comparison of number of nonattainment areas in Clean Air Act classification
categories with number of areas in classification ranges based on 1989-91 ozone
monitoring data 6-17
6-6. Sites studied in time series comparisons of design values 6-23
6-7. Results from 1000 time series simulations of three-year design value estimates
from alternative design value estimation methods in five selected metropolitan
areas 6-25
6-8. Comparison of two methods of applying the Breiman tail-exponential method to
estimation of three-year design values (1000 three-year simulations) 6-31
6-9. Simulation comparison of the effect of averaging period on new York area site
340230006 design values estimated by various methods 6-33
6-10. Comparison of limiting values for alternative design value estimators from
alternative design value estimation methods in five selected metropolitan areas. . 6-35
6-11. The 95-percentile as a surrogate ozone design value (3000 one-year and 1000
three-year simulations) 6-37
6-12. Estimation of ozone design value from the 95th percentile surrogate 6-40
7-1. Meteorological variables potentially associated with ozone formation 7-3
8-1. Transport Adjusted Ozone Design Values, 1988-90 8-10
8-2. Comparison between Air Quality Design Value and Transport Adjusted Design
Value Derived Ozone Area Classifications, 1988-90 8-11
Vll
-------
10-1. Number of expected exceedances at design value sites within ozone area
classification categories for areas initially designated nonattainment under the
Clean Air Act Amendments of 1990 10-2
10-2. Average Number of Expected Exceedances at All Sites in Designated
Nonattainment Areas, 1987-89 10-2
10-3. Alternative Air Quality Indicator Values for 1987-89 10-5
10-4. Correlation Matrix for Alternative Indicators, 1987-89 10-7
10-5. Relative Ranks of Alternative Air Quality Indicators 10-8
10-6. Correlation Matrix for Spatial Indicators in Los Angeles 10-17
11-1. Matrix of Public Meeting Written Comments by Subject Area 11-2
viu
-------
List of Figures
1-1. Map depicting ozone design values at all sites in the northcentral states of
Indiana, Illinois, Michigan and Wisconsin, 1987-89 1-5
1-2. Actual and adjusted trends in number of days on which ozone concentrations
exceed 0.12 ppm in the Chicago area (adapted from Kolaz and Swinford,
1990) 1-15
1-3. Adjustment of ozone trend based on number of days above 90° F (Source: Jones,
1992) 1-16
1-4. Actual and meteorologically adjusted ozone trends in the 99th percentile of the
daily maximum 1-hour ozone concentration for Chicago, 1981 - 1990 (adapted
from Cox and Chu, 1992) 1-17
2-1. Initial ozone nonattainment area designations and classifications under the Clean
Act Amendments of 1990 2-8
3-1. Typical base ozone chemistry evolution profiles for NO, NO2, and ozone, and
ozone chemistry with reduced reactive organic (ROG) emissions 3-4
3-2. Ozone monitoring network in the continental United States, 1990 3-7
3-3. Second highest daily maximum 1-hour ozone concentration by year for all sites
in the Los Angeles-Anaheim-Riverside, CA CMSA 3-10
3-4. 95th percentile ozone concentration by year for all sites in the Los Angeles-
Anaheim-Riverside, CA CMSA 3-10
3-5. Second highest daily maximum 1-hour ozone concentration by year for all sites
in the Boston-Lawrence-Salem, MA-NH CMSA 3-11
3-6. 95th percentile ozone concentration by year for all sites in the Boston-Lawrence-
Salem, MA-NH CMSA 3-11
3-7. Second highest daily maximum 1-hour ozone concentration by year for all sites
in the Chicago-Gary-Lake County, IL-IN-WI CMSA 3-12
3-8. 95th percentile ozone concentration by year for all sites in the Chicago-Gary-Lake
County, IL-IN-WI CMSA 3-12
3-9. Second highest daily maximum 1-hour ozone concentration by year for all sites
in the Dallas-Ft. Worth, TX CMSA 3-13
3-10. 95th percentile ozone concentration by year for all sites in the Dallas-Ft. Worth,
TX CMSA 3-13
3-11. Second highest daily maximum 1-hour ozone concentration by year for all sites
in the Houston-Galveston-Brazoria, TX CMSA 3-14
3-12. 95th percentile ozone concentration by year for all sites in the Houston-Galveston-
Brazoria, TX CMSA 3-14
3-13. Second highest daily maximum 1-hour ozone concentration by year for all sites
in the New York-Northern New Jersey-Long Island, NY-NJ-CT CMSA 3-15
3-14. 95th percentile ozone concentration by year for all sites in the New York-
Northern New Jersey-Long Island, NY-NJ-CT CMSA 3-15
3-15. Map depicting ozone design values at all sites in the northcentral states of
Indiana, Illinois, Michigan and Wisconsin, 1987-89 3-16
IX
-------
3-16. Map depicting ozone design values at all sites in the northeastern states of
Connecticut, Massachusetts, New York and Rhode Island, 1987-89 3-16
5-1. Map depicting location of the ozone monitoring sites that reported hourly ozone
concentration data for at least one year during the period 1980 - 1990 5-4
5-2. Map depicting location of the 323 ozone monitoring sites that monitored each
year during the period 1980 - 1990 5-4
5-3. Number of areas (CMSAs/MSAs/counties) reporting ozone data to AIRS by
year 5-5
5-4. Example of variability of ozone daily maximum 1-hour concentrations 5-10
5-5. Log transform of daily maximum 1-hour concentrations at sample site 5-13
5-6. LOWESS smoothing of the log of the daily maximum 1-hour concentrations at
the sample site 5-14
5-7. LOWESS fits to ten years of ozone data at the sample site 5-15
5-8. Probability plot of ARIMA residuals at a site in the New York metropolitan
area 5-16
5-9. Probability plot of ARIMA residuals at a site in the Houston metropolitan
area 5-17
6-1. Average and maximum design values for monitors in the Chicago metropolitan
area for alternative design value estimation methods 6-6
6-2. Average and maximum design values for monitors in the New York metropolitan
area for alternative design value estimation methods 6-7
6-3. Average and maximum design values for monitors in the Los Angeles-South
Coast Air Basin area for alternative design value estimation methods 6-8
6-4. Comparison of design values estimated using the California Air Resources Board
(CARB - Larsen) method with the fourth highest daily maximum 1-hour
concentration at all sites in the Chicago and New York metropolitan areas, 1989-
91 6-9
6-5. 5th and 95th percentile differences for comparisons among alternative ozone
design value estimation procedures 6-10
6-6. 5th and 95th percentile differences among multi-year ozone design value
estimates 6-12
6-7. Number of areas with multi-year tabular method ozone design values greater than
or equal to 0.125 ppm 6-14
6-8. Trends in multi-year table look-up ozone design values in Atlanta, GA, 1980 -
1990 6-19
6-9. Trends in multi-year table look-up ozone design values in Chicago, 1980 -
1990 6-19
6-10. Trends in multi-year ozone design values in New York for all sites and sites with
eleven years of data 6-20
6-11. Design value bias versus characteristic largest value 6-27
6-12. Standard deviation of design value affected by site and design value magnitude
and method of computation 6-29
6-13. Effect of averaging time on design value estimated by three different methods at
four locations 6-34
-------
6-14. Limiting design value (CLV) versus average 3-year 95th percentile 6-39
7-1. Annual number of days on which the average daily maximum ozone concentration
in the Philadelphia Ozone Network exceeds 0.085 ppb based yearly data run
down the CART tree grown on 1979-1988 data 7-13
7-2. Actual and adjusted trends in number of days on which ozone concentrations
exceed 0.12 ppm in the Chicago area 7-16
7-3. Actual and meteorologically adjusted ozone trends in Atlanta, Baltimore,
Bakersfield and Baton Rouge, 1981-1991 (adapted from Cox and Chu, 1991). . 7-17
7-4. Actual and meteorologically adjusted ozone trends in Beaumont, Birmingham,
Boston and Bridgeport, 1981-1991 (adapted from Cox and Chu, 1991) 7-19
7-5. Actual and meteorologically adjusted ozone trends in Charlotte, Chicago,
Cincinnati and Cleveland, 1981-1991 (adapted from Cox and Chu, 1991) 7-20
7-6. Actual and meteorologically adjusted ozone trends in Columbia, Dallas, Denver
and Detroit, 1981-1991 (adapted from Cox and Chu, 1991) 7-21
7-7. Actual and meteorologically adjusted ozone trends in El Paso, Fresno, Hartford
and Houston, 1981-1991 (adapted from Cox and Chu, 1991) 7-22
7-8. Actual and meteorologically adjusted ozone trends in Los Angeles, Louisville,
Miami and Milwaukee, 1981-1991 (adapted from Cox and Chu, 1991) 7-23
7-9. Actual and meteorologically adjusted ozone trends in Los Angeles, Louisville,
Miami, and Milwaukee, 1981-1991 (adapted from Cox and Chu, 1991) 7-24
7-10. Actual and meteorologically adjusted ozone trends in Muskegon, New York,
Philadelphia, and Phoenix, 1981-1991 (adapted from Cox and Chu, 1991). . . . 7-25
7-11. Actual and meteorologically adjusted ozone trends in Tampa, Tulsa, and
Washington, D.C., 1981-1991 (adapted from Cox and Chu, 1991) 7-26
8-1. Isopleths of ozone daily maximum 1-hour concentrations for June 17, 1987. ... 8-3
8-2. Isopleths of ozone daily maximum 1-hour concentrations for June 18, 1987. . . . 8-4
8-3. Isopleths of ozone daily maximum 1-hour concentrations for June 19, 1987. ... 8-5
8-4. Isopleths of ozone daily maximum 1-hour concentrations for June 20, 1987. . . . 8-6
8-5. Midwest region on July 17, 1987 8-8
8-6. Northeast region on July 7, 1988 8-8
8-7. Northeast region at 1400 EST on July 8, 1988 8-8
8-8. Northeast region at 2300 EST on July 8, 1988 8-8
9-1. Sample illustration of the use of confidence intervals to determine statistically
significant changes 9-3
9-2. Comparison of meteorologically adjusted, and unadjusted, trends in the composite
average of the second highest maximum 1-hour concentration for 43 MSAs, 1983-
1992 9-12
10-1. Average number of expected exceedances versus the EPA DV for all sites,
1987-89 10-4
10-2. Year to year variability in selected ozone air quality indicators 10-10
10-3. Temporal variability hi ozone indicators hi Los Angeles and Houston 10-11
10-4. Temporal variability in ozone indicators in Chicago and Milwaukee 10-12
10-5. Temporal variability in ozone indicators in New York and Baltimore 10-13
10-6. Temporal variability in ozone indicators in Philadelphia and Atlanta 10-14
XI
-------
10-7. Temporal variability in ozone indicators in Boston and Hartford 10-15
10-8. Number of days on which the federal ozone standard was exceeded in 1989. . 10-18
10-9. Design values (fourth highest value in each three year period) for selected
stations 10-19
10-10. Comparison of annual ozone indicators 10-20
10-11. Comparison of indicators for overlapping three year periods 10-22
10-12. Comparison of indicators for overlapping three year periods normalized to 1980
values 10-23
xn
-------
1 EXECUTIVE SUMMARY
INTRODUCTION
This report responds to the requirements of Section 183(g) of the Clean Air Act
Amendments (CAAA) of 1990, which requires that
"The Administrator shall conduct a study of whether the methodology in use by the
Environmental Protection Agency as of the date of enactment of the Clean Air Act
Amendments of 1990 for establishing a design value for ozone provides a reasonable
indicator of the ozone air quality of ozone nonattainment areas. The Administrator
shall obtain input from States, local subdivisions thereof, and others. The study shall
be completed and a report submitted to Congress not later than 3 years after the date
of the enactment of the Clean Air Act Amendments of 1990. The results of the study
shall be subject to peer and public review before submitting it to Congress." (PL
101-549, Sec. 183 (g))
Ground-level ozone, the primary constituent of smog, causes several adverse health
and environmental effects, such as respiratory problems, crop loss and materials damage.
EPA has established a national ambient air quality standard (NAAQS) for ground-level
ozone. According to EPA regulations, an area is not meeting the ozone standard
("nonattainment") if the expected number of days per year with daily maximum 1-hour
concentrations greater 0.12 ppm is greater than 1. As of October 1994, there are 91 areas of
the country that are designated as nonattainment areas for ozone.
The ozone design value is a surrogate measure of attainment status, a measure of
progress, and an indicator of how much concentrations must be reduced to meet the standard.
The EPA design value method yields an estimate for the ozone design value that is consistent
with the current ozone NAAQS. The current EPA design value method is simply to select
the fourth highest daily maximum 1-hour concentration as the design value during the 3-year
compliance period (Laxton, 1990). The fourth highest value is the design value, since if the
fourth highest day is reduced to the level of the standard, then there will be one day per year
above the level of the standard assuming three years of data.
With passage of the Clean Air Act Amendments (CAAA) of 1990, added emphasis
was placed on ozone design values. In addition to designating areas as nonattainment for
ozone, the CAAA introduced a classification process to further categorize nonattainment
areas according to the extent of their ozone problem. As shown hi Table 1-1, this area
classification was based upon the ozone design value. The CAAA stated that the design
value "shall be calculated according to the interpretation methodology issued by the
Administrator most recently before the date of the enactment." Before the 1990 CAAA,
designation of nonattainment areas simply involved a yes/no determination as to whether the
area met the standard. The additional classification step introduced by the 1990 CAAA
1-1
-------
placed greater emphasis on ozone concentration observations and on the methodology used to
determine the design value.
TABLE 1-1. Ozone classifications specified in the 1990 Clean
Air Act Amendments.
Area Class
Marginal
Moderate
Serious
Severe
Extreme
Design Value*
0.121 up to 0.138
0.138 up to 0.160
0.160 up to 0.180
0.180 up to 0.280
0.280 and above
Attainment Date**
3 years
6 years
9 years
15 years
20 years
*The design value is measured in parts per million (ppm).
**The primary standard attainment date is measured from the date of
the enactment of the Clean Air Act Amendments of 1990.
Another reference to the use of design values is contained in Section 181(b)(2) of the
Act, which states that EPA "shall determine, based on the area's design value (as of the
attainment date), whether the area attained the standard by that date." EPA's preliminary
interpretation of this Section is that the "average number of exceedances per year shall be
used to determine whether the area has attained" which is the attainment test for the ozone
National Ambient Air Quality Standard (NAAQS) (Federal Register, 1992).
National Ambient Air Quality Standard for Ozone
In 1979, EPA promulgated the ozone NAAQS at a level of 0.12 ppm that is attained
"when the expected number of days per calendar year with maximum hourly average
concentrations above 0.12 part per million (235 ptg/m3) is equal to or less than 1 as
determined by Appendix H" (40CFR 50.9). The attainment test specified in Appendix H
states that the "expected number" of days with concentrations above 0.12 ppm ("exceedance"
days) is to be estimated by calculating the average number of exceedances during the most
recent three years. Additional information is contained in Appendix H and the EPA Ozone
Guideline on procedures for dealing with missing data (EPA, 1979). The Guideline makes it
clear that the expected exceedance criterion is to be applied independently to each monitoring
site. For areas with multiple monitoring sites, all sites within the nonattainment area must
meet the standard for the area to be designated in attainment of the ozone NAAQS.
1-2
-------
Ozone Design Values
As noted above, compliance with the ozone NAAQS is judged on the basis of
expected exceedances, and becomes a "yes/no" decision. However, once it is established
that an area exceeds the standard, the next logical question to ask is, "By how much?" The
air quality design value is intended to provide a measure of how far concentrations must be
reduced to achieve attainment or, equivalently, how far out of attainment the area represented
by a monitoring site is. In this respect, the design value can be viewed as an air quality
indicator for a given location.
Given the expected exceedance form of the ozone NAAQS, the design value for this
standard is defined in the EPA guideline document as "the concentration with expected
number of exceedances equal to one" (EPA, 1979). Note that in this context the ozone
guidelines are referring to the unknown "true" number of expected exceedances per year
rather than the estimate of expected exceedances determined using the Appendix H
calculations. In statistical terms, this is the value which is exceeded once per year on
average. If the daily maximum ozone concentrations are assumed to be independent and
have the same distributions every day throughout the year, then the design value is the
characteristic largest value (CLV) of that distribution. The Ozone Guideline described
several different options for estimating design values, including a table look-up approach,
graphical procedures, and fitting statistical distribution. The current EPA design value
method is simply to select the fourth highest daily maximum 1-hour concentration as the
design value during the 3-year compliance period (Laxton, 1990). The fourth highest value
is the design value, since if the fourth highest day is reduced to the level of the standard,
then there will be one day per year above the level of the standard assuming three years of
data.
Strictly speaking, the design value is an unknown quantity depending on the
underlying distribution of ozone concentrations, and the EPA design value and alternatives
are estimators of the (true) design value. To retain the readability of this document the term
"design value" may refer to either the unknown population value or an estimator, depending
on the context in which it is used. Where appropriate, the term "true design value" is used
for clarification. The "EPA design value" always refers to the table look-up value.
Regulatory History of Design Values
Beginning hi the 1970's, air quality design values were used as the primary input to
simple air quality models, such as the "rollback model" and the Empirical Kinetic Modeling
Approach (EKMA) (deNevers and Morris, 1975; Meyer et al., 1977). These models were
used to estimate emission reductions needed to attain the NAAQS and to evaluate alternative
control strategy options (deNevers and Morris, 1975; Meyer et al., 1977; Wilson and
Scruggs, 1980). This use of ozone design values diminished following the development of
more complex photochemical modeling approaches (EPA, 1981; EPA, 1991b). In some
applications, design values used in estimating emissions reductions have been adjusted to
1-3
-------
account for factors such as the level of transported ozone, or air quality reductions expected
from future control measures (Meyer, Gipson, and Freas, 1977; Wilson and Scruggs, 1980).
Such adjusted design values have come to be known as "control strategy values" to
differentiate them from "air quality" design values, which are estimated directly from the
ambient monitoring data (Rhoads and Tyler, 1987).
To support the passage of the Clean Air Act Amendments, EPA began issuing annual
lists of areas failing to meet ozone and carbon monoxide NAAQS which contained their
corresponding air quality design values (EPA, 1987, 1988, 1989, 1990, 1991a, 1992b). It is
clear from the language of the CAAA of 1990, and the legislative history of the Act, that the
initial area classifications were to be based on the ah- quality design value, which is the
primary focus of this study (EPA, 1993). However, issues such as adjusting ozone design
value for transport, emissions trends, and meteorological variability are addressed in this
study within the context of "control strategy values."
EPA Design Value Methodology
The design value associated with the ozone NAAQS is an abstract quantity that can
only be estimated from available data. The Ozone Guideline suggests several methods for
estimating the design value, including a simplified table look-up procedure, approaches using
statistical distributions, and techniques based on conditional probabilities. No single
approach was required by the Guideline.
The table look-up procedure, summarized below in Table 1-2, has been designated as
the EPA design value estimation method (Laxton, 1990). Basically, the tabular method
identifies the lowest observed concentration that was not exceeded more than an average of
once per year during the measurement period. This methodology is essentially unchanged
from the State Implementation Plan (SIP) guidance issued in 1981, and is the method that
was used for all of the annual design value lists issued by EPA and the initial ozone area
classifications (EPA, 1981, 1987, 1988, 1989, 1990, 1991a,b, 1992, 1993; 40CFR58).
Using the tabular method focuses attention on a concentration that was actually observed, as
compared to a statistical fitting technique that could yield a design value that does not
correspond to a concentration observed on a particular day. The tabular approach has
several additional advantages not always shared by more complex statistical procedures.
First, estimates can be made quickly, and directly, from existing summaries of air quality
data. Second, the design value estimates are reproducible and verifiable with actual
monitoring data. Third, it provides a uniform approach for all areas. It is also worth noting
that current monitoring regulations do not require the reporting of hourly ozone data for all
sites across the nation (Federal Register, 1991). Thus, statistical approaches which require
fitting distributions to all the data, or even the upper 10 percent of the distribution, are not
applicable for sites that only report summary statistics and not the individual hourly
concentrations, or daily maximum 1-hour values.
1-4
-------
TABLE 1-2. Ozone design value rank based on number of years of data.
Number of Valid Years Ozone Design Value Rank
(at least 75% of days during (daily maximum 1-hr concentration)
designated ozone season)
less than one valid year highest daily maximum
1 year of data 2nd highest daily maximum
2 years of data 3rd highest daily maximum
3 years of data 4th highest daily maximum
THE OZONE DESIGN VALUE STUDY
Section 183(g) of the Act directs EPA to conduct a study of the methodology
currently in use for calculating design values to determine if the calculated design value
"provides a reasonable indicator of the ozone air quality of ozone nonattainment areas."
Thus, the focus of the study is on the design value methodology as initially developed in the
Ozone Guideline and later defined in current EPA guidance (40CFR50.9; EPA, 1979;
Laxton, 1990). Issues concerning the form of the current ozone NAAQS are more properly
treated within the existing mechanism for NAAQS review. EPA is in the midst of reviewing
the ozone NAAQS. The Agency intends to propose any change to the standard by Spring
1995 and, after taking public comment, will promulgate the final decision in Spring 1997.
The "reasonable indicator" evaluation is dependent on the intended application of the
design value. It is quite possible that a design value estimation procedure that provides a
reasonable indicator for the purpose of determining the nonattainment classification of a small
geographic area surrounding a monitoring site may not be suitable for the purpose of
estimating the required degree of emission reduction needed to achieve attainment or for the
purpose of estimating health risks to nearby populations. Therefore, it is necessary to
indicate the intended application of a design value estimation procedure before judging
whether it yields a reasonable air quality indicator. This issue can be examined in both a
temporal and a spatial framework.
Spatial Representativeness
Design values are estimated individually for each monitor in an area, and the
maximum value is used to determine the nonattainment classification of the entire area.
Ozone concentrations can also be locally depressed immediately downwind of a source of
nitrogen oxides (NOJ due to scavenging by nitric oxide (NO). Key concerns are (1) whether
the monitoring network is sufficiently dense and monitors are appropriately located to
represent air quality over the area in question and (2) how spatially uniform are design value
1-5
-------
estimates across metropolitan areas. The study examined the spatial distributions of ozone
ambient concentrations from existing monitoring networks within urban areas. Figure 1-1
shows the variability in 1987-89 ozone design values calculated for all monitoring sites in the
northcentral states of Indiana, Illinois, Michigan, and Wisconsin. As illustrated, design
values can vary from levels near the standard to levels near 0.20 ppm at sites across these
states. The study also described regional-scale ozone episodes and examined large-scale
features using spatial concentration distributions calculated by photochemical dispersion
models.
Temporal Representativeness
A major concern with respect to the temporal representativeness of design values is
the number of years of data used to calculate the design value. Current EPA guidance calls
for the use of three years of data, if available (40CFR50.9). The use of three years of data
is a compromise between the need to include as much data as possible to arrive at an
accurate estimate and the need to recognize nonstationarities in the data record resulting from
precursor emission trends. Setting the time period for judging compliance also sets an upper
limit on the number of exceedances that a site can experience in any one year and the area
still remain in attainment of the NAAQS.
Wide year-to-year variations in weather conditions can result in significant differences
in estimated design values from one three-year period to the next, even in the absence of
emission changes, as shown in recent design value lists that include 1988 ozone data.
Meteorological conditions in 1988 were highly conducive to ozone formation, especially in
the eastern half of the nation. Summer 1988 was the third hottest summer on record (Heim,
1988). Adding 1988 data to the three-year data window increased the number of areas not
meeting the ozone standard to 98, an increase of 37 areas (EPA, 1989, 1990). More recent
summers have been less conducive to ozone formation than the summer of 1988. In the
East, the period from January through July 1989 was among the wettest on record (Heim,
1988). In the Northeast, summer 1990 also had above-average precipitation (Heim, 1990).
However, summer 1991, which was the eighth warmest summer on record, saw the return of
ozone conducive conditions, especially in the Northeast (Heim, 1989). In addition to these
meteorological differences, volatile organic compound emissions have been reduced since
1988 levels. The volatility of gasoline, measured as Reid Vapor Pressure (RVP), was
reduced 11 percent between 1988 and 1989, and an additional 3 percent between 1989 and
1990 (Federal Register, 1989; MVMA, 1988a, 1988b, 1988c). As a result of both changing
meteorological conditions and emissions reductions, the latest design value listing, based on
1991-93 data, showed that 55 of the initial 98 nonattainment areas now meeting the ozone
NAAQS (EPA, 1994). This is the fourth update that does not include data from the 1988
peak ozone year. Seven of the original 98 nonattainment areas have already been
redesignated to attainment. Thus, factors such as the sequence of meteorological conditions,
and reductions in emissions can introduce temporal variability hi design values.
1-6
-------
OZONE
Design Value, 1987-89
Figure 1-1. Map depicting ozone design values at all sites in Indiana, Illinois, Michigan,
and Wisconsin, 1987-89.
1-7
-------
As another measure of temporal variability, Table 1-3 summarizes ozone area
classifications that would result from increasing the number of years used in the EPA design
value method. The table focuses on data windows ending hi 1989, because most of the area
classifications were based on data from that period (40CFR58). However, to maintain
historical consistency, Consolidated Metropolitan Statistical Areas (CMSAs), Metropolitan
Statistical Areas (MSAs), and counties are used to define the geographic area, and not the
nonattainment area boundaries of the currently designated nonattainment areas. Table 1-3
shows that the largest differences in ozone area classifications are associated with design
values based on a single year of data. There is close agreement between classifications based
on 3 and 4 years of data, while the longer data windows (5 and 6 years) have fewer
nonattainment areas (4 and 7 fewer, respectively) than the 3-year estimates. There is some
downward movement hi area classifications evident hi the longer tune periods. That is,
severe areas have moved downward to serious, serious areas to moderate, and moderate
areas to marginal.
TABLE 1-3. Impact on ozone area classifications of varying the number of years when
estimating the ozone design value using the EPA tabular method.
Number of Areas (CMSA/MSA/County)
Clean Air Ac
r Single ^
fear Desig
n Value
Ozone Classification
Extreme
Severe
Serious
Moderate
Marginal
Total
1987
1
6
17
22
22
68
1988
1
11
23
45
32
112
1989
Multi-year Design Value
1988-
1989
1 1
4 7
4 20
12 36
18 35
39 99
1987-
1989
1
9
16
33
39
98
1986-
1989
1
9
15
33
37
95
1985-
1989
1
7
14
33
39
94
1984-
1989
1
6
15
29
40
91
ALTERNATIVE OZONE DESIGN VALUE ESTIMATION METHODS
The Ozone Guideline introduced several techniques that could be used to estimate
design values including (1) a table look-up procedure (which evolved into the current EPA
method), (2) the use of fitted statistical distributions, and (3) the use of a conditional
probability approach.
1-8
-------
There are two distinct approaches to fitting distributions to air quality data:
(1) fitting parametric distributions to raw hourly or daily concentrations, and (2) fitting
extreme value distributions to the highest concentrations. A review of the literature
conducted for this study, including the reported goodness-of-fit results, suggests that a
growing consensus appears to favor the use of the tail exponential distribution to fit the
annual maximum hourly ozone concentrations. The use of the lognormal distribution to fit
the hourly and, possibly, daily maximum hourly, ozone concentrations was another popular
method (Mage, 1984; Curran and Frank, 1975). The selection of a "best" statistical
distribution for calculation of ozone design values may not be possible because such a
distribution probably varies according to the location studied and the time period of interest.
An approach developed by Breiman for EPA fits an exponential distribution to the upper 5 to
10 percent of the concentration distributions for each year (Breiman et al., 1978). Ozone
design values are estimated by combining the tail-exponential distributions for the three-year
compliance period. Another approach is to combine data years and fit a parametric
distribution to the upper tail of the three-year distribution. The tail exponential approach
developed by Larsen and others at the California Air Resources Board (CARB, 1992b) was
developed in response to the 1988 California Clean Air Act which allows highly irregular or
infrequent violations of the state ambient air quality standards to be excluded from the
attainment/nonattainment designation process (Larsen and Bradley, 1991; CARB, 1992a;
Larsen, 1991). In June 1990, the California Air Resources Board (CARB) determined that
exceedances expected to recur less frequently than once in seven years could be excluded.
The tail exponential approach was proposed as a method of estimating the one-in-seven-year
concentration. More recently, the CARB revised the exclusion frequency to be one in one
year (CARB, 1992b).
Comparisons have been made of design values estimated using the EPA tabular
approach and those estimated using exponential and Weibull distributions. These
distributions were fitted to the upper 5 percent of the three-year distribution for 1987-89 at
all ozone sites in the historical database using the Breiman tail-exponential procedure and by
a procedure that estimated the parameters by maximum likelihood using a Newton-Raphson
algorithm (Breiman et al., 1978; SAS; Freas, 1992a). The tabular method design value
estimates tended to be lower than those obtained with the tail-exponential and other
distribution fitting methods. Differences in the several parts per billion range were found
among the various distribution fitting methods. Table 1-4 presents the results for the
Chicago, New York and Los Angeles metropolitan areas of using five different methods for
estimating design values: (1) the EPA tabular approach, (2) simply using the appropriate
percentile from the empirical distribution, (3) the Breiman tail-exponential fitted to the upper
5 percent of the data, (4) a 10 percent tail-exponential fit, and (5) the CARB tail-exponential
approach developed by Larsen. The CARB approach was applied both with and without the
empirical calibration factor used with the method. The calibration factor was determined by
Larsen hi a way that recognized the expected discrepancy between the tail-exponential
method and the EPA tabular method, since the EPA method is expected to be biased low on
theoretical grounds. According to the CARB (McGuire, 1994), the calibration factor
estimate recommended by Larsen is based on ozone data from monitoring sites throughout
1-9
-------
California and was selected to produce design value estimates at a "suitable midpoint"
between the uncalibrated method and the EPA method.
Average and maximum values across all monitoring sites in these areas are listed in
Table 1-4. Details of the calculations, including the percentile definitions and confidence
intervals, are provided in Section 6. The maximum values indicate the design value which
would be assigned to the nonattainment area, assuming that design values from any monitors
not included in this analysis are smaller. Except for design values based on the 10 percent
tail exponential (and the 5 percent tail exponential in Los Angeles), the various methods
produced estimated average design values that are within 0.01 ppm of each other. This is
also true of the maximum design values in each area, except hi Los Angeles where the
differences were as large as 0.02 ppm. Of course, results at individual monitors may show
wider variations. Design values obtained by fitting a tail exponential distribution to the top
10 percent of the data values are higher than even the third highest concentration in each
area, both on average and for the maximum values. Lower design values were obtained
from tail exponentials fitted to the top 5 percent of the ozone values although, on average,
they are still higher than the third highest for the New York and Los Angeles areas. These
results indicate that the portion of the distribution of daily maximum concentrations to which
the tail exponential is fitted can have a significant impact on the estimated design value.
This is the primary motivation for using Larsen's approach, which uses multiple tails fitted to
various portions of the upper end of the distribution and weights the results toward those tails
which best fit the available data.
If design values are to be estimated by fitting distributions, the tail-exponential
distribution approach, using either the Breiman formulation or the CARB method, seems
preferable on the basis of its simplicity, ease of fitting, robustness and goodness of fit. The
goodness of fit for a large number of sites is likely due to the property that a wide variety of
daily maximum ozone concentration distributions have an approximately exponential tail.
1-10
-------
TABLE 1-4. Average and maximum estimated (ppm) design values for the period 1989-91 using
each estimation method (4th High = fourth highest concentration, 3d High = third
highest concentration, Pcntl = percentile method, 5%TlExp = 5 percent tail
exponential, 10%TlExp = 10 percent tail exponential, Larsen = CARS method,
LarNoCal = CARS method without calibration factor).
5% TIExp
95% Confidence
Interval
10% TIExp
95% Confidence
Interval
4th High 3rd High Pcntl
Lower Upper Lower Upper
5%TIExp Bound Bound 10%TlExp Bound Bound Larsen LarNoCal
Chicago Area
Average:
Max:
New York
Area
Average:
Max:
Los Angeles
Area
Average:
Max:
0.118
0.151
0.140
0.165
0.21
0.28
0.122
0.164
0.146
0.175
0.22
0.30
0.118
0.152
0.140
0.166
0.22
0.30
0.123
0.161
0.150
0.175
0.25
0.32
0.113
0.146
0.138
0.159
0.22
0.29
0.137
0.183
0.168
0.199
0.30
0.38
0.126
0.168
0.159
0.185
0.26
0.34
0.116
0.153
0.146
0.167
0.23
0.31
0.138
0.187
0.177
0.209
0.30
0.39
0.113
0.151
0.144
0.166
0.21
0.28
0.118
0.158
0.151
0.175
0.22
0.30
USE OF TIME-SERIES MODELS
In this study, time-series models were used as a tool for evaluating alternative design
value estimation methodologies. The ambient ozone database was used to develop a time-
series model of the behavior of daily maximum ozone concentrations. Given such a model,
large numbers of random simulations of single seasons of daily maximum ozone values can
be generated that allow the limiting CLV (the "true design value") and design values for any
number of methods to be calculated over a large number of years. Thus, both the inherent
biases and precision of alternative design value methods can be studied using a wide variety
of averaging years. These data sets have no missing values and therefore are free from this
source of error.
The time-series model has been applied hi five geographically diverse metropolitan
areas: Atlanta, GA; Charlotte, NC; Chicago, IL-WI; Houston, TX; and New York, NY-NJ-
CT. One hundred three-year sequences of ozone daily maximum concentrations were
generated for key sites in each area. The results are similar to those observed with the
ambient data comparisons. That is, the EPA tabular method gave lower design value
estimates, on average, than Breiman's tail-exponential method. However, tail exponential
estimates at some individual sites can be lower than the EPA tabular values, depending on
the shape of the concentration distribution.
1-11
-------
PEER AND PUBLIC REVIEW
Section 183(g) of the CAAA of 1990 requires the EPA to "obtain input from States,
local subdivisions thereof, and others." In conducting the Ozone Design Value Study, EPA
has made every effort to have an open process and to ensure full public input and
participation. These efforts focused on information exchange through participation in
professional meetings and conferences, involvement of interested parties on the study
working group, and holding a public meeting (Freas, 1992a, 1992b; Curran, 1992a, 1992b;
57FR34133). The study plan and the results of the multi-year analyses were presented in
technical papers at the Air and Waste Management Association's Tropospheric Ozone
Specialty Conferences. These papers were peer reviewed prior to publication in the
Conference Proceedings.
A study review group has been established to provide input on technical issues and
policy concerns. The group is composed of representatives from EPA program, research,
policy and legal offices. State and local air pollution control agency officials also serve on
the review group.
On September 10, 1992, EPA held a public meeting in Arlington, VA to obtain input
on technical considerations and on implementation and policy issues to be addressed within
the context of the Ozone Design Value Study. The meeting announcement was published in
the Federal Register, and to ensure that all interested parties were aware of the public
meeting, copies of the meeting announcement were sent to both individuals and organizations
that had previously expressed interest in ozone-related issues (57FR34133). At the public
meeting, presentations were made on behalf of the Motor Vehicle Manufacturers Association
and Ford Motor Company. Written comments were received from ten respondents,
representing State and local air pollution agencies, industry and private individual views.
On March 14, 1994, EPA published a Federal Register Notice announcing the
availability of a draft report on the study for public review and comment. Prior to that
announcement, copies of the draft report were mailed to all parties that previously expressed
an interest in the study. More than 250 copies of the report were mailed out in response to
requests. As of the close of the public comment period on April 14, 1994, comments had
been received from only two respondents. Requests were received from several parties to
extend the comment period. On April 28, 1994, a second Federal Notice was published that
extended the public comment period until May 31, 1994. Although many additional requests
for copies of the draft report were answered during this period, only eight additional parties
submitted comments by the close of the comment period. Technical peer review was
conducted under the auspices of the National Institute of Statistical Sciences. The report
responds to the public comments and was revised to address the technical corrections
identified during peer review.
Comments received during the public meeting and on the draft report can be grouped
into two major categories: (1) those relating directly to design value issues and (2) those that
1-12
-------
would require changes in legislation or a revision to the form of the ozone standard.
Those in the first category include issues concerning (1) the statistical robustness of the
current design value methodology, (2) the precision and accuracy of ozone monitoring data,
and (3) the use of other statistical techniques, such as fitting a tail-exponential model, for
determining the design value. The second category includes issues associated with changing
the form of the ozone standard to a more robust air quality indicator, or proposing to modify
the attainment test to incorporate a statistical test, such as a "t-test" for judging compliance
with the standard (Heuss, 1992; Chock, 1989, 1992; Heuss and Chock, 1992). Such changes
are beyond the scope of this study and are more properly addressed during the next ozone
NAAQS review.
OTHER CONSIDERATIONS
Many comments received raise issues related to the concept of a "control strategy"
design value, not the air quality design value. Adjusting design values for factors such as
transported ozone, meteorology, and emissions trends falls within the control strategy design
value concept, not the air quality design value methodology used to classify ozone
nonattainment areas under the CAAA of 1990.
Adjusting for Transported Ozone Levels
Transport of ozone and ozone precursors generated in one air basin can significantly
influence ozone concentrations in neighboring air basins located considerable distances
downwind. Some comments received on the original nonattainment area classifications
argued that EPA should have considered lowering the classification because of the impact of
transport from upwind areas (Federal Register, 1991; EPA, 1992a). The amendments
specifically acknowledge that transport across state boundaries plays a major role during high
ozone events in the northeastern urban corridor between Washington, D.C. and Boston.
Transport of ozone and precursors also plays a significant role in other parts of the country,
including the Gulf Coast region and Lake Michigan. Although the amendments call for the
establishment of a transport commission to study this issue, the Act does not provide for
adjusting the air quality design values for transport. The one instance that transport may be
considered during the initial classification process is if the design value is within 5 percent of
the classification level.
As a result of the strong influence of transported precursors and ozone in some areas,
design values at such locations may be heavily influenced by emission changes occurring
many kilometers away hi an upwind area. Adjusted design values differ from "current air
quality design values" in that they take into account the degree to which transport of ozone
and precursors from upwind metropolitan areas contributes to ozone concentrations at the
monitoring site in question.
This study described a computer model, Transported Ozone Design Value (TODV),
1-13
-------
which has been developed to assist in determining the likely source regions associated with
high ozone concentration events (Sabo and Hawes, 1990). TODV can only provide an
approximate location of the emissions source region likely to have influenced a particular
afternoon ozone peak. No estimate of the relative contributions of upwind vs. local
emissions to the peak is provided, and back-trajectory calculations based on routine wind
data can contain large uncertainties. Selection of the transport-adjusted design value requires
an experienced analyst to interpret the results, which introduces a subjective element to the
adjustment process. Application of this approach to 1988-90 ozone data yielded transport-
adjusted "control strategy" design values for 35 areas. These transport adjustments ranged
from a decrease of 0.05 ppm to increases of 0.04 ppm. Thus, these transport adjustments
can lead to both decreases and increases in an area's design value as the downwind impact is
attributed back to the source area, or the impact from upwind areas is subtracted out.
Adjusting for Meteorological Variability
Meteorological conditions have been shown to play a key role in explaining variations
in daily maximum ozone concentrations. Given similar precursor emissions, the basic
differences between days when ozone concentrations are average or below average and days
when concentrations are high (i.e., episode days) are in the prevailing meteorological
conditions. High ozone concentrations are likely to occur with low wind speeds, elevated
temperatures, intense solar radiation (i.e., no cloud cover), shallow mixing depths, and the
wind patterns that bring, keep, or return high background concentrations to the region. In
some years, such meteorological conditions occur more frequently and with greater intensity
than in others, leading to a greater number of high ozone days even if precursor emission
levels do not differ significantly from those in other years. Thus, design values determined
from a single year of data vary in accordance with weather conditions during the year in
question and may or may not be representative of design values that can be expected to occur
in the future, even in the absence of any precursor emission trends. To some extent, basing
design values on three years of data instead of one eliminates some of the meteorological
variability, but a single unusual year such as 1988 can still strongly affect the three-year
value. This has raised concern that meteorological variability must be considered when
assessing ozone air quality trends and judging progress toward attainment of the ambient
standards (NRC, 1991).
The influence of meteorological conditions, particularly temperature, on ozone
concentrations has been well established (NRC, 1991; Sweitzer and Kolaz, 1984; Jones,
1985; Jones, 1989; Kolaz and Swinford, 1990; Wakim, 1990; Zeldin and Meisel, 1978; Cox
and Chu, 1992). The most successful empirical models used in ozone trends adjustments
account for roughly 60-80 percent of the variance in daily maximum ozone concentrations
(see for example Kolaz and Swinford, 1990; Wakim, 1990; Cox and Chu, 1992). Due to
correlations of temperature with other variables, the daily maximum temperature is often the
single most important variable in explaining day-to-day ozone variations. However, since
high temperature by itself is not sufficient to produce high ozone concentrations, including
other meteorological variables in the analysis often produces better results. This is
1-14
-------
particularly true at locations where ozone and precursor materials transported from upwind
source regions account for a significant concentration increment on high ozone days. High
concentrations at such locations are primarily associated with weather conditions conducive to
both ozone formation and transport from the upwind source regions.
Much of the year-to-year variability in ozone design values and other summary
statistics is attributable to interannual variations in prevailing weather conditions during the
high ozone season. These fluctuations can mask underlying ozone trends associated with
changes in precursor emission patterns and can affect estimates of design values. As a
result, a great deal of attention has been given to the development of procedures for adjusting
summary statistics to remove the effects of meteorological fluctuations (Sweitzer and Kolaz,
1984; Jones, 1989; Jones, 1992; Kolaz and Swinford, 1990; Wakim, 1990; Zeldin and
Meisel, 1978; Cox and Chu, 1992). A wide variety of methods have been used, all of which
rely on the development of a mathematical relationship between ozone concentrations and
meteorological factors. This relationship is then used to estimate (predict) ozone
concentrations expected to occur under standardized meteorological conditions. Figure 1-2
illustrates actual and adjusted trends in the number of days the ozone NAAQS was exceeded
in Chicago (Kolaz and Swinford, 1990). The "adjusted" summary statistics calculated from
these predicted concentrations can then be examined for trends. Figure 1-3 shows how an
index of ozone conducive days (days with maximum daily temperature greater than 90° F)
can be used to adjust the trend in the number of exceedances of the ozone NAAQS (Jones,
1992). Although these approaches are very useful for assessing trends, one must consider
how meteorological adjustment affects the intended level of protection for the standard if
such an approach were to be used for assessing compliance with the ozone NAAQS.
It may be possible to improve the performance of meteorological adjustment
techniques by focusing on meteorological variables that describe the persistence of ozone-
conducive conditions over multi-day periods (Cox and Chu, 1992). The importance of
persistence and the day-to-day carryover of pollutants has been demonstrated by Kolaz and
Swinford among others (Kolaz and Swinford, 1990). EPA has initiated a program to
investigate techniques for adjusting ozone trends for meteorological influences. One of the
methods being studied is a statistical model developed by Cox and Chu in which the
frequency distribution of ozone concentrations is described as a function of meteorological
parameters (Cox and Chu, 1992). The results of application of the model to a number of
urban areas are encouraging. Figure 1-4 shows the actual and adjusted trends in the 99th
percentile concentrations in Chicago. EPA is seeking to review and expand the technical
basis for the methodology under a cooperative agreement with the National Institute of
Statistical Sciences (NISS).
1-15
-------
30-
25
to
Q 20-
8
I 15-
o>
0)
g
UJ 10
1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989
Year
Actual
Adjusted
Figure 1-2. Actual and adjusted trends in number of days on which ozone concentrations
exceed 0.12 ppm hi the Chicago area (adapted from Kolaz and Swinford,
1990).
1-16
-------
50
40
30
UJ a
UJ •-
20
m
>»
9
10
TRENDS IN O3 AIR QUALITY AND TEMPERATURE
N.Y./N.J./CT. REGION
1980-1992
A
f
f
I
\
A
/ \ Ttrnpcratura
• l
•
V
80818283848586878889909192
YEAR :
Monitoring Site: Fairfield County
60
50
40 Cb
§5
2§
01.0
30 If
20
10
Figure 1-3. Adjustment of ozone trend based on number of days above 90° F (Source:
Jones, 1992).
1-17
-------
CHICAGO OZONE TRENDS
99 TH PERCENTILE—DAILY MAXIMUM (JUNE-SEPT)
220
200
180
160
140
O
o 120
100
80
60
ACTUAL
ADJUSTED
95XCONF
1981 1982 1983 1984 1985 1986 1987 1988 1989 1990
YEAR
Figure 1-4. Actual and meteorologically adjusted ozone trends in the 99th percentile of the
daily maximum 1-hour ozone concentration for Chicago, 1981-1990 (adapted
from Cox and Chu, 1992).
1-18
-------
MAJOR FINDINGS
The Ozone Design Value Study has examined the current EPA method, as well as
alternative approaches, for calculating ozone design values. The key findings of the study
are as follows:
1. With passage of the Clean Air Act Amendments of 1990, the primary role of the
air quality design value is to establish the ozone classification of ozone nonattainment areas.
2. Although the year-to-year differences in maximum ozone concentrations can be
large, all of the different methods examined in this study for estimating air quality design
values exhibit less year-to-year variability. The EPA design value is slightly more variable
than lower percentile indicators and design values estimated from fitted tail exponential
distributions, although it remains highly correlated with these indicators.
3. Increasing the number of years used to estimate the design value reduces the
year-to-year fluctuations. Comparisons made for 3-year periods ending in 1988-90 had less
variability in the design value estimates than during previous 3-year periods. This is likely
due to the fact that there was a single dominant year (1988) for peak ozone levels during the
1988-90 time period.
4. The past decade has seen large year-to-year variability in ozone concentrations.
However, the relative variation in ozone concentrations recorded among monitoring sites
throughout large urban areas can be as great as, or greater than, the year-to-year variation in
ozone concentrations recorded at a particular monitoring location. Spatial variations in ozone
concentrations at smaller, sub-metropolitan-length scales are not well defined in many areas
due to the sparsity of ozone monitors.
5. The EPA tabular design value method tends to give lower, but more variable
estimates for the ozone design value than some of the statistical modeling methods, such as
the Breiman tail exponential approach. Results of the time series modeling analysis suggest
that the tail exponential approach provides the best compromise regarding bias and precision
in the estimate of the "true" design value.
6. Given the database available at the time, generally data through 1989, the use of
more robust (less variable) methods such as the tail exponential approach would not have
significantly changed the initial ozone nonattainment area designations and classifications.
Use of more years of data (i.e., 4 or 5 years) in estimating the design value would have
resulted in lower classifications in only a limited number of cases. However, more recent
data periods that do not include 1988 yield significantly different results. For the years
1989-91, the first 3-year compliance period that excludes the 1988 data, 42 of the original
classified 98 nonattainment areas have ambient ozone meeting the standard. Seven of these
areas have been redesignated to attainment. The most recent compliance period, 1991-93,
1-19
-------
has 48 of the remaining 91 classified nonattainment areas also meeting the ozone standard.
7. Since the "true" design value is in the tail of the ozone concentration distribution,
the EPA tabular design value method and more robust alternatives are perforce subject to
greater variability than estimators of the central part of the distribution. Like any statistical
estimator, errors in these estimated design values can lead to 1990 CAAA misclassification
of nonattainment areas, just as errors in the Appendix H estimated expected exceedance rate
can lead to misclassification of attainment areas as nonattainment areas and vice versa.
Analyses included in this study provide estimates of the theoretical misclassification rates, but
for a given site and monitoring period it is impossible to determine whether the estimated
classification is the (unknown) true classification.
8. The "air quality" design value differs in concept and application from the
"control strategy value." The former is based solely on the actual measured ozone air
quality data and relates directly to the form of the ozone NAAQS. Control strategy design
values have historically been used to evaluate emission control strategies, and may
incorporate adjustments for factors such as transported ozone levels and meteorological
variability. Use of the control strategy value concept to judge attainment under the Act
would require EPA to revise its preliminary interpretation of Section 181(b)(2) published in
the General Preamble to Title I.
9. For thirty-five areas modeled, the transport contribution to design values in areas
subject to transport was found to be as large as 0.05 ppm. Increases in the design value of
up to 0.04 ppm were estimated when the downwind impact was attributed back to the source
area.
10. EPA has initiated a program (Cox and Chu, 1991) to investigate techniques for
adjusting ozone trends for meteorological influences. One method being studied is a
statistical model in which the frequency distribution of ozone concentrations is described as a
function of meteorological parameters. EPA is seeking to review and expand the technical
basis for the methodology under a cooperative agreement with the National Institute of
Statistical Sciences (NISS). Preliminary results suggest that the bias and uncertainty
associated with long-trend estimates can be significantly reduced by including meteorological
covariates as parameters in the statistical modeling process.
11. The use of a simple linear function of the 95th percentile of the distribution of
daily maximum ozone concentrations as a surrogate design value is less satisfactory than any
of the four more direct estimators of the design value. It fails to significantly reduce the
variability of the associated estimated characteristic largest value (CLV) below that achieved
with the more direct methods. (From another perspective: controlling the 95th percentile
fails to improve control of the underlying CLV.) At the same time it introduces substantial
biases which vary with the site. The bias problem would result in uneven treatment of sites
relative to what would be achieved with the more direct measures. Nor would the use of the
95th percentile obviate the need to use 3-year data sets.
1-20
-------
CONCLUSIONS
The question for the Ozone Design Value Study is "Does the EPA design value
methodology provide a 'reasonable indicator of ozone air quality in ozone nonattainment
areas'?" The answer depends on the intended application of the design value. Each
nonattainment area was classified as a Marginal Area, a Moderate Area, a Serious Area, a
Severe Area, or an Extreme Area based on the design value for the area. The area's
classification establishes the primary standard attainment date and the requirements for State
Implementation Plans.
In responding to Section 183(g), EPA sought to focus this study on whether the
design value serves as a reasonable indicator of attainment status as defined by the current
NAAQS, progress in reaching attainment, and of how much concentrations must be reduced
to meet the standard. The EPA design value method yields an estimate for the ozone design
value that is consistent with the current ozone NAAQS. Given the findings of this study, the
EPA design value yields a "reasonable" estimate of the "true" air quality design value for the
area and of peak ozone levels within the nonattainment area for the initial three year
compliance period.
The EPA design value provides a reasonable estimate of peak levels within the urban
area, and the degree of nonattainment of the area. However, the design value cannot
describe the spatial variability in ozone concentrations across the area. More robust
indicators based on specific monitoring sites also have large spatial variability. Ozone design
values calculated with the EPA design value method are highly correlated with other more
robust indicators. However, due to the spatial variability observed across urban areas, one
cannot expect a single numerical value to adequately describe complex concentration
gradients across large metropolitan areas.
The current EPA design value method may not provide a reasonable indicator of
ozone levels in future years due to the large year-to-year variability in meteorological
conditions, or to reductions in emissions following implementation of control measures.
However, other more robust air quality indicators also exhibit similar year-to-year
variability.
The method used to adjust for meteorological influences on long-term ozone trends
could be adapted for use hi calculating meteorologically adjusted exceedance rates and design
values. While such adaptations are technically feasible, and would reduce the year-to-year
variability, the use of adjusted exceedance rates in NAAQS attainment and adjusted design
values for classification purposes would represent a major departure from current EPA policy
and NAAQS implementation guidelines. Also, a meteorologically adjusted design value may
not be the best indicator of the air that people breathed during a specific calendar year.
Concerns about the current ozone standard were raised during the public review
process. Although changes to the form of the ozone standard were outside the scope of this
1-21
-------
study, they are being considered within the context of the current review of the ozone
NAAQS. The knowledge gained from the input of all parties to this study during the public
review process will be used to address issues concerning the form of the ozone standard and
design value methodologies.
REFERENCES
Breiman, L., J. Gins, and C. Stone. 1978. "Statistical Analysis and Interpretation of Peak
Air Pollution Measurements," Work performed under EPA Contract 68-02-2857,
Technology Service Corporation, Santa Monica, CA.
CARB. 1992a. "Proposed Amendments to the Criteria for Designating Areas of California
as Nonattainment, Attainment, or Unclassified for State Ambient Ah- Quality
Standards." California Air Resources Board, Sacramento, California.
CARB. 1992b. Technical Support Document for Proposed Amendments to the Criteria for
Designating Areas of California as Nonattainment, Attainment, or Unclassified for
State Ambient Air Quality Standards; California Air Resources Board, Sacramento.
Chock, D. P. 1989. "The need for a more robust ozone air quality standard", Journal of
the Air Pollution Control Association. 39:1063.
Chock, D. P. 1992. "Statement of Ford Motor Company before the U.S. EPA on the
Ozone Design Value Study of the Clean Air Act Amendments of 1990", Washington,
DC.
Cox, W. M. and S. H. Chu. 1992. "Meteorologically Adjusted Ozone Trends hi Urban
Areas: A Probability Approach", in Transactions of the Tropospheric Ozone and the
Environment II International Specialty Conference. Air and Waste Management
Association, Pittsburgh, PA, pp 342-353.
Curran, T. C. 1992a. "Present Status and Current Program", panel discussion presented at
the Tropospheric Ozone: Nonattainment and Design Value Issues International
Conference. Air and Waste Management Association, Boston, MA.
Curran, T.C. 1992b. "The Clean Air Act Ozone Design Value Study", in Transactions of
the Tropospheric Ozone and the Environment II International Specialty Conference.
Ah" and Waste Management Association, Pittsburgh, PA, pp 354-364.
Curran, T. C., and N. H. Frank. 1975. "Assessing the Validity of the Lognormal Model
When Predicting Maximum Air Pollution Concentrations", 68th Air Pollution
Control Association Annual Meeting, Boston, Massachusetts.
1-22
-------
deNevers, N. and J.R. Morris. 1975. "Rollback Modeling: Basic and Modified", Journal of
the Air Pollution Control Association. Volume 25, No. 9.
EPA. 1979. Guideline for the Interpretation of Ozone Air Quality Standards. EPA-450/4-
79-003, U.S. Environmental Protection Agency, Research Triangle Park, NC,
February 1979.
EPA. 1981. Guideline for Use of City-specific EKMA in Preparing Ozone SIPs. EPA-
450/4-80-027, U.S. Environmental Protection Agency, Research Triangle Park, NC.
EPA. 1987. "EPA's Office of Air Quality and Planning has completed its review of 1986
air quality monitoring data of the air pollutants ozone and carbon monoxide", Press
Release. U.S. Environmental Protection Agency, Washington, DC.
EPA. 1988. "EPA Lists Areas Failing to Meet Ozone and Carbon Monoxide Standards",
Press Release. U.S. Environmental Protection Agency, Washington, DC.
EPA. 1989. "EPA Lists Areas Failing to Meet Ozone and Carbon Monoxide Standards",
Press Release. U.S. Environmental Protection Agency, Washington, DC.
EPA. 1990. "EPA Announces 96 Areas Failing to Meet Smog Standards, 41 Areas
Violating Carbon Monoxide Standards", Pressjtelease, U.S. Environmental
Protection Agency, Washington, DC.
EPA. 1991a. "EPA Announces the Availability of 1990 Data", Note to Correspondents.
U.S. Environmental Protection Agency, Washington, DC.
EPA. 1991b. Guideline for Regulatory Application of the Urban Airshed Model. EPA-
450/4-91-013, U.S. Environmental Protection Agency, Research Triangle Park, NC.
EPA. 1992a. "Designations/Classifications Corrections Notice and Ozone/Carbon
Monoxide Technical Support Document", U.S. Environmental Protection Agency,
Research Triangle Park, NC.
EPA. 1992b. "EPA Data Show Steady Progress in Cleaning Nation's Air", Press Release.
U.S. Environmental Protection Agency, Washington, DC.
EPA. 1994. "EPA Report Shows Continuing Progress in Cleaning Nation's Air", Press
Release. U.S. Environmental Protection Agency, Washington, DC.
Federal Register. 1989. "Volatility Regulations for Gasoline and Alcohol Blends Sold in
Calendar Years 1989 and Beyond", (54FR11868).
1-23
-------
Federal Register. 1991. "Air Quality Designations and Classifications; Final Rule",
(56FR56694).
Federal Register. 1992. "State Implementation Plans; General Preamble for the
Implementation of Title I of the Clean Air Act Amendments of 1990; Proposed Rule",
Federal Register. April 16, 1992.
Freas, W. P. 1992a. "Clean Air Act Ozone Design Value Study: Impact of Varying the
Number of Data Years on Estimates of Ozone Design Values", presented at the
Tropospheric Ozone: Nonattainment and Design Value Issues International
Conference. Air and Waste Management Association, Boston, MA.
Freas, W. P. 1992b. "EPA's Design Value Study", Tropospheric Ozone: Air Quality
Attainment Issues. Air and Waste management Information Exchange Workshop,
Research Triangle Park, NC.
Heim, R. H. Jr. 1988. "United States Summer Climate in Historical Perspective", National
Climatic Data Center, NOAA, Asheville, NC.
Heim, R. H. Jr. 1989. "United States Summer Climate in Historical Perspective", National
Climatic Data Center, NOAA, Asheville, NC.
Heim, R. H. Jr. 1990. "United States Summer Climate in Historical Perspective", National
Climatic Data Center, NOAA, Asheville, NC.
Heuss, J. M. 1992. "Statement of the Motor Vehicle Manufacturer's Association of the
United States, Inc. before the U.S. EPA on the Ozone Design Value Study of the
Clean Air Act Amendments of 1990", Washington, DC.
Heuss, J. M., and D. P. Chock. 1992. "The role of the design value in ozone compliance",
in Transactions of the Tropospheric Ozone and the Environment II International
Specialty Conference. Air and Waste Management Association, Pittsburgh, PA, pp
377-388.
Jones, K. H. 1985. "Urban Air Quality," in Environmental Quality 1984. Council on
Environmental Quality, Washington, D.C.
Jones, K. H. 1989. "Urban Air Quality," in Environmental Quality 1987-88. Council on
Environmental Quality, Washington, D.C.
Jones, K. H. 1992. "The 1990/91/92 O3 Data Base and Its Implications Relative to
Currently Designated Ozone Nonattainment Area Regulatory Programs." Air and
Waste Management Association Specialty Conference, Tropospheric Ozone:
Nonattainment and Design Value Issues, Boston Massachusetts, October 1992.
1-24
-------
Kolaz, D. J. and R. L. Swinford. 1990. "How to Remove the Influence of Meteorology
from the Chicago Area Ozone Trend", Presented at the Air and Waste Management
Association 83rd Annual Meeting and Exhibition, Pittsburgh, PA.
Larsen, L. C. 1991. "Evaluating the Performance of an Exponential-Tail Model for Use in
Determining Ozone Attainment Designations in California." Air and Waste
Management Association Tropospheric Ozone and the Environment II Specialty
Conference, Atlanta, Georgia.
Larsen, L. C., and R. A. Bradley. 1991. "Use of an Exponential-Tail Model to Estimate
Ozone Concentrations with an Infrequent Recurrence Rate in California," Air and
Waste Management Association 84th Annual Meeting and Exhibition, Vancouver,
British Columbia.
Laxton, W. G. 1990. "Ozone and Carbon Monoxide Design Value Calculations." U.S.
Environmental Protection Agency memorandum, Office of Air Quality Planning and
Standards, Research Triangle Park, NC.
Mage, D. T. 1984. "Pseudo lognormal distributions", Journal of the Air Pollution Control
Association. 31(4):374-376.
McGuire, T. 1994. Comments on draft ozone design value study report. Letter to Robert
Kellam, May 27, 1994.
Meyer, E. L., G. L. Gipson, and W. P. Freas. 1977. Procedures for Quantifying
Relationships Between Photochemical Qxidants and Precursors. EPA-450/2-77-021a,
U.S. Environmental Protection Agency, Research Triangle Park, NC, November
1977.
MVMA. 1988a. "National Fuel Survey: Gasoline and Diesel Fuel - Summer 1989", Motor
Vehicle Manufacturer's Association, Washington, DC.
MVMA. 1988b. "National Fuel Survey: Motor Gasoline - Summer 1988", Motor Vehicle
Manufacturer's Association, Washington, DC.
MVMA. 1988c. "National Fuel Survey: Motor Gasoline - Summer 1990", Motor Vehicle
Manufacturer's Association, Washington, DC.
NRC. 1991. Rethinking the Ozone Problem in Urban and Regional Air Pollution. National
research Council, National Academy Press, Washington, DC.
Rhoads, R. G. and D. D. Tyler. 1987. "Distinction between 'Ozone Design Value' and 'SIP
Control Value' for Ozone", U.S. Environmental Protection Agency, Research
Triangle Park, NC (memorandum dated April 29, 1987).
1-25
-------
Sabo. E. J., and J. T. Hawes. 1990. "User's Guide and Program Documentation for the
Transported Ozone Design Value Model", prepared under EPA Contract No. 68-02-
4393, U.S. Environmental Protection Agency, Research Triangle Park, NC.
SAS. SAS/STAT User's Guide. Volume 2, SAS Institute Inc., Gary, NC, p. 997.
Sweitzer, T. A., and D. J. Kolaz. 1984. "An Assessment of the Influence of Meteorology
on the Trend of Ozone Concentrations in the Chicago Area." Air Pollution Control
Association/American Society for Quality Control Specialty Conference on Quality
Assurance in Air Pollution Measurements, Boulder, CO.
Wakim, P. G. 1990. "1981 to 1988 Ozone Trends Adjusted to Meteorological Conditions
for 13 Metropolitan Areas", presented at the 83rd Annual Meeting of the Air and
Waste Management Association, Pittsburgh, PA.
Wilson, J. H. and M. A. Scruggs. 1980. Methodologies To Conduct Air Quality
Assessments of National Mobile Source Emission Control Strategies. EPA-450/4-80-
026a, U.S. Environmental Protection Agency, Research Triangle Park, NC.
Zeldin, M. D. and W. S. Meisel. 1978. Use of Meteorological Data in Air Quality Trend
Analysis. EPA-450/3-78-024, U.S. Environmental Protection Agency, Research
Triangle Park, NC.
1-26
-------
2 INTRODUCTION
The Clean Air Act Amendments (CAAA) of 1990 specified that each nonattainment
area in the U.S. be classified with regard to the severity of the ozone problem on the basis of
the ozone design value concentration. In the terminology of air quality analysts, a design
value is the concentration that must be reduced to the level of the standard to achieve
attainment of the standard. Classifications range from marginal to extreme depending on
how far above 0.12 ppm the design value is. In addition, the amendments call upon EPA to
conduct a study of the methodology currently in use for calculating design values to
determine if the calculated design value "provides a reasonable indicator of the ozone air
quality of ozone nonattainment areas" (CAAA, 1990).
This section provides background information on technical issues associated with the
formulation of the ozone NAAQS, and procedures for defining, calculating, and interpreting
ozone design values. This section concludes with a discussion of the scope of the Ozone
Design Value Study.
OZONE STANDARDS AND DESIGN VALUES
National Ambient Air Quality Standard for Ozone
In 1979, the U.S. Environmental Protection Agency (EPA) promulgated a National
Ambient Air Quality Standard (NAAQS) for ozone at a level of 0.12 parts per million (ppm).
This standard is attained "when the expected number of days per calendar year with
maximum hourly average concentrations above 0.12 part per million is equal to or less than
1 as determined by Appendix H" (40CFR50.9). Appendix H describes the manner in which
the "expected number" of days with concentrations above 0.12 ppm ("exceedance" days) is
estimated. Additional information is contained in the EPA Ozone Guideline document (EPA,
1979). Appendix H and the Ozone Guideline specify procedures for dealing with missing
data and make it clear that the expected exceedance criterion is to be applied independently
to each monitoring site. For areas with multiple monitoring sites, all sites must meet the
standard for the area to be designated in attainment.
The formulation of the ozone standard is designed to take into account the fact that
daily maximum ozone concentrations vary widely from one day to the next in response to
changing weather conditions. This in turn causes year-to-year variations in the number of
exceedances recorded at a monitoring site. A larger number of exceedances are recorded in
years when weather conditions conducive to ozone formation occur more frequently than
other years. In recognition of this variability, the ozone NAAQS is written in terms of an
expected exceedance rate. This is a statistical term which can be defined as the value
approached by the average number of exceedances per year as the number of years included
2-1
-------
in the average is increased without limit.1 For this limiting average to exist, it must be
assumed that the random processes which cause the number of exceedances to vary from one
year to the next are stationary, i.e., that there is no change in their nature over time that
causes the number of exceedances per year to permanently increase or decrease; instead, the
exceedances simply fluctuate randomly about some long-term, fixed average value which is
defined as the expected exceedance rate.
Since the amounts of precursor emissions can be expected to vary over time as a
result of emission control programs, and economic and demographic trends, the number of
exceedances per year observed at a monitoring site can be expected to exhibit
nonstationarities such as long-term trends. Thus, the expected exceedance rate at a given
point in time is the expected exceedance rate associated with emission levels prevailing at
that time. If emissions change (e.g., following implementation of a control program), the
expected exceedance rate is also likely to change. This raises an issue concerning the
representativeness of data collected in previous years when emission levels may have differed
from current values.
Implementation of the Ozone NAAQS: Estimating the Expected Exceedance Rate
From the above discussion it is clear that the expected exceedance rate is a
mathematical abstraction that cannot be directly observed (and is therefore unknown) but
must instead be estimated from what data are available. As required by Appendix H, the
simplest estimate is obtained by averaging exceedances over a three-year period, that is, by
adding up the number of exceedances observed during the three-year period and dividing by
3 (assuming no data are missing; see below). If there are no nonstationarities during the data
collection period, then the more years included in the average, the more accurate the estimate
of the underlying exceedance rate will be. In most cases, however, inaccuracies arise if
more than a few years of data are used since emission patterns cannot reliably be assumed to
remain unchanged over longer time periods. Taking this tradeoff into account, EPA
formulated procedures for estimating the expected exceedance rate in Appendix H and in the
Ozone Guideline that call for the use of a three-year averaging time. The procedure was
selected as a reasonable compromise between the need to use the longest possible time period
and the need to limit the number of exceedances that would be allowed in a single year and
still have the area remain in attainment. It must be kept in mind, however, that the average
number of exceedances over a three-year period is only an approximation to the expected
exceedance rate and that it varies randomly from one three-year period to the next just as the
annual number of exceedances varies from one year to the next, although the variations tend
to be of a smaller magnitude. As noted, the actual expected exceedance rate (which is
'Appendix H to 40CFR50.9 states that the expected exceedance rate is to be determined by
calculating the average number of exceedances over a three-year period, with suitable adjustments
for missing data. A statistical definition of the term "expected exceedance rate" is not provided.
The three-year period was selected to limit the number of exceedances that can occur in a single
year and the area still remain in attainment of the NAAQS.
2-2
-------
unknown) does not vary over time so long as the random processes generating the
exceedances are stationary.
Ozone Design Values
As noted above, compliance with the ozone NAAQS is judged on the basis of the
estimated expected exceedances, and becomes a "yes/no" decision. However, once it is
determined that an area exceeds the standard, the next logical question to ask is, "By how
much?" The design value is intended to provide a measure of how far concentrations must
be reduced to achieve attainment or, equivalently, how far out of attainment the area
represented by a monitoring site is. In this respect, the design value can be viewed as an air
quality indicator for a given location.
Given the expected exceedance form of the ozone NAAQS, the design value for this
standard is defined in the EPA guideline document as "the concentration with expected
number of exceedances equal to one" (EPA, 1979). Note that in this context the ozone
guidelines are referring to the unknown "true" number of expected exceedances per year
rather than the estimate of expected exceedances determined using the Appendix H
calculations. In statistical terms, this is the value which is exceeded once per year on
average. If the daily maximum ozone concentrations are assumed to be independent and
have the same distributions every day throughout the year, then the design value is the
characteristic largest value (CLV) of that distribution. Given that there are 365 days in a
year (ignoring leap years) and assuming independent identically distributed daily maximum
ozone concentrations, the CLV is simply the value which is exceeded with probability 1/365.
Conceptually, if the CLV is less than or equal to the level of the standard (0.12 ppm) for all
monitoring sites in an area, then the expected exceedance rates at all monitors will be less
than or equal to one per year as required by the ozone NAAQS.2 Thus, the air quality
indicator (design value) which the formulation of the ozone NAAQS identifies as being of
significance is the CLV. By limiting the CLV to a value not exceeding 0.12 ppm, the ozone
NAAQS indirectly limits the frequency and magnitude of peak daily maximum ozone
concentrations. That is to say, although an area just in attainment may experience an
occasional year with several daily maximum concentrations in excess of 0.12 ppm, these will
be balanced out by years with no exceedances such that, on average, no more than one day
per year exceeds 0.12 ppm. More precisely, a Poisson distribution approximation implies
that only 8 percent of years have three or more exceedances of 0.12 ppm for a site just in
attainment. If the CLV is significantly above 0.12 ppm, then in most years a number of
days can be expected to have concentrations well in excess of 0.12 ppm.
2 It must be noted, however, that in some cases with incomplete monitoring data the CLV
estimated from monitoring data will be less than the NAAQS and yet the site will be designated
nonattainment based on a calculation of the estimated expected exceedance rate (which is the
actual basis of the attainment test called for by the NAAQS). This situation can occur because the
CLV and true expected exceedance rate are unknown, and are estimated by the design value in
Appendix H procedures, respectively, which are both subject to error.
2-3
-------
Strictly speaking, the design value is an unknown quantity depending on the underlying
distribution of ozone concentrations, and the EPA design value and alternatives are
estimators of the (true) design value. To retain the readability of this document the term
"design value" may refer to either the unknown population value or an estimator, depending
on the context in which it is used. Where appropriate, the term "true design value" is used
for clarification. The "EPA design value" always refers to the table look-up value.
Estimating Design Values
Initial EPA Guidance
Just as the expected exceedance rate is a mathematical abstraction that cannot be
directly observed but must instead be estimated from available data, the design value
associated with the ozone NAAQS is also an abstract quantity that can only be estimated.
The Ozone Guideline suggests several methods for estimating the design value, including a
simplified table look-up procedure, techniques incorporating statistical distributions, and
techniques based on conditional probabilities. No single approach was required by the
guideline, so that the initial guidance contained a fair degree of flexibility in how these
design values could be estimated. However, both the Ozone Guideline and Appendix H
made clear that anainmenilnonattainme.nl determinations are to be based on estimates of the
expected exceedance rate, not on comparisons of the design value with the level of the
standard. Once an area has been determined to be in nonattainment on the basis of the
expected exceedance rate, then estimated design values can be used to assess the degree of
nonattainment for the purpose of formulating emission control strategies. Thus, the
flexibility in design value estimation techniques incorporated into the ozone guidelines did not
extend to flexibility in attainment determinations.
Additional EPA Guidance
In 1981, EPA issued guidance on the determination of ozone design values for the
purpose of precursor emission control strategy development (EPA, 1981). This guidance is
based on use of the Empirical Kinetic Modeling Approach (EKMA) to estimate the degree of
precursor emission controls needed to reduce the ozone concentration observed on a
particular high ozone day to the level of the standard. According to the guidance, EKMA is
to be used to calculate the percentage emission reduction (control requirement) required on
each of the five highest ozone days. The control requirement to be used in State
Implementation Plans (SIPs) should then be the day with the fourth highest control
requirement assuming three valid seasons of daily ozone measurements are available. The
third highest control requirement is used if two seasons are available, and the second highest
is used if only one valid season is available. The guidance is based on the fourth highest
control requirement as opposed to the fourth highest ozone concentration since the nonlinear
relationship between emission reductions and ozone concentrations raises the possibility that
the day with the fourth highest concentration may not always correspond to the day with the
fourth highest control requirement. In most cases, however, the two days are the same.
2-4
-------
Thus, the 1981 guidance indirectly suggested that the design value for the current ozone
NAAQS could be estimated as the (/z+l)th highest daily maximum concentration given n
valid seasons of monitoring data.
Using the («+l)th highest concentration as the design value focuses attention on a
concentration that was actually observed, as opposed to a statistical fitting technique that
could yield a design value that does not correspond to a concentration observed on a
particular day. The advantage of using an observed value as the estimated design value is
that the day on which the value was observed can be used as the basis for a computer model
exercise to determine the degree of emission controls needed to reduce the daily maximum
concentration on that day to the level of the NAAQS. However, the (n+l)th highest
concentration design value may be more variable and less desirable than alternative design
values, based on a statistical fitting technique.
During the late 1980s, EPA issued ozone design value lists that used a table look-up
approach patterned after the 1981 SIP guidance (e.g., EPA, 1987). This approach is
summarized in Table 2-1. Basically, it results in the identification of the lowest observed
concentration that was not exceeded more than an average of once per year during the
measurement period.
TABLE 2-1: Ozone design value rank.
Ozone Design Value
(Based on Rank-Ordered List of
Number of Valid Years3 Daily Maximum Concentrations)
0 highest
1 2nd highest
2 3rd highest
3 4th highest
On June 18, 1990, EPA issued a memorandum signed by William G. Laxton, Director of the
Technical Support Division, documenting the procedures used in determining both ozone and
carbon monoxide design values (Laxton, 1990). This memorandum essentially maintained
the status quo and documented the table look-up procedure summarized above as the EPA
design value estimation method. This was the last EPA policy statement on design values
prior to enactment of the 1990 Clean Air Act Amendments. Table 2-2 reproduces the
3 A valid year is one in which valid daily maximum ozone concentrations are available on at
least 75 percent of days during the officially designated ozone season. A valid daily maximum
concentration requires that at least nine hourly concentrations were collected between 9:00 a.m.
and 9:00 p.m. local standard time. These restrictions ensure that missing data do not result in
significantly underestimated design values.
2-5
-------
example contained in the Laxton memo for an area with two ozone monitors to illustrate the
basic points in ozone design value estimation. This example indicates that the design value
methodology is applied independently to the two sites in the area and that the highest design
value for the sites becomes the design value for the area. The memorandum discusses
additional details for special cases, but the general methodology is essentially unchanged
from the SIP guidance issued in 1981.
TABLE 2-2. Hypothetical example of ozone design value determination (case
with two ozone sites in an area, each year at least 75 percent
complete).*
Four Highest Daily Maximum Values
SITE 1
1986
1987
1988
Max
.127
.129*
.142
2nd Hi
.123
.124
.136
3rd Hi
.122
.121
.134
4th Hi
.110
.116
.115
* The design value for Site 1 is 0.129 ppm, the fourth highest daily
maximum value during the three-year period.
Four Highest Daily Maximum Values
Max 2nd Hi 3rd Hi 4th Hi
SITE 2
1986
1987
1988
.110
.110
.180
.100
.100
.175
.095
.095
.160
.090
.090
.110*
* The design value for Site 2 is 0.110, the fourth highest value during the
three-year period. Note that any of the days with .110 could have been selected.
0.129 ppm would be the design value for the area.
1990 Clean Air Act Amendments
With passage of the 1990 CAAA, a new dimension was added to the use of ozone
design values. In addition to designating areas as nonattainment for ozone, the CAAA
introduced a classification process to further categorize nonattainment areas according to the
extent of their ozone problem. As shown in Table 2-3, this classification was based upon the
ozone design value. The CAAA stated that the design value "shall be calculated according to
2-6
-------
the interpretation methodology issued by the Administrator most recently before the date of
the enactment." Thus the June 18, 1990 Laxton memo determined the methodology
underlying the ozone classifications published in the November 6, 1991 Federal Register.
These classifications, shown on the map in Figure 2-1, resulted in 37 marginal areas, 34
moderate areas, 17 serious areas, 9 severe areas, and 1 extreme area—Los Angeles.
TABLE 2-3. Ozone classifications specified in the 1990
Clean Air Act Amendments.
Area Class
Marginal
Moderate
Serious
Severe
Extreme
Design Value*
0.121 up to 0.138
0.138 up to 0.160
0.160 up to 0.1 80
0.1 80 up to 0.280
0.280 and above
Attainment Date**
3 years
6 years
9 years
15 years
20 years
*In parts per million (ppm).
**Measured from the date of the enactment of the Clean Air
Act Amendments of 1990.
Before the 1990 CAAA, designation of nonattainment areas simply involved a
"yes/no" determination as to whether the area met the standard. The additional classification
step introduced by the 1990 CAAA placed greater emphasis on ozone concentration
observations and on the methodology used to determine the design value.
THE OZONE DESIGN VALUE STUDY
In addition to the new use of ozone design values imposed by the 1990 CAAA
described above, Section 183(g) of the CAAA required EPA to conduct an Ozone Design
Value Study. In particular, Section 183(g) states:
The Administrator shall conduct a study of whether the methodology in use by
the Environmental Protection Agency as of the date of enactment of the Clean
Air Act Amendments of 1990 for establishing a design value for ozone
provides a reasonable indicator of the ozone air quality of ozone nonattainment
areas. The Administrator shall obtain input from States, local subdivisions
thereof, and others. The study shall be completed and a report submitted to
Congress not later than 3 years after the date of the enactment of the Clean
Air Act Amendments of 1990. The results of the study shall be subject to
peer and public review before submitting it to Congress.
2-7
-------
Areas Designated Nonattainment for Ozone
Note Initial nonattainment area status as of November 6,1991
Figure 2-1. Initial ozone nonattainment area designations and classifications under the
Clean Act Amendments of 1990.
2-8
-------
On the basis of this language, it is assumed throughout this report that the focus of
the Ozone Design Value Study is on EPA's design value methodology "as of the date of
enactment of the Clean Air Act Amendments of 1990." Concerns regarding the level of
protection provided by the current ozone NAAQS are more properly treated within the
existing mechanism for NAAQS review, and the wording of Section 183(g) clearly
establishes that Congress was interested in examining EPA's ozone design value
methodology, not the ozone NAAQS. With this viewpoint, it is possible to delineate the
scope of the design value study as an examination of EPA's design value methodology
within the framework of the existing ozone NAAQS. Thus, EPA has sought to focus this
study on whether the design value serves as a reasonable indicator of attainment status as
defined by the current NAAQS, progress in reaching attainment, and of how much
concentrations must be reduced to meet the standard.
As required by the 1990 CAAA, the design value study is to consider the extent to
which the design value estimation procedure "provides a reasonable indicator of the ozone
air quality of ozone nonattainment areas." This issue can be examined in both a temporal
and a spatial framework. One issue common to both frameworks is that the "reasonable
indicator" evaluation is dependent on the intended application of the design value. It is
quite possible that a design value estimation procedure which provides a reasonable
indicator for determining the nonattainment category of a small geographic area
surrounding a monitoring site may not be suitable for the purpose of estimating the
required degree of emission reduction needed to achieve attainment or for the purpose of
estimating health risks to nearby populations. Therefore, it is necessary to indicate the
intended application of a design value estimation procedure before judging whether it yields
a reasonable air quality indicator.
Spatial Representativeness
Currently, design values are estimated individually for each monitor in an area, and
the maximum value is used to determine the nonattainment category of the entire area.
Thus, the question is "How representative is the design value measured at an individual
monitoring site of 'air quality of the nonattainment area' at large?" There is also the issue
of whether or not the monitoring network is sufficiently dense to represent air quality
over the area in question. This study examines typical spatial distributions of ozone
concentrations. Large-scale features can be determined by examining monitoring data. To
identify smaller-scale features, it may be necessary to examine spatial concentration
distributions calculated by photochemical dispersion models or the results of special,
intensive field monitoring programs.
Transport of ozone and ozone precursors generated in one air basin can significantly
influence ozone concentrations in neighboring air basins located considerable distances
downwind. Title I of the 1990 Clean Air Act Amendments prohibits any activity from
emitting pollutants that contribute significantly to nonattainment in another state or
interfere with Prevention of Significant Deterioration requirements. The amendments
2-9
-------
specifically acknowledge that transport across state boundaries plays a major role during
high ozone events in the northeastern urban corridor between Washington, D.C. and
Boston. The amendments call for the establishment of a transport commission to study
this issue.
Temporal Representativeness
A major concern with respect to the temporal representativeness of design values is
the number of years of data used to calculate the value. Current EPA guidance on
estimating ozone design values calls for the use of three years of data, if available. As
discussed, the use of three years of data is a compromise between the need to include as
much data as possible to arrive at an accurate estimate of the underlying expected
exceedance rate and the need to limit the number of exceedances allowed in a single year
and still have the area remain in attainment. However, year-to-year variations in weather
conditions can result in significant differences in estimated design values from one three-
year period to the next, even in the absence of emission changes, as shown by McCurdy
and Atherton (1990).
Reasonable Indicator Criteria
To judge whether or not various design value estimation procedures provide a
reasonable indicator of air quality, criteria for determining what is reasonable need to be
developed. In addition to concerns about spatial and temporal representativeness, these
criteria must also address issues such as:
Symmetry: This concept is best explained by example. Consider a proposal to
incorporate information on the uncertainty in an estimated design value when
judging attainment status. If, for an area actually in attainment, the probability of
being falsely judged in nonattainment on the basis of the estimated design value is to
be considered, then, for an area actually in nonattainment, the probability of being
falsely judged in attainment must also be considered, and, conversely, if the
probability of a false attainment determination is considered, then so would the
probability of a false nonattainment determination.
Consistency/Equity: A design value estimation procedure should be consistent
from one area to the next. People in different locations should all receive the level
of health protection intended by the NAAQS. Any procedure that results in
different treatment for areas with similar air quality problems should be viewed
with caution.
Clarity: Technical details of a design value estimation procedure should be
sufficiently clear so that there is no ambiguity about how the procedure is to be
applied. It is also desirable that the results of the procedure can be easily
communicated to the general public.
2-10
-------
Other Design Value Issues
Another reference to the use of design values is contained in Section 181(b) of the
Act, which states that EPA "shall determine, based on the area's design value (as of the
attainment date), whether the area attained the standard by that date." EPA's preliminary
interpretation of this Section is that the "average number of exceedances per year shall be
used to determine whether the area has attained," which is the attainment test for the
ozone NAAQS and not the ozone design value (EPA, 1992). Thus, design values will not
be used to make this determination, although for complete monitoring data the two
determinations may be the same.
STRUCTURE OF THE REPORT
The remaining sections of this report include:
Section 3. BACKGROUND—background information on the nature of the ozone problem.
Section 4. OZONE DESIGN VALUE METHODOLOGffiS-discussion of alternative
methods for estimating ozone design values.
Section 5. EVALUATION APPROACHES FOR OZONE DESIGN VALUE
METHODS—ambient data base and time series modeling.
Section 6. COMPARISONS AMONG ALTERNATIVE DESIGN VALUE
METHODS—comparisons of alternative design value methodologies using the
ambient database and time series modeling approach.
Section 7. THE ROLE OF METEOROLOGY IN OZONE FORMATION-an assessment
of the impact of meteorology on ozone formation and methods for adjusting
ozone trends for meteorological variability.
Section 8. ASSESSING THE ROLE OF TRANSPORTED OZONE AND
PRECURSORS—an assessment of the impact of transported ozone and methods
for adjusting ozone design values for transport.
Section 9. DETECTING TRENDS IN OZONE DESIGN VALUES-methods for
determining trends in ozone design values.
Section 10. ALTERNATIVE AIR QUALITY INDICATORS-a review of alternative air
quality indicators and their relationship to ozone design values.
Section 11. PUBLIC OUTREACH EFFORTS-a summary of the public outreach efforts,
including public meetings, technical conferences and technical review.
Section 12. SUMMARY OF FINDINGS.
2-11
-------
REFERENCES
ARB. 1990. "Assessment and Mitigation of the Impacts of Transported Pollutants on
Ozone Concentrations within California." California Air Resources Board,
Sacramento, California.
CAAA. 1990. Clean Air Act Amendments, P.L. 101-549, Nov. 15, 1990, §183 g.
Cleveland, W. S., B. Kleiner, J. E. McRae, and J. L. Warner. 1976. Photochemical air
pollution: Transport from the New York City area into Connecticut and
Massachusetts. Science. 191:179-181.
Chock, D. P. 1989. The need for a more robust ozone air quality standard. J. Air Pollut.
Control Assoc.. 39:1063.
Chock, D. P., and J. M. Heuss. 1987. Urban ozone and its precursors. Environmental
Science and Technology. 21(12):1146.
Code of Federal Regulations. 40 CFR Part 50.9 (FR 44:8220).
Curran, T. C., and W. P. Freas. 1991. "Consequences of a More Statistically Robust
Ozone Air Quality Standard." Presented at the 84th Annual Meeting & Exhibition
of the Air & Waste Management Association, Vancouver, British Columbia (June
16-21, 1991).
EPA. 1987. [ref. footnote 3 of EPA press release 8/27/87, Dave Cohen, press office:
"EPA's Office of Air Quality and Planning has completed it's [sic] review of 1986
air quality monitoring data of the air pollutants ozone and carbon monoxide"].
U.S. Environmental Protection Agency.
EPA. 1987. Metropolitan Statistical Areas (MSA) Regulatory Analysis Air Quality Data
Base. 1983-1985. U.S. Environmental Protection Agency.
EPA. 1979. Guideline for the Interpretation of Ozone Air Quality Standards. U.S.
Environmental Protection Agency (EPA-450/4-79-003).
EPA. 1981. Guideline for Use of City-specific EKMA in Preparing Ozone SIPs. U.S.
Environmental Protection Agency (EPA-450/4-80-027).
EPA. 1991. National Air Quality and Emissions Trends Report. 1990. U.S.
Environmental Protection Agency (EPA-450/4-91-023).
2-12
-------
EPA. 1992. "State Implementation Plans; General Preamble for the Implementation of Title
I of the Clean Air Act Amendments of 1990; Proposed Rule." (57 FR 13498) U.S.
Environmental Protection Agency.
Fairley, D., and C. L. Blanchard. 1991. Rethinking the ozone standard. J. Air Waste
Manage. Assoc.. 41(7):928.
Heuss, J. M., and D. P. Chock. 1991. "The Role of the Design Value in Ozone
Compliance." Presented at the Air and Waste Management Association International
Specialty Conference, Atlanta, Georgia (November 1991).
Javitz, H. S. 1980. Statistical interdependencies in the ozone national ambient air quality
standard. J. Air Pollut. Control Assoc.. 30:58.
Langstaff, J. E., and A. K. Pollack. 1985. "Meteorological Characterization of High Ozone
Levels: A Pilot Study of St. Louis, Missouri." Systems Applications, Inc., San
Rafael, California.
Laxton, W. G. 1990. "Ozone and Carbon Monoxide Design Value Calculations." U.S.
Environmental Protection Agency memorandum, Office of Air Quality Planning and
Standards (June 18, 1990).
Lefohn, A. S., D. S. Shadwick, U. Feister, and V. A. Mohnen. 1992. Surface-level ozone:
Climate change and evidence for trends. J. Air Waste Manage. Assoc.. 42(2): 136.
Lippmann, M. 1989. Health effects of ozone. A critical review. J. Air Waste Manage.
Assoc.. 39(5):672.
McCurdy, T., and R. Atherton. 1990. Variability of ozone air quality indicators in selected
metropolitan statistical areas. J. Air Waste Manage. Assoc.. 40(4):477-486.
NRC. 1991. Rethinking the Ozone Problem in Urban and Regional Air Pollution. National
Research Council, National Academy Press, Washington, D.C.
Pollack, A. K., T. E. Stoeckenius, J. L. Haney, T. S. Stocking, J. L. Fieber, and
M. Moezzi. 1988. "Analysis of Historical Ozone Concentrations in the Northeast.
Volume I: Main Report." Systems Applications, Inc., San Rafael, California
(SYSAPP-88/192a).
Possiel, N. C., D. C. Doll, K. A. Baugues, E. W. Baldridge, and R. A. Wayland. 1990.
"Impacts of Regional Control Strategies on Ozone in the Northeastern United States."
Presented at the Air and Waste Management Association's 83rd Annual Meeting &
Exhibition, Pittsburgh, Pennsylvania (June 24-29, 1990).
2-13
-------
Rhoads, R. G., and D. D. Tyler. 1987. Distinction between "Ozone Design Value" and
"SIP Control Value" for Ozone. U.S. Environmental Protection Agency
memorandum (April 29, 1987).
Samson, P. J., and B. Shi. 1988. "A Meteorological Investigation of High Ozone Values in
American Cities." Department of Atmospheric, Oceanic, and Space Sciences, Space
Physics Research Laboratory, The University of Michigan, Ann Arbor, Michigan.
Stoeckenius, T. E. 1991. "Development of Procedures for Estimating Effects of Emission
Control Strategies on Long-Term Ozone Concentrations in the Presence of
Climatological Changes." Systems Applications International, San Rafael, California
(SYSAPP-91/121).
Wackier, D. J., and P. V. Bayly. 1987. "The Effectiveness of Emission Controls on
Reducing Ozone Levels in Connecticut from 1976 Through 1987." The Scientific and
Technical Issues Facing Post-1987 Ozone Control Strategies. Air & Waste
Management Association, Hartford, Connecticut, November 1987.
Whitten, G. Z., and M. W. Gery. 1986. "The Interaction of Photochemical Processes in
the Stratosphere and Troposphere." In Effects of Changes in Stratospheric Ozone and
Global Climate. Volume 2: Stratospheric Ozone. J. G. Titus, ed. U.S.
Environmental Protection Agency.
Zeldin, M. D., and W. S. Meisel. 1978. Use of Meteorological Data in Air Quality Trend
Analysis. U.S. Environmental Protection Agency (EPA-450/3-78-024).
2-14
-------
3 BACKGROUND
In this section, background information on the tropospheric ozone problem is
presented to familiarize the reader with the basic technical issues concerning the nature of the
ozone problem. Brief discussions are presented concerning the chemistry of ozone
formation, the spatial and temporal distribution of ozone, and the relationship of ozone to
meteorological conditions. A more extensive review of these topics is provided by NRC
(1991).
Ozone is a trace constituent of the earth's atmosphere that is produced in small
amounts through natural processes. In the higher atmospheric region known as the
stratosphere, a layer of relatively high concentrations (peaking at a mixing ratio of about 12
parts per million or ppm) exists at elevations ranging between about 25 to 45 km. This
naturally occurring stratospheric ozone layer serves the vital function of shielding the earth's
surface from harmful ultraviolet radiation emitted by the sun. Ozone is also found in smaller
amounts within a few kilometers of the earth's surface (the lower troposphere). In remote
locations far removed from human influence, these concentrations are quite small (roughly
0.02-0.04 ppm) and of little concern. In and around urban areas, however, relatively high
near-surface concentrations are often produced (as high as 0.2-0.3 ppm in Los Angeles, the
city with the most severe ozone problem in the U.S.). Exposure to concentrations
encountered within and downwind of many major U.S. urban areas is known to produce
adverse health effects in humans and is harmful to plant life as well. The following
discussion focuses on the characteristics of these elevated tropospheric ozone levels.
Tropospheric ozone is a secondary pollutant. That is to say, it is not emitted directly
into the atmosphere. Instead, the elevated ozone concentrations found within and around
major urban areas are the result of a complex series of photochemical (light driven) reactions
involving two types of precursor gases: nitrogen oxides (NOX) and volatile organic
compounds (VOCs). Both NOX and VOCs are emitted by transportation and industrial
sources. VOCs are emitted from automobiles, chemical manufacturing facilities, dry
cleaners, paint shops, and other sources where solvents are used. NOX is emitted when fuel
is burned at high temperatures such as in automobiles or at stationary sources such as utility
power plants and industrial steam boilers. Motor vehicles are the single biggest source
category of ozone precursors in many urban areas. Natural sources of VOCs and NOX have
also been identified. Under certain conditions, potentially significant quantities of VOCs are
emitted by plants, and NOX may be emitted as a metabolic byproduct of organisms living in
the soil. In recent years, much attention has focused on the contribution of such biogenic
VOC emissions to the ozone problem in certain parts of the U.S.
Ozone is a reactive gas that can cause health problems because it damages lung tissue,
reduces lung function, and sensitizes the lungs to other irritants. Scientific evidence indicates
that ambient levels of ozone affect not only people with impaired respiratory systems, such
as asthmatics, but healthy adults and children as well. Exposure to ozone for 6 to 7 hours at
relatively low concentrations has been found to significantly reduce lung function in normal,
3-1
-------
healthy people during periods of moderate exercise. This decrease in lung function often is
accompanied by such symptoms as chest pain, coughing, nausea, and pulmonary congestion
(EPA, 1993). Studies continue on what length of exposures under what circumstances and to
what concentration levels are harmful to humans. Although these studies provide new
insights into the effects of exposure to ozone, the ozone design value study mandated by the
1990 CAAA, which is the focus of this report, is based on the current NAAQS. The ozone
design value study is not intended to address issues related to the degree of health protection
afforded by the current NAAQS.
CHEMISTRY OF OZONE FORMATION
Although a detailed description of the chemistry of ozone formation is beyond the
scope of this section, a brief discussion of the basic chemical mechanisms involved is
provided to assist the reader in understanding the factors that determine the magnitude and
distribution of ozone concentrations and hence ozone design values.
The chemistry of ozone formation is described in detail by Whitten (1990); a general
review can also be found in the recent National Research Council Report (NRC, 1991). The
brief and simplified description given here presents the most important reactions in ozone
formation. First, ozone is created by the photolysis of NO2, as follows:
NO2 + h»> -» NO + O (1)
O + O2 + M -* O3 + M (2)
where hi> represents the ultraviolet light energy input and M is an inert compound. At the
same time, however, ozone is destroyed by the NO created in reaction 1:
O3 + NO ^ NO2 + O2 (3)
Under steady-state conditions, ozone concentrations would not increase, since one ozone
molecule would be created and one would be destroyed for each NO2 molecule photolyzed.
However, ozone accumulates in the atmosphere because another set of reactions involving
VOCs can preclude reaction 3 by converting the NO of reaction 1 to NO2 (which then
undergoes photolysis, leading to additional ozone formation):
"VOC" + OH -* RO2 (4)
RO2 + NO -» RO + NO2 (5)
where R is an organic fragment or hydrogen.
3-2
-------
Finally, NO2, rather than being photolyzed as in reaction 1, may be removed from further
atmospheric reactions by the following reaction:
NO2 + OH -» HNO3 (6)
An important consequence of these reactions is that reductions in VOC concentrations
will always result in reduction of the rate of ozone formation and the peak ozone
concentration in a manner similar to that illustrated in Figure 3-1. However, reductions in
NOX concentrations will either decrease or increase the ozone peak at a specific location.
The determining factor is the relative probabilities of the reaction of VOC with the hydroxyl
radical (the presence of which requires sunlight) and the reaction of NO2 with the hydroxyl
radical. Reactions of VOC with OH lead to the formation of the RO2 radical, which, in
turn, converts the NO portion of the NOX to NO2 (reaction 5); it is the photolysis of NO2
that leads to ozone formation (reactions 1 and 2). However, this same photolysis reaction
(reaction 1) also yields NO, which, in turn, might remove the ozone just formed (reaction 3)
unless further RO2 is available to sustain the reactions leading to ozone formation (reactions
5, 1 and 2). Moreover, when NO2 reacts with OH to form HNO3 (nitric acid), NO2 is
removed from the ozone formation process (reaction 6).1
One approach to understanding the complex behavior of NOX reductions toward ozone
formation, therefore, is to consider the competition between VOC and NO2 for hydroxyl
(OH) radicals, which is represented by reactions 4 and 6, restated again below with their rate
coefficients:
"VOC" + OH -» RO2 #! = 3,100 ppmC1 min'1 (4)
NO2 + OH ^ HNO3 *2 = 17,000 ppm'1 min'1 (6)
The rate coefficient for reaction 4 represents a VOC-weighted average for a VOC mix
considered representative of a typical urban atmosphere, i.e., one containing approximately
56 percent paraffinic carbon, 7 percent olefinic carbon, and 21 percent aromatic carbon
(Baugues, 1986).2 It is evident from the rate coefficients for reactions 4 and 6 that the
'it is also useful to note that while VOC, NOX and sunlight are required to generate ozone,
only sunlight (energy) and VOC are consumed when ozone is formed. The amount of ozone
formed is proportional to the NO2:NO ratio; VOC and sunlight govern the magnitude of this
ratio. The loss of NOX (e.g., to HNO3) does not lead to ozone formation.
2For an atmosphere that contains a greater fraction of paraffinic carbon (e.g., alkanes), the
weighted-average rate coefficient for reaction 4 would be less, causing the VOC:NOX ratio, which
balances the probabilities of reactions 4 and 6, to be greater. Similarly, an atmosphere that
contains more olefinic carbon or aromatic carbon would have a larger weighted-average rate
coefficient, causing the VOC:NOX ratio that balances the probabilities of reactions 1 and 2 to be
smaller.
3-3
-------
«
o
\
\ —
\ x' — s
\ NO NQ /
\*. (ROG reduced/
\ O
\ /'•-• •''
K /;x .
' ><
--" ^-^
>
\ • N02 (ROG reduced) A
\ -JPZ ^ /-
^•-"" 7
/' \ /
\ / 0;
\ /
^^ -""
^-^•^•C^-.
03 (ROG reduced )x
Time
Figure 3-1. Typical base ozone chemistry evolution profiles for NO, NO2, and ozone, and
ozone chemistry with reduced reactive organic (ROG) emissions.
3-4
-------
probability per unit time that the NO2 molecule reacts with OH is approximately 5.5
( = 17,000/3,100) times greater than the probability that "average" VOC molecule reacts with
OH. Thus, when the VOC concentration is approximately 5.5 times the NOX concentration,
the probability per unit time that VOC reacts with OH is equal to the probability per unit
time that NO2 reacts with OH.
From the foregoing discussion, it can be seen that when the VOC:NOX ratio is
substantially greater than 5.5 reaction 4 is favored. Under these conditions, a shortage of
NO exists that can be oxidized by the RO2 radicals thus formed, such that the probability that
these radicals react with each other (to form peroxides) is greater than the probability that
they react with NO to produce NO2, and thus ozone. Under these conditions-large
VOC:NOX ratios—ozone production is limited by the amount of NOX available; thus
reductions in NOX concentrations lead to reductions in ozone concentrations. However,
when VOC:NOX ratios are approximately 5.5 or less, the ubiquitous NOX retards ozone
formation. Under these conditions, NO2 reacts faster with the OH to form nitric acid than
VOC with OH to eventually form ozone. Reduction of NOX concentrations under these
circumstances increases the efficiency of ozone formation and accelerates the sequence of
chemical reactions, which, in turn, can result in increases in the peak ozone concentration.3
It should be noted that the influence of changes in NOX emissions on ambient ozone
concentrations in urban areas is subject to numerous factors and can only be predicted by
application of photochemical grid models.
Before deriving any generalizations from this brief description of the dependence of
ozone formation on VOC and NOX levels, a refinement must be added to the foregoing
discussion. The "critical" VOC:NOX ratio also depends on the duration of the photolytic
period (the period in the day with sunlight). When this is taken into account, the critical
VOC:NOX ratio that governs whether peak ozone levels will increase or decrease when NOX
reductions occur changes from approximately 5.5 to approximately 11 (which is
approximately equal to the prevailing ratio observed between 6:00 a.m. and 9:00 a.m. at
monitoring sites in the Los Angeles basin).
In summary, both theory and experimental evidence show that reductions in VOC
concentrations will always reduce the rate of ozone formation and decrease ozone peaks.
However, reductions in NOX concentrations may cause ozone levels to go up or down,
depending on the ambient VOC:NOX ratio.
3 The efficiency of ozone formation is increased because the reaction rate that removes the
OH radical (reaction 6) is reduced, resulting in more hydroxyl radicals being available to produce
RO2 radicals to oxidize NO to NO2, which in turn leads to ozone formation. See Figure 3 in
Chock and Heuss (1987), which illustrates the location of the 5:1 NMHC:NOX ratio on so-called
ozone isopleth curves.
3-5
-------
DISTRIBUTIONAL CHARACTERISTICS OF TROPOSPHERIC OZONE
Monitoring Ambient Ozone Concentrations
Routine observations of urban ozone concentrations were begun in Los Angeles in the
1950s. Today, most major urban areas are served by networks of monitors operated by state
and local government agencies and various federal agencies that routinely report hourly
ozone concentration measurements. Data collected at most of these monitoring sites are
reported to EPA's Aerometric Information and Retrieval System (AIRS) on a regular basis.
AIRS serves as a national archive of ambient air quality data.
Each monitoring site is classified into one of three categories: NAMS, SLAMS, or
SPMS. The NAMS, or National Air Monitoring Stations, were established to ensure a long-
term national network for urban-area-oriented ambient monitoring information and to provide
a systematic, consistent database for air quality comparisons and trends analysis. The
SLAMS, or State and Local Air Monitoring Stations, were established to allow state or local
governments to develop monitoring networks tailored more to their immediate ambient air
monitoring needs. The SPMS, or special purpose monitoring stations, are stations used by
state or local agencies to fulfill specific, short-term monitoring needs. Figure 3-2 shows the
distribution of NAMS, SLAMS, and other ozone monitor types across the U.S. As of 1990,
a total of 812 ozone monitors in 467 counties reported data to the AIRS. A total of 590 sites
reported data in each year from 1988 to 1990; a total of 471 sites reported data in at least 8
of the 10 years from 1981 to 1990 (EPA, 1991).
Spatial Distribution
Since ozone is formed over a period of hours from a mixture of primary pollutants
that are emitted by numerous widespread sources, its spatial distribution is more uniform
than that of many primary pollutants such as carbon monoxide. Unfortunately, information
on typical spatial gradients in ozone concentrations is limited due to the sparsity of the
monitoring network. Of course, concentration gradients decrease as one moves away from
major precursor source regions as a result of mixing within the atmosphere.
In discussing the spatial distribution of ozone, it is useful to conceptualize a series of
concentric rings encompassing areas of different sizes. Such a conceptual model of ozone
distribution—adapted from a recent National Research Council Report (NRC, 1991) in which
five canonical regions are identified—is defined as follows:
Urban Core: Comprises the primary business/commercial district of an urban area;
dominated by primary emissions from stationary sources with some mobile source
impacts. Ozone concentrations are depressed as a result of NO scavenging and young
air mass age.
3-6
-------
* - NRMS
a - SLRMS
Figure 3-2. Ozone monitoring network in the continental United States, 1990.
Urban Perimeter: Includes the heavily populated area immediately adjacent to the
urban core; typically characterized by major transportation arteries and high mobile
source emission densities. This area experiences the primary ozone impacts resulting
from urban core precursor emissions.
3-7
-------
Mesoscale Transport Region: The entire region affected by emissions from one or
more urban core areas. This region may be defined by topographical boundaries such
as mountains or shorelines. In some cases, it may be defined by the outer perimeter
of a major metropolitan area encompassing several urban cores. In other cases, the
outer boundary of this region may not be well defined. Under the sluggish
atmospheric circulation systems typically associated with high ozone episodes,
pollutants emitted in the urban core can stay within the mesoscale transport region for
several days at a time.
Synoptic Transport Region: Regions associated with long-range transport of ozone
and precursors usually controlled by the synoptic-scale atmospheric flows associated
with high and low pressure systems. These systems typically control areas measuring
several hundred to a thousand kilometers across—large enough to transport polluted
air masses a considerable distance.
Global Background Region: Remote areas far removed from the direct influence of
major source regions that are representative of planetary-scale background conditions.
Since photochemical pollutants in urban air masses are relatively short-lived,
planetary-scale background ozone concentrations are primarily a function of biogenic
precursor emissions and stratospheric ozone intrusions (see NRC, 1991 and Lefohn et
al., 1992).
Rural locations in the eastern U.S. occasionally experience extended periods of
widespread elevated ozone concentrations. On the other hand, western rural ozone
concentrations tend to be much lower (NRC, 1991). Exceedances of the NAAQS are
generally confined to areas within the mesoscale transport region associated with major urban
areas. It is these regions, therefore, that are the primary focus of this report; information on
rural ozone can be found in numerous references (e.g., Lefohn et al., 1992, and references
therein).
On the basis of both routine ozone observations and tracer experiments, it has been
demonstrated that there can be a significant impact of precursor emissions from one urban
core on downwind population centers within either the same mesoscale or synoptic-scale
transport region but which may be in different topographically or politically defined air
basins. For example, it is generally recognized that transport along the urban corridor
stretching from Washington, D.C. to New York and Boston contributes significantly to the
severity and geographic extent of ozone episodes along the eastern seaboard (Pollack et al.,
1988; Cleveland et al., 1976; Wackter and Bayly, 1987; Samson and Shi, 1988; and Possiel
et al., 1990). Tracer experiments conducted in California have identified interbasin transport
there as well (CARB, 1990).
Within the outer perimeter of a large urban area, distributions of daily maximum
ozone concentrations can vary markedly from one location to the next. For example, Figure
3-3 shows the year-to-year variations in the annual second highest daily maximum 1-hour
concentrations at monitoring locations in the Los Angeles metropolitan area. On this figure,
3-8
-------
a different symbol (letter or number) is used to identify each individual monitoring site.
Missing symbols for specific monitoring sites may indicate that either (1) the monitoring site
stopped sampling during this period or (2) its symbol may be hidden from view, because
only one symbol can be displayed at a given concentration level. In the westernmost
(coastal) parts of the Los Angeles basin, average concentrations are low while in the more
inland parts of the basin (Glendora/Pomona, downwind of the major urban core source
regions), concentrations are higher and variability is greater. Figure 3-4 reveals that the
spatial variability among monitoring sites in the basin is similar for the 95th percentile
concentrations, but the magnitude of the annual differences is smaller than that found for the
annual second maximum concentrations. Figures 3-5 through 3-14 present these same two
ozone statistics for five additional Consolidated Metropolitan Statistical Areas (CMSAs);
Boston, Chicago, Dallas, Houston, and New York. As these figures illustrate, the relative
variation in ozone concentrations recorded among monitoring sites throughout these large
urban areas can be as great as or greater than the year-to-year variation in ozone
concentrations recorded at a particular monitoring location.
Spatial variations in ozone concentrations at smaller, sub-metropolitan length scales
are not well defined in many areas due to the sparsity of ozone monitors. Ambient
monitoring data from the Los Angeles basin indicate that concentrations are fairly uniform
over distances on the order of 10 km. However, ozone concentrations can be locally
depressed immediately downwind of a NOX source due to NO scavenging (NRC, 1991).
Because ozone design values are estimated individually for each monitor in an area, and the
maximum value is used to determine the nonattainment category of the entire area, key
concerns are (1) whether the monitoring network is sufficiently dense and monitors are
appropriately located to represent air quality over the area in question and (2) how spatially
uniform are design value estimates across metropolitan areas. Figures 3-15 and 3-16 show,
respectively, the spatial variability in 1987-89 ozone design values at all monitoring sites in
the four northcentral states of Indiana, Illinois, Michigan, and Wisconsin and for the four
northeastern states of Connecticut, Massachusetts, New York, and Rhode Island. As
illustrated, design values can vary from levels near the standard to levels near 0.20 ppm at
sites across these states.
Temporal Distribution
Seasonal Patterns
Since ozone is formed through photochemical processes, sufficient sunlight is required
to produce appreciable quantities. In the latitudinal band covering most of the continental
U.S., wintertime solar input is low and cool temperatures reduce reaction rates. As a result,
ozone concentrations remain low. It is only between late spring and early fall that the sun is
high enough in the sky for enough hours each day and temperatures are warm enough to
generate ozone concentrations that exceed the NAAQS. In southern cities such as Los
Angeles and Houston, however, exceedances can occur during any month, although they are
much more limited during the winter. Several other factors contribute to the summertime
3-9
-------
0.44
0.43
0.42
0 41
0.40
O.J9
0 58
0.37
O.It
0.35
0.34
0.35
0.32
0.31
0.30
0.29
0.28
0.27
0.26
0.2S
0.24
fl.'Z
0.22
0.21
0.20
a. 11
0.18
0.17
o.u
0.15
0.14
0.13
0.12
0.11
0.10
0.09
0.04
0.07
0.06
0.05
0.04
0.03
0.02
0.01
0.00
4
>
3
J
E
7
C
C
B
0
H
0
9
J
0
it
V
A
B
K
S
•
*
f
f
H
S
Z
a
A
U
F
J
9
H
J
7
0
A
0
»
B
0
»
S
M
•
V
f
•
4
3
A
9
G
J
E
II
7
A
0
V
H
•
«(
Z
P
a
p
D
3
a
F
X
J
N
N
7
N
9
t
A
S
+
0
K
0
F
P
M
I
«
P
4
J
U
3
G
F
9
T
A
C
7
1
K
M
0
P
«
»
*
H
t
P
*
J
3
I
T
F
A
J
1
9
R
C
t
H
K
D
C
P
•
+
H
V
P
A
4
3
X
U
C
7
J
N
F
H
9
J
4
0
D
S
1
H
P
F
4
3
I
F
e
A
7
C
9
*
C
0
H
H
S
K
H
P
V
T
4
3
I
U
X
A
M
7
4
2
D
C
0
N
H
H
S
*
P
t
I
X
E
4
I
U
J
I
N
F
C
C
U
7
2
H
0
M
K
0
L
S
4
0
4
2
I
#
N
9
7
J
U
A
4
D
T
L
K
5
«
0
Figure 3-3. Second highest daily maximum 1-hour ozone concentration by year for all sites
in the Los Angeles-Anaheim-Riverside, CA CMSA.
PCT95
0.33
o.52
0.31
0.30
0 29
0.28
0.27
0.24
0.25
0.24
0.23
0.22
0.21
0.20
0.19
0.18
0.17
O.U
0.15
0.14
0.13
0.12
0.11
0.10
0.09
o.os
0.07
0,04
0.05
0.04
0.03
0.02
0.01
0.00
Z
8 9
i
f *
e N
a
F J
7 E
A
I 7
9 0
v 9
H 0
J Y
R A
A S
• U
D 0
M K
K P
P
P
*
4
3
X
N
G
E
F
B
7
0
9
R
H
A
0
S
M
a
p
0
4
3
6
J
F
7
J
N
N
9
2
A
U
0
F
V
I
P
*
P
4
3
I
G
+
F
1
C
7
9
A
0
0
C
K
S
p
p
4
X
3
0
G
N
F
7
1
9
R
0
6
0
K
V
P
A
4
4
3
I .
G 3
J
G
f
7
C
J
9
V R
0 t
K <
H D
P H
K
F
T
4
3
I
J
Z
7
A
2
J
9
t
0
•
K
H
K
E
3
X
o
F
C
7
J
2
C
4
0
U
K
S
*
4
X
J
3
N
I
2
7
J
Q
4
V
T
E
K
0
Figure 3-4. 95th percentile ozone concentration by year for all sites in the Los
Angeles-Anaheim-Riverside, CA CMSA.
3-10
-------
SCOK
0.200
0.050
0.000 «
I
Figure 3-5. Second highest daily maximum 1-hour ozone concentration by year for all sites
in the Boston-Lawrence-Salem, MA-NH CMSA.
PCTM
O.U
0.00 »
I
1WO 1M1 1982 1961
1W5 1W* 1M7
1M« I'M
Figure 3-6. 95th percentile ozone concentration by year for all sites in the Boston-
Lawrence-Salem, MA-NH CMSA.
3-11
-------
SECMX I
0.225 «
o.ozs
N
L
X
F
J
H
H
«
U
»
7
S
B
Z
A
1
1WO
0
H
X
a
j
s
2
V
B
F
B
0
L
1WI
0
0
c z
X
i
X 5
9 X B
E F
H 4 I
J K T
A H X H
J C 0 C
H 01
K H F E
Z A H V
B I i
0 C I
T t
V T
II B
A
1WZ IMS 1W> IMS I'M 1M7
P
0
i
X
s
N
I
X
1
H
B I
V H
T D
B 1
C
I
t
r
i
i«« i*et
i
p
q
H
C
I
1
B
*
1«0
Figure 3-7. Second highest daily maximum 1-hour ozone concentration by year for all sites
in the Chicago-Gary-Lake County, IL-IN-WI CMSA.
PCT95
0.1 to
K 0
H
* J M
F 9 F
> X 0
7 Z K
a B
3 A I
F Z
S
A
z e v
p
H
E K
e x
0 L
I I
s
a F
B C
I T
0
i
0
X
4
K «
) 5
« H
H I
0 B
4 T
F
a
X
K
S
p
4
101
I 0 P
F H
H
B H
T a
E
H
1
t
i a
F
0.00 •
I
Figure 3-8. 95th percentile ozone concentration by year for all sites in the
Chicago-Gary-Lake County, IL-IN-WI CMSA.
3-12
-------
SECMX
0.18
0.17
O.lt
a.it
0.14
0.15
0.12
0.11
0.10
0.0*
0.0«
0.07
0.0*
0.05
0.04
0.01
0.02
0.01
0.00
1W1 1W2
Figure 3-9.
Second highest daily maximum 1-hour ozone concentration by year for all sites
in the Dallas-Ft. Worth, TX CMSA.
fCT« I
I
0.130
0.125
0.120
0.115
0.110
0.105
0.100
0.095
O.OM
0.085
0 080
0.07S
0.070
0.065
0.040
0.055
0.050
0.045
0.040
0.015
O.OiO
0.025
0.020
0.015
O.OIO
0.005
0.000
Figure 3-10. 95th percentile ozone concentration by year for all sites in the Dallas-Ft.
Worth, TX CMSA.
3-13
-------
StCMX
0.30
0.25
0.20
0.15
0.10
0.05
0.00
"
S
s
a
u a
H
K 0
HO 9 T
r » H
* 0 F A L 5
E 0 D 9 8 »
F I UNA
63 51
7 U S DA P
E 7 I » J J
3
1 T 2 t
0 L 8 4 T
1 1 F T
I E 7 Z •
1
0
R S
P 2
o 7 a s
2
C J
H R »
J
T H
A T
»
D J N
N L D
L N t
t a H
A
1980 1W1 1982 1983 1984 1985 US* 1M7 1988 1W« 1990
Figure 3-11. Second highest daily maximum 1-hour ozone concentration
in the Houston-Galveston-Brazoria, TX CMSA.
PCT 95
0.1*
0.14
0.12
0.10
0.08
O.Oi
0.04
0.02
0.00
c v
B
F 0 S K
E 0 K 9 »
I 9 7 M 9 S L
3
I 33
1 » 7 V 0 S 0
4
F E 1 8 8 J 8
1 U 1 3 2 2
07 *8 2
A
3
2 S
1
by year for all sites
0 9
9 T D
A H N
2 9 «
a 2
1980 1981 1982 1983 19O* 1985 198* 1987 1988 1989 1990
Figure 3-12. 95th percentile ozone concentration by year for all sites in the Houston-
Galveston-Brazoria, TX CMSA.
3-14
-------
SECMX
0.50
0.00 <
I
1M3
Figure 3-13. Second highest daily maximum 1-hour ozone concentration by year for all sites
PCTM I
0.225 »
in the New York-Northern New Jersey-Long Island, NY-NJ-CT CMSA.
p.000 »
I
Figure 3-14. 95th percentile ozone concentration by year for all sites in the New
York-Northern New Jersey-Long Island, NY-NJ-CT CMSA.
3-15
-------
OZONE
Design Value, 1987-89
Figure 3-15. Ozone design values at all sites in Indiana, Illinois, Michigan, and Wisconsin,
1987-89.
OZONE
Design Value, 1987-89
Figure 3-16. Ozone design values at all sites in Connecticut, Massachusetts, New York, and
Rhode Island, 1987-89.
3-16
-------
ozone maximum, including increased biogenic emissions, increased evaporative emissions
from motor vehicles and other sources, and increased fuel combustion required to power air
cooling systems that are associated with higher temperatures in some locations. To account
for this seasonal ozone pattern and to accommodate differences in local climates, EPA has
designated specific "ozone seasons" for each State consisting of a contiguous set of months
during which minimal ambient air quality monitoring requirements must be met (see Table 3-
1). In southern locales such as Los Angeles and Houston, the official ozone season spans all
12 months, while in northern states such as Montana, the monitoring season spans only the
summer months June through September.
Diurnal Patterns
Ozone concentrations in most locations follow a predictable pattern over the course of
a day. Maximum concentrations occur in the afternoon when solar radiation peaks and the
morning emissions have had sufficient time to react and build up a reservoir of ozone. In
some cases, the afternoon maximum may be further enhanced by the downward transport of
ozone from aloft as the layer of well-mixed air near the ground thickens during the day in
response to increased solar heating of the earth's surface. The ozone being transported down
may have formed the previous day and become trapped above the mixed layer during the
evening as the amount of mixing decreased after sunset or it may have been transported aloft
from other locations overnight.
An ozone minimum is typically observed at night when there is no photochemical
production and ozone is lost both through dry deposition (i.e., reaction on surfaces such as
vegetation) and reaction with freshly emitted NO. Near strong nighttime NO source regions,
sufficient NO may exist to essentially eliminate all ozone in the shallow nighttime mixed
layer.
Year-to-Year Variation
Daily maximum ozone concentrations vary widely from one day to the next primarily
as a result of changes in weather conditions. Given similar precursor emissions, the basic
differences between days when ozone concentrations are average or below average and days
when concentrations are high (i.e., episode days) are in the prevailing meteorological
conditions. High ozone concentrations are likely to occur with low wind speeds, elevated
temperatures, intense solar radiation (i.e., no cloud cover), shallow mixing depths, and the
wind patterns that bring, keep, or return high background concentrations to the region. In
some years, such meteorological conditions occur more frequently and with greater intensity
than in others, leading to a greater number of high ozone days even if precursor emission
levels do not differ significantly from those in other years (see Figures 3-3 through 3-14).
Thus, design values determined from a single year of data vary in accordance with weather
conditions that occurred during the year in question and may or may not be representative of
design values that can be expected to occur in the future, even in the absence of any
3-17
-------
TABLE 3-1. Ozone monitoring season by state.
State
Alabama
Alaska
Arizona
Arkansas
California
Colorado
Connecticut
Delaware
District of Columbia
Florida
Georgia
Hawaii
Idaho
Illinois
Indiana
Iowa
Kansas
Kentucky
Louisiana
Maine
Maryland
Massachusetts
Michigan
Minnesota
Mississippi
Missouri
Montana
Nebraska
Nevada
New Hampshire
New Jersey
New Mexico
Begin Month
March
April
January
March
January
March
April
April
April
January
March
January
April
April
April
April
April
April
January
April
April
April
April
April
March
April
June
April
January
April
April
January
End Month
November
October
December
November
December
September
October
October
October
December
November
December
October
October
October
October
October
October
December
October
October
October
October
October
November
October
September
October
December
October
October
December
Continued
3-18
-------
TABLE 3-1. Concluded.
State
New York
North Carolina
North Dakota
Ohio
Oklahoma
Oregon
Pennsylvania
Puerto Rico
Rhode Island
South Carolina
South Dakota
Tennessee
Texas AQCR 4, 5, 7, 10, 11
Texas AQCR 1, 2, 3, 6, 8, 9, 12
Utah
Vermont
Virginia
Washington
West Virginia
Wisconsin
Wyoming
American Samoa
Guam
Virgin Islands
Begin Month
April
April
May
April
March
April
April
January
April
April
June
April
January
March
May
April
April
April
April
April
April
January
January
January
End Month
October
October
September
October
November
October
October
December
October
October
September
October
December
October
September
October
October
October
October
October
October
December
December
December
3-19
-------
precursor emission trends. To some extent, basing design values on three years of data
instead of one eliminates some of the meteorological variability, but a single unusual year
such as 1988 can still strongly affect the three-year value. Numerous procedures designed to
take into account the meteorological component of ozone concentration variability have been
proposed. These procedures are reviewed, and a more detailed description of meteorological
influences on ozone is provided, in Section 7.
Other Technical Issues
As pointed out by Heuss and Chock (1991), year-to-year variations in the number of
exceedance days can introduce significant uncertainty into attainment determinations and
design value calculations. For locations that typically record no more than a handful of
exceedances each year, the number of exceedances can vary widely from one three-year
period to the next even if emissions remain stable. However, an area just in attainment will
only have a 23 percent chance of being differently classified in two consecutive years,
according to one reviewer's calculations using a Poisson approximation. In theory, these
areas could jump in and out of attainment of the NAAQS (i.e., have fewer than or greater
than three exceedances in three years) as each new year is incorporated into the running
three-year averaging period used to determine the attainment status. In practice, however,
areas do not routinely jump in and out of attainment after each new year because (1) there
has been a historical pattern of a dominant peak year and (2) that peak year remains in the
attainment test period for three years. Javitz (1980) calculated that an area that is just in
attainment (i.e., has a true expected exceedance rate of exactly one per year) will experience
four or more exceedances (and thus be falsely designated nonattainment according to the
attainment criteria described in Appendix H) in 35 percent of all three-year periods. In the
past, this has generally not been a major issue since most areas were well above or well
below the standard. Of course, this percentage decreases as the true expected exceedance
rate drops below one per year. It should also be noted that an area that is just out of
attainment (i.e., has a true expected exceedance rate greater than or equal to 1.05) will
experience an even more significant number of periods during which the three-year average
exceedance rate turns out to be less than 1.05, thus resulting in a false determination of
attainment.4 Using a Poisson approximation, this probability is approximately 61 percent.
Several investigators (Chock and Heuss, 1987; Heuss and Chock, 1991; Chock, 1989;
Fairley and Blanchard, 1991) have pointed out that more stable seasonal summary statistics
can be formulated, such as the seasonal mean daily maximum concentration or the 90th or
95th percentile of the daily maximum concentration (i.e., the concentration exceeded on 10
percent or 5 percent, respectively, of days during the season). These statistics are based on a
larger number of daily observations and are therefore less variable. However, as illustrated
4 Appendix H includes a rounding convention that requires the estimated expected exceedance
rate to be specified to within one decimal place. Thus the minimum estimated expected
exceedance rate required for nonattainment is 1.1, which corresponds to a true expected
exceedance rate of 1.05 or greater.
3-20
-------
by Figures 3-4, -6, -8, -10, -12, and -14, these upper percentiles also exhibit relatively large
year-to-year variation in concentration levels. It should also be noted that these other
indicators bear no direct relationship to design values for the current ozone NAAQS.
Therefore, use of these indicators as design values for ozone nonattainment area designations
and classifications would require revision to the current NAAQS and to Section 181 (a) of
the CAAA of 1990. For example, Chock (1989) suggested a revised standard requiring the
95th percentile to be less than or equal to 0.095 ppm. The design value for this standard
would be the 95th percentile, which could be estimated in a number of different ways.
According to Chock, this standard would be "equivalent" in some sense to the current
NAAQS. As pointed out by Curran and Freas (1991), however, "equivalency" can be
defined in many ways and does not mean that the two standards would result in identical
attainment/nonattainment determinations in all cases. In other words, there is no
deterministic, one-to-one relationship between design values for the current NAAQS and the
95th percentile that could be used to calculate one given the other. The two design values
can only be linked via statistical relationships, which may be inadequate for some locations.
If a universally applicable statistical relationship can be shown, then a function of the 95th
percentile, rather than the 95th percentile itself, might serve as a useful estimated design
value for the current NAAQS.
SUMMARY
This section has provided background information on technical issues associated with
the tropospheric ozone problem. The chemistry of ozone formation was described as well
as its relationship to emission control strategies. The spatial and temporal characteristics of
ambient ozone data were presented and their consequences for estimating ozone design values
was discussed.
The past decade has seen large year-to-year variability in ozone concentrations.
However, the relative variation in ozone concentrations recorded among monitoring sites
throughout large urban areas can be as great as or greater than the year-to-year variation in
ozone concentrations recorded at a particular monitoring location. Spatial variations in ozone
concentrations at smaller, sub-metropolitan length scales are not well defined in many areas
due to the sparsity of ozone monitors.
Finally, the issues raised by Chock and other researchers concerning the attainment
test for the current ozone NAAQS and the use of alternative indicators or standards were
described.
REFERENCES
CARB. 1990. "Assessment and Mitigation of the Impacts of Transported Pollutants on
Ozone Concentrations within California." California Air Resources Board,
Sacramento, California.
3-21
-------
Clean Air Act Amendments, P.L. 101-549, Nov. 15, 1990, §183 g.
Cleveland, W. S., B. Kleiner, J. E. McRae, and J. L. Warner. 1976. Photochemical air
pollution: Transport from the New York City area into Connecticut and
Massachusetts. Science. 191:179-181.
Chock, D. P. 1989. The need for a more robust ozone air quality standard. J. Air Pollut.
Control Assoc.. 39:1063.
Chock, D. P., and J. M. Heuss. 1987. Urban ozone and its precursors. Environmental
Science and Technology. 21(12):1146.
Curran, T. C., and W. P. Freas. 1991. "Consequences of a More Statistically Robust
Ozone Air Quality Standard." Presented at the 84th Annual Meeting & Exhibition of
the Air & Waste Management Association, Vancouver, British Columbia (June 16-21,
1991).
EPA. 1979. Guideline for the Interpretation of Ozone Air Quality Standards. U.S.
Environmental Protection Agency (EPA-45 0/4-79-003).
EPA. 1981. Guideline for Use of City-specific EKMA in Preparing Ozone SIPs. U.S.
Environmental Protection Agency (EPA-450/4-80-027).
EPA. 1987. [ref. footnote 3 of EPA press release 8/27/87, Dave Cohen, press office:
"EPA's Office of Air Quality and Planning has completed it's [sic] review of 1986 air
quality monitoring data of the air pollutants ozone and carbon monoxide"]. U.S.
Environmental Protection Agency.
EPA. 1987. Metropolitan Statistical Areas (MSA) Regulatory Analysis Air Quality Data
Base. 1983-1985. U.S. Environmental Protection Agency.
EPA. 1991. National Air Quality and Emissions Trends Report. 1990. U.S. Environmental
Protection Agency (EPA-450/4-91-023).
EPA. 1993. National Air Quality and Emissions Trends Report. 1992. U.S. Environmental
Protection Agency (EPA 454/R-93-031).
Fairley, D., and C. L. Blanchard. 1991. Rethinking the ozone standard. J. Air Waste
Manage. Assoc.. 41 (7):928.
Heuss, J. M., and D. P. Chock. 1991. The role of the design value in ozone compliance.
Air Waste Manage. Assoc.. November, 1991.
3-22
-------
Javitz, H. S. 1980. Statistical interdependencies in the ozone national ambient air quality
standard. J. Air Pollut. Control Assoc.. 30:58.
Langstaff, J. E., and A. K. Pollack. 1985. "Meteorological Characterization of High Ozone
Levels: A Pilot Study of St. Louis, Missouri." Systems Applications, Inc., San
Rafael, California.
Laxton, W. G. 1990. "Ozone and Carbon Monoxide Design Value Calculations." U.S.
Environmental Protection Agency memorandum, Office of Air Quality Planning and
Standards (June 18, 1990).
Lefohn, A. S., D. S. Shadwick, U. Feister, and V. A. Mohnen. 1992. Surface-level ozone:
Climate change and evidence for trends. J. Air Waste Manage. Assoc.. 42(2): 136.
Lippmann, M. 1989. Health effects of ozone. A critical review. J. Air Waste Manage.
Assoc.. 39(5):672.
McCurdy, T., and R. Atherton. 1990. Variability of ozone air quality indicators in selected
metropolitan statistical areas. J. Air Waste Manage. Assoc.. 40(4):477-486.
NRC. 1991. Rethinking the Ozone Problem in Urban and Regional Air Pollution. National
Research Council, National Academy Press, Washington, D.C.
Pollack, A. K., T. E. Stoeckenius, J. L. Haney, T. S. Stocking, J. L. Fieber, and
M. Moezzi. 1988. "Analysis of Historical Ozone Concentrations in the Northeast.
Volume I: Main Report." Systems Applications, Inc., San Rafael, California
(SYSAPP-88/192a).
Possiel, N. C., D. C. Doll, K. A. Baugues, E. W. Baldridge, and R. A. Wayland. 1990.
"Impacts of Regional Control Strategies on Ozone in the Northeastern United States."
Presented at the Air and Waste Management Association's 83rd Annual Meeting &
Exhibition, Pittsburgh, Pennsylvania (June 24-29, 1990).
Rhoads, R. G., and D. D. Tyler. 1987. Distinction between "Ozone Design Value" and
"SIP Control Value" for Ozone. U.S. Environmental Protection Agency
memorandum (April 29, 1987).
Samson, P. J., and B. Shi. 1988. "A Meteorological Investigation of High Ozone Values in
American Cities." Department of Atmospheric, Oceanic, and Space Sciences, Space
Physics Research Laboratory, The University of Michigan, Ann Arbor, Michigan.
3-23
-------
Stoeckenius, T. E. 1991. "Development of Procedures for Estimating Effects of Emission
Control Strategies on Long-Term Ozone Concentrations in the Presence of
Climatological Changes." Systems Applications International, San Rafael, California
(SYSAPP-91/121).
Wackter, D. J., and P. V. Bayly. 1987. "The Effectiveness of Emission Controls on
Reducing Ozone Levels in Connecticut from 1976 Through 1987." The Scientific and
Technical Issues Facing Post-1987 Ozone Control Strategies. Air & Waste
Management Association, Hartford, Connecticut, November 1987.
Whitten, G. Z. 1990. "The Atmospheric Chemistry of Ozone Formation" (SYSAPP-
90/013). Systems Applications International, San Rafael, California.
Whitten, G. Z., and M. W. Gery. 1986. "The Interaction of Photochemical Processes in
the Stratosphere and Troposphere." In Effects of Changes in Stratospheric Ozone and
Global Climate. Volume 2: Stratospheric Ozone. J. G. Titus, ed. U.S.
Environmental Protection Agency.
Zeldin, M. D., and W. S. Meisel. 1978. Use of Meteorological Data in Air Quality Trend
Analysis. U.S. Environmental Protection Agency (EPA-450/3-78-024).
3-24
-------
4 OZONE DESIGN VALUE METHODOLOGIES
Appendix H to Part 50 discusses how to determine whether the standard is attained
and cites the Guideline for Inteipretation of Ozone Air Quality Standards (EPA, 1979) as
providing additional implementation details. Both Appendix H and the Guideline were issued
simultaneously with the National Ambient Air Quality Standard (NAAQS) for ozone when it
was revised by EPA in 1979. As discussed in Section 2, the guideline defines the ozone
design value in probabilistic terms as that concentration with an expected exceedance rate of
once per year. This is referred to in statistical literature as the characteristic largest value
(CLV) (Gumbel, 1958). After these discussions on the definition of a design value, the
guideline introduced several techniques that could be used to estimate design values. These
included a simplified table look-up procedure, different statistical distributions, and the use
of a conditional probability approach.
OZONE GUIDELINE DESIGN VALUE METHODS
EPA Table Look-up Method
In 1981, EPA issued the Guideline for the Use of City-specific EKMA in Preparing
Ozone SIPS, which used a table look-up approach for selecting the appropriate control
estimate as determined from photochemical modeling of peak ozone days (EPA, 1981). The
Empirical Kinetic Modeling Approach (EKMA) required that the design value be a
concentration that was actually measured in order to obtain the necessary modeling
parameters. Thus, the simple tabular procedure shown in Table 4-1 was selected because it
gave a design value that was observed, as opposed to a statistical fitting technique that could
yield a design value that was not actually measured, but rather a statistical estimate. The
tabular approach has several practical, implementation advantages as compared to more
complex statistical techniques. First, ozone design value estimates could be made quickly,
and directly, from existing ambient data reporting systems. Second, it was understandable
by control agency personnel, and the actual ambient air quality values could be reaffirmed
through existing quality assurance procedures. Third, the approach could be applied using
reported summary statistics which were readily available. This last feature was important
given the existing ambient air quality data reporting requirements in the monitoring
regulations (40CFR58). The reporting requirements for the State and Local Ambient
Monitoring Stations (SLAMS), as specified in the regulations, only require the reporting of
the four highest daily maximum 1-hour concentrations, the number of observations, and the
number of estimated exceedances in the Annual SLAMS Report. Hourly data were not
required to be reported to EPA. Thus, distribution fitting techniques that required daily
maximum data were not feasible for such sites. With the advent of newer, automated data
reporting and capture methods, most of the hourly ozone data collected in this country are
now reported to EPA.
4-1
-------
TABLE 4-1. Ozone Design Value Rank Based on Number of Years of Data
Number of Valid Years Ozone Design Value Rank
(at least 75% of days during designated ozone (daily maximum 1-hr concentration)
season)
Less than one valid year Highest daily maximum
1 year of data 2nd highest daily maximum
2 years of data 3rd highest daily maximum
3 years of data 4th highest daily maximum
During the late 1980s in support of the Clean Air Act debate, EPA issued ozone
design value lists that continued to use this tabular technique (EPA, 1987, 1988, 1989, 1990,
1991). On June 18, 1990, EPA issued a memorandum documenting the table look-up
approach as the EPA design value estimation method (Laxton, 1990). This policy
memorandum essentially maintained the status quo and documented the tabular approach
shown in Table 4-1 as the EPA design value method. The memorandum provided specific
example calculations and clarified several technical details. It restated that air quality design
values are estimated on a site-by-site basis, and that the design value for the "area" is the
highest design value estimated at any site within the area. This was the last EPA policy
statement issued prior to passage of the CAAA of 1990.
Use of Statistical Distributions
This section contains a review of important articles dealing with the fitting of
statistical distributions to air quality data. Appendix A contains a quick reference tabular
summary of each numbered reference which appears in brackets "[ ]". This section also
summarizes the parametric distributions fitted to the data, the fitting method, and the
applicability of the method to the current ozone NAAQS.
The Ozone Guideline (EPA, 1979) allows the use of fitted distributions to calculate
design values, provided that the fit is acceptable. In this case, the design value is often
estimated as the 100 x 3647365th percentile of the distribution of the daily maximum since
that value is exceeded an average of once per 365 days (assuming that the distribution is the
same throughout the year). In general, if the parametric distribution is correct, or a very
good approximation to reality, then the use of the fitted distribution is likely to lead to better
estimates of the expected number of exceedances, and of the design value, than estimates
using the empirical distribution (raw data). As in the case of the tabular design value
approach, design values estimated from ambient monitoring data by this statistical modeling
approach may be less than the level of the standard even though the area has failed the
attainment test based on estimated exceedances. However, in contrast to the tabular
4-2
-------
approach, these statistical methods can also yield design values that are greater than the
NAAQS even though the area has an expected exceedance rate less than or equal to 1.
There are two distinct approaches to fitting distributions to air quality data. In the
first, one of the standard parametric distributions, such as the lognormal distribution, is fitted
to the raw concentrations. The second approach is based on extreme value theory, which
uses the fact that if we have a large number of approximately independent measurements
with approximately the same distribution, there are only three types of limiting parametric
distributions for the highest concentrations (maximum, second maximum, etc.) and for the
upper tail of the distribution. The limiting extreme value distribution depends on two or
three unknown parameters.
Both approaches (fitting parametric distributions to raw hourly or daily
concentrations, or fitting extreme value distributions to the highest concentrations) assume
weak serial dependence and only small trends in the raw concentration data (i.e.,
approximate stationarity). Weak serial dependence means that two consecutive daily
maximum concentrations may be highly correlated, but daily maxima separated by a long
time interval are almost independent. The proviso of "small trends" means that the
distribution, and in particular the mean, of the daily maximum is almost constant in time.
Alternative approaches that directly take dependence and trends into account are described
elsewhere in literature reviews of time series and trend modeling.
The extreme value theory approach is particularly important for ozone NAAQS issues
because it focuses attention on the extreme (maximum) concentrations that are primarily
responsible for determining the design value. Furthermore, the applicability of the extreme
value approximations across a variety of raw concentration distributions, even in the presence
of serial dependence (Leadbetter et al., 1983) and small trends, is an advantage over
approaches that fit parametric distributions to the raw data or that use the observed
concentration distribution. The main limitation of the theory is the requirement for large
numbers of raw concentrations, or for only using those concentrations in the extreme upper
tail of the distribution. In particular, use of the extreme value distribution for the annual
maximum is based on applying the asymptotic theory to 365 daily maxima instead of an
infinite number (assuming a 365-day ozone season). It is possible that 365 (or the length of
the ozone season) may not be large enough for the extreme value approximation to be
sufficiently accurate since convergence to the limiting extreme value distribution is typically
very slow (see, for example, Cohen, 1982). However, Cohen (1982) demonstrated that in
many cases a much better approximation can be obtained using the three-parameter extreme
value distribution rather than the two-parameter extreme value distribution (Gumbel) even if
the Gumbel distribution is the theoretical limiting distribution. Furthermore, the fitted
distribution may give an even better approximation than the theoretical limiting distribution
using the shape parameters calculated for an assumed raw concentration distribution.
Given this literature review, including the reported goodness-of-fit results, a growing
consensus appears to favor the use of the tail exponential distribution to fit the annual
4-3
-------
maximum hourly ozone concentrations. The use of the lognormal distribution to fit the
hourly and, possibly, daily maximum hourly ozone concentrations is another popular method.
This discussion primarily focuses on these two statistical distributions. Other statistical
distributions that have been used for air quality data include the normal, gamma, Weibull,
beta, powers of normal distributions (Box-Cox transformations), and exponential
distributions. A comparison of distributions for hourly ozone data in Wisconsin ([4]) found
that the lognormal distribution fit best. A comparison for daily mean ozone concentrations in
Australia ([15]) found that the gamma distribution fit best. In any case, the selection of a
"best" statistical distribution for calculation of ozone design values may not be possible
because such a distribution probably varies according to the location studied and the time
period of interest.
Fitting Methods
The first step in fitting statistical distributions to air quality data is the selection of the
parametric family. The second step is the fitting method in which the specific distribution in
the family that best fits the available data is selected. The specific distribution is usually
defined by specific values of location, scale, and/or shape parameters. For example, the
selection of a lognormal distribution defines a family of distributions with different location
and scale parameters. The maximum likelihood method ([4, 8, 9, 10, 12, 14, 15, 16, 19,,
20, 21, 22, 25]) has certain well-known optimal properties, particularly when the number of
observations is very large. In this method the values of the parameters are selected so as to
maximize the joint density at the observed data values (very roughly speaking, the probability
of obtaining the observed concentrations). A less powerful but often simpler approach is the
method of moments ([9, 10, 16, 26]) in which the sample moments (means, variances, etc.)
are matched to the theoretical moments based on the fitted parametric form. A similar
method is the method of quantiles, in which the sample quantiles or percentiles (median, 90th
percentile, etc.) are matched to the theoretical quantiles ([1, 2, 7, 9, 10, 11, 13, 16, 17,
24]).
In many cases the fitted distribution can be transformed so that the unknown
parameters are scale and location parameters. The method of quantiles is then equivalent to
creating a probability plot for the transformed distribution and fitting a straight line. For
example, the lognormal probability plot is usually created so that one axis contains the
concentrations plotted on a logarithmic scale and the other axis contains percentages (or
probabilities) plotted on a scale that converts the percentage IQOp to the value x that is the
corresponding standard normal percentile (i.e., the value x such that IQOp percent of values
from a standard normal distribution are less than x). Essentially, the /th highest
concentration is plotted against the probability iln. Improved fits can be obtained by using
alternative plotting positions to iln (see Cunnane, 1978). The optimal plotting position for
the /th highest concentration depends on the statistical distribution to be fitted and the
properties required of the parameter estimates. Cunnane (1978) shows that replacing iln by
(/ - a)/(n + 1 - 2a), with a = 3/8 for the normal-case and a = 0.44 for the exponential
4-4
-------
or Gumbel (extreme value) distributions, gives approximately unbiased minimum mean
square error estimates for quantiles.
The method of fitting two quantiles ([1, 2, 9, 10, 24]) is based on drawing a line
through two points on the probability plot to estimate the scale and location parameters.
Some of the investigators fitted the straight line using least squares to more than two
quantiles ([13, 16]), or to all the quantiles ([7, 11, 17, 24]). An approach that uses a plot of
the distribution function rather than the inverse distribution function is described in paper 24.
In general, selection of the fitting method is less crucial to the ultimate use of the
results than selection of the parametric form. A problem that is common to all of the
methods described in the cited papers is the use of the inherent assumption that all the
observations used in the distributional fit are independent and identically distributed. (The
maximum likelihood method does not explicitly require these assumptions, but all of the
reviewed articles that used maximum likelihood methods also used these two assumptions in
deriving the likelihood function.) Alternative methods that allow for nonstationarity and
dependence are discussed in the time series and trends literature reviews. Because ozone
design values are expressed in terms of concentrations in the upper tail of the distribution
(i.e., extreme values), it is reasonable to assume that the applications that fit the
concentration distribution to the entire hourly concentration data will be less consistent with
the ozone NAAQS than are the applications that fitted the distribution to the upper tail only.
(Extreme value theory shows that the limiting distribution of the highest values depends only
on the extreme upper tail of the distribution.) However, the accuracy of the fitting method
decreases if fewer upper tail concentrations are used. Thus for the purpose of estimating
design values, the maximum likelihood method, least-squares fitting to all quantiles method,
and the method of moments should be applied only to the upper tail of the hourly
concentrations rather than the complete set of concentrations. Furthermore, it is appropriate
that the quantile method be applied to the high rather than the low quantiles.
Lognormal Distributions
The pioneering papers by R. Larsen ([1, 2]) describe how the lognormal distribution
appeared to fit air pollutant concentration data for a wide variety of cities, pollutants, and
averaging times. Furthermore, Larsen found empirically that the median and annual
maximum were proportional to powers of the averaging time, and hence was able to derive
formulae to convert medians and annual maxima between two different averaging times.
Several authors have criticized the Larsen paper. In particular, Patel ([3]) commented
on the invalidity of the independence assumption. Patel also commented on the invalidity of
Larsen's equating the expected log maximum with the log expected maximum, which Larsen
used to derive the conversion formulae.
4-5
-------
Perhaps the most crucial issue is the inconsistency of Larsen's approach caused by the
fact that if the distribution is lognormal for one averaging time, and if the concentrations are
independent, then the distribution for a longer averaging time cannot theoretically be
lognormal. One resolution of this issue is to argue as in Larsen's response to Patel ([3]) that
Larsen's results are just simply observations about the approximate behavior of air quality
data and are not intended to provide a theoretical model. A second response is found in the
paper by Bencala and Seinfeld ([17]) in which the lognormality of CO data is derived
assuming a lognormal wind speed distribution. Further, we have found arguments made in
several papers not included in this survey that the concentrations can be thought of as
products of a large number of independent variables, which implies a limiting lognormal
distribution by the central limit theorem (see, for example, some of the papers and
discussions in EPA (1974)).
A most interesting treatment of this problem appears in the article by Mitchell ([18]),
in which a lognormal distribution is shown to fit averages of a small number of independent
lognormal concentrations better than the normal distribution. One reason for this effect is
probably that although the central limit theorem implies the normality of an average of a
large number of independent lognormal concentrations, for small samples the lognormal
distribution is better at fitting the skewness of the distribution of the average. (The average
cannot be symmetric since it ranges from 0 to plus infinity.) In practice, another reason for
this may be the effect of autocorrelation: Bencala and Seinfeld ([17]) give an example to
show the approximate lognormality of the average of correlated lognormal concentrations.
One difficulty with the lognormal distribution is the problem of estimating the
moments (such as the mean) of the lognormal concentrations from the parameters of the
normal distribution fitted to the logged data. If the concentrations are lognormal, then the
obvious approach is to exponentiate the mean of the log concentrations, which gives the
geometric mean concentration. However, it is shown in [22], for example, that the
geometric mean is biased for the true mean concentration. An unbiased estimate would be
given by the arithmetic mean concentration, but this estimate does not use the assumed
lognormality and is not the best (minimum variance unbiased) estimate.
Extreme Value Theory Approximations
The theory of extreme values (see [6], [7], and the textbooks cited in the reference
section) stems from the result that if we have a large number of approximately independent
measurements with approximately the same distribution, then there are only three types of
limiting parametric distributions for the highest concentrations (maximum, second maximum,
etc.). In particular, the limiting distribution of the maximum is either the two-parameter
Gumbel distribution or the three-parameter Type n or Type ffl extreme value distribution.
This result provides theoretical support for the use of one of these extreme value theory
distributions to fit the annual maximum hourly ozone concentration distribution. Similar
extreme value distribution limits apply to the second maximum, third maximum, ... and £th
maximum, where k is small. However, it is important to keep in mind the distinction
4-6
-------
between the fah highest hourly value and the fcth highest daily maximum hourly value. The
latter value is more relevant to the present ozone NAAQS.
The papers by Georgopolis and Seinfeld ([9]) and by Johnson and Symons ([13])
present a hybrid approach in which the lognormal or Weibull distribution is fitted to the
hourly or daily maximum hourly concentrations and then extreme value theory is applied to
estimate the distribution of the annual maximum. The main problem with this approach is
that it fails to take advantage of the general applicability property of extreme value theory
under which, for a general daily maximum distribution, the limiting distribution of the
maximum is restricted to only the three extreme value distributions. This result applies even
under mild departures from the independence ([23], Leadbetter et al. (1983)) and identical
distribution assumptions. In fact, Leadbetter et al. (1983) showed that under quite strong
local dependence (between measurements at most a few days apart), the same limiting
distribution applies, although the shape parameters may be shifted. An alternative approach
that uses the general applicability property (because it does not use any specific distributions
for the daily maximum concentrations) is to directly fit the extreme value distribution to the
annual maxima.
Tail Exponential and Related Approaches
An extreme value theory approach that takes advantage of the general applicability
property is presented in several papers ([12], an important and frequently cited paper
describing original work by Breiman and others performed under contract to the
Environmental Protection Agency, and [14], [19], [20], [21], and [25]). These papers use
the result that the limiting tail distribution can only be one of three distribution functions
known as the family of Generalized Pareto Distributions (GPD—the limiting distribution of
the concentrations that exceed a given threshold as the threshold increases). The most
commonly used member of this family is the tail exponential distribution, which corresponds
to the Gumbel distribution in the first classical extreme value theory approach. The tail
exponential extreme value theory approach takes the highest 5, 10, or 20 percent of the daily
maximum concentrations and fits an exponential distribution to those concentrations above
the corresponding concentration threshold c, the 95th, 90th, or 80th percentile of the entire
daily maximum concentration data set. The exponential distribution has the probability
density a exp{-a (x — c)}. No assumption is made about the distribution below c. The
estimated value of a using the maximum likelihood method is the reciprocal of the mean
upper tail concentration. Given the tail exponential distribution, the expected number of
exceedances per year of the ozone NAAQS is 365 (0.05) [exp {-a (0.125 - c)}], assuming a
365-day ozone season, a 5 percent upper tail fit, and assuming that 0.125 exceeds the 95th
percentile of the daily maxima (i.e., the upper tail threshold). The term in the single
brackets is the probability of exceeding 0.125 given that c is exceeded, and 0.05 is the
probability of exceeding c. The exceedance threshold used is 0.125 ppm rather than the 0.12
ppm ozone NAAQS since the rounding convention in Appendix H specifies that an
exceedance occurs only if the concentration is greater than or equal to 0.125 ppm.
4-7
-------
A major advantage of the tail exponential or tail GPD approach is that much more of
the data can be used than, say, the annual maxima. The approach has general applicability
analogous to those of the classical theory. The crucial choice is the selection of the high
concentration threshold, or, equivalently, of the percentage of observations in the tail used to
fit the tail exponential (or GPD). Several authors have found fairly good results using the
upper 5, 10, or 20 percent tails. Larsen and others ([19, 20, 25]) have used an approach that
combines estimates for a range of tail percentages between 5 and 20 percent, either using a
simple average ([19]) or using a weighted average that gives greater weight to the better
fitting tail exponential distributions ([20]).
Since the GPD includes all the limiting tail distributions, the tail GPD approach may
be more robust although the application is more complex than for the tail exponential. Smith
([21]) fit the GPD to 72-hour ozone maxima that exceed the selected thresholds; such an
approach takes into account clustering of daily maxima due to very strong serial correlations.
A more theoretical study of clustering was given by Leadbetter (1991).
Conditional Probability Approach
The conditional probability approach is described in the Ozone Guidelines (EPA,
1979). This approach is an alterative to the approach of fitting a single distribution to all
data collected during the attainment period. Suppose that the attainment period is N years
(typically N = 3). A statistical distribution is selected and fitted for each year separately
using the approaches described earlier in this section. If Ff(x) is the distribution fitted to the
z'th year of data, and if each year gets equal weight, UN, then the distribution fitted to all N
years is
N
F(x) = £ FWN-
/=i
The conditional probability approach design value is the solution of the equation F(x) =
1/365, which would usually have to be solved by an iterative process.
The same approach could be applied to an even finer subdivision of the attainment
period (e.g., into quarters rather than years), which would lead to a larger value of N. One
advantage would be that ozone trends within a year could be accounted for. A disadvantage
would be that there may be insufficient monitoring data to accurately estimate Ft(x).
A more complex version of the conditional probability approach assumes that the
daily maximum ozone distribution is different for different meteorological classes and that
the frequencies of the different meteorological classes are known (either the observed
frequencies or the expected frequencies according to an appropriate model may be used). It
then follows that
4-8
-------
K
F(x) = £ F(x\class i) P(class i)
1=1
where F(x \ class i) is the estimated daily maximum ozone distribution for the fth
meteorological class and P(class i) is the relative frequency for the fth meteorological class.
The design value would be estimated by iteratively solving the equation F(x) = 1/365.
OTHER DESIGN VALUE APPROACHES
California Method for Computing Recurrence Rate Values
The tail exponential approach of Larsen and others ([19, 20, 25]) was developed
primarily to implement criteria in the 1988 California Clean Air Act specifying conditions
under which highly irregular or infrequent violations of the state ambient air quality
standards could be excluded from the attainment/nonattainment designation process. In June
1990, the California Air Resources Board determined that exceedances of the California
ozone standard expected to recur less frequently than once in seven years could be excluded.
More recently the Board determined that exceedances expected to recur less frequently than
once in one year could be excluded. The tail exponential approach was proposed as a
method of estimating the one-in-one-year or one-in-seven-year concentration. In [19], the
proposed approach used the methods of Breiman and others ([12]) described above to
estimate the one-in-seven-year concentration (based on three years of data) for approximately
166 different exponential tails, corresponding to all possible upper tail percentages from the
top 5 percent (55 values) to the top 20 percent (220 values). If the exponential tail
probability density is a exp {— a(x — threshold)} corresponding to a lOOp percent upper tail,
then the estimated one-in-seven-year concentration for that upper tail is threshold +
(l/a)log(7(365)/7) assuming a 365-day ozone season. The implementation in [19] used the
simple average of the 166 one-in-seven-year concentration estimates to give the overall one-
in-seven-year concentration estimate.
The one-in-seven-year-concentration estimate in [19] ignores the fact that the
goodness of fit of the exponential tail model varies with the size of the upper tail. The
goodness of fit is affected partly by the overall accuracy of the exponential tail model and
partly by the accuracy of the parameter estimates. The exponential tail assumption is an
approximation that can be assumed to be more accurate for the more extreme upper tails.1
The precision of the exponential tail slope parameter increases with the number of
observations in the upper tail. It follows that the goodness of fit for a given upper tail
percentage is a compromise between the competing errors of the exponential approximation
1 This assumption is based on extreme value theory, which applies if the daily maximum
concentrations can be assumed to be approximately independent and identically distributed with a
common distribution that has the appropriate exponential type tail behavior; possible common
distributions include the normal and lognormal distributions.
4-9
-------
(generally improving with lower upper tail percentages) and of the parameter estimates
(improving with higher upper tail percentages). However, McGuire (1994) pointed out that
in some cases, the higher upper tail percentiles actually produce better fits to the data than
the lower percentages.
The approach in [20] uses a weighted average of the 5 to 20 percent tail exponential
estimates of the one-in-seven-year concentration. Each tail exponential estimate of the one-
in-seven-year concentration is weighted by the inverse of the corresponding chi-square
goodness-of-fit statistic. The chi-square goodness-of-fit statistic is based on partitioning the
tail into 10 classes with equal expected frequencies and comparing the observed and expected
frequencies. Another modification in [20] of the implementation in [19] was the use of the
threshold estimate given by the average of the lowest five tail concentrations and the four
concentrations just outside the tail, instead of using the corresponding upper tail percentile
(estimated as the highest concentration not included in the upper tail). Although the ideas of
giving greater weight to better fitting models and using more than one concentration value to
estimate the upper tail percentile are intuitively sound, it should be noted that no theoretical
justification is provided in [20] for the particular procedures used.
EPA Tabular Method Extended to Multiple Years
EPA guidance specifies that the three most recent years of ambient ozone
monitoring data be used to determine compliance with the ozone NAAQS and to estimate
ozone design values. One possible way to reduce the year-to-year variability of the ozone
design value estimates is to lengthen the "data window" used with the tabular procedure.
The "number of years plus 1" rule is easily extended to more than 3 years, i.e., the design
value for 4 years of data is simply the fifth highest daily maximum 1-hour concentration.
While design value estimates based on more years of data may be more stable over
time, they also are less responsive to underlying changes in the distribution of ozone
concentrations. Nationally, the peak ozone years were 1980, 1983, and 1988. Thus, design
values based on the three-year period 1984-86 show much greater decreases in ozone
concentrations than 4-, 5-, or 6-year design values which incorporate ozone data from 1983.
Conversely, design value estimates based on a longer time period may not increase as much
as those based on fewer years of data whenever a peak ozone year occurs. If design values
are based on periods longer than three years, a decision may have to be made on the
representativeness of design values estimates from monitoring sites that had not operated for
several years.
Control Strategy Design Values
Beginning in the 1970's, air quality design values were used as the primary input to
simple air quality models, such as the "rollback model" and the Empirical Kinetic Modeling
Approach (EKMA). These models were used to estimate emission reductions needed to
attain the NAAQS and to evaluate alternative control strategy options (deNevers and Morris,
4-10
-------
1975 and EPA, 1977). This use of ozone design values diminished following the
development of more complex photochemical modeling approaches (EPA, 1991).
In some applications, design values used in estimating emissions reductions have
been adjusted to account for factors such as the level of transported ozone, or air quality
reductions expected from future control measures. Such adjusted design values have come to
be known as "control strategy values" to differentiate them from "air quality" design values
which are estimated directly from the ambient monitoring data (Rhoads and Tyler, 1987).
Recent attention has been focused on adjusting ozone design values for the year-to-year
fluctuation in meteorological conditions. Meteorological influences on ozone concentrations
and adjustment methods are discussed in Section 7. The role of transported ozone and
precursors is the subject of Section 8 of this report.
SUMMARY
The Ozone Guideline introduced several approaches that could be used to estimate
ozone design values, including a table look-up method, a graphical procedure, the use of
different statistical distributions, and the use of a conditional probability approach. No single
approach was required by the Guideline, so that the initial guidance contained a fair degree
of flexibility in how design values could be estimated. Since the early 1980's, EPA has used
a tabular approach for estimating ozone design values that have been reported in annual air
quality assessments. Prior to enactment of the 1990 Clean Air Act Amendments, EPA issued
a policy memorandum documenting the procedures used in determining both ozone and
carbon monoxide design values (Laxton, 1990). This memorandum essentially maintained
the status quo and documented the table look-up procedure summarized above as the EPA
design value estimation method. This was the last EPA policy statement on design values.
A wide variety of statistical distributions for air quality data have been investigated.
For ozone data, the tail exponential and lognormal distributions have been widely used, but
the literature does not provide definitive guidance on the best distribution to use for ozone
design value calculations. However, there appears to be a growing consensus in favor of the
tail exponential, on the basis of its general applicability and good fit, simplicity, and ease of
fitting.
4-11
-------
REFERENCES
Cohen, J. P. 1982. Convergence rates for the ultimate and penultimate approximations in
extreme value theory. Adv. Appl. Prob.. 14:833-854.
Cunnane, C. 1978. Unbiased plotting positions—a review. J. Hydrol.. 37:205-222.
deNevers, N. and J.R. Morris. 1975. "Rollback Modeling: Basic and Modified". Journal
of the Air Pollution Control Association. Volume 25, No. 9.
EPA. 1974. Proceedings of the Symposium on Statistical Aspects of Air Quality Data.
U.S. Environmental Protection Agency, Research Triangle Park, North Carolina
(EPA-650/4-74-038).
EPA. 1977. Procedure for Quantifying the Relationship Between Photochemical Oxidants
and Precursors. U.S. Environmental Protection Agency, Research Triangle Park,
North Carolina (EPA-450/4-77-021a).
EPA. 1979. Guidelines for the Interpretation of Ozone Air Quality Standards. U.S.
Environmental Protection Agency, Research Triangle Park, North Carolina (EPA-
450/4-79-003).
EPA. 1987. National Air Quality and Emissions Trends Report. 1985. U.S. Environmental
Protection Agency, Office of Air Quality Planning and Standards, Research Triangle
Park, North Carolina (450/4-87-001).
EPA. 1988. National Air Quality and Emissions Trends Report. 1986. U.S. Environmental
Protection Agency, Office of Air Quality Planning and Standards, Research Triangle
Park, North Carolina (450/4-88-001).
EPA. 1989. National Air Quality and Emissions Trends Report. 1987. U.S. Environmental
Protection Agency, Office of Air Quality Planning and Standards, Research Triangle
Park, North Carolina (450/4-89-001).
EPA. 1990. National Air Quality and Emissions Trends Report. 1988. U.S. Environmental
Protection Agency, Office of Air Quality Planning and Standards, Research Triangle
Park, North Carolina (450/4-90-002).
EPA. 1991. Guidelines for Regulatory Application of the Urban Airshed Model. U.S.
Environmental Protection Agency, Research Triangle Park, North Carolina (EPA-
450/4-91-013).
Gumbel, E.J. 1958. Statistics of Extremes. Columbia University Press.
4-12
-------
Laxton, W. G. 1990. "Ozone and Carbon Monoxide Design Value Calculations." U.S.
Environmental Protection Agency memorandum, Office of Air Quality Planning and
Standards, Research Triangle Park, NC.
Leadbetter, M.R. 1991. On a basis for extremes over threshold modeling. Stat. & Prob.
Letters. 12:357-362.
Leadbetter, M.R., G. Lindgren, and H. Rootzen. 1983. Extremes and Related Properties of
Random Sequences and Processes. Springer-Verlag, New York, pp. 127-129.
McGuire, T. 1994. Comments on draft ozone design value study report. Letter to Robert
Kellam, May 27, 1994.
Rhoads, R.G. and D.D. Tyler. 1987. "Distinction between 'Ozone Design Value' and
'SIP Control Value' for Ozone". U.S. Environmental Protection Agency, Research
Triangle Park, North Carolina.
Whitten, G. Z. 1990. "The Atmospheric Chemistry of Ozone Formation" (SYSAPP-
90/013). Systems Applications International, San Rafael, California.
4-13
-------
Numbered References
1. Larsen, R. I. 1969. A new mathematical model of air pollutant concentration,
averaging time, and frequency. J. Air Pollut. Control Assoc.. 19:24.
2. Larsen, R. I. 1971. A Mathematical Model for Relating Air Quality Measurements
to Air Quality Standards. U.S. Environmental Protection Agency.
3. Patel, N. R. 1973. Comment on "A new mathematical model of air pollution
concentration." J. Air Pollut. Control Assoc.. 23:4.
4. Holland, D. M., and T. Fitz-Simons. 1982. Fitting statistical distributions to air
quality data by the maximum likelihood method. Atmos. Environ.. 16(5): 1071-
1076.
5. Gleit, A., M. R. Leadbetter, and A. R. Manson. 1984. Estimation for lognormal
data with multiplicative errors. J. Air Pollut. Control Assoc.. 31(3):237-241.
6. Roberts, E. M. 1979. Review of statistics of extreme values with applications to
air quality data. Part I. Review. J. Air Pollut. Control Assoc.. 29(6):632-637.
7. Roberts, E. M. 1979. Review of statistics of extreme values with applications to
air quality data. Part II. Applications. J. Air Poliut. Control Assoc.. 29(7):733-
740.
8. Stoline, M. R. 1991. An examination of the lognormal and Box and Cox family of
transformations in fitting environmental data. Environmetrics. 2(1):85-106.
9. Georgopoulos, P. G., and J. H. Seinfeld. 1982. Statistical distributions of air
pollutant concentrations. Environ. Sci. Technol.. 16(7):401A-416A.
10. Mage, D. T., and W. R. Ott. 1984. An evaluation of the methods of fractiles,
moments and maximum likelihood for estimating parameters when sampling air
quality data from a stationary lognormal distribution. Atmos. Environ.. 18(1):163-
171.
11. Rao, S. T., G. Sistla, and J. Y. Ku. 1987. "Temporal and Spatial Variability of
Ozone Concentrations in the New York Metropolitan Region." Air and Waste
Management Association International Specialty Conference, Hartford, Connecticut
(November 1987).
12. Breiman, L., J. Gins, and C. Stone. 1978. "Statistical Analysis and Interpretation
of Peak Ak Pollution Measurements." Technology Service Corporation, Santa
Monica, California. Work performed under EPA contract 68-02-2857.
4-14
-------
13. Johnson, T. R., and M. J. Symons. 1980. "Extreme Values of Weibull and
Lognormal Distributions Fitted to Ambient Air Quality Data." 73rd Air Pollution
Control Association Annual Meeting, Montreal, Quebec (June 1980).
14. Curran, T. C. 1984. "Data Screening for Large Air Quality Data Sets." 77th
Annual Meeting of the Air Pollution Control Association, San Francisco, California
(June 24-29, 1984).
15. Taylor, J. A., A. J. Jakeman, and R. W. Simpson. 1986. Modeling distributions
of air pollutant concentrations—I. Identification of statistical models. Atmos.
Environ.. 20(9): 1781-1789.
16. Jakeman, A. J., J. A. Taylor, and R. W. Simpson. 1986. Modeling distributions
of air pollutant concentrations—II. Estimation of one and two parameter statistical
distributions. Atmos. Environ.. 20(12):2435-2447.
17. Bencala, K. E., and J. H. Seinfeld. 1976. On frequency distributions of air
pollutant concentrations. Atmos. Environ.. 10:941-950.
18. Mitchell, R. L. 1968. Permanence of the lognormal distribution. J. Optical
Society of America. 58:1267-1272.
19. Larsen, L. C., and R. A. Bradley. 1991. "Use of an Exponential-Tail Model to
Estimate Ozone Concentrations with an Infrequent Recurrence Rate in California."
Air and Waste Management Association 84th Annual Meeting and Exhibition,
Vancouver, British Columbia (June 16-21, 1991).
20. CARB. 1992. "Proposed Amendments to the Criteria for Designating Areas of
California as Nonattainment, Attainment, or Unclassified for State Ambient Air
Quality Standards." California Air Resources Board, Sacramento, California.
21. Smith, R. L. 1989. Extreme value analysis of environmental time series: An
application to trend detection hi ground-level ozone. Statist. Science. 4:367-393.
22. Gilbert, R. O. 1983. Statistical distributions for contaminant studies, and the
estimation of average concentrations. TRAN-STAT Statistics for Environmental
Studies. 25. Battelle Memorial Institute, Pacific Northwest Laboratory, Richland,
Washington.
23. Bennan, S. M. 1964. Limit theorems for the maximum term in stationary
sequences. Annals of Mathematical Statistics. 35:502-516.
4-15
-------
24. Curran, T. C., and N. H. Frank. 1975. "Assessing the Validity of the Lognormal
Model When Predicting Maximum Air Pollution Concentrations." 68th Air
Pollution Control Association Annual Meeting, Boston, Massachusetts.
25. Larsen, L. C. 1991. "Evaluating the Performance of an Exponential-Tail Model
for Use in Determining Ozone Attainment Designations in California." Air and
Waste Management Association Tropospheric Ozone and the Environment II
Specialty Conference, Atlanta, Georgia.
26. Mage, D. T. 1984. Pseudo lognormal distributions. J. Air Pollut. Control Assoc..
31(4): 374-376.
4-16
-------
EVALUATION APPROACHES FOR OZONE DESIGN VALUE METHODS
INTRODUCTION
Two different but complementary approaches have been taken in the comparisons of
alternative ozone design value methodologies. In the first, a database was constructed that
contained the daily maximum 1-hour ozone concentrations for all sites in the AIRS database
for the years 1980 through 1990. Using this database, design values were calculated for
alternative procedures on the same data sets and the results compared. These data, obtained
over a variety of sites, give an indication of the variability of each method and suggest the
relative biases of the methods. The drawback of this first approach is that the underlying
"true design value" is not known and the absolute biases cannot be determined.
The second approach attempts to overcome these difficulties. The ambient air quality
database provides the base data for the derivation of a time-series model. Given such a
model, large numbers of random simulations of single seasons of daily maximum ozone
values can be generated, which allow the limiting characteristic largest value (CLV), and
design values for any number of methods, to be calculated over a large number of years.
Thus, both the inherent biases and precision of alternative design value methods can be
studied using a wide variety of averaging years. These data sets have no missing values and
therefore are free from this source of error. Data can be removed from the individual years
to study the sensitivity of the methods to missing values.
There are also problems with the time-series approach. First, the study did not
develop a single time-series model equally applicable to all site-years of data. Instead, two
models were developed, which between them seem to do reasonably well in all cases studied.
They tend to bracket the actual behavior in most situations with one of the series usually
being the better of the two in a particular case. There may be an inherent problem with
fitting a time series to actual ambient data. The parameters of the time series reflect the total
behavior of the data, whereas the interest here is in the behavior at the extremes of the
seasonal distribution of the ozone data. If the behavior of the extreme values is not a
reflection of the random behavior of the data as a whole, then the procedures used to derive
the time series will be in question. This does not appear to have been a serious problem
with this model.
OZONE AIR QUALITY DATABASE
State and local governments throughout the nation operate ozone monitoring stations
that collect actual direct measurements of pollutant concentrations. The vast majority of
these measurements represent the country's heavily populated urban areas. These permanent
monitoring stations produce hourly concentration data using monitoring instruments that
operate continuously, producing a measurement every hour for a possible total of 8,760
hourly measurements per year. These air quality data are submitted to EPA's Aerometric
5-1
-------
Information Retrieval System (AIRS) by State and local governments, as well as federal
agencies. EPA and other federal agencies also operate some air quality monitoring sites on a
temporary basis as a part of air pollution research studies. These data may also be submitted
to AIRS.
The national monitoring network conforms to uniform criteria for monitor siting,
instrumentation, and quality assurance (40CFR58). The monitoring sites are classified into
one of three categories: NAMS, SLAMS, or SPMS. The NAMS, or National Air
Monitoring Stations, were established to ensure a long-term national network for urban area-
oriented ambient monitoring information and to provide a systematic, consistent database for
air quality comparisons and trends analysis. The SLAMS, or State and Local Air Monitoring
Stations, were established to allow State or local governments to develop monitoring
networks more tailored to their immediate ambient monitoring needs. The SPMS, or special
purpose monitors, are stations used by the State and local agencies to fulfill very specific or
short-term monitoring goals. Data from all three types of monitoring sites are included in
this multi-year database.
Hourly ozone data were retrieved from AIRS for any site that reported data for the
years 1980 through 1990. Valid daily maximum 1-hour concentrations were calculated for
each day that met the required daily validity criteria as defined in Appendix H to 40CFR,
Part 50. That is, a valid daily maximum means that at least 75 percent of the hourly values
from 9:01 a.m. to 9:00 p.m. (Local Standard Time) were measured, or at least one hourly
value exceeded the level of the ozone standard. Additional summary statistics retrieved from
AIRS included in the database are the annual number of estimated exceedances, the required
number of sampling days, and the percent data completeness. The estimated exceedances
were adjusted for missing data during the ozone season following the prescribed procedures
in Appendix H. (See Table 3-1 in Section 3 for a list of the designated ozone monitoring
seasons, which generally vary by state.)
Spatial Distribution of Ozone Monitors
The map in Figure 5-1 displays the location of the 1,500 monitors that reported ozone
data in at least one year during the study period of 1980 through 1990. The number and
location of ozone monitors can vary from year to year due to many factors. New monitors
may be added, or existing monitors may be discontinued, as a result of network reviews of
monitoring objectives. Some monitoring locations are lost when existing building leases are
terminated, or existing space is renovated. The map in Figure 5-2 displays the spatial extent
of the 323 monitors that operated in each year during the period 1980-90. A comparison of
these two maps reveals a paucity of long-term monitoring sites in the smaller cities and areas
throughout the Southeast, West, and Northwest.
The 323 sites that operated continuously each year since 1980 are contained within
221 counties. In contrast, the current monitoring network of 818 monitors that reported data
to AIRS for 1990 includes 469 counties.
5-2
-------
Trends in Ozone Monitoring Coverage
The number of ozone monitoring sites reporting data to AIRS varied from year to
year during 1980-1990. The most recent year with data, 1990, had the largest monitoring
network with 818 monitors reporting to AIRS, up 114 monitors or 16 percent, from a low of
704 monitors in 1986. Focusing on counties as the geographical unit of interest, 1990 was
again the peak year with 469 counties reporting ozone data to AIRS. This count is up 70
counties, or 18 percent, from the 399 counties reporting data in 1986. A third way of
looking at the geographical coverage of the monitoring network is to focus on the number of
distinct "areas" reporting data annually to AIRS. For these comparisons, an "area" is
defined as the larger of the geographical unit of a county, a Metropolitan Statistical Area
(MSA), or a Consolidated Metropolitan Statistical Area (CMSA). In some cases, the
geographical scope of the areas designated nonattainment under provisions of the Clean Air
Act Amendments of 1990 differ from the boundaries of existing MSAs and CMSAs. To
facilitate comparisons using the historical ozone air quality monitoring data, MSAs and
CMSAs are used to summarize many of the analyses that follow. However, when the focus
is on analyses of current ozone air quality data, summaries are presented using the designated
ozone nonattainment areas. The chart in Figure 5-3 shows the temporal pattern in the
number of areas reporting ozone data and in the distribution of area types (CMSA, MSA, or
county). Even though the number of CMSAs reporting ozone data has remained constant
during this period, the mix of sites within these areas may have changed. As expected, most
of the variability in the spatial coverage of the ozone monitoring network has occurred in
non-metropolitan areas.
TIME-SERIES MODELING APPROACH
This section begins with a literature review of papers that address issues associated
with time-series modeling of air quality data. These include discussions of various time-
series models used to model exceedance days and ozone hourly and daily maximum
concentrations. A tabular summary of the papers included in this literature review is
contained in Appendix B. The section concludes with a discussion of time-series models
developed for this study to evaluate alternative ozone design value methodologies. Model
evaluation results are provided in Appendix C.
Time-Series Literature Review
This literature review is a summary of important articles (cited as bracketed numbers
from the list at the end of this section) that address the statistical modeling of ozone
concentrations taking into account their serial dependence. Serial dependence implies that the
peak ozone concentration on a given day is correlated with the ozone concentrations on
previous days; otherwise, the concentrations would be independent. A related literature
review of statistical distributions fitted to air quality data focuses on cases where the
concentrations are assumed to be approximately independent. The computation of a design
value that corresponds to the statistical model for the serially dependent or independent cases
5-3
-------
FIGURE 5-1.
Ozone monitoring sites that reported hourly ozone concentration data
for at least one year during the period 1980-90.
FIGURE 5-2.
Location of the 323 ozone monitoring sites that monitored each year
during the period 1980-90.
5-4
-------
Number of Areas
300
250 -
200 -
1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990
CMSA's MSA's Counties
Source: Aerometric Information Retrieval ,System
FIGURE 5-3.
Number of areas (CMSAs/MSAs/counties) reporting ozone data to
AIRS by year.
5-5
-------
is discussed later on in this summary. This literature review discusses articles in which the
main focus is on the time-series model.
The strong serial dependence that has been found in ozone concentrations, and in
ozone exceedance days ([1], [4], [11]) has important implications for ozone NAAQS
attainment issues. An illustration of the serial dependence in daily maximum ozone
concentrations is found in [1]; high concentrations and NAAQS exceedances tend to be
repeated on the succeeding day. Design value estimates that realistically take into account
the serial dependence are likely to be more reliable, since those estimates are more precise
than estimates that assume independence. One problem with the time-series approach is that
a multitude of statistical models are available, often with different policy implications (since
different time-series models tend to lead to different estimated design values), and the
available data are too noisy to allow us to identify any one of these models as providing a
clearly superior description of the data. To some extent, the same comment applies to the
multitude of statistical distributions that have been fitted to ozone concentration data in cases
where independence has been assumed. However, the similarity of the tail behavior of the
fitted distributions implies that the design values tend not to be strongly affected by the
choice of the common distribution function, at least for a hypothetically independent
sequence of hourly or daily maximum hourly concentrations.
Concentration Time Series Models
Several authors have examined the autocorrelations between successive hourly and
daily maximum hourly concentrations ([1, 4, 11] for example). Autocorrelation is the
statistical correlation between one value and the kth previous value, where k is the order or
lag (the first-order autocorrelation is between one value and the previous value). Because of
this serial dependence, several authors have considered time-series models that explicitly take
the serial dependence into account.
The simplest concentration time series model ([1, 19]) is the stationary normal first-
order autoregressive model. The term "stationary" refers to an assumption that the
distribution of each concentration is constant (given no information about past or future
concentrations) and, more generally, that the joint distribution of any finite number of
consecutive values is also constant. The term "normal" corresponds to an assumption that
the concentrations have a normal distribution. The term "first-order autoregressive" implies
that the concentration on a given day or hour is given by a constant plus a multiple of the
previous concentration plus an independent error term.
Various authors have used more complex versions of this model for the daily
maximum hourly ozone concentrations. One extension of the model ([2, 12, 16]) is to use a
higher-order stationary autoregressive model (so that the present concentration is a linear
combination of several previous values and the independent error) or the even more general
autoregressive moving average (ARMA) time-series models. A further extension treats the
nonstationary cases where the mean is changing in time ([3, 5, 12, 16]). The Box-Jenkins
5-6
-------
approach takes day-to-day, month-to-month, and/or year-to-year differences until the
resulting difference series appears to be stationary. An alternative approach to
nonstationarity is to assume the mean is a certain function (linear, quadratic, or cubic
polynomial) of time, which is often estimated using multiple regression.
An important use of ozone time-series models has been in forecasting the daily
maximum hourly concentrations ([2, 3, 5, 16]). A typical ozone forecast is an advance
estimate of the next day's daily maximum hourly ozone concentration based on the present
day's ozone concentrations and meteorological measurements. Such estimates are often used
to alert the public about the predicted ozone level so that people particularly susceptible to
high ozone concentrations may take appropriate actions. Investigators have found that
forecasts based on time-series models using only the daily maximum ozone concentration
perform poorly([2]). Significantly improved forecasts include trend terms and various
meteorological factors in the time-series models, such as the most recent temperature
measurements.
A similar extension of the simple time-series model is in intervention analysis ([17,
18]). In the application of this theory in papers 17 and 18, one of the intervention events
was the opening of the Golden State Freeway in 1960, and the model thus assumes that this
event caused a step change in hourly ozone concentrations.
Exceedance Time Series Models
An alternative approach to serial correlations between consecutive daily maxima is to
ignore the actual measurement and only record whether or not that value is an exceedance of
a given threshold. The threshold can be, but does not have to be, the ozone NAAQS. Since
the exceedance either occurs or does not occur, this is a binary time series. The serial
dependence can be tested by comparing the observed exceedance process with the expected
behavior if consecutive exceedances were independent. Runs tests ([1, 4]) or exceedance
duration tests ([11]) have shown that there is statistically significant dependence for daily
maximum ozone exceedances. A runs test compares the observed number of runs of
consecutive exceedances with the expected number in the independent case. An exceedance
duration test compares the observed numbers of consecutive exceedances with the expected
values in the independent case.
The simplest binary time-series model, apart from independence, is probably the first-
order Markov chain model. This model ([1, 4, 6, 7, 8]) implies that the probability of an
exceedance depends only on whether or not the previous day had an exceedance, rather than
on the complete history of exceedance days.
The theory of extreme values shows that in the case of no trends, under very general
conditions, the distribution of the number of exceedances of a high threshold concentration is
approximately the compound Poisson distribution ([14]; see also Hsing and others (1988)).
Clusters of exceedance days occur "randomly" according to a Poisson process. The number
5-7
-------
of daily exceedances in each cluster has some random distribution. In the special cases of
independence or weak dependence, the limiting approximation to the exceedance series has
exactly one exceedance per cluster and the number of exceedances has a Poisson distribution.
A special case of the compound Poisson limiting distribution was investigated by Thrall and
Burton ([6]); their approximation corresponds to clusters with a geometrically distributed
number of exceedances per cluster.
Davison and Hemphill ([15]) proposed an exceedance model that incorporates
meteorological factors. This model, applied to Texas ozone data, assumes that exceedances
follow a Poisson process, but the rate of the Poisson process has a logarithm that is a linear
combination of meteorological measurements. The proposed application of this model was to
estimate the numbers of exceedance days when measurements are missing. This approach
differs from the missing value adjustment of Appendix H to the ozone standard, that assumes
that the proportions of exceedance days are the same for days with and without valid
measurements. However, it is assumed in Appendix H that missing concentrations on days
outside the ozone season, and on days sandwiched between valid daily maxima below 75
percent of the standard, are in both cases less than the standard.
Computation of the Distribution of the fcth Highest Value
The literature on time-series models for ozone includes analyses of the concentration
series (hourly or daily maximum hourly) and of the exceedance day binary series. These two
approaches are related. In principle we can easily compute the occurrence probabilities of
the exceedance days from a time-series model of the daily maximum hourly concentrations.
On the other hand, important properties of the concentration series can be computed from the
distributions of the numbers of exceedances of various threshold concentrations, rather than
simply the exceedances of the ozone NAAQS (0.12 ppm). Several of the cited papers exploit
the result that the kth highest value is greater than x only if the number of exceedances of x
is greater than or equal to k. In particular, it follows that the distribution function of the
annual maximum is the same as the probability of zero exceedances of the (dummy) variable
x, and the distribution function of the second highest value is the same as the probability of
zero or one exceedance of x.
Several papers use a time series model for the daily maximum concentrations to
compute the distribution of the annual maximum, second maximum, and so on ([7, 8, 9, 10,
12, 13]). This computation is very difficult to do analytically, and is more easily done either
by simulation ([8, 8a, 12]) or by using an extreme value approximation ([8, 9, 10, 12, 14]).
As discussed in the paper [8a], the calculations in the papers [7] and [8] contain
mathematical errors because the first-order Markov chain model for exceedances is
inconsistent with the assumption that the daily maximum concentrations form a Markov
process. The Markov chain model for exceedances roughly states that the conditional
probability of an exceedance given the previous exceedance record depends only on whether
or not the previous day had an exceedance. The Markov process model for daily maximum
5-8
-------
concentrations, which is a basic assumption in [7] and [8], roughly states that the conditional
probability distribution of the daily maximum given the past monitoring record depends only
on the record for the previous day. A fallacy that is easily made assumes that the Markov
process assumption implies the Markov chain assumption. Although the calculations in [7]
and [8] are not mathematically exact, because of this issue, Chock [8a] reported the results of
a small simulation study that implied that the calculations in [8] were approximately correct.
The most relevant feature of extreme value theory is that if the dependence is weak
and there is little or no trend, the limiting distribution of the maximum, second maximum,
and so on, is the same as the limit in the independent case ([7, 8, 8a, 13]). The accuracy of
the extreme value approximation for the &th highest value has been investigated by Chock
([8, 12]) and others (see [20] for example). Chock found that the approximation is best
when the rank is low (i.e., for the &th highest for small K) and when the autocorrelation is
low. (Using the results from Herman's paper [13], the dependence in the AR(1) process is
weak enough that the limiting distribution does not depend on the autocorrelation.) If the
dependence is too strong (so that the condition D' of Leadbetter and others (1983) is not
satisfied) but there is little or no trend, then the limiting distribution is affected by the
clustering of exceedances ([6, 14]). Typically the general form of the limiting extreme value
distribution of the maximum is the same as for the independent case, but the shape
parameters that define the specific limiting distribution is different from the independent
case. However, the distribution of the second highest, third highest, etc., have quite a
different form from the independent case if clustering occurs ([4]).
If the dependence is weak and there is a significant trend (i.e., the mean concentration
is not constant in time), then the limiting distribution depends on the trend and has been
derived by Horowitz and others ([9, 10]). Note, however, that Leadbetter and others [20]
have shown that the theoretical results in [9] and [10] do not apply to all trend functions
described in Horowitz's exposition, because a mathematical error resulted in the use of the
wrong normalizing constants for rapidly increasing trends.
Time-Series Model Development
The above literature review and the strong serial correlation of daily maximum ozone
concentrations, as illustrated by Figure 5-4, provide the motivation for developing a time-
series model as a tool for evaluating alternative ozone design value methodologies. Given
such a model, large numbers of random simulations of ozone seasons can be generated in
which the "true" design value is known. Thus, the inherent biases and precision of
alternative design value methods can be determined. These data sets have no missing values
and therefore are free of this source of error.
The following sections discuss the two time-series models developed in this study.
They are relatively simple but appear in most cases to adequately capture the significant
behavior of daily maximum ozone concentrations over a variety of sites. In this section we
5-9
-------
Variability of Ozone Dally Maxima
260
200 -
160 -
100 -
50
50 100 _ 150
Day of Ozone Season
200 .. 250.
Figure 5-4. Example of variability of ozone daily maximum 1-hour concentrations.
5-10
-------
describe the application of the time- series models for two site years only. In Section 6, a
more complete analysis applies these time series to 11 sites.
Basic Form of the Daily Maximum 1-hour Ozone Time Series Model
The two time-series models have the same basic form. They differ only in the
probability distribution associated with the shock term. The base model represents an
autoregressive random process of order one designated by AR(1). This process is applied
after the daily maximum concentrations are transformed to their natural logarithms and the
seasonal effect is removed. Each season a given monitor site generates a single realization of
daily maximum 1-hour ozone concentrations:
c(l), c(2), . . . , c(f-l), c(r), . . . ,c(/0
where c(t) is the daily maximum 1-hour ozone concentration on day t of an ozone season
containing n days. A time-series model that appears to apply to monitor sites in a variety of
geographic areas is as follows:
x(t) - m(r) = is a measure of the autocorrelation between the adjusted log daily maximum con-
centration on a given day and the previous day's adjusted log daily maximum concentration.
The term z(t)a is the shock term. In the model it generates a random shock which added to
the previous day's autocorrelated contribution yields the current day's adjusted log
concentration. The term z(f) is a standardized variate which is randomly sampled each day
from a distribution, with zero mean and unit standard deviation; o is the standard deviation
of the resulting random shocks.
The time-series model given in equations 1 and 2 has three major features. First, the
daily maximum concentrations are transformed to their natural logarithms. Second, a
running average log daily maximum concentration is subtracted from each log daily maxi-
mum to adjust for the seasonal effect. Third, the present day's adjusted log daily maximum
is obtained as the sum of an autoregressive term of order 1 which accounts for the previous
day's contribution plus a random shock term generated from a defined probability distribu-
tion.
5-11
-------
The use of the log transform is suggested by the increasing variability of the daily
maximum concentrations as the concentrations increase (see Figure 5-4). Figure 5-5 shows
how the variability tends to be more uniform over the season under the log transform.
Several approaches to computing the moving average for a season x(t) were studied
including simple running averages and fitting second and third and fourth order polynomials
to a season of log daily max data. These were all abandoned in favor of the LOWESS
procedure, which employs distance-weighted least squares to produce a smoothed average.
The power of this method is that a tension factor can be adjusted to give complete control
over the degree of smoothing achieved. After considerable experimentation, a tension factor
of 1/3 was applied to all sites studied. Figure 5-6 shows the generated LOWESS curve for
the same site-year featured above. Lowering the tension produces increasing fine structure in
this curve, while increasing the tension leads to an increasingly smooth curve. Figure 5-7
shows the LOWESS fits for the years 1981 through 1990 for the same site. There is a
substantial yearly effect on the shapes and heights of the fitted curves. Only a portion of this
variation can be accounted for by day-to-day variability in the adjusted log concentrations.
The terms and o are obtained by application of standard statistical software which
fits the Box-Wilson ARIMA (AutoRegressive Integrated Moving Average) model to a set of
time ordered data. In this case the series was the adjusted log daily maximum concentrations
for single seasons. This analysis permits a thorough exploration of the complex ARIMA
random process which includes both autoregressive and moving average processes of any
order as well as taking into account trends and seasonal effects. The exploration clearly
showed that a simple AR(1) process was sufficient in all cases studied. The inclusion of
moving average processes and higher order autoregressive and moving average processes
occasionally improved fit but never enough to justify inclusion of additional terms. Seasonal
effects were not investigated because of the large differences in seasonal effect from one year
to the next. The appearance of the series plots, autocorrelation function plots (ACF) and
partial auto correlation function plots (PACF) all strongly signaled the predominance of an
AR(1) process.
It is customary in fitting a time series to think in terms of the standardized variate z
as being associated with the normal distribution N(0,l). If so, the residuals from the
ARIMA fit should be normally distributed. Figure 5-8 is a probability plot of the residuals
from the ARTMA fitted to the same example site-year. The plot assumes the standardized
variate is normally distributed (or a lognormal distribution of the untransformed daily
maximum concentrations). In this case a straight line should fit the data in the figure.
Clearly this is not the case. However, in many cases the fit is quite good, as shown in
Figure 5-9 for another site in a different geographic area for the same year. In the former
case it was found that the use of a modified extreme value distribution provided a good
representation of the data. The success of the extreme value distribution is an indication that
the daily maximum concentrations approximately have Weibull distributions. This is because
the logarithm of a variate that has a Weibull distribution has an extreme value distribution.
5-12
-------
Log Transform of Ozone Dally Max
6
60 100_ 160
Day of Ozone Season
200
260
Figure 5-5. Log transform of daily maximum 1-hour concentrations at sample site.
5-13
-------
LOWESS SmootNng of Log Dally Max CTens.- ,333)
6
s
60 100, 160
Day of Ozone Season
200
250
Figure 5-6. LOWESS smoothing of the log of the daily maximum 1-hour concentrations at
the sample site.
5-14
-------
LOWESS Fits to Ten Years at a Site
5
3
3 -
50
100 160
Days of Season
200 250
Figure 5-7. LOWESS fits to ten years of ozone data at the sample site.
5-15
-------
PPLOT of ARIMA Residuals - NYC0006-88
CO
-1
-2-1 0
ARIMA Aaattuata - InQpptt
Figure 5-8. Probability plot of ARIMA residuals at a site in the New York metropolitan
area.
5-16
-------
PPLOT of AR1MA Residuals - HOU10034-88
0
-1
-2
-1
AHMA ReakJLBb - InCppb)
Figure 5-9. Probability plot of ARIMA residuals at a site in the Houston metropolitan
area.
5-17
-------
It was found that all the site years studied could be represented by one or the other or both
distributions within a standard deviation of the design value. This finding is reasonable since
a single season's ozone daily max concentrations can usually be fitted by a lognormal or
Weibull distribution with one or the other fitting better.
Model Evaluation
Table 5-1 compares several alternative design values measures for two different site-
years computed first from the ambient ozone database (AIRS) and then from a simulation
using the ARIMA and LOWESS fits. Each simulation produced 250 realizations. Several
alternative design value measures were computed for each realization and then averaged over
the 250 realizations. Standard deviations of individual realizations from the mean values
were also calculated. The design value designated by EPA is the EPA table look-up
procedure. Since the application here is to single years and there are no missing data, the
second largest concentration is the EPA table look-up design value. The columns headed by
EXPO5 and EXPO 10 are tail-exponential design values which take into account the behavior
of the upper 5 percent and 10 percent of the data respectively. The PCNT95 and PCNT90
columns are the 95-percentile and 90-percentile points on the distribution of the daily
maximum concentrations. These are potentially less variable surrogates for the design value.
Here they are used as a measure of how well the upper tail of the distributions are simulated.
The CHV column has two interpretations. It is the annual largest concentration. When
averaged over the 250 simulations it is the most reliable estimate of the "true" design value,
and thus a basis for calculating the biases of the other measures. The averaged simulated
design values are in excellent agreement with the ambient measurements at these sites. In all
cases the single season AIRS values are within a standard deviation of the averaged simulated
values. It is noted that single-season comparisons are appropriate in comparing simulated
average values and design values computed from the ambient ozone database. If the
comparison is good for single years, it will be good for averages of multiple years of real
data. Appendix C contains example time-series model simulation results for 30 years and
summary statistics for the entire set of 250 simulations.
SUMMARY
Two basic approaches have been proposed for evaluating alternative methodologies
for estimating ozone design values: (1) comparing ozone design value estimates using
historical ambient ozone data, and (2) comparing ozone design value estimates using time-
series models. The first approach allows us to compare alternative design value estimates on
a national basis using historical ozone monitoring data. In the time-series approach, the
"true" design value is known from the underlying time series, and we. can get an estimate of
the precision and bias associated with each method for a limited number of cities.
5-18
-------
Table 5-1. Comparison of simulated and ambient ozone design values for sites in New
York and Houston, 1988.
SITE RUN DATA CHV EPA EXPOS EXP10 PCNT95 PCNT90
HOU1034 AIRS O3DV .220 .190 .186 .197 .110 .090
LOG SIM 03DV .215 .183 .191 .190 .113 .093
STDV .042 .025 .022 .019 .008 .005
NYC0006 AIRS 03DV .216 .212 .220 .205 .135 .119
EVD SIM 03DV .214 .192 .200 .206 .140 .118
STDV .031 .023 .023 .024 .012 .009
Notes:
1. Single year of daily maximum ozone data.
2. LOG SIM is AR(1) process with lognormal shock term.
3. EVD SIM is AR(1) process with extreme value shock term.
4. In each case 250 simulations were run.
5. In the data column, the rows labeled O3DV are the design value estimates.
STDV is the standard deviation for the 250 simulations.
6. In the AIRS runs, CHV is the largest daily maximum in one year. In the
simulations, this number is the best estimate of the "true" design value.
5-19
-------
REFERENCES
1. Thrall, A. D., T. Permutt, and M. M. Moezzi. 1985. "The Pattern of High
Ambient Ozone Concentrations in Selected U.S. Urban Areas." American Statistical
Association Annual Meeting, Las Vegas, Nevada (August 5-8, 1985).
2. Zelden, M. D., and J. C. Cassmassi. 1979. "Development of Improved Methods for
Predicting Air Quality Levels in the South Coast Air Basin." Technology Service
Corporation, Santa Monica, California.
3. Ristroph, J. H. 1989. "Analysis and Modeling of Ozone: Baton Rouge, 1986-88."
Louisiana Department of Environmental Quality.
4. Chock, D. P. 1982. On the non-randomness of high-pollution days. Atmos.
Environ.. 16(12):2855-2862.
5. Aron, R. H., and I.-M. Aron. 1978. Statistical forecasting models: n. Oxidant
concentrations in the Los Angeles basin. J. Air Pollut. Control Assoc.. 28(7):684-
688.
6. Thrall, A. D., and C. S. Burton. 1987. The probability of NAAQS nonattainment
when exceedance outcomes constitute a two-state Markov process. Atmos. Environ..
21(12):2695-2702.
7. Hirtzel, C. S., R. B. Corotis, and J. E. Quon. 1982. Estimating the maximum value
of autocorrelated air quality measurements. Atmos. Environ.. 16(11):2603-2608.
8. Chock, D. P. 1984. Statistics of extreme values of a first-order Markov normal
process: An exact result. Atmos. Environ.. 18(11V.2461-2470.
8a. Hirtzel, C. S., and D. P. Chock. 1985. Discussion [of paper 8]. Atmos. Environ..
19(7):1207-1210.
9. Horowitz, J. 1980. Extreme values from a nonstationary stochastic process: An
application to air quality analysis. Technometrics. 22(4):469-482.
10. Horowitz, J., and S. Barakat. 1978. Statistical analysis of the maximum
concentration of an air pollutant: Effects of autocorrelation and non-stationarity.
Atmos. Environ.. 13:811-818.
11. Hirtzel, C. S., and J. E. Quon. 1981. Statistical analysis of continuous ozone
measurements. Atmos. Environ.. 15(6): 1025-1034.
5-20
-------
12. Chock, D. P. 1985. Statistics of extreme values of air quality—A simulation study.
Atmos. Environ.. 19(10): 1713-1724.
13. Herman, S. M. 1964. Limit theorems for the maximum term in stationary
sequences. Annals of Mathematical Statistics. 35:502-516.
14. Cohen, J. 1989. On the compound Poisson limit for high level exceedances.
J. Appl. Prob.. 26:458-465.
15. Davison, A. C., and M. W. Hemphill. 1987. On the statistical analysis of ambient
ozone data when measurements are missing. Atmos. Environ.. 21(3):629-639.
16. Robeson, S. M., and D. G. Steyn. 1990. Evaluation and comparison of statistical
forecast models for daily maximum ozone concentrations. Atmos. Environ..
24B(2):303-312.
17. Box, G.E.P., and G. C. Tiao. 1975. Intervention analysis with applications to
economic and environmental problems. J. American Statistical Assoc.. 70(349):70-
79.
18. Tiao, G. C., G.E.P. Box, and W. J. Hamming. 1975. Analysis of Los Angeles
smog data: A statistical overview. J. Air Pollut. Control Assoc.. 25(3):260-268.
19. Fairley, D., and C. L. Blanchard. 1990. "Rethinking the Ozone Standard." Air and
Waste Management Association, 83rd Annual Meeting, Anaheim, California.
20. Leadbetter, M. R., G. Lindgren, and Rootzen. 1983. Extremes and Related Properties
of Random Sequences and Processes. Springer-Verlag, New York, pp. 127-129.
Additional Reference
Hsing, T., J. Hiisler, and M. R. Leadbetter. 1988. On the exceedance point process for a
stationary sequence. Prob. Theory Rel. Fields.78:97-112.
5-21
-------
6 COMPARISONS AMONG ALTERNATIVE DESIGN VALUE METHODS
The Ozone Guideline defines the ozone design value as that concentration with a true
expected exceedance rate of once per year. Thus, the design value associated with the ozone
NAAQS is an abstract quantity that can only be estimated from available data. The Ozone
Guideline suggests several methods for estimating the design value, including a simplified
table look-up procedure, approaches using statistical distributions, and techniques based on
conditional probabilities.
The EPA ozone design value method in effect as of the date of enactment of the 1990
Clean Air Act Amendments is the table look-up procedure. As noted previously, for
monitoring stations meeting the 75 percent data completeness criterion in each of the three
most recent years, this method consists of selecting the fourth highest daily maximum 1-hour
concentration during the three-year period.
This section describes the results of the two complementary approaches used within
this study to evaluate alternative ozone design value methods: (1) comparisons of design
values estimated from ambient ozone data and (2) comparisons of alternative design values
derived from a time-series model. Details of the historical ambient ozone air quality
database, and the ozone time-series model development are described in Section 5 this report.
ALTERNATIVE DESIGN VALUE METHODS
The Ozone Guideline allows the use of fitted distributions to calculate ozone design
values, provided the fit is acceptable. The steps involved in fitting statistical distributions to
air quality data include selecting the specific distribution and selecting the fitting method.
There are two distinct approaches to fitting distributions to air quality data: (1) fitting
parametric distributions to raw hourly or daily concentrations and (2) fitting extreme value
distributions to the highest concentrations. Since the ozone design value is expressed in
terms of concentrations in the upper tail of the distribution, most applications have focused
on methods which fit the upper 5 to 10 percent of the distribution.
Breiman Tail Exponential Method
Under this method, two parameter tail exponential distributions were fitted to the
upper lOOp percent of the daily maximum concentrations using the maximum likelihood
estimators as described by Breiman et al. (1978). Separate calculations were performed for 5
percent and 10 percent exponential tails. Approximate 95 percent confidence bounds for
design values were obtained from each of the tail exponential distributions. Assuming that
the upper tail distribution is approximately exponential, it follows that the mean upper tail
concentration has approximately a gamma distribution. Therefore, an approximate 95
percent confidence interval for x is given by
6-1
-------
, r •, -
lower limit = x +
p
0 G~l(Q.915)
... .
upper limit = xp +
0
where G"!(0) is defined as the 100 0th percentile of the gamma distribution, with density
proportional to x™'1 e'x (the sum of m independent exponentially distributed variables, each
with mean 1).
California Air Resources Board Method
This method, developed by Larry Larsen of the California Air Resources Board (see
CARB, 1992), is an extension of the tail exponential approach developed by Breiman in
which multiple tail exponential distributions are fitted to the top lOOp percent of the daily
maximum concentrations for lOOp ranging from 5 percent to 20 percent. A separate design
value is estimated for each distribution by calculating the 100 x (364/365)th percentile of the
distribution. A weighting factor is assigned to each tail estimate based on a chi-square
goodness-of-fit statistic for the tail. The final design value estimate is then taken to be the
weighted average of the individual estimates. Software for implementing this method was
obtained directly from Larsen. The software includes a provision for applying a calibration
factor to the estimated design value. The calibration procedure consists of raising the
estimated value (in pphm) to the 0.983 power. The calibration factor was determined by
Larsen in a way that recognized the expected discrepancy between the tail-exponential
method and the EPA tabular method, since the EPA method is expected to be biased low on
theoretical grounds. According to the CARB (McGuire, 1994), the calibration factor
estimate recommended by Larsen is based on ozone data from monitoring sites throughout
California and was selected to produce design value estimates at a "suitable midpoint"
between the uncalibrated method and the EPA method. For comparison purposes, the Larsen
method was applied both with and without this calibration factor.
Distribution Fitting Method
The Breiman and Larsen methods are just two possible approaches to fitting
parametric distributions to the upper tail of air quality distributions. In the comparisons that
follow, exponential and Weibull distributions were fitted to the upper 5 percent of the three
6-2
-------
year distribution for 1987-89 using a procedure that estimated the distribution parameters by
maximum likelihood using a Newton-Raphson algorithm (SAS, 1990).
Percentile Method
Although not widely used, this method was included for purposes of comparison with
the other methods. The 100 x 364/365th percentile was estimated directly from the sorted
daily maximum concentrations using the SAS UNTVARIATE procedure. This method is
identical to the table look-up method if no days are missing during the ozone monitoring
season.
COMPARISONS USING AMBIENT OZONE DATA
A database was constructed that contained the daily maximum 1-hour ozone
concentrations for all sites in the AIRS database for the years 1980 through 1990. Daily
maximum concentrations were extracted from these data. This database was used for those
comparisons that focused on the changes in attainment status and on the impact of varying
the number of years used to estimate the design value. Since the "true" design values are
unknown, design value estimates obtained from each method can only be compared relative
to each other.
This database was supplemented with additional ozone monitoring data from the
period 1989-91 to provide an indicator of current status (EPA, 1992). For the metropolitan
area comparisons using the 1989-91 data subset, only stations with valid daily maxima on at
least 75 percent of days during the federally designated ozone season (April - October for
both areas) were used. All other stations were dropped from the analysis to focus attention
on how the various design value estimation procedures differ when reasonably complete data
sets are available. For stations meeting the seasonal data completeness criteria, any daily
maxima not meeting the above validity criteria were set to missing if they occurred during
the ozone season.
Selected Metropolitan Area Results, 1989-91
Design values were estimated for four different design value estimation methods: (1)
the EPA tabular method, (2) the percentile method, (3) the Breiman tail-exponential method,
and (4) the California Air Resources Board (CARB-Larsen) tail-exponential method. For the
Breiman method, estimates were made using both the upper 5 percent (5% tail) and upper 10
percent (10% tail) of the concentration distributions. For the CARB method, estimates were
obtained both with and without the calibration factor. These comparisons were made using
monitoring data from three groups of nonattainment areas:
1. New York-New Jersey-Long Island and Greater Connecticut nonattainment areas
2. Chicago-Gary-Lake County and Milwaukee-Racine nonattainment areas.
6-3
-------
3. Los Angeles-South Coast Air Basin nonattainment area.
Design values were calculated by the above methods for monitoring sites in the three
selected areas. Only data for the high ozone season (May through October) were used for
the South Coast Air Basin analyses, although the designated ozone season is the entire year.
Average and maximum values across monitoring sites in each region are listed in Table 6-1
and depicted graphically in Figures 6-1 through 6-3. Differences in the average value
obtained by each method are equivalent to the mean of the differences between the methods
across sites, while the maximum values indicate the design value which would be assigned to
the nonattainment area, assuming design values from any monitors not included in this
analysis are smaller.
Except for design values based on the 10 percent tail exponential (and the 5 percent
tail exponential in Los Angeles), the various methods produced estimated average design
values that are within 0.01 ppm of each other. This is also true of the maximum design
values in each area, except in Los Angeles the differences were as large as 0.02 ppm. Of
course, results at individual monitors may show wider variations (see Appendix D for a
tabular summary of the estimates for each monitoring site in these three metropolitan areas).
A review of Figures 6-1 through 6-3 indicates that the percentile method produces results
essentially identical to the fourth highest. This result is expected since the selection criteria
were established so that missing data would not be a major factor. Design values obtained
by fitting a tail exponential distribution to the top 10 percent of the data values are higher
than even the third highest concentration in each area, both on average and for the maximum
values. Lower design values were obtained from tail exponentials fitted to the top 5 percent
of the ozone values although, on average, they are still higher than the third highest for the
New York and Los Angeles areas. These results indicate that the portion of the distribution
of daily maximum concentrations to which the tail exponential is fitted can have a significant
impact on the estimated design value. This is the primary motivation for using Larsen's
approach, which uses multiple tails fitted to various portions of the upper end of the
distribution and weights the results toward those tails which best fit the available data.
Larsen's method, on average, produces estimated design values that fall between the
third and fourth highest concentration in the New York area, are equal to the fourth highest
in Los Angeles, while in the Chicago area, the Larsen method produces slightly lower values
on average. In all cases, maximum values from the Larsen method are essentially identical
to those of the fourth highest. Recalculation of design values using Larsen's method with the
calibration factor removed yields estimated design values that fall between the third and
fourth highest concentrations in Chicago, equal the third highest in Los Angeles, but are
higher, on average, than the third highest concentration in New York.
A further comparison of Larsen's method with the fourth highest concentration is
presented in Figure 6-4, which depicts the results for each monitoring site in the New York
and Chicago areas both with and without Larsen's calibration factor. The effect of excluding
the calibration factor is simply to shift the cloud of points a little to the right, with the large
6-4
-------
TABLE 6-1. Average and maximum estimated design values (ppm) for the period 1989-91
using each estimation method (4th High = fourth highest concentration, 3d
High = third highest concentration, Pcntl = percentile method, 5%TlExp = 5
percent tail exponential, 10%TlExp = 10 percent tail exponential, Larsen =
CARS method, LarNoCal = CARB method without calibration factor).
4th High 3rd High
Chicago Area
Average: 0.118 0.122
Max: 0.151 0 164
New York Area
Average: 0.140 0.146
Max: 0.165 0.175
Los Angeles
Area
Average: 0.21 0.22
Max: 0.28 0.30
5%TlExp
95% Confidence
Interval
Lower Upper
Pcntl 5%TlExp Bound Bound 10%TlExp
0.118 0.123 0.113 0 137 0.126
0.152 0.161 0.146 0.183 0.168
0.140 0.150 0.138 0.168 0.159
0.166 0.175 0.159 0.199 0.185
0.22 0.25 0.22 0.30 0.26
0.30 0.32 0.29 0.38 0.34
10% TIExp
95% Confidence
Interval
Lower Upper
Bound Bound
0.116 0.138
0.153 0.187
0.146 0.177
0.167 0.209
0.23 0.30
0.31 0.39
Larsen LarNoCal
0.113 0.118
0.151 0.158
0.144 0.151
0.166 0.175
0.21 0.22
0.28 0.30
6-5
-------
Chicago Area
4th High Pcntl 10%TIExp LarNoCal
3d High 5%TIExp Larsen
Figure 6-1. Average and maximum design values for monitors in the Chicago metropolitan
area for alternative design value estimation methods.
6-6
-------
New York Area
4th High Pcntl 10%TIExp LarNoCal
3d High 5%TIExp Larsen
Figure 6-2. Average and maximum design values for monitors in the New York
metropolitan area for alternative design value estimation methods.
6-7
-------
Los Angeles Area
4th High Pcntl 10%TIExp LarNoCal
3d High 5%TIExp Larsen
Figure 6-3. Average and maximum design values for monitors in the Los Angeles-South
Coast Air Basin area for alternative design value estimation methods.
6-8
-------
(including calibration adjustment)
0.17-
0.16-
0.15'
0.14
0.13
O)
X
f 0-12
0.11
0.1
0.09
0.08
All Sites
6.08 0.09 bTT
0.11 0.12 0.13 0.14 0.15 0.16 0.17
0.17
0.16
0.15
0.14
o> 0.1
« 0.12
^
0.11
0.1
0.09
0.08
(excluding calibration adjustment)
All Sites
0.08 0.09 0.1 0.11 0.12 0.13 0.14 0.15 0.16 0.17
Figure 6-4. Comparison of design values estimated using the California Air Resources
Board (CARB - Larsen) method with the fourth highest daily maximum 1-hour
concentration at all sites in the Chicago and New York metropolitan areas
1989-91.
6-9
-------
values being shifted farther than the small values. (The Los Angeles data are not shown due
to the scaling that would be required to accommodate their higher concentration values.)
Recall that Larsen's calibration factor is incorporated into the CARB method as a means of
targeting the EPA fourth high concentration, on average. In interpreting these figures, it is
important to keep in mind that the spatial correlation in ozone concentrations implies that
each point is not an independent observation. With this caveat in mind, these results suggest
that, at least for the cases examined here, Larsen's method produces design value estimates
that tend to be slightly lower on a site-by-site basis than the fourth highest concentration for
values below about 0.13 ppm and slightly higher for values above 0.13 ppm. Removing the
calibration factor from Larsen's method results in smaller differences for the lower design
values but greater differences in values above 0.13 ppm.
Comparisons for All Ozone Sites, 1987-89
Because almost all of the ozone nonattainment area classifications were based on
1987-89 ozone data, design value estimates were made for all sites for this three-year period.
Figure 6-5 presents the 5th and 95th percentile differences for comparisons among 1987-89
design values estimated using the EPA tabular method, the Breiman tail exponential
approach, and Weibull and exponential distributions, which were fitted to the combined
three-year distribution of ozone concentration data using the maximum likelihood approach
1987-89
(3 years)
(0.015)
0.005
0.009
0.007
(0.03) (0.02) (0.01) (
•I EPA - 5% Tailx
5% Tailx - Weib Fit
0.01 0.02 0.03
5% Tailx - Exp Fit
Exp Fit - Weib Fit
Figure 6-5. 5th and 95th percentile differences for comparisons among alternative ozone
design value estimation procedures.
6-10
-------
described above. Differences in the several parts per billion range were found among the
various distribution fitting methods. As the figure indicates, the tabular method yields design
values that tend, on average, to be less than those obtained with the tail exponential method.
Design values estimated using the Breiman tail exponential approach, which fits a separate
distribution to each year prior to estimating the design value, tend to be slightly higher than
those obtained from an exponential distribution fitted to the combined three-year distribution.
The Weibull fit gave higher estimates than the exponential fitted to the combined distribution.
Comparisons Among Multi-year Tabular Method Design Value Estimates
Appendix H to the ozone standard specifies that the three most recent years of
ambient ozone data should be used to judge compliance with the standard. EPA guidance
also requires that design values be estimated for the same three-year compliance period. As
noted previously, one potential way to reduce the year-to-year variability in design values is
to use more years of data. Several different evaluation criteria can be used to compare the
results of varying the number of years of data in the tabular method of estimating ozone
design values. One measure is to examine the variability in the ozone design value
concentration at a site, among sites, and among geographical areas.
Figure 6-6 presents the 5th and 95th percentile differences for the comparisons
between the 3-year estimate of the ozone design value and the estimates based on 4, 5, and 6
years of data at all sites reporting data. The site differences are grouped by the three end
years of 1988, 1989 and 1990. Ninety percent of the differences fall within the range shown
on the chart. Differences plotted to the right of zero indicate that the 3-year design value
was larger than the 6-year estimate, and conversely, differences to the left of zero indicate
that the 6-year estimate was larger. The largest differences are found for the period ending
in 1988, with the three-year estimate ranging from 0.01 ppm greater than the longer periods,
to 0.023 ppm less than the longer-term estimates. The largest differences are found between
the 3-year and 6-year estimates ending in 1988, likely due to the fact that the 6-year
estimates ending in 1988 span two peak ozone years, 1988 and 1983. In every case except
one (i.e., 3-year versus 4-year ending in 1989) the 3-year ozone design value estimates tend
to yield lower values than the estimates based on longer time periods.
The previous comparisons have looked at the variability among monitoring sites of
design values estimated using the multi-year tabular method. Table 6-2 shows the variability
in the number of areas (i.e.. CMSAs, MSAs, and counties) with design values equal to or
greater than 0.125 ppm ozone as estimated by the multi-year tabular method. In these
comparisons, the highest design value at any site in the area is selected as the design value
for the area for that time period. All sites reporting data for the period 1980 through 1990
are included in the analysis, even if they stopped reporting data at some point during this
time period. The results are also presented graphically in Figure 6-7. In both the table and
the graph, the number of areas with design values greater than or equal to 0.125 ppm is
displayed at the end-year of the designated time period.
6-11
-------
5th and 95th Percentile Differences
End-year is 1988
End-year is 1989 -
End-year is 1990 -
(0.03) (0.02) (0.01) 0 0.01 0.02 0.03
H DV(3yr) - DV(4yr) jj] DV(3yr) - DV(5yr)
DV(3yr) - DV(6yr)
Figure 6-6. 5th and 95th percentile differences among multi-year ozone design value
estimates.
6-12
-------
Table 6-2. Comparison of Number of Areas with Multi-year Tabular Method Ozone
Design Values Equal to or Greater than 0.125 ppm
Number of
Years in
Ozone Design
Value
Estimate
1 -YEAR
2-YEARS
3-YEARS
4-YEARS
5-YEARS
6-YEARS
Number of Areas
With Ozone Design Values Greater Than NAAQS
End-year of Time Period Used
80 81 82
97 73 61
97 75
98
-
.
.
83
90
85
88
101
-
-
84
53
77
80
86
98
-
to Estimate Design Value
85
49
56
74
75
83
96
86
53
55
59
74
74
84
87
68
69
70
71
82
84
88
112
108
104
103
100
104
89
39
99
98
95
94
91
90
48
55
98
96
93
92
Given the pattern in ozone data that we have seen during the past 11 years—several
peak years followed by years with much lower ozone concentrations—it is not surprising that
the greatest variability in the number of areas with design values greater than or equal to
0.125 ppm is associated with the 1-year estimates. The largest number of such areas is
associated with the 1-year design values for 1988, and the smallest with the 1-year estimate
for 1989. The number of areas with 1-year ozone design values greater than or equal to
0.125 ppm in 1983 is also larger than the number of areas for the 2-year and 3-year
estimates. The reason the number of areas based on the 4-year design values for 1983 is
greater than the estimates for shorter time periods is that two peak ozone years (1980 and
1983) fall in this time period. Finally, it is worth noting that there is less variability in the
number of areas with design values greater than the level of the ozone NAAQS for the
periods ending in 1988-90 than during previous years. This is likely due to the fact that
there was a single dominant year (1988) for peak ozone levels for all multi-year data periods
(except the 6-year period ending in 1988, which includes the 1983 peak ozone year).
6-13
-------
Number of Areas
125
100
75
50
25
80 81 82 83 84 85 86 87 88 89
End-year of Time Period Used for Design Value
1-year 2-years 3-years 4-years 5-years 6-years
90
Figure 6-7. Number of areas with multi-year tabular method ozone design values greater
than or equal to 0.125 ppm.
6-14
-------
The Clean Air Act Amendments of 1990 required that each nonattainment area for
ozone be classified as a Marginal Area, a Moderate Area, a Serious Area, a Severe Area, or
an Extreme Area based on the design value for the area. These nonattainment area
classifications define the primary standard attainment date and the emissions control
requirements for the area. The design value levels associated with each classification were
listed below in Table 6-3. The attainment dates associated with each classification were
listed previously in Table 1-1.
Table 6-3. Ozone Nonattainment Area Classifications
Area Class
Marginal
Moderate
Serious
Severe
Extreme
Ozone Design Value
(ppm)
0.121 up to 0.138
0.138 up to 0.160
0.160 up to 0.180
0.1 80 up to 0.280
0.280 and above
On November 6, 1991, EPA published the ozone nonattainment area designations and
classifications (40CFR Part 81). In most cases, the area classifications were based on ozone
design values estimated using the tabular method with data from 1987-89 (3 years of data).
A few areas had already submitted their 1990 ozone data, and the area's classification was
based on data from 1988-90. The potential variability in ozone nonattainment area
classifications resulting from varying the number of years of data used to estimate design
values was examined. Table 6-4 presents the impact on ozone area classifications of varying
the number of years used in the multi-year tabular design value method. The table focuses
on data windows ending in 1989, because most of the area classifications were based on data
from this period. However, to maintain consistency with the summary presented in Table 6-
2, this table continues to use CMSAs, MSAs, and counties to define the geographic area, and
not the nonattainment area boundaries of the designated nonattainment areas.
As in the case of the other indicators, the largest differences in ozone area
classifications are associated with design values based on a single year of data. There is
close agreement between classifications based on three and four years of data, while the
longer data windows (5- and 6-year) have fewer nonattainment areas (four and seven fewer,
respectively) than the 3-year estimates. There is some downward movement in area
classifications evident in the longer time periods; that is, severe areas have moved downward
to serious, serious areas to moderate, and moderate areas to marginal.
6-15
-------
Table 6-4. Impact on Ozone Area Classifications of Varying the Number of Years when
Estimating the Ozone Design Value Using the EPA Tabular Method
Number of Areas (CMSA/MSA/County)
Clean Air Act Single Year D.V.
Ozone
Classification
Extreme
Severe
Serious
Moderate
Marginal
Total
1987
1
6
17
22
22
68
1988
1
11
23
45
32
112
1989
1
4
4
12
18
39
1988-
1989
1
7
20
36
35
99
Multi-year Design Value
1987-
1989
1
9
16
33
39
98
1986-
1989
1
9
15
33
37
95
1985-
1989
1
7
14
33
39
94
1984-
1989
1
6
15
29
40
91
The potential variability in area classifications resulting from changing the three-year
data window used to assess compliance is shown dramatically in Table 6-5, which contrasts
the initial nonattainment area classifications (which were generally based on 1987-89 data)
with those that would result from the 1989-91 air quality update, which is the first 3-year
compliance period to exclude 1988. Of the original 98 nonattainment areas, 42 nonattainment
areas have 1989-91 ozone data that demonstrate compliance with the standard; 12 of 33
moderate and 30 of 42 marginal nonattainment areas meet the ozone standard based on
1989-91 data (EPA, 1992). Since that listing, seven areas have been redesignated to
attainment; Charleston, WV, Greensboro-Winston-Salem-High Point, NC, Kansas City, KS-
MO, Knoxville, TN, Parkersburg, WV, Raleigh-Durham, NC and Cherokee County, SC. For
the current 1991-93 period, 48 of the remaining 91 nonattainment areas meet the ozone
standard (EPA, 1994). A complete listing of 1991-93 ozone design values for the original 98
nonattainment areas can be found in Appendix E.
6-16
-------
Table 6-5. Comparison of number of nonattainment areas in Clean Air Act classification
categories with number of areas in classification ranges based on 1989-91
ozone monitoring data.
Number of
Areas with
Clean Air Act 1990 CAAA
Classification Classification
Number of Areas in
Classification Range
for 1989-91 Design
Values
Extreme
Severe
Serious
Moderate
Marginal
Less than O3
NAAQS
Incomplete data
Total
1
9
12
33
43
n/a
n/a
98
1
2
7
21
24
42
1
98*
* Kansas City and Cherokee County, SC have already been redesignated
to attainment.
Multi-year Design Value Estimates in Selected Cities
The above comparisons have focused on differences in design value estimates at the
individual site level and among geographic areas. Figures 6-8 through 6-10 display multi-
year ( 2-, 3-, 4-, and 5-year) table look-up ozone design value estimates for the period 1980
through 1990 in three large metropolitan areas; Atlanta, Chicago, and New York. These
areas were selected because they each have several sites in their monitoring networks that
have operated during this time period, and because they illustrate several different design
value scenarios.
6-17
-------
The trend in multi-year ozone design values in Atlanta is displayed in Figure 6-8.
The ozone design values were estimated using all sites that reported data to AIRS for the
years 1980 through 1990. The overall trend appears relatively flat during this period. The
2-year design value estimates exhibit the largest year-to-year differences. Focusing on the
last three time periods, those ending in 1988, 1989 and 1990, reveals that the 2-year design
value estimate was higher than the three longer-term estimates for 1988 and 1989, but
significantly lower than the other three values in 1990, due to dropping 1988 out of the data
window. The ozone design values based on three or more years track closely together
during this 11-year period. The 3-year ozone design value estimates in 1989 and 1990 are
slightly lower than the design values based on four and five years of data.
Chicago
Figure 6-9 presents the trend in multi-year ozone design values in Chicago for all
sites reporting data to AIRS for 1980 through 1990. The "fishbone" pattern in the early
years arises because the longer-term data windows continue to capture a special study site
with the highest concentration recorded in 1980, that stopped monitoring early in 1981. The
design value estimates are the same level in 1988 and 1989 for all four multi-year periods.
The 2-year estimate for 1990 is significantly lower (by more than 0.02 ppm) than the longer-
term estimates, reflecting the absence of the 1988 peak values and the much lower values
recorded in 1989 and 1990.
New York City
Figure 6-10 presents the trend in ozone design values for all sites reporting data for
the years 1980 through 1990 in the New York CMS A. As in Chicago, the "fishbone"
pattern results from a discontinued site that recorded peak ozone levels in 1980. This
becomes apparent when one focuses on the additional plot that displays ozone design values
estimated only at sites with data in each year of the 11-year period. This latter plot also
reveals a general downward trend in ozone design values in New York City. As in the other
two cities, the 2-year design value estimates for 1988 and 1989 are greater than, or equal to,
the 3-, 4-, and 5-year estimates. However, unlike these other two cities, in New York City,
the 5-year design value for 1990 is lower than the 3-year and 4-year estimates.
6-18
-------
wi uiauui i,
0.18
0.16
0.14
0.12
2-year 3-year 4-year 5-year
81
82
83
84
85
86
87
88
89
90
Figure 6-8. Trends in multi-year table look-up ozone design values in Atlanta, GA,
1980-90.
0.22
0.2
0.18
0.16
0.14
0.12
Concentration, ppm
Chicago
2-year 3-year 4-year 5-year
81
82
83
84
85 86
87
88 89
90
Figure 6-9. Trends in multi-year table look-up ozone design values in Chicago, 1980-90.
6-19
-------
Concentration, ppm
0.3
0.28
0.26
0.24
0.22
0.2
0.18
0.16
New York City
All Sites
2-year 3-year 4-year 5-year
81
82
83
84
85
86
87
88
89
90
Concentration, ppm
0.3
0.28
0.26
0.24
0.22
0.2
0.18
0.16
New York City
Only Sites With Data in All Years, 1980-91
2-year 3-year 4-year 5-year
81
82
83
84
85
86
87
88
89
90
Figure 6-10. Trends in multi-year ozone design values in New York for all sites and sites
with 11 years of data.
6-20
-------
Summary of Findings from the Air Quality Design Value Comparisons
1. Although the "true" design value is not known, air quality monitoring data can be used to
examine the relative differences among alternative methods for estimating ozone design
values.
2. The EPA tabular design value method tends to give lower but more variable estimates for
the ozone design value than some of the statistical modeling methods, such as the Breiman
tail exponential approach. The tail exponential approach developed by the California Air
Resources Board (CARB-Larsen) includes a calibration factor that was estimated to target the
annual fourth highest concentration. In the selected study areas, the CARB-Larsen method
was applied both with and without this adjustment. These comparisons suggest that, at least
for the cases examined here, Larsen's method produces design value estimates that tend to be
slightly lower on a site-by-site basis than the fourth highest concentration for values below
about 0.13 ppm and slightly higher for values above 0.13 ppm. Removing the calibration
factor from Larsen's method results in smaller differences for the lower design values but
greater differences in values above 0.13 ppm.
3. Design values were estimated for all ozone monitoring sites that collected data during the
three-year period 1987-89, the time period used for most of the original ozone nonattainment
area designations/classifications. Comparisons made among the EPA tabular method and
more robust approaches (e.g., Breiman tail exponential, fitting Weibull and exponential tail
distributions using maximum likelihood) revealed that the tabular approach tended to give
slightly lower estimates of the design value.
4. Increasing the number of years used to estimate the design value reduces the year-to-year
fluctuations in its value. Comparisons made for 3-year periods ending in 1988-90 had less
variability in the design value estimates than during previous 3-year periods. This is likely
due to the fact that there was a single dominant year (1988) for peak ozone levels during the
1988-90 time period.
5. Given the database available at the time, generally data through 1989, the use of more
robust methods such as the tail exponential approach would not have significantly changed the
initial ozone nonattainment area designations and classifications. Use of more years of data
(i.e., 4 or 5 years) in estimating the design value would have resulted in lower classifications
in only a limited number of cases. However, more recent data periods that do not include
1988 yield significantly different results. For the years 1989-91, the first 3-year compliance
period that excludes the 1988 data, 42 of the original 98 nonattainment have ambient ozone
meeting the standard. Seven of these areas have been redesignated to attainment. The most
recent compliance period, 1991-93, has 48 of the remaining 91 nonattainment areas also
meeting the ozone standard.
6-21
-------
TIME-SERIES MODELING SIMULATION RESULTS
In this study we are using time-series models as a tool for evaluating alternative
design value estimation methodologies. The ambient ozone database was used to develop a
time-series model of the behavior of daily maximum ozone concentrations at key monitoring
sites in selected metropolitan areas. This approach yields a time-series model that is
representative of the variability of ozone concentrations at a site during the 11 year period
1980-1990. Given such a model, large numbers of random simulations of single seasons of
daily maximum ozone values can be generated which allow the "true" design value and
estimated design values for any number of methods to be calculated over a large number of
years. The true design value is also known as the characteristic largest value and is
abbreviated as "CLV" throughout this section. The CLV is the concentration exceeded once
per year on average over a hypothetically infinite monitoring period. Thus both the inherent
biases and precision of alternative design value methods can be studied using a wide variety
of averaging years. These data sets have no missing values and therefore are free from this
source of error.
Methodology
The time-series simulation model described in Section 5 was applied to monitoring
sites in five geographically diverse metropolitan areas: Atlanta, GA; Charlotte, NC; Chicago,
IL-WI; Houston, TX; and New York, NY-NJ-CT (Table 6-6). Three sites each were
examined in the Charlotte, Chicago, and New York areas and a single site each in the
Atlanta and Houston areas. Of the eleven sites, six were best described by a time series with
a shock term that has an extreme value distribution and five sites by a normally distributed
shock term. The site-years at a given site, and the sites in a given metropolitan area, tended
to be dominated by a single shock term distribution type. As discussed in Section 5, the
extreme value shock term implies that the daily maximum ozone concentrations have Weibull
distributions. The normal shock term implies a normal distribution of the daily maximum
ozone concentrations.
The behavior of six design value estimation methods and one surrogate design value
estimation method was studied using time-series models derived for each site. A single
simulation generated one ozone season of daily maximum ozone concentrations. To compute
a single design value required TV years (seasons) of daily maximum data. The EPA table
lookup method is a 3-year method. But for the purposes of this study cases with N = 1,2,
3, 4, 5, 6, 10, and 250 were studied. In each of these cases about 3,000 seasons of
simulated data were generated. Contiguous simulations would be grouped into N simulations
each and AT-year design values calculated by each method. The average and standard
deviation of the N-year design values were calculated from the Af-year simulations for each
site. The first 35 calculated JV-year design values for each method were also recorded.
These data are displayed in Appendix C.
6-22
-------
Table 6-6. Sites studied in time-series comparisons of design values.
Metropolitan
Area
Atlanta
Charlotte
Chicago
Houston
New York
States
Covered Site Code
GA 13
NC 37
37
37
IL-WI 55
17
17
TX 48
NY-NJ-CT 34
36
09
089
119
119
119
059
097
031
201
023
085
001
0002
0034
1005
1009
0002
1002
7002
1034
0006
0067
3007
Shock Term
Distribution
Extreme Value
Extreme Value
Extreme Value
Extreme Value
Lognormal
Lognormal
Lognormal
Lognormal
Extreme Value
Extreme Value
Lognormal
Site
Years
11
10
10
11
11
11
11
10
10
11
11
6-23
-------
The 250-year design values were taken as sufficiently accurate estimates of the
limiting values for each of the computed design values. Their standard errors were about
0.0003 ppm (0.2% of the design value) for the six extreme value shock sites and about 0.001
ppm (0.5-0.7% relative standard error) for the normal shock sites.
Two table look-up design value estimation methods were studied: the EPA designated
procedure which in the case of no missing data takes the (AH-l)th largest value in N years of
daily maximum ozone data as the design value estimate and an alternative procedure which
takes the Mh largest value in N consecutive years of data. Four different methods of
estimating the tail-exponential design value were studied. A method was defined by whether
it used the upper 5 percent or 10 percent of the distribution of daily maximum ozone
concentrations and which of two methods it used to handle N years of data. The two
methods were (1) combining the N years into a single distribution and (2) fitting a tail-
exponential distribution to each year of data, expressing the N-year distribution as a sum of
equally probable individual year distributions, and iteratively determining the concentration
whose expected exceedance rate is once per year. The second method is the conditional
probability approach detailed in the 1979 EPA Guideline for the Interpretation of Ozone Air
Quality Standards (EPA, 1979) and described in Section 4.
In the case of both table look-up methods the limiting design value as N increases
without limit is the true underlying ozone design value, the characteristic largest value
(CLV). The limiting design values for the tail-exponential methods are not necessarily the
CLV since the 5 percent and 10 percent upper tails in general are only approximated by an
exponential distribution.
A single surrogate design value, the 95th percentile of the daily maximum
distribution, was investigated. A surrogate design value does not directly estimate the
underlying design value. Instead, a parameter of the underlying distribution of daily maxima
is estimated which has less variability than the direct design value estimators. The less
variable surrogate estimator can then be used either in place of the direct estimators or to
calculate a more stable estimate of the underlying design value, the CLV.
For any site the design value estimates for the same N years are calculated from the
same simulated data set. Thus, the pattern of design value estimates within a site is quite
sensitive to subtle differences.
Comparison of Alternative Design Value Estimation Methods
Table 6-7 compares the 3-year design value estimates for the eleven study sites using
the EPA table look-up, the alternative table look-up, and the 5 percent and 10 percent tail-
exponential employing the conditional probability method to compute multi-year design
values. The four methods, for the most part, are in fair agreement both in average value and
standard deviation. The EPA table look-up design value is always the smallest. When the
time series shock term for a site has an extreme value distribution, the 10 percent
6-24
-------
Table 6-7. Results from 1,000 time-series simulations of three-year design value estimates
from alternative design value estimation methods in five selected metropolitan
areas.
Distribution Average
Metropolitan Characteristic EPA Tabular
Area Site Largest Value Method
Average
Alternate
Average
Breinan 5%
Average
Breinan 10%
Tabular Method Tail-Exponential Tail-Exponential
Atlanta
Charlotte
Chicago
Houston
New York
0002
0034
1005
1009
0002
1002
7002
1034
0006
0067
3007
C
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
:LV
156
133
130
139
157
132
137
218
172
160
218
Desigi
w/(stc
0.152
0.128
0.126
0.135
0.149
0.128
0.132
0.212
0.166
0.154
0.206
i Value
1. dev.)
(.015)
(.011)
(.011)
(.010)
(.025)
(.014)
(.015)
(.030)
(.018)
(.014)
(.032)
Desigi
w/(stc
0.157
0.132
0.129
0.138
0.157
0.134
0.138
0.225
0.172
0.159
0.217
i Value
1. dev.)
(.017)
(.011)
(.012)
(.011)
(.028)
(.016)
(.018)
(.035)
(.020)
(.015)
(.035)
Desigi
w/(stc
0.157
0.131
0.129
0.138
0.155
0.132
0.136
0.217
0.171
0.159
0.213
» Value
1. dev.)
(.015)
(.010)
(.011)
(.010)
(.025)
(.014)
(.016)
(.028)
(.018)
(.014)
(.033)
Desigr
W/(8t
-------
tail-exponential design value exhibits the largest design value. The alternative table look-up
design value tends to be the next highest, while the 5 percent tail is either equal to the
alternative table look-up value or slightly smaller. When the shock term is normally
distributed the alternative table look-up design value is always the largest. The 10 percent
tail-exponential value is next highest. The 5 percent tail-exponential is either equal to the 10
percent exponential tail value or slightly smaller.
Comparison of Biases of 3-Year Design Values
The bias of the design values in any row (site) of Table 6-7 is obtained by subtracting
the CLV from the design value in the same row. The biases appear to vary with size of the
design value, nature of the shock term distribution, and possibly other site characteristics.
The largest biases occur with the New York #3007 and Houston #1034 sites, both of which
exhibit design values above 0.20 ppm. The biases for the other sites form a more consistent
pattern and are summarized in the following table:
Summary of Biases of Alternative Ozone Design Value Estimates
(Sites with design values below 0.20 ppm)
Median Range*
EPA Table Look-up -0.005 -0.004 to -0.008 ppm
Alt. Table Look-up 0.000 -0.001 to 0.002 ppm
5% Tail-Exponential -0.001 -0.002 to 0.001 ppm
10% Tail-Exponential 0.001 -0.003 to 0.003 ppm
*New York #3007 and Houston #1024 sites excluded
Inclusion of the New York and Houston sites does not affect the median biases but
significantly extends their range. The EPA tabular method displays a consistent negative
bias. The other three measures exhibit relatively little bias. There is a suggestion that the
5 percent tail-exponential bias tends to be slightly negative and the 10 percent tail-exponential
slightly positive. The range over sites of the 10 percent tail-exponential bias is slightly
greater than that of the 5 percent tail-exponential and the alternative table look-up. This
observation suggests a slight advantage of the 5 percent tail-exponential over the 10 percent
tail-exponential in that it leads too greater uniformity in design value over sites.
Figure 6-11 provides another view of the data in Table 6-7. The four design value
biases for the same site fall along the same vertical line in the figure since the CLV has a
single value for each site. To obtain separation between the Houston and New York sites the
Houston site's CLV was offset slightly to 0.22 ppm. The regression lines in Figure 6-11
serve primarily as visual aids. The biases tend to vary with site and possibly with the
6-26
-------
0.008
0006
0004
0002
0000
> -0.002
U
> -0.004
S
-------
magnitude of the design value. The figure clearly shows the consistent negative bias of the
EPA table look-up procedure. For CLV values below 0.18 ppm the alternative table look-
up design value exhibits the least bias. The 5 percent and 10 percent tail-exponential design
values biases are only slightly greater. At design values above 0.20 the spread in biases may
be bigger than at levels closer to the ozone NAAQS design value (0.12 ppm). However, too
few sites have been tested at these levels.
The 3-year design value standard deviations in Table 6-7 are plotted against their
respective design values in Figure 6-12. There is a strong dependence of standard deviation
on the magnitude of the design value: the larger the design value the larger the associated
standard deviation. The data points for a given city tend to cluster together in the figure
because the four design values methods are estimating the underlying design value fairly
well, and because the standard deviations are primarily a function of the magnitude of the
design values. There is also an effect of shock term distribution type. A wider spread exists
in the standard deviations of the different design values at sites whose shock term is normally
distributed compared to sites whose shock term has an extreme value distribution.
At the normal shock term sites, the greatest standard deviation belongs to the alternate
table look-up design value, which also produces the largest design value of the four methods.
The 10 percent tail-exponential design value has the smallest standard deviation. The EPA
table look-up and 5 percent tail-exponential have intermediate standard deviations and on
average have close to the same standard deviation.
At the extreme value sites, the alternate table look-up procedure continues to exhibit
the largest standard deviation but not the largest design value. The standard deviations of the
EPA table look-up and the 5 percent and 10 percent tail exponential design values are all
about equal.
The relative positions of the data points in Figure 6-12 for both the alternative and
EPA table look-up methods roughly conform to the slope of the regression line. The
difference is largely explained by the relative magnitudes of the respective design values.
The positions of the data points of the 5 percent and 10 percent tail-exponential design values
relative to the two table look-up design values do not conform to the slope. Their standard
deviations are clearly lower than expected based on their design values. In this comparison
the relative standard deviation of the 10 percent tail-exponential design value is smaller than
that of the 5 percent tail exponential design value.
These observations are consistent with the expectation that the tail-exponential methods
are more robust measures of the design value than the table look-up methods. The
computation of tail-exponential methods uses the largest 5 percent or 10 percent of the daily
maximum data to compute design vales whereas the table look-up procedures are based on
the location of a single daily maximum value in the distribution. The tail-exponential
6-28
-------
0.04
0.03
c
o
• ^H
•*->
(0
• r- 4
>
Q)
a
«5 0.02
TJ
C
* 0.0!
-------
methods can be said to make greater use of the available data. The fact that the 10 percent
tail-exponential uses twice the data used by the 5 percent tail-exponential accounts for it
being the more robust of the two measures.
The improvement in robustness over the EPA table look-up is not large, however.
Part of the reason is that the tail-exponential method provides a larger, less biased estimate
of the design value than the EPA table look-up method. The smaller EPA table look-up
value means a somewhat smaller standard deviation than it would otherwise have. The
greater robustness of the tail-exponential measures is more clearly seen when comparison is
made with the alternative table look-up procedure, which exhibits very little bias error.
The other reason probably has to do with the fact that the upper 5 percent and 10
percent tails do reflect to a fair extent the fluctuations in the sampled extreme values from
one period to the next. They are to some extent measuring the three-year sample design
value. This effect is illustrated in the section below, which treats a surrogate design value
measure.
Considering bias and standard deviation together, the tail-exponential yields a
relatively small but demonstrable improvement in robustness over the tabular methods and
has a smaller bias than the EPA table look-up method. The 5 percent tail exponential
method has a slightly smaller bias but slightly greater standard deviation than the 10 percent
tail-exponential method. While the alternative table look-up procedure appears to exhibit the
least bias, it has the greatest standard deviation and is therefore the least robust measure.
Comparisons of Tail-Exponential Multi-Year Methods
Table 6-8 compares the two methods for treating N years of daily maximum ozone
data in application of the tail-exponential method: (1) fitting separate tail-exponential
distributions to each year of data and solving iteratively for the concentration that
corresponds to the estimated CLV (conditional probability method) and (2) fitting one tail-
exponential distribution to the combined N years of data (N-yeai distribution method).
Comparing the two sets of 5 percent tail-exponential values in the table, the conditional
probability method yields design values from 0.000 to 0.003 ppm higher than the 3-year
distribution method. In six of the eleven cases the difference is 0.001 or less. There are
three cases in which the difference is 0.003. Two of these cases are accounted for by the
Houston 1034 site and the New York 3007 sites, which have 0.22 ppm design values.
Comparing the two sets of 10 percent tail-exponential design values, the 3-year distribution
method is lower than the conditional method by 0.000 to 0.004 ppm. In six of the eleven
cases the difference is less than or equal to 0.002 ppm. Two cases differ by 0.003 ppm and
one by 0.004 ppm. The latter case is the Houston 1034 site.
As with the conditional probability 3-year 5 percent and 10 percent tail-exponential
methods, the 3-year distribution 10 percent tail-exponential design values are somewhat
6-30
-------
Table 6-8.
Comparison of two methods of applying the Breiman tail-exponential method
to estimation of three-year design values (1,000 three-year simulations).
- Conditional Probability - 3-Year Distribution
Distribution Average
Metropolitan Characteristic Breiman 5%
Area Site Largest Value Tail-Expo.
CLV Design Value
w/(std. dev.)
Houston
New York
1034
0006
0067
3007
Average
Breiman 10%
Tail-Expo.
Design Value
w/(std. dev.)
Average
Breiman 5%
Tail-Expo.
Design Value
w/(std. dev.)
Average
Breiman 10%
Tail-Expo.
Design Value
w/(std. dev.)
Atlanta
Charlotte
Chicago
0002
0034
1005
1009
0002
1002
7002
0.
0.
0.
0.
0.
0.
0.
156
133
130
139
157
132
137
0.157
0.131
0.129
0.138
0.155
0.132
0.136
(.015)
(.010)
(.011)
(.010)
(.025)
(.014)
(.016)
0.159
0.133
0.131
0. 140
0.154
0.133
0.136
(.015)
(.010)
(.011)
(.010)
(.024)
(-013)
(.015)
0.155
0.131
0.128
0.137
0.152
0.131
0.134
(.014)
(.010)
(.011)
(.010)
(.023)
(.013)
(.015)
0.158
0.133
0.130
0.140
0.151
0.131
0.135
(.014)
(.010)
(.011)
(.010)
(.022)
(.013)
(.015)
0.218
0.172
0. 160
0.218
0.217 (.028) 0.215 (.025) 0.214 (.025) 0.211 (.021)
0.171 (.018)
0.159 (.014)
0.213 (.033)
0.174 (.018)
0.162 (.014)
0.170 (.018)
0.158 (.014)
0.213 (.032) 0.210 (.032)
0.172 (.017)
0.161 (.014)
0.210 (.031)
() indicates standard deviation
6-31
-------
larger than the 5 percent tail. The differences at individual sites closely parallel those
observed with the conditional probability set.
The bias of the conditional probability 5 percent tail-exponential was smaller in eight
cases, the same in one, and larger in two of the eleven cases. The conditional probability 10
percent tail-exponential has a smaller bias in four cases, the same in three cases, and larger
in four cases. The conditional probability 5 percent and 10 percent exponential tail biases
were uniformly small for sites with normal shock terms. The standard deviations are
generally in good agreement between the two sets. In those few cases where the design
values differ by 0.003 to 0.004 ppm, the higher of the two design values exhibits a slightly
higher standard deviation.
Summarizing, the two methods for combining 3 years of data yield essentially the
same design values in most cases. The conditional probability method is sb'ghtly superior
with respect to small bias.
Effect of Averaging Period on Alternative Design Value Methods
The time-series simulations are a particularly convenient way to study the effect on
estimated design values of the time period over which design values are estimated.
Simulations were run for 1, 2, 3, 4, 5, 6, 10, and 250-year periods for each of the sites. For
each W-year period approximately 3,000 simulations were run of an ozone season of data.
The 250-year design values (10 250-year simulations) are taken as equal to the limiting
values of their respective design values. Both the EPA table look-up and the alternate table
look-up methods converge to the CLV value. The results for the New York area site 34 023
0006 are given in Table 6-9. Results for the remaining sites, which include all the measures
studied, can be found in Appendix C. From the table it is seen that the standard deviations
fall to roughly two-thirds of their 1-year value in three years and one-third their one-year
values at 10 years for all but the 95th percentile surrogate. Its initial decline is more rapid.
The design value data for three estimation methods at each of four sites are plotted in
Figure 6-13. The EPA table look-up shows the greatest change in average estimated design
value with increasing years. The alternative table look-up procedure exhibits the least drift
in bias from 1-year to 250-year periods.
Table 6-10 compares the estimated limiting design values of four estimators. These
numbers are actually the average of 10 runs each of 250-year simulations (involving 2,500
annual simulations). Their standard errors vary from 0.0003 to 0.0006 ppm. The limiting
design values of EPA and modified table look-up are the same and equal the CLV. The
limiting W-year 5 percent and 10 percent tail distributions are closer to their respective CLVs
that those computed from the iterative conditional probability method. In each case the 5
percent tail-exponential is slightly closer to the CLV than the 10 percent tail-exponential.
6-32
-------
Table 6-9. Simulation comparison of the effect of averaging period on New York area site
340230006 design values estimated by various methods.
Averaging
Period
(years)
1
2
3
4
5
6
10
Average
EPA Tabular
Method
Design Value
w/(std. dev.)
0.157 (.027)
0.164 (.021)
0.166 (.018)
0.169 (.016)
0.168 (.015)
0.170 (.013)
0.171 (.011)
Average
Alternate
Tabular Method
Design Value
w/(std. dev.)
0.172 (.033)
0.173 (.024)
0.172 (.020)
0.173 (.017)
0.172 (.016)
0.173 (.014)
0.173 (.011)
Average
Breinan 5%
Tail-Exponential
Design Value
w/(std. dev.}
0.166 (.029)
0.170 (.021)
0.171 (.018)
0.173 (.016)
0.172 (.014)
0.173 (.013)
0.174 (.010)
Average
Breiman 10%
Ta i 1 - Exponent i a 1
Design Value
w/(std. dev.)
0.169 (.029)
0.173 (.021)
0.174 (.018)
0.176 (.016)
0.175 (.014)
0.176 (.013)
0.177 (.010)
Average
Distribution
95-Percentile
Surrogate D.V.
w/std. dev.)
0.121 (.016)
0.121 (.011)
0.120 (.009)
0.122 (.008)
0.121 (.007)
0.121 (.006)
0.121 (.005)
250 0.171 (.002) 0.172 (.002)
() indicates standard deviation
0.174 (.002)
0.176 (.002)
0.121 (.001)
6-33
-------
Altern. Tabular Design Value 5% Tail-Exponential Design Value EPA Tabular Design Value
U.^^
0.21
0.20
0.19
0.18
0.17
0.16
0.15
0.14
0.13
0.12
C
0.22
0.21
0.20
0.19
0.18
0.17
0.16
0.15
0.14
0.13
0.12
C
0.22
0.21
0.20
0.19
0.18
0.17
0.16
0.15
0.14
0.13
n 1 7
i i i I i i i i i i
7 ' "^1 — 0 — "°~ NYC-3007
NYC-0008
V^'\ " ' '. ' '• CHT40D02 ' '
i a
""'?'"
- jr--*--*^ ~ . .
; CHR-0031
- ^-* — * * T * , , , ,
)1 23456789 10
1 ' ' ' A A ' l ' rt
'_ ^ T. . .NYC-3007
-- • ' • •" NYC-i-oooe
^^"^ CHI-0002 :
CHR-0034
^ « • • — *
1 1 1 1 I 1 ! 1 1 1
) 1 23456789 10
i i rh_ ' rh rh ' ' 1 r*i
L ? . ° NYC-3007
'NYC-OC06
cra-oooz
CHH-0034
A A * • A
1 1 1 1 1 1 1 ! 1 1
H-
250
Kn
T '
7-
250
hri
111'
012345678910 250
Number of Years Averaged
Figure 6-13. Effect of averaging time on design value estimated by three different methods
at four locations.
6-34
-------
Table 6-10.
Comparison of limiting values for alternative design value estimators from
alternative design value estimation methods in five selected metropolitan areas.
Metropolitan
Area Site
Atlanta
Charlotte
0002
0034
1005
1009
Alternate
Tabular
Method
0.156
0.133
0.130
0.139
EPA
Tabular
Method
0.156
0.133
0.130
0.139
N-Year Distribution
5% Tail- 10% Tail
Exponential Exponential
Method Method
0.157
0.133
0.131
0.139
0.159
0.135
0.132
0.141
Conditional Probability
5% Tail- 10% Tail
Exponential Exponential
Method Method
0.159
0.134
0.131
0.140
0.162
0.136
0.133
0.142
Chicago
Houston
0002
1002
7002
1034
0.157
0.132
0.137
0.218
0.157
0.132
0.137
0.218
0.156
0.132
0.137
0.214
0.155
0.132
0.137
0.211
0.160
0.134
0.139
0.219
0.159
0.134
0.139
0.217
New York
0006
0067
3007
0.172
0.160
0.218
0. 171
0.160
0.218
0.172
0. 161
0.217
0.174
0.163
0.216
0.174
0.163
0.220
0.176
0.165
0.220
All limiting values are the average of 10 runs of 250-year simulations
6-35
-------
Surrogate Design Values
The current ozone NAAQS has been criticized as specifying statistical measures of
attainment whose estimators exhibit an unnecessarily high degree of variability. It has been
argued that the ozone NAAQS should be specified in terms of a statistical measure whose
estimators exhibit less variability. A NAAQS specifying the 95th percentile of the
distribution of daily maximum ozone concentrations has been proposed. Its estimators
would exhibit less variability than estimators of the expected exceedance rate. The
standard level for the suggested NAAQS would be set at a value that gives protection
equivalent to that provided by the present standard. This suggestion assumes that such a
standard level exists.
The consideration missed in this argument is that the form of a NAAQS should be
determined by the nature of the adverse effects it is intended to protect against. An
efficient NAAQS specifies that particular aspect of air quality which best correlates
through an exposure-response relationship with the adverse effect(s) it is designed to protect
against. Such a standard provides a solid basis for establishing a level of air quality at
which the risk of adverse effects is reduced to an acceptable minimum. (A multiple
number of possible adverse effects for a given criteria pollutant may require a multiple
number of NAAQS.) It also provides the same degree of protection to all sites just in
attainment.
The current primary ozone NAAQS is designed to protect against rapidly reversible
adverse health effects caused by short term exposure to ambient ozone. Since the greater
the concentration of a pollutant the greater the chance of adverse effects and the greater
their intensity, the ozone NAAQS seeks to control the magnitude of the extreme values
which can occur at a location. To control the extreme values it sets an upper limit on the
concentration level whose expected annual exceedance rate is equal to once per year.
A 95th percentile NAAQS, because it is a more central measure than the point on
the distribution of daily maximum ozone concentrations represented by the current
standard, would not necessarily be a more efficient NAAQS for controlling extreme values
even though the variability of its air quality measure is decidedly smaller than that of the
current standard. Since it is the extreme values which must be controlled, a 95th percentile
standard needs to be judged by how well it controls the extreme values.
Since this report is primarily concerned with the design value of the current form of
the ozone NAAQS, the appropriate way to consider the 95th percentile is as a potential
surrogate measure of the ozone design value. The question is: does the estimation of the
95th percentile lead to an indirect estimation of the ozone design value (the CLV) that
exhibits less bias and more precision than other more direct estimators? If the answer is
to be yes, there minimally must be a reasonably tight relationship between the 95th
percentile and the CLV that holds over all sites.
6-36
-------
Table 6-11. The 95-percentile as a surrogate ozone design value (3000 one-year and 1000
three-year simulations).
Distribution Average
Metropolitan Characteristic EPA Tabular
Area Site Largest Value Method
CLV Design Value
w/(std. dev.)
Average
One-Year
95-Percentile
Surrogate
w/(std. dev.)
Estimated
Average
Characteristic
Largest Value
w/(std. dev.)
Atlanta
Charlotte
Chicago
Houston
0002
0034
1005
1009
0002
1002
7002
1034
0.
0.
0.
0.
0.
0.
0.
0.
156
133
130
139
157
132
137
218
0.
0.
0.
0.
0.
0.
0.
0.
152
128
126
135
149
128
132
212
(.015)
(.011)
(.011)
(.010)
(.025)
(.014)
(.015)
(0.30)
0.111
0.103
0.100
0.109
0.099
0.089
0.094
0.115
(.012)
(.010)
(.011)
(.010)
(.016)
(.010)
(.012)
(.013)
0.
0.
0.
0.
0.
0.
0.
0.
165
149
144
161
142
123
132
172
(.023)
(.019)
(.021)
(.019)
(.030)
(.019)
(.023)
(.025)
New York 0006 0.172 0.166 (.018)
0067 0.160 0.154 (.014)
3007 0.218 0.206 (.032)
0.121 (.016)
0.116 (.014)
0.133 (.028)
0.183 (.030)
0.174 (.027)
0.206 (.053)
Note 1: () indicates standard deviation
6-37
-------
look-up design value as an estimator of the CLV. The 95th percentiles based on both one
and three years of data were calculated (only the one-year values are shown). It is seen that
the standard deviations for the one-year 95th percentiles are close in size to those of the
EPA table look-up method. Since there is no advantage in precision, the one-year 95th
percentile will not be considered further. The three-year 95th percentile on average has a
standard deviation of about half that of the table look-up method for the same site. It is
therefore a candidate surrogate. Figure 6-14 displays the CLV versus the average 3-year 95
percent percentile for the eleven sites. A regression line and 95 percent confidence limits
are fitted to the data. It is apparent there is no one-to-one relationship between the
variables. As would be expected, the 3-year 95th percentile increases as the CLV increases,
but the scatter of the points indicates a strong dependence on site. The regression line may
be the best that can be done without introducing added factors. Thus the use of the 95th
percentile surrogate leads to significant uneven treatment among sites.
To complete the analysis of the 95th percentile as surrogate measure, the regression
analysis was used to predict the average of the 3-year design value based on the average
three-year 95th percentile at a site. The results of the calculation are given in Table 6-12.
For comparison purposes the CLV, the average EPA look-up design value and the average
3-year 5 percent tail-exponential design value are included in the table. The design value
derived from the 95th percentile using the regression model is not an accurate estimate of
the CLV, exhibiting both positive and negative biases of varying magnitudes. The average
three-year tail-exponential is less biased. The EPA table look-up design value is more
biased than the regression estimate because of the strong tendency of the EPA design value
to underestimate the CLV. The standard deviations of the 3-year 95th percentile derived
design values are somewhat smaller than those of the EPA look-up and 3-year 5 percent tail
exponential. But this advantage in lower variability is largely related to the negative biases
at several of the sites, and does not compensate for the larger bias of this estimate
compared to the other estimates. It is possible that a much more complicated nonlinear
function of the 95th percentile would lead to a more accurate estimated design value.
An interesting feature noted by one reviewer is that the regression line is close to the
theoretical line CLV = 2S applicable if the tail exponential model was precisely valid.
Summary of Conclusions from the-Time Series Modeling Study
1. The time series for individual years of the same site were dominated by a single shock
term distribution type: either normal or extreme value. These distributions imply that the
respective distributions of the daily maximum ozone concentrations are approximately
lognormal or Weibull. The shock term was a normal distribution for five of the eleven
sites studied and an extreme value distribution for six of the eleven sites. The time-series
models of sites associated with a given metropolitan area tended to have the same shock
term type.
6-38
-------
5
o
0.24
0.22
0.20
0.18
> 0.16
fl
0.14
GO
•S 0.12
0.10
0.08
CLV = 1.896S - 0.046
0.08 0.09 0.10 0.11 0.12 0.13 0.14
Avg. 3—Year 95—Percentile Surogate (ppm)
Figure 6-14. Limiting design value (CLV) versus average 3-year 95th percentile
6-39
-------
Table 6-12. Estimation of ozone design value from the 95th percentile surrogate.
Distribution Average
Metropolitan Characteristic EPA Tabular
Area Site Largest Value Method
CLV Design Value
w/(std. dev.)
Average
Three-Year
5% Tail-Expon.
Design Value
w/(std. dev.)
Estimated Average
Design Value
From Three Year
95-Percentile
w/(std. dev.)
Atlanta
Charlotte
Chicago
0002
0034
1005
1009
0002
1002
7002
0.
0.
0.
0.
0.
0.
0.
156
133
130
139
157
132
137
0.
0.
0.
0.
0.
0.
0.
152
128
126
135
149
128
132
(.015)
(.011)
(.011)
(.010)
(.025)
(.014)
(.015)
0.157
0.131
0.129
0.138
0.155
0.132
0.136
(.015)
(.010)
(.011)
(.010)
(.025)
(.014)
(.016)
0.
0.
0.
0.
0.
0.
0.
163
151
145
161
140
123
132
(.013)
(.011)
(.013)
(.011)
(.017)
(.011)
(.013)
Houston
New York
1034
0006
0067
3007
0.218
0.172
0.160
0.218
0.212 (0.30) .217 (.028)
0.166 (.018)
0.154 (.014)
0.206 (.032)
0.171 (.018)
0.159 (.014)
0.213 (.033)
0.172 (.013)
0.182 (.017)
0.174 (.015)
0.211 (.032)
Note 1: () indicates standard deviation
2: Estimated design value: DV = 1.896*5 - 0.046 where S is the
3-year 95-percentile.
3: Standard deviation of estimated design value is 1.896 times the
standard deviation of the 3-year 95-percentile.
6-40
-------
2. The bias of a given design value estimation method can vary with the nature of the shock
term and other unidentified characteristics of the site. If the estimator is biased, the bias
may tend to increase in the negative or positive direction with increasing magnitude of the
design value. Bias can also vary with the number of years over which the design value is
estimated.
3. The year-to-year variability of estimated design values increases as the magnitude of the
design value estimate increases. The variability is also affected by the type of shock term
associated with the site and other unidentified site characteristics. The variability decreases
as the number of consecutive years over which the design value is estimated increases. The
standard deviation decreased by 31-39 percent in proceeding from one- to three-year
measures, 44-52 percent in five years, and 59-67 percent in ten years over four different
measures at one site.
4. Based on 1,000 simulations of 3-year sets of daily maximum ozone data, the table look-
up method has significant negative bias. For design values below 0.2 ppm, the alternative
table look-up method is essentially an unbiased estimator. The 5 percent and 10 percent tail-
exponential procedures have small to negligible biases comparable to the alternative table
look-up procedure.
5. The alternative table look-up procedure exhibits the least precision of the methods tested
but is not substantially less precise than the other methods. The 5 percent and 10 percent
tail-exponential and EPA table look-up procedures have essentially the same precision.
Given that the magnitudes of the 5 percent and 10 percent tail-exponential design values are
greater than the EPA table look-up design value and closer to the unbiased alternative table
look-up procedure, it is clear that the tail-exponential methods are comparatively more robust
measures than the two table look-up methods.
6. The EPA table look-up method exhibits somewhat higher precision than it would
otherwise have because of its negative bias, which results in smaller design value estimates
than the other methods.
7. To achieve the precision level of the present EPA table look-up procedure, all the
alternative methods would need to be based minimally on 3 years of daily maximum ozone
data.
8. Given the preceding conclusions and the observation that the bias of the 3-year 5 year
tail-exponential design values are somewhat less variable over sites than the 10% tail-
exponential design values, it is concluded that the 3-year 5 percent tail-exponential represents
the best compromise of bias and precision of the four methods tested.
9. The conditional probability method of estimating 3-year 5 percent tail-exponential design
values exhibits slightly less bias than the combined 3-year distribution method. As the
6-41
-------
number of years over which the estimate is computed increases the situation reverses. Both
methods are essentially equivalent.
10. The use of the 95th percentile of the distribution of daily maximum ozone
concentrations as a surrogate design value is less satisfactory than any of the four more
direct estimators of the design value. It fails to significantly reduce the variability of the
associated estimated characteristic largest value (CLV) below that achieved with the more
direct methods. (From another perspective: controlling the 95th percentile fails to improve
control of the underlying CLV.) At the same time it introduces substantial biases which
vary with the site. The bias problem would result in uneven treatment of sites relative to
what would be achieved with the more direct measures. Nor would the use of the 95th
percentile obviate the need to use 3-year data sets.
REFERENCES
Breiman, L., J. Gins, and C. Stone. 1978. "Statistical Analysis and Interpretation of Peak
Air Pollution Measurements." Technology Service Corporation, Santa Monica,
California. Work performed under EPA contract 68-02-2857.
GARB. 1992. "Technical Support Document for Proposed Amendments to the Criteria for
Designating Areas of California as Nonattainment, Attainment, or Unclassified for
State Ambient Air Quality Standards." California Air Resources Board, Sacramento,
California (March, 1992).
EPA. 1979. "Guideline for the Interpretation of Ozone Air Quality Standards",
EPA-450/4-79-003. U.S. Environmental Protection Agency, Research Triangle Park,
NC, DC, February 1979.
EPA. 1992. "EPA Data Show Steady Progress in Cleaning Nation's Air", Press
Release.U.S. Environmental Protection Agency, Washington, DC, October 19, 1992.
EPA. 1994. "EPA Report Shows Continuing Progress in Cleaning Nation's Air", Press
Release, U.S. Environmental Protection Agency, Washington, DC, October 19,
1994.
Larsen, L. C., R. A. Bradley, and G. L. Honcoop. 1990. "A New Method of
Characterizing the Variability of Ozone Air Quality-Related Indicators."
Transactions Tropospheric Ozone and the Environment, Air and Waste Management
Association International Conference, Pittsburgh, Pennsylvania.
SAS. 1990. SAS/STAT User's Guide. Volume 2, SAS Institute Inc., Gary, NC, p. 997.
6-42
-------
7 THE ROLE OF METEOROLOGY IN OZONE FORMATION
A major issue in the development and application of ozone design value estimates
centers on the influence of meteorological conditions. High ozone concentration events
are known to be associated primarily with certain combinations of meteorological factors.
In general, warm, clear days with light winds that bring, keep, or return high precursor
and ozone concentrations into an area coincide with the highest ozone concentrations.
Ozone design values estimated from data collected in years during which such conditions
are unusually frequent are unusually high due to the greater frequency of high ozone
days. Conversely, in years during which such weather conditions are infrequent,
estimated design values are lower. This effect can be quite pronounced, as evidenced by
the large number of high ozone days observed during the unusually warm summers of
1983 and 1988 (EPA, 1990). Meteorological influences of this sort tend to obscure
ozone trends resulting from trends in precursor emissions and can cause areas to fluctuate
between attainment and nonattainment in the absence of any changes in precursor
emissions.
The following sections present a brief review of the literature on meteorological
influences on ozone concentrations and methods for adjusting annual ozone summary
statistics (such as design values) to account for meteorological influences. Much has
been written on this subject. This review is not intended to cover all of these
publications, but rather to describe those which, from a historical perspective, are of
primary importance to our current understanding of meteorological influences and
adjustment methods. This section begins with a discussion of the meteorological factors
that have been found to correlate with ozone concentrations, and then surveys the various
adjustment methods that have been devised.
In the following sections, numbers enclosed in square brackets refer to the
numbered references in the tabular summary of studies on trend adjustment methods in
Appendix G. Full citations are provided in numerical order at the end of this review.
Unnumbered references do not appear in the table; full citations for these appear in
alphabetical order at the end of this section.
METEOROLOGICAL INFLUENCES ON OZONE CONCENTRATIONS
Numerous studies of the relationship between meteorological variables and ozone
concentrations have been conducted (see in particular the summaries of Zeldin and Meisel
(1978) and Pollack et al. (1988; [16]), from which much of the following material is
adapted). These studies have shown that the particular meteorological conditions most
closely associated with high ozone events vary somewhat from one location to the next.
In general, however, high ozone has been found to be associated with high temperatures,
clear skies, and light winds. Contrary to the expectations of most investigators, analyses
of the effect of mixing height on ozone concentrations have been inconclusive. Chock et
al. (1982; [10]) pointed out that this may be a result of the fact that accurate
7-1
-------
measurements of mixing height are difficult to make and that ozone may be carried over
from one day to the next in layers of stable air above the inversion top, thus confounding
the relationship between mixing height and surface ozone concentrations.
Meteorological variables can be classified according to their role in the ozone
formation process as indicated in Table 7-1. Since there is a strong interrelationship
between many of the variables (e.g., daily maximum temperature and average wet bulb
temperature), the variables could have been grouped in many ways; the classifications
shown in Table 7-1 are merely one example. Each of the variables can also be expressed
in many forms (e.g., daily maximum temperature, daily average temperature, etc.).
Pollack et al. (1988; [16]) discussed the relative merits of several possible forms. Each
of the groups of meteorological variables and their relationship to ozone is briefly
described in the following sections.
Insolation
Ozone formation is governed by the amount of solar radiation that is available for
NO2 photolysis. NO2 photolysis rates can be obtained from Demerjian et al. (1980) for
each daylight hour, and adjustments for cloud cover can be made according to the method
of Maul (1980) as adapted by Scire et al. (1983). These values can be integrated over
the 0800-1400 LST "ozone formation period" suggested by EPA (1985) or, alternatively,
over all daylight hours. Ceiling height is loosely related to the amount and type of clouds
and therefore the amount of solar radiation reaching the lowest layers of the atmosphere.
The number of daylight hours provides a measure of the extent of solar insolation in the
day and also serves as an indicator of time of year.
Ventilation
Certain variables are associated with the advection and dispersion of ozone and
precursors. Previous studies (Zeldin and Meisel, 1978; Langstaff and Pollack, 1985;
Pollack et al., 1988; [16]) have shown an inverse relationship between wind speed and
ozone concentrations. Wind speed is a measure of the rate of advection and transport,
and wind speed and wind direction are often good indicators of the weather regime
affecting an area. Whether wind speed is a good indicator of transport and dispersion
depends on wind direction. Low wind speeds are typically associated with slow-moving
high pressure systems and stagnant air masses, which are often characterized by above
average ozone concentrations. High wind speeds are often associated with weather
disturbances or the advection of new air masses into an area; both are characterized by
relatively low ozone concentrations. Wind speed may affect both the early morning
buildup and afternoon dispersion of ozone and precursors. To account for differences in
wind direction, a vector average wind speed defined as the magnitude of the resultant
vector can be calculated in addition to the scalar average wind speed.
7-2
-------
TABLE 7-1. Meteorological variables potentially associated with ozone formation.
Insolation
Sky cover
Integrated clear sky photolysis rate
Integrated cloudy sky photolysis rate
Ceiling height
Number of daylight hours
Ventilation
Vector average wind speed
Scalar average wind speed
Wind fluctuation (ratio of scalar to vector average wind speed)
Mixing height
Ventilation coefficient (mixing height times wind speed)
Transport
Wind direction
Indirect Measures
Previous day's daily maximum ozone concentration
Temperature (daily maximum, range)
Humidity indicators: dewpoint temperature, wet bulb temperature, relative
humidity
Surface pressure (corrected to sea level)
Surface pressure range (daily max. minus min. surface pressure)
Precipitation
850 mb temperature
1000 mb-850 mb temperature difference
500 mb temperature
850 mb height
500 mb height
850 mb dewpoint
Synoptic weather pattern type
7-3
-------
A measure of the persistence of wind direction during the day is provided by the
wind fluctuation ratio, which is defined as the ratio of the scalar average wind speed to
the vector average wind speed. The scalar average wind speed is the average of the
numerical wind speeds ignoring the wind direction. The vector average wind speed is
calculated by averaging the northerly and easterly components and computing the length
of the resultant vector. A ratio near one indicates a highly persistent wind direction
whereas a larger ratio indicates variability in wind direction.
Mixing height describes the vertical dimension of the volume through which
dispersion of ozone and precursors can occur. Generally, restricted mixing depths are
conducive to the buildup of pollutants, resulting in higher ozone concentrations. Elevated
mixing heights allow a relatively large amount of vertical dispersion to occur, resulting in
lower ozone concentrations. As in the case of wind speed, the mixing height may affect
both the nocturnal buildup and afternoon dispersion of ozone. The degree of difference
between morning and afternoon mixing heights may provide an indication of the type of
diurnal weather pattern to which a particular day belongs. Mixing height in many areas
and on many occasions may be hard to define because there is no sharp discontinuity in
lapse rate.
Ventilation, which is defined as the product of the mixed layer wind speed and
depth of the mixed layer, is a measure of the overall ability of the atmosphere to
transport and disperse pollutants, both vertically and horizontally. It is noi a primary
meteorological variable, but it has a strong influence on the buildup of pollutants during
the night; during the day, atmospheric ventilation directly influences the maximum
concentrations of ozone in the mixed layer.
Transport
Trajectories of polluted air masses are a function of wind speed and direction.
Ozone and precursors may be transported both into and out of a particular region. High
ozone concentrations at sites outside an urban area can occur only when trajectories are
such that a polluted air mass is advected toward the site. As a result, concentrations at
such sites are highly correlated with wind speed and wind direction.
Indirect Measures
Many meteorological variables are associated with combinations of factors that
either promote or hinder the formation of high ozone concentrations, although they are
not a direct measure of any of the factors themselves. For example, the daily range of
surface pressure does not in itself affect ozone concentrations. However, large pressure
ranges are most likely indicative of windy days with possible frontal passage and
associated precipitation. Such days generally have low ozone concentrations due to the
dispersing action of the winds, lack of solar insolation, scavenging of pollutants by
precipitation, and low temperatures. Conversely, high ozone days may be characterized
7-4
-------
by small pressure ranges. In particular locations (such as the Los Angeles Basin),
pressure differences between two stations (e.g., coastal and inland) are correlated with
ozone concentrations because of their effect on local wind patterns.
One of the most significant meteorological variables in this group is temperature.
Nearly all studies of ozone-meteorology relationships identify temperature as the single
most important factor associated with fluctuations in ozone concentration levels. While
some photochemical reactions do proceed more rapidly as temperatures increase, resulting
in a net increase in ozone formation (Whitten and Gery, 1986), temperature is also (and
perhaps most significantly) related to a number of other factors that influence ozone
formation. For example, very warm days generally occur only in conjunction with clear
skies and light winds, both of which are conducive to high ozone concentrations. In
addition, high temperature may promote changes in the amount and chemical nature of
precursor emissions (e.g., an increase in evaporative VOC emissions) which may also
promote ozone formation. In certain locations, such as in the Los Angeles basin and, to
some extent, other coastal air basins in California, 850 mb temperature1 has been found
to correlate strongly with ozone. This is thought to be a result of the fact that the 850
mb temperature over Los Angeles provides a measure of the strength and height of the
subsidence inversion normally located over the region. The condition of this inversion
exerts a particularly strong influence on ozone concentrations in the basin.
Humidity indicators (dew point, wet bulb temperature, and relative humidity) may
be related to observed ozone concentrations because they are indicative of overall weather
patterns that either promote or hinder ozone formation. Humidity also plays a role in the
photochemical reactions that result in ozone formation (Whitten and Gery, 1986).
Humidity can affect the rate of ozone destruction during the night, but its primary effect
is on ozone production during the day. Pollack et al. (1988; [16]) found a significant
relationship between ozone and relative humidity: ozone concentrations in excess of 0.08
ppm were significantly more likely (in a statistical sense) to occur in conjunction with
relative humidities below 80 percent. Cox and Chu (1991; [20]) also identified a
negative correlation between relative humidity and ozone. However, scatter plots
presented by Pollack et al. for monitoring sites in the Northeast suggest that this
relationship is reversed for relative humidities below about 55 percent. This may reflect
the fact that such low humidities are associated with the advection of relatively dry, cool,
and clean continental air masses over the region. For relative humidities above 55
percent, it is likely that the inverse relationship with ozone reflection of the inverse
relationship between relative humidity and temperature at constant mixing ratios.
Humidities at the lower end of this range are likely to be associated with higher
temperatures and lack of precipitation—both of which are indicators of high ozone.
'Air temperature at the height above ground at which the atmospheric pressure is equal to 850
millibars (typically at an altitude of about 5000 feet).
7-5
-------
At least one account of a positive relationship between ozone and seasonal
precipitation levels has been published. Sandberg et al. (1978; [9a]) described a
regression of the number of high ozone days against total precipitation from the preceding
two winter rainy seasons in the San Francisco Bay Area. The apparently strong
relationship between summer ozone and rainfall during the two preceding winters
identified by the regression was hypothesized to result from an increase in summer
biogenic emissions from trees whose growth was stimulated by the winter rains. A later
report (Sandberg et al., 1991; [9b]) indicated that the relationship, while still significant,
has weakened in recent years, possibly as a result of changes in anthropogenic emission
patterns.
A number of other variables can be included in the "indirect measures" group.
The inclusion of the previous day's ozone concentration is based on the fact that weather
patterns tend to persist from one day to the next and the possibility that ozone is carried
over from one day to the next. Temperature lapse rates (e.g., 1000-850 mb, 1000-700
mb temperature differences) are indicative of overall weather patterns and atmospheric
stability. The presence of precipitation (occurrence of thunderstorm, rain or drizzle)
generally reduces ozone concentrations due to increased cloudiness, decreased solar
insolation, washout by rain, higher ventilation by increased winds, and lower
temperatures. Atmospheric thickness (e.g., the vertical separation between the 850 and
500 mb pressure surfaces) is a direct function of the mean temperature in the layer and is
thus related to ozone concentrations for the reasons discussed above.
Although it is possible to demonstrate that many of the meteorological factors
discussed above are correlated with ozone, peak ozone concentrations are typically
observed to occur in conjunction with certain combinations of these factors. Thus,
temperature is correlated with ozone, but high temperature by itself is not a sufficient
condition for the occurrence of the highest ozone concentrations. The situation is further
complicated by the fact that temperature is typically correlated with a host of other
meteorological variables such as insolation, wind direction, etc. Many investigators have
suggested that the various combinations of weather conditions observed at a particular
location can best be summarized in terms of synoptic-scale weather patterns; therefore,
these patterns control a significant fraction of the day-to-day variability in ozone
concentrations. Synoptic weather systems, such as blocking high pressure systems and
stationary fronts and troughs are key indicators of regional ozone buildup. For local high
ozone, local effects should also be considered. Pollack et al. (1988; [16]) and
Stoeckenius (1991) evaluated two synoptic classification systems that have been proposed
for the northeastern United States. Although a statistically significant relationship
between the patterns and daily maximum ozone concentrations at individual locations was
confirmed, only a small fraction of the day-to-day variability in ozone was accounted for
by the patterns due to the lack of specific information on conditions at a particular
monitoring site. Since these findings are by no means conclusive, further research is
needed.
7-6
-------
ADJUSTING OZONE CONCENTRATIONS TO ACCOUNT
FOR METEOROLOGICAL INFLUENCES
As summarized above, ozone concentrations are affected by meteorological
conditions in a variety of ways, and therefore meteorology exerts a significant influence
on annual ozone summary statistics such as design values. By definition, the actual
ozone design value (i.e., the value which is exceeded once per year on average) is
representative of the full distribution of meteorological conditions, including normal as
well as unusually warm or cool years. However, estimates of the design value based on
data collected during an unusually warm or cool period may not be representative of the
underlying design value in years closer to the long-term climatic norm. Thus, an
estimate based on 1988 data may be significantly higher than one based on a cooler year
such as 1986, even though precursor emission levels, and therefore the actual, underlying
design value, may have remained unchanged. One way around this problem is to average
the annual summary statistics over a time period sufficiently long to be representative of
typical meteorological fluctuations. Unfortunately, the 3-year attainment test specified in
Appendix H to the ozone standard (40CFR50) and described in the Ozone Guideline
(EPA, 1979) is not always sufficient for this purpose. However, averages over longer
time periods are likely to include years with significantly different emission levels and
thus obscure any emissions-related trend. The National Academy of Sciences (NRC,
1991) has recommended the development of methods to normalize ozone trends for
meteorological variation.
Numerous papers and reports have been written on methods for adjusting ozone
concentration data to a common meteorological basis (see Appendix G for the listing of
the numbered references). Most of these investigations were motivated by the need to
detect underlying trends in ozone concentration data that would provide a historical
indication of the effectiveness of emission control programs relative to increases in
emissions generating activities. These trends are frequently obscured by the "noise"
induced by meteorological fluctuations and are therefore difficult to detect. Although
trends in estimated design values are of particular interest, various other summary
statistics have been used in these studies, as described more fully in Appendix G.
Zeldin and Meisel (1978) surveyed a number of alternative trend adjustment
methods. Most methods rely on the development of an empirical model (linear
regression models and classification systems are the most common) that relates ozone
concentrations to meteorological conditions. The influence of changing precursor
emission levels is either incorporated into the model explicitly in the form of a term
involving a time index, or implicitly via multiple sets of model parameters, with each set
estimated from data collected over a short (one- to three-year) time interval during which
emissions are assumed to remain constant. The model is then used to estimate (i.e.,
predict) ozone concentrations under a fixed set of meteorological conditions. These
conditions may represent those observed in a particular reference year or they may be
averages representative of "normal" or "typical" conditions. Assuming the model fits the
7-7
-------
data well, any year-to-year changes in the predicted ozone values indicate the influence of
changes in precursor emission patterns.
The most successful empirical models used in ozone trends adjustments account
for roughly 60 to 80 percent of the variance in daily maximum ozone concentrations.
Fluctuations in annual summary statistics adjusted using the models (see, for example,
Stoeckenius and Hudischewskyj, 1990; [17]) exhibit a residual correlation with the
unadjusted summary statistics, thus suggesting that not all of the meteorological
influences have been fully accounted for by the model. Cox and Chu (1991; [20])
presented adjusted summary statistics that vary smoothly in time and do not appear to be
correlated with unadjusted values. Adjustment methods can be categorized as being
based on regression models, classification models, and other models as indicated in Table
2 of Appendix G and described in the following sections.
Regression Methods
Many trend adjustment studies are based on the use of linear regression models to
relate ozone concentrations to meteorological conditions ([2], [6], and [10-13] in Los
Angeles and ([9], [16], and [18] hi other cities). Regression models are of the form:
Y = a + £ b& + btT + e
where
Fis the daily or seasonal ozone summary statistic (e.g., daily maximum hourly
average or seasonal mean of the daily maximums);
Xi are daily or seasonal meteorological variables (e.g., daily maximum
temperature or seasonal mean temperature);
Jis a time index (e.g., year);
e is the error term (residual); and
bj, bt, and a are regression parameters determined via the least-squares fitting
procedure.
The regression parameter for the time index term (bt~) is interpreted to represent the
(assumed) linear, meteorologically adjusted trend. This interpretation is based on the
assumption that the bfa terms in the regression equation fully account for the
meteorological influences on ozone concentrations (i.e., the residuals, e, are independent
of meteorology). If this is not the case, then bt may be subject to meteorological
influences and give a false indication of the effects of changes in precursor emissions. In
the extreme, if no attempt is made to account for meteorological influences in formulating
the model, then the "adjusted" trend is simply the slope of the least-squares regression
line obtained using the time index as the only regressor. Generally speaking, such
7-8
-------
trends, when calculated over time periods of five years or less, may be an artifact
generated by the presence of one or more years with unusual meteorological conditions.2
An additional problem with this approach is that it only handles linear trends, although
other smooth trends (i.e., quadratic) can be treated through appropriate transformation of
the time index.
Most authors do not include a time index term in the regression model and instead
fit a separate set of regression parameters using data from each year. Differences in
parameters from one year to the next are assumed to result from differences in precursor
emissions, although in practice sampling error results in additional differences. This
approach eliminates any a priori assumptions about the nature of the underlying trend and
does not assume that errors are independent of meteorology; the adjusted trend is simply
interpreted as the trend that appears when that portion of the meteorological variability
explained by the regression model is eliminated. Thus, although the adjusted trend may
still incorporate some meteorological variability, it is not subject to the sort of potential
misinterpretation noted above.
Zeldin et al. (1990; [6]) developed a regression equation for daily maximum ozone
using data from a 3-year base period. For subsequent periods, differences between ozone
predicted from the equation and observed ozone concentrations were assumed to indicate
the influence of deviations in precursor emissions from base period levels.
Chock et al. (1982; [10a]) and Kumar and Chock (1984; [10b]) developed separate
linear regression equations for each year during 1971-79 in which daily maximum ozone
concentration was the dependent variable and five different meteorological variables
served as regressors. These equations were then used to predict daily maximum ozone
concentrations using a standardized set of values for the meteorological variables. The
resulting trends in annual average daily maximum ozone were interpreted to represent
trends resulting from precursor emission changes. A similar method was employed by
Duckworth et al. (1980; [11]) and Davidson et al. (1985; [12]).
Kuntasal and Chang (1987; [13]) took a slightly different approach to trend
adjustment. They regressed monthly mean daily maximum ozone against 850 mb
temperature using data from the entire trend period (1968-85). Adjusted ozone
concentrations for each month were then obtained by substituting the climatoldgical mean
temperature into the regression equation and adding in the residual for that month. Thus,
in effect, the slope of the temperature term is used to adjust the observed ozone
concentration to the value one would expect to observe under climatological mean
temperature conditions. Wakim (1990; [18]) employed a similar approach using multiple
2Longer-term trends are much less likely to be influenced in this way since the effects of a
few unusual years are spread out over a longer time period.
7-9
-------
linear regression and annual mean residuals to adjust the annual mean of the 30 and 60
highest daily maximum ozone concentrations.
Pollack et al. (1988; [16]) pointed out that linear or log-linear regression of daily
maximum ozone against meteorological variables may lead to underprediction on high
ozone days (i.e., the expected value of the error term on high ozone days is not equal to
zero). Since high ozone days are of particular importance from a regulatory viewpoint,
this bias could produce undesirable results. To mitigate the bias, Pollack et al. attempted
to develop an improved regression fit on high ozone days (defined as days with maximum
concentrations greater than 0.08 ppm) by identifying a joint set of meteorological criteria
which a day must satisfy to be put into the high ozone category. A multiple log-linear
regression was then developed using only data from the high ozone days. Thus, Pollack
et al. employed a two-stage model consisting of a screening stage and a regression stage.
A separate set of regression parameters was estimated for each year in the study period to
allow for variations in the ozone-meteorology relationship resulting from changes in
precursor emissions. Adjusted ozone concentrations were calculated by substituting daily
meteorological conditions corresponding to a particular (reference) year into the
regression equations. As a result, the meteorologically adjusted exceedances calculated
by Pollack et al. are interpreted as the number of exceedances that would have been
observed had meteorological conditions in each year been identical to those in the
reference year. Any remaining year-to-year changes in ozone are interpreted to result
from changes in precursor emissions.
Classification Methods
As pointed out above, regression models can underestimate concentrations on high
ozone days and, furthermore, they do not lend themselves easily to the calculation of
meteorologically adjusted trends in exceedance summary statistics such as the number of
days per year exceeding the ozone NAAQS. Given the regulatory significance of high
ozone days and exceedance summary statistics, alternative models for use in trends
adjustment have been considered by a number of investigators. A common approach is to
classify days on the basis of meteorological conditions into two or more categories that
define their ozone formation potential ([3-5], [15]). Thus, warm stagnant days may be
assigned to a high ozone category while cool, windy days are assigned to a low ozone
category.
Various methods used to define such categories are described below. This
discussion is followed by a description of the procedures used to calculate adjusted trends
based on these categories.
Methods for Defining Categories
Meteorological categories of ozone formation potential can be identified in a
number of different ways. Burkhart et al. [17] defined years with similar numbers of
7-10
-------
days exceeding a preset temperature threshold as having similar ozone formation potential
("similar summers" concept). Changes in the annual ozone summary statistic from one
such season to the next are then presumed to arise almost entirely as a result of changes
in precursor emissions. Other investigators define ozone formation potential on a daily
basis. Jones developed one of the earliest approaches (Jones, 1985, 1989) which
classified any day with a temperature in excess of 90°F (32.2°C) as an ozone-conducive
day. Pollack and Moezzi (1985; [15]) defined an ozone-conducive day as any day in
which the temperature exceeds, and wind speed and cloud cover do not exceed, preset
cutoff values. Cutoffs are independently determined for each variable by noting the value
exceeded (or not exceeded) on 90 percent of high ozone days (defined as days above 0.10
or 0.12 ppm). Although this method is both objective and simple, it results in the
identification of many more ozone-conducive days than actual high ozone days since it
does not consider the joint distribution of meteorological conditions on high ozone days.
Sweitzer and Kolaz (1984; [4]) recognized the importance of the effect of joint
conditions on meteorological variables and developed a two-category system consisting of
"ozone-conducive" days and non-ozone-conducive days using a set of joint criteria
(arrived at by inspection) based on typical values of five different meteorological
variables characteristic of high ozone events in Chicago. Kolaz and Swinford (1990; [5])
extended this approach to include a numerical measure of the degree of conduciveness
associated with each ozone-conducive day.
Zeldin and Breiman (1979; [2]) employed a slightly different approach, regressing
daily maximum ozone against various meteorological variables and then defining
categories of meteorological conditions based on ranges of ozone concentrations predicted
by the regression equation. This approach is less likely to be affected by subjective
judgment than the method of Sweitzer and Kolaz but requires careful selection of the
predicted ozone ranges assigned to each category and an analysis of the regression results
(including an analysis of residuals) to ensure that the selected categories are appropriate.
Zimmerman et al. (1987; [3]) presented an example of categories developed on the
basis of a synoptic typing scheme that seeks to relate ozone formation potential to each of
several synoptic weather patterns. This approach requires judgments by a meteorologist
familiar with the synoptic weather patterns characteristic of the region under study but is
potentially powerful in that it deals directly with the atmospheric processes responsible
for generating ozone-conducive conditions on a regional scale.
Most of the category development procedures described above rely on a
meteorological classification of days by ozone formation potential which is arrived at
through an inspection of joint distributions of ozone and key meteorological variables.
The selection of key variables and category definitions is a manual process involving
some subjective judgment that cannot be consistently applied in an automated fashion to
different locations since ozone-meteorology relationships vary from one location to the
next. Stoeckenius and Hudischewskyj (1990; [17]) demonstrated the use of a subjective,
7-11
-------
automated procedure for generating meteorological categories. Their method is based on
an alternative regression model developed using the Classification and Regression Tree
(CART) methodology of Breiman et al. (1984). In this model, a data set consisting of
daily maximum ozone concentrations and daily meteorological variables is subdivided into
a finite number of categories corresponding to different regions in the n-dimensional data
space defined by the meteorological variables. The divisions are constructed by the
CART software to minimize the within-category mean square deviations while
maximizing the between-category mean square deviations. Thus, a high temperature, low
wind speed category might be identified within which primarily high ozone days can be
found together with a low temperature, high wind speed category associated with
primarily low-ozone days.
Adjustment Procedures Based on Meteorological Categories
A number of different approaches for obtaining meteorologically adjusted annual
summary statistics from a classification model have been developed. Sweitzer and Kolaz
(1984; [4]) calculated a meteorologically adjusted trend in the number of days on which
ozone concentrations exceeded 0.12 ppm by noting the ratio of the number of exceedance
days to the number of ozone-conducive days in each year. The number of exceedances
per conducive day was found to decrease between 1977 and 1983, thus suggesting a
decrease in ozone independent of meteorological factors. A similar approach has been
used by Jones (1992a, b [21]; see also CEQ, 1985, 1989; NRC, 1991, pp. 62-63), in
which ozone-conducive days are simply defined as days with temperatures in excess of
90°F. However, as pointed out in the NRC report and by Stoeckenius and
Hudischewskyj (1990; [17]), large year-to-year variations were observed in this ratio,
making trend detection difficult. Jones presented his adjusted exceedances in terms of
running 3-year averages, thus producing a smoother trend (Figure 7-1). Years especially
conducive to ozone formation such as 1983 and 1988 are readily identifiable by this
method even though it is less reliable than more sophisticated approaches.
Another simple approach to calculating meteorologically adjusted trends using a
classification model is to compare an annual summary statistic from one year to the next
only on days with similar meteorological conditions. Chock et al. (1982; [10]) used days
matching meteorological criteria that define "high," "moderate," and "low" ozone
formation potential and calculated the year-to-year trend on days falling in each category.
Pollack and Moezzi (1985; [15]) used a similar approach. Burkhart and co-workers
(1990, 1991, 1992; [7]) attempted to account for meteorological variability by grouping
years according to the frequency of occurrence of ozone-conducive days and calculating
the trend hi annual summary statistics only for those years with similar numbers of
conducive days.
Several authors (Zeldin and Thomas, 1975; [1], Zeldin and Breiman, 1979; [2],
Wackier and Bayly, 1987; [14]) used meteorological categories in regression relationships
to develop adjusted trends. Zeldin and Thomas (1975; [1]) defined a seasonal
7-12
-------
TRENDS IN O3 AIR QUALITY AND TEMPERATURE
N.Y./N.J./CT. REGION
1980-1992
50
40
0)
30
z§
£o
QX)
Si
20
10
A
f
I
I
I
\
\T*mp*ratur«
A
•4
60
50
UUL
40 X°Q
H?
<>
B O
uj£
Q.«
2«
30 S|
20
10
80 81 82 83 84 85 86 87 88 89 90 91 92
YEAR [ -JSS"
Monitoring Site: Fairfield County
IfMT
Figure 7-1. Adjustment of ozone trend based on number of days above 90°F (Source:
Jones, 1992).
7-13
-------
meteorological index based on the average value of a daily score of ozone potential. The
daily ozone potential score is determined on the basis of 10 categories representing
combinations of atmospheric stability, temperature, inversion height, horizontal pressure
gradient, day of week, and month. Zeldin and Thomas regressed the number of hours
exceeding 0.20 ppm against the seasonal mean score for a three-year base period and
used the resulting relationship to predict the number of hours exceeding 0.20 ppm during
each year of the study period. Differences between the predicted and actual trends
indicate the influence of meteorology.
Wackter and Bayly (1987; [14]) developed a definition of "ozone-conducive days"
based on a combination of temperature, wind speed, and wind direction conditions. A
linear regression was then used to relate the annual ozone summary statistic to the
number of ozone-conducive days. An additional variable representing the current year
was included in the regression equation to model the trend assumed to be associated with
changes in precursor emissions. The regression parameter for the year variable was
interpreted as representing the meteorologically adjusted trend. As noted in the previous
section, this interpretation assumes that the regression model fully accounts for all
meteorological influences on ozone concentrations.
Stoeckenius et al. (1986), in an independent derivation similar to that presented by
Zeldin and Breiman (1979; [2]), presented a generalized framework for using
meteorological categorization schemes comprising two or more categories to compute
adjusted trends in exceedances. This method is based on the use of the conditional
probability approach for estimating ozone design values described in the Ozone
Guideline. An adjusted exceedance trend is obtained by multiplying the probability of
exceedance in each meteorological category (conditional probabilities) by the long-term
average (i.e., climatological) frequency of occurrence of each category and summing over
all categories. If the categorization scheme does a good job of accounting for
meteorological influences, the conditional probabilities vary only in response to non-
meteorological factors, and the adjusted exceedance rate is a good estimate of the number
of exceedances that one would have expected to observe if meteorological conditions
during the year in question had been "typical" or "average." Assuming a multinomial
distribution for the exceedances, it is shown how the adjustment method partitions the
variance in exceedance rates into within-category and between-category components.
The fact that the adjusted annual summary statistics developed via the above
method can be interpreted as the values that would have been observed had
meteorological conditions during the year followed a "typical" or "normal" pattern
represents an advantage over other adjustment procedures that do not allow for such an
interpretation. For example, meteorologically adjusted exceedances calculated in this way
can be interpreted as a reasonable estimate of the expected (i.e., long-term average)
exceedance rate since, at least to some extent, any peculiarities in meteorological
conditions during the year in question have been accounted for.
7-14
-------
Other Adjustment Methods
While the adjustment methodology demonstrated by Stoeckenius and
Hudischewskyj (1990; [17]) is theoretically sound, flexible, and easy to implement, the
meteorologically adjusted trends calculated with this method, as well as those based on
linear regression models (with residuals added back in), exhibit some remaining
correlation with the unadjusted trends, suggesting that the models do not account for all
of the meteorological variability they seek to explain. A typical example of this is shown
in Figure 7-2, which presents results obtained by Stoeckenius and Hudischewskyj for
Philadelphia. The unadjusted exceedances of 0.085 ppm in this figure show the influence
of unusually warm weather in 1983 and 1987-88. However, unless one believes that
emission increases unrelated to the unusual weather conditions occurred in these years, it
appears that not all of the increase in exceedances in these years are properly accounted
for by the adjustment method. Kolaz and Swinford (1990; [5]) attempted to address these
shortcomings by calculating the probability of exceeding 0.12 ppm as a function of a
meteorological index of ozone potential defined on the basis of wind speed, temperature,
and solar radiation. These probabilities are then used in conjunction with long-term
average index values to compute an adjusted exceedance trend. This extension of the
meteorological categorization procedure provides a means of taking into account not just
fluctuations in the frequency of ozone-conducive conditions, but also in the intensity of
those conditions, while at the same time avoiding reliance on simple linear relationships
such as those used in the regression-based methods discussed earlier. Results obtained by
Kolaz and Swinford using this method are presented in Figure 7-3. Although the adjusted
trend is considerably less "noisy" than the unadjusted trend, some of the remaining year-
to-year fluctuations, particularly the increases in exceedances in 1983, 1985, 1987, and
1988, match fluctuations in the unadjusted trend during those years, suggesting that the
meteorological influences are not fully accounted for by this method either.
Alternative statistical models that also seek to address the shortcomings noted
above have been proposed by Shively (1990; [19]) and Cox and Chu (1991; [20]). In
both papers, a statistical distribution is selected to describe the daily maximum ozone
concentration. Parameters of the distribution are written as functions of meteorological
conditions and a time index (to allow for gradual trends such as those that might be
associated with changes in precursor emission patterns). Shively (1990; [19]) modeled
exceedances of 0.12 and 0.16 ppm as a nonhomogeneous Poisson process in which the
log of the intensity parameter (equal to the log of the number of exceedances occurring in
a given time interval) is a linear function of several meteorological variables and a year
index. Model parameters for the meteorological variables and the year index are
determined via maximum likelihood. Given the standard errors of the parameter
estimates, the null hypothesis that the parameter for the year index is equal to zero can be
tested. Rejection of the null hypothesis indicates a statistically significant underlying
trend unaffected by fluctuations in meteorological conditions (or at least those fluctuations
accounted for by the model). Shively did not calculate meteorologically adjusted annual
7-15
-------
90 r-
80 r
20.
10.
O AnnuaJ Average
Annual Average Adjusted tor Meteorology
79
00
01
02 03 fl-1
Base Year
05
06
07
00
= Annual Average (unadjusted)
= Annual Average Adjusted for Meteorology
Figure 7-2. Annual number of days on which the average daily maximum ozone
concentration hi the Philadelphia Ozone Network exceeds 0.085 ppm based
on yearly data run down the CART tree grown on 1979-88 data (adapted
from Stoeckenius and Hudischewskyj, 1990).
7-16
-------
1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989
Year
Actual
Adjusted
Figure 7-3. Actual and adjusted trends in number of days on which ozone
concentrations exceed 0.12 ppm in the Chicago area (adapted from Kolaz
and Swinford, 1990).
7-17
-------
summary statistics and therefore his results cannot be directly compared to year-by-year
results obtained via regression or classification procedures.
Cox and Chu (1991; [20]) modeled the daily maximum ozone concentration using
a Weibull distribution with a fixed shape parameter and a scale parameter the logarithm
of which varies as a linear function of several meteorological variables and a year index.
Model parameters are fitted via maximum likelihood. The fitted distribution can be used
to estimate percentiles and threshold exceedance probabilities for the daily maximum
ozone concentration given a fixed set of meteorological conditions. Cox and Chu tested
for a statistically significant trend term to determine if an underlying meteorologically
adjusted trend can be detected. They also used their model to calculate "meteorologically
adjusted" estimates of the upper percentiles of daily maximum concentrations in each
year. These estimates are obtained by substituting a one-year sequence of adjusted daily
meteorological variables into the Weibull scale parameter equation and finding the
concentration value exceeded on 1 percent of days over the course of the year. The
adjusted meteorological variables are obtained by normalizing each of the actual daily
values observed over the 10-year period being analyzed with respect to their long-term
mean and standard deviation. This process rescales each meteorological variable so that
it becomes centered over its long-term distribution. Cox and Chu demonstrated both that
year-to-year variations due to sampling effects are minimal, and that the adjusted 95th
percentiles are primarily influenced by the trend term for the 31 cities shown in Figures
7-4 through 7-11. The smoothing introduced by the meteorological adjustment is
especially evident in the ozone trends for eastern cities where peak ozone years, such as
1983 and 1988, have been followed by years less conducive to ozone formation. Results
of application of the model to a number of urban areas are encouraging. EPA is seeking
to review and expand the technical basis for the methodology under a cooperative
agreement with the National Institute of Statistical Sciences (NISS).
SUMMARY AND CONCLUSIONS
The influence of meteorological conditions, particularly temperature, on ozone
concentrations has been well established. Meteorology has been found to account for as
much as 80 percent of the variance in daily maximum ozone concentrations at many
locations in the eastern and midwestern U.S. and in California coastal cities. Although
the particular combination of meteorological variables most closely associated with high
ozone events varies from location to location, high temperatures, clear skies, light winds
and limited vertical mixing generally result in the highest ozone events. Due to
correlations of temperature with other variables, the daily maximum temperature is often
the single most important variable in explaining day-to-day ozone variations. However,
since high temperature by itself is not sufficient to produce high ozone concentrations,
including other meteorological variables in the analysis often produces better results.
This is particularly true at locations where ozone and precursor materials transported
from upwind source regions account for a significant concentration increment on high
7-18
-------
ATLANTA OZONE TRENDS
tS Hi KRCENT1U — OAJLT MAXINM (MAY-OCT)
220
200
180
'"'
140
120
100
90
60
TREND (X/YR): 0.7
STDERR: 0.2
1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991
YEAR
BALTIMORE OZONE TRENDS
IS Hi PCKCNTU — OALT MAXMIM (JUNE-SOT)
220
200
180
160
uo
120
100
80
(0
TREMO (XA»): - ' 1
STO ERR: 0 3
1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991
TEAR
220
200
ISO
"°
HO
120
100
80
60
BAKERSFIELO OZONE TRENDS
95 Hi PERCEMTIIX — DALY MAXIMUM (UAY-OCT)
BATON ROUGE OZONE TRENDS
93 III FERCENTLE — DALY MAXIMUM (MAT-OCT)
'-0
STDERR: 0.2
220
200
180
|,60
„ 140
£
S 120
100
80
60
TREND (XA») -0 1
STD ERR. 0.3
1161 1182 198J 1984 1985 1166 1*87 1M6 19*9 1990 1991
YEAR
1981 1982 1983 1984 198S 19M 1987 1968 1989 <990 1991
YEAR
Figure 7-4. Actual and meteorologically adjusted ozone trends in Atlanta, Baltimore,
Bakersfield, and Baton Rouge, 1981-91 (adapted from Cox and Chu,
1991).
7-19
-------
BEAUMONT.TX OZONE TRENDS
95 Hi KRCCMTIU — DULY MAXHUM (MAY-OCT)
2JO-
200
leo
"
:,«
8 120
too
80
CO
TREND ( VYR): -1.4]
STO ERR: 0.41
1981 1982 198] 1984 1985 1986 1987 1988 1989 1990 1991
YEAH
BIRMINGHAM OZONE TRENDS
95 ft KRCtNTU — DM.Y MAXIMUM (MAT-OCT)
220
200
180
1981 1982 19B3 1984 19«5 198C 1987 1988 1989 1990 1991
TEAR
BOSTON OZONE TRENDS
»5 Hi PtHCOITUI — DAILY MAXIMUM (JUNE-SEPT)
BRIDGEPORT OZONE TRENDS
95 Hi PtRCENTIU— OAIT MAXMJM (JUNE-SEPT)
220
200
180
120
100
80
80
TREND (XA"): -I-01
STO ERR: 0.3
1981 1982 1983 1984 198$ 1988 19*7 1M8 1989 1990 1991
YEAR
1981 1982 1983 1984 198S I9«« 1987 1988 1989 1940 1991
Figure 7-5. Actual and meteorologically adjusted ozone trends in Beaumont,
Birmingham, Boston, and Bridgeport, 1981-91 (adapted from Cox and
Chu, 1991).
7-20
-------
CHARLOTTE OZONE TRENDS
95 Hi PERCENTItC— OALY MAX6AM (MAY-OCT)
220
200
160
u 140
§ 120
100
M
60
*^^^y^
•*• ACTUAL
»•« ADJUSTED
— tSXCONF
TREND (X/YR): -0.8
STD ERR: 0.2
\^
1*81 1982 1983 1984 1969 1988 1987 1988 IMS 1990 19
TEA*
1
CINCINNATI OZONE TRENDS
IS Hi PERCENTILE — DAILY MAXIMUM (JUNE-SEPT)
220
200
180
„ 140
g 120
100
60
60
^ ^J-4
•** ACTUAL
•^ADJUSTED
— 95XCONF
1961 1*82 11*3 1(64 1*6S 1*86 11*7 11
YEAR
TREND (X/YR)-. -1.3
STD ERR: 0.3
N_
^H^^M
188 1*09 1*90 1991
220
200
160
^ 140
§ 120
100
80
(0
220
200
180
£• 160
u MO
g 120
100
M
60
CHICAGO OZONE TRENDS
*i *i KRCtNTU— DAIY MAXMUV (JUNE-SEPT)
THEN!) (XAR): -0.5
STO ERR: 0.4
^£^4\^
V_^^
*»• ACTUAL
••• ADJUSTED
- - *SX CONF
, i i , 1 ,.,..,....,-... ( .... ( ..,.,....! J i i .,....,....,
1961 1*82 1983 1984 1989 1*66 1967 1988 1989 1990 1991
YEAR
CLEVELAND OZONE TRENDS
9! Id PERCENTILE— OAIY MAXIMUM (JUNE-SEPT)
TREND (X/YR). -1.6
STD ERR: 0.2
^\
^^fcijT4-*--^^
»*« ACTUAL
••« ADJUSTED
- 9SXCONT
1961 1962 1983 1984 1*89 196* 1*67 1986 1969 19*0 1991
YEAR
Figure 7-6. Actual and meteorologically adjusted ozone trends in Charlotte, Chicago,
Cincinnati, and Cleveland, 1981-91 (adapted from Cox and Chu, 1991).
7-21
-------
220
180
1"°
u **'
s 1M
100
80
to
220
' 200
180
S 180
a.
g"0
3 120
100
80
80
COLUMBIA.SC OZONE TRENDS
H Mi rtRCEHTlU— OAIY UAXMM (MAY-OCT)
TREND (X/YR): -0-2
SID ERR: 0.2
'
t=*=--4-==ir-±^+==^^
i*=*:— — i^ — *"^ x^ SI
•*• ACTUAL • ">
•M ADJUSTED
— 95XCONF
1*81 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991
YEAR
DENVER OZONE TRENDS
99 Hi P€RC£NTIU — DAILY MAXIMUM (JUNE-SEPT)
TREND (X/YR): -1.5
STD ERR.' 0.2
*^^ ^ — -^^fcr —4—i
~*'^ ^fc—4
»*«Acnua
•M ADJUSTED
— 95XCONF
19(1 1982 1983 1*84 1989 1*88 1987 1*88 1989 19*0 1991
YEAR
220
200-
180
g 140
100
80
M
220
200
180
1"°
* M°
o
100
go
(0
DALLAS OZONE TRENDS
95 ft PUCtMTU — DALY MAXIMUM (UAY-OCT)
TREND («Al): -Z-*
STO ERR: 0.3
>^^_
^"^ f ' ^"^"^-J
»•« ACTUAL
*•« ADJUSTED
9SXCOHT
1981 1982 1983 1984 198S 19W 1*87 1*88 1989 1990 199,
YEAH
DETROIT OZONE TRENDS
95 Hi PERCENTLE— DALY MAXNUU (JUNE-SEPT)
TREND (XAR): -1.3
STO ERtt: 0.2
*.
>< ~~~J~^-* I^Sx^ }
^*^-^"^
»** ACTUAL
•*• AOJUSTEO
95XCOW
1981 1*82 1983 1984 1985 1*86 1*87 1988 1989 1990 1991
YEAR
Figure 7-7. Actual and meteorologically adjusted ozone trends in Columbia, Dallas,
Denver, and Detroit, 1981-91 (adapted from Cox and Chu, 1991).
7-22
-------
220
180
r-
„ 140
8 120
100
80
60
220
180
I'"
o
o 120
100
80
60
EL PASO OZONE TRENDS
99 Hi KRCtNTIlE — DAILY MAXIMUM (APRM.-OCT)
TREND (X/YR): 1.2
STDERR: O.I
1
1— fc^£^T— K-4- i^ * » ^
»»« ACTUAL
••• ADJUSTED
— 99XCONF
1961 1962 1983 1984 1985 1986 1987 1988 1989 1990 1991
TEA!
HARTFORD OZONE TRENDS
95 th PERCENTILE — DAILY MAXIMUM (JUNE-SEPT)
TREND {x/YR): -Z.B
STD ERR: 0.3
i-*4\ ^
^\^/^
«*• ACTUAL
M« ADJUSTED
95XCONF
1981 1982 1983 1984 1989 198t 1987 1988 19*9 1990 19(1
TEAR
220
180
|...
u 140
z
o
N |20
100
80
to
300
240
t 210
0 180
M
0
190
120
90
FRESNO OZONE TRENDS
99 Hi PCRCENTLE — DALY MAXIMUM (MAT-OCT)
TRtNO (XA»): -O.J
STD ERR: 0.2
^ ^~^^U^
»*« ACTUAL
•*• ADJUSTED
- 95X CONF
1981 1982 1983 1984 1989 1986 1967 1988 1989 1990 1991
TEAR
HOUSTON OZONE TRENDS
99 Ih PERCENTU— DA«.Y MAXIMUM (UAT-OCT)
TREND (X/YR): -2 8
STD ERR: 0.3
.^^
^^^=^/^4
*»« ACTUAL
•«« ADJUSTED
99XCONF
1961 1982 1983 1(64 1989 1981 1(67 1(86 1(8( 1990 1991
TEAR
Figure 7-8. Actual and meteorologically adjusted ozone trends in El Paso, Fresno,
Hartford, and Houston, 1981-91 (adapted from Cox and Chu, 1991).
7-23
-------
LOS ANGELES OZONE TRENDS
95 * KMXNT1U—DM.T MAXIMUM (APML-OCT)
TREND (X/YR): -2.8
STDtRR: 0.2
1911 1M2 1983 1984 1989 1986 1987 1988 1989 1990 1991
MIAMI OZONE TRENDS
99 Hi PERCENTIU—OA«.Y MAXKUM (MAY-OCT)
1981 1981 1983 1984 198S 1988 1987 1988 1989 1990 1991
LOUISVILLE OZONE TRENDS
95 Hi POKCNTU—DALY NAXUUM (JUKE-SEPT)
220
200
180
1"°
w 140
z
S 120
100
80
M
STOERR: 0.51
I • • ' • I
1981 1982 1983 1984 198S 1986 1987 1988 1989 1990 1991
YEAR
MILWAUKEE OZONE TRENDS
99 Hi PERCENTU—DAILY MAXIMUM (JUNE-SEPT)
220
200
180
I'"
u 140
8 (20
100
80
80
UMNO (XA»)- -0.91
STOERR: 0.3
1981 1982 1983 1984 1989 1986 1987 1988 1989 19*0 1991
YEAR
Figure 7-9. Actual and meteorologically adjusted ozone trends in Los Angeles,
Louisville, Miami, and Milwaukee, 1981-91 (adapted from Cox and Chu,
1991).
7-24
-------
220
200
180-
I"0
u '40
g
S '20
100
80
CO
220
200
180
a ICO
£
~.40
8 120
100
80
60
MUSKEGON OZONE TRENDS
95 Hi PERCENTILE — DAILY MAXIMUM (JUNE-SEPT)
TREND (X/YR): 0.0
STDERR: 0.4
XT^>^
\r^ ' N! — l^* J * ~f
»»« ACTUAL
••• ADJUSTED
— tSXCONF
1981 1982 1963 1984 1965 1986 1987 1988 1989 1990 1991
YEAR
PHILADELPHIA OZONE TRENDS
95 Hi PERCENTILE — DAILY MAXIMUM (JUNE-SEPT)
TREND (X/YR): -1.4
STDERR: 0.2
T—4W /\
CJ^rN' — '~~~~^^f~~~4\\~^
^~\1X^ ^ V~^S — ~^
»** ACTUAL
•••ADJUSTED
— 95XCONF
1981 1982 1963 1984 1985 I96( 1987 1988 1969 1990 1991
YEAR
220
200
160
„ 140
S
8 120
100
80
CO
220
200
180
SIM
t
„ 140
o
S 120
too
80
*0
NEW YORK OZONE TRENDS
95 Hi PCRCENTU— DALY MAXMJM (JUNE-SEPT)
, TREND (xA»): -2-5
STD ERR: 0.3
fr-""^ ^^:;^H~-7K-i\_L
\ / ^~Y^ri~rl
v
**• ACTUAL
•« ADJUSTED
9SX CONF
1961 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991
YEAR
PHOENIX OZONE TRENDS
95 Hi PERCENTILE — DAILY MAXIMUM (APRIL-OCT)
TREND (X/YR). -I.Ol
STDERR: 0.2|
^-^==^*___
V— — —*-^-"^ ~~~*
»»« ACTUAL
•M OJUSTEO
93XCONT
1961 1982 1983 1984 1985 1961 1987 1988 1989 1990 1991
YEAR
Figure 7-10. Actual and meteorologically adjusted ozone trends in Muskegon, New
York, Philadelphia, and Phoenix, 1981-91 (adapted from Cox and Chu,
1991).
7-25
-------
TAMPA OZONE TRENDS
95 Hi PERCCHTILC—OAIT UAXMIM (MAY-OCT)
220-
200
IN
S 160
£
^ 140
120
100
80
60
TREND (X/YR): -V.7|
STOEIW: 0.3
1981 1982 1983 1984 1>8S 19B6 1967 1988 1989 1990 1991
YEAR
TULSA OZONE TRENDS
95 ft PCRCCNTU—MIT IttXMUH (MAT-OCT)
220
200
180-
100
80
to
fTRtNB (*/«): -0 1
I sro ERR: 0.2
1981 1987 1983 1984 1985 1996 1987 1988 1989 1*90 1991
TEAR
220
200
180
•£ 160
a.
a.
^ 140
§ 120
100
80
60
WASHINGTON OZONE TRENDS
95 Hi I>ERC£N71L£—DAILY MAXIMUM (JUNE-SEPT)
I TREND (X/YR)- -06]
STB ERR' 0.3
1981 1982 1963 1984 1185 1966 1987 1968 1989 1990 1991
YEAR
Figure 7-11. Actual and meteorologically adjusted ozone trends in Tampa, Tulsa, and
Washington, D.C., 1981-91 (adapted from Cox and Chu, 1991).
7-26
-------
ozone days. High concentrations at such locations are primarily associated with weather
conditions conducive to both ozone formation and transport from the upwind source
regions.
The year-to-year variability in prevailing weather conditions during the high ozone
season can mask underlying ozone trends associated with changes in precursor emission
patterns and can distort estimated design values. As a result, a great deal of attention has
been given to the development of procedures for adjusting summary statistics to remove
the effects of meteorological fluctuations. A wide variety of methods have been used, all
of which rely on the development of a mathematical relationship between ozone
concentrations and meteorological factors. This relationship is then used to estimate
(predict) ozone concentrations expected to occur under standardized meteorological
conditions. "Adjusted" summary statistics calculated from these predicted concentrations
can then be examined for trends or, if the standardized meteorological conditions are
representative of typical or average conditions, to provide estimates of design values less
affected by year-to-year meteorological fluctuations.
Regression models have been widely used as the basis for calculating
meteorologically adjusted summary statistics. These models are particularly well suited
for concentration summary statistics such as the annual mean daily maximum but are
more difficult to apply to exceedance summary statistics such as the number of days
exceeding the ozone NAAQS.
Classification models require fewer assumptions concerning the form of the
relationship between meteorological factors and ozone concentrations than do regression
models and allow for greater flexibility in defining summary statistics. They are
particularly well suited for calculating adjusted exceedance summary statistics. However,
development of the category definitions requires at least some manual inspection of data
and subjective judgment.
EPA has initiated a program (Cox and Chu, 1991) to investigate techniques for
adjusting ozone trends for meteorological influences. One of the methods being studied
is a statistical model in which the frequency distribution of ozone concentrations is
described as a function of meteorological parameters. EPA is seeking to review and
expand the technical basis for the methodology under a cooperative agreement with the
National Institute of Statistical Sciences (MISS). Preliminary results suggest that the bias
and uncertainty associated with long-trend estimates can be significantly reduced by
including meteorological covariates as parameters in the statistical modeling process.
The method used to adjust for meteorological influences on long-term ozone
trends could be easily adapted for use in calculating meteorologically adjusted exceedance
rates and design values. While such adaptations are technically feasible, the use of
adjusted exceedance rates in NAAQS attainment and adjusted design values for
classification purposes would represent a major departure from current EPA policy and
7-27
-------
NAAQS implementation guidelines. Such a major change should be considered within
the context of the ozone NAAQS review process.
REFERENCES
Breiman, L., J. H. Friedman, R. A. Ohshen, and C. J. Stone. 1984. Classification and
Regression Trees. Wadsworth, Belmont, California.
Code of Federal Regulations, 40CFR, Part 50.9 and Appendix H.
Cox, W.M. and S.H. Chu. 1991. "Meteorologically Adjusted Ozone Trends in Urban
Areas: A Probabilistic Approach". Presented at the Tropospheric Ozone and The
Environment II Air and Waste Management Association Specialty Conference.
Atlanta, GA.
Demerjian, K. L., K. L. Schere, and J. T. Peterson. 1980. Theoretical estimates of
actinic (spherically integrated) flux and photolytic rate constants of atmospheric
species in the lower troposphere. Adv. Environ. Sci. Technol.. 10:369-460.
EPA. 1979. Guideline for the Interpretation of Ozone Air Quality Standards. U.S.
Environmental Protection Agency (EPA-450/4-79-003).
EPA. 1985. "Accounting for Meteorological Variations When Interpreting Ozone Air
Quality Standards. Appendix A. Procedures for Develop ing an Index of
Meteorological Conditions Conducive to Exceedances of the Ozone NAAQS."
U.S. Environmental Protection Agency.
EPA. 1990. National Air Quality and Emissions Trends Report, 1990. U.S.
Environmental Protection Agency, Office of Air Quality Planning and Standards
(EPA-450/4-91/023).
Jones, K. H. 1985. "Urban Air Quality," in Environmental Quality 1984. Council on
Environmental Quality, Washington, D.C.
Jones, K. H. 1989. "Urban Air Quality," in Environmental Quality 1987-88. Council
on Environmental Quality, Washington, D.C.
Jones, K. H. 1992a. "The Truth About Ozone and Urban Smog." In Policy Analysis,
No. 168, February 19, 1992. Zephyr Consulting, Seattle, Washington.
7-28
-------
Jones, K. H. 1992b. "The 1990/91/92 03 Data Base and Its Implications Relative to
Currently Designated Ozone Nonattainment Area Regulatory Programs." Air and
Waste Management Association Specialty Conference, Tropospheric Ozone:
Nonattainment and Design Value Issues, Boston Massachusetts (October 27-30,
1992).
Langstaff, J. E., and A. K. Pollack. 1985. "Meteorological Characterization of High
Ozone Levels: A Pilot Study of St. Louis, Missouri." Systems Applications Inc.,
San Rafael, California.
Maul, P. R. 1980. "Atmospheric Transport of Sulfur Compound Pollutants." Central
Electricity Generating Bureau MID/SD/80/0026/R. Nottingham, England.
NRC. 1991. Rethinking the Ozone Problem in Urban and Regional Air Pollution.
National Research Council, National Academy Press, Washington, D.C.
Scire, J. S., F. W. Lurmann, A. Bass, and S. R. Hanna. 1983. "User's Guide to the
MESOPUFF-II Model and Related Processor Programs." Environmental
Research & Technology, Inc., Concord, Massachusetts.
Stoeckenius, T. E., A. D. Thrall, A. K. Pollack, and L. R. Chinkin. 1986.
"Recommendations for the Analysis, Development, and Testing of a Method for
Relating the Frequency of Occurrence of Meteorological Conditions to
Exceedances of the Ozone NAAQS." Systems Applications Inc., San Rafael,
California (SYSAPP-86/016).
Whitten, G. Z., and M. W. Gary. 1986. "The Interaction of Photochemical Processes
in the Stratosphere and Troposphere." In Effects of Changes in Stratospheric
Ozone and Global Climate. Volume 2: Stratospheric Ozone. J. C. Titus, ed.
U.S. Environmental Protection Agency.
Zeldin, M. D., and W. S. Meisel. 1978. Use of Meteorological Data in Air Quality
Trend Analysis. U.S. Environmental Protection Agency (EPA-450/3-78-024).
7-29
-------
8 ASSESSING THE IMPACT OF TRANSPORTED OZONE AND PRECURSORS
Since the early 1970's, there has been a growing awareness that ozone and ozone
precursors are transported beyond the political jurisdiction of source areas and can
significantly influence ozone concentrations in neighboring air basins located considerable
distances downwind. To address this aspect of the problem, the 1990 Clean Air Act
Amendments established a transport commission for the Northeast, and they allow the
establishment of commissions in other parts of the country. The transport of ozone
concentrations generated from urban manmade emissions of precursors in numerous areas to
locations downwind can result in elevated levels of ozone across regional spatial scales.
Studies of the relationship between high ozone concentrations hi downwind impact
zones and meteorological conditions have demonstrated that ozone peaks in areas such as
southwestern Connecticut are heavily influenced by emissions from upwind sources in
neighboring states (see, for example, Possiel et al., 1990; Sampson and Shi, 1988; Wackter
and Bayly, 1987). Transport of ozone and precursors also plays a significant role hi other
parts of the country, including the Gulf Coast region and Lake Michigan. Transport of
material between ah" basins in California has been recognized as significant by that state and
is the subject of an active research program. As is the case with federal law, the California
Clean Air Act requires upwind ah- basins to mitigate downwind impacts.
In response to Section 184(d) of the 1990 Amendments, EPA issued criteria for
assessing the role of transported ozone and precursors in ozone nonattainment areas (EPA,
1991). The Act requires States to ensure that prescribed emissions reductions are sufficient
to meet the ozone NAAQS in nonattainment areas. It is in meeting the Act's air quality
management requirements that proper consideration of transport is important. The guidance
describes the use of meteorological and air quality data to support modeling analyses,
including the use of trajectory models and the Urban Airshed Model.
CHRONOLOGY OF A MULTI-REGIONAL OZONE EPISODE, JUNE 17-20, 1987
An example of a regional scale ozone episode occurred during June 1987 and is
displayed below using a geographic information system (GIS). The system was used to
generate isopleths based on the daily maxima of hourly ozone concentrations from all sites in
the northeast and north central areas of the United States. Data were also obtained from
Canadian sites, in Ontario, during the same tune period. The Canadian data helped define
the isopleths near the Great Lakes. The GIS was used to display the isopleths as levels of
grey shading (Figures 8-1 through 8-4). These isopleths involve a certain amount of
smoothing, so that the maps provide a simplified overview and are not intended to provide
precise city specific concentrations. The episode begins on Wednesday, June 17 and ends on
Saturday, June 20.
8-1
-------
Wednesday, June 17: A strong high pressure system is located in southeastern Canada,
and the whole study area is experiencing high temperatures from the high 70s to the low 90s,
with small amounts of precipitation in New England. Ozone readings above the standard are
observed in the Chicago and Milwaukee areas and at one site in Parkersburg, West Virginia
(Figure 8-1).
Thursday, June 18: The high pressure system is now over Pennsylvania, and a strong
low pressure system has moved into southern Canada. No precipitation has been observed in
the entire region. Areas of high ozone concentrations are centered on Chicago, Milwaukee,
northwest Indiana, eastern Ohio, western Pennsylvania, and central Maryland (Figure 8-2).
Friday, June 19: The high pressure system has moved off the coast of New Jersey,
with high temperatures in the 80s to 90s. Again, no precipitation has been observed in the
entire area. High ozone concentrations are now observed in eastern Michigan, western
Pennsylvania, and from eastern Pennsylvania through the northeast corridor (Figure 8-3).
Saturday, June 20: A weak cold front moving into the area from the northwest has
reached Pennsylvania. Precipitation is observed along this front, and maximum temperatures
have reached the mid-90s to the east of the front. High ozone concentrations are observed in
central New Jersey, with very high concentrations centered on New York City (Figure
8-4).
These displays indicate an initiation of an episode in the Chicago-Milwaukee area which
is followed by a general eastward movement ending in New York City. The episode tracks
the basic meteorological events occurring during this period. By Sunday no elevated
readings of ozone are reported, and the episode has ended. Recognition of such episodes in
the Great Lakes region lead to development of the Lake Michigan Ozone Study (LMOS), a
cooperative State and Federal effort to study ambient ozone levels on a regional scale in the
Lake Michigan area. The LMOS data collection efforts consisted of a major field study
during summer 1991, which collected air quality and meteorological data.
As illustrated by this example, visualizing the regional nature of ozone transport using
monitoring data requires spatial interpolation among a discrete number of individual sites
which are generally concentrated in or near major cities. Photochemical modeling, which
produces three-dimensional predictions of ozone air quality, can provide added insight into
transport patterns. Several examples of urban ozone plumes and their interaction on a
regional scale are shown from predictions of the Regional Oxidant Model (ROM) for the
Midwest and Northeast U.S. in Figures 8-5 through 8-8. Each figure displays ROM ozone
predictions for a single hour extracted from simulations that encompass multi-day episodes.
On July 17, 1987 (Figure 8-5), urban ozone plumes extend from the major Midwest cities
downwind with the southerly wind flow on this day. Note the high ozone predicted along the
western shore of Lake Michigan and offshore over the lake resulting from urban areas along
8-2
-------
MAXIMUM ONE HOUR OZONE FOR JUNE 17, 1987
Ozone Concentrations in ppm
Figure 8-1. Isopleths of ozone daily maximum 1-hour concentrations for June 17, 1987.
8-3
-------
MAXIMUM ONE HOUR OZONE FOR JUNE 18, 1987
, 00-.06
Ozone Concentrations in ppm
D. OT-.I2 9 -13-. 16 • .17-.20
rajjlai ^^^
Figure 8-2. Isopleths of ozone daily maximum 1-hour concentrations for June 18, 1987.
8-4
-------
MAXIMUM ONE HOUR OZONE FOR JUNE 19, 1987
.00-•06
Ozone Concentrations in ppm
Figure 8-3. Isopleths of ozone daily maximum 1-hour concentrations for June 19, 1987.
8-5
-------
MAXIMUM ONE HOUR OZONE FOR JUNE 20, 1987
Ozone Concentrations in ppm
,00-.06
Figure 8-4. Isopleths of ozone daily maximum 1-hour concentrations for June 20, 1987.
8-6
-------
the western shore. Also, ozone plumes from cities along the Ohio River extend northward as
far as Detroit. In the Northeast, predictions for July 7, 1988 (Figure 8-6) reveal a
continuous "river" of moderately high ozone extending from eastern Ohio southeastward to
Richmond, VA then northeastward along the Northeast Corridor from Washington, D.C. to
coastal Maine. Ozone plumes with concentrations exceeding the level of the NAAQS are
embedded within this area over and downwind of major ozone-precursor emissions areas. In
fact, a continuous area with ozone levels at or above 0.12 ppm is predicted from southeastern
Pennsylvania into Rhode Island. Finally, comparing ozone patterns for 1400 EST and 2300
EST on July 8, 1988 (Figures 8-7 and 8-8) shows the transport of ozone exceeding 0.15 ppm
to the Boston area from sources to the southwest during this time period.
The prevailing winds along the eastern seaboard during this episode were out of the
southwest as is typical of high ozone days in this region. The long plumes of elevated ozone
concentrations stretching downwind from the major urban centers such as Washington-
Baltimore and Philadelphia suggest the extent of transport in this region.
ADJUSTING OZONE DESIGN VALUES FOR TRANSPORT
Section 181 (a) of the Clean Air Act Amendments of 1990 established ozone
classifications for nonattainment areas based on the area's air quality design value. After the
initial classification process, EPA could, at its discretion, adjust the classification if the
design value was within 5 percent of the level for another category. In making such an
adjustment, EPA could consider such factors as the number of exceedances, the level of
transport, and the mix of sources and air pollutants in the area. Given this language, and the
legislative history including air quality design value lists for urban areas, EPA determined
that no direct adjustment for transport was to be made to the ozone design value used for the
initial classifications (FR, 1991). However, it is clear that as a result of the strong influence
of transported precursors and ozone in some nonattainment areas, design values at such
locations may be heavily influenced by emission changes occurring many kilometers away in
upwind areas.
In early SIP guidance for the EKMA Model, EPA addressed the issue of how design
values are to be used in assessing the need for emissions reductions in both the downwind
and upwind areas (EPA, 1977). EPA recognized the need to identify the transport-related
component of design values that are to be used for the purpose of designing local emission
control strategies (so called "SIP control values"; see Rhoads and Tyler, 1987). These
"control strategy" design values may differ from "current air quality design values" estimated
in the table look-up procedure in that they take into account the degree to which transport of
ozone and precursors from upwind metropolitan areas contributes to ozone concentrations at
the monitoring site hi question.
8-7
-------
JUL 07, 1988 18 EST
Figure 8-5. Midwest region on July 17,
1987.
Figure 8-7. Northeast region at 1400 EST
on July 8, 1988.
JUL 08, 1988 23 EST
1 H I I I I I I II '
JLL 08 1988 14 EST
ti i ii ii in 111 MI i in 111 N ign 111 iii
Figure 8-6. Northeast region on July 7,
1988.
Figure 8-8. Northeast region at 2300 EST
on July 8, 1988.
8-8
-------
A computer model (TODV: Transported Ozone Design Value; Sabo and Hawes,
1990) has been developed to assist in determining the likely source regions associated with
high ozone concentration events. This program uses a combination of hourly ozone
concentrations and routine wind speed and direction data to calculate back trajectories and
estimate the point of origin (i.e., early morning location) of air parcels arriving at a
monitoring site at the time of the afternoon ozone peak on the 20 highest ozone days. This
information is then used to determine if the estimated design value at a monitoring site is
likely to have been strongly influenced by emissions from upwind metropolitan areas as
opposed to the metropolitan area in which the monitor is located. TODV can only provide
an approximate location of the emissions source region likely to have influenced a particular
afternoon ozone peak. No estimate of the relative contributions of upwind vs. local
emissions to the peak is provided, and back trajectory calculations based on routine wind
data can contain large uncertainties. Selection of the transport adjusted design value requires
an experienced analyst to interpret the trajectory results and concentration estimates, which
introduces a subjective element into the adjustment process. A more sophisticated evaluation
of the relative contributions of upwind and downwind areas requires application of a
photochemical grid model such as the Urban Airshed Model. The TODV model is used here
solely for the purpose of demonstrating the potential effect of transport on design values.
Application of the TODV Model to 1988-90 ozone data yielded transport adjusted
"control strategy" design values for 35 areas. As indicated in Table 8-1, these adjustments
ranged from a decrease of 0.05 ppm to increases of 0.04 ppm. Thus, these transport
adjustment can lead to both decreases and increases in an area's design value as the
downwind impact is attributed back to the source area.
As noted above, the ozone nonattainment area classifications under the 1990
Amendments were determined using the air quality design value, not a "control strategy"
value that was adjusted for transport. However, the design values for the 35 areas in Table 8-
1 provide an opportunity to ask the question, "What would be the difference in classification
if ozone design values were adjusted for transport?" Table 8-2 summarizes the comparison
between ozone area classifications that would result from the 1988-90 air quality design value
and those obtained from the transport adjusted design value. In twelve of the classified areas
(1 serious, 2 moderate, 9 marginal), the area's transport adjusted design value is less than the
level for a marginal area. Ten areas recorded increases in design values after accounting for
transport, i.e., transport was attributed back to the areas. Only one of these areas with upward
adjustments in design values would change classification; Beaumont, TX from moderate to
severe.
8-9
-------
Table 8-1. Transport Adjusted Ozone Design Values, 1988-90
Area
Atlantic City, NJ
Beaumont-Port Arthur, TX
Boston CMSA
Charlottesville, VA
Chicago CMSA
Cumberland, MD-WVA
Harrisburg-Lebanon-
Carlisle, PA
Hartford CMSA
Janesville-Beloit, WI
Johnstown, PA
Lake Charles, LA
Lancaster, PA
Lewiston-Auburn, ME
Manchester, NH
Milwaukee CMSA
Muskegon, MI
New Bedford, MA
New Haven, CT
New London-Norwich, CT-RI
New York CMSA
Philadelphia CMSA
Pittsburgh CMSA
Pittsfield, MA
Portland, ME
Portsmouth-Dover-Rochester,
NH-MA
Poughkeepsie, NY
Providence CMSA
Rockford, IL
San Diego, CA
Sharon, PA
Sheboygan, WI
Springfield, MA
Washington, DC-MD-VA
Worcester, MA
York, PA
1988-90
Design Value
0.148
0.150
0.165
0.113
0.187
0.072
0.131
0.172
0.113
0.133
0.132
0.122
0.137
0.128
0.187
0.151
0.150
0.157
0.151
0.197
0.187
0.149
0.102
0.154
0.156
0.134
0.148
0.105
0.190
0.128
0.153
0.167
0.165
0.127
0.129
Adjusted
Design Value
0.119
0.190
0.170
0.106
0.190
0.107
0.128
0.158
0.096
0.123
0.120
0.113
0.108
0.105
0.150
0.125
0.108
0.153
0.128
0.216
0.196
0.157
0.103
0.139
0.137
0.103
0.158
0.113
0.170
0.118
0.113
0.116
0.163
0.107
0.111
Increase/
(Decrease)
(0.029)
0.040
0.005
(0.007)
0.003
0.035
(0.003)
(0.014)
(0.017)
(0.010)
(0.012)
(0.009)
(0.029)
(0.023)
(0.037)
(0.026)
(0.042)
(0.004)
(0.023)
0.019
0.009
0.008
0.001
(0.015)
(0.019)
(0.031)
0.010
0.008
(0.020)
(0.010)
(0.040)
(0.051)
(0.002)
(0.020)
(0.018)
8-10
-------
Table 8-2. Comparison between Air Quality Design Value and Transport Adjusted Design Value
Derived Ozone Area Classifications, 1988-90.
Number of Areas
Ozone Classification ~
1988-90 1988-90
Ozone Design Value Transport Adjusted
Design Value
Severe
Serious
Moderate
Marginal
Not classified
Total
5
4
11
10
5
35
4
3
6
5
17
35
SUMMARY
It is clear that the transport of ozone and ozone precursors generated in one air basin
can significantly influence ozone concentrations in neighboring air basins located
considerable distances downwind. EPA has issued guidance on criteria for assessing the role
of ozone transport in ozone nonattainment. The Amendments call for the establishment of a
transport commission to study this issue. Although the Amendments specifically
acknowledge that transport across state boundaries can be an important factor during high
ozone episodes, Section 181(a) of the Clean Air Act does not include a provision for
adjusting the initial ozone nonattainment area classification for ozone transport. Section
181(a)(4) states that the "Administrator may, in the Administrator's discretion, within 90
days after the initial classification" adjust the classification // the design value in the area is
within 5 percent of another category. "In making such an adjustment, the Administrator may
consider the number of exceedances of the national primary ambient air quality standard for
ozone in the area, the level of pollution transport between the area and other affected areas,
including both intrastate and interstate transport, and the mix of sources and air pollutants in
the area."
REFERENCES
EPA. 1977. "Uses, Limitations and Technical Basis of Procedures for Quantifying
Relationships Between Photochemical Oxidants and Precursors." U.S. Environmental
Protection Agency, Research Triangle Park, North Carolina.
EPA. 1991. "Criteria for Assessing the Role of Transported Ozone/Precursors in Ozone
Nonattainment Areas." EPA-450/4-91-015. U.S. Environmental Protection Agency,
Research Triangle Park, North Carolina.
8-11
-------
Federal Register. November 6, 1991. 40CFR Part 81, "Air Quality Designations and
Classifications; Final Rule." 56 FR 56694-56858.
House Committee Report, H.R. No. 101-490, Part 1, 101st Congress, second Session.
Possiel, N.C., D.C. Doll, et. al. 1990. "Impacts of Regional Control Strategies on Ozone
in the Northeastern United States." Air and Waste Management Association 83rd
Annual Meeting, Pittsburgh, Pennsylvania.
Rhoads, R.G. and D.D. Tyler. 1987. "Distinction between 'Ozone Design Value' and 'SIP
Control Value for Ozone'", U.S. Environmental Protection Agency, Research
Triangle Park, North Carolina.
Sabo, E.J., and J.T. Hawes. 1990. "User's Guide and Program Documentation for the
Transported Ozone Design Value Model", Prepared under EPA Contract Number 68-
02-4393, U.S. Environmental Protection Agency, Research Triangle Park, North
Carolina.
Sampson, P.J., and B. Shi. 1988. "A Meteorological Investigation of High Ozone Values
in American Cities." Department of Atmospheric, Oceanic, and Space Sciences,
Space Physics Research Laboratory, The University of Michigan, Ann Arbor,
Michigan.
Wackter, D.J., and P.V. Bayley. 1987. "The Effectiveness of Connecticut's SIP on
Reducing Ozone Levels from 1976 through 1987." Air Pollution Control Association
Specialty Conference, "Scientific and Technical Issues Facing Post-1987 Ozone
Control Strategies." Hartford, Connecticut.
8-12
-------
9 DETECTING TRENDS IN OZONE DESIGN VALUES
This section presents a literature review of articles on the estimation of trends in
ozone design values. As discussed in Section 2, design value for ozone can be defined hi
statistical terms as the concentration that is exceeded once per year on average.
Procedures for estimating the design value from ambient ozone concentrations were
described in Section 4. For many of the articles in the present literature review, the
design value for a given site and year is estimated by the annual highest or second highest
daily maximum hourly ozone concentration.
Unless otherwise noted in this review, the term "trend" refers to year-to-year
variations in the design value. Ozone design values are typically estimated from three
years of data rather than on a single-year basis. However, it is conceptually easier to
evaluate trends in single-year design value estimates because estimates based on
overlapping multi-year periods are correlated. The assumed trend at some sites may be
realistically represented as a constant increase between one year and the following year,
or as a constant percentage increase. However, ozone data at some sites have shown
long-term increases and long-term decreases for different periods; such nonlinear ozone
trends may be due to nonlinear meteorological or emissions trends. The important role
of meteorology in ozone formation and trends was addressed in detail in Section 7.
Although meteorological variability is not the primary focus of this section, it will be
addressed within the context of trends detection.
The methods that have been used for trend detection are in many cases applicable
to any annual summary statistic, rather than being restricted to ozone design values.
Therefore, this literature review includes articles that discuss trend detection techniques in
ozone design values, annual summary statistics and design values for other air pollutants,
water quality data, and stratospheric ozone data. Furthermore, various articles on ozone
trend detection methods have been based on trends hi the annual exceedances of the
ozone NAAQS. Although the numerical trend rate for the design value is not simply
related to the numerical trend in the annual exceedance rate, a decreasing rate of
exceedances implies a decreasing design value; thus approaches based on trends hi
exceedance rates are included in this literature review.
This summary review begins with a discussion of the definition of the annual
design value for a single site or group of sites when there is an annual trend (variation
from year to year). Later sections discuss linear regression and general linear model
parametric methods of trend estimation, nonparametric methods, methods based on
extreme value theory approximations, and methods that use time-series models. A
tabular summary of the papers cited in this literature review is provided in Appendix H.
The articles are listed by the same reference number (hi brackets) used hi this summary.
The ordering is fairly arbitrary except that related papers tend to appear consecutively.
9-1
-------
STATISTICAL APPROACHES TO TREND ANALYSES
Most of the trend analyses in the cited articles use only annual summary statistics
in trend estimation. Thus the trend represents overall changes from year to year, and the
annual summary statistics incorporate seasonal variation. If, instead, the trend analysis is
based on fitting a model to monthly, daily, or hourly concentrations, then the general
approach taken in the cited articles is to assume that the concentration distribution varies
both seasonally (within the year) and annually (across years). This is usually achieved by
allowing the parameters of the fitted distribution to depend on the calendar month and on
the year.
In most cases the trend analyses ignore the year-to-year serial dependence; the
design values for two consecutive years are assumed to be independent. A general
problem is that it is very difficult to distinguish between a trend (not necessarily linear)
and serial dependence since both phenomena can lead to very similar concentration series.
It is important to recognize a distinction between methods of trend detection and
trend estimation. A trend detection method simply determines whether or not a trend
exists. A trend estimation method estimates the size of the trend effect. A trend estimate
can often be used for trend detection using the rule that an estimated non-zero trend
implies that a trend exists; a trend is usually considered "non-zero" if the appropriate
confidence interval for the trend does not contain zeio. The distinction is more important
for nonparametric methods rather than parametric methods. Several nonparametric
methods (such as the Spearman's rho and chi-square tests described in [18]) are used for
trend detection but cannot be used for trend estimation.
Trend analyses often use averages or sums of annual summary statistics across
multiple sites. A trend in the composite average across sites does not imply the same
trend at every site. Further, it is possible that the site averages do not show a significant
trend but there are large increasing and decreasing trends at many sites. One useful
result is that the central limit theorem implies that in cases where a large number of site
design values are combined, the distribution of the sum or average is approximately
normal. The normality of the annual summary statistic is an assumption made for the
linear regression and general linear model analyses in the cited articles.
Various types of trends are considered in the cited articles. The simplest type is a
linear trend, which means that the annual summary statistic increases or decreases on
average by the same amount every year. In this case the mean annual summary statistic
is a linear function of the calendar year. In some cases higher order polynomial trend
functions for the annual mean have been used, such as a quadratic or cubic functions.
These results are usually displayed by simply plotting the observed annual means against
the year and superimposing the estimated trend function (a straight line for linear trends).
9-2
-------
Instead of making specific assumptions about the form of the trend, a more
general approach allowing for arbitrary year effects has also been used. In this approach,
a trend is indicated by statistically significant differences between the mean annual
summary statistics for a pair of years. An often useful plot is given by graphing the
annual means with simultaneous confidence intervals defined such that two means are
significantly different if the corresponding confidence intervals do not overlap ([1, 15,
22, 32-40].) For example, Figure 9-1 (from [15]) illustrates simultaneous confidence
intervals for four years of data. Since the plotted confidence intervals overlap for years 1
and 2 but not for years 1 and 3, years 1 and 2 are not significantly different, but years 1
and 3 are significantly different.
COMPOSITE MEAN
RELATIONSHIPS (MULTIPLE COMPARISONS)
YEARS 1 AND 2 ARE NOT SIGNIFICANTLY
DIFFERENT.
YEARS 2 AND 3 ARE NOT SIGNIFICANTLY
DIFFERENT.
YEARS 1 AND 3 ARE SIGNIFICANTLY
DIFFERENT.
YEAR 4 IS SIGNIFICANTLY DIFFERENT FROM
ALL OTHERS.
95% CONFIDENCE
INTERVAL ABOUT
COMPOSITE MEAN
YEAR1
YEAR 2
YEAR 3
YEAR 4
Figure 9-1. Sample illustration of the use of confidence intervals to determine
statistically significant changes.
9-3
-------
The Tukey studentized range technique used to derive the simultaneous confidence
intervals in Figure 9-1 was described by Pollack and others ([15, 22]). Since the
number of possible simultaneous comparisons is k(k—1)12, where k is the often large
number of years of data analyzed, testing each pair at the usual 5 percent significance
level would lead to a high probability that at least one difference would be declared
significant when in fact there are no real trends (since each comparison would then have
a 5 percent probability of being declared statistically significant). To treat this problem,
the confidence intervals were computed in such a way that if there is no real trend, the
probability that every pair of annual confidence intervals overlaps is 95 percent. Thus
the probability of erroneously determining one or more significant differences is 5
percent. The same Tukey studentized range technique is applicable in other cases where
a general linear model is used to estimate the year effects.
If the assumptions of normal distributions and constant variances for every year
implicit in the general linear model are not satisfied, then the Bonferroni approach to
simultaneous confidence intervals has been used ([15, 22, 32-40]). For example, the
constant variance assumption is inappropriate for the number of annual exceedances of
the ozone NAAQS. The Bonferroni approach is used in these situations to compute
confidence intervals such that the probability of erroneously determining one or more
significant differences is 5 percent or less. In general these Bonferroni intervals are
wider than the unknown width needed to exactly attain an overall 5 percent error
probability, i.e., the Bonferroni intervals are an upper bound approxim?tion to the exact
95 percent simultaneous confidence intervals.
When considering trends in ozone design values, the effects of varying
meteorology are important since these effects can often mask underlying trends in the
design values that may be attributable to hydrocarbon or nitrogen oxides emissions
reductions. Various methods have been used to adjust annual summary statistics for
meteorological effects prior to the trend estimation so that the effects of the
meteorological "noise" is reduced. These methods are used in some of the papers cited
in this literature review ([4, 5, 7-10]) but were described in greater detail in Section 7.
The paper by Wackter and Bayly ([10]) also includes adjustments for emissions effects.
This literature review focuses on methods of detecting trends in annual summary
statistics, regardless of any adjustments made to the annual summary statistics to account
for these meteorological or emissions effects.
Linear Model Approaches
One of the simplest methods of trend detection and estimation is simple linear
regression ([4, 5, 10, 11, 18]). In this method the expected annual summary statistic is
assumed to be linear in the year, so the estimated means plot as a straight line. This
statistical model is usually fitted by least squares, which is defined as finding the straight
line such that the squared errors about that line (squared differences between the observed
annual summary statistic and the estimate from the straight line trend) are minimized.
9-4
-------
That method is most appropriate when the annual summary statistics are normally
distributed with a constant variance. For ozone concentration data at a single site, the
assumption of normality may be less tenable than an assumption of log-normality (see the
literature review on statistical distributions), so one can apply simple linear regression to
the logged concentrations and then transform from the straight line in log space back to
an exponential curve for the mean concentrations ([4, 5]).
Simple linear regression can be used for annual summary statistics from a single
site or from a group of sites. If the data set is complete, i.e., if data are available for
every site every year, then the straight line fitted by the least-squares method to the
complete data set (with one value for each site/year combination) is the same as the line
fitted to the yearly means (with one value for each year). In effect the model would
assume that for any given year the values for the different sites came from a single
distribution. However, if the assumptions of independent, normally distributed errors
about the trend line and equal variances for each site/year statistic apply, then the
complete data least-squares analysis provides much better estimates of the confidence
intervals for the fitted trend line. Under these assumptions it follows that there is an
equivalence between the use of simple linear regression for trend estimation and the use
of the Pearson correlation coefficient for trend detection: The slope of the straight line is
significantly different from zero (indicating a non-zero trend) only if the Pearson
correlation coefficient is significantly different from zero.
An alternative to simple linear regression that does not make the assumption of a
linear trend is one-way analysis of variance ([15]). In this model the underlying mean
value for a given site/year concentration depends only on the year but is an arbitrary
value for each year.
If annual summary statistics are available for every site, a better approach than
simple linear regression or one-way analysis of variance is the use of two-way analysis of
variance ([1, 2, 3, 15, 22, 32-40]). The two-way analysis of variance method assumes
that the annual summary statistic for a given site and year is the sum of a site effect, a
year effect, and a normally distributed error term. The errors are assumed to be
independent, with mean zero and a constant variance. The main advantage of this
method is that it accounts for the dependence between annual composite site averages
caused by site effects. Simple linear regression and one-way analysis of variance ignore
the site-to-site variation by assuming this variation is part of the error; these site effects
introduce dependencies between the errors for the same site. A significant trend
determined from the two-way analysis of variance corresponds to a statistically significant
year effect. The different year effects can be displayed by plotting the composite site
averages for each year with Tukey simultaneous confidence intervals (as in Figure 9-1).
The two-way analysis of variance method can also be applied in cases where some
annual summary statistics are missing for some sites (often because the available data are
insufficient to satisfy the validity criteria for the annual summary statistic). In this case
9-5
-------
the approach is equivalent to using the general linear model on the available data to
estimate the site and year effects, and then estimating any missing value as the sum of the
estimated site and year effects corresponding to the particular missing site and year ([!]).
One disadvantage of this method is that data from all sites are used to fill in the missing
values, rather than data from local sites. Cohen and Pollack ([2]) provide an extension of
the two-way analysis of variance approach that deals with this problem by allowing the
year effects to depend on the region. Regions are defined by combining nearby sites in
such a way that the mean squared error in the fitted general linear model is minimized.
The two-way analysis of variance approach can be modified to allow for specific
trend functions by assuming that the year effects are certain functions of the year, but the
site effects are arbitrary. The case of a linear trend but arbitrary site effects is used by
Capel and others ([3]).
Nonparametric Methods
The assumptions of independent normally distributed errors with the same constant
variance used in several trend analyses may not be applicable. To test for and estimate
the trend without making such distributional assumptions, various nonparametric methods
have been used. Nonparametric methods proposed in the literature include a simple chi-
square test of a step trend ([18]), the use of Spearman's rho ([7, 8, 9, 12, 18]), and
Kendall's tau ([13, 14, 21]) for trend detection, and the use of the Theil/Sen slope
estimator for linear trend estimation ([13, 21]). Since only very general distributional
assumptions are made (concerning the dependence structure), the results are valid under
very general conditions but the methods have lower power (trend detection probability)
compared to parametric tests in cases where the parametric model assumptions are
reasonable approximations. The nonparametric tests can have much greater power than
parametric tests when the distributional requirements of the parametric test are violated
([12]). A disadvantage of the nonparametric methods used in the cited articles is that
likely site-to-site variability is not included in the analyses.
The chi-square test of trend ([18]) is a simple test to compare exceedance rates for
two different years. A simple two-by-two table is created giving the number of NAAQS
exceedance days and NAAQS non-exceedance days for each year. If there were no
trend, then the proportions of exceedance days per year would be approximately equal for
both years. The differences between the observed numbers of exceedance days and the
expected numbers in the case of no trend can be used to compute a chi-squared statistic.
Because of the minimal amount of information used to compute this trend test statistic,
the test has the disadvantage of having a very low trend detection probability which hi
most cases outweighs the advantage of simplicity.
The Spearman's rho test of trend ([7, 8, 9, 12, 18]) is based on Spearman's rho
statistic, which is the standard Pearson correlation coefficient between the rank of the
annual summary statistics and the year. The rank is 1 for the highest summary statistic,
9-6
-------
2 for the second highest, and so on. If there is no trend and all observations are
independent, then all rank orderings are equally likely. This fact is used to calculate the
statistical significance of the Spearman's rho statistic; a value significantly different from
zero implies a significant trend. If ties in the annual summary statistics are present, then
the significance level has to be adjusted to account for the number of ties. In paper 12 a
comparison of the power (trend detection probability) of Spearman's rho with the power
of simple linear regression shows that the nonparametric test can be almost as efficient as
the simple linear regression t test even when the normality assumption holds. The linear
regression power calculations hi [12] are based on formulae that are incorrect for small
samples but approximately correct for large samples (see formula 3b hi [12]). Thus the
reported results in that paper may be Inaccurate for small samples and should be used
with caution.
Kendall's tau is an alternative nonparametric statistic that can be used to test for
trend ([13, 14, 21]). This statistic can be calculated as the number of possible pairs of
years for which the ordering of the years is the same as the ordering of the annual
summary statistics (the lower annual statistic occurs hi the earlier year) less the number
of possible pairs of years with the reverse ordering. If there is no trend and all
observations are independent, then all rank orderings of the annual statistics are equally
likely; this result is used to compute the statistical significance of the tau statistic.
Adjustments for tied annual summary statistics are described in the cited articles.
Adjustments of Kendall's tau for seasonality ([13]) and serial dependence ([14])
have been proposed and investigated in the context of water quality data analysis. A
seasonally adjusted Kendall's tau ([13]) allows for different annual means and trends hi
different calendar months by adding up the 12 Kendall's tau statistics from each month.
In paper 13 the null distribution of this statistic (when there are no trends) is calculated
assuming that values from different calendar months are independent. In paper 14 the
null distribution is calculated assuming that values hi different months can be correlated.
Both papers include calculations of the power of these tests for simulated data. The
power of the seasonal and serial dependence adjusted Kendall's tau is greater than the
power of the simpler seasonal dependence adjusted Kendall's tau if there is serial
dependence, but is less in the independent case. One difficulty hi application of these
approaches to ozone design values is that the ozone design value depends upon the
highest daily maximum ozone concentrations, and these are more likely to occur in
certain calendar months. Possible trends in the maximum concentrations for those
months are much more important than trends for other months that tend to have lower
maximum concentrations.
Kendall's tau test of trend is related to the Theil/Sen nonparametric slope
estimator ([21]), which gives an estimate of the assumed linear trend. This estimator is
the median of all possible ratios of the change in the annual summary statistic from one
year to a later year divided by the number of years separating the two values. If the
trends differ by calendar month, then the same calculation can be applied to the monthly
9-7
-------
summary statistics by only considering ratios for values in the same month, i.e., that
differ by an exact multiple of 12 months ([13]).
Extreme Value Theory Approaches
The theory of extreme values is summarized in Section 4 and Appendix A. This
theory can be used to estimate the distribution of the annual maximum hourly
concentration and/or the second up to the Mi highest daily maximum hourly
concentration, and to estimate the distribution of the number of days for which the daily
maximum exceeds a high threshold (that may or may not be the ozone NAAQS). This
section considers some trend analyses based on these approximations. We first present
the use of the Poisson process approximation for the daily exceedances. This relatively
simple approach is now being routinely applied for the EPA Trends Reports ([32-40]).
Other applications of extreme value theory found in this literature review are also
presented. These other methods are potentially very useful in analyzing ozone design
value trends but may be too complex for routine application.
The simplest approach that uses extreme value theory is based on the result that
exceedance days follow a Poisson process in the limit, provided that the dependence
between daily concentrations separated by a given number of days decreases sufficiently
fast as the separation increases. Assuming that the numbers of exceedances for different
sites are approximately independent, it follows that the total number of exceedance days
for a given year (summed over the sites) approximately have a Poisson distribution. A
more accurate, but more complex, approach does not assume independence of
exceedances between sites, which leads to a compound Poisson rather than Poisson
distribution for the total number of exceedances. The Poisson distribution has a variance
equal to the mean, and the maximum likelihood estimate of this parameter is the observed
number of exceedances. If annual exceedances are averaged across a large number of
sites, then the annual average number of exceedances per site are approximately normally
distributed, with a mean estimated by the annual average number of exceedance days per
site and a variance estimated by the annual average number of exceedance days per site
divided by the number of sites. This result can be used to calculate simultaneous
confidence intervals for the annual mean number of exceedances per site using the
Bonferroni method, as described previously. (The Tukey studentized range method is not
applicable in this case because the variance varies from year to year.) This approach is
derived in paper 15 and has been applied in several of the annual EPA Trends Reports
([32-40]).
Shively ([20]) used a related approach to estimate the long-term trend hi ozone
exceedance rates for Houston daily maxima. The sequence of daily exceedances of the
selected high ozone threshold was modeled as a nonhomogeneous Poisson process. Thus
the exceedances were assumed to follow a Poisson process with a rate that was not
constant. The logarithm of the exceedance rate for a given day is the sum of multiples of
9-8
-------
certain meteorological measurements for that day and of a multiple of the calendar year.
The calendar year multiple gives the estimated trend.
In another paper, Shively ([19]) used the limiting joint extreme value distribution
for the k highest daily maximum hourly ozone concentrations for each of several years.
This limiting distribution assumes that all daily maxima are approximately independent,
and that for each year the daily maxima have the same distribution. These
approximations may not be sufficiently realistic. The location parameter (a parameter
related to the mean of the limiting distribution) was assumed to change linearly with the
year. The maximum likelihood method (described in the literature review on statistical
distributions) was used to estimate the parameters, but a bootstrap method was used to
determine the statistical significance of the trend, since the amount of data used in the
analysis was too small to apply asymptotic theory for the significance test.
The methods of Smith ([6]) use the latest advances in extreme value theory to
derive a very complete description of the sequence of daily maxima that incorporates the
most general limiting extreme value distribution for the upper tail, the possible clustering
of exceedances, seasonal trends (within year), and annual trends (across years). Since
exceedance days often cluster together in cases of strong serial dependence (see the time
series literature review for more details), Smith fitted the trend model to all hourly ozone
concentrations greater than a high threshold separated in time by more than a cluster
interval; if more than one hourly exceedance of the threshold occurred within the cluster
interval, only the highest of the cluster exceedances was used to fit the model. To fit
ozone data from Houston, Texas (1973-86) various thresholds (0.08, 0.10, 0.12, 0.16,
0.20, 0.26, 0.28, and 0.30 ppm) and two alternative cluster intervals (24 and 72 hours)
were used with somewhat different results.
According to the limiting extreme value theory model, the cluster exceedances
occur according to a Poisson process, and the distribution of the cluster maximum
concentration is the tail generalized Pareto distribution (GPD). The tail GPD has a
location, scale, and shape parameter. To treat seasonality, which is variation within the
year, the scale and shape parameters differ by the calendar month (or pah" of months),
but are the same for every year. To treat trend, which is variation from year to year, the
location parameter was assumed to be an Intercept plus a slope parameter multiplied by
the calendar year; both intercept and slope vary by calendar month (or pair of months).
This complex model was fitted by the maximum likelihood method.
In one application of the model at a relatively low threshold, the annual trend was
only significant for February and June. At higher thresholds the trend was examined by
comparing the fitted model with no trend term for 1973-80 with the fitted model with no
trend term for 1981-86. The results showed that the estimated number of exceedances of
each high threshold was lower in the later period, which implies a decreasing trend hi the
high concentrations. However, Smith found no such trends for the low concentrations.
9-9
-------
The extreme value theory model in Smith's paper ([6]) is the most realistic
application of extreme value theory to ozone data since it incorporates serial dependence,
seasonal dependence, annual trends, and the most general limiting extreme value
distribution. This is much more complicated than the independent tail exponential
distribution model that is recommended for routine use over short monitoring periods (see
the literature review on statistical distributions). The routine application of Smith's
model by air quality managers is difficult because of the computational difficulties in
fitting the model and the very difficult problem of selecting reasonable choices for the
threshold and cluster interval; Smith's selections for Houston are likely to be
inappropriate for many other cities. An even more complete analysis would include
terms representing meteorological effects into the extreme value theory model and allow
for nonlinear trend functions.
The ad hoc approaches of Smith [6] have been investigated on a more formal theoretical
basis in several recent papers. In particular, the mean cluster size is the reciprocal of the
"extremal index," described in the textbook by Leadbetter and others (1983) and the
focus of this theoretical research. Relevant papers are Leadbetter (1991), Leadbetter and
others (1989), Rootzen and others (1994), and Smith and Weisman (1994).
TIME-SERIES MODELS
The extreme value model of Smith ([6]) is an example of a model that includes
trends and serial dependence (i.e., the mean design value is not assumed to be constant
and the hourly ozone concentrations are not assumed to be independent). In principle,
any of the time series models described in the literature review on time-series models
could be adapted to treat trends using a modification which allows some or all of the
parameters of the daily maximum hourly ozone distribution to vary from year to year.
Analyses of this type are described in the time-series literature review. Time-series
methods including trend analyses have also been applied to stratospheric ozone ([16, 17]);
these methods could be adapted to treat tropospheric ozone daily maxima instead of
monthly average stratospheric ozone measurements. The analysis in paper 16 is a
frequency domain analysis. Such an analysis represents the time series in a more indirect
manner as a combination of frequency distributions (low frequencies correspond to long-
term trends). A more directly interpretable time-series analysis in the time domain ([17])
represents the series as a first-order autoregressive time series plus sinusoidal terms
representing seasonal variation within the year plus a multiple of the number of months
from the initial month.
9-10
-------
OZONE TRENDS ADJUSTED FOR METEOROLOGY
The important role that meteorological conditions play in determining ground-level
ozone concentrations was discussed in Section 7. Because annual variations in these
conditions can be substantial, year-to-year fluctuations in ozone concentrations can be
quite large. The effect of such variations can be to mask any long-term trends in ozone
that might be occurring as a result of changes hi precursor emissions.
Methodologies developed to account for meteorological variability within the
context of trends assessments were also presented hi Section 7. EPA's latest Trends
Report (41) featured a comparison of the unadjusted national ozone ah" quality trend with
the composite trend hi 43 metropolitan areas based on the model developed by Cox and
Chu (42). The "adjusted" trend indicator shown hi Figure 9-2 is the composite mean of
the meteorologically adjusted 99th percentile daily maximum 1-hour concentrations across
each of the 43 individual metropolitan areas. The smoothing introduced by the
meteorological adjustment is especially evident in the ozone trends where the peak ozone
years, such as 1983 and 1988, have been followed by years less conducive to ozone
formation. The general pattern is clear, a steady downward trend. The composite
average of the 99th percentile daily maximum 1-hour concentrations hi 1992 is 10 percent
lower that the 1983 level. This composite trend captures the spatial and temporal
variability hi meteorological conditions among these 43 metropolitan areas. As illustrated
by this figure, the composite trend in the unadjusted 99th percentile daily maximum 1-
hour concentration for these 43 metropolitan areas tracks the national composite ozone
trend hi the second highest daily maximum 1-hour concentration. Thus, the
meteorologically adjusted trend is likely to be a reasonable indicator of the composite
national ozone trend.
SUMMARY AND CONCLUSIONS
For routine trend analyses of data from a single site, simple linear regression of
the annual maximum or second maximum of the daily maximum hourly concentrations
against the year is convenient and relatively easy to understand. Nonparametric
alternatives using Spearman's rho and Kendall's tau for trend detection, and the Sen/Theil
slope estimate for trend estimation, can also be applied to determine the extent to which
the results of the parametric trend analysis depend on the normality assumption. With a
long series of hourly ozone data at a single site, the complex extreme value theory model
of Smith ([6]) yields a more realistic analysis but one that requires a substantial
computing and interpretative effort by the investigator. When averages of ozone design
values across multiple sites are taken, the central limit theorem provides some theoretical
justification for the two-way analysis of variance with site and year effects. Useful year-
to-year comparisons can be made using Tukey's simultaneous confidence intervals. To
analyze trends hi total exceedance days for multiple sites, the use of the Poisson
approximation to the distribution and Bonferroni simultaneous confidence intervals is
9-11
-------
Concentration, ppm
U.1B
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
1
. '
--.. ..""'**•
-• '4
I
Met Adjusted Trend-43 MSA's
(99th percentile daily max 1 -hr cone.)
X* i
x - — L
Unadjusted Ozone Trend-43 MSA's^x^
•»,(99th percentile daily max 1-hr cone.) ^"^^
**»..
National Composite Mean Ozone Trend "*••...
(Annual 2nd Daily Max 1-hr)
Actual (43 MSA's) Met Adjusted (43 MSA's) National (509 sites) |
(99th Percentile) (99th Percentile) (2nd Daily Max 1-hr) |
i i i i I i
983 1984 1985 1986 1987 1988
I I I I
1989 1990 1991 1992
Figure 9-2. Comparison of meteorologically adjusted and unadjusted trends in the
composite average of the second highest maximum 1-hour concentration for
43 MSAs, 1983-92.
9-12
-------
preferable to the two-way analysis of variance because the variance of the number of
exceedances usually varies from year to year.
The important role of meteorology in determining the year-to-year variability in
ozone concentrations cannot be ignored. EPA has initiated a cooperative program to
review and expand the development of models for meteorological adjustment. Based on
initial encouraging results, EPA incorporated a comparison of meteorologically adjusted
and unadjusted ozone trends in 43 metropolitan areas in its 1992 Trends Report.
REFERENCES
EPA. 1979. Guidelines for the Interpretation of Ozone Air Quality Standards. U.S.
Environmental Protection Agency, Research Triangle Park, North Carolina
(450/4-79-003).
Leadbetter, M. R., G. Lindgren, and Rootzen. 1983. Extremes and Related Properties of
Random Sequences and Processes. Springer-Verlag, pp 127-129.
Leadbetter, M. R. 1991. On a basis for extremes over threshold modeling. Stat. &
Prob. Letters. 12.
Leadbetter, M. R., I. Weissman, L. de Haan, and H. Rootzer. 1989. On clustering of
high levels in statistically stationary series. Proc. 4th Int. Meeting Statist.
Climatology (ed., J. Sansom). Wellington; New Zealand Meteorological Service.
Rootzer, H., M. R. Leadbetter, and L. de Haan. 1994. "Tail and Quantile Estimation
for Strongly Missing Stationary Sequences." Technical Report.
Smith, R. L., and I. Weissman. 1994. Estimating the extremal index. J. Royal Statist.
Soc. B. 56:3.
Numbered References
1. Pollack, A. K., and T. S. Stocking. 1989. "General Linear Models Approach to
Estimating National Air Quality Trends." Systems Applications, Inc., San Rafael,
California (SYSAPP-89/098).
2. Cohen, J. P., and A. K. Pollack. 1991. "General Linear Models Approach to
Estimating National Air Quality Trends Assuming Different Regional Trends."
Systems Applications International, San Rafael, California (SYSAPP-91/035).
3. Capel, J., T. R. Johnson, and T. McCurdy. 1983. "Analysis of Ozone Trends
for Selected Indices of Daily Maximum Air Quality Data." Air Pollution Control
Association Annual Meeting, Atlanta, Georgia (June 19-24, 1983).
9-13
-------
15. Pollack, A. K., W. F. Hunt, Jr., and T. C. Curran. 1984. "Analysis of
Variance Applied to National Ozone Air Quality Trends." 77th Annual Meeting
of the Air Pollution Control Association, San Francisco, California (June 24-29,
1984).
16. Bloomfield, P., G. Oehlert, M. L. Thompson, and S. Zeger. 1983. A frequency
domain analysis of trends in Dobson total ozone records. J. Geophysical Res..
88(C13):8512-8522.
17. Reinsel, G., G. C. Tiao, M. N. Wang, R. Uwis, and D. Nychka. 1981.
Statistical analysis of stratospheric ozone data for the detection of trends. Atmos.
Environ.. 15(9): 1569-1577.
18. EPA. 1974. Guideline for the Evaluation of Air Quality Trends. Office of Air
Quality Planning and Standards, U.S. Environmental Protection Agency.
19. Shively, T. S. 1990. An analysis of the long-term trend in ozone data from two
Houston, Texas monitoring sites. Atmos. Environ.. 24B(2):293-301.
20. Shively, T. S. 1991. An analysis of the trend in ground-level ozone using
nonhomogeneous Poisson processes. Atmos. Environ.. 25B(3):387-395.
21. Freas, W. A., and E. Sieurin. 1977. "A Nonparametric Calibration Procedure
for Multi-Source Urban Air Pollution Dispersion Models." Fifth Conference on
Probability and Statistics in Atmospheric Sciences, American Meteorological
Society, Las Vegas, Nevada.
22. Pollack, A. K., and W. F. Hunt 1985. "Analysis of Trends and Variability in
Extreme and Annual Average Sulfur Dioxide Concentrations." Air Pollution
Control Association Specialty Conference on "Quality Assurance in Air Pollution
Measurements," Boulder, Colorado.
23. EPA. 1973. The National Air Monitoring Program: Ah" Quality and Emissions
Trends - Annual Report. U.S. Environmental Protection Agency, Office of Air
Quality Planning and Standards, Research Triangle Park, North Carolina
(450/1-73-OOla and b).
24. EPA. 1973. Monitoring and Air Quality Trends Report. 1972. U.S.
Environmental Protection Agency, Office of Air Quality Planning and Standards,
Research Triangle Park, North Carolina (450/1-73-004).
25. EPA. 1974. Monitoring and Ah- Quality Trends Report. 1973. U.S.
Environmental Protection Agency, Office of Air Quality Planning and Standards,
Research Triangle Park, North Carolina (450/1-74-007).
9-15
-------
4. Chock, D. P., S. Kumar, and R. W. Herrmann. 1982. An analysis of trends in
oxidant air quality in the South Coast Air Basin of California. Atmos. Environ..
16(ll):2615-2624.
5. Kumar, S., and D. P. Chock. 1984. An update on oxidant trends in the South
Coast Air Basin of California. Atmos. Environ.. 18(10):2131-2134.
6. Smith, R. L. 1989. Extreme value analysis of environmental time series: An
application to trend detection in ground-level ozone. Statist. Science. 4:367-393.
7. Kolaz, D. J., and R. L. Swinford. 1989. "Ozone Trends in the Greater Chicago
Area." Ozone Conference on "Federal Controls for Ozone Around Lake
Michigan," Lake Michigan States' Section and Wisconsin Chapter of the Air and
Waste Management Association (October 12-13, 1989).
8. Kolaz, D. J., and R. L. Swinford. 1988. "Ozone Air Quality: How Does
Chicago Rate?" 81st Annual Meeting of the Air Pollution Control Association,
Dallas, Texas (June 1988).
9. Sweitzer, T. A., and D. J. Kolaz. 1984. "An Assessment of the Influence of
Meteorology on the Trend of Ozone Concentrations in the Chicago Area." Air
Pollution Control Association Specialty Conference on "Quality Assurance in Air
Pollution Measurements," Boulder, Colorado (October 14-18, 1984).
10. Wackter, D. J., and P. V. Bayly. 1987. "The Effectiveness of Connecticut's SIP
on Reducing Ozone Levels from 1976 through 1987." Ah" Pollution Control
Association Specialty Conference on "The Scientific and Technical Issues Facing
Post-1987 Ozone Control Strategies," Hartford, Connecticut (November 1987).
11. SCAQMD. 1991. "Final Air Quality Management Plan 1991 Revision. Final
Appendix II-B: Air Quality Trends in California's South Coast and Southeast
Desert Air Basins, 1976-1990." South Coast Air Quality Management District.
12. Lettenmaier, D. P. 1976. Detection of trends in water quality data from records
with dependent observations. Water Resources Res.. 12(5)-.1037-1046.
13. Hirsch, R. M., J. R. Slack, and R. A. Smith. 1982. Techniques of trend
analysis for monthly water quality data. Water Resources Res. , 18(1): 107-121.
14. Hirsch, R. M., and J. R. Slack. 1984. A nonparametric trend test for seasonal
data with serial dependence. Water Resources Res.. 20(6):727-732.
9-14
-------
37. EPA. 1989. National Air Quality and Emissions Trends Report. 1987. U.S.
Environmental Protection Agency, Office of Air Quality Planning and Standards,
Research Triangle Park, North Carolina (450/4-89-001).
38. EPA. 1990. National Air Quality and Emissions Trends Report. 1988. U.S.
Environmental Protection Agency, Office of Air Quality Planning and Standards,
Research Triangle Park, North Carolina (450/4-90-002).
39. EPA. 1991. National Air Quality and Emissions Trends Report. 1989. U.S.
Environmental Protection Agency, Office of Air Quality Planning and Standards,
Research Triangle Park, North Carolina (450/4-91-003).
40. EPA. 1991. National Air Quality and Emissions Trends Report. 1990. U.S.
Environmental Protection Agency, Office of Air Quality Planning and Standards,
Research Triangle Park, North Carolina (450/4-91-023).
41. EPA. 1993. National Air Quality and Emissions Trends Report. 1992. U.S.
Environmental Protection Agency, Office of Air Quality Planning and Standards,
Research Triangle Park, North Carolina (454/R-93-031).
42. Cox, W.M. and S.H. Chu. 1991. "Meteorologically Adjusted Ozone Trends in
Urban Areas: A Probabilistic Approach". Presented at the Tropospheric Ozone
and The Environment II Air and Waste Management Association Specialty
Conference. Atlanta, GA.
9-17
-------
26. EPA. 1974. Monitoring and Air Quality Trends Report. 1974. U.S.
Environmental Protection Agency, Office of Air Quality Planning and Standards,
Research Triangle Park, North Carolina (450/1-76-001).
27. EPA. 1975. National Air Quality and Emissions Trends Report. 1975. U.S.
Environmental Protection Agency, Office of Air Quality Planning and Standards,
Research Triangle Park, North Carolina (450/1-76-002).
28. EPA. 1977. National Air Quality and Emissions Trends Report. 1976. U.S.
Environmental Protection Agency, Office of Air Quality Planning and Standards,
Research Triangle Park, North Carolina (450/1-77-002).
29. EPA. 1978. National Air Quality and Emissions Trends Report. 1977. U.S.
Environmental Protection Agency, Office of Air Quality Planning and Standards,
Research Triangle Park, North Carolina (450/2-78-052).
30. EPA. 1981. 1980 Ambient Assessment - Air Portion. U.S. Environmental
Protection Agency, Office of Air Quality Planning and Standards, Research
Triangle Park, North Carolina (450/4-81-014).
31. EPA. 1983. National Air Quality and Emissions Trends Report. 1981. U.S.
Environmental Protection Agency, Office of Air Quality Planning and Standards,
Research Triangle Park, North Carolina (450/4-83-011).
32. EPA. 1984. National Air Quality and Emissions Trends Report. 1982. U.S.
Environmental Protection Agency, Office of Air Quality Planning and Standards,
Research Triangle Park, North Carolina (450/4-84-002).
33. EPA. 1985. National Air Quality and Emissions Trends Report. 1983. U.S.
Environmental Protection Agency, Office of Air Quality Planning and Standards,
Research Triangle Park, North Carolina (450/4-84-029).
34. EPA. 1986. National Air Quality and Emissions Trends Report. 1984. U.S.
Environmental Protection Agency, Office of Air Quality Planning and Standards,
Research Triangle Park, North Carolina (450/4-86-001).
35. EPA. 1987. National Air Quality and Emissions Trends Report. 1985. U.S.
Environmental Protection Agency, Office of Air Quality Planning and Standards,
Research Triangle Park, North Carolina (450/4-87-001).
36. EPA. 1988. National Air Quality and Emissions Trends Report. 1986. U.S.
Environmental Protection Agency, Office of Air Quality Planning and Standards,
Research Triangle Park, North Carolina (450/4-88-001).
9-16
-------
9-18
-------
10 ALTERNATIVE AIR QUALITY INDICATORS
Design values are air quality indicators that may be viewed as a means for
characterizing the ozone air quality of a metropolitan area in a single number. The design
value can serve as a surrogate measure of air quality status, a measure of progress, and an
indicator of how much concentrations must be reduced to meet the standard. However, the
design value for an area is, by definition, the highest design value among all sites in the
monitoring network. This raises the question of how well the spatial and temporal
distribution of ozone concentrations in the area are represented by the design value.
There are two air quality indicators that are directly related to the ozone National
Ambient Air Quality Standard (NAAQS): the expected exceedance rate and the air quality
design value. Throughout this section the expected exceedance rate is defined as the
expected number of exceedances per year estimated using the Appendix H procedures, and
the air quality design value is the EPA design value (table look-up procedure). Attainment
of the ozone NAAQS is achieved when the expected exceedance rate is less than or equal to
1 at every site within the area. The ozone air quality design value is an estimate of the
concentration expected to be exceeded once per year on average. The ambient air quality
database was used to examine the relationship among these two NAAQS related statistics and
other air quality indicators. The comparisons focus primarily on the 3-year period 1987-89,
which was the database used to establish the majority of the initial nonattainment area
classifications. However, the comparisons of temporal variability of these indicators include
data from the entire 11-year period, 1980-90.
OZONE NAAQS RELATED INDICATORS
The two ozone NAAQS air quality indicators included in these comparisons represent
the highest value recorded among all monitoring sites within the area. For the 98 areas
initially designated nonattainment for ozone under the Clean Air Act Amendments of 1990
(CAAA of 1990), each statistic was estimated as follows:
Avg Est. Exc: Estimated exceedances from each of the three years, 1987,
1988, and 1989, are simply averaged to obtain the Appendix H
expected exceedance rate for 1987-89.
EPA D.V. The EPA table look-up procedure was used to estimate the air
quality design value for the years 1987-89. With complete
monitoring data (i.e., greater than 75 percent of the days during
the ozone season), the EPA design value is simply the fourth
highest daily maximum 1-hour ozone concentration.
As required by the CAAA of 1990, the EPA ozone design value was used to establish
the initial ozone classification of each nonattainment area. Table 10-1 displays these area
classifications and the corresponding number of expected exceedances per year at the design
10-1
-------
value site in each area. As expected, the general pattern is for decreasing numbers of
exceedances as one goes from a higher category to the next lower category. However, there
is some overlap between adjoining categories in exceedance rates at the design value
monitors. Table 10-2 shows similar statistics for all sites within these nonattainment areas,
not just the design value site. Again, the pattern is for lower exceedance rates as one goes
from a higher classification to a lower classification category. The wider range in
exceedance rates for all sites is likely due to the fact that both downwind and upwind sites
are included.
Table 10-1. Number of expected exceedances per year (defined by Appendix H) at design value
sites within ozone area classification categories for areas initially designated
nonattainment under the Clean Air Act Amendments of 1990.
Clean Air Act
Classification
Extreme
Severe 17
Severe 15
Serious
Moderate
Marginal
Number
of Areas
1
5
4
12
33
43
Number of expected
site
Maximum
137.5
59.6
38.8
44.2
9.9
3.8
exceedances per
Minimum
137.5
12.2
8.8
3.7
1.5
1.1
year at design value
Average
137.5
23.0
17.7
10.6
4.5
2.1
Table 10-2. Average Number of Expected Exceedances at All Sites in Designated Nonattainment
Areas, 1987-89
Clean Air Act Number Average Number of Expected Exceedances
Classification of Sites
Extreme
Severe 17
Severe 15
Serious
Moderate
Marginal
40
79
43
105
187
97
Max
137.5
59.6
38.8
44.4
9.9
4.4
Min
0.0
0.0
0.0
0.0
0.0
0.0
95th
%-tile
124.2
16.6
15.6
11.1
6.7
3.8
90th
%-tile
118.1
13.1
12.6
8.3
5.4
3.1
50th
%-tile
46.7
7.8
7.7
3.4
1.4
1.1
25th
%-tile
14.7
2.0
5.4
1.0
0.3
0.3
Avg
54.5
7.9
8.9
4.9
2.1
1.4
10-2
-------
ALTERNATIVE AIR QUALITY INDICATORS
Five additional air quality indicators were also calculated for each of the 98
nonattainment areas. In each case, these are site-specific indicators; network average
indicators are described later in the section on spatial variability. The highest indicator value
at any monitoring site hi the nonattainment area was calculated as follows:
Avg. 2nd Daily Max: The average of the second daily maximum 1-hour ozone
concentration in each of the three years, 1987, 1988, and
1989.
Avg. 95th %-tile: The average of the 95th percentile concentration hi each
of the three years.
5% Tail Exp. D.V.: The ozone design value estimated by fitting a tail
exponential distribution to the upper 5 percent of the
annual ozone concentration distribution. The fitting
method is the Breiman procedure (Breiman, 1978).
Avg. 10 High Days: The average of the ten highest daily maximum 1-hour
ozone concentrations in each of the three years.
Avg Met Adj 99th %-tile The average of the 99th percentile ozone concentrations
from each year that were adjusted for year-to-year
changes in meteorology using the Cox and Chu model
(Cox and Chu, 1991). Values are only available for 35
of the 98 areas.
The air quality values for each of the seven air quality indicators are listed below in
Table 10-3 hi decreasing order of the EPA design value. Table 10-4 presents the linear
correlation matrix for these seven indicators. As one would expect, the strongest association
is found between the EPA D.V. and the design value estimated using the 5% tail exponential
approach since they both target the same statistic. The EPA D.V. is also highly correlated
with the average second maximum concentration, the average of the ten highest days, and the
meteorologically adjusted 99th percentile concentration. The lowest correlation (0.84) was
found for the relationship between the two NAAQS indicators, the EPA D.V. and the
expected exceedance rate. Figure 10-1 displays a plot of these two NAAQS related
indicators for all monitoring sites that reported data for the period 1987-89.
Table 10-5 lists the relative ranking (from largest [rank=l] to smallest [rank=98]) of
each air quality indicator for the original 98 nonattainment areas. Although there is good
agreement between the rank of the EPA design value and the ranks of the other air quality
indicators, there are several noticeable differences. In Dallas, Seattle, and Monterey, CA,
10-3
-------
MgnVdw
1M7-M
33 •
32 •
J1 -
JO •
at •
X -
24 •
33 •
32
.21
2O
.18
-
#•'
• • • •
10 20
40 GO go 70 «> go 100 110 120 130 14
Average Number of Estimated Exceedances, 1987-89
Figure 10-1. Average number of expected exceedances versus the EPA DV for all sites,
1987-89.
10-4
-------
Table 10-3. Alternative Air Quality Indicator Values for 1987-89
Nonattainment Area
Los Angeles South Coast Air Basin
Southeast Desert Modified AQMD
Houston-Galveston-Brazoria NA
New York-N. New Jersey-Long Is
Baltimore NA Area
Chicago-Gary-Lake County NA Area
Milwaukee-Racine NA Area
Sen Diego NA Area
Philadelphia-Wilmington-Trenton
Muskegon NA Area
San Joaquin Valley NA Area
Sheboygan NA Area
Greater Connecticut NA Area
El Paso NA Area
Ventura Co NA Area
Manitowoc Co NA Area
Springfield (W. Mass) NA Area
Boston-Lawrence-Worcester NA Area
Portsmouth-Dover-Rochester, NH
Washington NA Area
Baton Rouge NA Area
Hunt ington-Ash I and NA Area
Atlanta NA Area
Providence (all of RI) NA Area
Beaumont-Port Arthur NA Area
Dallas-Fort Worth NA Area
Sacramento Metro NA Area
Charlotte-Gastonia NA Area
Knox Co and Lincoln Co NA Area
Cincinnati-HamiIton NA Area
Cleveland-Akron-Lorain NA Area
Portland NA Area
St. Louis NA Area
Parkersburg-Marietta NA Area
Greensboro-Winston-Salem-High Pt
San Francisco-Bay NA Area
Louisville NA Area
Pittsburgh-Beaver Valley NA Area
Kewaunee Co NA Area
Atlantic City NA Area
Detroit-Ann Arbor NA Area
Smyth Co NA Area
Dayton-Springfield NA Area
Grands Rapids NA Area
Jefferson Co NA Area
Salt Lake City-Ogden NA Area
Richmond-Petersburg NA Area
Jersey Co NA Area
Raleigh-Durham NA Area
Reading NA Area
Kent County and Queen Anne's Co
Memphis NA Area
Santa Barbara - Santa Maria -
Seattle - Tacoma NA Area
Toledo NA Area
Charleston NA Area
Miami-Fort Lauderdale-W. Palm B.
Monterey Bay Unified NA Area
Nashville NA Area
Allentown-Bethlehem-Easton NA
Lewiston - Auburn NA Area
Owensboro NA Area
EPA
D.V.
0.330
0.240
0.220
0.201
0.194
0.190
0.190
0.190
0.187
0.181
0.180
0.176
0.172
0.170
0.170
0.167
0.167
0.165
0.165
0.165
0.164
0.164
0.162
0.162
0.160
0.160
0.160
0.158
0.158
0.157
0.157
0.156
0.156
0.152
0.151
0.150
0.149
0.149
0.147
0.145
0.144
0.144
0.143
0.143
0.143
0.143
0.142
0.141
0.141
0.141
0.140
0.140
0.140
0.140
0.140
0.138
0.138
0.138
0.138
0.137
0.137
0.137
AVG 2nd
Dai ly
Max
0.327
0.223
0.203
0.187
0.164
0.163
0.174
0.183
0.168
0.167
0.160
0.159
0.172
0.160
0.167
0.154
0.138
0.165
0.158
0.146
0.154
0.144
0.149
0.154
0.137
0.133
0.153
0.136
0.137
0.139
0.143
0.148
0.147
0.142
0.151
0.140
0.134
0.142
0.133
0.135
0.132
0.129
0.136
0.138
0.129
0.132
0.133
0.128
0.129
0.126
0.117
0.128
0.127
0.101
0.128
0.124
0.130
0.116
0.139
0.129
0.121
0.125
AVG
95th
%-TILE
0.240
0.170
0.120
0.140
0.128
0.128
0.129
0.140
0.135
0.123
0.140
0.115
0.121
0.100
0.140
0.121
0.110
0.124
0.106
0.120
0.100
0.111
0.110
0.115
0.100
0.110
0.120
0.112
0.117
0.114
0.113
0.119
0.113
0.112
0.115
0.100
0.105
0.116
0.111
0.114
0.109
0.104
0.109
0.112
0.105
0.109
0.110
0.105
0.112
0.105
0.102
0.102
0.080
0.083
0.100
0.105
0.078
0.092
0.112
0.111
0.097
0.108
5% TAIL
Expan.
D.V.
0.343
0.244
0.231
0.212
0.192
0.199
0.199
0.222
0.186
0.184
0.173
0.182
0.179
0.175
0.176
0.170
0.166
0.176
0.166
0.164
0.160
0.163
0.170
0.167
0.157
0.143
0.172
0.150
0.174
0.159
0.158
0.166
0.163
0.152
0.158
0.148
0.148
0.152
0.153
0.146
0.153
0.146
0.143
0.143
0.144
0.147
0.146
0.139
0.145
0.142
0.140
0.138
0.136
0.111
0.138
0.143
0.136
0.116
0.141
0.146
0.135
0.139
AVG 10 HI
Dai ly
Max 1-hr
0.291
0.205
0.166
0.163
0.150
0.153
0.153
0.163
0.146
0.140
0.154
0.141
0.145
0.135
0.159
0.136
0.127
0.141
0.128
0.134
0.128
0.129
0.134
0.130
0.122
0.122
0.146
0.122
0.121
0.124
0.126
0.127
0.132
0.123
0.133
0.120
0.118
0.125
0.121
0.126
0.122
0.115
0.120
0.123
0.120
0.118
0.121
0.114
0.120
0.117
0.107
0.112
0.107
0.090
0.112
0.113
0.106
0.100
0.123
0.118
0.108
0.115
AVG MET ADJ AVG
Adj. 99th
%-tile
0.309
t
0.223
0.211
0.170
0.154
0.172
0.184
0.191
0.144
0.171
_
o!l45
0.134
_
_
o!l49
_
o!l64
0.153
0.140
0.149
0.147
0.145
0.149
0.165
0.134
_
Oll44
0.145
_
_
.
m
0.150
0.145
.
f
0.148
.
.
.
.
_
0.162
0.117
.
,
_
O!l19
.
a
0.118
_
.
.
.
.
Est.
Exceed 's
137.5
59.6
14.6
17.4
10.7
13.0
13.0
32.9
15.6
9.5
44.4
9.1
9.0
7.9
38.8
9.9
6.7
10.0
5.3
8.1
4.8
5.5
9.3
6.4
3.9
3.5
15.8
4.2
7.4
5.4
5.5
8.7
6.2
6.7
7.2
3.5
3.2
7.0
5.5
4.0
4.4
2.4
3.1
4.4
3.4
2.0
4.4
3.7
4.2
3.4
2.2
2.0
1.1
1.4
2.7
2.3
1.7
8.2
5.6
4.4
1.5
3.7
Cont i nued
10-5
-------
Table 10-3. Alternative Air Quality Indicator Values for 1987-89—Concluded
Nonattainment Area
Harrisburg-Lebanon-Carlisle NA
Canton NA Area
Hancock Co and Waldo Co NA Area
Knoxville NA Area
Poughkeepsie NA Area
Youngstown-Warren-Sharon NA Area
Birmingham NA Area
Johnstown NA Area
Cherokee Co NA Area
Buffalo-Niagara Falls NA Area
Columbus NA Area
Lake Charles NA Area
Edmonson Co NA Area
Norfolk-Virginia Beach-Newport
Sussex Co NA Area
Altoona NA Area
Erie NA Area
Scranton-WiIkes-Barre NA Area
Tampa-St. Petersburg-Clearwater
UaI worth Co NA Area
York NA Area
Albany-Schenectady-Troy NA Area
Manchester NA Area
Essex Co NA Area
Lexington-Fayette NA Area
Greenbrier NA Area
Lancaster NA Area
Portland-Vancouver AQMA NA Area
Evansvilie NA Area
Paducah NA Area
Indianapolis NA Area
Phoenix
South Bend-Mishawaka NA Area
Kansas City NA Area
Reno
Door Co NA Area
EPA
D.V.
0.136
0.135
0.135
0.135
0.134
0.134
0.133
0.133
0.132
0.131
0.131
0.131
0.130
0.130
0.130
0.129
0.129
0.129
0.129
0.129
0.129
0.128
0.128
0.127
0.126
0.125
0.125
0.125
0.124
0.124
0.121
0.121
0.121
0.120
0.111
0.108
AVG 2nd
Dai ly
Max
0.127
0.130
0.137
0.117
0.108
0.123
0.126
0.120
0.118
0.128
0.124
0.128
0.114
0.118
0.127
0.120
0.132
0.120
0.124
0.115
0.121
0.112
0.112
0.112
0.114
0.119
0.117
0.110
0.119
0.113
0.120
0.110
0.116
0.124
0.096
0.108
AVG
95th
%-TILE
0.106
0.106
0.102
0.100
0.083
0.105
0.105
0.102
0.098
0.103
0.097
0.094
0.094
0.099
0.100
0.099
0.110
0.100
0.094
0.096
0.105
0.097
0.090
0.096
0.100
0.103
0.106
0.096
0.103
0.096
0.102
0.093
0.096
0.095
0.086
0.096
5% TAIL
Expan.
D.V.
0.138
0.139
0.157
0.133
0.130
0.136
0.141
0.139
0.137
0.139
0.132
0.140
0.130
0.129
0.143
0.132
0.143
0.135
0.134
0.128
0.134
0.134
0.127
0.129
0.128
0.131
0.128
0.131
0.130
0.127
0.126
0.121
0.128
0.130
0.110
0.117
AVG 10 HI
Daily
Max 1-hr
0.117
0.115
0.118
0.105
0.094
0.114
0.119
0.110
0.110
0.115
0.107
0.117
0.106
0.108
0.113
0.106
0.120
0.110
0.109
0.104
0.111
0.106
0.101
0.105
0.107
0.108
0.111
0.099
0.111
0.105
0.109
0.105
0.107
0.106
0.088
0.102
AVG MET ADJ AVG
Adj. 99th Est.
%-tile Exceed's
2.3
2.0
3.8
1.8
1.3
2.1
0.128 3.0
2.5
2.4
3.8
1.7
2.0
2.1
2.0
3.6
2.0
4.0
3.0
0.120
0.122
o!l20
1.4
3.8
1.4
1.8
2.0
1.4
1.3
2.3
0.0
0.0
10-6
-------
Table 10-4. Correlation Matrix for Alternative Indicators, 1987-89
Pearson
DVMAX89
EPA DV( 1987-89)
SECMAX89
AVG 2NDMXC1987-89)
PCTHAX89
AVG 95XTILE(1987-89)
T5XHAX89
5% TAIL EX( 1987-89)
HI10MX89
AVG 10 HI DAYS
(1978-89)
MET8789
AVG MET ADJ 99%TILE
(1987-89)
EXCMAX89
AVG EST EXC( 1987-89)
Correlation
DVMAX89
1.00000
0.0
98
0.96643
0.0001
98
0.89095
0.0001
98
0.97224
0.0001
98
0.95618
0.0001
98
0.93137
0.0001
35
0.84195
0.0001
98
Coefficients
SECHAX89
0.96643
0.0001
98
1.00000
0.0
98
0.91855
0.0001
98
0.97819
0.0001
98
0.98329
0.0001
98
0.92046
0.0001
35
0.86541
0.0001
98
/ Prob > ]R]
PCTMAX89
0.89095
0.0001
98
0.91855
0.0001
98
1.00000
0.0
98
0.90923
0.0001
98
0.95859
0.0001
98
0.89142
0.0001
35
0.90394
0.0001
98
under Ho:
T5XHAX89
0.97224
0.0001
98
0.97819
0.0001
98
0.90923
0.0001
98
1.00000
0.0
98
0.97087
0.0001
98
0.92273
0.0001
35
0.83534
0.0001
98
Rho=0 / Number
HI10MX89
0.95618
0.0001
98
0.98329
0.0001
98
0.95859
0.0001
98
0.97087
0.0001
98
1.00000
0.0
98
0.92449
0.0001
35
0.90477
0.0001
98
of Observations
MET8789
0.93137
0.0001
35
0.92046
0.0001
35
0.89142
0.0001
35
0.92273
0.0001
35
0.92449
0.0001
35
1.00000
0.0
35
0.83855
0.0001
35
EXCMAX89
0.84195
0.0001
98
0.86541
0.0001
98
0.90394
0.0001
98
0.83534
0.0001
98
0.90477
0.0001
98
0.83855
0.0001
35
1.00000
0.0
98
10-7
-------
Table 10-5. Relative Ranks of Alternative Air Quality Indicators
Nonattainment Area
Los Angeles South Coast Air Basin
Southeast Desert Modified AQHD
Houston-Galveston-Brazoria NA
New York-N. New Jersey-Long Is.
Baltimore NA Area
Chicago-Gary-Lake County NA Area
Milwaukee-Racine NA Area
San Diego NA Area
Philadelphia-Wilmington-Trenton
Muskegon NA Area
San Joaquin Valley NA Area
Sheboygan NA Area
Greater Connecticut NA Area
El Paso NA Area
Ventura Co NA Area
Manitowoc Co NA Area
Springfield (W. Mass) NA Area
Boston-Lawrence-Worcester NA
Portsmouth-Dover-Rochester, NH
Washington NA Area
Baton Rouge NA Area
Huntington-Ashland NA Area
Atlanta NA Area
Providence (all of RI) NA Area
Beaumont-Port Arthur NA Area
Dal las-Fort Worth NA Area
Sacramento Metro NA Area
Charlotte-Gastonia NA Area
Knox Co and Lincoln Co NA Area
Cincinnati-Hamilton NA Area
Cleveland-Akron-Lorain NA Area
Portland NA Area
St. Louis NA Area
Parkersburg-Marietta NA Area
Greensboro-Winston-Salem-High Pt
San Francisco-Bay NA Area
Louisville NA Area
Pittsburgh-Beaver Valley NA Area
Kewaunee Co NA Area
Atlantic City NA Area
Detroit-Ann Arbor NA Area
Smyth Co NA Area
Dayton-Springfield NA Area
Grands Rapids NA Area
Jefferson Co NA Area
Salt Lake City-Ogden NA Area
Richmond-Petersburg NA Area
Jersey Co NA Area
Raleigh-Durham NA Area
Reading NA Area
Kent County and Queen Anne's Co
Memphis NA Area
Santa Barbara - Santa Maria -
Seattle - Tacoma NA Area
Toledo NA Area
Charleston NA Area
Miami-Fort Lauderdale-W. Palm B.
Monterey Bay Unified NA Area
Nashville NA Area
Allentown-Bethlehem-Easton NA
Lewiston - Auburn NA Area
Owensboro NA Area
Harrisburg-Lebanon-Carlisle NA
Canton NA Area
RANK FOR
EPA
D.V.
sin 1
2
3
. 4
5
ea 7
7
7
n 9
10
11
12
13
14.5
14.5
16.5
16.5
19
19
19
21.5
21.5
23.5
23.5
26
26
26
28.5
28.5
30.5
30.5
32.5
32.5
34
Pt 35
36
37.5
ea 37.5
39
40
41.5
41.5
44.5
44.5
44.5
44.5
47
49
49
49
;o 53
53
53
53
53
57.5
B. 57.5
57.5
57.5
61
61
61
I 63
65
RANK FOR
AVG 2nd
Daily Max
1
2
3
4
12
13
6
5
8
9
14.5
16
7
14.5
10
19
34
11
17
26
18
27
23
20
38
43
21
39
37
32.5
28
24
25
29
22
31
42
30
44.5
41
46
52
40
35
54
47
44.5
59
52
64
81.5
57
60
97
57
68.5
49.5
6%
32.5
52
72
65
61.5
49.5
RANK FOR
AVG 95th
%-TILE
1
2
16
4.5
9.5
9.5
8
4.5
7
12
4.5
22
13.5
70
4.5
13.5
38
11
46.5
16
70
34
38
22
70
38
16
28
19
24.5
26.5
18
26.5
30.5
22
70
52.5
20
34
24.5
42
57
42
30.5
52.5
42
38
52.5
30.5
52.5
63
63
97
95.5
70
52.5
98
92
30.5
34
79
44
46.5
46.5
RANK FOR
5% TAIL
Exp. D.V
1
2
3
5
8
6.5
6.5
4
9
10
17
11
12
15
13.5
19.5
23
13.5
23
25
28
26.5
19.5
21
32.5
50.5
18
38
16
29
30.5
23
26.5
36.5
30.5
39.5
39.5
36.5
34.5
43.5
34.5
43.5
50.5
50.5
47
41
43.5
61
46
54
57.5
65
69
97
65
50.5
69
96
55.5
43.5
71.5
61
65
61
RANK FOR RANK FOR Avg RANK FOR
AVG 10 HI MET ADJ 99th AVG Est.
Daily Max %-tile Exceed's
1
2
3
5
10
8.5
8.5
4
11
16
7
15
13
18
6
17
27
14
25
20
26
24
19
23
38.5
38.5
12
36
42
32
30
28
22
33
21
43
50
31
40.5
29
37
58.5
46.5
35
44
49
40.5
60
46.5
53
79
64.5
81
97
64.5
62
83
94
34
52
76
56
54
58.5
1
2
3
8
12
6
5
4
24
7
22!
28
16.5
10"
13
26
16.5
19
20.5
15
9
27
25'
20.5
14
22.5
18
11
35
33
34
1
2
9
6
12
10.5
10.5
5
8
15
3
17
18
22
4
14
26.5
13
35
21
36
31
16
28
45
52
7
41.5
23
34
32.5
19
29
26.5
24
53
56
25
32.5
43
39.5
62.5
57
39.5
55
72
37.5
50
41.5
54
67
72
91
83
60
64
79
20
30
37.5
82
49
65.5
75.5
Continued
10-8
-------
Table 10-5. Relative Ranks of Alternative Air Quality Indicators—Concluded
Nonpttainment Area
Hancock Co and Waldo Co NA Area
KnoxviUe NA Area
Roughkeepsie NA Area
Youngstown-Warren-Sharon NA Area 67.5
Birmingham NA Area
Johnstown NA Area
Cherokee Co NA Area
Buffalo-Niagara Falls NA Area
Columbus NA Area
Lake Charles NA Area
Edtnonson Co NA Area
Norfolk-Virginia Beach-Newport
Sussex Co NA Area
Altoona NA Area
Erie NA Area
Scranton-Uilkes-Barre NA Area
Tampa-St. Petersburg-Clearwater
Wai worth Co NA Area
York NA Area
Albany-Schenectady-Troy NA Area
Manchester NA Area
Essex Co NA Area
Lexington-Fayette NA Area
Greenbrier NA Area
Lancaster NA Area
Portland-Vancouver AQMA NA Area
Evansvilie NA Area
Paducah NA Area
Indianapolis NA Area
Phoenix
South Bend-Mishawaka NA Area
Kansas City NA Area
Reno
Door Co NA Area
RANK FOR
EPA
D.V.
i 65
65
67.5
>.a 67.5
69.5
69.5
71
73
73
73
76
76
76
80.5
80.5
80.5
• 80.5
80.5
80.5
i 84.5
84.5
86
87
89
89
i 89
91.5
91.5
94
94
94
96
97
98
RANK FOR
AVG 2nd
Dai ly Max
36
83
96
70
63
75
79
57
66.5
55
88
80
61.5
73.5
48
73.5
66.5
86
71
90
91
92
87
78
81.5
93
77
89
76
94
85
68.5
98
95
RANK FOR
AVG 95th
%-TILE
63
70
95.5
52.5
52.5
63
77
59
79
89
89
75.5
70
75.5
38
70
89
83.5
52.5
79
93
83.5
70
59
46.5
83.5
59
83.5
63
91
83.5
87
94
83.5
RANK FOR
5% TAIL
Exp. D.V.
32.5
76
82.5
69
55.5
61
67
61
77.5
57.5
82.5
85.5
50.5
77.5
50.5
71.5
74
88.5
74
74
91.5
85.5
88.5
79.5
88.5
79.5
82.5
91.5
93
94
88.5
82.5
98
95
RANK FOR
AVG 10 HI
Dai ly Max
51
87
96
61
48
70
69
57
77
55
85
74
63
84
45
71
72
91
67
86
93
89
78
75
68
95
66
88
73
90
80
82
98
92
RANK FOR Avg RANK FOR
MET ADJ 99th AVG Est.
%-tile Exceed's
48
77.5
88
68
29 58.5
61
62.5
47
80
72
69
75.5
51
72
44
58.5
32 85
88
81
46
85
77.5
72
85
88
65.5
95
93
92
30 96
94
31 90
97.5
97.5
10-9
-------
contrast, the tail exponential design value ranks for Hancock County, Sussex County, and
Erie, PA were relatively higher than the EPA D.V. ranks. Other large differences were
found between the relative area rankings based on the EPA design value and rankings
obtained for the 95th percentile concentration in some areas, including Houston, El Paso,
Beaumont, Portsmouth and Sheboygan.
TEMPORAL VARIABILITY IN OZONE INDICATORS
The National Academy of Sciences (NRC, 1991) recommended that more statistically
robust indicators be developed to track progress in reducing ozone. The report described the
use of different percentiles as compared to maximum values. Figure 10-2 presents the trend
in the peak concentrations and selected percentiles for all sites reporting data during the
period 1980-90. The percentiles presented in the figure range from the 50th percentile (the
median) to the 95th percentile. The year-to-year variation is similar among all the indicators,
with a tendency to become flatter (less variable) in the lower percentiles. Figures 10-3
through 10-7 display the temporal variation in five air quality indicators for ten selected
nonattainment areas. The year-to-year variability is similar, especially for the concentration
based indicators. Less agreement hi year-to-year variation is evident between the
concentration based indicators and the expected exceedance rate.
(All sites reporting data)
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
Concentration, PPM
80
81
82
83
84
85
Year
86
87
88
89
90
Max 1-hr 2nd Max 1-hr95th %-tile90th %-tile50th %-tile Mean
Figure 10-2. Year-to-year variability in selected ozone air quality indicators.
10-10
-------
Los Angeles
Concentration, ppm
0.5
0.4
0.3
0.2
0.1
Estimated Exceedances
300
J_
_L
_L
.L
_L
_L
JL
JL
_L
250
200
150
100
50
1982 1983 1984
1985 1986 1987
End Year
1988 1989 1990
EPA D.V. 5% TAILX D.V. AVG SEC MAX AVG 95th PCTILE EST EXCEED
Houston
Concentration, ppm
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
_L
Estimated Exceedances
50
JL
J_
t
_L
JL
JL
40
30
20
10
1982 1983 1984 1985 1986 1987
End Year
1988 1989 1990
Figure 10-3. Temporal variability in ozone indicators in Los Angeles and Houston.
10-11
-------
Chicago
Concentration, ppm
0.25
0.2
0.15
0.1
0.05
Estimated Exceedances
25
j_
J_
J_
20
15
10
1982 1983 1984
1985 1986 1987
End Year
1988 1989 1990
EPA D.V. 5% TAILX D.V. AVG SEC MAX AVG 95th PCTILE EST EXCEED
Milwaukee
Concentration, ppm
0.25
0.2
0.15
0.1
0.05
Estimated Exceedances
25
20
15
10
1982 1983 1984
1985 1986 1987
End Year
1988 1989 1990
Figure 10-4. Temporal variability in ozone indicators in Chicago and Milwaukee.
10-12
-------
New York City
Concentration, ppm
Estimated Exceedances
0.3
0.25
0.2
0.15
0.1
0.05
0
j_
1982 1983 1984
1985 1986 1987
End Year
1988 1989 1990
70
60
50
40
30
20
10
0
EPA D.V. 5% TAILX D.V. AVG SEC MAX AVG 95th PCTILE EST EXCEED
Baltimore
Concentration, ppm
0.2
0.15
0.1
0.05
Estimated Exceendances
50
40
30
20
10
1982 1983 1984
1985 1986 1987
End Year
1988 1989 1990
Figure 10-5. Temporal variability in ozone indicators in New York and Baltimore.
10-13
-------
Philadelphia
Concentration, ppm
Estimated Exceedances
0.2
0.15
0.1
0.05
-
.
_
•
-
V-v
fc
.;•:;• ..
--. :>'!
--.. .. .;
.-,- .- •
:; ; ;'
19
*••*
**''•'•
\ '.-'.<
;:•:-:-
'!'
lil:
•\xV
ll-:-
tv
82
E
-&!
?;::-V:::::x::Y:.- '.^ ^
T-: ::.>,C" :(:'':::x
.'-i >.;:--V-;;-'. K-.-.-: •
1983 198^
PAD.V. 5%TAILX
if
;l
•:•:
• •:.
\
D.V.
~^^y--^^^::-,:<>r^r:::::^
lur:^:!
i--^.y • M::. , ? .
t i : . 1 1 <
1985 1986 1987 1988 1989 1
End Year
AVG SEC MAX AVG 95th PCTILE EST EXCEED
.
•M* ••
'
•
-
1
990
ou
20
15
10
5
Atlanta
Concentration, ppm
0.2
0.15
0.1
0.05
Estimated Exceedances
16
!£.
1982 1983 1984 1985 1986 1987 1988 1989 1990
End Year
14
12
10
8
6
4
2
0
Figure 10-6. Temporal variability in ozone indicators in Philadelphia and Atlanta.
10-14
-------
Boston
Concentration, ppm
0.2
0.15
0.1
0.05
Estimated Exceedances
16
t
14
12
10
8
6
4
2
1982 1983 1984 1985 1986 1987 1988 1989 1990
End Year
EPA D.V. 5% TAILX D.V, AVG SEC MAX AVG 95th PCTILE EST EXCEED
Greater Connecticut (Hartford)
Concentration, ppm
0.3
0.25
0.2
0.15
0.1
0.05
Estimated Exceedances
40
s,
30
20
10
1982 1983
1984 1985 1986 1987
End Year
1988 1989 1990
Figure 10-7. Temporal variability in ozone indicators in Boston and Hartford.
10-15
-------
SPATIAL VARIATIONS IN OZONE CONCENTRATIONS
Analysis of Monitoring Data
Ozone monitoring networks in most metropolitan areas are quite sparse. A notable
exception is the Los Angeles Basin where data are routinely collected over a large network
(35 sites in 1989), with 17 sites reporting data to AIRS (as of 1991). Many of these sites
have been reporting continuously since 1980, making the South Coast Air Basin (SOCAB)
one of the most comprehensively instrumented areas in the world. This extensive database
can be used to characterize the spatial variations in ozone concentrations to an extent which is
not possible in most other metropolitan areas.
Hourly ozone concentration data at stations in the South Coast (Los Angeles) Air
Basin were obtained from AIRS and local sources for the period 1980-91 for the high ozone
season (April-October). ' Daily maximum concentrations were computed and flagged as
valid for days with at least nine valid hourly readings between 0900 and 2100 LST. Only
stations with at least 75 percent valid daily maximums were retained. Various ozone
indicators ("yearly" or "annual" indicators) were calculated for each monitor using the data for
each year. Similar indicators were calculated using data combined into overlapping three year
periods ("3-year" indicators). These indicators are defined as follows:
EPA DV: The second highest (fourth highest for 3-year) daily maximum
concentration.
90%tile: The 90th percentile of the daily maximum concentrations.
95%tile: The 95th percentile of the daily maximum concentrations.
An additional set of ("network") indicators was calculated by combining data for all sites. In
this case, EPA DV is the maximum second highest concentration over the monitoring network
(fourth highest for 3 -year indicators). In addition, two exposure indicators were calculated for
the network:
I: The number of pphm-days per station-day above 12 pphm, defined by
where D^ is the daily maximum concentration for they'th monitor (j=l,... , Nj) for the /th day
(i=i,... , Jv)) and (D,. - 12)+ equals DtJ -\2\fDij>\2 and equals 0 otherwise.
^he Federally designated ozone season for the Los Angeles basin is January-December, but peak
ozone concentrations are generally confined to April-October period.
10-16
-------
/ (Pop Wt): A population weighted version of / defined by
where Pj is the population associated with the /th monitor.
Figure 10-8 displays the number of days on which concentrations exceeded the NAAQS in
1989. The relatively low concentrations along the coast and high values inland are
characteristic of the SOCAB. Spatial variations are also evident in the design values
calculated for each station (Figure 10-9). Design values at the coastal and metropolitan
stations (North Long Beach and Lynwood) are significantly lower than at stations further
inland. In addition, the station with the maximum design value varies from one year to the
next and these stations can be located at considerable distances from one another (Pico Rivera
and Pomona are 30 km apart, Pomona and Bur bank are 50 km apart).
Given the large spatial variations in concentration distributions, care must be taken
when interpreting data from sparse monitoring networks. For example, in regions with only a
few monitoring stations, a single station located in an area of high concentrations may
consistently be the design value station for the region. However, this does not mean that the
location of the highest concentrations does not vary from year to the next; it may simply
mean that the alternate locations are not near any monitors.
Spatial Indicators of Air Quality
One way of examining the adequacy of the design value as an air quality indicator is
to compare the design value with other ozone air quality indicators, including indicators which
are more broadly based in both temporal and spatial dimensions. Figure 10-10 compares the
design value for the SOCAB (calculated using data from the same stations shown in Figure
10-9) with three other network indicators. This comparison indicates a high degree of
correlation between the design value and the broader indices, at least for the period
1980-1988. The design value departs somewhat from the other indicators during the last
three years of the period. Linear correlations between the indicators are shown in Table
10-6.
Table 10-6. Correlation Matrix for Spatial Indicators in Los Angeles.
Design Value 90th Percentile 95th Percentile /
90th Percentile
95th Percentile
7
/ (Pop Wt.)
.72
.822
.842
.825
.975
.939
.926
.967
.962
.992
10-17
-------
IIIIIMMIIMMMI
SCAQMO
Air Monilonng
Network
I , I , I , I
-•- liMomplele. ) V
N« measured. \_ _ ^
Figure 10-8. Number of days on which the federal ozone standard was exceeded in 1989.
10-18
-------
80 81 82 83 84 85 86 87 88 89
North Long Beach
La Habra
Lynwood
Pomona
Pico Rivera
Burbank
Figure 10-9. Design values (fourth highest value in each three year period) for selected
stations.
10-19
-------
80 81 82 83 84 85 86 87 88 89 90 91
-*- EPA D.V. -H— 90%tile -»
-*- | -B- | (Pop Wt)
«- 95%tile
Figure 10-10.
Comparison of annual ozone indicators.
10-20
-------
These results indicate an overall high degree of association between the design value
and the broader air quality indicators for this period in the SOCAB. Indicators based on
overlapping three-year periods are significantly less variable than those based on individual
years. This point is illustrated in Figure 10-11. As hi the yearly results, generally similar
trends are evident in each indicator.
Although the yearly design value is highly correlated with the broader air quality
indicators, some of the indicators exhibit greater year-to-year variability than others, as
indicated hi Figure 10-12, which is identical to Figure 10-10 except that the indicator value
for each year is plotted as a fraction of its value in 1980. This places the indices on an equal
footing and shows that the exposure indicator, /, is both the most volatile and exhibits the
largest percent improvement since 1980. The design value is slightly more volatile than the
90th and 95th percentiles. The greater volatility and greater overall downward trend in the
exposure indicator is characteristic of indicators based on the number of NAAQS
exceedances.
These results show that, although all of the indicators follow similar trends in the
SOCAB, trends for some indicators may differ from the design value trend hi other ah*
basins. For example, hi the San Francisco Bay Area, the design value shows no change
between 1986 and 1988 while the population-weighted exceedances decreased 50 percent and
the 90th percentile and risk-weighted exposures increased. In other words, the number of
NAAQS exceedances in heavily populated areas decreased but the occurrences of
concentrations slightly below the NAAQS increased. Similar differences hi trends between
indicators are evident in the other air basins. This suggests that no single indicator can be
relied on to fully characterize trends in all of the different aspects of ozone concentrations,
which may be of interest to air quality managers.
SUMMARY
Comparisons between ozone design values estimated by the EPA table look-up method
and other ozone ah" quality indicators revealed a high degree of correlation among these
alternative indicators. Although the year-to-year variability was similar between the EPA
design value and more robust indicators based on lower percentile concentrations, the
magnitude of the change hi the percentiles was less. The exposure indicators showed a
greater degree of year-to-year fluctuations than the concentration-only indicators. Give the
spatial variability hi ozone observed across urban areas, one cannot expect a single numerical
value, such as the design value, to adequately describe concentration gradients.
10-21
-------
10
3-Year Ozone Indicators
SOCAB Sites
82 83 84 85 86 87 88 89 90 91
EPA D.V.
90%tile
1 (Pop Wt)
95%tile
Figure 10-11.
Comparison of indicators for overlapping three-year periods.
10-22
-------
Yearly Ozone Indicators (Normalized)
SOCAB Sites
82 83 84 85 86 87 88 89 90
80 81
EPA DM.
I
90%ti!e
I (Pop Wt)
95%tile
Figure 10-12.
Comparison of indicators for overlapping three-year periods normalized
to 1980 values.
10-23
-------
REFERENCES
Breiman, L., J. Gins, and C. Stone. 1978. "Statistical Analysis and Interpretation of Peak
Air Pollution Measurements," Work performed under EPA Contract 68-02-2857,
Technology Service Corporation, Santa Monica, CA.
Cox, W.M. and S.H. Chu. 1992. "Meteorologically Adjusted Ozone Trends in Urban Areas:
A Probabilistic Approach." Presented at the Tropospheric Ozone and The
Environment n Air and Waste Management Association Specialty Conference.
Atlanta, GA.
NRC. 1991. Rethinking the Ozone Problem in Urban and Regional Air Pollution.
National Research Council, National Academy Press, Washington, D.C.
10-24
-------
11 PUBLIC OUTREACH EFFORTS
Section 183(g) of the CAAA of 1990 requires the EPA to "obtain input from States,
local subdivisions thereof, and others." In conducting the Ozone Design Value Study, EPA
has made every effort to have an open process and to ensure full public input and
participation. These efforts have focused on information exchange through participation in
professional meetings and conferences, involvement of interested parties on the study
working group, holding a public meeting, and making a draft copy of this report available
for public comment. EPA also arranged to have experts at the National Institute of Statistical
Sciences conduct a peer review of the draft report.
A study review group has been established to provide input on technical issues and
policy concerns. The group is composed of representatives from EPA program, research,
policy and legal offices. State and local air pollution control agency officials also serve on
the review group. EPA has also sought input from State and local control agency officials
through involvement with the Standing Ambient Monitoring Work Group (SAMWG).
Early in the study, EPA staff met with interested parties and participated in technical
conferences and meetings. On March 22, 1991, EPA staff met with representatives of the
American Petroleum Institute (API), and their member companies, to hear their concerns
about the current ozone design value methodology. API representatives expressed their
desire to have EPA (1) adopt a more statistically robust form of the ozone standard (i.e., one
that is subject to less variability) and (2) use additional indicators, such as measures of
exposure, to evaluate control strategy progress. EPA staff have chaired technical sessions,
participated in panel discussions, and presented technical papers at the Air and Waste
Management Association's Annual Meetings, specialty conferences, and workshops on design
value issues (Curran, 1992a,b; Curran and Freas, 1991; Freas, 1991, 1992a,b).
OVERVIEW OF PUBLIC MEETING COMMENTS
On September 10, 1992, EPA held a public meeting in Arlington, VA to obtain input
on technical considerations, and implementation and policy issues to be addressed within the
context of the Ozone Design Value Study. The meeting announcement was published in the
Federal Register. "Notice of Public Meeting" (57FR34133), August 3, 1992. To ensure that
all interested parties were aware of the public meeting, copies of the meeting announcement
were sent to both individuals and organizations that had previously expressed interest in
ozone related issues. At the public meeting, presentations were made on behalf of the Motor
Vehicle Manufacturer's Association by Mr. Jon Heuss of General Motors Corporation, and
on behalf of Ford Motor Company by Dr. David Chock of Ford. Written comments were
received from ten respondents, representing State and local air pollution agencies, industry,
and private individual views. The comments received have been grouped into the seven
general subject areas listed in Table 11-1. Some of these comments relate directly to issues
concerning the methodology for estimating ozone design values, while other comments
express concerns about either the form of the ozone standard or implementation of the
11-1
-------
Table 11-1. Matrix of Public Meeting Written Comments by Subject Area
Increase Adjust for Compliance Measurement Design of Transport Refine
Robustness Annual Test Interference Monitoring Meteoro-
Meteorology Networks logical
Models
Motor Vehicle
Manufacturer's XXX X
Association
Ford XXX X
American
Petroleum X XX
Institute
US Dept. of the X XXX
Interior
California Air
Resources Board X X
Dept. of
Environmental X X
Protection,
Connecticut
Environmental
Protection XX X X
Agency, Illinois
New York State
Dept. of X X
Environmental
Conservation
Air Pollution
Control District, X
San Diego
Zephyr X XX
Consulting
11-2
-------
CAAA of 1990. A brief summary of each major subject area follows, and an EPA provided
synopsis of the public comments can be found in Appendix I. A verbatim transcript of the
public meeting, and the written comments and technical papers submitted in response to the
Notice, will be included in the public record of this study.
Increase Robustness of the Design Value
Five commenters expressed concern over the reliance on an extreme value for the
design value. Several commented on the instability associated with using the fourth highest
daily maximum hourly value from the 3-year period and several proposed a summary statistic
based on more information such as the 95th percentile, the mean of the three annual 95th
percentiles, or the mean of the highest 30 values. The rationale for these proposed
modifications is to increase the robustness and the stability in the design value.
EPA comment: The ozone design value is "the value that should be reduced to the standard
level thereby ensuring that the site will meet the standard. With the wording of the ozone
standard the appropriate design value is the concentration with expected number of
exceedances equal to 1." Thus, by definition, the design value is an extreme value. More
robust indicators also exhibit year-to-year fluctuations similar to those observed for the
design value concentrations (Curran and Frank, 1991; Curran and Freas, 1991; NRC, 1991).
It should also be noted that the mean of the highest 30 values is closely related to a tail
exponential design value estimate. Furthermore, the alternative "more robust" statistics may
not give equivalent public health protection at all sites.
Adjust for Annual Meteorological Differences
Seven commenters said that there is a need for a method to account for temporal
variation in meteorology which can mask the underlying ozone trend. Although the 3-year
time period used in the current design value method lessens the impact of a particular year
when severe heat, low precipitation, and light winds contribute to elevated ozone
concentrations, the effect of a year with particularly severe meteorology, such as 1988, is
still strongly felt. Several suggested modifying the current procedure, which excludes the
three highest values in the 3-year period, to either one that can exclude one or more values
each year based on meteorological information or one that selects the second highest from the
set of three annual second highest values. Interest was expressed in a method to minimize
the temporal variation in ozone design values. Methods were suggested to adjust the design
value for the annual variation hi the number of ozone conducive days, using temperature data
and data for other meteorological parameters. In each case, the intent is to decrease the
impact of a single high-ozone year on the design value.
EPA comment: Compliance with the ozone standard is judged on the basis of the actual
ambient air quality measurements. The report by the National Academy of Sciences (NAS)
stated that more "statistically robust methods be developed to assist in tracking progress in
reducing ozone (NRC, 1991)." The NAS did not suggest using meteorologically adjusted air
11-3
-------
quality indicators to judge compliance with the ambient standard. Thus, the meteorological
adjustments are not appropriate for the "air quality" design value but may be appropriate for
a "control strategy" design value. Analyses must communicate trends relevant for the health
based standard. EPA has initiated a program to investigate techniques for adjusting ozone
trends for meteorological influences (Cox and Chu, 1992).
Compliance Test for Attainment
Concern was expressed for the need to have consistency between the design value and
the compliance test for attainment (either to achieve attainment initially or to maintain it).
Some commented that the current design value is not the best target because it doesn't
incorporate the variability in the ozone measurements and requires compliance 100 percent of
the time. Some have suggested modifying this part of the procedure with a two-component
testing approach which would be stringent in first accepting an area as in attainment, but for
areas that have been in attainment for several consecutive 3-year periods and have no
significant increases in ozone-precursor emissions, a less stringent test would be used, which
would reduce the probability of misclassifying an attainment area as nonattainment in the
test. This modification is intended to reduce the oscillation of areas between attainment and
nonattainment due principally to annual fluctuations in meteorological conditions.
EPA comment: The compliance test for the current ozone standard is given in Appendix H
to Part 50.9 of the Code of Federal Regulations (44 FR 8220). Changes in the form of the
current ozone standard, or the attainment test, are beyond the scope of this study. These
issues are more properly addressed within the context of the standard review process.
Measurement Interference and Bias
Two commenters expressed concern about potential measurement interference caused
by conditions of high temperature, humidity, and ozone precursor concentrations. Although
some factors might cause bias in either direction, commenters felt that most of the factors
will result in a positive bias. Two commenters said that both of the major instrument types
used for the ozone monitoring networks, the ethylene chemiluminescence (CL) instrument
and the ultraviolet absorption (UV) instrument, appear to be affected; the UV instrument
interferences seem to be more complex and larger than in the CL instrument.
EPA comment: In accordance with revisions to Appendix A, 40CFR Part 58, site-specific
precision and accuracy data have been submitted to EPA beginning in 1987. EPA has
studied the issue of interferences from humidity and photochemical pollutants, and based on a
1991 study and earlier studies, EPA does not concur with the commenters' conclusion of
widespread interference. EPA is, however, attempting to respond even further to concerns
in this area by supporting a project titled "Investigation of Interferences and Other Anomalies
in the Measurement of Atmospheric Ozone Concentrations." EPA is looking to jointly
conduct this project with the American Petroleum Institute.
11-4
-------
Size and Location of Ozone Monitoring Network
Four comments were received stating that there is a need for an ozone monitoring
network of adequate size and appropriate locations to characterize ozone measurement
variation over small distances, particularly in metropolitan areas. Suggested factors to
consider include local terrain features, usual wind roses, proximity to ozone precursor
emissions, and seasonal meteorological variation.
EPA Comment: EPA has revised the ambient air quality surveillance regulations (58 FR
8452, February 12, 1993) to include provisions for the enhanced monitoring of ozone and its
precursors. Included in these revisions are minimum criteria for network design, monitor
siting, monitoring methods, operating schedules, quality assurance, and data submittal.
Transport from Nearby Areas
Three commenters said there is a need to develop techniques to account for transport
of ozone and/or precursors from nearby areas. In extreme situations, an air quality region
with in-transport may not be able to demonstrate attainment no matter how rigorous the
proposed emission control strategy. A modified approach was suggested, such as
consideration of a larger regional effort to design control options.
EPA Comment: The air quality design value estimate is determined from the actual ambient
monitoring data observed at a monitoring station. In some locations, the transport of ozone
and its precursors can have a significant impact on local air quality levels. Section 181 (a)(4)
gives the EPA Administrator the discretionary authority to modify the area's initial
classification if the area's design value is within 5 percent of another category. The
Administrator may consider other factors such as "the level of pollution transport." The Act
has provisions for treating an ozone nonattainment area as a rural transport area and for
establishing Transport Commissions.
Refine Meteorological Models
Four commenters expressed concern about the adequacy of existing meteorological
models to perform their role in the current design value methodology. A stated limitation of
using photochemical grid models to demonstrate attainment is that the required detailed input
information for emissions, initial conditions, boundary conditions, and meteorology may not
be available, particularly for a specific design value day in the past. Commenters stated that
updated emission inventories are needed in some cases and additional guidance should be
developed to clarify this issue. Also, some commenters stated that the Regional Oxidant
Model (ROM) outputs are not on a scale sufficient to characterize the more detailed spatial
patterns encountered for individual metropolitan areas.
EPA Comment: Research programs to develop/validate models and search for an optimum
national ozone strategy are currently under way within EPA. Development and refinement
11-5
-------
of the Emission-Based Modeling (EBM) methods required by the Clean Air Act are being
pursued within the base research program of EPA. Consistent with expected levels of
resources, EPA plans are for extensive field, laboratory, emission inventory and modeling
studies to be completed by 1998.
SUMMARY OF PUBLIC COMMENTS ON DRAFT REPORT
On March 14, 1994, EPA published a Federal Register Notice announcing the
availability of a draft report on the study for public review and comment. Prior to that
announcement, copies of the draft report were mailed to all parties that previously expressed
an interest in the study. More than 250 copies of the report were mailed out in response to
requests. As of the close of the public comment period on April 14, 1994, comments had
been received from only two respondents. Requests were received from several parties to
extend the comment period. On April 28, 1994, a second Federal Notice was published that
extended the public comment period until May 31, 1994. Although many additional requests
for copies of the draft report were answered during this period, only eight additional parties
submitted comments by the close of the comment period. The report was also sent to the
National Institute of Statistical Sciences for peer review. This section presents a summary of
the public and peer review comments received and EPA's responses.
Several public reviewers provided extensive and detailed comments. In addition, the
peer review effort resulted in the formulation of over 90 individual comments and
recommendations. A synopsis of all comments made by each organization is presented in
Appendix J. Comments of an editorial nature were reviewed and any necessary changes to
the text were made. All non-editorial comments were carefully considered and appropriate
responses prepared which, in many cases, resulted in changes to the text.
Written public review comments were received from 10 organizations:
Ford Motor Company (Ford)
General Motors (GM)
The Chevron Companies (Chevron)
New York State Department of Environmental Conservation (NYDEC)
Mobile Oil Corporation (Mobile)
E.I. du Pont de Nemours and Company (DuPont)
American Petroleum Institute (API)
California Air Resources Board (CARB)
11-6
-------
State of Ohio Environmental Protection Agency (Ohio EPA)
Zephyr Consulting (Dr. Kay H. Jones)
Most of the public comments received on the draft report fall into the same major
subject areas as those made during the public meeting, as discussed hi the previous section.
These are summarized again here and include any significant new aspects raised with regard
to the draft report. Additional subject areas were also raised in the public comments and are
also summarized below together with EPA's responses. Peer review comments and
responses are summarized in a later subsection.
Distinction Between Statistical Population Parameters and Estimated Parameters
At least two commentators (Ford and DuPont) indicated some confusion regarding the
distinction between theoretical quantities related to the statistical population of daily
maximum ozone concentrations and estimates of these quantities made on the basis of
observations from networks of ozone monitors. For example, there appears to be some
confusion about whether or not the term "EPA design value" refers to the theoretical quantity
known as the characteristic largest value (CLV) or the commonly used estimator of this
value, the fourth highest concentration in three years. Ford also makes the comment that the
CLV is not directly related to the expected value (i.e., average over many realizations) of the
EPA design value.
EPA Response: Changes have been made to the text of the report to make the distinction
between population parameters and estimators of those parameters more clear. The "true"
design value, or CLV, is an unknown quantity depending upon the unknown true distribution
of daily maximum ozone concentrations. The EPA design value is an estimate of the CLV.
Similarly, the "true" expected exceedance rate is an unknown quantity which, for compliance
purposes, must be estimated using the Appendix H calculations. The "true" design value
corresponds exactly to the "true" expected exceedance rate in the sense that the "true"
expected exceedance rate exceeds one per year if and only if the "true" design value exceeds
the standard. Similarly, the EPA design value and the ozone NAAQS exactly correspond in
the case of 100 percent complete three-year monitoring data. Any inconsistency between the
CLV and the EPA design value is due to the fact that the EPA design value is an estimator
of the CLV, and, like all other design value estimators, is subject to error. Furthermore,
while Ford is correct that the expected value of the fourth highest is not the same as the
CLV, the accuracy of an EPA design value for a given attainment period is assessed by the
difference between the EPA design value and the CLV, rather than by the difference between
the EPA design value and the expected value of the EPA design value. The bias of the EPA
design value is the expected difference between the EPA design value and the CLV across
infinitely many three-year attainment periods, which equals the expected value of the EPA
design value minus the CLV.
11-7
-------
Increase Stability of the Design Value Estimator
Nearly all commentators discussed concerns related to the year-to-year variability in
the current EPA design value estimator (i.e., the fourth highest daily maximum concentration
in three years). This variability can cause areas to change attainment status over consecutive
overlapping three-year periods, thus leading to uncertainties in the regulatory process.
A number of alternatives designed to reduce variability were suggested, including the
three-year mean (or median) of the annual second highest daily maximum, measures based
on the 95th percentile of daily maximum concentrations, and the mean of the 10 or 30
highest daily maximums.
Ford commented that the comparison of design values and 95th percentiles in the draft
report are not valid due to discrepancies in sample sizes, treating combined instead of
individual sites, misinterpretation of correlation diagrams, and misattribution of statistical
bias.
Ford also commented that design value estimators based on distributional assumptions
are very sensitive to the form of the assumed distribution and that the calibration factor used
in the CARB-Larsen tail exponential method to correct an apparent bias between design
values estimates obtained from this method and the EPA design value may not be equally
applicable to other sites or time periods.
EPA Response: As noted above, the ozone design value is, by definition, "the value that
should be reduced to the standard level thereby ensuring that the site will meet the standard.
With the wording of the ozone standard the appropriate design value is the concentration with
expected number of exceedances equal to 1" (EPA, 1979). Thus, by definition, the design
value is an extreme quantity which, inevitably, is subject to variability. The degree of
variability is compared to the variability of other air quality indicators in this report. For
example, it is noted in Section 3 that the annual 95th percentile exhibits smaller year-to-year
differences than the annual second maximum, although the site-to-site variability is similar
for both statistics. Furthermore, Finding #6 in Section 12 explicitly points out that "the EPA
design value is slightly more variable than the lower percentile estimators." However, the
reduced variability comes at the expense of bias relative to the "true" design value and
therefore, the alternative indicators cannot serve as a substitute for the design value.
In response to Ford's comments on the comparisons of 95th percentiles and the fourth
highest in three years (i.e., the "EPA design value"), we note:
1. As pointed out above, the appropriate comparisons to make are between the 95th
percentile and the CLV and between the EPA design value and the CLV, since these
indicate the biases of the design value estimators. The CLV has an infinite rather
than three-year sample size because the CLV is an unknown population parameter.
11-8
-------
2. Fitting the same regression line (CLV against three-year 95th percentile) to all
sites is appropriate because it would probably be unacceptable to calculate design
values using different regression estimates at different sites; estimating site-specific
regressions equations to provide equal public health protection at all sites would be
difficult and complex. Although nonlinear regression models might provide a better
relationship between the 95th percentiles and the design value, there is no guarantee
that a specific relationship would be equally applicable to all monitoring sites.
3. A perfect correlation between the 95th percentile and the CLV at a given site is
not expected.
4. Ford's observation that the sign of the bias of the regression estimates depends on
which of the two time-series models was fitted is interesting. Ford deduces that the
pattern in the estimated CLVs in Figure 14 may be entirely explicable from the
properties of the fitted time-series models, rather than reflecting a property of the
unknown true CLVs. This may be correct. However, since the best fitting of the
two candidate time series models was used at each site, it may instead be reasonable
to conclude that Figure 14 demonstrates the difficulty of obtaining a single universally
applicable design value estimate based on the 95th percentile because of significant
differences in time-series structures at different sites.
With regard to Ford's and GM's suggestion to use the 3-year mean of the annual
second highest daily maximum concentrations in place of the fourth highest over three years
(EPA design value), we note that this statistic is discussed in Section 10. A problem with
this statistic may be that it is more biased with respect to the CLV than the EPA design
value. That is to say, this statistic tends to underestimate the CLV more than the EPA
design value. This follows from the observation that, if the unknown underlying daily
maximum ozone distribution has an approximately exponential tail and the daily maximum
values are approximately independent, then the expected value of the (AM-l)th highest in N
years tends to increase with N, approaching the CLV for a sample of size N in the limit as N
grows to infinity. Thus in many (though not all) cases, the expected annual second
maximum is less than the expected EPA design value, and the expected EPA design value is
less than the CLV. The median of the annual second maxima, not specifically discussed in
the report, suffers from similar biases, greater variability, and is much less sensitive to the
observed air quality. As an extreme example, note that for many areas the median second
maximum for 1987-89 would have been the same if the 1988 concentrations had all been ten
times as great.
With regard to Ford's comment on the sensitivity of design value estimates to
assumptions about the underlying distribution of daily maximum concentrations, we note that
it is certainly true that the estimated design value from a fitted distribution may vary
significantly depending upon the selected form of the distribution. On the other hand, if a
good fit is found (especially in the tail of the distribution), then it is surely more reasonable
to use an estimated design value based on a well-fitting distribution than to use a different
11-9
-------
estimate based on a poorly-fitting distribution. Discussions in Section 4 include the point
that a fit to the tail of the distribution may well be more appropriate for estimating extreme
values such as the design value. Several commentators (Chevron, Mobil, CARB) and the
NISS peer reviewer support the use of the tail exponential distribution to generate design
value estimates.
With regard to Ford's comment about the CARB calibration factor, we note the
following comment made by CARB (letter from T. McGuire, May 27, 1994):
"The calibration factor was not based on data from the Bay Area monitoring network
only, although those data played a role along with data from many other sites around
the State. The calibration factor was not set so that our method would precisely
match EPA's fourth high estimate. In fact, we knew that EPA's method was biased
low (theoretically), and we chose not to make the two methods coincide. We also
acknowledged the possibility that our method could be biased high, and the Bay Area
data along with the other data throughout California helped determine a suitable
midpoint between the impressions from the EPA and uncalibrated CARB method."
Attainment Test
A number of commentators expressed concern with the NAAQS attainment test
specified in Appendix H of 40 CFR 50. These concerns were generally the same as those
expressed during the public meeting, i.e:
• Lack of consistency between the test used to determine attainment/nonattainment
status (which is based on exceedances) and comparisons of the design value with the
level of the ambient standard (0.12 ppm) which can arise when monitoring records
are incomplete (thus requiring the use of the missing data adjustments discussed in
Appendix H).
• Lack of consideration for year-to-year fluctuations in the design value in the current
attainment test leads to attainment/nonattainment flip-flops and the requirement for a
design value target that is "significantly" below the level of the standard (0.12 ppm)
to maintain attainment in the long run. In other words, the actual stringency of the
standard is greater than the "nominal" stringency implied by the level of the standard.
With regard to the first point, GM suggested that an analysis be included in the report of the
extent to which areas are classified as nonattainment as a result of the adjustment of observed
exceedances to account for missing values. GM also commented that the study should
include analyses of "too close to call" transitional attainment categories. API notes that it
supports the development of such a "transitional nonattainment" designation.
With regard to the second point, Ford suggested the use of a statistical attainment
test. Such a test, based on the student's-t distribution, has been proposed by Chock (1991).
11-10
-------
Alternatively, Fairley and Blanchard (1990) suggested a two-part test based on the number of
exceedances in recent years which is designed first to require a high level of proof in
demonstrating attainment for the first time, and then shifts to require a high level of proof
that attainment is no longer valid, thus reducing the probability of declaring an area that was
in attainment as nonattainment if, based on the theoretical value of the expected exceedance
rate, the area is actually still in attainment. Mobil suggested using estimates of the value that
is expected to be exceeded once in seven years (obtained from the CARB tail exponential
fitting procedure) to exclude extreme concentrations from data used to determine attainment.
GM commented that the attainment test applied to areas reaching post-1990 attainment
deadlines should be based on a comparison of the area's design value with the level of the
standard, and not on the expected exceedance rate, and that, therefore, the Design Value
Study should address issues related to the use of design values in determining attainment.
GM also suggested that the study should include a discussion of nonattainment area
classification methodology.
NYDEC and others commented that variability in design values and attainment
determinations could be reduced by increasing the length of the attainment period beyond
three years. DuPont suggested using periods of at least six years and noted that this is
"conservative" in areas with long-term downward trends in ozone. NYDEC noted, however,
that using longer attainment periods would increase the length of time before an area could
be declared in attainment. Furthermore, Ohio EPA noted that the use of longer attainment
periods would have only limited effect on reducing the influence on design values of an
extreme year such as 1988.
EPA Response: As noted above, the test used to determine attainment of the current ozone
NAAQS is specified in Appendix H of 40 CFR 50 (see 44 FR 8220). EPA was aware of the
"actual" stringency of the standard at the time of promulgation in 1979. EPA selected a
three-year attainment period to specifically limit the number of exceedances that could occur
in a single year in areas that are in attainment of the standard. Changes in the form of the
current standard, including the attainment test, are beyond the scope of this study. These
issues will be addressed in the forthcoming Staff Paper being prepared as part of EPA's
review of the ozone NAAQS. Thus, the analysis of missing values suggested by GM is
beyond the scope of this study. With regard to GM's comment on the attainment test for
post-1990 attainment deadlines, we refer to the General Preamble for the Implementation of
Title I of the Clean Air Act Amendments of 1990, which states that the attainment test for
areas reaching then- attainment deadlines will be based on the "average number of
exceedances per year" (57 FR 13506).
Adjust Design Values for Annual Meteorological Fluctuations
Ford, GM, NYDEC, Mobil, DuPont, Jones, and API recommended that EPA adjust
design value estimates to account for year-to-year fluctuations in meteorological conditions
since these fluctuations are not associated with changes in the level of controllable emissions,
11-11
-------
cause areas to flip-flop in and out of attainment, and can result in uncertainties about the
degree of emission reductions needed to achieve or maintain attainment.
Ford and GM recommended against using the meteorological adjustment method
developed by Cox and Chu (1992) as presently formulated since the adjusted annual design
values are largely determined by the trend term and the influence of changing emissions over
the trend period on the ozone-meteorology relationship is not accounted for. Both Ford and
GM recommended using a method that treats each year individually.
EPA Response: As noted in the previous section, compliance with the ozone standard is
judged on the basis of the actual ambient air quality measurements. It is the actual ambient
air quality, not a hypothetical adjusted value, which is of concern with respect to the
potential for adverse health impacts. EPA has long recognized the need to develop reliable
meteorological adjustment techniques for the analysis of ozone trends. A trend adjustment
procedure is currently under development as described by Cox and Chu (1992). With regard
to the concerns about the Cox and Chu method raised by Ford and GM, Cox (1994, personal
communication) acknowledges this possibility and notes in a preliminary analysis that
relationships between ozone and some meteorological variables do change over time in some
of the urban areas studied. Investigation into this issue is ongoing.
Influence of Transport from Nearby Areas
GM and NYDEC commented that the impact of ozone and precursors transported
from upwind jurisdictions should be taken into account in attainment/nonattainment
determinations. Ford commented that the TODV (Transported Ozone Design Value) model
used to separate out the upwind contribution to design values is inadequate and that the UAM
(Urban Airshed Model) should be used for this purpose. Ohio EPA noted that some regions
may not be able to demonstrate attainment via local emission reductions due to the
overwhelming influence of transport from upwind areas. It suggested that control strategy
design values that have been adjusted for transport and meteorological effects be adopted for
use in classifying nonattainment areas.
EPA Response: As a measure of local air quality, the design value is calculated using data
from monitors in the area in question. Section 8 of this report discusses transport issues and
notes that local air quality can be significantly affected by transport from upwind areas. As
noted above, the Clean Air Act gives the EPA Administrator some discretionary authority to
modify the initial design value classification based on factors such as "the level of pollution
transport" (Clean Air Act, Section 181(a)(4)). In addition, the Act provides for the
identification of rural transport areas and Transport Commissions for dealing with transport-
related matters. EPA agrees that the UAM is a more scientifically rigorous (though labor-
intensive) approach to estimating the impact of transport; the inclusion of the TODV model
results in this study is purely for purposes of illustrating the potential impacts of transport on
design values.
11-12
-------
Influence of Spatial Variability in Ozone Concentrations
Ford, GM, Chevron, Mobil, DuPont, and API noted that ozone concentrations vary
significantly from one part of a nonattainment area to another and that the EPA design value
does not adequately account for this variability. They recommended that an alternative
design value, based on measures of population exposure, be used. Chevron commented that
the design value does not provide reasonable information about current exposures and that
trends in design values do not necessarily reflect trends in exposures.
EPA Response: Spatial variability in ozone concentration is discussed extensively in Section
10. EPA agrees that population exposure measures based on data from multiple monitoring
sites provide a broader-scale indication of ozone levels than the maximum estimated design
value over the monitoring network. However, given the current form of the ozone NAAQS
and the Clean Air Act requirement that all population groups must be protected, the ozone
design value must be estimated at each monitoring site in an area, and the site with the
highest design value determines the level of emission controls required.
Response to Previous Public Comments
GM commented that the draft report did not include an adequate discussion of and
response to the public comments received.
EPA Response: The final report includes an expanded review of all comments received (see
synopses in Appendix I), and an expanded discussion and response to comments in this
section. In addition, a large number of changes have been made to the draft report in
response to both the public comments received and the extensive peer review comments (as
discussed in the next subsection).
SUMMARY OF PEER REVIEW COMMENTS ON DRAFT REPORT
Dr. Malcom R. Leadbetter of the National Institute for Statistical Studies prepared a detailed
peer review of the draft report. A complete synopsis of his comments is presented in
Appendix J. Overall, Dr. Leadbetter noted that "The EPA study is an extremely well
considered document which addresses issues raised and evaluates them fairly on the basis of
their scientific merit." He further noted that the EPA table look-up estimator, although not
perfect, is "reasonable" and has the advantages of being simple, statistically valid, and unlike
some estimators based on fitting data to parametric distributions, can be applied with equal
validity to any monitoring site. Dr. Leadbetter agrees with the public commentators in
pointing out that development of more "robust" design value estimators and air quality
indicators would be advantageous and suggests that alternative indicators, such as those
described in this report, continue to be evaluated and considered for use. He particularly
favors indicators based on the tail exponential distribution as these are both simple and robust
11-13
-------
but cautions that the universal applicability of the tail exponential distribution to all
monitoring situations has yet to be demonstrated.
Additional significant comments provided by Dr. Leadbetter and EPA's responses are
discussed below. Where appropriate, changes to the text of the draft report have been made
in response to these comments. Dr. Leadbetter supplied numerous additional specific
suggestions for changes to the text to improve readability and scientific accuracy as indicated
in Appendix J. Changes to the text of the report have been made incorporating these
comments.
Comment: One undesirable feature of the table look-up approach is the "higher than
desirable" probabilities of misclassifying an attainment area as nonattainment and vice versa
(see example calculations in Appendix K. For a given "true" design value, alternative
approaches may result in lower misclassification probabilities. However, since
misclassification will always be a problem for areas near the attainment/nonattainment
boundary, implementation of a "too-close-to-call" (i.e., transitional nonattainment) category
would be advantageous.
EPA Response: As noted in Dr. Leadbetter's detailed comments, the high variability (and
resulting misclassification probabilities) of the EPA design value and most of the proposed
alternative design value estimators is primarily a result of its relationship to the ozone
NAAQS, which causes the estimated design value to depend only on the extreme data values.
Changes in the form of the current ozone standard, including changes to the 1990 CAAA
ozone design value classifications, might lead to lower variability of the estimated design
value and reduced misclassification probabilities. Such changes, including the "too close to
call" category, are beyond the scope of this study but are being addressed in the forthcoming
Staff Paper being prepared as part of EPA's review of the ozone NAAQS.
Comment: A distinction needs to be made between the true design value, expected to be
exceeded once per year, and estimated design values.
EPA Response: This comment relates to the apparent confusion in terminology as discussed
in connection with some of the public review comments above. The "true" design value is
an unknown parameter of the underlying distribution of ozone concentrations and is the value
expected to be exceeded with a probability of 1/365 assuming a 365-day ozone season. The
true design value, assuming independent, identically distributed daily maximum ozone
concentrations, is also known as the characteristic largest value (CLV). The EPA design
value (based on the table look-up procedure) is indeed an estimator of this unknown
parameter, as are other design value estimators discussed in the study. Changes have been
incorporated into the text of this report to alleviate potential confusion between unknown
"true" parameters and estimators of those parameters. To retain the readability of the
document, the text does not always distinguish estimated and true design values where the
meaning is clear from the context. Where there is potential for confusion, the terminology
"true design value" is used to indicate the unknown population parameter.
11-14
-------
Comment: The study focuses on the fundamental issue of comparing methods of estimating
the design value but inadequately treats the important issue of how to take spatial variation
into account when using design values at different sites to assess area air quality. In
particular, under the current EPA practice of focusing on the monitor with the highest
estimated design value, a determination of nonattainment may be more likely for areas with
more monitoring sites, thus resulting in potential inequities between areas with different
monitor densities.
EPA Response: The spatial variation in ozone concentrations is an important subject for
continued research, and several aspects of this problem are considered in the report. The use
of the highest monitoring site design value for classification and nonattainment determination
is justified by the CAAA requirement to protect public health throughout the attainment area,
including the health of people who might be exposed at or near the site with the highest
concentrations.
Comment: Comments on alternative forms of the ozone NAAQS may be relevant to the
design value study since a revised ozone NAAQS might better correspond to a different
design value.
EPA Response: While a NAAQS different from the one currently in place would require the
formulation of different design value estimators, EPA's interpretation of section 183(g) of the
Clean Air Act is that the design value study is to focus on design value methodologies
consistent with the current NAAQS. Thus, design values consistent with alternative NAAQS
are beyond the scope of this study. Potential changes to the NAAQS and associated design
value issues are being addressed in the forthcoming Staff Paper being prepared as part of
EPA's review of the ozone NAAQS. Obviously, findings from this study will be an
important consideration in the review process.
Comment: The conclusion section should discuss the crucial issues of the performance of the
design value estimates and their ability to accurately classify nonattainment areas.
EPA Response: Appendix K includes some additional analyses of misclassification
probabilities prepared by Dr. Leadbetter and the "Major Findings" section has been modified
to discuss these issues.
Comment: It is not clear why using alternative design values for nonattainment area
designations and classifications would require revision to the NAAQS and the CAAA.
Although consistency between the design value and the NAAQS attainment test is "natural,
aesthetic, and traditional," departures from this consistency would be justified if
improvements in statistical properties (i.e., "robustness") could be achieved. Thus, the
arguments made in Section 6 that a poor regression fit between the 95th percentile and the
CLV implies that the 95th percentile gives a poor design value estimate are not convincing.
11-15
-------
EPA Response: Although the EPA Guideline for the Interpretation of the Ozone NAAQS
suggests several alternative procedures for estimating the design value, including the table
look-up procedure and methods based on fitting distributions, Appendix H of 40 CFR 50
precisely specifies the calculation of expected exceedances used for the attainment test.
Using an alternative design value such as the 95th percentile for the attainment test definition
would therefore require a revision to the current ozone NAAQS. Further, the 1990 CAAA
specified that the 1990 nonattainment area classifications were to be made using the approved
design value methodology in place at the time of enactment, i.e. the EPA tabular approach.
Strictly speaking, use of an alternative design value for classifying new ozone nonattainment
areas would not require a revision to the ozone NAAQS. However, according to the Ozone
Guideline, a design value has to be a reasonable estimator of the value expected to be
exceeded once per year ("true" design value). Since most of the summary statistics proposed
on page 3-19 of the draft (including the 95th percentile) are not reasonable estimators of the
"true" design value, they cannot be used directly as design value estimators. Furthermore,
the poor regression fit suggests that a simple linear function of the 95th percentile will not
provide a suitable design value estimator applicable over all sites and monitoring periods.
The scatter evident in Figure 6-14 suggests that site-specific factors would have to be taken
into account, thus a simple method could not be applied with equal validity at all sites and
attainment periods.
Comment: At the end of Section 3, the text wrongly implies that a deterministic relationship
to the current NAAQS design value, rather than just a statistical relationship, is required to
justify the use of the 95th percentile.
EPA Response: This is a misreading of the text, which has been rewritten for clarity.
Comment: Figure 5-4 demonstrates the fact that the distribution of ozone concentrations is
not the same throughout the ozone season, which suggests that the process of estimating a
design value using a single distributional fit as in, for example, the tail exponential
approximation, is inherently contradictory.
EPA Response: It is reasonable to assume that the distribution of ozone concentrations is not
the same throughout the ozone season. However, the EPA design value approach, the tail
exponential approach, and several of the other preferred design value approaches are based
only on the tail of the ozone distribution which is likely to contain values from only a small
period within the ozone season. Thus these design methods are essentially based on a single
distributional fit to a small portion of the ozone season. The conditional probability approach
described in Section 4 is a relatively complex method that might be used to directly treat the
nonstationarity issue.
Comment: The procedure for estimating exceedances on days with missing monitoring data
suggested by Davison and Hemphill (see reference 15 in Section 5) represents a potential
improvement over the current Appendix H method and suggests the need for a review of the
Appendix H methodology.
11-16
-------
EPA Comment: As noted above, such changes are beyond the scope of this study.
However, the treatment of missing values is being addressed in connection with the ozone
NAAQS review currently underway.
REFERENCES
Chock, D. P. 1991. Issues regarding the ozone air quality standards. J.Air Waste Manage.
Assoc.. 41:148-152.
Cox, W.M., and S.-H. Chu. 1992. "Meteorologically Adjusted Ozone Trends in Urban
Areas: A Probability Approach," in Transactions of the Tropospheric Ozone and the
Environment n International Specialty Conference. Air and Waste Management
Association, Pittsburgh, PA, 1992.
Curran, T.C. 1992a. "The Clean Air Act Ozone Design Value Study," in Transactions of
the Tropospheric Ozone and the Environment n International Specialty Conference.
Air and Waste Management Association, Pittsburgh, PA.
Curran, T.C. 1992b. "Present Status and Current Program," panel discussion presented at
the Tropospheric Ozone: Nonattainment and Design Value Issues International
Conference. Air and Waste Management Association, Boston, MA, October 28,
1992.
Curran, T.C., and N.H. Frank. 1991. "Ambient Ozone Trends Using Alternative
Indicators," in Transactions of Tropospheric Ozone and the Environment. Air and
Waste Management Association, Pittsburgh, PA.
Curran, T.C, and W.P. Freas. 1991. "Consequences of a More Statistically Robust Ozone
Air Quality Standard," presented at the Air and Waste Management Association
Annual Meeting, Vancouver, BC, June 18, 1991.
EPA. 1979. Guideline for the Interpretation of Ozone Air Quality Standards. U.S.
Environmental Protection Agency (EPA-450/4-79-003).
EPA. 1992. National Air Quality and Emissions Trends Report. 1991. U.S. Environmental
Protection Agency, Research Triangle Park, NC.
Fairley, D., and C. L. Blanchard. 1990. "Rethinking the Ozone Standard." Air and Waste
Management Association, 83rd Annual Meeting, Anaheim, California.
Freas, W.P. 1991. "Ozone Data Analysis: Design Values, Meteorological Adjustment and
Trends" (session chair), Tropospheric Ozone and the Environment n International
Specialty Conference. Air and Waste Management Association, Atlanta, GA,
November 6, 1991.
11-17
-------
Freas, W.P. 1992a. "EPA's Design Value Study," in Tropospheric Ozone: Air Quality
Attainment Issues. Air and Waste management Information Exchange Workshop,
Research Triangle Park, NC, September 16, 1992.
Freas, W.P. 1992b. "Clean Air Act Ozone Design Value Study: Impact of Varying the
Number of Data Years on Estimates of Ozone Design Values," presented at the
Tropospheric Ozone: Nonattainment and Design Value Issues International
Conference. Air and Waste Management Association, Boston, MA, October 29,
1992.
NRC. 1991. Rethinking the Ozone Problem in Urban and Regional Air Pollution. National
Research Council, National Academy Press, Washington, DC.
11-18
-------
12 SUMMARY OF FINDINGS
This report responds to the requirements of Section 183(g) of the Clean Air Act
Amendments (CAAA) of 1990, which requires that
"The Administrator shall conduct a study of whether the methodology in use
by the Environmental Protection Agency as of the date of enactment of the
Clean Air Act Amendments of 1990 for establishing a design value for ozone
provides a reasonable indicator of the ozone air quality of ozone nonattainment
areas. The Administrator shall obtain input from States, local subdivisions
thereof, and others. The study shall be completed and a report submitted to
Congress not later than 3 years after the date of the enactment of the Clean
Air Act Amendments of 1990. The results of the study shall be subject to
peer and public review before submitting it to Congress."
With passage of the Clean Air Act Amendments (CAAA) of 1990, added emphasis
was placed on ozone design values. In addition to designating areas as nonattainment for
ozone, the CAAA introduced a classification process to further categorize nonattainment
areas according to the extent of their ozone problem. This area classification was based
upon the ozone design value. The CAAA stated that the design value "shall be calculated
according to the interpretation methodology issued by the Administrator most recently before
the date of the enactment." Before the 1990 CAAA, designation of nonattainment areas
simply involved a yes/no determination as to whether the area met the standard. The
additional classification step introduced by the 1990 CAAA placed greater emphasis on ozone
concentration observations and on the methodology used to determine the design value.
Another reference to the use of design values is contained in Section 181(b) of the
Act, which states that EPA "shall determine, based on the area's design value (as of the
attainment date), whether the area attained the standard by that date." EPA's preliminary
interpretation of this Section (EPA, 1992) is that the "average number of exceedances per
year shall be used to determine whether the area has attained," which is the attainment test
for the ozone NAAQS. Thus, design values are not used to make this determination,
although the determinations based on Appendix H and based on the EPA design value are
identical in the case of complete monitoring data.
The focus of the Ozone Design Value Study is on EPA's design value methodology as
initially developed in the Ozone Guideline and later defined in current EPA guidance "as of
the date of enactment of the Clean Air Act Amendments of 1990." Issues concerning the
form of the current ozone National Ambient Air Quality Standard (NAAQS) are more
properly treated within the existing mechanism for NAAQS review, and the wording of
Section 183(g) clearly establishes that Congress was interested in examining EPA's ozone
design value methodology, not EPA's ozone NAAQS.
12-1
-------
Section 183(g) of the CAAA of 1990 requires the EPA to "obtain input from States,
local subdivisions thereof, and others." In conducting the Ozone Design Value Study, EPA
has made every effort to have an open process and to ensure full public input and
participation. These efforts have focused on information exchange through participation in
professional meetings and conferences, involvement of interested parties on the study
working group, and holding a public meeting in Arlington, VA on September 10, 1992.
On March 14, 1994, EPA published a Federal Register Notice announcing the
availability of a draft report on the study for public review and comment. Prior to that
announcement, copies of the draft report were mailed to all parties that previously expressed
an interest in the study. More than 250 copies of the report were mailed out in response to
requests. As of the close of the public comment period on April 14, 1994, comments had
been received from only two respondents. Requests were received from several parties to
extend the comment period. On April 28, 1994, a second Federal Notice was published that
extended the public comment period until May 31, 1994. Although many additional requests
for copies of the draft report were answered during this period, only eight additional parties
submitted comments by the close of the comment period. The report was also sent to the
National Institute of Statistical Sciences for peer review.
Comments received during the public meeting and in the draft report can be grouped
into two major categories: (1) those relating directly to design value issues and (2) those that
would require changes in legislation or a revision to the form of the ozone standard. Those
in the first category include issues concerning (1) the statistical robustness of the current
design value methodology, (2) the precision and accuracy of ozone monitoring data, and (3)
the use of other statistical techniques, such as fitting a tail-exponential model, for
determining the design value. The second category includes issues associated with changing
the form of the ozone standard to a more robust air quality indicator, or proposing to modify
the attainment test to incorporate a statistical test, such as a "t-test" for judging compliance
with the standard. Such changes are beyond the scope of this study and are more properly
addressed during the next ozone NAAQS review.
Many comments were received that raise issues related to the concept of a "control
strategy" design value, not the air quality design value. Recall from earlier discussions, that
adjusting for factors such as transported ozone, meteorology, and emissions trends falls
within the control strategy design value concept, not the air quality design value methodology
used to classify ozone nonattainment areas under the CAAA of 1990.
MAJOR FINDINGS
The Ozone Design Value Study has examined the current EPA method, as well as
alternative approaches, for calculating ozone design values. The key findings of the study
are as follows:
1. With passage of the Clean Air Act Amendments of 1990, the primary role of the
air quality design value is to establish the ozone classification of ozone nonattainment areas.
12-2
-------
2. Although the year-to-year differences in ozone concentrations can be large, all of
the different methods for estimating air quality design values show similar patterns in the
year-to-year variability.
3. Increasing the number of years used to estimate the design value reduces the
year-to-year fluctuations. Comparisons made for 3-year periods ending in 1988-90 had less
variability in the design value estimates than during previous 3-year periods. This is likely
due to the fact that there was a single dominant year (1988) for peak ozone levels during the
1988-90 time period.
4. The past decade has seen large year-to-year variability in ozone concentrations.
However, the relative variation in ozone concentrations recorded among monitoring sites
throughout large urban areas can be as great as, or greater than, the year-to-year variation in
ozone concentrations recorded at a particular monitoring location. Spatial variations hi ozone
concentrations at smaller, sub-metropolitan length scales are not well defined in many areas
due to the sparsity of ozone monitors.
5. The EPA tabular design value method tends to give lower, but more variable,
estimates for the ozone design value than some of the statistical modeling methods, such as
the Breiman tail exponential approach. Results of the time-series modeling analysis suggest
that the tail exponential approach provides the best compromise regarding bias and precision
in the estimate of the "true" design value.
6. The EPA design value is highly correlated with some less variable air quality
indicators. However, the EPA design value is slightly more variable than lower percentile
indicators.
7. Given the database available at the time, generally data through 1989, the use of
more robust (less variable) methods such as the tail exponential approach would not have
significantly changed the initial ozone nonattainment area designations and classifications.
Use of more years of data (i.e., 4 or 5 years) hi estimating the design value would have
resulted in lower classifications in only a limited number of cases. However, more recent
data periods that do not include 1988 yield significantly different results. For the years
1989-91, the first 3-year compliance period that excludes the 1988 data, 42 of the original
classified 98 nonattainment have ambient ozone meeting the standard. Seven of these areas
have been redesignated to attainment.1 The most recent compliance period, 1991-93, has 48
of the remaining 91 classified nonattainment areas also meeting the ozone standard.
8. Since the "true" design value is hi the tail of the ozone concentration distribution,
the EPA tabular design value method and more robust alternatives are perforce subject to
greater variability than estimators of the central part of the distribution. Like any statistical
1 Areas must apply for and meet certain administrative requirements to be redesignated in addition
to meeting the expected exceedance limitation spelled out in the ozone NAAQS.
12-3
-------
estimator, errors in these estimated design values can lead to 1990 CAAA misclassification
of nonattainment areas, just as errors in the Appendix H estimated expected exceedance rate
can lead to misclassification of attainment areas as nonattainment areas and vice versa.
Analyses included in this study provide estimates of the theoretical misclassification rates, but
for a given site and monitoring period it is impossible to determine whether the estimated
classification is the (unknown) true classification.
9. The "air quality" design value differs hi concept and application from the
"control strategy value." The former is based solely on the actual measured ozone air
quality data and relates directly to the form of the ozone NAAQS. Control strategy design
values have historically been used to evaluate emission control strategies, and may
incorporate adjustments for factors such as transported ozone levels and meteorological
variability. Use of the control strategy value concept to judge attainment under the Act
would require EPA to revise its preliminary interpretation of Section 181(b)(2) published in
the General Preamble to Title I.
10. For thirty-five areas modeled, the transport contribution to design values in
areas subject to transport was found to be as large as 0.05 ppm. Increases in the design
value of up to 0.04 ppm were estimated when the downwind impact was attributed back to
the source area.
11. EPA has initiated a program (Cox and Chu, 1992) to investigate techniques for
adjusting ozone trends for meteorological influences. One of the methods being studied is a
statistical model hi which the frequency distribution of ozone concentrations is described as a
function of meteorological parameters. EPA is seeking to review and expand the technical
basis for the methodology under a cooperative agreement with the National Institute of
Statistical Sciences (NISS). Preliminary results suggest that the bias and uncertainty
associated with long-trend estimates can be significantly reduced by including meteorological
covariates as parameters in the statistical modeling process.
12. The use of a simple linear function of the 95th percentile of the distribution of
daily maximum ozone concentrations as a surrogate design value is less satisfactory than any
of the four more direct estimators of the design value. It fails to significantly reduce the
variability of the associated estimated characteristic largest value (CLV) below that achieved
with the more direct methods. (From another perspective: controlling the 95th percentile
fails to improve control of the underlying CLV.) At the same tune it introduces substantial
biases which vary with the site. The bias problem would result in uneven treatment of sites
relative to what would be achieved with the more direct measures. Nor would the use of the
95th percentile obviate the need to use 3-year data sets.
12-4
-------
CONCLUSIONS
The question for the Ozone Design Value Study is "Does the EPA design value
methodology provide a 'reasonable indicator of ozone air quality in ozone nonattainment
areas'?" The answer depends on the intended application of the design value. Each
nonattainment area was classified as a Marginal Area, a Moderate Area, a Serious Area, a
Severe Area, or an Extreme Area based on the design value for the area. The area's
classification establishes the primary standard attainment date and the requirements for State
Implementation Plans.
The ozone design value is a surrogate measure of attainment status, a measure of
progress, and an indicator of how much concentrations must be reduced to meet the standard.
The EPA design value method yields an estimate for the ozone design value that is consistent
with the current ozone NAAQS. Given the findings of this study, the EPA design value yields
a "reasonable" estimate of the "true" air quality design value for the area and of peak ozone
levels within the nonattainment area for the initial three-year compliance period.
The EPA design value provides a reasonable estimate of peak levels within the urban
area, and the degree of nonattainment of the area. However, the design value cannot describe
the spatial variability in ozone concentrations across the area. More robust indicators based on
specific monitoring sites also have large spatial variability. Due to the spatial variability
observed across urban areas, one cannot expect a single numerical value to adequately describe
complex concentration gradients.
The current EPA design value method may not provide a reasonable indicator of ozone
levels in future years due to the large year-to-year variability in meteorological conditions, or
to reductions in emissions from implementation of control measures. Other more robust air
quality indicators also exhibit similar year-to-year variability.
The method used to adjust for meteorological influences on long-term ozone trends
could be adapted for use in calculating meteorologically adjusted exceedance rates and design
values. While such adaptations are technically feasible, and would reduce the year-to-year
variability, the use of adjusted exceedance rates in NAAQS attainment and adjusted design
values for classification purposes would represent a major departure from current EPA policy
and NAAQS implementation guidelines. Also, a meteorologically adjusted design value may
not be the best indicator of the air people breathed during a specific calendar year. Such a
major change is more appropriately considered within the context of the ozone NAAQS review
process as this adjustment could affect the level of health protection intended by the standard.
Concerns about the current ozone standard were raised during the public review
process. Although changes to the form of the ozone standard were outside the scope of this
study, they are being considered within the context of the current review of the ozone
NAAQS. The knowledge gained from the input of all parties to this study during the public
review process will be used to address issues concerning the form of the ozone standard and
design value methodologies.
12-5
-------
REFERENCES
EPA. 1992. "State Implementation Plans; General preamble for the Implementation of Title
I of the Clean Air Act Amendments of 1990; Proposed Rule" (57FR13498), April 16,
1992.
Cox, W.H. and S.H. Chu. 1992. "Meteorologically Adjusted Ozone Trends in Urban
Areas: A Probability Approach," in Transactions of the Tropospheric Ozone and the
Environment n International Specialty Conference. Air and Waste Management
Association, Pittsburgh, PA, 1992, pp. 342-353.
12-6
-------
APPENDIX A
Summary of Articles Discussing the Fitting of Statistical Distributions
to Air Quality Data
A-l
-------
Summary of Articles Discussing the Fitting of Statistical Distributions
to Air Quality Data
The articles selected for this literature review are summarized in Table A-l. In the table, the
column headed "No." gives the reference number of the cited paper(s) used both in this
summary text and in the list of references. The ordering is arbitrary, except that related
papers appear consecutively in most cases. The column headed "Distribution fitted" gives
the distribution(s) analyzed in the paper, and, in most cases, fitted to an example data set
described in the columns headed "Species/Averaging time," "Region," and "Years." In some
cases the paper is theoretical only and does not describe applications to real data. Note that
"Ox" has been used as an abbreviation for oxidant data, since for many of the earlier papers
direct measures of ozone (O3) were not available. The column headed "Fitting method"
describes the statistical method used to fit the distribution to the data (see the next section for
descriptions of the terms used.) The column headed "Comments" briefly describes the key
issues discussed in the paper and the conclusions reached.
A-2
-------
CQ
G
tt)
2
55
u
"o
•C
4-1
tt)
01
c
•rl
4J.
-rl
to
| Years
c
o
•rl
Ol
tt)
tt
1
-rl
H
0)
-^ c
CQ -rl
tt) O)
-H id
U rl
tt) tt)
ft >
co ft
T)
tt)
4J
4J
-rl
to
C
0
•H
4J
3
XI
-H
rl
4J
m
-rl
Q
0
53
>, 8
rH 3
rH g rH
18 -H (d
U X G tt)
-H Id 0 Q) E
i-l g -rl X! -H
-rl 4-1 4-) 4J
&T3 M
E C 0 id ft rl Id
4J -rl ID 1-1
4J "0 Q) S 0)
•rl tt) rl O >
to g id ftid
(Q
0)
i-l
-H
4J
1
o
s
4J
•H
f
s «
H !~)
a
id co
- CQ -H
-HO X! » - G
4-> hH ft CQ 0 O
- Id -rH -H CJ 4J
OG * CQ tt) 3 CO Ol
cn G ri CD 'd o -H G
flj -H tt) r-l Id |4 U -H
CJ CJ > tt) rH C XI
-H c C O)-H • id cn
r| .,_) (JJ rf rj 4J ^J (^
UCJQ 3 & W to IS
*
v X
U ° rH
rl C?
/-\ rsj 0) 'H
r) O ^ tJl •
£5 d) (d CQ
„ C0 JN| 4)
rs - fl) 5
O O x > -H
co % o (d -u
•
rH
flj
£
Q
o
G
g1
^
CN
H
'd T3
m G tt)
G <« G 4J
mo o u
rl -H 0) -H 'd tt)
0) 4J CJ 4J tt) ft
ft ft C Id 4-> rT
id g tt) g u tt) •
ft 3 *d -H at E
ca G X ft S 3
m co tt) 0 rfX) E
tt) id ft W tt) -H
3. 0) ft g X
& D x) ft 01 3 id
-H rl G Id O g g
4-1 -rl rH -rl
-H tN CU X Ol
)H «I*H X! <4H Id O
U H 0 -U 0 g rH
"tt7
O
•8
CQ 0
'tt)
o
id
0)
tt)
CO
"oT
0
XI
id
tt)
tt)
m
•
rH
id
C
Q
o
G
I
tJ)
^
ro
id
CO -d 4J
§ rH S'o -S
rl Id
Oi u CQ cn d) i
0 -H R CO rH Ol
rl 4J 0 4) ft 0
ft CO -H C E rH
•H 4J T3 Id
rl 4-1 3 0 X 4-1
tt) Id XI 0 H -rl •
4J 4J -r| Ol i
-rl 4J 0) X)
Id > rl
4-> x; -H o -d
CJ 4J 4J rl tt)
0) Id rl 4J
g 0 u tt) id
tt) CO -H g
Vl r-l 4-> -rl MH
3 73 ft G 4-) O
CO 0) -H tt) CO
Id 4J 4J g tt) CQ tt)
0) (d rH tt) -H CJ
g U 3 rl 0) CO C
-H g 3 XI >i id
XI rH 01 rH -H
CJ Ql tt) Id C Id rl
(d cu Xi tt) id G id
U lH 4J g CJ Id >
•a •
0) C
ca id
id D
•rt E
-rl
G 0) 0>
-rl U 4J
6 § g
CO -rt -rl
tt) rl 4J
co id ca
D > tt)
1
1
1
,
co
0)
0) 0) g
AJ rH -H
O X) 4->
•d CJ -rl
tt) CQ U
•d '»-H o
G to > -H
tt) TJ m
ft-H c m
CO rH tt) -H
3 0 > E
CO CQ O tt)
•
rH
i O 3 O
tt) rl -rl XI §
X! 0 4J -H 3 XI CO
4-> 0) Id rl g 01 CJ
XI g 4J -H -H -H
CQ 4J -rl CQ X XI 4J
S X -H (d CO
fl) tt) O "d g rl -H
-H 3 rl tt) 4->
> H ft tt) tt) X! Id
at id ax; x; -u 4->
K > id 4-> 4J o m
1
;
1
1
•d
§
•
H id
id 4->
•d-H id
D 4J "d
4-> C
id a) I-H
rH G
-------
•o
I
-H
4J
g
O
to
4J
G
CO
i
u
T?
o
*G
4J
3
Fitting
ears
G
0
-H
O)
to
a
a
.r
EH
Ol
^ G
CO -H
CO O)
-H id
U rl
co co
a>
co <
TJ
(U
4J
4J
•r
b
j:
-r
4.
*•
X
•r
^
4.
a
•r
Q
C
Z
rt >,
ast squa
extreme
obabilit
0) O r)
! 1 JJ Pj
' «H
H H
rl
-H
4J
to
O
U
.C G
4J -H
3 to
o (d
CO CQ
t
fa
_rl
co K
to
(d
to
to •
E rH
to to
rl XJ
jj 6
wl
r-
(d M Q) >, •
4J tU 0) CO ,C 4J CO
Id JC S -H 4-) -H C
"0 EH 0 rH 0
a c OI-H -H
rl 0) O G X) 4J
OX! • rl -H -H (d Id
4-1*0 '004-)tQX)rH
D) 0) CO Id 3 O tO
4-1 Cfl r-H X) H 0 0) 0
-H tO tO rH -H -~ C to g 4->
Ono2o-Hto2'tOOrH
HI aid c -oxi 4J to G a
•§
O 4-*
.£ -rl
•r4 ^W
rH
to g o-o
ii W 4J CO
•H 4J
rH CO E U
X! £ CO
E 4J 4-> 4-> •
3 -H CO CO
g Ol rl73 CO
-H G 0 • 3
X -H Ol C rH
id CQ M o
S«
H H
t
£
ttf
Ol
•H
rf
U
-H
E!
^
CO
4J
id S to
* w co
I 4-> rH
•a co x>
§-H id
E -H
O CO rl
rl £! (d
O 0 >
•O
0)
e
0
4-1
to
§
n
4J
X •
O rH
U *
X «
O 0
m c
CO
Oi
to G 0
3-H 0 CO 4J
O 4J 4-1 • E
-H 4J T3 co to m
rl -H G IB G rl G
* u-i 0-H X -H
4-1 i-H -H 4J 0) 4J
CQ o id 4-> id id
CO E 3 rl CQ E
X)COrli-HX)4Ja) -H •
•H'OOrH-HG'O X«J
rlOG3rlt03tOOE
O .^j ' *Q 4J U *H ^3 W "H
m 4-> OI-H m c u I-H a x
totooto-HOGfdan
QgrHS-OUH^ldg
to
4->
C 4-*
CO -H •
E t* to
O CO
E • rH
T3-H
4-1 O 4J
0 0 G
E si id
•o 3 -H 5
0 SrH &
XJ-H CO
4J X M 0
SgrH 4J
H
r-
CTl
,_)
d
^
id
G
CO
•o
id
CO
id
o<
T3
G X
tti itf
r4 fc^ ^ i>^
CJ r* rl
(V) rH f*irH
rl rH rl
. 3-H 3
KO Id 0
o as -ox:
to
tO >i
to to
rl 4J
4-1 id
-W -H
rH X
id 0
E * rl
M rH a *
O rH a^H
G 3 rtl 0)
' X) Xj
OI-H to g
o to n 3
en
CO i-H
to 4-) id
x: id 3 g
OI-H G E co
(4H G 4J Id -H >
O -H CO X
U -H O 4J 73 rH
G W 4J U 03
to id to • 0 to
Q) g co to a E »G to
Id 0 O 0 tO E rH
am -H j; -H to jj
Ocog4JErHXJ
to
4J 4J
G -H
to b •
E to
o to
g • rH
•0-H
m O 4J
O O C
T3§--S3
0 g rH &
X3-H CO
4J X J«! 0
Q) (ti 'H ^
SSrH 4J
i
i
i
i
i
Ol
0 •
rH It)
4J
T) id
CO T3
4->
(d I-H
C p
-H O
CO G
.
rH
C
^
O
G
t
0)
S
o
rH
-------
T3
4)
3
G
•rl
4J
§
U
g
03
4J
C
d)
5
u
y?
O
4J
S
S
Oi
G
•H
AJ
4-)
-H
to
CO
M
(d
0)
>H
g
-H
Di
Q)
M4
4)
E
-H
H
Oi
-•v G
CQ-rl
4) 01
-H id
U rl
4) 4)
a^
•o
0)
4J
•rl
b
G
0
-H
4J
•H
^|
4J
CQ
•rl
Q
0
55
C "O CQ
O 4) A T3 G rH
•H Vi U 4) 4) O Id
4J 4) -id .G tQ -H 3 •
3 4-1 CO 0 4J CJ ' 4J G CO
XI 4H ^i in 4) 3 G >i
-ri-ri4Jid a in S XI id id
rl 4J T5 G 4) a 0 • 4) -rl -d
4J3 4> >i id id CD 4H CQ JH >H 4)
CO XI CQ Vl 4J4) 414J4J4J4)
•H M4)4-i'da 4-1 "O-Hld
4> S £ TJ a COCa-rlVlrl4>4>4)ca4>
XI (d 304J4>4H4>4>O4J34)4)
g 4-1 lH |H 0 01 0 C 4J CQ >H 4J i-H U
3-HldOrlrHOOOCXt4>-HldOX
O 4-1 ft 4-1 Dirt! XI 4J U -H 0 ft4H > 4J 4)
4J
-rt 4) 4J
Id ft
4) >
H >i
id co 4J
D1 4) rH
CO M -rl
[I Q
jj J^ (Q
CQ 4) XI
Id 0
4) O H
Hi 4J a
I
GO
SH
•a u
§ -
4J
13 in
H 0
o a
M-l 0)
4J 01
id TJ
^_| -rH
li ^_t
00 CQ
X
id
E *
^
^1 r- 1
i-H i-l
. -H 3
m Id O
O Q .G
4)
id
id
g
-rl
S
r-l
^i 4}
r-l Xj
•rl c
id 3
Q O
H
H
T3 01 *H
G G 0 •
Id -H 4-1 CO
rH 4) 4J 4J
•rl CO JH J5 CQ U
Id G Id O14J 4)
4J 0 -n CQ 4H
•H CQ 4) 3 4-1
tH 4J 0 S -r-i 4)
0) Id rH 13
ft M -rl rtj id rH
ft4J 4J Id
3 C C 4) G
4) 4) • 6 O
>, U U T3 4) CQ
rH G W 4) J3 Id
G 0 4) CQ U 4)
O U ft 3 CQ CQ
•
•o
8
R T-l
S rH
-rl 4)
IJ ^
Irl
. ^
f*- ^"
H H
T)
c
id
CQ ft,
CD U
r-l
0) ~
01 73
G G
f^ id
rH
CQ ,V
o id
rl 0
rH
IN*
rn 4) C71
> G
^ 4) -H
r*i CO Ol *
O id to
• in 4)
- ft 4) S
O W > -rl
U EH Id 4J
rH
id
•H
4J
G
4)
c
O
4)
i-H
-H
id
H
^
4) T3 £
4J 4) 4J rH
4J 4J • 4J -H 4)
-H 4) rH U S XI
m XI id 4) §
6 ft Id rH 3
Id G ft rfErH O
4J I-H O O 0) -H 4)
Id rH -H G X S U •
73 3 -U ' "O id -rl CQ
X) 3 Ol 4) 6 'O 4J 4)
S-HX1O4J 4)O4J
^H ^ VH £ id id ft c
4J G -rl 3 ft g -H
4) 4) CD Id 4J G Is ^i 4J
rn5''d5w§8SS
4->
-H VH
4-1 4)
a
m ft
4) 3 •
in CD
id G 0)
&-rl rH
Id -H
ca 4J JJ
rl G
JJ (D D
CQ U U
id VH
4) O 4)
j jj a
vo
^
H
Q
§
*.
^
4)
CQ
CQ
^,
rH
. rl
2 X
*
rH
id
B rH
0 rH
C 3
' X)
Ol-H
0 4)
JIZ
n
H
-------
•o
0)
g
-H
JJ
§
U
I
nments
§
\J
U
•o
0
X!
JJ
cu
55
01
G
•H
JJ
JJ
-H
CO
M
id
CU
G
0
-H
01
UJ
K
CU
E
Oi
~*^ C
CQ -H
0) C
U r-
cu a
co a
•a
0)
jj
jj
•r
fa
G
C
-r
4-
i
•r
4_
a
C
0
S5
<*>
cu o
01 CU X!
G X! EH CO
rH -H JJ Id «*> •
nj ca S in jJ
3 X! 3
X) Id 0 V< XI -H X-H 0) H tH
O g U a JJ IM CUT) E -"—
t
•g
o
3 -H
£ rH
-H 0)
iN
CO
H H
id
JJ
id
I~j Q
c o
O-H
JJ JJ
id cu
O J£
c{ H G
ex cu id
w «; CQ
flj
. g •
r— 1 V-f
«-H 3
o id o
U QX!
.
rH
id
-H
JJ
a
cu
c
0
8-
0)
rH
•H
id
rH
0 0
U) G I cu 0 00
-H o m r-J JJ •
IM -H 01 m r* •
JJ to Ol O U
jj 3 >, G w -H
CO XI rH m -H O JJ
CU -H rH M T3 Ol CQ
XI M Id » H O -H
JJ 3 O E JJ
CU CO CQ _ U rH (d
XI -H 3 m U O JJ
H T3 -"O id fee; CQ
T)
I
3 -H
grH
•H 0)
I'rH
i
r^ ??
H H
^ .
a) id
C-H
J_( r— (
3 id
O M
XI JJ
rH CD
CU 3
S <
«.
,lsf«fs
CU rH J . Q)
•a 3 . o s
G OQ y
1
ttJJ U .rH
CQ to , _ri-H
3 Id mO (d
co no 2; Q
. *
rH ffl
id JJ
g cu •
H X! 0)
0 !H
C - 3
c rH JJ
OlrH (d X
O rH >-H -H
rH 3 U 4J E
XI -H G
»-H JJ 0) rH
id cu CQ c id
| IS -H 0 g
O JJ rH CD G
in
H
CQ Ol •
T) C CU T3
O Id -H rH 0
XI -H tt) JJ -H 0
JJ r» CU ^ Id JJ X! •
tt) CXIrdEC-HjJ
g t3 0 JJ 3,-H CU rH CO
0) -H O1 JJ U d) 0)
(J) ^1 JJ IM CQ CO M J^ XI
G id id 0 cu cu -H
•H ftrH G QM ca g u > g O
0) -H JJ O -H -H IM
x;cuxicaoi-i S ai
-O
0 JJ
c* m
t! id
CD cu o
g rH JJ
« « JJ CQ
•O CO -H V4 CD
O JJ i U
X pi4 3 C M
fd -H IM & id CD
g rH O tt E D.
1
1
1
•d
cu
id
rH
3
g
-H
CO
^
H
id
c
P
o
G
1 rH
Ol Id
0 -H
rH - JJ
rH G
-rH CU
id 3 G
EXl 0
g -H Q,
S CD X
O f. CD
H
-------
-H
4J
O
U
CO
4J
C
11)
5
n
u
•a
£
4J
s
en
•rH
4J
4-1
•H
b
m
rl
id
0)
6
0
•H
en
0)
0)
E
-H
Oi
\ C
CQ-H
cu Cn
•H (d
U rl
CU CU
cn <
•0
0)
4->
•H
tu
C
0
4J
XI
-H
H
4-1
CO
•H
Q
0
rH i
(d * td cu en co
Cn E Vi O T3
CQ 1 CO O M Id rH 0)
(d Cn i -H rH O (1)
SO Cn C co E tt
rH 0 CO >, 1 C O CO
CtlC 0) rH C rH 01 0 rl
COUx! 0 CU 0 -H W T3
•OO-HCUOOfl)O
4J
0 rH
hH 4-1 a
gS
H H
.8
id cn
- CO -H
-HO x! • » C
4-> 1-5 tt CO 0 0
-Id - rH -H U 4J
CnCMCBTJO-HC
(d -H Q) H Id ij U -H
U U > 0) rH C X!
-H C C Cn-H • id CQ
CJ CJ fl rX f\| ^| |"TJ [J
t
^^
rH
Vl
o
.
o
u
n
•o
S j
"-' i
^ (d
r-H-.Cn
id VH
S
4-> O rH
CO rH XI >, rl
tQ fl) (d rH (D
cu to 4J -H cu x;
4J (d M-l C ^1 4J 4J
rd u o cu (d id (d
rl Ti > E rl
4J >, CU C -H
COCCnCUrHXrHrH
OE^icDEMEE
a) a > c o So o
Q -rl i -H30MH CUC
CQ3rH'O rHMH CU>^ •
-H -H CU H (d J«; in -H 3 CU
cuidco K^CQ (dCn4-)4-)
rH04J X!C4Ji-Hc>|oO)HE
ns-H cura4JWidXi 4J -H
-H4JOX1C 36 -HO rH4J
rH4-)34-)4-'(dd}4-) -H tQ CN 'O Id tQ
-rICX) 4J(l)4JCfl 01 C 0)
(dcu-H'OmnJidtHCQOO cn-n
4->C^d)OEE (UOi4Jid|4Ha>
0 4-1 4J -H -H »H >H 3
£1 X -rl -H JStflCUXJrH >£(d
f
•o
8
B X*
E rH
(Q -ri
SrH
|
03 °°
H H
t
•H
C
o
-rl
rH
U
X
16
E •
Sr?
rH )H
. -H 3
m Id O
o ox;
.
rH
id
•H
C
CU
C
o
a)
rH
•H
id
H
-------
•a
0)
c
•H
4-1
G
0
U
«<
w
I
CO
4J
G
0)
O
U
-o
o
X!
4J
0)
S
01
c
-H
4J
4-1
-H
ears
G
O
-H
O)
S
0)
S
-H
O)
^- G
CO -H
0) tJl
-H fl
U rl
CD 0)
ft >
-a
0)
4J
4-1
-r
b
G
c
4.
X
•r
i«
4-
03
-r
a
0
a
cn Q) a) ii ca
H CO X3 -H H-l < -H
3 4J x: o
rl 0) U ' S
0) O 01 CO 0) CO • fl
fl rl CU 4J 0 0) U Ol
a-d cu x: S-n o
0-Hfl iHrHOCDft
4J <4H CO fl fl 0 -H
•H *O CD U CO 4J rH 'O
rH g £< g X! Pi in CO 3 -rl
t^
o
k
.
I
§ H
g rH
•H 0)
LJ [^
if
1-1
-H
c
M W
O ff^ iT***
4-1 H
-H g
rH - iH •
fl co cu ca
t_) CD 4J 0)
4J tO 4-1
t- -H fl -H
CO CD W CO
^
(d
6 *
^*i
i>i r— t
rH U
. -rl 3
m fl 0
o ax:
*
rH
(Q
•rH
4J
C
0)
G
0
fi-
0)
rH
-H
fl
H
o
CN
rl XI
co 3 Cn
H -H CD CD 0 0 -H
•H x; x; x! to x;
fl 4J EH 4J rH
4-) 3 >iCN fl G
CO X) 0 rH C- CO
CO CD • 4-> H 4-> >
flX!r-H4J 3Vlfl-H
V -H 3 -H X! > 4-1 T)
CO 4J XI "
H
CN
g
0
rl C CU T3
0) 14-1 Q) • fl 0)
X! XI G rH g CD
4J GrHSCUfl33fl
fl fl xi cu e g -H
co CU6G4Jgfi-H£!
CU 0) g C 0 OGG
4JX! O-rlOlCtH-rlS
fl4JUC4JGO S
k -H i 3 -H -H CO 0) fl)
caO4JO-HflfllOxlGfl
G (UHhgrH-H4Jld6
0 CO g 4J -H 3 > -H -rl
cu-Hcux:-H(BO)HOflca
QX) CD4J T3 CD Q,O< M-l > CD
.
0
§ XI
3 -H
-H CU
13
i
i
i
i
H
(0
6
H
0
d
(T)
Q
H
H
(d
Cj
^
0
a
M
-------
•o
0)
G
-H
4J
g
U
nments
5
\J
u
"8
•G
4J
$
01
G
-H
4-1
4J
-rl
fc
CO
^1
id
0)
*
c
o
-rl
01
0)
Oi
0)
E
-H
EH
Cn
\ G
Cfl-H
Q) 01
•H fd
U M
0) 0)
a >
•a
m
4J
-H
Cn
C
0
-H
4J
X)
•H
4J
CO
-H
Q
0
S5
rl 4->
rH 0 i-l Id 4->
a) w m id £ u co
0) Xj 0 £• 4J G idOi
J3EIMCQT3M O MH G
4J 3 0) 0) 0 T3 -H -rl
OG-H4JGO) 4-> O S 0)
4J OrHld T) Id Wl rH Id
Id U -rl arH S-rl rH 0> 4J Q)
4J 4J 3 Id ^ Id 0 iH 0) O
co 4J -H id o o a o 4J u -H
o) a IH E u -H u -H
o|1m'x4j1d4JO)4j'ci
r-» «J i-*
flj 4J
a a-rl -rl rH -rl O iH
OOiG30)3 SE3 U
ft 4-1 ca 0) r-i -H id O-HidT3
GOU3-HM 'Grl r<
J3 T3 O G -rl rH -rt 0 -H rH a
EH Id 4-> O 0)
-H CO -H H
tH G 4J -H
Id -H Id 4->
Ol G Id rl G
O rH 0 Ol 4-> Id
rH -H -rl Id C 3.
Id 4-> 0) 51
0) 4J 3 G U
J3 XI 0 G O
4J 0) -H -H O £
,G ^ 4-1 U 4->
4J 4J 4J U
0 CO G 0) 4J
rH UH -rl 3 J3 -rl
CU O *O HH 4J f*i
i •
CN CN
VO C~
CT\ CT>
H H
tn
0
rd"
-rl -
ft Id
H 43 - •
0) Ol CO 0
•O G 0) Oi
Id -rl rH Id
rH E 1) U
-H H Ol-H
43 -rl G »G
Pi pf) ri| ft
O rH
u id
» 0) O1
~ lT r*
D4 r* *-•
M * CJ1
(d to
rl 0)
-H
co o id 4J
*
*- rH
rH (d
id -H
rl GrH
0 0) rH
G G 3
0 XI
0 ?S CD
tJ i > 0)
0) 4J J3 CO M 0) U
i-l C 4JCQ 0) 0)COC
0)0) G'H'OrH > Id
^U Ofl) C 3 GT3
COO) COEHO)JC 0 0 0)
T) cu a CD ft 4-> >i 4-i s a) u
OJ.-H rH Id U>,COGtJX
4J -ri CN t3 -ri • a fl)G4-*Ur^OGQ)
4JJJ 0)4-10) 4Jfl)-H6rH-HO
-HCG>aoia>id oirH g (d co^-^
cu i-i 4J m in a) idk(d4JidUhid
^a">oa^oS;fflra^5a)!H?!,
(
•0
8
E H
-H 0)
M WJ
SrH
CN O
oo cn
o> cr»
H H
id
•H
g
0
•H
rH
(d
u
E *
3 0)
-H 0
Xl-3
M
id 0
e ^
^i rH
rH rl
-H 3
id O
D Si
,_j
id
-H
4J
G
0)
G
0
0)
i-H
-rl
id
^
in
CN
-------
•8
u
g
u
w
rj;
1
CD
Comment!
1
4J
3
?
-rl
4J
4-1
-rl
b.
CO
rl
R)
CD
M
C
0
-rl
Ol
oj
Q!
0)
6
-rl
EH
Ol
CO -H
0) 01
•rl «J
U r!
0) (U
a >
w<
•a
/•
U
4J
JJ
•r
[i.
£
-r
4_.
X
-r
M
4J
CO
-r
Q
d
53
0) tfl >
> CO rH r! i O
ft) to n) a) -H u
J3 C e 4J nj £ ^
Jt ft 4-> 4J U -H
CD 0) 0 0) (0 g
C Jrf C ^3 T) O CO
0 CO i 4-1 i
•o -ri 01 a cu T) >
CU4->HO J301CO
4->3(t!rHCU4JC!nSk
4J ^3 rH (> -H O
-H -H -rl Q) rlV4OCa
4J -rl 4-> O (d 6 4J
co co m 4-> 4J o :5 r-n co
JS-H O3-HO&OCO
E-IT) (tj 4J J3 4-1 (0 CO W 4J
CO
4J
C
IH
0
"S
X!
4J
s
CN
r~
as
rH
u
Q
n
u
4J
Ol
C
-H
A
CO
1
•
>i
rH
V-l
3
o
K
.
o
U
n
0 rH
•H (0
Hj rH 5 "^
C It! CO 0 CO
-H -rl C C fi
JO 4J O O! 0
g fl -H 0 -H
O CD 4-1 rH 4J
°g36.H
Sfi'-ri'S-a
Q) 0) 4J tt) 4J
C CO CO CO
•H M-l -H O. -H
A 0 -O — "O
VO
-------
References
1. Larsen, R. I. 1969. A new mathematical model of air pollutant concentration,
averaging time, and frequency. J. Air Pollut. Control Assoc.. 19:24.
2. Larsen, R. I. 1971. A Mathematical Model for Relating Air Quality Measurements
to Air Quality Standards. U.S. Environmental Protection Agency.
3. Patel, N. R. 1973. Comment on "A new mathematical model of air pollution
concentration." J. Air Pollut. Control Assoc.. 23:4.
4. Holland, D. M., and T. Fitz-Simons. 1982. Fitting statistical distributions to air
quality data by the maximum likelihood method. Atmos. Environ.. 16(5): 1071-1076.
5. Gleit, A., M. R. Leadbetter, and A. R. Manson. 1984. Estimation for lognormal
data with multiplicative errors. J. Air Pollut. Control Assoc. , 31(3):237-241.
6. Roberts, E. M. 1979. Review of statistics of extreme values with applications to air
quality data. Part I. Review. J. Air Pollut. Control Assoc.. 29(6):632-637.
7. Roberts, E. M. 1979. Review of statistics of extreme values with applications to air
quality data. Part H. Applications. J. Air Pollut. Control Assoc.. 29(7):733-740.
8. Stoline, M. R. 1991. An examination of the lognormal and Box and Cox family of
transformations in fitting environmental data. Environmetrics. 2(1):85-106.
9. Georgopoulos, P. G., and J. H. Seinfeld. 1982. Statistical distributions of air
pollutant concentrations. Environ. Sci. Technol.. 16(7):401A-416A.
10. Mage, D. T., and W. R. Ott. 1984. An evaluation of the methods of fractiles,
moments and maximum likelihood for estimating parameters when sampling air
quality data from a stationary lognormal distribution. Atmos. Environ.. 18(1): 163-
171.
11. Rao, S. T., G. Sistla, and J. Y. Ku. 1987. "Temporal and Spatial Variability of
Ozone Concentrations in the New York Metropolitan Region." Air and Waste
Management Association International Specialty Conference, Hartford, Connecticut
(November 1987).
12. Breiman, L., J. Gins, and C. Stone. 1978. "Statistical Analysis and Interpretation of
Peak Air Pollution Measurements." Technology Service Corporation, Santa Monica,
California. Work performed under EPA contract 68-02-2857.
94069. A A-11
-------
13. Johnson, T. R., and M. J. Symons. 1980. "Extreme Values of Weibull and
Lognormal Distributions Fitted to Ambient Air Quality Data." 73rd Air Pollution
Control Association Annual Meeting, Montreal, Quebec (June 1980).
14. Curran, T. C. 1984. "Data Screening for Large Air Quality Data Sets." 77th
Annual Meeting of the Air Pollution Control Association, San Francisco, California
(June 24-29, 1984).
15. Taylor, J. A., A. J. Jakeman, and R. W. Simpson. 1986. Modeling distributions of
air pollutant concentrations—I. Identification of statistical models. Atmos. Environ..
20(9): 1781-1789.
16. Jakeman, A. J., J. A. Taylor, and R. W. Simpson. 1986. Modeling distributions of
air pollutant concentrations—EL Estimation of one and two parameter statistical
distributions. Atmos. Environ.. 20(12):2435-2447.
17. Bencala, K. E., and J. H. Seinfeld. 1976. On frequency distributions of air pollutant
concentrations. Atmos. Environ.. 10:941-950.
18. Mitchell, R. L. 1968. Permanence of the lognormal distribution. J. Optical Society
of America. 58:1267-1272.
19. Larsen, L. C., and R. A. Bradley. 1991. "Use of an Exponential-Tail Model to
Estimate Ozone Concentrations with an Infrequent Recurrence Rate in California."
Air and Waste Management Association 84th Annual Meeting and Exhibition,
Vancouver, British Columbia (June 16-21, 1991).
20. CARB. 1992. "Proposed Amendments to the Criteria for Designating Areas of
California as Nonattainment, Attainment, or Unclassified for State Ambient Air
Quality Standards." California Air Resources Board, Sacramento, California.
21. Smith, R. L. 1989. Extreme value analysis of environmental time series: An
application to trend detection in ground-level ozone. Statist. Sciences. 4:367-393.
22. Gilbert, R. O. 1983. Statistical distributions for contaminant studies, and the
estimation of average concentrations. TRAN-STAT Statistics for Environmental
Studies. 25. Battelle Memorial Institute, Pacific Northwest Laboratory, Richland,
Washington.
23. Berman, S. M. 1964. Limit theorems for the maximum term in stationary
sequences. Annals of Mathematical Statistics. 35:502-516.
24. Curran, T. C., and N. H. Frank. 1975. "Assessing the Validity of the Lognormal
Model When Predicting Maximum Air Pollution Concentrations." 68th Air Pollution
Control Association Annual Meeting, Boston, Massachusetts.
94069.A A-12
-------
25. Larsen, L. C. 1991. "Evaluating the Performance of an Exponential-Tail Model for
Use in Determining Ozone Attainment Designations in California." Air and Waste
Management Association Tropospheric Ozone and the Environment n Specialty
Conference, Atlanta, Georgia.
26. Mage, D. T. 1984. Pseudo lognormal distributions. J. Air Pollut. Control Assoc..
31(4):374-376.
94069. A A-13
-------
A-14
-------
APPENDIX B
Summary of Articles Discussing Applications of Time Series Models
to Ozone Concentrations
94069.B B-l
-------
n
g
[Q
AJ
G
g
H
K
CJ
tn
|H
id
0)
1*
g
•H
01
tt)
K
rH
T3
0
g
m
tt)
-H
0)
co
0)
S
-H
U
a
0
u
.
u
X
ta
o
2
s
O • AJ
rH 0) G
rH D id
O-H U
i 01 •
G 13 id a) m
0 rl -O AJ G
•HO O
AJ tt) m -H
id tt) u c AJ
AJ -rl 3 rH rH
G ra "b a)
0) m d) id IH
U 0) tt) rl
G w o x: o
0 OiX AJ U
u a) tt) o o
r- ^ -° tl
6 0 IH 3
3 AJ 0 0 id
63 AJ
-H id 01 m
X c tnu
id tt) -H G C
6 r-H M -H Id
f*V (U frj rrj
>, 6 AJ In tt)
r-H -H 01 O 0)
-H m 3 u u
Id rH 0 X
Q id u id tt)
^H
£ *
H H
co
id"
01 -H
tt) ^x: a)
rH O tti-H
0) U rH H
Cn tn 0) -H
q -H -o >
*( u id ra
§rH -H
-H 3
o MX: o
J b 0. J
£ d)
0 rl
rH Id
rH
O D)
G •
0) >-l O 0) — .
0) -H M OH
0 IH M G ^"*
M id 53 U ri
•0
§
• O
O G
X 0
W U
H
c
3 '
6 6
-H 3
X 6 tt)
Id -H rH >
6 X id id
Oi id u 01
>i CJ 6 -H
rH -H Oi Ol
•H 6 tn 0 C
Id 3 - rH -H
T3 01 >, O rH
m id k a)
1) id -o o T)
x: tt) o
AJ m at AJ 6
C 3 tt)
rl 0 0 6rH
O -H -H Id
IH AJ !> Oi U
U tt) G -H
01 -H M -H M
jj T! P[ m rp
tn tt) 36
id H a) 3
U OifC 01 G *
tt) AJ rH 01
rl G tt) T) AJ
O id IH T) C ra
IHX! O 0 id id
AJ 6 u
CQ (D C CJ
0) fl) U G 0 rl
•rl 03 G O -H 0
)H )H tt) -H jj IH
0) 0 AJ tn id
to U in tn 6 W
-H tt) fi tt)
0) tt) 01 IH o AJ
g i-l M OlM-l AJ
-H 0) tt) tt) G tt)
H * ft tt -H S3
«£
£ <*
r-i H
6
v
•s
id
rH
&
•
0)
•H
l-i m
tt) ra
•0 H Id rH 4J tt)
lij D > -H U TJ
a id a) o
>, rH T3 0 0 rH N
c a) ra IH >,.c o
5 M IH 0 rH AJ
-H D< 3 H H G >i
5 0 0 3 0 rH •
2 4^ k AJ s* 3 6
fj AJ pi tt) 0 -H
e* 6 tt) * JTi X
m Ol tt) X! tt) Id
nCC'dAJcaie
g -H 0 C 0 X!
R 3 O AJ O rH
*• 0 rl -H
O rH G C rH IH "O
tt) 0 0 M rl
X! T) -H -H O 3 AJ tt)
-H g OJ Id X! Id AJ
? M rl rH U
a AJ AJ rH 6 -* I*
rH 0 0) 0) * g -H M-l
tt) -H U U -H C
T3 In G C AJ X DiAJ
o id o o -H id -H o
^s
*^
H H
3
0)
Oi
g
Oi
G
0
AJ
id
CQ
T3
rH
tt) Id rH
> C id G
-H O U O
iH 01 -H -H -H
0) 10 4J Oi AJ
•a D -H o id
IH M T3 rH k
O 01T3 0 AJ
tt) Id rl G
AJ rl 0 tt) •
ra o x! tt) u id
rl AJ AJ AJ G JJ
•H 3 -H a> o id
Cn Id * 6 U TJ
u
G
0
u
ro
rH O
W AJ
3
O ft
X! 3
•O E AJ
§ g §
•H U
tn X-H
>,id IH
id 6 -H
*>,&
d) rH -H
O-H m
J-S >,
rH
tt) 0) rH
tt) X! Id
U AJ U
X -H
0) rl AJ
O m
tt) •
O-H id
IH AJ 4) T)
id )H
tn r-H 0) (N
AJ 0) S
m IH IH
tt) rl AJ O
AJ O G
u id 01
m o T) id
G AJ-H rH
o! id o id
-£
«B
H "
d
id"
ra
3
1
1
•o
§ .
• U
U G
X 0
U U
01
ra G
3 -H
O AJ
, -H G
>i > 0) tt) ra
rH tt) M 01 Id
-H (H tt) Q) £
id a !s M
t3 a x^
0) id D AJ
a) x: AJ ri c
X! AJ_g 0
01 0 H tt) tt)
G id AJ x!
•H T3 U 4J
AJ 0) -H flj
tn a Cn H •
id id O m o AJ
U XI rH tt)^ G
0) o m T) id
^ AJ vi n> G u
0 c o o id-H
IH id 0) IH
•O AJ Si! -H
to -H 4) G, tt) C
O X 6 3 tt) Oi
IH O _ 6 Se -H
•O m
ra >, G G tt)
rH rH «J H £! |>,
tt) rl AJ rH
T3 3 0) rH
003 • AJ >i AJ
33 G id m
O 5 ra tt) -O -H
-H -H > m AJ
to X >, C id
-H 0 O
IH m -H -H
0) m AJ Oi
•O tt) -H 0
M IH 73 rH
0 O1T3 O
tt) id ri
AJ M O •
ra o x! tt) id
rl AJ AJ AJ AJ
•H 3 -H 0) Id
fe id ? 6 *o
CJ
G
0
u
in
-------
to
0)
u
co
fl
fl)
c
0
-H
Oi
0)
BJ
rH
0)
o
2
ta
0)
-H
0)
C/)
0)
g
-H
u
G
O
U
a
X
H
g
fl) tO tH
XI S 0)
fl) 4J 0 X)
0) X! rH CO g
CO U IH rH i-l 3
fl fl O 0 fl) G
fl) )H m 4J
W CO O 0) CO 0)
O co IH to to 3 x!
G > fl -rl rH U
-H -H G U X! U
01 O EH V4
TJ -H rH CO
rH CO 4J fl -rl IH
0 G 3 -n •
XI OX) U G in G
CO -rl -rl fl) O 3 0
fl) 4J Vl a-H V -rl •
Xi > CO 3030)
4J U -rt fl XI XI W
fl) T3 -H CO -H CO
0) CO > N fl) H 3
U XI -H tO 4J U4JrH
COrHCUtQCmU
It) ft U -rl fl -H
fl) O fl) fl 0) 0)
0) < "o G CD u a
U V4 1 0) O O-rl
XvfloimXrito
6 H X -H fl) U
fl)3Ofl)04>EC
4J IH 4J fl) TJ
fl) Oi 0 T) 01 0)
OiXI G GO) 0)
•H u o> o 3 X
S-rl -H 3 O fl) -H IH
* rH C U X) S O
1
1
>
0
IH
(rt
S
M
0)
•o
o
4J G
CO -H
IH n)
u
X
vo
H
rH 0)
4J
o
4J •« fl
CO
•Q 4J fl) k
fl) G XI O
-H d) 4J >H
rH g
a a) •* o N Q)
fl 3 -rl -O
CO CO CO -H
fl) fl 4J >
X) fl) O fl) O
g rH rH k
d cu a a
U 4J fl fl)
•H • m VH
CO rH O fl
4J fl U fl)
CO O 4J-H
fl) M H 03
0) H 0) -H fl
X! 0 H iJ
w o a u G
0 E 3 0)
fl) 4J (3 "D >
XI 3 X OJ-rl
£H fl fl) iH Ol
«"*
ss
J
H
^
0
01
fl
u
•rl
XI
u
D
* fl)
O CO
0) rH
k rH
fl O k •
IH 0) G
CO T3--H
0) 0 w fl
u u O x; •
d u fl
(3 T3 4J CO
13 fl) to ^
fl) g H O k
fl) 3 -rl J^ fl)
u ta m n ft
X to fl fl
W fl fl g Q
u
X
r-
4J « G
10 tO 0 4J t-~
0) fl) --H CO •
XI N fl) 4J D O
01 -H N fl XI
-H > CO -iH rH 01 G
XI C tQ 0) -rl fl
O 0) fl) in x; X!
3 ftrH rH O fl) 0
tO 4J to Oi
4J 0)
fl)>l D CD P CQ iH fl
XI !H T3 g fl fl
iJflGfl)fl) xicoco
>fl)HiJX:G-rl
M W -rl AJ g 0
mo IH -H fl -H
CO •rlOlCCOXrH.U
0 Q) fl -H CQgHX)
-rH.Ui-H.UCfl M -rl
UrHHE WX104J
3UOrHOG— '.U-rl
IH rH U IH -rl 3 "O •
fl 0 fl) A! fl C
GOUXIGfl)C 010
0 3iJOUflwG-ri
-H 0) fl -rl G M 0 -H 4J
4JH 0)4Jfl Cli U fl
3flfl)G3-HXl -Hg
XI > -H XI iH 4J g -rH
•rl CQ -rl T3 -rl fl -H • -rl X
i) 3 fl rH 4J 2 iH
torn oiutocu'OG tua
1
1
fl)
? 0)
O CO
0) rH
iH rH
fl O )H •
IH fl) G
tO T) -rl
fl) O H fl
U 4J 0 Si •
fl T3 4J CO
'O fl) to t>
fl) g k 0 rl
0) 3 -rl X fl)
u to IH H a
X to fl fl
w fl fl S ft
o
X
U)
CO
H • CO fl)
fl d u xi
g -rl rH 4J
H fl t> 3
O X! G to k
fl) 0) "O 0 IH m
jj ^ u > G A:
(UflHGODrH -fl)AJ
SHrH-'iSA; ftftf tQJC! CJ
fl 3 Oi 'O VH fl) S] fl) 4J fl)
U ^ fl) fl "O U V4
COrH fl)S (UCUU
flGU -^XlflflO
'O 0 fl X k rH JJ Ty XI U
C fl) fl) "~* tt) 4J
i3 co IH *O rt to tt)
r-fl) X!0 -HXSrH
ft tn 4J d)X)a)Ofl)
CO fl 0) 4JX!-rl X!4J
fl) 0) fl W OX! g
aa> wxl-rij-i W4J >I-H
fl tfl iJ 4J M-l fl ft TJX
aa>x wxswVH3o
Xlfl)Olt»iU>iO4Jk
itaoo33coa)
ETJ-HX) fl)tng ca
3fl)rH flflXltO-HWfl
co co -H tn co4Jfltaa>u
co co XI to 3 a
UX1 OrH 0) (U-rlrH ar7
a) to o o I-H a*o fl fl •"
X!-rlHrHOflOX:g
-------
co
jj
c
0)
CJ
CQ
rl
id
0)
G
o
-H
cn
CU
K
^
(1)
T3
2
to
0)
-rl
0)
co
0)
g
-H
U
c
8
.
y
x
u
f\
£
JJ en
G G
Vi 0) 0) 0) O
0 13 X! «*> T3 X! -rl
4-1 Q) EH G EH JJ
JJ >i O 0) U
G 3 r-n ma G
O XI Vl • 0) • 3
•rl -H -H >, CQ tJ Id 4-1
JJ rl Id JJ -H G g
Id JJ 4H -H -H -H T3
•H -H vi id 3 fl) id cu
XTJcDViGgggVi
0 "OOO-H3 JJx^
a<-i 3 jj (0 m rH 0) -^
arH Olldgld-rlgrO
id (d 13 C JJ id O co
g 0) -rl CO rH JJ 73 01 O\
0) VI 4-1 > i Id Id H
SOidOGSXiTIVi —
1-HGi-HrHOCJJCDO
Id ' 3 rH C 2 JJ 4H H
> Oi u id id to 3 id
O rH *O CD X) JJ
fl) rH (Ct 03 C *O ^ 'H O ^J
£ OCnlQJftiV-lfDQ)
0) G O JJ g JJ rl
M CO -H G CJ -rl CO VI VI
JJ4H-rlJJOCDJJ-rlOCD
xo -H -H a to -a o jj
cu idTJjJrTa) GW
g g G Id Q) >1-rl 0)
Ol 3 -H O rH CrH X)
•H -H 3 lH 0) X! Id -rl (d
JJXgrHVlJJJJtJ 0)
-H id (d o id -H jj j
-rl rHO)0-HO)C30)
rH d) ~rS C t-i^ 4^ S fl) CQ (U
r] (fl 0) 3 CQ 0 ^3 fl) 03
«I JJ T3 Olid OJrH-H VI-
VO
f^
H
.
"
..
-rl
3
0
•
JJ
CO
f—
H **
r-< 'O
6 fl)
3m -M
G ill
•H 'H rH S
X m CD rl
g > rl G G
>irH 000
>irH Id U G-H
rH VI g 0 JJ
-H 3 C JJ TJ id
« 0 O 3 C U
Q XI C id id CQ
u
c
0
o
e G
3 O CQ
£ -rl G
-H 4J O
X id -H
E 0) Oi rd CQ
H G H >i
>i VI -H 3 (fl
r-l 0 Vl T3 T3
-H U 3
id 0 'O 0) ^
T3 JJ U •
3 w G H
T3 < >, id
§id 13 O
t3 0) JJ
0)
S"O If) O i-H
•H 0) H X
3 fd MH O
0 rH 0 0) VI
Xi 3 014H
u m id
o id id cu cu
4H U H > -H
(rf M
co cu vi id
G Vi 0 !>
O Id 4-1 •
-H CO XI
jj Q) G x; a
id G cu jj a
rH 0 > C
0) M 0) 0 O
rl 0 E CN
O >i 01 Vi ^
U rH -H 0) 0)
o nx: g >
JJ 3 §O
3 0 to 3 X)
ri! xl -H to nJ
' CO
f^ C^"
H H
J
H
o"
Ol
id
u
•H
fj
u
1
1
•0
§ ,
• u
O G
X 0
w o
H
rH
G 0
O 4H
CQ 0) C-H 0)
(d g -H o) JJ JJ >i
4J 0)'O4JrH g grH
C "O 0) 3 0)-H id
-HGOJldO-HJJCQU
JJ-HVlrH CQXCUO)
u^jJ3ta o) 1-1 a
CGOJEGO) coco
4H 1-3 to -H EH H 0) 0) T3 4J
jj rt| 4J jj 3 3 0) id
0) m rH i-H T3 i-H
U T3Vi-HididG3
•H •0)0)T3>>COg
Xj T) JJ S -U -H
3 0) 4J 0) rH 0) tO
CJ r* *rl CO Xi Id CO G
04-lO)4J30)OO)
CD g -H G X! -H X! •
X! CDCOVlCDGJJJJJJtO
jj Vi (d 0) JJ id id A;
£c co 3 "O g G C
id'HrH CD i'mSxxi ri
Q)(t)d)«CO(l)(dO^'
T3 0 Ol-H a CD 0)
x!o)g o -H jj a vi xi
Id Id CQ CQ CD g -H
CDgCDCDVlC'CD X!
•H -H -H CD tt) Vl 3 G
OtQtt)JJ
<^^
** ^
o 2
rH "
.
s
X
id
CQ
3
5
0)
Id - rl
CO 73 Id
•H en G
id 0) to T3
g Xi Vl Vl 0)
3 jj o jj
e ^ M id
-H i-H 0) Vl i-H
X Id g CD 0)
id g -H vi
g VI JJ 0) VI
0 X! 0
>i G O JJ U
rH I -rl 0
-rl OlX) *O JJ
nl 0 3 5 3
Q rH U id Id
CJ
G
0
u
cs
H
•
CO
0 Di
jj id
rH
•o
G 01
(d CD G
JJ JJ -H
rl Id CQ
o "C to id
r-l 4H CO)
(1) rH 0 VI
X) to id -H u
g CD g JJ C
O rH 0 rH
OiG CDXi
o a vi jj
-rl Id >< VI -H
JJ VI 0 *
o id id u
JJ E C O JJ
a-H o JJ CQ
& X -H 3 (d
>i ID JJ id 4-1
co g id
VI 10 X! rH
0) O JJ JJ
JJ CD JJ 0)
G JJ id -H
jj o id x; cj
Id -H i-H JJ -H
X! 4J 0) 4-1
XI vi o> 3
CO -H O T3 CQ
CD VI CJ-H
> 4J 0 > 0
0 CO JJ O VI
Vl -rl 3 Vl 0)
& TJ id a N
1
1
1
t
r—|
id
g
£4
o
c
CT
id
c
0
-rl
JJ
id
JJ
CO
u
c
0
u
CO
H
01
0) CO G CO)
xl CD o id H x!
JJ 3 rH — JJ
rH 0)
CQ Id (0 > CD
G !> id • to
0 Vl 4H X! C id
-H 4J CD 0 0 U
4J CQ > i-H -H
-H 0) O rl rH JJ JJ
T3 x! 0) -H 3 C
G Ol CO .Q S X) CD
O -H CD g -rl "O
CJ Xl O 3 TJ Vl G
G G r-l JJ CD
rH (D dj 0 CQ Of
id x: ti CD x! -H jj to JJX! vi •
3 Vi Vi T> O —
vi xi CD id id G G
CD -H .Q Qi 3 JJ O
•O Vi g , i CD
VI Ol CJ
id G G
G 0 CD
O rH T)
-H G
JJ M 0)
id id a
JJ 0) 0)
CO S "O
•0
§ .
• u
U G
X O
w u
*
H
-------
S
QD
to
4J
G
0)
Q
U
CO
VJ
id
0)
*
S
-rl
0)
IV
CX
i-l
0)
•3
0
m
fl)
-H
rl
fl)
CO
0)
g
-rl
H
U
c
0
U
.
u
X
M
d
0 G
4J 0
-H •
13 4-) XI "4-1
Q) Id 4-> O
EGG
3 -H o id d)
tn xj g to >i
ca S in o r-H
rd 0 4-> id ft-ri
U (d X rl (d
to xl 3 t3
id id 01
SO) HTJ G C
Id G O -H -rl
rrj -H IH VH id co
rH 0) g 0)
>i en a -H
§id cu a) E
r-H CO XI
tnxi a> E-« G
rl -rt f3 U -rl
0 -rl G
IH 4J ri id • H
Id Id T3 G rH
fl) X! > fl> 0 -H
4J 4J fl) -rl «H
rl g "id X 3 O
£ U 0) X! 4J
fl) 4J -rt -rl
U -H OT4H lH CO
G rl 0 O JJ -H •
id id r-H ca tn
*O 0) 0 rl -rl "O fl)
0) 0 rl tt) T3 0 O
d) r-H 0 XI X! G
u cu § G 4-1 id
X Id JJ 3 O fl> "O
d) a> c en E fl)
d) g ca ai
fl) > 0) -H tt) U
xi id 0)
H3J
CO °°
CP\ ~J
H
CO
J4
0)
£-t
w
co
ta ta
3 tn
0 0)
fl) U
G 0
0) M
0) a
o
0 O
£ ca
< tn
C -H
gs
.
u
H
in
H
Vl (H
g CD 0) E
3 rH 'O T3 T3 CO 3
gel) tt) in 0 - E
-rl Ti 4-) 0 E >1-rl
X 0 id id X
g fl) to id m E
Ol rl ri T3 0) CO
>lG Id rl -rl 1 Xl 3 S
H -H 0 MH 0 0 rH
-rl 4-> "O U 4J 0) -H -H
•O -rl X! J3 4J >nt-l 0) *D
M-l 4J 3 fl) Id rl
rl 4J Id rl -O ftJJ
0 4J M-I c 4-lO
O.Grl ttOMH-HG
EHidid B ca-ricu
O) -O r-i U id x!
01 0 CD C rl fl) 0) D14J
-rl tJ R rl 4J O T3
4-)fl)fl)-H4J4JgrOCG
m 4-1 xl E a) COS
(fl 4J 4J fl)CQfl)id-r<
U -H G g > tn E
fl)<4H4->0-rl -H >CQ3 •
ri id-H4J -tnmcDEa)
0 0) XI 4J gCOCUrl-Hrl
Q) i^ -H ID V-I c gj nj jj
CQ ^ t3 4JJJXJ t71d)W £ (d
2 0) C (d OJ M J-(
•H C 9 O 'O 0 0 M^ rH CU
riOtnGidri4JKHca-Hg
i>oS8&ei5rd;o:j'32
00 J2
r>
CT\ ~j
H "
U
ca
Q)
«
^
V4
0)
ca
id
rl
tLi
i
0)
>
-H •
ca tn
CO rH
0) 0)
rl 13
tn 01 0
3 0) g
0 rl
•H O 0)
M 4-> ft
, id JJ
U
§
u
\D
rt
C fl)
TJ G CU 4J l-H rl
ftrH td C rH 0
C a) o - 0) fl) -HOrlrlJJfO'
rl CO fl) rl M-l r-H MH fl) (d Id *H JJ
fl) rH (1) O 3= 3.4J rl
4JQ)Q<^ Ui-l CTtaCBrH
G g g7 Oi O fl) ca co CQ id
-H -H S3 m G > 4J M cu 3
4JXC-H Gfl)JJX)rOco
T3 COOCC-rljJCO G3
rdMrlJJQj "^ g 0) 4J 0)
OCr-Hfl) Q)fl)rl T3JJJJ-
JJO3>fl)O)gldl-lG-rl JJ
-rlUrlX!CEft(d3rHtnU
CQ 4J "H 0) 4-J fd ^ Q)On3Q)Q)
•HtdJJjJ xima)G'4-iC4JM-i
JJrlGOU g-H 0*^
O Qt T) VD 1 ^1 Id -rl
XI id fl)fl)Gcn C 0) 0) r-H G
JJ 0) XI 3 id H • 0 ^ co id 0
EJJJJ SGcutn GJJ
r-Htn^iGvo^ GX10JJ cu
idcuOOcnaicord -H c cu >
Vi-Hti)-HHtl)D(j3T)JJ(oyrl
0)rl JJ MEOCDGUCCU
GO) ldGt-13 4Jfl)-Hfl)4J
DM -U-rl -nGfd>»WrOC
01 rH -rl fl) OI E IH -H C -rl
fl) fl) r-H a JJ Qj-H -rl fl) G 0)
fl) g T3 Pi g (d Q) CQ JJ JJ Ol Pi fl)
H JJ E id -n COCQ*OCU-HCnt34J
1 t
in CN
in o
C^ CTl
rH i-H
ca
cu
i-H
fl)
tn
c
(^
ca
0
^
01 C CQ r-l
' G 0 -rl XI
o -H m -H cu JJ
JJ r* CQ 4J X* r*i G K*I
3 0 fl) G E"* rl O rH
Id g U fl) O g rl
0 > fl) 3
•a a) wi ri • js o o
fl) K* Oi fl) CO JJ JJ X!
JJ -rl JJ fl)
idcocuGr-ii-i'oa)
rl CO Ol-rl XI Id fl) Ol •
01 0) It) Id rl -rl Id 0)
CU rl ^H X! *H Q) I— 1 Vl C
JJ 01 fl) JJ ^H G pi. QJ Q
C fl) r^ *rl (d tt) Ql p^ N
0
s
u
£3
-------
CO
4J
c
|
u
ca
rl
id
(U
*
c
0
-rt
Oi
CO
ttl
rH
CU
• -0
o
g
m
0)
-H
M
CO
CO
II)
g
-H
EH
U
8
u
.
g
u
0
z
cu
rj
4J
4-1
0
id
T3 CO
G g
id ft cu
0 (£<
T3 U-l EH
J_| ^|
id cu
f n o, •
0)
-rl
CU
rl
id
CO
-rl
rl
cu
ft
C T)
id w cu
4J Id 4J
mA id
ij T
4J rH
c-o m
CU rl >
CD ftf 0)
CU T3
rl C Id
a Id-rl
4J rl
CU CO CO
£ 4J
&4J *-H
CO
4J
0
cu
g
0)
.a
4J
1
?
a\
H
r-
ts
6
T3
C
0
»C
o
-rl
0)
U
§
•o
cu
cu
o
X
cu
CO
0
H
O
•o
3
,
g
M
(U rl
m a o
o
id cu
ca A
CO Vj 4J
cu o
rl |^ J^
CU O
K* rH MH
-H id
4-1 CO 1-1
U 0 CO
CU ftJJ
4-104-1
«H rl CO
CO ft XI
CO
CO
•
CO
cu
4J
•H
m
r-H
id
G
0
-rl
4J
8
CU
>
id ca
* §
CO -H
4J 4J
§3
C »C
0 0-H
U CO rl
m 4-i
>1-H CQ
& 0-H
"0 cu *c
o
g
u
o
4->
a
-H
CO
C
0
•H
4-1
id
H
0)
rl
J3
4J
• *
CU
cu
•a
ca
3
id
-rl
rl
CU
-H
rl
U
g
1
-H
^
(^
g
>•
rH
•r
id
0
rl
-H
(d
0
4J
-rl
^
-H
4-1
-H
m
c
0)
m
«
^
4-1
-rl
U
•H
r-H
i1
-rl
CO
^
r|
4J
rH
id
CU
A
CO
-H
CU
c
0
N
O
>
,rH
rl
£
A
4J
c
CO
g
-H
id
4-1
4->
id
4-1
O
s^
4-1
-H
H
•rl
•9
4-1
ca
^
CO
0)
01
§
A
CJ
E**"i
p
-rl
rH
i
rl
CU
73
rl
0
1
4-1
m
rl
-H
•4-1
rH
id
C
-rl
C
T3
TJ
CU
m
0
8-
rl
0
4-1
S,
rH
4J
C
CO
rl
§rl CO
ft 4-1
«
id
4->
id
d)
IP
id
rH
-H
id
>
U-t
o
cu
CO
3
^
CO
3
4J
id
4J
ca
•
>
-rl
CO
CO
cu
rl
cu
£H
s^
g
0)
01
c
-rl
rl
4->
CO
rH
id
4J
U
id
ca
3
1 vers
-H
•o
m
C
0
•H
id
rH
O
-rl
^
m
cu
C
•H
4-1
CO
T>
,rj
U
id
0
CO
£-1
•
m
C
0
-H
O)
0)
rl
4-1
G
CU
2
•H
id
4J
id
C
o
c
*o
c
id
4J
C
cu
g
2
-H
rl Id
ft4J
T)
CO
4J
id
rH
3
g
-H
CO
I
rH
id
cu
c
0
•o
CO
m
0
4J
CO
id
A
L)
id
o
rl
s
id
4-1
id
T3
Oi
CU
rl
0
3
H
-------
APPENDIX C
Example Time Series Model Simulation Results
94069.C
-------
TEST SIMULATION FOR AR(1)/LOWESS FIT WITH LOGNORMAL SHOCK
Simulations: 250
HOUSTON CMSA JAN - DEC Season days: 366
Site: HOU1034 48 201 1034 0 Year: 1988
MMMMMMMMMMMMMMMMMMMMMMMMMMMMM^
2nd 1st Tail-Expo
Yr AVG 50% 65% 75% 85% 90% 95% 97% High High 5% 10%
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
0
0
0
0
0
0
0
0
0
"0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
.052
.055
.052
.054
.052
.054
.049
.052
.053
.051
.049
.057
.052
.055
.049
.050
.048
.054
.053
.050
.057
.054
.05-.
.052
.051
.057
.050
.050
.051
.05-i
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
r\
\j
0
0
0
0
0
0
0
0
0
0
0
0
.043
.045
.044
.047
.047
.047
.044
.044
.044
.045
.042
.048
.047
.047
.043
.044
.042
.C46
.042
.046
.049
.047
.046
.045
.044
.050
.046
. 044
. 045
. 047
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0 .
0.
0.
0.
053
060
057
057
058
056
055
055
056
057
052
062
058
059
055
054
054
055
056
056
059
062
062
056
056
063
055
055
055
058
0.067
0.071
0.063
0.068
0.066
0.066
0.062
0.071
0.063
0.067
0.063
0.073
0.066
0.071
0.064
0.063
0.061
0.067
0.068
0.063
0.075
0.070
0.072
0.065
0.066
0.072
0.063
0.062
0.064
0.070
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
.084
.085
.077
.084
.079
.082
.076
.085
.081
.082
.082
.092
.077
.085
.077
.077
.075
.084
.087
.077
.095
.084
.085
.083
.080
.089
.072
.075
.078
.082
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
.097
.100
.091
.099
.086
.097
.088
.092
.097
.095
.092
.108
.088
.099
.086
.084
.081
.104
. 101
.086
. 108
.093
.095
.097
.091
.098
.090
.084
.089
.091
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
o.
0.
0.
124
123
113
119
106
122
112
112
118
108
112
125
110
120
104
094
101
126
128
100
123
115
120
122
111
121
115
105
110
116
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
140
151
135
126
121
137
132
114
143
119
135
135
120
140
114
106
112
143
137
113
140
133
128
132
120
129
130
117
124
129
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
197
212
219
163
150
201
164
173
189
150
163
204
177
202
148
179
212
230
:i5
151
223
224
196
162
182
202
168
212
189
185
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
207
219
240
209
250
237
180
176
215
152
172
232
205
207
157
209
243
236
242
168
230
252
245
165
262
287
203
234
272
190
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
.212
.230
.237
.175
.180
.206
.189
.163
.214
.160
.191
.194
.183
.219
.150
.180
.194
.236
. 194
. 172
.224
.204
. 182
.173
. 175
.221
.180
. 190
.195
.195
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0-
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
21
22
22.
18.
18-
20.
19.
17.
20!
15 (
19:
19(
is:
20:
16:
161
is:
22J
211
167
206
201
19:
19f
183
211
192
187
191
203
&™m^2^™™™™MMM^
0.052 0.045 0.057 0.067 0.082 0.093 0.113 0.127 0.183 0.215 0.191 0.190
^'??vl?r''I/?r-0^^^^2vC^lMZ^7^^
SD 0.002 0.002 0.003 0.003 0.00-^ 0.005 0.008 0.010 0.025 0.042 0.022 0.019
/*^-friM™^™:?:7*™iM7^^
-------
APPENDIX D
Ozone Design Value Comparisons
94069.D
D-l
-------
a
0!
C
0
o
t%
u
.X
3
b
at
9
o
a?
o
j:
U
fe
C
8
fr
R
3
5
.1
4)
O
T
U
CO
-S §
">;§[£
Ju
3 v. 1 -0
c .§• ^ ,2
a «r>
>< ON
°nj w g ^j
SS Q vs> °
O i-J =*• 03
ON
1 * !•§
c &.U §
U C- o
rj »— . KO _x
i. "~>
X ON
w
'« «- g T3
* *a|
>3 ^ DO
"^ t^
O
8O §
— "* "
en to
a.
1
"2 •§)
3C
g
b
*_j
Cr^ «»* CW
2 H W
y^ "3
a.
0 P x-^
H^!
K
B
3
U
S
W5
OO
S
o
^^
0
oo
S
o
NC
d
00
g
o
6
8
o
i
o
S
o
8
d
|
o
§
S
-1
I S
O 0
r» oo
r» o
O O
OO fS
2 S
o o
O vn
en o
O 0
00 ON
O 00
O O
— 00
0 0
S §
o o
O 00
*— ( O
o o
S $
O 0
~ g
0 O
»-> oo
O 0
i i
S S
J -J
2 2
o o
ON en
O O
8 8
o o
2 S
0 O
o\ o
O 0
8 S
0 O
SC**1
0
o o
r~ Q
g 2
o o
OO en
O —
O O
S g
o o
8 S
o o
O m
i §
o o
J _>
oo r-
O ON
0 O
«N —
O 0
f- 00
2 g
0 O
en en
CS "•
0 C
2 g
O 0
O 0
0 O
rs o
r) cs
O 0
•* en
2 g
O 0
2 S
O 0
cs en
O O
^^ ^D
o o
1 1
en en
O 0
J -J
o rs
o o
"" 2
0 O
o -<
0 0
00 O
— en
0 O
1 E
O 0
V*> ON
O O
0 O
00 1/1
O —
o o
1— \D
O —
O 0
00 ON
O C-J
0 O
r"" 25
o o
V> ON
0 0
0 0
en cs
S S
J J
S 8
o o
rs —
o o
c^ ts
o o
0 O
CS (S
0 O
2 2
o o
t~- en
o o
2 2
0 0
8 5
o o
2 g
o o
— ts
o o
r~ en
o —
0 O
i i
Tfr 00
0 0
J J
00
o
o
fS
o
8
o
S
o
r-
o
o
(S
o
12
o
S
o
2
o
vO
o
fS
o
§
g
J
NO en
(S O
0 0
\r> m
"* —
0 0
S £
0 O
00 —
TJ- m
0 O
§ 2
o o
25 2:
O 0
NO m
o o
rs ON
o o
en O
O O
en -*
O O
25 S
O O
cs —
ON ON
0 0
J J
S S
o o
— rs
0 0
8 |
0 0
en en
rs rs
O O
2 2
0 0
^D "^
o o
— 2
o o
ON «
g 2
o o
0 =
o o
o —
o o
vo en
0 0
O O
i i
— ON
J J
"S
2 I
d §
U
o
en
o
en
O
rs
rs
O
S
o
rs
o
en
O
ON
O
O
S
O
en
O
CJ
o
00
00
o
z
Q
-------
O
P.
I
r-
^
V)
S
CD
-5 §
fls
S .0 £
14.:
s g-g
o, u 5
So
«— -. KD -it
v>
ON
ill
J # ca
ON
v. 1 -0
O. _, o
£} 6S CQ
IT)
ON
fco'H
J3 * CO
ON
V) —
•* o
w
en .2P
BC
c
J
„ .
O t* *"
^H C* W
a? vis °-
SJ £ x
m H a
D_
ijf
CO
B
!
O
55
es
0
-
o
8
o
2
o
OS
O
es
es
o
en
es
o
8
0
ON
ts
O
^
£J
o
ts
o
8
ts
z
8
o
tn
o
en
0
oo
O
s
0
£
o
en
— •
O
o
0
a
o
^
"•"
o
2
0
Tt
r-
es
£
en
o
3
o
s
0
in
0
8
o
8
o
^
en
O
ON
O
3
o
^
2
0
tN
tn
o
Tt
ON
O
£
en
0
8
o
?!
o
12
o
r-
es
o
00
es
o
ON
O
tn
o
3
o
CO
o
00
0
i
ON
8
^
fc
o
00
o
en
o
ON
O
VO
O
en
•t
O
VD
O
8
o
00
VO
o
ON
O
9
0
ON
8
ON
V)
O
$
8
o
3
o
Ov
0
VC
O
a
o
ts
o
en
es
o
£
o
o
o
t
ts
o
8
0
i
i
£
Tf
o
0
o
ON
tn
o
VC
O
en
O
VO
en
O
ON
O
00
o
5!
o
NO
2
o
VO
o
i
o
%
£
o
-
c
i
o
VO
o
oo
o
8
o
VO
0
§
o
en
O
^_^
JQ
O
&
o
1
ON
O
^
en
0
•t
o
ts
o
en
O
r-
o
t-
ts
o
tn
en
O
00
O
en
en
O
^
!-J
o
ts
o
i
ON
£
00
o
00
o
V)
0
en
oc
o
5?
o
ts
o
V)
V~t
o
K
o
VO
o
t
VO
o
V)
0
oo
8
ON
O
%
tn
o
ON
O
ts
en
O
r-
o
tn
o
9
o
•*
o
en
o
3
o
00
2
o
9
o
1
ON
0
^
•<*
O
es
o
§
o
VO
O
tn
o
VO
en
O
^
VD
O
ON
en
O
Tfr
o
^
2
o
vO
en
O
f-
8
o
ge
00
o
VO
tn
o
VO
o
r-
en
o
2
o
ts
o
en
es
O
en
o
TJ-
ts
O
tn
ts
o
es
0
i
ts
£
00
o
tn
o
VI
o
VO
en
0
ts
o
ts
o
V)
es
0
tn
o
V)
ts
O
w
ts
o
ON
0
8
en
en
£
Q
-------
13
1?
c
a
X
w
S
O
•a
1
c
a
X
W
8 iS
« £§
•s "•>
.3
g
c
c
o
U
i
r
"— ' >
^
u
Z
i
0
•£
u
Z
£
[0
C
o
.52
(3
E
o
o
u
•a
M
•O
0
N
O
t
w
J
cc
o n
jj .2 i_
<= 2 £
J2 js t£
J3 u
«-^
c
O- a
D ^ oa
m
Sr9 "§
> ^ S
in
& °-§
D. §
rj BV QQ
•n
ON
fe §•§
£ U §
,3 ^ 03
m
ON
in JJ
* £ '5
8 S g
^"* ^) U*
r^ 1>
^ 0-
•*^
*o 22
CO .M
E
c
^
^ rs fj.
2 H BJ
*S -a O<
«"H £
&
jj D ,->
ejs
35
^
c
o
U
u
1
^^
r-
o
00
0
vo
o
00
DC
o
in
0
o
0
^^
VO
0
VO
0
t^.
J2;
o
^
VO
0
5?
0
f-
8
^J
8
u
in co
O O
in oo
oo t-»
O 0
s? *
O 0
co O
00 OO
O O
CO CO
O 0
ON VO
O 0
VO ON
•n **
O 0
oo f-«
O O
** ON
vo in
0 0
ON oo
in in
O O
ON vo
O 0
r^i c^
0 ^
— «
8 8
H H
U U
•n
r-
o
s
0
^
O
$
o
£
o
$
o
•n
o
s
o
in
oo
o
in
•*•
O
VO
0
8
^_
8
u
es
0
p:
o
£
o
(S
0
(S
o
(N
O
(N
0
VO
CO
O
co
•n
o
c^
2
o
ON
O
|
CO
8
z
00
r-4
0
£
0
a
o
5
o
E
o
CN
O
m
CM
o
tx
o
f^
CO
O
r-
o
PS
o
8
CO
O
Z
VO ON
in ^t
0 0
vO t-
00 VO
0 0
CO —
O O
oo r-
vc *n
O O
Tt CO
O O
VD CO
O O
r^ CN
vo co
O O
ON CN
O O
r^ CN
vo in
0 0
fN CO
in rj-
O O
VD fO
O O
§ |
t^ ON
O O
z 2
CN —
in vo
O 0
ON ON
r~ oo
0 0
r- vo
0 0
in r-
vc oc
O 0
co in
O 0
CO —
O 0
ON —
Tt in
O 0
$ In
O 0
— o
vo r-
O 0
oo in
Tj1 VO
o o
fN ON
Tf CO
O O
vo in
co in
(N fN
O 0
Z z
r- TJ-
o o
ON in
VO <*3"
0 0
co fN
0 0
VO O
in Tt"
0 0
co r-
co •— i
0 0
t~~ O
co fN
0 0
ON Tfr
co fN
0 O
— ON
0 0
Tf fN
V~> CO
O O
fN vo
O O
r~ o
CO fN
O O
8^?
^^
co in
r~ o>
fN CO
O O
Z Z
m
0
&
0
ON
0
co
0
s
o
8
o
vO
O
0
—
0
-------
c
'I
n
>_
<
y— S
W
ft)
^n
u
00
1
(A
,3
Ni^^
-fc*
%
3
$
s
&
CO
1
1
1,
o
o
V
s
T3
>
.1
'8
•8
o
O
•f*.
D
U
J
n
<
H
-2 e
^ '^ n
§||
|3fc
*c3 C rj
•j2 it o ^
§ &u §
Si O- o
I »g*
X O>
ui
'a v. If -o
&? ^
^5 fc^M *^ CO
" o!
^
"c3 u £ T3
"c D. U §
ft) O* o O
1 D^tt
u
1 6 1 "g
H > u §
""* IT)
* «3
8 en C
^ 8
" ^ (£
„
2
*P f-j
en SP
BC
§
U
J
S ^ S"
2 H w
* ^ &
>o r* BJ
cx
»S ^7*
•^ M "S
H J ^
ti
*»
&o
ft)
55
o
en
O
Os
en
O
en
o
en
O
Os
ts
o
o
en
O
O
O
00
O
en
O
ts
en
o
00
ts
o
o
s
^.
U
ts
ts
o
00
ts
o
es
o
oc
ts
o
ts
o
fN
O
o
ts
o
,_
tN
O
X
o
ts
0
o
c^
6
Os
®
,
U
ts — intsinvovnts — r-~tst-^oo
— tststs — eststsfs — tststs
ooooooooooooo
oooets — o^rts — oenoot-vo
— tsenentsenenenentstsenen
OOOOOOOOOOOOO
entsmtnvot^vo^'»~|ooenosOs
"-tststs — tstststN — fstses
ooooooooooooo
tnoc — oots»nenOO'a-Ooot-~
fStSentstStnenenentsenenen
OOOOOOOOOOOOO
^^ C1^ ^^ ^^ ^O \Q ^O C^l ^3 t^" C^ 00 ^*i
•— tststs— •tsfststs-Htststs
ooooooooooooo
•
-------
APPENDIX E
Ozone Air Quality Update, 1990-92
94069.E
E-l
-------
m
01
ot •
!OOVOOVeOOinOVDOOiHOOOm'9'OOOOOOOOOOOOOf\)CNOOc*)r»)coofl'r*)mt~t~
H .OWWOOliHOCOiHOlUlt-iHiHOO'
C ,, O iH r-l Ifl tH
Ss H
lrHm«<\Or O
gi^r^OCDOOO^rnOOODC^CCVDO^lflOJO^fOHr^OO^^U^Mf^fHOr^OJ^VDr^CN^rH^'CJO^
Oaro^OO^^tNl/ir^OiniflrnHOrH^^rHiHOO^MOl^OOrnf^fOVU^rniHr^O^CSVVrHrnrH
^^^f^rtrHr^^H^C^»MiHf^HiH»Hr^rHrHrH^rSOrHrHH»Mr^rHHrHfHr^^rHiHTH
iJoooooooooooooooooooooooooooooooooooooooooooo
CO
O)
0)
(D
T)
a
*>§
3°
"• JJ
n *
*«W
C"B
0 n
H ,_)
i-H ft)
(B JJ
C ID
01 o)
ID 0
ft)
01 JJ
E (B
01 ^
h 0)
JJ T3
X 0
to
3
0
•H
M
01
in
jj
0) ID
k >j
01 0)
> T3
0) O
CO JJ
3 IB
0 fc
.-( 0)
H T3
0> 0
r- in
^
01 0>
n M
o> o>
> >
01 01
r-4 (f)
r~^ u^
to m ID JJ
3 3 C ID
0 O-H h
•H -H cr> o)
Vj W W T)
QJ 0) IB C
IB E
C 3
•H 0
tJl-H
U Kv
ID D
r-
0)
lj
01
>
0
f^ TO
c c
-H -H
tn en
^ k-J
IB ro
n ID ID
c c c
•H -H -H
Oi cn tri
^ ti ^
IB IB IB
JJ
c
•H h -H -H 0 -H 0 0 >-l -H -H ^ KJ ^i ^1 l-l l-l -H U M
01 0) 01 CVH Ol-H •HOIOIOlOlOlOlOJOlOIIBOJO)
tiT) MT) VJ h U n kJ M > l-i U'O'D'DTJ'O'OJJTS'O
.,__________..-.._--- ---OIBO(BIBO)IB010)0)IBIDOOOOOOJJOO
WSWWSWSWWWWEESWWEESSSSSESSWZWMWESSSSSEEieCSS
(6
a
(D
0)
I
I
(0
o
a
o
N
O
CO
IB
CQ
CO Z
IB <
OTJ
U 0)
z
E
n!
tnx
c -H
•H c
E u
M o
•H .e
CD Ou
JJ -H
3 C
OD
CO
0 IB
01 ffi
i—l
0) >i
010)
5 w
< 0)
jj
to c
o o
(I)
(B CJ
D IJ
l-l «£
<
rt
< Z
Z
ss
« >, ID
01 (0 rH
V4 CO r-l
< I (0
O >
< O
z o c
-H -H
0 O 3
pi c tr
4) ID (0
•H l-i 0
D It, 10
u
o
aa
E £
O O
•J <
<
rt! "D
-rt V
tj -H
(0 «J
S-H
Ti
10 0
jj s: ID
c D
RJ 4J ^4
W M <
I 0)
ID O «£
VJ 01 Z
<« a
ja o
^ JJ U
ID m
CD ID ID
4) U
IB JC 3
f
V
IB
ft)
CO
IB
0)
I-l
IB
ft)
C C C C 7 C
ID ID ID ID O U
CO C/1 M W W >
ID
0>
ti
<
i
3 ID
U 01 ID
•H H 0)
JJ < M
U <
D <
C Z <
C Z
0 C
U O O
JJ U
M tn
oi c x
JJ -H D
ID £ O
D (D 10
k-i IB 3
ID
a,
c
3
O (B
U 01 IB
M 01
ro < 1-.
U
i i
01
ID
D >•
JJ
C
O
u
XI 01
to x
i-i ID IB
ft) 0) J
JJ J-i i
D < >.
CX U
< IB
• Z O
W IB 0
i JJ 01
ID C IB
a ID u
E ^H-rt
HrfU
ID
0)
IB
0) IB
l-l 01
ID Z
01
>-i JJ
0 C
a oi
ID CO
c
IB £
•ri JJ
T3 3
C 0
l-l Cfl
z<
IB Z
0)T> IB
U C 01 0)
< IB JJ Wi
i-l JJ <
< £ 0>
Z D) ><
< ID Z
0 i h.
U C i Ol
O C-H
C JJ OrH
o tn jj -H
to c 01 >
c -H c m
O JJ -H -H
E C X 3
•C 3 0) 0
W I J J
ID ^l
0) <
M ID
< 41 <
«$*
Z 0)
0 Z
0 £ B
XI IB
OJ U C
C 3 O
01 TJ JJ
? ID ID
o eu m
0) <
JJ Z
CO
01 —
O CO
^ CO
o 10
S £ ID
0) • VJ
U 2 <
0) <
ID 0) W
iJ-H M
i IM 0
c me
O C-H
JJ -H JJ
10 ^ i-l
o a ID
n 01 m
c z
c
<< o
u
c
01 O
VO
2 r-t
O IB
S
•o
C-O
IB C
ID
>-
JJ 0
C U
o A:
u u
0
JJ O
ce
0) IB
X S
IS
5^
eg
l-l
0 C
U ^
C 3
•H JQ
ID
0)
ID
01
C i Z
10
c-o
0 0 C
U JJ ID
Br-l
X -H jj
O S U
C 0) 0
X J CU
ID <
< ft) Z
Z VJ ID
< 0) IB
O < ID < ft) C
JQ Z 0) ^0
< m < Z to
•D < IB
C-H < >,Z O
c az jj '
< IB -H B O
i K C U
JJ O
•H to tJi to
O T3 0> ID
t-i C J«. to
jj IB to C
3 JJ
0 0
0) ^ 3 ID JJ J3
a o 2 * w u
'<
O
(a
-------
•o •
0 °
*
O^IOrHOOOOOOOOOOOC.OOOOOOOOOOfNOOOOfliOOOi-IOmiHVOOiH
M
ooooooooooooooooooooooooooooooooooooooooooo
CO
* 9
«*
(4 OOOfSiHOOOOOVOHO
5u
HTHOOOOOOOOOOOOOOOOVOOOH(NCSr>VOOHiH
ooooooooooooooooooooooooooooooooooooooooooo
O)
O)
0)
4-1
(0
•o
Q.
Lr Act
cation
Clean Ai
Claasifi
jj
c
Attainme
jj
c
Attainme
Margina]
Serious
Moderate
Margina]
Margina]
Margina]
Margina]
Margina]
r-
Severe 1
Margina]
Margina]
Moderate
Moderate
Margina]
Moderate
Moderate
Margina]
ui c C
re IBrtCC3CCCro
-H -t-t -H -H -H -H -H ^t ±4 ^ -H -H O -H -H -H ^j
010^01010101010) 0) V 01 01-H IB re 01 4)
reitiB(Bi8iB(B4)OOrere4)jjjjreo
ESSZESrwiEESEc/XKtEs:
Serious
Moderate
Serious
t-
Severe 1
Moderate
Margina]
Moderate
re
d
v>
to
0)
0)
c
1
(D
o
Q>
O
N
O
•
§
o
0,
J2
01
IB
41
<
IB
< 0>
<
IB <
X Z
t-l
D 0)
i JJ
J= CO
01 4)
•H f
01 U
PH C
IB IB
K S.
01
0)
jc re
u o)
o u
a <
oi z
0 >.
D JJ
>.Z
O
3 U
0-H
E JJ
to c
JJ IB 0
>-i .-I C
O JJ O)
a < a
>. IB
T3 U-
IB IB
JJ IB 01
v a <
C 01
41 IB <
•C *•"' Z
o z
to i o
•D
C
re
01
o
0)
IB
0) -H
< tn
rf-H
10
01
TO V-i
4) IB <
< 0)
M 3 r-l (B
IB
U
re
C 01
c
0) <
C0)
£2
c
o ro ro E ^
C 0) 4) rH 4)
re >-i M -H >
2
4)
IB
CQ
a
IB
14-1 0)
O t-i
<
a) m u jj i-i o) >OIMJ^
^ 2 coi4)r-(W C
(03 i JJCOcB>«J=0-H4)EO'O OiiH do -Hoirerein-Hc'
«>3CC4l»-i>"-i3^i>-t4J-Hljj;C-Hjji8h>
Q
w
Q
JB 2
o
C 4) i
4) 0) i
•H 0
O 4)
IB
01
0)
•H < 01 V4
o< ?^
re oz
CQ oi o> Oi
• TI a b
CO! 3
O O IB A
V ft 0 4) to X C ^
M jj 2 >J 0) JJ -H JJ
<«iM <>-HOlJJ
2 0 JJ rH U ^ Oi
24) i 0 Z O 4) > '
i-H jj U. i JX i T)
coi-ic i o c ro >: c
j= > E re re JJ
Q..C 3 rH ft CO
E 01 IB rH 3 i
0 E
a>
B
a
JJ
«
JJ
CO
-------
CO
o>
»»
0)
0)
05
n
a
3
a
4£
in
to
«
zone Nonattainment Ai
O
H
1991-93 Update
Daily Eatimated
1-hr Exceedances
« *
- V
* X
• 4J
?.§
Clean Air Ac
Claasificati
„
Nonattaimnent Area Nan
oooooooooo
goooorsooooo
oooooooooo
O f-4
I-l 1-1
*-* *--
oooooooooo
JJ JJ
r- c C
•H -H -H rl M rJ tH -H -H -rt -H
re «
01 4>
Smyth County NA Area
Seattle - Tacoma NA Ai
Door Co NA Area
Kewaunee Co NA Area
Manitowoc Co NA Area
Milwaukee-Racine NA Ai
Sheboygan NA Area
Walworth Co NA Area
Charleston NA Area
Greenbrier NA Area
Parkersburg NA Area
>1 Nonattainment Areas
supplemental data from
jj
-H
3
w
tt
n Retrieval System (Al
o
-H
4J
TO
•rometric Infon
QJ
4)
JJ
fc
E
01
JJ
CD
CD
EPA' s air quality data
EPA Regional Offices.
deral Register,
ing.
0) "H
AJ.tt
c S
•H i
*j
t areas as published
are not included in
™ tn
E «
C (^
• ozone nonatta
'jonattainment i
w —
0 ~
•"re
oU
•H "
m »
fO r
si
Designations and classi
81. Unclassified and i
jj
. Jj
or calculating design
plete years of data is
t is important to note
is of the design value
«r.s
C n m o)
re m £
25 *«
3-5 jc c
CD * U 0
93 period using EPA
hest monitored value
one exceedance for ea
areas be classified
)d was used .
rH .^ B •U -H
rH £, iH £ O.
imated for the
ally, the fourt
the standard al
hat O, nonattai
Ly the 1987-89
m 4) W " 'a
S C 2 *O u
to 0 g " C
•H ft) "— 4)
ID ' *^ cr
3 o ^ JJ -
The updated air quality
:on Memorandum, June 18,
the updated air quality i
0 Clean Air Act Amendmen
the Amendments were pas
X to cr, 0)
> TO re rH ^
naximum 1-hour average
Dlumn shows the number
: quality value. This
3t two columns contain
last column shows the
listed in the previous
f* v H ~
Q< ni W«rri ^ (y
I parts per million (p
timated number of exo
cording the highest t
3-year period, 1991-9
r concentration in 1
id maximum 1-hour cone
^ CO 4) 3 c
' " ^ £. A 8
0 01 0) JJ ' 4)
CO 01JJ ^ ** to 5 ^
0 " S E ° °
01 C ° 4) 4)
S0~ rl CO rl
jj jj o
rH « "O - tO J-1
m 4) "*J ftj flj -rl
SrS « JJ J5 tO
i. S f Cn ai
'he National Ambient Air
reeded more than once pei
0.12 ppm standard was ej
r adjustment for incompl
ic site recording the hi
ceedances for 1993 at thi
f s 01 •"- x
Jj 4> jj -*-1 41
C
O
zone monitoring seas
0
41
£
JJ
rating during
41
a
0
s
a
M
""
S
Special purpose monitor
a County, WI .
in Kenosh
to
re
4)
z
o
Cn
TO
U
£
u
4)
JJ
0
UH
41
jj
•H
CO
0)
3
1-1
re
jj
-H
rH
re
^i
>H
TO
•c
4)
The nonattainment/updat
ON
J2
Bl
based on data throu^
re
4)
re
i status of the
41
JJ
D
•H
^
4)
•H
4)
r<
The Regional Office is
.2
I
CO
< < I
I > > >
u
o
w
o
tu
U
CO JJ 0)
01 U J3
3 0) JJ JJ
re 41.
> to .
JJ
TO
CD 41
•O TJ « -H 3
JJ JJ JJ rH
0
-------
rH
en
a*
r-4
C
•H
•o
01
JJ
n
complete data repc
c
M
r>
IB
M-l
0
a
3
jj
r-l
CE
JJ
. «o
tl
0
U-l
JJ
c
3
O
U
U
10
0
JJ
"O
01
JJ
to
3
*^t
•a
18
to
0)
u
c
IB
T)
CU
01
U
X
01
•o
01
JJ
IB
E
•H
JJ
to
HI
•a
c
IB
0)
3
rH
IB
>
iX
JJ
-H
,—4
(8
tr
M
.,-j
IB '
rH
•O o>
0) cr\
JJ iH
IB
•c c
a-H
"
p>,
01 rH
£ C
JJ 0
<* n
0 jj
§•8
'55
<0 -H
•3*
-£
m-H
u £
•V
. »>•
ffi"0
3
jj
W
w
o
£
0)
U
c
V
r-l
01
M-l
M
0)
JJ
c
•^
•c
01
N
•H
rH
IB
u
O
rH
O
JJ
01
•c
•o
V
CO
3
JJ
0
c
JJ
c
IB
rH
D.
jj
C
01
E
jj
IB
0)
U
JJ
r-l
01
JJ
IB
3
01
.C
jj
JJ
18
•o
CO
JJ
IB
O
O
rH
01
JJ
-H
to
in
•H
r-l
O
JJ
-H
0
«
!
• ^
M-l
IB
JJ
IB
a
CT\
o a
(N 0
1/1 JJ
1/1 01
^J
tO -H
•HJ:
*
c
0 M-l
-H 0
JJ
IB C
> 0
OJ «H
rH JJ
0) M
oiR
CO 10
JJ
>.to
•0-H
JJ 10
n 01
r-l
•c m
,2 B
u o)
c c
-H -i-l
IB IB
JJ JJ
C JJ
3 ro
0 C
£ 0
01
jr o)
jj s:
H
M-l
O
JJ IB
M 01
(8 ^
a n
CO JJ
IB r-l
0
< a
> 0)
c
- IB
C »H
•H JJ
IB
JJ rH
C ffl
3 rJ
O 3
£ V4
a IB
o
jj CO
0) "H
-H CO
ss
SH
a
o
JJ • C
«S °
; site was located
reported after 19
! 4500 feet elevati
j= « ?
^m°
•S-S
f
t-i 0
°^-H-
18
JJ C
0) 3
eu o
w
K
IH
<
C
•H
•o
V
JJ
u
u
rH
U-4
V
r^
JJ
41
»X
JJ
0
o
to
IB
01
CO
OJ
o
N
o
c
CO
c
0
u
to
-H
3
r-l
O
M-l
T3
0>
jj
CO
3
•n
•D
IB
CO
01
U
c
IB
•o
01
0)
u
X
01
•o
01
jj
IB
Iculation of estim
IB
u
rH
-------
APPENDIX?
Ozone Design Values From Time Series Simulations
94069.F F-l
-------
-L-iuc ocj: j.e.=> a j.iuuj.awj.on or ^.no-cago oite jri"/
Design Values (ppm)
********* N-Year Distribution
1
2
3
4
5
6
10
250
Standard
Table
CLV Lookup
137 .121
135 .126
134 .128
134 .130
133 .130
133 .130
132 .130
132 .132
Deviations
Tail-Expo
5%
.128
.130
.131
.131
.132
.132
.131
.132
(ppm)
10%
.128
.130
.131
.131
.132
.132
.131
.132
********* N-Year Distribution
1
2
3
4
5
6
10
250
Table
CLV Lookup
029 .020
020 .016
016 .014
015 .013
012 .012
012 .011
009 .009
002 .002
Tail-Expo
5%
.021
.016
.013
.012
.010
.010
.008
.002
10%
.020
.015
.013
.011
.010
.009
.008
.002
********
Percentile
95%
089
089
089
089
090
089
089
089
90%
.077
.077
.077
.077
.077
.077
.077
.077
********
Percentile
95%
010
007
006
005
005
004
004
001
90%
.008
.006
.005
.004
.004
.003
.003
.001
* Cond.
Prob.*
Tail-Ex
5%
.128
.131
.132
.133
.134
.134
.133
.134
* Cond.
10%
.128
.132
.133
.133
.134
.134
.133
.134
Prob.*
Tail-Ex
5%
.021
.016
.014
.013
. .011
.011
.008
.001
10%
.020
.015
.013
.012
.010
.010
.008
.002
Time Series: Log/LOWESS/AR(l); Shock: Lognormal distribution
chi!002b.cal
-------
Sites Studied in Time Series Comparisons of Design Values
CMSA
Atlanta
Charlotte
Chicago
Chicago
Chicago
Houston
New York
New York
New York
Site Code
13
37
55
17
17
48
34
36
09
089
119
059
097
031
201
023
085
001
0002
0034
0002
1002
7002
1034
0006
0067
3007
Abrev.
ATL0002
CHR0034
CHI0002
CHI1002
CHI7002
HOU1034
NYC0006
NYC0067
NYC3007
Shock Site-yean
Extreme
Extreme
Lognorm
Lognorro
Lognorm
Lognorm
Extreme
Extreme
Lognorm
11
10
11
11
11
10
10
11
11
-------
Time Series Simulation of Atlanta Site #13 089 0002 1
Design Values (ppm)
********* N-Year Distribution ******** * Cond. Prob.*
Table Tail-Expo Percentile Tail-Ex
CLV Lookup 5% 10% 95% 90% 5% 10%
1
2
3
4
5
6
10
.155
.155
.157
.157
.156
.156
.156
.144
.148
.152
.153
.153
.153
.155
.152
.153
.155
.156
.156
.156
.157
.155
.156
.158
.159
.158
.158
.159
.111
.110
.110
.111
.111
.111
.111
.097
.096
.097
.097
.097
.097
.097
.152
.154
.157
.158
.157
.157
.159
.155
.157
.159
.161
.160
.160
.161
250 .156 .156 .157 .159 .111 .097 .159 .162
Standard Deviations (ppm)
********* N-Year Distribution ******** * Cond. Prob.*
Table Tail-Expo Percentile Tail-Ex
CLV Lookup 5% 10% 95% 90% 5% 10%
1
2
3
4
5
6
10
. 026
. 020
.017
.015
. 014
.013
.009
.022
.017
.015
.014
.012
.012
.009
.023
.017
.014
.013
.011
.011
. 008
.023
.016
.014
.012
.010
.010
.007
.012
.008
.007
.005
.005
.005
.003
.009
.007
.005
.004
.004
.004
.003
.023
.018
.015
.014
.012
.012
.009
.023
.017
.015
.014
.012
.012
.009
250 .001 .001 .001 .001 .001 .001 .001 .001
Time Series: Log/LOWESS/AR(1); Shock: extreme value distribution
-------
Time Series Simulation of Chicago Site #55 059 0002 1
Design Values (ppm)
********* N-Year Distribution ******** * Cond. Prob.*
Table Tail-Expo Percentile Tail-Ex
CLV Lookup 5% 10% 95% 90% 5% 10%
1
2
3
4
5
6
10
.159
.159
.157
.158
.157
.157
.158
.139
.147
.149
.151
.152
.153
.155
.148
.152
.152
.154
.153
.155
.155
.148
.151
.151
.152
.152
.153
.154
.099
.099
.098
.099
.099
.099
.099
.084
.084
.084
.084
.084
.084
.084
.148
.154
.155
.156
.156
.158
.159
.148
.153
.154
.156
.156
.157
.158
250 .157 .157 .156 .155 .099 .084 .160 .159
Standard Deviations (ppm)
********* N-Year Distribution ******** * Cond. Prob.*
Table Tail-Expo Percentile Tail-Ex
CLV Lookup 5% 10% 95% 90% 5% 10%
I
2
3
4
5
6
10
. 047
.033
. 028
. 024
.021
.020
.015
.033
.028
.025
.021
.019
.019
.014
.035
.027
.023
.020
.018
.017
.014
.033
.026
.022
.018
.016
.016
.012
.016
.011
.009
.008
.007
.007
.005
.011
.008
.006
.005
.005
.005
.004
.035
.028
.025
.022
.019
. 018
.014
.033
.027
.024
.020
.018
.018
.014
250 .003 .003 .003 .003 .001 .001 .003 .003 111
Time Series: Log/LOWESS/AR(l); Shock: Lognormal distribution
chi0002b.cal
-------
Time Series Simulation of Chicago Site #17 031 7002 1
Design Values (ppm)
********* N-Year Distribution
1
2
3
4
5
6
10
250
Standard
CLV
138
137
138
137
137
138
137
137
Table
Lookup
.123
.129
.132
.133
.134
.135
.135
.137
Deviations
Tail-Expo
5%
.130
.133
.134
.135
.135
.136
.136
.137
(ppm)
10%
.131
.134
.135
.135
.135
.136
.136
.137
********* N-Year Distribution
1
2
3
4
5
6
10
250
CLV
028
021
018
015
013
012
010
002
Table
Lookup
.021
.018
.015
.014
.012
.012
.009
.002
Tail-Expo
5%
.022
.018
.015
.013
.011
.011
.009
. 002
10%
.021
.017
.015
.012
.010
.010
.009
.002
********
Percentile
95%
094
094
094
094
094
095
094
095
90%
.082
.082
.082
.082
.082
.082
.082
.082
********
Percentile
95%
012
009
007
006
005
005
004
001
90%
.009
.007
.005
.005
.004
.004
.003
.001
* Cond .
Prob.*
Tail-Ex
5%
.130
.134
.136
.136
.137
.138
.138
.139
* Cond .
10%
.131
.134
.136
.136
.137
.138
.138
.139
Prob.*
Tail-Ex
5%
.022
.018
.016
.014
.012
.011
.009
.002
10%
.021
.017
.015
.013
.011
.011
.009
.002
Time Series: Log/LOWESS/AR(l); Shock: Lognormal distribution
chi7002b.cal
-------
Time Series Simulation of Charlotte Site #37 119 0034 1
Design Values (ppm)
********* N-Year Distribution ******** * Cond. Prob.*
Table Tail-Expo Percentile Tail-Ex
CLV Lookup 5% 10% 95% 90% 5% 10%
1
2
3
4
5
6
10
.129
.130
.132
.131
.131
.132
.132
.122
.126
.128
.129
.129
.130
.131
.127
.129
.131
.131
.131
.131
.132
.129
.131
.133
.133
.133
.133
.134
.103
.103
.104
.104
.104
.104
.104
.094
.094
.094
.094
.094
.094
.094
.127
.129
.131
.131
.132
.132
.133
.129
.131
.133
.133
.134
.134
.135
250 .133 .133 .133 .135 .104 .095 .134 .136
Standard Deviations (ppm)
********* N-Year Distribution ******** * Cond. Prob.*
Table Tail-Expo Percentile Tail-Ex
CLV Lookup 5% 10% 95% 90% 5% 10%
1
2
3
4
5
6
10
.016
.013
.011
.010
.009
. 008
.006
.014
.012
.011
.009
.008
. 008
.006
.015
.012
.010
.009
.008
.008
.006
.015
.012
.010
.009
.008
. 008
.005
.010
.007
.006
.005
.004
.004
.003
.008
.005
.004
.004
.004
.003
.002
.015
.012
.010
.009
.008
.008
.006
.015
.012
.010
.009
.008
.008
.006
250 .001 .001 .001 .001 .000 .000 .001 .001
Time Series: Log/LOWESS/AR(1); Shock: extreme value distribution
-------
Time Series Simulation of New York Site #09 001 3007 l
Design Values (ppm)
********* N-Year Distribution
1
2
3
4
5
6
10
250
Standard
CLV
213
215
217
215
217
217
217
218
Table
Lookup
.186
.200
.206
.207
.210
.211
.213
.218
Deviations
Tail-Expo
5%
.198
.206
.210
.212
.213
.213
.214
.217
(ppm)
10%
.198
.206
.210
.211
.212
.212
.213
.216
********* N-Year Distribution
1
2
3
4
5
6
10
250
CLV
065
064
035
031
029
023
018
003
Table
Lookup
.049
.040
.032
.029
.027
.022
.018
.003
Tail-Expo
5%
.053
.040
.032
.028
.026
.021
.017
.002
10%
.051
.038
.031
.027
.025
.020
.016
.002
********
Percentile
95%
133
134
136
137
136
136
136
137
90%
.113
.114
.114
.115
.114
.114
.114
.115
********
Percentile
95%
028
020
017
015
013
Oil
009
001
90%
.022
.016
.013
.012
.010
.009
.007
.001
* Cond.
Prob.*
Tail-Ex
5%
.198
.208
.213
.214
.216
.216
.217
.220
* Cond.
10%
.198
.208
.213
.214
.216
.216
.217
.220
Prob.*
Tail-Ex
5%
.053
.041
.033
.029
.027
.022
.017
.002
10%
.051
.039
.032
.028
.026
.021
.017
.002
Time Series: Log/LOWESS/AR(l); Shock: Lognormal distribution
nyc3007b.cal
-------
Time Series Simulation of Houston Site #48 201 1034 1
Design Values (ppm)
********* N-Year Distribution
1
2
3
4
5
6
10
250
Standard
CLV
239
228
225
223
221
220
220
218
Table
Lookup
.200
.208
.212
.213
.214
.214
.216
.218
Deviations
Tail-Expo
5%
.212
.213
.214
.214
.214
.214
.215
.214
(ppm)
10%
.210
.209
.211
.211
.211
.210
.211
.211
********* N-Year Distribution
1
2
3
4
5
6
10
250
CLV
066
039
035
027
024
022
018
003
Table
Lookup
.040
.031
.030
.023
.022
.020
.016
.003
Tail-Expo
5%
.039
.026
.025
.019
.018
.016
.013
.002
10%
.034
.023
.031
.017
.015
.014
.011
.002
********
Percentile
95%
115
115
115
115
115
115
115
115
90%
.093
.093
.093
.093
.093
.094
.093
.093
********
Percentile
95%
013
008
007
006
005
005
004
001
90%
.010
.006
.005
.004
.004
.004
.003
.000
* Cond.
Prob.*
Tail-Ex
5%
.212
.215
.217
.217
.218
.217
.219
.219
* Cond.
10%
.210
.212
.215
.215
.216
.215
.217
.217
Prob.*
Tail-Ex
5%
.039
.028
.028
.021
.020
. 018
.015
.002
10%
.034
.025
.025
.019
.018
.016
.013
.002
Time Series: Log/LOWESS/AR(1); Shock: Lognormal distribution
hou!034b.cal
-------
Time Series Simulation of New York Site #34 023 0006 1
Design Values (ppm)
250
********* N-Year Distribution ********
Table Tail-Expo Percentile
CLV Lookup 5% 10% 95% 90%
* Cond. Prob.*
Tail-Ex
5% 10%
1
2
3
4
5
6
10
.172
.173
.172
.173
.172
.173
.173
.157
.164
.166
.169
.168
.170
.171
.166
.169
.170
.172
.171
.172
.172
.169
.172
.172
.174
.173
.174
.174
.121
.121
.120
.122
.121
.121
.121
.105
.105
.105
.105
.104
.105
.105
.166
.170
.171
.173
.172
.173
.174
.169
.173
.174
.176
.175
.176
.177
.172
.171
.172
.174
.121
.105
.174
.176
Standard Deviations (ppm)
250
********* N-Year Distribution ********
Table Tail-Expo Percentile
CLV Lookup 5% 10% 95% 90%
* Cond. Prob.*
Tail-Ex
5% 10%
1
2
3
4
5
6
10
.033
.024
.020
.017
.016
.014
.011
.027
.021
.018
.016
.015
.013
.011
.029
.021
.018
.016
.014
.013
.010
.029
.020
.017
.015
.013
.012
.009
.016
.011
.009
.008
.007
.006
.005
.012
.008
.007
.006
.005
.005
.004
.029
.021
.018
.016
.014
.013
.010
.029
.021
.018
.016
.014
.013
.010
.002
.002
.001
.002
.001
.001
.002
.002
Time Series: Log/LOWESS/AR(1); Shock: Extreme value distribution
nyc0006b.cal
-------
Time Series Simulation of New York Site #36 085 0067 1
Design Values (ppm)
********* N-Year Distribution
1
2
3
4
5
6
10
250
Standard
CLV
158
160
159
160
160
160
159
160
Table
Lookup
.146
.153
.154
.156
.157
.158
.158
.160
Deviations
Tail-Expo
5%
.154
.157
.158
.159
.160
.160
.160
.161
(ppm)
10%
.157
.160
.161
.162
.162
.162
.162
.163
********* N-Year Distribution
1
2
3
4
5
6
10
250
CLV
025
019
015
013
Oil
Oil
008
001
Table
Lookup
.021
.017
.014
.013
.011
.011
.008
.001
Tail-Expo
5%
.022
.017
.014
.012
.010
.010
.007
.001
10%
.023
.017
.014
.012
.010
.010
.007
.001
********
Percentile
95%
116
117
116
117
117
117
117
118
90%
.102
.103
.102
.103
.103
.103
.103
.103
********
Percentile
95%
014
010
008
007
006
006
004
000
90%
.011
.008
.007
.006
.005
.005
.003
.000
* Cond .
Prob.*
Tail-Ex
5%
.154
.158
.159
.160
.161
.161
.161
.163
* Cond.
10%
.157
.161
.162
.163
.163
.163
.163
.165
Prob.*
Tail-Ex
5%
.022
.017
.014
.012
.010
.010
.008
.001
10%
.023
.017
.014
.012
.011
.010
.007
.001
Time Series: Log/LOWESS/AR(l); Shock: Extreme value distribution
nyc0067b.cal
-------
APPENDIX G
Summary of Studies on Meteorological Adjustment of Ozone Trends
94069.G G-l
-------
Adjustment Method
1 Variables Used
for Adjustment
S fr.a
Oet «
E M
•a g "i
g j£
g t/3
<
c _
.2 H
ta -5
u p
aS
(U
1
g
<£
o
Z
Classification: generated ozone potential
score based on tabulated meteorological
conditions and regressed cumulative score
against number of exceedances - applied
regression to climatological mean
meteorological conditions
^
inversion strength,
950 mb temperature
inversion height,
horizontal pressure
gradient
00
C
11
<4-
0 0 g
j- 8 a.
4) CL
E s S
P O •
C J= 0
°
o t 'C ^ -o > -^ |
~ g
sl
« r*
fc t
"c3
00
c
< .g
O c§
_) CO
c
rt
'5
CO
•a
'•5 ^
U as
cs
Classification: number of days in each
meteorological category re-weighted
based on ratio of long-term average
category frequency to observed frequency
•a
synoptic typing
scheme based on
regional surface
pressure patterns ani
local weather
W '
2 ° ' ' E
o ^~* g o ^- "Q §*
| 1 1 o o § o
C «Ci *S ^^ ^*^ ^^ <**^
3 fc. ^ • * *A
c & c o o o A
X
c"
o
3
O
SC
•a
^j
4J
e
1
4i V*s
1 *°
N O
^
Classification: "ozone conducive" days
defined on basis of joint meteorological
criteria determined by inspection; ratio of
number of exceedance days to number of
conducive days in each year provides
adjusted trend
daily maximum
temperature, dew
point, precipitation,
wind speed, percent
sunshine
•o d
o oo
i- .£
4) "O
O Q
E 8 E
3X0.
CUD.
^
o"
00
_u
j5
U
"o
•o
I
•1 is
£o
^
-------
1
o
U
O
W
J
CD
<
Adjustment Method
"S ^
V Q)
D £
5/5 ^
4> 3
S ^
.2 <
l£
0)
S b.H
O 2 *i
E S2
1 E'S
i«g*
<
o ^
•*-» 'rf
cQ ^3
O p
Q *3
O
1
(2
6
Z
^
C ^ O
O *^ ^v
"i s 4
Classification: defined daily
meteorological index of ozone form;
potential; normalized exceedance ra
function of index for each year to 2
average index values
w>
03
3 _,
.1 £ -o ,o
X 3 ^ C **j
CQ •*•* ^ O ^
1 111 !
« 1 f S 8
™ .is s> •— ex
10 rs
•S o
(_)--,
Ill
E 8 E
^
o"
0
S
T3
o
1
t/5
T3
§
^-*s
O ^^
M o
IT)
l_
S r^
^..B -9 1
Regression: step-wise multivariate
regression of daily maximum ozone
regression parameters estimated froi
year base period and applied to all
subsequent years; adjustment factor
difference between observed and
predicted concentrations in subsequ(
years
Ji« wJ
•S ta
"2 T3 .2
•^ ^_» ^
c o £
""" rt ***
— "e s
-° « ^
S 11
"©
•S 2
^» • *• *) 1
« 3 r. c
illlg
^§3X0.
c c c n> ex
—
&
C
< c
o 'i
i-l CO
§
0
13
4)
2
"S
N
VO
If
Classification: summary statistics
compared for all years with similar
numbers of days above 30 and 32 d
E
3
.§ £
X 3
E *-
> ex
11
•>-> & p „ *j C 4J~ *S_
t/5 « •»-* ^ LO V5 •— fe£) ^*
sj | g 2 »2 o o. « ^
(•in .v« »C .«• O hf] rt 4) ' [ f
C «^ •*• ti« c O 3 f-i 4j ^^
O ^* V^ ^* *"* Q 1 Q cj 4) O4
ri"-ir3*->fiCr-E3u^H
DrtoeeeSOT^SX
«-o<*:-o.= !« coo c ^ ex
ll
M
•o o
• [ ,
o *5
o 55
E 8 E
3 x; ex
c ex
s<
f a"'
^5
o1
ON
^,
a
1
«
on
oo
-------
I
+->
o
O
u
-J
CQ
Adjustment Method
"S c
D |
8 §
3 •o'
.2 <
J3 •-
^. O
s
8 bo
OcS -S
E •—
*S g 'i
i § 00
C ^
Is
C3 *O
o 2
t-5 ^
4)
1
U*4
H
ff»
6
Z
•I i
o " c
c e o -jf
.og £j J3
Regression: simple linear regress
San Francisco Bay Area; correlat
with ambient HC, CO, and daily
maximum temperature suggest wi
to summer HC to summer ozone
adjusted trend calculated
fc ««
O i-
Q
C- C
.2 '^
.•£ c
•S" .5
o M
cL o
3'i
S o.
l§ i.
T3 o Q.
**"" K ^
fe 53 =>
•g 8oT
C 4> O
O
U
GO
H
i £•
CO CD
oe"
**-*
13
t>
go
'E "=
UH ,~
B >, &
ea -C^
<5 |5 —"
i*'| -g>
& f, x:
>> » -
j; ^-"i c .S
^i c^ rO W)
~ E ^ -o E £ c
^ 3 TJ 2 3 00
^ E S | E fe S
S 'R 8 .i 'x "8 s -a
EESoE^ES
__
"S
< <=
s 'i
U CQ
00 -jj
c^ o
^a
•a -o
s §
W /— s.
*fj cd r^
o E °°
63 -
o
u
4) '5b
'-C "o
Regression: annual step-wise mu
linear regression applied to climal
mean meteorological conditions
"rt «j
?f c c
b o S
-Zi ^ -5
* '§ 2
P "O U
fl^ C *""
*- re 2
to |
Rll
00 > S
>>
^ 1
c!
c5 x
O K!
E E
C/}
13
c
< G
Vi 'tfl
O W
J CQ
.
15
j-
O
& — ^
^ O
o oo
•3 ON
Q O
—
c
_o
Regression: annual linear regress
applied to climatological mean
temperature
o
h:
^
re
C.
E
*->
t
o
oo
w rs)
^> ^_ (—
c -S • ^
**^ fll ^ '* i f^-i
O r— ff f~) ^C
§ « E 1 2 .s s
• —60-^g g 4J "O •
*>S*-o£c—
1
0
IT)
OO
C
<4- 2 >•>
ill!.!
rn 55 E E E
WI
"S
< c
s'i
J CQ
JS
U
•o
CT3
^ -^
CC r —
c oo
3 ON
m
-------
I
o
U
O
U
CQ
Adjustment Method ||
"8 i
(« 4>
=> £
£ 3
i?
| s
> «.
^
C
0 «"'"
z E|
CO E «
3 3 ^3*
C ri Cfl
s °°
'^d
.11
OB T3
o 5
1
*§
6
z
Classification: days conducive to high
ozone determined from joint distribution
of meteorological variables; annual ozone
statistics regressed against number of
conducive days and year; year term
assumed to represent adjusted trend
"S •« S
§ .S.2
E * %
3 « .fc
E »•= cLis
= .E E g
•o E S &
^.^ E
fe1 — 3
2 d s •-
o M g x
<5 ^3 E E
E g E •— ' *
3 X O.'w
c S ex-o
•4-*
S
'§
V
c
o
U
Jj>
^*
ea
CD
1
^J ^^\
T^ QQ
O OO
CO ^^
^ o
—
Classification: ozone conducive and non-
conducive days identified on basis of
meteorological data; adjusted trend based
on frequency of exceedances on ozone
conducive days
•sS
E & 8
i s"B
'§!•§
c j- *
rt
'G
c«ri
_•
'a
o
1
° irT
JS oo
•5 ON
<£ O
V)
Regression: elimination of days not
conducive to high ozone based on joint
distribution of meteorological parameters;
multiple log-linear regressions of daily
maximum ozone for each year; reference
year of meteorological data substituted
into each regression model and adjusted
expected exceedance rate calculated from
regression prediction allowing for random
error terms
« o
— (U V
E 8 E
3 X O.
C 4> 0.
I
re
0)
JG
O •
Z D
OO
00
o\
15
u
s
O.
S
Decision Tree: classification and
regression trees (CART) used to
categorize ozone concentration on each
day on basis of meteorological variables;
reweighting of observations based on
climatological frequency of occurrence of
each meteorological category; adjusted
annual summary statistics calculated from
the reweighted observations
i_r
> > o
« 8 'o
6"73 T3 "O i_
<- 3 c ^
3 O 'S ^
E 4>" "o ^ T3
'5s ^ c' '1
S « ^ 2 ^
E S3 =5 ta ,3
— * c* 'E "o ^
^ _
^— 3
•° 6 _ E
U— i ^" '^-
o c° S g
fcl E 1
E K E ~"
3 X O. 'w
C (L> O- -O
I I
1 &
.2 ..-o1
iS < 'C E-
£ (X CO U
f
ON
^^
•o 1
i •§
^ o
u -5
O 3
oo X
t-
-------
I
•w
I
o
U
CO
Adjustment Method
"8 I
35 4)
3 §
WJ "3
.2 -S,
S -51
.2 <
c3 «-
>£
V
fi *-».
N C1 0
Ort \C
E 2
"« g 'ii
"5 *2 4-»
c ,5 v*
c tf>
^^
•is
CS TJ
83
j^j »*
4)
O
1
JU[
«J
PC
o
Z;
O C3
Regression: step-wise multiple linear
regression using data from all years;
adjusted annual mean summary statistii
obtained by applying regression model
with climatological mean meteorologic
conditions and adding errors back in.
Logistic regression used to adjust
exceedance summary statistics
•cT >•> ft
ij \j O
<*- a5 "> 3
O "O Q. - W
C W5 >\ •*-" 'P
'3 g'l 1 f% g
.£ 3 - 3 * 'C
l/^ S .13 "73 O k- t-
TJ- a -o c o ex ex
•0 >> x
CQ CTJ CS "~
•0 "O O
*° ^ e «3 -5
c ••"• •£ JN cS
05 •*• S E o E
e* ^? e 3 X CX
fc \6 fc c
& §1 1
«*- o- * E
° >-. C >^ o
« Xi » r= 5
|g|-l
1 1 •§ 1 1
V-i J~ *^ O fc
ci^s-B &l
IA
w CN \O
^ "&"§
«3 -5 rt
-0 «< S E
e (D t* *^
e S J= x;
3 X CX CX
c cx ex
X
c"
2
3
O
K
,—
o
Os
T^ t
^-^
^>
«
f
2
•a c *j K
u c ir™ *T3 TL
Statistical Distribution Fit: Daily
maximum ozone modeled by Weibull
distribution with invariant shape
parameter and scale parameter expressi
as a function of meteorological conditii
and an annual trend term; parameters 1
via maximum likelihood; model applies
with normalized meteorological variabi
* •*-*
bo !fi •£•
c - o > .5?
'£ "S u •- Q j=
E 1 & g f "B .c1
'x 3 5 c a -^ ^
1 ^ •" o & • - c1
>> ex 2 S -a •£ 'c
= E « £ .S | o
JS S 2; "6 > 2 c:
T3 *2 W 03 !> *C C
—
4_>
§ s=
u c
u 3r> c
^* *CC •«
j: T3 'x
*^ ert
ON »4- 2
ON O E
C
V- O O
52 to "^
M « 'o
as ^
^^ *m *O CX5
^_ " w «»-• .
(S § 2 p
ON
ON
N^^^
S
•a
g
x
a
o
c
o
Classification: Ozone conducive days
defined as days with temperatures in
excess of 90° F. Adjusted trend basec
frequency of exceed ances on ozone
conducive days.
E
3
.§ £
« 1
E £
D
>> CX
11
W3
CC ^~
•o o
'o w>
4) *Q
^ c
E 8 E
3X0.
c
-------
NUMBERED REFERENCE SECTION
[1] Zeldin, M. D., and D. M. Thomas. 1975. "Ozone Trends in the Eastern Los
Angeles Basin Corrected for Meteorological Variations." San Bernardino County
Air Pollution Control District, San Bernardino, California.
[2] Zeldin, M. D., and L. Breiman. 1979. "Trend Analysis of Emissions and
Meteorologically Adjusted Ozone Air Quality in Houston and Los Angeles. Task
1: Application and Refinement of Methodology for Adjusting Ozone Air Quality
for Meteorological Influences." Technology Service Corporation, Santa Monica.
California.
[3] Zimmerman, K. A., R. Tropp, and R. Barta. 1987. "The Relationship of
Weather Patterns to Ozone in the Houston, Texas Area." Presented at the 80th
Annual Meeting of the Air Pollution Control Association, New York, New York,
June 21-26, 1987.
[4] Sweitzer, T. A., and D. J. Kolaz. 1984. "An Assessment of the Influence of
Meteorology on the Trend of Ozone Concentrations in the Chicago Area." Air
Pollution Control Association/American Society for Quality Control Specialty
Conference on Quality Assurance in Air Pollution Measurements, Boulder,
Colorado (14-18 October).
[5] Kolaz, D. J., and R. L. Swinford. 1990. "How to Remove the Influence of
Meteorology from the Chicago Area Ozone Trend." Presented at the Air and
Waste Management Association 83rd Annual Meeting and Exhibition, Pittsburgh,
Pennsylvania, June 24-29, 1990.
[6] Zeldin, M. D., and J. C. Cassmassi. 1990. "Ozone Trends in the South Coast Air
Basin: An Update." South Coast Air Quality Management District, El Monte,
California.
[7a] Burkhart, R. P., N. J. Beloin, and A. W. Oi. 1990. "A Decade of Ozone
Monitoring in New England: Analysis and Trends." Presented at the Air & Waste
Management Association 83rd Annual Meeting and Exhibition, Pittsburgh,
Pennsylvania, June 24-29, 1990.
[7b] Burkhart, R. P., and N. J. Beloin. 1991. "Analysis of New England Ozone
Levels: 1981-1990." Air and Waste Management Association Seventh Joint
Conference on Applications of Air Pollution Meteorology, New Orleans, Louisiana,
January 14-18, 1991.
[7c] Burkhart, R. P. 1990. "Analysis of New England Ozone Levels and Trends:
1981-1991." Presented at the Air & Waste Management Association Specialty
Conference, Tropospheric Ozone: Nonattainment and Design Value Issues, Boston,
Massachusetts, October 27-30, 1992.
G-8
-------
[8] Savukas. R. A. 1990. "Allegheny County Ozone in the 1980s." Presented at the
Air and Waste Management Association 83rd Annual Meeting and Exhibition.
Pittsburgh, Pennsylvania, June 24-29, 1990.
[9a] Sandberg, J. S., M. J. Basso, and B. A. Okin. 1978. Winter rain and summer
ozone: A predictive relationship. Science, 200:1051-1054.
[9b] Sandberg, J. S., M. J. Basso, A. Okin, and D. Fairley. 1991. "The Winter
Rain and Summer Ozone Relationship Re-examined." Presented at AWMA
Specialty Conference: Tropospheric Ozone and the Environment II, Atlanta.
Georgia, 4-7 November 1991.
[lOa] Chock, D. P., S. Kumar, and R. W. Herrmann. 1982. An analysis of trends
in oxidant air quality in the South Coast Air Basin of California. Atmos. Environ.,
16(ll):2615-2624.
[lOb] Kumar. S., and D. P. Chock. 1984. An update on oxidant trends in the South
Coast Air Basin of California. Atmos. Environ.. 18(10):2131-2134.
[11] Duckworth, S., J. Kinney, and F. Granum. 1980. "The Effect of Meteorological
and Other Factors on Air Quality Trends and Pollution Episodes." In Air Quality
Trends in the South Coast Air Basin Through 1979. Air Resources Board,
Technical Services Division.
[12] Davidson, A., M. Hoggan. and P. Wong. 1985. "Air Quality Trends in the
South Coast Air Basin 1975-1984." South Coast Air Quality Management District,
El Monte, California.
[13] Kuntasal, G., and T. Y. Chang. 1987. Trends and relationships of 03, NOX,
and HC in the South Coast Air Basin of California. J. Air Pollut. Control Assoc..
37:1158-1163.
[14] Wackter, D. J., and P. V. Bayly. 1987. "The Effectiveness of Connecticut's
SIP on Reducing Ozone Levels from 1976 through 1987." Air Pollution Control
Association Specialty Conference, "Scientific and Technical Issues Facing Post-
1987 Ozone Control Strategies." Hartford, Connecticut.
[15] Pollack, A. K., and M. M. Moezzi. 1985. "Application of a Simple
Meteorological Index of Ambient Ozone Potential to Ten Cities." Systems
Applications, Inc., San Rafael, California (SYSAPP-85/197).
[16] Pollack, A. K., T. E. Stoeckenius, J. L. Haney, T. S. Stocking, J. L. Fieber,
and M. Moezzi. 1988. "Analysis of Historical Ozone Concentrations in the
Northeast." Systems Applications Inc., San Rafael, California (SYSAPP-88/192).
[17] Stoeckenius, T. E., and A. B. Hudischewskyj. 1990. "Adjustment of Ozone
Trends for Meteorological Variation." Systems Applications, Inc., San Rafael,
California (SYSAPP-90/008).
G-9
-------
[18] Wakim, P. G. 1990. "1981 to 1988 Ozone Trends Adjusted to Meteorological
Conditions for 13 Metropolitan Areas." Paper 90-97.9, presented at the 83rd
Annual Meeting of the Air and Waste Management Association, Pittsburgh,
Pennsylvania, June 24-29, 1990.
[19] Shively, T. S. 1990. "An Analysis of the Trend in Ground-Level Ozone
Using Nonhomogeneous Poisson Processes." Paper 90-97.10, presented at the 83rd
Annual Meeting of the Air and Waste Management Association, Pittsburgh,
Pennsylvania, June 24-29, 1990.
[20] Cox, W. M., and S.-H. Chu. 1991. "Meteorologically Adjusted Ozone
Trends in Urban Areas: A Probability Approach." Presented at the Air and Waste
Management Association Specialty Conference on Tropospheric Ozone and the
Environment II, Atlanta, Georgia, November 4-7, 1991.
[2la] Jones, K. H. 1992. "The Truth About Ozone and Urban Smog." In Policy
Analysis. No. 168, February 19, 1992, Zephyr Consulting, Seattle, Washington.
[21b] Jones, K. H. 1992. "The 1990/91/92 O3 Data Base and Its Implications Relative
to Currently Designated Ozone Nonattainment Area Regulatory Programs."
Presented at the Air and Waste Management Association Specialty Conference,
Tropospheric Ozone: Nonattainment and Design Value Issues, Boston,
Massachusetts, October 27-30, 1992.
G-10
-------
APPENDIX H
Summary of Articles on Detecting Trends in Ozone Design Values
H-l
-------
H-2
-------
i
13
.£?
'33
•o
o
o
N
O
_c
•o
§
*-*
bA
.S
o
"S
•o
i_
<«
4}
3
.2"
'c
x:
!
1
W>
JO
6
(D
•o
M
!S
*0
|
V3
^
ffi
1
CA
P
O
o
V)
1
£
"el
o
en
'*!*
a
CO
2
5
§
'&
*
to
b
2
s
BJ
d
v oo i-
*- Sr r>
This methodology can be used for any
national average annual air quality measu
The least squares estimates of model
parameters can be used to estimate missit
site values and adjust the annual means f<
those missing values. Trends can be
evaluated by comparing annual mean
confidence intervals for different years.
U 4^
~ C
^ S|
U * °
t£ o ,£
.§ I s
t« w ">
Hi
M
b^ 1
wot)
E ~ 71
— s rt S
§ S - tl
c S o
lllS
ob r-
f~ 00
ON O\
g
I
z
T3 >>
O "3 e
1 1 1
1 •& H
« .SP rt
* «,
S. .— u £
C8 t^ eil rt rt
This paper provides an enhancement of p
1 in which missing annual summary statii
are estimated using trends in local sites
rather than the national trends. Regions i
defined by an agglomerative algorithm th
minimizes the error in the fitted least squ
model.
o
.E £ *
O .** C
TO 5? [/*
fc •*- =
£ O 60 "S.
« E .5 .
o = "b o
.M 3 — , *-*
"w M fc '55 e
3 J 8" ^ 1
^f, ^ *^ •*•'
g B S 1 fc
p -o % S -s
- 1 ^ 8 i
3 K * n O-
C P «-- O
5 JJ C M t»
oo r»
t~ 00
Os Os
§
«
Z
"S •£•
c s i
3 43 *jj<
«5 00 rt
S £ E
^
§ 8
tn •-- •••
Only 14 percent of the straight line slope
were statistically significant (those slopes
were also all negative). Normal distribut
and constant variance assumptions are
unlikely to be valid for exceedance statist
^
£
li^
P ~O fc
in
« o u
I = "o
— *- E
I1N .
c/: . ^
in -p £. 3
c« C <*- O^
Eel t/>
P P fa to
P S g (8
c/) S3 >> —
i
t— 00
O^ t^>
O
O .
5|
C/2 'w
§ i.
g*8 S"8
ll^ll
XS JM C *O
9 J3 *P p v
S S E g. g
en
00 M
.S "3
In model a, trends were evaluated by plo
annual confidence intervals for the mean
daily maximum oxidant for each group ol
days. In b, the regressions were used to
calculate and plot annual confidence inter
based on the midvalues of each range of
meteorological parameters used in model
Trends were essentially flat for each
meteorological regime.
^
P U5
1 8-3 g
i!1 ]!
FTi 5^ O *"* w
o j5 £2
j; - P E
* £ 2 £ —
§ 1 '"8 . I .§
>-\ ^" £ *^s Wj
1 if §|1
w 4j p- •*— ,~ f*
c- 4.5 -^ Q r* y
•C Z 0 Q. « &
B. 4i •>- W) 4)
15 Ja "§ •! ~ E
Jg k. tS ^5 >c e
03 o. n E t '""
"1? S "« td
.__
t- OO
Os ON
i .
O P
U '53
I .b <
W < U
E
^
«
«
D
"1
-------
§
u
ffl
Comments
Statistical methods
e
Ctf
>
c
o
•§>
06
Q
'i
In
rt
c«
fc*
{Q
§
3
O
1
i --iii
•JJ, P '§ g tc u. jg
S §£* **•'"' ^ 'E* §
ill iff if
e
0 0
'5 ~ c «
The tail distribution of the concentn
for the cluster maximum is assumed
be the generalized Pareto distributio
with a location parameter that is a
straight line function of the year (thi
intercept and slope depend on the
month).
Cluster maxima occur as a Poisson
process.
o NO
r** oo
«-* «-H
rS
(S
o
to
§
K
52 S J
§ ® *a ifi ^ S
•5 *c <*c ^i jj g
a o c *> « S
3 o 5^ JS
B G j^ ^3 ,. . J*
*. -3 60-3 o » .•
ii V a! J2 w c 5
U W «S •£ x II
vO
"8
The unadjusted exceedance rates showed a
very weak downward trend. The
meteorologically adjusted exceedance rates
showed a stronger downward trend.
Statistical tests of trend in papers 8 and 9
showed a non-significant trend for unadjust
exceedances and a statistically significant
trend after adjustment.
w
**- 3
O •=
Trends are adjusted for average
meteorological conditions using an
ozone formation potential index to
define ozone conducive days. The
Spearman's rho non-parametric test
trend (correlation between annual va
and the integers) was used.
r- r-
r- oo
Os Os
J
hM
-
W)
.§
Lc
U
0
o
"U
S
8
X
V
r-*
i «
1 ^
< T3
ON
oo'
r-"
J I)
The trends were shown to be statistically
significant at the 1 percent significance leve
Normal distribution and constant variance
assumptions are unlikely to be valid for sor
of these statistics.
S *u
£ C
Trend lines for the raw data were fr
by simple linear regression. Trend
lines were also fitted using multiple
linear regression against emissions a
"exceedance conducive days."
r~- t-
VO 00
ON Os
3
O
Q
O
U
•o u
c a - "
^ o 8 g S
TH M S3 ^ *X
3 o E 8
"w E «S ^
^j! <-» ^ S ^
§ !S .§ ? "3 •
irlll §1
0
. S
Observed decreases were found to be
statistically significant at the 5 percent levei
Constant variance assumptions are unlikely
be valid for the exceedance statistics.
Normality assumptions may be reasonable
approximations due to the large expected
numbers of exceedances.
03
For each site, trends were evaluated
using simple linear regression to
estimate the percentage trend and thi
statistical significance of the trend.
^
t^*- O^
Q\ ON
^?! ^j *•
c§ & C
o -S & '5
o ^ oo
1 * « § <
cfl i [U < U
u
B «
•a o
O -C *
g "3 g
§£ P
c9
3 . "8 "3
§
-------
3
'€
O
U
Comments 1
Statistical methods
>>
§
So
(2
O
'OT
a
V)
j-j
S
s
c
C/3
6
Z
Although the methods were derived in the
context of water quality trend analysis they
can be applied to trends in ozone design
values and summary statistics. The power
calculations for linear regression use
approximate formulae that are invalid for
small samples.
«1
•n T .—
Approximations for the power (tren
detection probability) of parametric
(linear regression) and non-paramet
(Spearman's rho) trend tests are der
mathematically and by simulation.
Cases of independent and dependen
sequences are considered.
ts m
•* t^
Cv O\
£
8^
> < T3
3* S
a
3
£•
*«
cr
b.
&
ed
£
fS
Trend detection probabilities are derived by
simulation for various distributions (including
autoregressive - moving average time series)
using the seasonal Kendall tau test, simple
linear regression, and linear regression
applied to normalized monthly values. One
advantage of the non-parametric approach is
the ease in which censored data can be
analyzed.
te 0
The Kendall's tau non-parametric te
for annual trends is adjusted to deal
with seasonal data by summing 12
monthly Kendall's tau statistics. A
related non-parametric slope estimat
is also derived.
c~
60 ,-— ° v>
.5 z s §
1 efrilo'So'l
% 1 =S 8 1 % -2 g \
z55cx
"«
cr
u
0
CO
£
f>
The enhanced robustness of the test to cases
of serial dependence from month to month
implies a lowering of the power for the
independent case.
c
The seasonal Kendall's tau statistic
developed in paper 13 is adjusted to
include possible correlations betwee
the 12 monthly values.
c~-
3 »
W C
rt O
U'I
O cd
§1
sl
a
•€
^
•a
O"
cd
£
•
y ^.
The mean second highest daily
maximum is analyzed by a repeated
measures one-way analysis of variai
(same as paper 1). The mean
exceedance numbers are analyzed b
fitting a Poisson distribution.
w>
« <8
§2
5s
!Z T3
11
•g >, 8 "3
S 'a - u
§ -5 6 |
w £ "^
B "bh a o
^ .?" 3 x
^ j= E o
v>
-------
I
o
U
(U
I
a>
"e
o
u
Statistical methods
|M
cd
>
•I
&
o
Ul
3
V)
£j<
i/i
o
S-S -8
Trend estimation for autocorrelated annual
monthly ozone values has usually been bas
on fitting a stationary time series model to
the detrended data. The approach develop
in this paper is based on separating the
significant frequencies in the temporal and
spatial series.
"a *
The methods used assume both spati
and temporal correlations as well as
time trend. The analysis is in the
frequency domain.
^f Q\
\Q f^
C^ ^^
§
tt
II
B
•Si
i-o g
3 I'V o -P
Hill
s
t/3
S
i
•a
•o
i
s
"o
li
£ 1
o —
Monthly values are assumed to be t\
sum of a monthly linear trend, a sur
sinusoidal curves representing seasoi
(within year) effects, and an
autoregressive error term.
OO 00
^O t^1
2 2
B
,O w
*o 3
to
B
•8 j
>,j§ Q § X
Hill
S~ H o w
M O *5 £-
E
o B
Issues discussed: representativeness and
weighting of site statistics, effects of
measurement protocol changes, biases due
intermittent sampling, need for
meteorological and emissions data in order
extrapolate trends. The chi-square test is
simple to perform but has very low trend
detection power.
« C T3
Techniques suggested are Spearman
rho, the Pearson correlation coeffici
(equivalent to a linear regression tre
test), a chi-square test based on
exceedance rates at two consecutive
time points.
o —
ON ON
B N
> <
, §
B "8
Using maximum likelihood estimatic
the limiting extreme value joint
distribution for the k highest values
assuming a linear yearly trend is fitt
t^ 00
g
3
O
f
'S •
tl
2
•o
The daily exceedance rate was assumed to
depend on meteorological measurements ar
on the long term linear yearly trend.
CS O
te 15
Exceedances are assumed to occur a
non-homogeneous Poisson process.
Thus the instantaneous exceedance r
does not depend on the past history.
— t-
OO 00
o\ o\
t
•i
§
rt
c
"S
Q
S
-------
9
O
U
Comments [|
w
1
1
s
'•a
1
e
1
B
O
'5b
U
o
'is
•B
3
en
i
co
o
Z
8.
o
o
£
to i-
II
il
11
« -2
3 (fl
O ' M
W £2 1)
O
i !
"w S
fij C3
*3 O O
S ^ c
2 j- o
31 =
13 # S
t -p *
O u ^
o > "
• £ 4)
« £
•*•* *o n* ^
£ -| ||
"1 ? ° «
* j> §• £
list
!
1
8,
1
1
1
£ E
rt O D
^ -B ^
s | s. |
§ 1 * t!
1 ll 1
c^- ri
t- 00
d
5
"3
Z
"S >> ^» f<
B ^ O >-O
0 ^ CO J.J C/3
8 *^ u *o t>
» W) p 60
S Ji ^ S S
a
The non-parametric methods in papers 12,
13, and 21 were often applied in the earlier
reports (1970s). Later reports used the
parametric methods in papers 1, 15, and 22.
One advantage of the ANOVA approaches is
that the year to year effects are not always
assumed to follow a linear trend.
<4—
O
•"• f_ 3 re <£
"S H __ M «
S' f ^ rv Cd
w ^5 3 ° >
d a '> g 's S
o -a „ S * -s
'Si « 0 Jr i •-
w •" c ,O w o
o « 53 <*- . - g
fck-o -C « ^ J-
»i> c cd (5 c *^
w £ ^ s o g
.a ~ *o « .si c
« 1 .2 -s t£ 8 „•
s 1 -^ g « | |
O. •- es "2 «*(£"§
£ « ?^ c D £ o
>$ B S 0 > .2 ><
Z .5 js o 5 oa «
i .
^1
Q
«t-J
Hi
z
& *
ii <*. 'S ft** R ^
Ills o""^
1 8 1" | S^
1 a Il2i
en
-------
APPENDIX I
Summary of Oral and Written Comments from the
Public Meeting on the Ozone Design Value Study,
Arlington, Virginia, September 10, 1992
94069.1 1-1
-------
LIST OF PANEL MEMBERS
William Hunt, Branch Chief
U.S. Environmental Protection Agency
Office of Air Quality Planning and Standards
Technical Support Division
Monitoring and Reports Branch (MD-14)
Research Triangle Park, NC 27711
919/541-5559
Warren P. Freas, Senior Statistician
U.S. Environmental Protection Agency
Office of Air Quality Planning and Standards
Technical Support Division
Monitoring and Reports Branch (MD-14)
Research Triangle Park, NC 27711
919/541-5469
Tom Helms, Branch Chief
U.S. Environmental Protection Agency
Office of Air Quality Planning and Standards
Air Quality Management Division
Ozone/Carbon Monoxide Programs Branch (MD-15)
Research Triangle Park, NC 27711
919/541-5527
Valerie Broadwell, Policy Analyst
U.S. Environmental Protection
Office of Air Quality Planning and Standards
Air Quality Management Division
Ozone/Carbon Monoxide Programs Branch (MD-15)
Research Triangle Park, NC 27711
919/541-3310
LIST OF SPEAKERS
Jon Heuss
Motor Vehicle Manufacturers Association
of the United States, Inc.
David P. Chock
Ford Motor Company
Dearborn, Michigan
94069.1 1-2
-------
LIST OF ATTENDEES
Tim Hogan
American Petroleum Institute
1220LSt.,NW
Washington, DC 20025
202/682-8323
Jon Heuss
General Motors/Motor Vehicle
Manufacturers Association
30400 Mound Road
Warren, MI 48090-9015
Hether Macfarlene
Hunter & Williams
2000 Pennsylvania Ave., NW
Washington, DC 20086
202/995-1500
George Lauer
Atlantic Richfield
515 S. Flower
Los Angeles, CA 90071
213/486-0777
Jon Gallinger
American Gas Association
1515 Wilson Blvd.
Arlington, VA 22209
703/787-1735
David P. Chock
Ford Motor Company
Dearborn, MI 48121
313/845-4777
Dirk Herkhof
US Department of Interior
Minerals Management Service
381 Elden Street
Herndon, VA 22071
Susie Brunige
BNA
Washington, DC
202/452-5558
Phyllis Hinterer
Interstate Natural Gas Association
555 13th Street, NW
Washington, DC 20004
202/626-3200
Subroto Mitro
US Navy, Bldg. 212
Washington Naval Yard
Washington, DC 20374
202/433-3777
Victoria Schobel
Clean Air Report
1225 Jefferson Davis Highway
Arlington, VA 22202
703/892-8516
Rob Smith
US Navy, Bldg 212
Washington Naval Yard
Washington, DC 20374
202/433-3777
Amy Steeley
Chemical Manufacturers Association
2501 M Street, NW
Washington, DC 20037
202/887-1357
Kent Avery
US Navy
Naval Air Systems
Command Air-0943G
Washington, DC 20361
703/692-8562, ext. 2313
94069.1
1-3
-------
LIST OF COMMENTERS
92-4 American Petroleum Institute
92-5 Connecticut Department of Environmental Protection
92-6 U.S. Department of the Interior
92-7 Testimony: Motor Vehicle Manufacturer's Association
(MVMA) - Jon Heuss, General Motors Corporation (GM)
92-8 MVMA-Ford (Dr. David Chock - JAWMA - 1991)
92-2 MVMA-Ford (J. Heuss and D. Chock - AWMA Tropospheric Ozone Specialty
Conference, Atlanta, GA, 1991)
92-10 MVMA-Ford (Dr. David Chock - Atmospheric Environment - 1984)
92-11 MVMA-Ford (Dr. David Chock - JAPCA - 1989)
92-12 MVMA-GM (Jon Heuss - AWMA Annual Meeting, Vancouver, BC, 1991)
92-13 Testimony: Ford Motor Company (Chock, Ford)
92-14 Illinois Environmental Protection Agency
92-15 San Diego County, CA Air Pollution Control Board
92-16 Zephyr Consulting
92-17 General Motors - Jon Heuss (addenda to testimony)
92-18 New York State Department of Environmental Connversation
92-19 California Air Resources Board
94069.1 1-4
-------
#92-4
SYNOPSIS
AMERICAN PETROLEUM INSTITUTE COMMENTS
"Comments on the Proposed Ambient Air Quality Surveillance Rule, 57 FR 7687
(March 4, 1992)" and resubmitted by API as public input for consideration within the context
of the Ozone Design Value Study, August 12, 1992.
API expressed concern about bias and interferences in the ozone measurements. API
notes there are interference issues with both of the predominate instrument types comprising
the ozone monitoring networks, namely the Federal Reference Method ethylene
chemiluminescence (CL) instrument and the equivalent ultraviolet absorption (UV)
instrument. API stated that the potential problems relate to specific conditions of high
temperature, humidity and ozone precursor concentrations which are especially present in
design value situations. Although some of the bias may be in either direction, API feels that
most of the factors will result in a positive bias.
API also expressed concern about the robustness of the current design value method.
API expressed concern that the current methodology utilizes a small number of high ozone
measurements and encourages the development of a more robust technique, which will
involve a larger portion of the available ozone data base.
API expressed concerns about the adequacy of the ozone monitoring network.
A region by region review of the current monitoring network size, location,
efficiency, and effectiveness is suggested. Flexibility is encouraged in network design to
take advantage of local geographical and terrain characteristics, frequencies of wind roses,
and current understanding ozone formation mechanisms in the area. API suggested that
ozone monitoring might be more efficient and cost-effective if it were more intense in the
high season.
94069.1 1-5
-------
#92-5
SYNOPSIS
CONNECTICUT DEPARTMENT OF ENVIRONMENTAL PROTECTION
COMMENTS
Written comments submitted by Kevin McCarthy, Assistant Director, Radiation and
Monitoring Division, Bureau of Air Management, Department of Environmental Protection,
State of Connecticut, September 9, 1992.
CT DEP has concern about bias and interferences in the ozone measurements.
Connecticut previously submitted a report to EPA describing unstable response for the
ultraviolet absorption ozone analyzers, due to presence of water vapor. There was, and still
is, concern about potential bias in the ozone data. Also, the State of Virginia has observed
similar instability problems causing the invalidation of a significant portion of its 1991 ozone
data. Other published reports indicate the ultraviolet ozone analyzers can be biased on the
high side due to organic compounds. A summary of the potential interference issue for
ozone monitors has been prepared by the American Petroleum Institute for the enhanced
ozone monitoring rules and has been resubmitted for the ozone design value study.
CT DEP has concern about the robustness of the current design value method.
Connecticut encourages EPA to consider developing a method which relies less heavily on
relatively few highest hourly values and which will utilize more data points from other than
the extreme tail of the ozone distribution.
94069.1 1-6
-------
#92-6
SYNOPSIS
U.S. DEPARTMENT OF THE INTERIOR COMMENTS
Written comments submitted by John Goll, Chief, Environmental Policy and Programs
Division, Minerals Management Service, U.S. Department of the Interior, Washington, DC,
September 9, 1992.
The DOI has concern with small-area resolution of the Regional Oxidant Model
(ROM). DOI staff expressed disagreement with the EPA interpretation that ozone isopleths
predicted by the ROM Model for the Northeastern United States appear uniform across a
broad geographic region. DOI believes the ROM model is not adequate to characterize the
more detailed spatial patterns found within individual metropolitan areas.
The DOI also expressed concern about the adequacy of the ozone monitoring
network. DOI emphasized the importance of having a monitoring network with sufficient
size and density to account for ozone variations which can occur over small distances within
metropolitan areas, particularly in areas of complex terrain. Increased consideration of
network design issues is encouraged.
DOI stated that there is a need for a method to account for temporal variation in
meteorology. Recognition that weather conditions conducive to ozone formation can differ
greatly year-to-year, although "high" ozone seasons may recur with an estimateable
periodicity, suggests that the design value methodology should be able to adjust in some way
for this annual variation. DOI expressed concern that, for a 3-year period of especially good
weather for preventing ozone formation, the present form of the design value strategy may
not be protective enough.
DOI raised concerns about ozone transport from an adjacent region. DOI points out
the issue where ozone or precursors are transported into an air quality control region, such
that the region may not be able to demonstrate attainment no matter how rigorous the
proposed emission control. DOI stated that some modified approach is necessary, such as
consideration of a larger regional effort to design control strategies.
94069.1 1-7
-------
#92-7
SYNOPSIS
MOTOR VEHICLE MANUFACTURERS ASSOCIATION STATEMENT
"Statement of the Motor Vehicle Manufacturers Association of the United States, Inc.
on the Ozone Design Value Study of the Clean Air Act Amendments of 1990," Jon Heuss,
Motor Vehicles Manufacturers Association, September 10, 1992.
MVMA expressed general concern about the need for increased robustness in the
design value, inconsistency between the nominal and actual stringencies of the standard,
requiring a site to average significantly fewer than one exceedance per year to have a
reasonably high chance of being designated an attainment site, the need to adjust the current
design value method to account for annual fluctuations in meteorology, and the implicit time-
lag for providing evidence of emissions improvements for moving toward attainment (i.e.,
due to the 3-year average criterion, the current procedure begins counting exceedances 3
years before the applicable deadline), which appears inequitable.
The seven criteria proposed by Curran for evaluation of "reasonableness" are
discussed by MVMA and endorsed as necessary attributes. Also, the six properties listed by
Fairley and Blanchard as desirable are described. MVMA criticized the current design value
procedures as being deficient in several of these including stability, sensitivity to changes in
air quality, use of available data, and consistent nominal and acaial stringency. Heuss
suggests several modifications which might improve the method such as using the 95th
percentile or a 95th percentile-centered mean, as the mean of the highest 30 daily
concentrations instead of the presently used design values in order to increase robustness.
MVMA also expressed concern about the outputs of the deterministic models used to
evaluate the effectiveness of control strategies. MVMA stated that there is a need to
investigate and understand them in probabilistic form so that informed decisions can be made
about the efficacy of selected emissions control scenarios. Heuss notes an inconsistency
between the planning tool used to aim at the attainment target, photochemical grid modeling,
and the target, the present design value.
These comments are supplemented by the concurrent submission of related reports
and technical papers previously prepared by MVMA member company scientists.
94069.1 1-8
-------
#92-8
SYNOPSIS
FORD MOTOR COMPANY TECHNICAL PAPER
"Issues Regarding the Ozone Air Quality Standards," Dr. David P. Chock, Ford
Motor Company, Journal of the Air and Waste Management Association, Vol. 41, No. 2,
February, 1991.
The rationale for this report is to develop a statistically-based form of the ozone air
quality standard that is based on an order statistic, such as a percentile, and to develop a
robust compliance test that reduces the impact of unusual meteorological situations (such as
the 1988 heat wave) on the attainment status of a region. A two-component testing
procedure is proposed. The first component is a statistical test of an area for attainment such
as the t-test with a relatively large a level, which is the probability of misclassifying an area
actually in attainment as nonattainment. The second component would be used for those
areas which have been previously tested and determined to be in attainment for some time
and would consist of the same t-test with a smaller a level. Such areas, to be eligible for
this less stringent test, would also have to demonstrate that no significant increases in ozone-
precursor emissions had occurred. Chock states that this approach would have the benefit of
reducing the impact of unusual weather fluctuations such as the 1988 heat wave on the
attainment status of long-time attainment areas while not protecting their ranking category
from an upward ozone trend. This approach is similar to that proposed by Fairley and
Blanchard. It requires compliance with the current methodology a high percentage of the
time (about 95%), rather than the current procedure of 100% compliance for the attainment
category.
94069.1 1-9
-------
#92-9
SYNOPSIS
TECHNICAL PAPER
"The Role of the Design Value in Ozone Compliance" Jon Heuss and David Chock,
presented at the Air and Waste Management Association International Specialty Conference,
"Tropospheric Ozone and the Environment, n: Effects, Modeling, and Control," Atlanta,
GA, Novembers-?, 1991.
This report provides a general discussion of the many issues surrounding the
application of the Design Value Methodology in Ozone Compliance. It notes the present
limitations of the procedure such as lack of stability resulting in frequent changes in
attainment status, non-robustness since only 4 hourly concentrations are considered from a 3-
year data base, sensitivity to air quality changes, and inconsistency between the nominal and
actual stringencies. No single solution is advocated. The several positive suggestions by
Chock, Heuss, Fairley and Blanchard, Curran and Freas are cited and should be further
evaluated. Several desirable properties of the improved procedures are suggested, however,
for example, the use of a more robust statistic such as the 95th percentile, which are
consistent with other reports by the authors.
94069.1 1-10
-------
#92-10
SYNOPSIS
TECHNICAL PAPER
"Statistics of Extreme Values of a First-Order Markov Normal Process: An Exact
Result," Dr. David P. Chock, General Motors Research Laboratories, Atmospheric
Environment, Vol. 18, No. 11, pp2461-2470, 1984.
This highly mathematical report examines the impact of autocorrelation on extreme
values. This paper derives the exact distribution for the maximum, second-highest, third-
highest, and fourth-highest values of a first-order Markov normal process. It is found that a
high positive autocorrelation (> 0.7 at log 1) lowers the values of the extremes and near-
extremes while increasing their variability. The means, medians, and modes together with
their deviations, for several autocorrelation values and several numbers of observations are
presented. As discussed in Section 5, a later paper by Hirtzel and Chock (1985) describes
some mathematical errors in these calculations because of inconsistencies in the assumptions
made. The relevance of these results to the ozone design value issues is discussed. Concern
is expressed that a few high ozone days from a single year or even a single season may
dominate the design value procedure for a 3-year period. The author states that the potential
efficiency and cost-saving for monitoring more or only in the high ozone season should be
further considered. The need to understand and adjust for multiple meteorological variables
is noted.
94069.1 I-11
-------
#92-11
SYNOPSIS
TECHNICAL PAPER
"The Need for a More Robust Ozone Air Quality Standard," Dr. David P.
Chock, General Motors Research Laboratories, Journal of the Air Pollution Control
Association, Vol. 39, No. 8, pp. 1063-1072, August, 1989.
This report notes limitations in the current ozone air quality standard such as the
instability in the annual second-highest value and the fourth highest value in a 3-year period,
an implicit compliance test that is not capable of accounting for the fluctuation in ozone
concentrations due to severe meteorology conditions, and inconsistencies between the
number-of-exceedances criterion and the design value for control implementation. A more
robust statistic such as the annual 95th percentile is proposed for this standard. This measure
is more stable than the extreme values used at present and yet is still characteristic of
repeated high ozone exposures. The author suggests that this statistic can serve both as the
compliance criterion and as the design value, eliminating the use of the number of
exceedances in the standard. Also, Chock states that the proposed t-test for compliance
allows the standard to be enforced simply and without ambiguity. He notes that by
determining the mean and standard deviation of the 95th percentile concentrations over 3 or
more years in the compliance test, the average concept of the current standard is preserved
while accounting for the stochastic nature of ozone concentrations.
94069.1 1-12
-------
#92-12
SYNOPSIS
TECHNICAL PAPER
"Implications of Attaining and Maintaining an Extreme Value Ozone Air Quality
Standard," Jon M. Heuss, presented at the 84th Annual Meeting of the Air and Waste
Management Association, June 16-21, 1991, Vancouver, British Columbia, Canada.
This report expresses concern about whether the current design value methodology
provides a reasonable indicator of air quality in nonattainment areas and hence is an effective
target for the development and assessment of local emissions control strategies.
Heuss states that one problem with the procedure to utilize photochemical grid models
to demonstrate attainment is that the required detailed input information for emissions, initial
conditions, boundary conditions, and meteorology may not be available, particularly for a
specific design value day in the past. Heuss notes that guidance has not been fully developed
for this situation. He states that this question was not an issue with the Empirical Kinetic
Modeling Approach (EKMA) that was previously used to model ozone for SIPs since EKMA
is a box model that does not require detailed input data. Updated emission inventories are
needed. Heuss suggests that consideration of statistical estimation techniques would be useful
to predict future ozone concentrations, even in the case where many data values are missing.
A second issue raised in the paper is a test for maintaining the standard for those
areas currently in attainment. The author states that perhaps a less stringent compliance test
would avoid the oscillations caused by fluctuations in meteorology and still maintain
reasonable detection of increasing ozone precursor emissions. To improve the present
situation, a more robust statistic is suggested, which provides greater stability as a design
value and, then combined with a compliance test, recognizes that ozone concentrations are
probabilistic in nature.
94069.1 1-13
-------
#92-13
SYNOPSIS
FORD MOTOR COMPANY STATEMENT
Statement of the Ford Motor Company on the Ozone Design Value Study of the
Clean Air Act Amendments of 1990, Dr. David Chock, Ford Motor Company, September
10, 1992.
Ford Motor Company has concerns about the current compliance test for the ozone
standard. Chock expressed particular concern about the compliance test portion of the design
value issue. At present, the compliance test requires complete (100 percent) adherence for
all time periods. Chock stated that this feature has resulted in the standard being more
stringent than is generally recognized, since an area with a long-term average of one
exceedance per year will have a 35 percent chance of having more than three exceedances in
3 years. He also noted that the present method results in instability problems as regional
attainment status can oscillate due to the presence of unusual meteorology conditions. Chock
suggests developing a statistical compliance test that takes into account the fluctuation of the
ozone concentrations. He suggests using the second highest of the three annual second
highest values in three consecutive years as the practical design value. Chock stated that this
form of design value would increase stability across years and use a more robust procedure
that adjusts for a particular year of unusually severe meteorological conditions.
94069.1 1-14
-------
#92-14
SYNOPSIS
ILLINOIS ENVIRONMENTAL PROTECTION AGENCY COMMENTS
Written comments submitted by Bharat Mathur, Chief, Bureau of Air, Environmental
Protection Agency, State of Illinois, September 10, 1992.
Illinois EPA stated that there is a need for a method to account for temporal variation
in meteorology. Illinois observes that the design value methodology needs to adjust
somehow for the impact of extreme meteorological conditions in a given year. Such an
adjustment is needed to provide increased stability so that an air quality region will not
fluctuate between attainment and non-attainment. They suggest two possible approaches,
which limit the contribution of a single year's data to the 3-year total, as examples of the
type of "formalizing" adjustment they believe is needed.
Illinois EPA has concern for the necessary time-lag in demonstrating the impact of
emissions control measures on ozone air quality for determining attainment. Illinois observes
that the requirement to use the most recent 3-year quality assured ozone data base to
determine attainment may be inequitable in some cases because of the time-lag in observing
the benefit of newly instituted control measures. They suggest that if a State can justify
when a 3-year period is not representative for a metropolitan area, the method allow some
flexibility by permitting consideration of some alternative time-table, such as a year-by-year
basis, in order to appropriately consider emission controls implemented more recently than 3-
4 years previously.
Illinois also expressed concerns about the adequacy of the ozone monitoring network.
Illinois observes the importance of having a monitoring network which is sufficiently large to
represent the spatial area.
Illinois suggests that the design value protocol be modified to incorporate the results
of urban airshed modeling which will be available in 1994 and will indicate whether the SIP
revisions will provide attainment by the applicable attainment projected date.
94069.1 1-15
-------
#92-15
SYNOPSIS
SAN DIEGO COUNTY, CA COMMENTS
Written comments submitted by R. J. Sommerville, Air Pollution Control Officer, Air
Pollution Control District, County of San Diego, September 14, 1992.
The San Diego APCD expressed concern about how ozone transport from an adjacent
region is treated. San Diego County, being located next to the South Coast Air Basin, an
area with the highest U.S. ozone levels, is concerned that the design value methodology be
able to determine the correct severity of the ozone problem within San Diego County, so that
the area is not unfairly and inequitably required to address the pollution problem in the
SCAB. The transport issue has been documented and discussed previously in the San Diego
County Air Pollution Control District's 1978 State Implementation Plan, which was approved
by EPA. The San Diego APCD feels that it is not clear how the current ozone design value
methodology will be modified to take into account ozone transport.
94069.1 1-16
-------
#92-16
SYNOPSIS
ZEPHYR CONSULTING COMMENTS
Written comments submitted by Dr. Kay H. Jones, President, Zephyr Consulting,
September 15, 1992.
Dr. Jones raised the issue of the ozone concentration/frequency relationship. He states
that to be technically consistent with the legal attainment target of frequency, i.e., no more
than 1 day per year on the average, the primary design value should be defined as a frequency
(days/year) rather than as a concentration. He also suggests that if a concentration design
value is required for classification reasons, then the frequency/concentration relationship
should be determined on a region by region basis, since, in his view, the
frequency/concentration correlation varies sufficiently between different locations.
Dr. Jones also stated the need for a method to account for temporal variation in
meteorology. He observes that the design value methodology needs to adjust in some manner
for the impact of extreme meteorological conditions in a given year. Summary data for the
annual frequency of daily maximum temperatures exceeding 90° in Baltimore during 1967-
1992 are presented as an illustration of a possible approach for an adjustment procedure.
Additionally, other meteorological factors besides temperatures, such as humidity.
rainfall, wind speed, and direction, should also be considered. Dr. Jones recommends that
when evaluating data for appropriate meteorological adjustment, the methodology should
require identification and use of a single (worst-case) ozone monitoring site location. He
states that in the past, some air quality regions have (inappropriately) used different monitor
locations for SIPS planning for different years.
Dr. Jones commented on ozone transport from an adjacent region. He states that areas
such as rural counties downwind from major urban areas should not be evaluated as unique
impacted regions. He suggests that some type of adjustment for transport seems appropriate.
Dr. Jones expressed concerns about the missing data procedures used to calculate
estimated exceedances of the ozone standard. He expressed concern that the current missing
data procedures do not adequately consider seasonality patterns to equitably estimate missing
data for different ozone "seasons." He suggested that improved analysis and understanding of
the temperature/ozone relationship is needed for the proper prediction of expected
exceedances. Dr. Jones states that this comment is consistent with
94069.1 1-17
-------
#92-16
recommendations by the National Academy of Sciences report "Rethinking the Ozone
Problem in Urban and Regional Air Pollution" and the 1987/88 CEQ Annual Report Chapter
on Urban Air Quality.
94069.1 1-18
-------
#92-17
SYNOPSIS
MVMA / GENERAL MOTORS ADDITIONAL COMMENTS
Written comments submitted by Jon Heuss, General Motors Research and
Environmental Staff, September 15, 1992.
In an additional comment submitted following his presentation, Heuss notes an
example which he believes illustrates an inconsistency between the current ozone design
value methodology and the current modeling guidance that requires all modeled episodes to
be below the ozone standard. He feels that EPA guidance may provide an emission control
target that is more stringent than the design value resulting in over control. Heuss also notes
that the Regional oxidant Model (ROM) may over predict peak ozone concentrations which
may also result in an over control situation, without some adjustment.
94069.1 1-19
-------
#92-18
SYNOPSIS
NEW YORK STATE DEPARTMENT OF ENVIRONMENTAL CONSERVATION
COMMENTS
Written comments submitted by Edward O. Sullivan, Deputy Commissioner, New
York State Department of Environmental Conservation, September 11, 1992.
The NY DEP feels that there is a need for a method to account for temporal variation
in meteorology. New York observes that the year-to-year variation in ozone design values
for the NJ-NY-CT CMS A during 1988-92 is due more to meteorological annual differences
than to precursor emissions changes. New York suggests that the design value methodology
should be able to adjust in some way for meteorological annual fluctuations in order to avoid
misclassification of areas as to compliance or non-compliance category.
NY DEP also expressed concern about the robustness of the current design value
method. New York encourages EPA to explore the development of a more robust statistic,
such as a specified quantile of the daily maximum 1-hour ozone concentrations in a 3-year
period, as a means of classifying non-attainment areas and specifying their emissions
reduction requirements. New York recommends using this new statistic, the Control
Strategy Design Value, to identify control requirements.
NY DEP expressed concern about the interpretation 2of non-urban ozone monitoring
site data. New York observes that there exist some air quality regions where the only high
ozone data contributing to the non-attainment classification are from non-urban sites, such as
a mountaintop location, where the contribution of local anthropogenic emissions is negligible.
94069.1 1-20
-------
#92-19
SYNOPSIS
CALIFORNIA AIR RESOURCES BOARD
Written comments submitted by Terry McGuire, Chief, Technical Support Division,
Air Resources Board, State of California, August 25, 1992.
The California Air Resources Board (CARB) stated that there is a need for a method
to account for temporal variation in meteorology. CARB observes that the design value
methodology could be improved by an adjustment that provides greater stability by
accounting for annual meteorological fluctuations. CARB suggests a possible approach, the
"recurrence rate" method, which can be set to varying levels of stringency. That is, 0, 1, or
more observed high observations could be excluded per year based on the meteorology
observed. A 1-in-l year "recurrence rate" is similar to the stringency in the current EPA
design value attainment method. CARB states that in these years with extreme
meteorological conditions, a different "recurrence rate" could be used to "normalize" such
time periods. In years of favorable meteorology (preventing high ozone formation), it is
possible no high values would be excluded.
94069.1 1-21
-------
APPENDIX J
Summary of Written Public and Peer Review Comments
on the Draft Ozone Design Value Study
94069. J J-l
-------
SUMMARY OF COMMENTS
This summary encompasses all substantive comments, excluding those on minor stylistic and
editorial matters. For clarity in references to lengthy comments, paragraph numbers are
given in brackets after each comment.
AMERICAN PETROLEUM INSTITUTE
API incorporates comments by Chevron of 27 April 1994, by API of 30 September 1986,
and by API of 17 March 1989.
1 As the current design value methodology does not characterize ozone air
quality temporally or spatially throughout nonattainment areas, it does not
answer the 1990 CAAA's requirement in Section 183(g) and should be
replaced with a more suitable indicator.
2 The present indicators for ozone—design value and maximum one-hour
concentrations—do not represent actual exposures of people, thus a new air
quality indicator should be used which incorporates population exposure levels
and activity scenarios in addition to other factors that affect exposure levels.
3 Because a very small number of extreme measurements, possibly highly
localized or due to natural fluctuations, can result in nonattainment status, the
EPA should develop an air quality indicator for the NAAQS review that makes
use of the entire body of monitored data throughout an MSA/CMSA and
focuses on the influence of anthropogenic emissions.
4 Because the air quality indicators, present and alternative, are unstable and
could bounce a region between attainment and nonattainment despite a long-
term trend toward attainment, the EPA should establish a "transitional
nonattainment" designation for areas that are near attainment.
94069. J J-2
-------
CALIFORNIA AIR RESOURCES BOARD
1 The CARB recommends use of the Larsen tail exponential method for
attainment designations and design value calculations.
2 Clarifications submitted for the discussion of the development of the CARB
design value method [Corrections 1,2].
3 Amplification suggested for the treatment of the exponential-tail assumption
[Suggestion 1].
4 Suggest combining findings number 2 and 6 on page 1-16 and clarifying
[Suggestion 2].
5 While many common air quality indicators are simple to compute, they are
complex to interpret and can give misleading results, whereas the CARB
method, although more complex to compute, is simple to interpret properly,
even for nontechnical people, and thus may serve the public better
[Suggestion 3].
94069.J J-3
-------
CHEVRON
1 The study should review, discuss, and support air quality indicators other than
the design value, especially ones that provide information on potential health
impacts.
2 The EPA should adopt the Larsen methodology used by the CARB to
determine the design value, which removes highly irregular or infrequent
meteorological events.
3 The EPA should identify indicators of air quality improvement besides the
design value because when used alone, the design value (1) overstates severity
of population ozone exposures by applying ozone levels at the design value
monitor to an entire region, (2) does not indicate improvement in population
exposures due to improvements in peak ozone, and (3) does not show how
efficiently (in terms of reducing health risk) progress in air quality is made.
4 Indicators that consider population exposure should be part of the air quality
planning process to assess the effectiveness of ways to reach ozone attainment
through NOX and hydrocarbon emission reductions.
5 The EPA should explore methods of determining attainment that avoid
weaknesses of the present method based on the design value, which relies on a
single monitor that is not weighted according to population and does not take
data from other monitors into account.
94069.J J-4
-------
DUPONT
1 As the draft report points out, the current design value cannot account for
spatial and temporal variations in ozone concentration that are necessary for
evaluating human exposures, so the section on exposure measures should be
expanded and the usefulness of exposure measures emphasized [2-3].
2 Because emissions across the United States have been decreasing, the use of
long-term averaging periods is a conservative procedure, and the report should
be amended to advocate a design value estimation method that uses averaging
periods of six years or longer [4-7].
3 Just as the EPA is studying techniques to adjust ozone trends for
meteorological variability, it should look for similar techniques to adjust the
design value [8].
4 Contrary to the report's assertion, the Clean Air Act's Section 110(k)(6)
provides for adjustments to initial ozone classifications due to transport [9].
5 The report should be amended to include a position on the use of adjusted
design values for ozone classification and attainment demonstration,
recognizing that because they currently cannot be adjusted they are
incompatible with "control strategy design values" used in developing emission
control plans, which can be adjusted [10-11].
6 The report should include a more detailed discussion of the EPA's
interpretation of CAA Section 181(b), which uses the words "design value"
whereas the EPA interprets it to mean "average number of exceedances per
year" [12].
Lesser comments [13]:
7 Request background on the basis for the breakpoints in table 1-1.
8 The report should note on page 1-4 that placement of ozone monitors is an
additional key concern.
9 Page 3-10 should describe the requirements for density and siting of ozone
monitors.
10 Supply a source for the assertion on page 6-30 that the NAAQS is designed to
protect against effects of short-term exposure.
11 Explain why design value estimates shown in figure 6-13 rise as averaging
time increases.
94069.J J-5
-------
FORD MOTOR COMPANY
1 The study considers only the reasonableness of the present design value
methodology and does not consider alternatives; indeed, alternatives are
deferred to the next ozone NAAQS review yet the document on which that
NAAQS will be based (the Criteria Document) does not address alternatives
[2].
2 The EPA's definition of the design value as the "characteristic largest value"
or CLV corresponding to the NAAQS (exceeded once per year on average) is
inconsistent with that of the present EPA design value (fourth highest in three
years) [3].
3 This inconsistency undermines the quality of the design value methodology and
the logic underlying comparisons of different methodologies [4].
Jtofrilf~iic,> t£ IL-
4 The CLV is determined with the assumption X)f infinite sample size, which is
inapplicable to the problem of assessing the^rank order statistic of a finite-size
sample [5].
5 The CLV cannot be determined without making assumptions such as those
made in Section 6 that cannot be rigorously justified [5].
6 The report should note that 95th percentiles are more robust statistics than the
present design value, as supported by Figures 3-3 through 3-14 and Table 6-9
[6].
7 Exceptional years such as 1988 tend to dominate the design value for the
three-year periods they belong to under the present methodology, causing large
variabilities between non-overlapping three-year periods, whereas alternative
design values could avoid this problem [7-8].
8 The comparison of design values and 95th percentiles in the report is not valid
due to discrepancies in sample sizes, treating combined instead of individual
sites, misinterpretation of correlation diagrams, and misattribution of statistical
bias [7-13].
9 The compliance test should take the variability of the test statistic into account
[14].
10 The report should discuss the alternative approach of using the three-year
mean or median of the annual second highest values [15].
94069. J J-6
-------
11 Extreme values are unduly affected by the distributions chosen, such as
lognormal; a "calibration factor" used by the CARB corrects this to an extent,
but the factor is site- and time-specific [16-17].
12 Increasing the multiyear sampling period would increase the design value and
the NAAQS stringency [18].
13 Because design values based on multiyear sampling periods become unduly
influenced by long-term ozone concentration trends, meteorologically-adjusted
ozone trends should be determined by examining ozone concentrations under a
fixed set of meteorological conditions for different years, with the
ozone-meteorology relation determined separately for each year (in contrast to
the Cox and Chu method) [19-20].
14 Adjustment of design values due to transport of ozone and precursors into and
out of an area should be done with a photochemical grid model, not the
Transported Ozone Design Value model [21].
15 Given the poor correlation between the EPA design value and the expected
number of exeedances, the upcoming revision of the NAAQS should use
rational, consistent indicators to represent the NAAQS and design value [22].
16 Even though there are more robust alternatives to the EPA design value,
compliance flip-flops will remain a problem that is best addressed by a
statistical compliance test incorporating design value fluctuations, in order to
balance public health needs against inevitable fluctuations in natural conditions
[23].
17 As the ideal design value for purposes of public health should take into
account human exposures to ozone, indoor as well as outdoor monitors should
be part of the NAAQS specifications [24].
94069. J J-7
-------
GENERAL MOTORS
General Comments
1 The roles of the design value in attainment classification, attainment
designations, and the SIP process have changed under the 1990 CAAA
Amendments and are a fit subject for review in the Design Value Study even if
the EPA wishes to replace design value with number of exceedances as an
attainment criterion in section 181(b)(2)(A) of the Act [9-12].
2 The study report should review the method and data used to arrive at the 1990
nonattainment area classifications and technically justifiable modifications
should be evaluated [14].
3 The EPA should revisit the CAAA deadlines for attainment after the 1994 SIPs
are submitted [15].
4 Since the current EPA design value is a poor measure of health impacts and
temporal trends, air quality indicators (such as the 95th percentile or mean of
the top ten) that are better correlated with number of exceedances than the
current design value should be used [16].
5 The EPA should continue to study more robust indicators of ozone levels that
are less sensitive to fluctuations in meteorological conditions and that better
reflect changes in emissions [17].
6 The three pages of discussion of the issues raised in public review of Chapter
11 do not represent a sufficient response to the comments received [19].
7 The study acknowledges a discrepancy between the nominal stringency of the
present attainment test (one exceedance per year) and its actual stringency
(never more than three exceedances per three years, hence significantly less
than one exceedance per year), but does not address the full ramifications of
this discrepancy [20-22].
8 Given that the EPA is considering the use of statistical attainment tests and the
use of a transitional attainment category in its NAAQS review, the study
should include analyses of these approaches [23-24].
9 The impact of transport of ozone and its precursors into an area should be
discussed in the context of attainment determination as well as classification
determination [25-28].
10 Because the design value methodology weights data disproportionately from
early in the three-year period of record (when extreme values are more likely),
94069.J J-8
-------
the study should analyze a methodology that is based on three one-year periods
rather than one three-year period for use in determining attainment at CAAA-
specified deadlines [29],
11 The study should evaluate ways to adjust the design value and attainment
schedule in light of the statistical discrepancy between nominal and actual
stringencies of the attainment standard and in light of the change in ozone
planning tools from EKMA to photochemical grid modeling [30-34].
12 The study should evaluate the exceedance database with respect to how often
the missing data convention is used and how often it affects attainment
classifications [35-36].
13 Once having made the additional analyses suggested in these comments, the
study will better answer the questions Congress asked regarding uses of design
values [37].
14 The Executive Summary and Conclusions should be revised when the rest of
the report is modified in accordance with these comments [38].
Specific Comments
15 Chapter 2 should acknowledge the difference between nominal and actual
stringencies and should discuss Fairley and Blanchard's six desirable properties
in attainment criteria [39].
16 Chapter 3 discussion of background ozone should be changed to be consistent
with the latest Criteria Document [40].
17 The discussion of human health effects should be scaled back to what is
supported by the data as given in the External Review Criteria Document [41].
18 Discussion of the VOC/NOX ratio at which NOX changes from a net ozone
former to a net ozone destroyer should include model simulations to account
for the complex interactions that affect this ratio [42].
19 Using one of Figures 3-3 to 3-14 in the Executive Summary, rather than
Figure 3-15, would better illustrate the extra margin of safety in the present
ozone standard resulting from the practice of using the largest design value at
any monitoring site to classify an area [43].
20 Chapters 4-6 should give more emphasis to the issues raised by Dr. Chock at
the September 10, 1992 public meeting, relating to deriving design values
from multiple years of data [44].
94069 .J J-9
-------
21 Chapter 4 should reflect the new recurrence rate (every year, not every seven
years) used in CARB's tail exponential method [45].
22 Chapter 7 should investigate analyses that treat each year separately rather than
combining ozone and meteorological data for several years, and should
recognize secondary influences of meteorological parameters on emissions
[46].
23 Chapter 8 should discuss the use of the UAM and coupled urban and regional
models to evaluate transport effects [47].
24 Chapter 9 should discuss the recent work of S. T. Rao [48].
25 Chapter 10 should have more discussion of the data in the tables with respect
to robust indicators [49].
26 Chapter 11 should address issues related to the form of the standard or the
attainment test because these are not treated in the External Review Criteria
Document, where the EPA would prefer they be addressed [50].
94069 .J J-10
-------
DR. KAY H. JONES, ZEPHYR CONSULTING
• The report did not properly acknowledge the early work and analyses by Dr.
Jones of the adjustment of ozone trends to account for year-to-year variation in
in the frequency of ozone conducive days. The adjustment of ozone
exceedances for temperature should be shown.
• The report did not properly reference the earlier work of the Council of
Environmental Quality on the adjustment of ozone trends to account for
temperature that was contained in their annual reports, Environmental Quality
1984 and Environmental Quality 1987-88.
94069.J J-11
-------
MOBIL OIL CORPORATION
1 The EPA design value does not provide a reasonable estimate of "area-wide
nonattainment" since spatial variability in concentrations is not accounted for.
2 Because year-to-year fluctuations cause significant variability in ozone
concentrations, the design value calculation should account for meteorological
variability and thus minimize the effect of aberrant years like 1988.
3 As the EPA recognizes the value of using multiyear data, which are more
stable statistically than one-year values, the design value calculation should use
methods like the CARB 7-year method to remove outliers from the database.
4 As the EPA acknowledges that a tail exponential function is a more robust
statistic of extreme 1-hour ozone concentrations, it should include a
probabilistic (tail exponential) estimation tool or a tune-series tool based on a
characteristic (extreme value) distribution.
5 The current design value's variability adds cost and uncertainty to the task of
planning air quality policy and control technology.
94069 .J J-12
-------
NATIONAL INSTITUTE OF STATISTICAL SCIENCES
General Comments (numbers are paragraph numbers')
• The current procedure is reasonable but improved robustness should be sought
by considering alternatives such as those based on exponential tail assumptions
[1-2].
• A distinction needs to be made between the true design value, expected to be
exceeded once per year, and estimated design values [3-4].
• The design value is not used as the ozone NAAQS compliance test but rather
as a classifying statistic that indicates air quality in nonattainment areas [3].
• The study focuses on the fundamental issue of comparing methods of
estimating the design value but inadequately treats the important issue of how
to take spatial variation into account when using design values at different sites
to assess area air quality [5-9],
• Desirable properties of a design value that are shared by the EPA estimator are
that it is easy to understand, the definition is the same at every site, and that
the general form of the distribution of the (estimated) design value is
statistically well determined (i.e. the general shape is approximately known,
and the parameters that define the specific shape at a given site can be well
estimated) [10-12].
• A desirable property of a design value that may not be properly shared by the
EPA estimator is low variability, which is needed to reduce misclassification
probabilities [10, 12-15].
• Example calculations of misclassification probabilities for attainment/-
nonattainment designations (based on estimated expected exceedances) and for
nonattainment classifications (based on estimated design values) suggest the
need to consider wider ranges for the nonattainment classification definitions
and the use of "too close to call" categories for attainment/nonattainment
designations [13-15].
• Design value estimators based on fitting a parametric distribution to the entire
data set are less desirable because they may tend to be site specific and to give
a relatively poor fit in the tail of the distribution [16].
94069 .J J-13
-------
• The fitting of the tail exponential (or more general tails) should be considered
an important alternative to the EPA procedure, although it is more complex
and may not apply at all sites [17-18].
• Exclusion or downweighting of exceedances based on meteorological
adjustment needs careful scientific justification, which may be obtained with
improved knowledge of the dependence of ozone on meteorology and ozone
precursors [19].
• The study is generally an extremely well considered document [19-20].
Detailed Comments (numbers are referenced pages from the draft document)
• Attainment determinations are based on estimated rather than true design
values [1-2].
• The advantages of reproducibility and uniformity can also apply to other
methods [1-3].
• The advantage that the EPA design value only uses summary statistics is not
important with the use of modern computers [1-3].
• Mention of any disadvantages of the EPA design values should be made here
[1-3].
• Different evaluations of the ability of EPA design values to indicate
nonattainment area ozone air quality are based on the site-specific
misclassification probabilities or on measures of spatial uniformity [1-4, 2-9].
• Using three years of data for design values is more stable than using one year
[1-6].
• The statistical modeling approach design value may correctly indicate
nonattainment when the attainment test wrongly indicates attainment [1-9].
• Comments on alternative forms of the ozone NAAQS may be relevant to the
design value study since a revised ozone NAAQS might better correspond to a
different design value [1-11].
• Finding 4 of high spatial variability suggests the problem that nonattainment is
more likely for areas with more monitoring sites, which may result in
inequities between different areas [1-16, 2-1, 2-10, 2-11, 11-4].
94069 J J-14
-------
• An explanation is needed for the fact that 38 sites met the standard in
1989-1991 but were not redesignated to attainment [1-17].
• The conclusion section should discuss the crucial issues of the performance of
the design value estimates and their ability to accurately classify nonattainment
areas [1-18].
• The design value definition should be kept consistent and explicit throughout
the report as "the concentration with an expected exceedance rate of one per
year," rather than "the value that should be reduced to the standard level
thereby ensuring that the site will meet the standard," which defines the design
value for nonattainment areas only [2-1, 4-1, 6-1, 9-1, etc.].
• Based on a Poisson approximation, an area just in attainment will have two or
more exceedances 26 percent of the time and three or more exceedances 8
percent of the time [2-3].
• If the three-year EPA design value is used it can be less than the NAAQS for
a nonattainment area only if the monitoring record is incomplete [2-3].
• NAAQS attainment/nonattainment criteria are equivalent to comparing the
fourth highest in three years with the standard [2-4].
• The same percentage emission reduction on different days can lead to different
estimated changes in ozone concentrations according to the EKMA [2-4].
• Computer modeling should be adaptable to treat estimated design values that
did not actually occur [2-4].
• The single («+l)th highest value is likely to be more variable than estimates
based on more values or based on fitted distributions, as illustrated in
McCurdy and Atherton (1990) [2-4, 2-10].
• For an area just in attainment, reasonable Poisson approximations lead to a
probability of only 0.23 for having different designations in two consecutive
years [3-19].
• The same Poisson approximation shows that it is much more likely (probability
0.61) that a borderline nonattainment area is misclassified compared to the
0.35 probability of misclassifying a borderline attainment area [3-19].
• The 95th percentile has less variability, but it may not be statistically similar
to the EPA design value in all locations [3-19].
94069. J J-15
-------
• It is not clear why using alternative design values for nonattainment area
designations and classifications would require revision to the NAAQS and the
CAAA [3-19].
• The text wrongly implies that a deterministic relationship to the current
NAAQS design value, rather than just a statistical relationship, is required to
justify the use of the 95th percentile [3-20].
• Use of the 100 x 364/65th percentile as the design value assumes the same
distribution throughout the year [4-3].
• A reference to the "comprehensive" book by Leadbetter et al. (p. 5-22) is
appropriate here [4-3, 4-7].
• The extreme value approximation is often improved using fitted rather than
theoretical constants or by fitting the three-parameter extreme value
distribution rather than the limiting two-parameter distribution [4-4].
• The reviewer agrees with the "growing concensus" in favor of the tail
exponential approach because it requires fewer assumptions than a total
parametric fit, probably gives a less variable design variable, and may be more
universally applicable than other candidate distributions [4-4].
• In some cases even strong serial dependence can still lead to the same form of
the limiting extreme value distribution [4-6, 5-9].
• Use of a lognormal distribution requires that the log transformed
concentrations are regarded as the fundamental measurement, and/or that the
geometric mean is regarded as the fundamental average [4-6, 4-7].
• Clusters of more than one exceedance rarely occur in practice but have been
analyzed in a cited paper [4-8].
• More discussion of the conditional probability approach is suggested [4-1,
4-9].
• Robustness seems to be defined inconsistently as either (a) the property of not
being overly dependent on distributional assumptions, (b) the use of a less
variable statistic, or (c) the property that the distribution does not vary
significantly across sites [4-11, 6-22, 11-1, etc.].
• Warnings should be given that some of the cited references (e.g., references
[7] and [8]) erroneously assume that if the daily maximum concentrations are
Markov (have a conditional distribution that depends only on the most recently
94069 .J J-16
-------
observed concentration), then the binary exceedances form a Markov chain [5-
8,1-11].
• The missing value procedures used in Davison and HemphilTs paper (15)
might be more generally applicable, and suggest the need for a review of the
missing value procedures in the NAAQS [5-8].
• Figure 5-4 demonstrates the fact that the distribution of ozone concentrations is
not the same throughout the ozone season, which suggests that the process of
estimating a design value using a single distributional fit is inherently
contradictory [5-15].
• Section 6.0 appears to have the same objective as Table 5-1 [6-1].
• The fitting of a single tail exponential distribution to a year of daily maximum
concentrations is inappropriate because of nonstationarity (trends within the
year) unless either nonstationarity is regarded as negligible, or the tail only
contains data from a single "high season," or the procedure seems "sensible
though ad hoc" [6-1].
• It would be interesting to know how many of the areas given in Table 6-4
were correctly classified [6-17].
• The term CLV is not widely used and is easily confused with the design value;
a better approach would be to refer to the CLV as the (true) design value and
the design value as the estimated design value [6-25, 6-29].
• The referenced 3-year 95th percentile does not appear in Table 6-11 and
"estimated average CLV" is not clearly defined [6-30].
• The arguments made that a poor regression fit implies the 95th percentile gives
a poor design value estimate are not convincing [6-30, 6-31].
• The regression line CLV = 1.8965 - 0.046 closely approximates the
exponential case result CLV = 25 [6-41].
• The role of meteorology in ozone formation is of interest because future
control requirements may take it into account, and also because it should be an
important part of any stochastic models for exceedances [7].
• Some discussion about the purposes of meteorological modeling should be
included here [7-1].
• Scalar and vector average wind speed should be defined [7-4].
94069 .J J-17
-------
Complex models such as those of Cox and Chu (1991) and Shively (1990) may
tend too much to be tailored to fit the specific local data [7-15].
Those complex models usefully take into account the variation of the
concentration distribution by assuming that the parameters of the distribution
on consecutive days depend on meteorological variables [7-15].
The purpose of Cox and Chu's linear transformation of meteorological
variables is unclear [7-15].
As noted, the Bonferroni approach can lead to wide, and then perhaps not very
useful, confidence intervals [9-4].
The absence of trend does not imply exactly equal annual exceedance rates
[9-7].
Independence between sites may not be too realistic an assumption; if there is
dependence then the total number of exceedances summed across sites could be
modeled as compound Poisson rather than simple Poisson [9-8].
The assumption made by Shively (ref 19) of identical daily distributions within
a year seems tenuous [9-9].
Several papers by the reviewer and others give a more formal theoretical
treatment than Smith's of the clustering of exceedances [9-9, 9-10].
Sufficiency of the 75 percent completeness criterion used for the EPA table
lookup procedure could perhaps be easily evaluated [10-1].
The "expected exceedance rate" should be the "estimated expected exceedance
rate" [10-1, 10-4, etc.].
The percentiles of the daily maximum concentrations are presumably from a
mixture of distributions applying to different parts of the ozone year [10-15].
The 95th percentile should intuitively be somewhat less variable than the
fourth highest, but this is not supported by the evidence presented [11-2].
The mean of the highest 30 values is a close relative of Hill's estimator for the
exponential tail parameter and therefore may be worth consideration [11-2].
Less variable statistics may behave equivalently to exceedances at one site but
not at another [11-2].
94069.J J-18
-------
• Meteorological adjustment has a role in assessing the effects of controls, but
the EPA correctly points out that health is affected by actual rather than
meterologically adjusted ozone [11-2],
• Consistency between the design value and the compliance test is natural,
aesthetic, and traditional but modest departures from consistency would be
justified if good statistical properties were then achieved [11-3],
• It does not appear that the 95th percentile has compelling statistical advantages
[11-3].
• The oscillation between attainment and nonattainment, and the suggested "too
close to call" procedures both deserve study [11-3, 1-13].
• Design value estimates could be used to determine compliance if matched to
the NAAQS; e.g., if the fourth highest value in three years is used [12-1].
• Regarding Finding 8, the control strategy value is likely to be more uncertain
than the design value because of imprecise knowledge about ozone transport
and the use of models; a corresponding control strategy based attainment test
would thus have further uncertainty problems [12-3].
• A search for simple but more "robust" procedures or modifications to the
NAAQS is valuable [12-4].
• An important test of the reasonableness of the EPA design value methodology
is the degree of correct classification [12-4].
• Data from years of small variability may also not provide reasonable estimates
of ozone levels in future years since the time period for the design value needs
to include both large and small fluctuations [12-4].
• Although the study authors' work has been exceptionally thorough, a reader
should realize that inclusion of a paper in the summaries is not an endorsement
and does not guarantee correctness [Appendix].
• Spatial variability is an important issue that fortunately can to an extent be
treated separately from the problem of single-site design value estimation;
ORD and AREAL work on monitoring network design using techniques of
geostatistics may be relevant to ozone [1-7].
• This summary may need further clarification [1-8].
94069. J J-19
-------
Chock's papers represent impressive scientific inquiry; Chock's t-test proposal
may lack simplicity and may not be uniform across sites [1-9,1-12].
The cited paper has a fundamental mathematical error [1-11].
Since there is a 65 percent probability of misclassifying a borderline,
noncompliant site as compliant using the current NAAQS (the converse case
has a 35 percent probability), the argument should lead to more stringent,
rather than less stringent, alternatives to the NAAQS [1-13,1-14].
Ozone transport is an important concern and current adjustment procedures are
somewhat ad hoc [1-16,1-17].
Meteorological effects on ozone variability should be studied but the actual
ozone levels are crucial to the standard [1-16,1-17, 1-20, 1-21].
More explanation about "modeling guidance" is needed to make this summary
more generally understandable [1-19].
94069.J J-20
-------
NEW YORK STATE DEPARTMENT OF ENVIRONMENTAL CONSERVATION
1 Conclusions stated in the report regarding the comparison of "alternative"
design value methodologies are vague.
2 The issues of meteorological variability and transport are considered only in
connection with what the report refers to as the "control strategy design value"
and consideration of them is relegated to the NAAQS review process rather
than the study, despite the fact that the current design value methodology is
used to classify nonattainment areas and hence determine control requirements
[7-8].
3 The report should clarify the relation between determination of attainment and
determination of a design value, particularly the treatment of "expectations" in
the two procedures, since some relevant factors in these expectations (such as
temperature) would be categorized as "control strategy design values" and
hence could not be considered in nonattainment classification [9-11].
4 While the CAA explicitly mentions use of design values in connection with
attainment determinations, the report reiterates the EPA's position that
attainment is based strictly on exceedances [9].
5 While the use of longer periods of data to determine design values would
reduce the effects of meteorological variability, it would also mask indications
of progress toward attainment and could lengthen a nonattainment region's
wait for reclassification [12].
94069.J J-21
-------
STATE OF OHIO ENVIRONMENTAL PROTECTION AGENCY
1 The report holds that adjustments to the design value, to allow for transported
ozone for example, should be called "control strategy adjustments" as
diffrentiated from the "air quality design values" estimated from monitoring
data [1-2].
2 Although the report shows that use of a fitted statistical distribution instead of
the look-up table procedure for determining the design value does not produce
significantly different design values, the use of a fitted distribution can result
in areas being reclassified nonattainment that are presently designated
attainment on the basis of the look-up table [3-4].
3 Use of longer averaging periods would have little effect on reducing the
impact of an unusual year such as 1988 [5].
4 Methods that adjust for transport and meteorological variability cannot be used
under the CAA in judging attainment, and hence must be considered under the
NAAQS review to correct design values for Ohio that are extremely high [6-
8].
5 Control requirements should be reviewed on a rolling three-year basis, which
would avoid the effect of the extreme 1988 air quality measurements on setting
control requirements (that are inappropriate given the last five years of near-
attainment ozone levels) and would be more consistent with the methods used
to evaluate redesignation requests [9].
6 If Section 181(b)(2) of the CAA does not allow for adjustment of the design
value to account for transport and meteorological variability, then a "control
strategy design value" that permits such adjustment should be adopted [10].
94069.J J-22
-------
APPENDIX K
Misclassification Probabilities for Design Value Classes
94069 .K K-l
-------
P = Pr{N1 > 4, N2 < 4} = P{Nj > 4} - P{N2 > 4}
where N, and N« axe the numbers of exceedances in three years of the
levels .121, .138. These are Poisson random variables with xespective
means 3Aj = 9.0, 3A2 = 5.4, so that P{Nj > 4} = -979, P{N2 > 4} =
.787 and hence P = .192. Thus there is a 19% probability that the area -
actually borderline between moderate and serious — will be misclassified as
marginal (and corresponding probabilities for other misdassifications).
Example 2: Suppose the area is in the middle of the serious region, with
D.V. u = 1.7. Then a = 27 and the expected numbers of
exceedances per year of u, = .16, u« = .18 are respectively A, = 1.3, A«
= .78. Hence the numbers N, and N« of exceedances of u, and u2 in 3
years are Poisson with respective means. 3.9, 2.34 giving the probability
of correctly classifying the area (as serious) as
P{NX > 4} - P{N2 > 4} = -547 - .209 = .338.
Example 1 shows a significant probability that a D.V. will be estimated in
a class which is not contiguous with the true 1. Example 2 shows
substantial probability (66%) of a wrong classification.
Assuming arithmetic correctness these examples suggest that
misclassification probabilities by the D.V. estimate may be rather high. If
this is the case two options are
(i) to use less groups in the classification;
or (ii) to investigate other D.V. definitions, of the type considered in
the study, based on more observations than just the fourth
largest.
Finally, these examples assume a distribution of totally exponential form
(not just in the tail). Certainly other cases should be considered, with
tail behavior approximating some found in practice, before regarding the
above calculations as at all conclusive. It may be noted incidentally that
the exponential tail form with a = 29 is dose to that actually observed in
Philadelphia data, as shown on Attachment D.
-------
Attachment A
Miscl ossification probabilities
for given true annual expected exceedance rates,
under the expected exceedance criterion
True annunl
expected exceedance
rate m
Misclas5«fication
probability
Actually
compliant
cases
(mil)
1
0.9
0.8
0.5
.35
.29
.22
.07
Actually
non-compliant
cases
marginally > 1
1.1
1.2
1.5
2.0
.65
.58
.52
.34
.15
-------
Attachment B
Misdassification Probabilities for Design Value Classes
(EPA estimator)
The ozone classification quoted in the study define classes as follows:
Class Design value (pom)
Marginal .121 to .138
Moderate .138 to .160
Serious .160 to .180
Severe .180 to .280
Extreme .280 and above
A nonattainment area is classified in a given class if the estimated D.V. -
the fourth highest daily maximum in 3 years — falls in the indicated
range. These calculations given two examples of the probability that the
D.V. estimate will give an incorrect classification.
For calculation it is assumed for simplicity that the distribution is
exponential with tail e~CK so that the expected number of exceedances of
the D.V. u in an assumed 100 day ozone season is 100 e~mi. Since this
is 1 by definition of the D.V. it follows that the
a = -u"1 to. .01 = 4.6 u""1.
Example 1: Area on the border between moderate and serious.
Here the D.V. u = .16 so that a = 4.6/.16 = 29.
The expected number of exceedances per year of a level v is 100 e~oai.
This is
Aj = 3.0 if v = .121
A2 = 1.8 if v = .138.
The probability P that the area will be declared marginal is the
probability that the fourth largest daily maximum in 3 years will lie
between .121 and .138, i.e.
-------
o
T-
o'
I
0)
0>
o
g
q
o
01
A}i|iqeqojd
-------
TECHNICAL REPORT DATA
(Please read Instructions on the reverse before completing)
1. REPORT NO.
EPA-454/R-94-035
3. RECIPIENT'S ACCESSION NO
4. TITLE AND SUBTITLE
Clean Air Act Ozone Design Value Study:
Report
A Report to Congress
Final
5. REPORT DATE
December 1994
6. PERFORMING ORGANIZATION CODE
7. AUTHOR(S)
Warren P. Freas
8. PERFORMING ORGANIZATION REPORT NO.
9. PERFORMING ORGANIZATION NAME AND ADDRESS
U.S. Environmental Protection Agency
Office of Air and Radiation
Office of Air Quality Planning and Standards
Research Triangle Park, NC 27711
10. PROGRAM ELEMENT NO
11. CONTRACT/GRANT NO.
12. SPONSORING AGENCY NAME AND ADDRESS
13. TYPE OF REPORT AND PERIOD COVERED
Final Report
14. SPONSORING AGENCY CODE
15. SUPPLEMENTARY NOTES
Mr. T. Stoeckenius and Dr. J. Cohen, Systems Applications International, provided
support for the statistical and meteorological analysis. Dr. W. Biller performed the
16. ABSTRACT
time series analysis.
Under Section 183 of the Clean Air Act, EPA was directed to conduct a study of the
methodology currently in use for calculating design values to determine if the
calculated design value "provides a reasonable indicator of the ozone air quality of
ozone nonattainment areas." The focus of the study was on technical and statistical
issues relating to the methodology contained in existing rules and guidance. Issues
concerning the current form of the National Ambient Air Quality Standard (NAAQS) for
ozone were not dealt with in the study.
The study found that the EPA ozone design value method does indeed provide a reasonable
estimate of the "true" air quality design value for ozone nonattainment areas and of
peak ozone levels within those nonattainment areas for the initial 3-year compliance
period. However, the design value cannot describe the spatial variability in ozone
concentrations across the area. More robust indicators based on specific monitoring
sites also have large spatial variability. Ozone design values calculated with the
EPA design value method are highly correlated with other more robust indicators.
However, due to the spatial variability observed across urban areas, one cannot
expect a single numerical value to adequately describe complex concentration.
gradients across large metropolitan areas. ______
7.
KEY WORDS AND DOCUMENT ANALYSIS
DESCRIPTORS
b.lDENTIFIERS/OPEN ENDED TERMS C. COSATI Field/Group
Design values
Ozone
Clean Air Act
Nonattainment
Air Pollution Regulation
18. DISTRIBUTION STATEMENT
Release Unlimited
19. SECURITY CLASS (This Report!
Unclassified
21. NO. OF PAGES
380
20. SECURITY CLASS (This page I
Unclassified
22. PRICE
EPA Form 2220-1 (R»y. 4-77) PREVIOUS EDITION is OBSOLETE
------- |