EPA/600/A-92/124
THE SPATIAL AND TEMPORAL ANALYSIS OF
NON-URBAN OZONE CONCENTRATIONS OVER THE EASTERN UNITED STATES
USING ROTATED PRINCIPAL COMPONENT ANALYSIS
Brian K. Eder*
Atmospheric Sciences Modeling Division
Air Resources Laboratory
National Oceanic and Atmospheric Administration
Research Triangle Park, NC, USA 27711
1. INTRODUCTION
Traditionally, ozone (03) has been considered an
urban-scale pollutant. More recently, however, it has been
recognized by scientists as a regional (Logan, 1989) and even
global-scale phenomenon (Liu et al., 1987) as high
concentrations are routinely observed over large non-urban
areas of most industrialized countries where forest retardation
and crop injury are becoming growing environmental
concerns (Lefohn and Lucier, 1991). Daily maximum 03
concentrations in these areas are often comparable to those
found in urban areas (Meagher et al., 1987) and daily
average levels can even exceed urban levels due to a lack of
nitric oxide (NO) scavenging.
Clearly, a better understanding of the spatial and
temporal variability of non-urban 03 concentrations is critical
to the development and implementation of standards designed
to mitigate any adverse effects on forest and crop
productivity as well as ecosystem well-being. In an attempt
to enhance this understanding, the eastern United States will
be segregated, using a Rotated Principal Component Analysis
(RPCA), into areas whose 03 concentrations exhibit unique,
homogeneous characteristics. Examination of these
characteristics will then be achieved using a variety of
interpretive analyses.
The RPCA approach has been used successfully in the
examination of other aerometric data including SO„=
concentrations in precipitation (Eder, 1989) and S02 ambient
air concentrations (Ashbaugh et al., 1984). The advantages
of using such an approach are numerous (Cox and Clark,
1981; Crutcher et al., 1986; Eder, 1989). First, because of
the copious amount of data often resulting from such large-
scale studies (nearly 100,000 observations for this study),
and because individual data tend to be erratic or noisy, it is
advantageous to employ an analysis technique which
identifies, through a reduction of data, the recurring and
independent modes of variation within the larger data set.
Secondly, RPCA will allow comparison of 03 characteristics
(i.e. trends, distributions, etc.) between regions whose
segregation is statistically and physically based as opposed to
those based on arbitrary or geo-political boundaries. And
* On assignment to the Atmospheric Research and Exposure
Assessment Laboratory, U.S. Environmental Protection
Agency.
finally, the analysis of 03 characteristics and trends within
these regions would be based upon an aggregation of data
from many stations over a long time period, as opposed to
individual stations, minimizing the effects of anomalous or
even erroneous data often associated with a particular station.
2. DATA
The data employed in this analysis were obtained
from the United States Environmental Protection Agency's
Aerometric Information Retrieval System (AIRS). This
database contains a multitude of hourly aerometric data,
including 03 concentrations, collected by local agencies at
nearly 800 stations nationwide. The data are monitored
using criteria established by the EPA (1985), which includes
multipoint calibrations, independent audits and data
validation based upon frequent zero, span and precision
checks and a rigorous quality assurance program. The 03
measurements were made during the 03 "season" (April
through October for this study) using either
chemiluminescence analyzers, which are sensitive to light
emitted by the reaction between 03 and ethylene or
ultraviolet photometers, which measure the absorption of
light by 03
A major consideration of this study was the
establishment and utilization of a complete, regionally-
representative 03 data base, unencumbered by either missing
data or local-scale variability. Attainment of this goal was
achieved by employing numerous strict selection criteria.
First, to avoid NO scavenging affects found in close
proximity to urban areas, only those stations classified as
either rural or suburban and reporting a land use of either
forest, agriculture or residential were selected. Rural stations
received highest priority. However, to meet the second
criteria, spatial completeness, several suburban stations were
also included. Third, only those stations reporting a capture
rate of 90.0% or better for the study period were considered.
Finally, to keep the amount of data manageable, only the
daily maximum 03 concentrations were used. Use of this
statistic is justified by Vukovich and Fishman (1986) who
have shown that at the time of maximum surface
concentration (typically between 1 and 3 pm LST), the
boundary layer is uniformly mixed and the surface
concentration is very nearly equal to the mean boundary
layer concentration.

-------
These selection criteria resulted in the inclusion of 77
stations across the eastern half of the United States, the
majority of which (55) are classified as rural. Several
combinations of station-"seasons" were examined before the
optimum period of 1985-1990 was selected. The total
capture rate for the period was 95.7% (94,617 out of a
possible 98,868 observations), which translates into an
average of 205 out of 214 days per station-'season".
Missing data were estimated using a linear interpolation
scheme.
3. METHODOLOGY
Mathematically, this analysis began with the
extraction of a square, symmetrical correlation matrix R
(having dimensions of 77 x 77) from the original data matrix
(having dimensions of 77 stations x 1284 days and containing
98,868 observations). Selection of a correlation matrix (as
opposed to a covariance matrix) has two advantages. First,
use of a correlation matrix is much more suitable for
resolving spatial oscillations (Overland and Preisendorfer,
1982), which is a major goal of the analysis. Second, use of
a correlation matrix allows isopleths of component loadings
(which can be regarded as the correlation coefficient between
the component and the individual station) to be drawn. By
using R and the identity matrix (I), of the same dimensions,
77 eigenvector-eigenvalue pairs were derived. The
eigenvectors represent the mutually orthogonal linear
combinations or modes of variation of the matrix, while their
respective eigenvalues represent the amount of variance
explained by each of the eigenvectors.
By retaining only the first few eigenvector-eigenvalue
pairs and their corresponding principal components, a
substantial amount of the total variance can be explained.
The higher order principal components, which explain
minimal amounts of the variance, are viewed as noise. The
exact number of components that should be retained remains
a subject of some controversy and uncertainty. Retaining
too few components results in the blending of spatial patterns
which are actually more discrete; whereas retaining too many
components will delineate ares which represent nothing more
than randomly generated variations (Richman, 1981). The
method selected for this analysis was developed by Overland
and Preisendorfer (1982) and is based upon a significance
test at a 95% Confidence Level (C.L.) using a random
Monte Carlo simulation. When applied to this data set, the
significance test indicated that the first six principal
components should be retained and that the higher order
principal components are considered noise. The Scree Test,
which entails the plotting of the percentage variance
explained by each component in the order extracted and then
looking for a discontinuity in the curve, also supported a six
principal component solution. The fraction of total variance
explained by these six components are presented in Table 1.
The original data set, which contained 77
intercorrelated and noisy variables (stations) has therefore
been reduced to six orthogonal and thereby independent
variables (principal components) and yet still explains almost
two-thirds of the total variance of the original data set.
Table 1. Statistics for the first six unrotated principal
components.

PC 1
PC 2
PC 3
PC 4
PC 5
PC 6
Eigenvalue
27.61
7.44
5.51
3.52
2.86
2.36
% Variance
35.86
9.66
7.15
4.58
3.71
3.06
Cumul. %
35.86
45.52
52.67
57.25
60.96
64.02
These results are very promising, especially given the highly
variable nature of 03 and the fact that daily, non-smoothed
data were used.
Since one of the major goals of this paper is to define
areas of homogeneous 03 concentrations, a rotation was
performed on the components in order to better segregate the
areas that have similar concentration characteristics. Of the
many types of rotation possible, an orthogonal method
developed by Kaiser (1958) was selected because it rigidlv
rotates the predetermined principal components while
retaining the constraints that the individual components
remain orthogonal or uncorrelated to each other. This
method increases the segregation between component
loadings, which in turn better defines a distinct grouping or
clustering of intercorrelated data, thereby making spatial
interpretation easier (Horel, 1981).
Comparison between the statistics associated with the
six rotated (Table 2) and unrotated (Table 1) components
reveals that the total explained variance remains the same,
64.02%, but the distribution of the variance explained over
the individual components is more uniform.
Table 2. Statistics for the first six orthogonally rotated
principal components.

PCI
PC2
PC3
PC4
PC5
PC6
Eigenvalue
13.55
10.82
8.58
7.58
5.46
3.30
% Variance
17.60
14.05
11.14
9.84
7.09
4.30
Cumul. %
17.60
31.65
42.79
52.63
59.72
64.02
4. RESULTS AND DISCUSSION
When the elements of each eigenvector are multiplied
by the square root of the associated eigenvalue one obtains
the component loading, which represents the correlation
between the component and the station. (The square of a
component loading indicates the proportion of variance at the
individual station that can be attributed to that component).
When these principal component loadings are spatially
mapped onto their respective stations for each component,
isopleths of component loadings can be drawn. Because of
space limitation, individual maps of each of the rotated
components are not presented. However, all six maps can be
incorporated into one by plotting the maximum loadings for
each station and their respective rotated principal component.
The resulting map (Figure 1) depicts six separate, contiguous
subregions, each exhibiting statistically unique 03
concentration characteristics. Their uniqueness is likely

-------
attributable to commonality of forcing factors such as
meteorological and emissions patterns.
Loadings associated with the first rotated component
define an area encompassing the Great Lakes region north
and west of the Ohio River. The second component
encompasses the Northeast States from Pennsylvania and
northern New Jersey, northward. The highest loadings
associated with the third rotated component are found over
the South, while the fourth component defines the Mid-
Atlantic region from North Carolina to Maryland. The fifth
component defines the Southwest part of the study area from
Texas and Louisiana northward to Kansas, and loadings
associated with the sixth component highlight stations in
Florida.
'NORTHEAST g
\J\ 82
75 8X
BS-
65-£v
76*80
g3 *MiOrATLANTiC
59
SOUTH'
SOtSTHWESr
FLORIDA' 76"
60 ?4
66
Maximum Compontnt loadings (xlOO)
Figure 1. Six homogeneous 03 concentration
subregions as defined by maximum component loadings.
In order to determine the stability of this rotated
principal component solution (i.e. stability of the six
homogeneous subregions), and to determine the amount, if
any, of correlation between subregions, the orthogonality
constraint was relaxed. This "relaxation" was achieved
through the application of an Harris-Kaiser case II Oblique
rotation, which has the same general objective as a Varimax
rotation (maximization of the variance of the component
loadings between components for a given station) but it
entirely relaxes the orthogonality constraint. If there is little
correlation between subregions, the obliquely rotated solution
will closely resemble the orthogonally rotated solution. For
this analysis, the oblique solution was very similar to the
orthogonal solution with only 4 of the 77 stations transferring
subregions. With only a few exceptions, which were found
between several adjacent subregions, cross-correlations were
quite weak, ranging mostly from -0.20 to 0.20. This
suggests that the physical factors determining the uniqueness
of 03 concentrations for each of these subregions are indeed
different.
Examination of the temporal characteristics of these
six homogeneous subregions can be accomplished with
principal component scores, which are weighted summed
values for the days over the stations, the weights being the
component loadings. The component scores are
standardized, so they have a mean of zero and standard
deviation of one. When plotted as a time series, the scores
provide excellent insight into the spectrum of temporal
variance experienced by each subregion - insight that would
prove very useful for the modeler trying to determine the
optimum time for model simulations.
For the sake of brevity, only three of the six time
series are presented (Figure 2). They reveal several
interesting features, most notably the presence and timing of
strong seasonalities. The Northeast Subregion (Fig. 2a)
exhibits a strong seasonality in which the highest principal
component scores (corresponding to high 03 concentrations)
occur much more frequently during the months of June
through August. Low scores occur predominantly during the
months of April, September and October. This strong
seasonality is not nearly as evident in the Southwest time
series (Fig. 2b) as the timing of high and low scores is more
sporadic. Florida's time series (Fig. 2c) reveals another
strong seasonality, however, one which is out-of-phase with
that occurring in the Northeast. For this subregion, the
highest concentrations tend to occur most frequently during
April and May, with low concentrations dominating the
remainder of the season.
UJ
DC
CO
UJ
z
2
E
O
LiJ
-------
Insight into the day-to-day variability occurring within
each subregion can also be gained from the time series. For
illustrative purposes lets examine 1988, when the Northeast
subregion recorded 10 days with principal component scores
greater than 3,0 standard deviations above the six year
normal. No other year for this subregion recorded more than
one score greater than 3.0. Additionally, eight of these days
occurred within a 31-day period, from June 16 through July
16; signifying the most intense episode experienced by the
subregion. Similar insight can be made upon quick visual
examination of the time series associated with the other
subregions. Modelers can easily select, for each subregion,
the worst period (3-day, 7-day, month, season, etc.) on
which to perform their simulations.
Having delineated the eastern United States into areas
that exhibit homogeneous 03 characteristics, further
elucidation and comparison of these characteristics can now
be accomplished through a variety of simple analyses
involving the raw ozone data found at each station within
subregions. Table 3 reveals that in comparison to the entire
study domain (mean of 56.06 ppb), the Florida Subregion
experiences considerably lower concentrations (48.74 ppb),
and that concentrations exceeding 80 ppb (120 ppb) occur
only one-half (one-fifth) as often as the domain
concentrations. The mean concentrations for the Great Lakes
(53.83), Northeast (54.61) and Southwest (55.15) are all
slightly lower than the domain average, while those for the
South (58.64) are slightly higher. Mean concentrations for
the Mid-Atlantic Subregion (63.23) are considerably higher
than the domain, as are the percentage of days exceeding 80
ppb (20.3%) and 120 ppb (1.9%).
Examination of the Coefficient of Variation (%)
reveals the compared to the entire domain (39.02), less
variability is found in the Mid-Atlantic (35.97), South
(37.04), and Great Lakes (37.48) Subregions, while more is
found in the Florida (40.17), Southwest (40.60) and
Northeast (40.93) Subregions.
Table 3. Summary statistics of the 0} (ppb) concentrations
associated with each subregion and the entire domain.
SUBREGION
MEAN
C. V.
% > 80
% > 320
Great Lakes
(11 = 26,964)
53.83
37.48
9.7
0.6
Northeast
(19,260)
54.61
40.93
13.1
1.3
South
(17,976)
58.64
37,04
15.0
0.9
Mid-Atlantic
(14,124)
63.23
35.97
20.3
1.9
Southwest
(14,124)
55.15
40.60
14.0
3.5
Florida
(6,420)
48.74
40.17
7.1
0.2
Domain
(98,868)
56.06
39.02
13.9
1.1
Examination of the monthly distributions associated
with each subregion is also revealing (Figure 3). Statistics
are presented in the form of a boxplot and include: 1" and
99th percentiles (open circles), 5th and 95th percentiles (bars),
25lh and 75th percentiles (rectangle ends) and median (center
line) and mean (filled circle). Figure 3(a) supports the
earlier analysis in that a strong seasonality, centered around
July, is observed in the Northeast Subregion. (This
seasonality is also observed in the Mid-Atlantic, South and
Great Lakes subregions; however, there are several
interesting variations in its timing, strength and symmetry).
Statistics for the Northeast Subregion reveal that
concentrations in July are substantially higher than the next
two closest months June and August (which are very
similar). The next highest concentrations are observed in
May, with April and September following closely, although
September exhibits considerably more variability than April.
Concentrations observed during October are much lower than
any other month.
1
u 80
M 60
APR MAY JUN JUL AUG SEP OCT
1
u SO
8 60
APR MAY JUN JUL AUG SEP OCT
m 80
APR MAY JUN JUL AUG SEP OCT
MONTH
Figure 3. Boxplots of the monthly Os concentrations
associated with three subregions: (a) Northeast, (b)
Southwest and (c) Florida.
-------
Also supporting the earlier analysis is the lack of a
strong seasonality in the Southwest Subregion (Fig. 3b) and
an out-of-phase seasonality observed in the Florida Subregion
(Fig. 3c). The monthly statistics remain virtually the same
through the Southwest Subregion's season, with perhaps a
slight decrease in September and October. Monthly Oj
concentrations observed in the Florida Subregion are
completely different than the other subregions. Highest
concentrations are found during the months of April and
May. Concentrations for the remaining months are
considerably lower and exhibit little variation.
Having examined the monthly variability within
subregions, we can now focus on the annual (April through
October) variability as seen in Figure 4. The annual
boxplots for the Northeast Subregion (Fig. 4a) reveal that the
highest concentrations (means and especially the higher
percentiles) occurred during 1988. 1985 provided the next
highest concentrations, with the remaining years, 1986,
1987, 1989 and 1990 all experiencing more similar
concentrations.
160
140
120
two
s
!£ 80
S 60
S
40
20
0
(a)
1
r
160
140
120
100
80
60
40
20
1985
1986
1987
1988
1969
1990
uj 80
S 60
1985
1986
1987
1988
1989
1990
160
140
120
I
100
s


80
1
60
0

40

20

0
(C)
JL
0
T
r160
•¦140
f-120
-100
80
60
40
20
0
1985
1986
1987
YEAR
1988
1969
1990
Figure 4. Boxplots of the annual 0} concentrations
associated with three subregions: (a) Northeast, (b)
Southwest and (c) Florida.
Mean concentrations for the Southwest Subregions
(Fig 4b.) were similarly high during 1987 and 1988.
Concentrations for the years 1985 and 1990 were also similar
and the next highest, followed by 1986 and 1989. Florida
(Fig. 4c) exhibited the smallest annual variability. With the
exception of 1985, which recorded slightly lower
concentrations, annual statistics remained virtually constant.
In order to determine if these annual summary
statistics exhibited trends within this limited six-year period,
a simple linear regression was performed for all six
subregions and the estimated slopes (and their standard
errors) of the models calculated (Table 4). Although only
two of the trends were statistically significant (90% C.L.),
they did generally exhibit within-subregion consistency. For
instance, all 03 statistics associated with the Southwestern,
Mid-Atlantic and Northeast Subregions show a slight
decreasing tend. (Only the median 03 concentrations
associated with the Mid-Atlantic Subregion was significant).
Conversely, the South Subregion has experienced a slight
increase in all statistics over the six year period. Both the
Great Lakes and Florida Subregions exhibit mixed results,
with the means and medians increasing over the period and
the higher percentiles generally decreasing. (The 99th % in
Florida was significant). Statistics associated with the entire
domain (all stations) indicated a general decreasing though
not statistically significant trend.
Table 4. Estimated trend line slopes (ppb year'') for
selected statistics for each subregion and the entire domain.
(Standard error of estimate provided in parenthesis.)
SUBREGION
MEAN
MEDIAN
75" S
95" *
99" %
Great Lakes
0.16
(1.00)
0.23
(0.93)
-0.09
(1.36)
0.26
(2.19)
-0.23
(2.92)
Northeast
-0.57
(0.67)
-0.54
(0.29)
-0.83
(0.90)
-0.51
(2.55)
-0.57
(3.55)
South
0.19
(0.92)
0.00
(0.73)
0.23
(1.23)
0.77
(1.88)
1.46
(2.46)
Mid-Atlantic
-0.87
(0.64)
•0.94*
(0.32)
-1.20
(0.86)
-0,69
(2.49)
-0.23
(3.54)
Southwest
-0.18
(0.60)
-0.06
(0.90)
-0.14
(0.81)
-1.31
(0.73)
-0.46
(1.28)
Florida
0.34
(0.35)
0.69
(0.55)
0.06
(0.47)
-0.31
(0.49)
-1-37—
(0.54)
Domain
-0.16
(0.66)
-0.17
(0.62)
-0.26
(0.94)
-0.20
(1.83)
0.06
(2.51)
Significant at 95%
' Significant at 90®
5. SUMMARY
The spatial and temporal variability of 03
concentrations over the eastern United States during the
period 1985 through 1990 was examined through the use of
a multivariate statistical technique called Principal
Component Analysis. The original data set, which contained
77 correlated variables (monitors) was reduced to six
uncorrelated principal components, while still explaining
almost two-thirds (64.02) of the total variance. Application
-------
of Kaiser's Varimax rotation led to the identification of six
separate, contiguous subregions which each exhibit
statistically unique 03 concentration characteristics.
When compared to the entire domain, the Great
Lakes, Northeast, Southwest and Florida Subregions
observed lower mean 03 concentrations. Conversely, the
South and Mid-Atlantic Subregions recorded higher than
domain means. Variability, as defined by the Coefficient of
Variation, was highest in the Northeast, Southwest and
Florida Subregions, and lowest in the Great Lakes, South
and Mid-Atlantic Subregions, The percentage of
observations exceeding concentration thresholds of 80 and
120 ppb, were higher in the Mid-Atlantic and Southwest
Subregions, lower in the Great Lakes and Florida Subregion
and near the domain average in the Northeast and South
Subregions.
A strong seasonal cycle was observed in the Great
Lakes, Northeast, Mid-Atlantic and South Subregions. The
timing, strength and symmetry varied across these
subregions. The Southwest Subregion exhibited little
seasonality, while the Florida Subregion contained a strong,
out-of-phase seasonality with the maximum occurring during
the months of April and May.
Annually, the highest 03 concentrations generally
occurred during 1988; however, several subregions (Florida
and the Southwest) recorded equally high concentrations in
other years. No one other year stood out statistically across
all subregions. Trend analyses indicated a slight, though not
statistically significant, decrease in 03 statistics for the
Northeast, Mid-Atlantic and Southwest Subregions and a
slight increase over the South Subregion. Ozone
concentration trends for the Great Lakes and Florida
Subregions were mixed indicating slight increases in mean
and median concentrations and slight decreases in the higher
percentiles.
These results have provided a statistically and
physically based rationale for choosing distinctive
geographical areas for interpreting 03 air quality distributions
and trends. Since data from stations within subregions
exhibit homogeneous variability, we have been able to
develop regionwide 03 indicators which have provided
meaningful insight into the seasonal and annual concentration
trends of the six subregions. The analysis has also suggested
that trends analyses for determining general progress in
improving 03 air quality could be based on aggregate
statistics from clusters of monitors rather than from
individual stations.
Dbclaimer
This paper has been reviewed in accordance with the U.S.
Environmental Protection Agency's policies and approved for
presentation and publication. Mention of trade names or
commercial products does not constitute endorsement or
recommendation for use.
6, REFERENCES
Ashbaugh, L. L., L. 0. Mynip, and R. G. Flocchini
(1984) A principal component analysis of sulfur
concentrations in the western United States. Amos.
Environ., 18, 783-791.
Cox, W. M. and J. Clark (1981) Ambient ozone
concentration patterns among eastern United States urban
areas using factor analysis. J. Air Pollut. Control Assoc.,
31, 762-766.
Crutcher, H. L., R, C. Rhodes, M. E. Graves, B.
Fairbairn, and A. C. Nelson (1986) Application of cluster
analysis to aerometric data. J. Air Pollut. Control Assoc-,
36, 1116-1122.
Eder, B. K. (1989) A principal component analysis
of S04= precipitation concentrations over the eastern United
States. Atmos. Environ., 23, 2739-2750.
Horel, J, D. (1981) A rotated principal component
analysis of the interannual variability of the Northern
Hemisphere 500 mb height field. Mon. Wea. Rev., 109,
2080-2092.
Kaiser, H, F. (1958) The Varimax criterion for
analytical rotation in factor analysis. Psychometrika, 23,
187-201.
Lefohn A. S. and A. A. Lucier (1991) Spatial and
temporal variability of ozone exposure in forested areas of
the United States and Canada: 1978-1988. J. Air Waste
Manage. Assoc., 41, 694-701.
Liu, S. C,, M. Trainer, F. C. Fehsenfeld, D. D.
Parrish, E. J. Williams, D. W. Fahey, G. Hubler, and P. C.
Murphy (1987) Ozone production in the rural troposphere
and the implications for regional and global ozone
distribution. J, Geophys. Res,, 92, 4191-4207.
Logan, J. A. (1989) Ozone in rural areas of the
United States. J. Geophys. Res. 94, 8611-8532.
Meagher, J. F., N. T. Lee, R. J. Valente and W. J.
Parkhurst (1987) Rural ozone in the southeastern United
States. Atmos. Environ., 21, 605-615.
Overland, J. E. and R. W. Preisendorfer (1982) A
significance test for principal components applied to a
cyclone climatology. Mon. Wea. Rev., 110, 1-4.
Rich man, M.B. (1981) Obliquely rotated principal
components: An improved meteorological mapping
technique? J. Appl. Meteor., 20, 1145-1159.
United States Environmental Protection Agency
(1985) Quality Assurance requirements for state and local
air monitoring stations (SLAMS). 40 CFR, Part 58,
Appendix A.
Vukovich, F. M. and J. Fishmann (1986) The
climatology of summertime 03 and SO:. Atmos. Environ.,
20, 2423-2433.
-------
TECHNICAL REPORT DATA
(Please read Instructions on the reverse before comple'1
1. REPORT NO. 2.
EPA/600/A-92/124
3.
4. TITLE AND SUBTITLE t
THE SPATIAL AND TEMPORAL ANALYSIS OF
NON-URBAN OZONE CONCENTRATIONS OVER THE EASTERN UNITED
STATES USING ROTATED PRINCIPAL COMPONENT ANALYSIS
5. REPORT DATE
6. PERFORMING ORGANIZATION CODE
7. AUTHORiS)
Brian K. Eder
8. PERFORMING ORGANIZATION REPORT NO.
9. PERFORMING ORGANIZATION NAME AND ADDRESS
Atmospheric Research & Exposure Assessment Laboratory
Office of Research and Development
U. S. Environmental Protection Agency
Research Triangle Park, North Carolina 27711
10. PROGRAM ELEMENT NO.
A101T/G/51
11. CONTRACT/GRANT NO.
12. SPONSORING AGENCY NAME ANO ADDRESS
Same as in 9
13. TYPE OF REPORT ANO PERIOD COVERED
14. SPONSORING AGENCY COOE
EPA/600/09
15. SUPPLEMENTARY NOTES _ , , . +h „ „ 	 ,
To be presented at the 12 Conference On Probability and
Statistics In The Atmospheric Sciences, June 22-26, 1992, Toronto, Ontario, Canada
16, ABSTRACT
The spatial and temporal variability of 03 concentrations over the eastern United States during the period 1985 through 1990 was
examined through the use of a multivariate statistical technique called Principal Component Analysis. The original data set, which contained
77 correlated variables (monitors) was reduced to six uncorrected principal components, while still explaining almost two-thirds (64.02) of the
total variance. Application of Kaiser's Varimax rotation led to the identification of six separate, contiguous subregions which each exhibit
statistically unique Oj concentration characteristics.
When compared to the entire domain, the Great Lakes, Northeast, Southwest and Florida Subregions all observed lower mean 03
concentrations. Conversely, the South and Mid-Atlantic Subregions recorded higher than domain means. Variability, as defined by the
Coefficient of Variation, was highest in the Northeast, Southwest and Florida Subregions, and lowest in the Great Lakes, South and Mid-Atlantic
Subregions. The percentage of observations exceeding concentration thresholds of 80 and 120 ppb, were higher in the Mid-Atlantic and
Southwest Subregions, lower in the Great Lakes and Florida Subregion and near the domain average in the Northeast and South Subregions.
A strong seasonal cycle was observed in the Great Lakes, Northeast, Mid-Atlantic and South Subregions. The timing, strength and
symmetry varied across these subregions. The Southwest Subregion exhibited little seasonality, while the Florida Subregion contained a strong,
out-of-phase seasonality with the maximum occurring during the months of April and May.
Annually, the highest 03 concentrations generally occurred during 1988; however, several subregions (Florida and the Southwest)
recorded equally high concentrations in other years. No one other year stood out statistically across all subregions. Trend analyses indicated
a slight, though not statistically significant, decrease in 03 statistics for the Northeast, Mid-Atlantic and Southwest Subregions and a slight
increase over the South Subregion. Ozone concentration trends for the Great Lakes and Florida Subregions were mixed indicating slight
increases in mean and median concentrations and slight decreases in the higher percentiles.
These results have provided a statistically and physically based rationale for choosing distinctive geographical areas for interpreting
O, air quality distributions and trends. Since data from stations within subregions exhibit homogeneous variability, we have been able to develop
regionwide Oj indicators which have provided meaningful insight into the seasonal and annual concentration trends of the six subregions. The
analysis has also suggested that trends analyses for determining general progress in improving 03 air quality could be based on aggregate
statistics from clusters of monitors rather than from individual stations.
17, KEY WORDS AND DOCUMENT ANALYSIS
t. DESCRIPTORS
b. IDENTIFIERS/OPEN ENDED TERMS
c, COSATI Field/Group



18. DISTRIBUTION STATEMENT
RELEASE TO PUBLIC
19. SECURITY CLASS (rhis Report)
UNCI ASSTFTFfl
21. NO, or PAGES
7
20. SECURITY CLASS (This page)
UNCLASSIFIED
22. PRICE
EPA Form 2220-1 (R«». 4-77) previous edition is obsolete
-------