You are on page 1of 11

ARTICLE IN PRESS

Atmospheric Environment 41 (2007) 7127–7137


www.elsevier.com/locate/atmosenv

The effects of meteorology on ozone in urban areas and their use


in assessing ozone trends
Louise Camaliera,, William Coxa, Pat Dolwickb
a
Office of Air Quality, Planning, and Standards, US Environmental Protection Agency, RTP, NC 27711, USA
b
Office of Air Quality, Planning, and Standards, National Oceanic & Atmospheric Administration, RTP, NC 27711, USA
Received 28 February 2007; received in revised form 23 April 2007; accepted 25 April 2007

Abstract

The United States Environmental Protection Agency issues periodic reports that describe air quality trends in the US.
For some pollutants, such as ozone, both observed and meteorologically adjusted trends are displayed. This paper
describes an improved statistical methodology for meteorologically adjusting ozone trends as well as characterizes the
relationships between individual meteorological parameters and ozone. A generalized linear model that accommodates
the nonlinear effects of the meteorological variables was fit to data collected for 39 major eastern US urban areas. Overall,
the model performs very well, yielding R2 statistics as high as 0.80. The analysis confirms that ozone is generally increasing
with increasing temperature and decreasing with increasing relative humidity. Examination of the spatial gradients of these
responses show that the effect of temperature on ozone is most pronounced in the north while the opposite is true of
relative humidity. By including HYSPLIT-derived transport wind direction and distance in the model, it is shown that the
largest incremental impact of wind direction on ozone occurs along the periphery of the study domain, which encompasses
major NOx emission sources.
Published by Elsevier Ltd.

Keywords: Ozone trends; Generalized linear model; Meteorological adjustment; HYSPLIT; Spatial patterns

1. Introduction variations have on ozone trends (US Environmental


Protection Agency, 2005). Results from these peri-
The United States Environmental Protection odic reports have typically been based on a relatively
Agency (US EPA) issues periodic reports that limited suite of meteorological parameters. The
describe the status and trends of air quality through- purpose of this paper is to describe an improved
out the US (US Environmental Protection Agency, version of the statistical model for meteorologically
2005). Because inter-annual meteorological varia- adjusting ozone trends that (1) accommodates the
tions are known to affect daily and seasonal average non-linear effects of the meteorological variables, and
ozone concentrations, EPA often uses statistical (2) includes a much more comprehensive suite of
techniques to reduce the effect that meteorological meteorological variables than were used in previous
analyses (Cox and Chu, 1993, 1996).
Corresponding author. The meteorological adjustment analysis focuses
E-mail address: Camalier.Louise@epa.gov (L. Camalier). on urban areas located in the eastern US primarily

1352-2310/$ - see front matter Published by Elsevier Ltd.


doi:10.1016/j.atmosenv.2007.04.061
ARTICLE IN PRESS
7128 L. Camalier et al. / Atmospheric Environment 41 (2007) 7127–7137

because the number and spatial distribution of functions indicates that a log link is the most
urban areas is sufficient to identify geographic appropriate for these data. A natural spline (Hastie
patterns of the ozone response to meteorology. and Tibshirani, 1990) was employed to allow for a
The initial analysis identifies the effect of each non-linear response between each meteorological
meteorological parameter on ozone at the 39 parameter and ozone concentration. A natural
selected eastern urban areas. Also examined is spline was also applied to a term used to account
the spatial distribution of these meteorological for seasonal changes.
effects on ozone, including the effects of tempera- As noted in the Introduction, previous analyses
ture, relative humidity and wind direction. The 39 (Cox and Chu, 1993, 1996) have been confined to a
urban areas used in the analysis are a subset of 53 relatively small array of meteorological parameters.
metropolitan statistical areas (MSAs) that have To expand upon the suite of meteorological vari-
been used in an EPA report (US Environmental ables that may have some impact on ozone
Protection Agency, 2004) on recent air quality concentrations, a more comprehensive data base
trends. has been assembled by EPA containing an extensive
array of both hourly and daily meteorological
parameters. The enhanced meteorological database
2. Technical approach
consists of nearly 700 meteorological sites across the
US and covers the period from 1995 to 2006. The
There are numerous publications that describe
raw surface meteorological data are extracted from
methods for adjusting measured ozone for the
the integrated surface hourly (ISH) database while
effects of meteorology (Bloomfield et al., 1996;
the raw upper air data is extracted from the
Thompson et al., 2001; Davis et al., 1998). As
Integrated Global Radiosonde Archive (IGRA)
previously used in Zheng et al. (2006), this analysis
databases, both of which are maintained by
employs a generalized linear model (GLM) to
National Climatic Data Center (NCDC). The sur-
describe the relationship between urban ozone and
face data and upper air data are joined by pairing
selected meteorological parameters taken from an
each surface site with its nearest upper-air neighbor.
extensive array of candidate meteorological vari-
The data pairing was only done for upper-air sites
ables. A separate model was fit for each urban area
considered to be ‘‘spatially representative’’ of the
using the GLM modeling function in the R software
nearest surface site. A complete description of
environment (R Development Core Team, 2006).
the omnibus data base along with a map showing
The GLM can be written as follows:
the locations of meteorological stations can be
gðmi Þ ¼ bo þ f 1 ðxi;1 Þ þ . . . f j ðxi;j Þ þ . . . f p ðxi;p Þk found on the scram website at: http://www.epa.
gov/scram001/meteorology/omnibus_meteorological_
þ Wd þ Y. ð1Þ
data_set.pdf. The daily meteorological parameters
The subscript, i, indicates the ith day’s observa- considered in this analysis are shown in Table 1.
tion, j, indicates the jth meteorological variable, Some of the variables are directly observed, while
where j ¼ 1, y, p, and the subscript, k, indicates the others are calculated based on hourly data or other
kth year. The parameter bo represents the overall observed parameters. In addition, ‘‘transport-
mean and f ( ) is the smoothing function where fj related’’ variables were created based on the hybrid
(xi,j) is the value of the smoothing function single-particle Lagrangian integrated trajectory
associated with the ith value of the explanatory (HYSPLIT) trajectory model simulations (Draxler
variable j. The term, Wd, represents the effect of the and Hess, 1997). The HYSPLIT model was run for
dth day of the week, where d ¼ 1, 2, y, 7 (Sunday each day of the data record to calculate 24-h
to Saturday, respectively). The term, Yk, represents backward trajectories from each surface site. The
the effect of the kth year on ozone, i.e. the trajectories were started at noon LST at a height of
meteorologically adjusted value of ozone for that 300 m (i.e., within the mixed layer). Fig. 1 shows
year, where k ¼ 1997, 1998, y, 2005. the results of a single HYSPLIT trajectory for
The element, g (mi), represents the ‘‘link’’ function Cleveland on 7 June 2005 and illustrates how
(McCullagh and Nelder, 1989), which specifies the transport wind direction and transport distance
relationship between the linear formulation on the are determined.
right side of Eq. (1) and the expected response, The ozone air quality data used in the analysis
the mi. Diagnostic evaluation of alternative link was taken from EPA’s air quality system (AQS)
ARTICLE IN PRESS
L. Camalier et al. / Atmospheric Environment 41 (2007) 7127–7137 7129

Table 1
List of daily meteorological parameters that comprise the
expanded meteorological data base and considered as part of
the adjustment model

Parameter Parameter
type

Temperature Maximum surface temperature


( 1C) Morning and afternoon average temperature
Diurnal temperature change
Minimum, maximum, and average apparent
temperature
1200 UTC temperatures at 925, 850, 700, and
500 mb
Deviation in temperature from a 10 year
monthly mean at 850, 700, and 500 mb
24-h change in 1200 UTC 850 mb temperatures
Fig. 1. HYSPLIT trajectory (red) diagram for the Cleveland
Wind (m s1) Average daily u and v wind vectors vicinity on 7 June 2005.
Average daily wind speed and direction
Morning and afternoon average u and v wind
vectors maximum 8-h ozone concentration is extracted
Morning and afternoon average wind speed and
from AQS for each monitoring site within each
direction
urban area. For analysis purposes, the highest 8-h
Humidity Average daily relative humidity (%) average among the monitoring stations in an urban
Midday and nighttime average relative humidity
area was selected to represent the ozone air quality
(%)
Average and maximum dew point temperature for each day. The data included the months from
( 1C) May to September (i.e. the ozone season) for each
Maximum water vapor mixing ratio (g kg1) year from 1997 to 2005. Finally, the modeling
Morning 850 mb dew point temperature ( 1C) database for each urban area was created by
24-h change in 1200 UTC 850 mb dew point
merging the daily ozone data with the daily
temperatures ( 1C)
meteorological data described above. Data from
Pressure (mb) Average station and sea-level pressure the meteorological station nearest to the center of
Morning geo-potential height at 850, 700, and
an urban area was chosen to represent the
500 mb
Deviation in geo-potential height from a 10 year meteorology for that urban area. The resulting
monthly mean at 850, 700, and 500 mb modeling data base for each urban area was a
Stability Difference in 1200 UTC temperatures between
matrix of approximately 1300 days (9 years times
surface and 850 ( 1C) 153 summer days) by 60 meteorological variables.
Maximum afternoon mixing height (m)
Maximum rate of mixing height increase 3. Model development
(m h1)

Transport 24-h HYSPLIT transport direction and distance Standard, non-automated methods were used to
trajectories (1, km) identify the most important meteorological vari-
X, Y, and Z components of the 24-h HYSPLIT ables. The selection process included ‘‘backward
trajectory
24-h scalar wind run (m)
one variable deletion’’ based on the F-statistic
(Venables and Ripley, 2002) along with diagnostic
Synoptic Average morning and afternoon fractional checks such as the examination of model residuals.
weather cloud cover (%)
Variables that were highly correlated with one
Total precipitation (in)
Binary indicators of the occurrence of rain, another and those which offered little explanatory
haze, and fog power are excluded early in the screening process
(Harrell, 2001). The screening process is applied for
each urban area separately and examined for
(http://www.epa.gov/ttn/airs/airsaqs/index.htm). consistency among all 39 areas. For example, daily
For consistency with EPA’s ozone National Ambi- maximum 1-h temperature was statistically signifi-
ent Air Quality Standards (NAAQS), the daily cant for 36 of the 39 urban areas and therefore was
ARTICLE IN PRESS
7130 L. Camalier et al. / Atmospheric Environment 41 (2007) 7127–7137

retained. The screening process resulted in a subset 4. Meteorological effects on ozone


of eight variables from the list of candidate
meteorological variables (Table 1) and this subset The parameter estimates obtained from the
was then used in all subsequent modelings. The model provide insight into the nature of the
eight variables used in the model are listed in ozone’s response to each meteorological variable.
Table 2. For each meteorological parameter, the The individual effects of the most important
‘‘Number of cities’’ column represents the number meteorological parameters on ozone are exa-
of cites for which the parameter is statistically mined for an example area (Cleveland, OH),
significant at the a ¼ 0.001 level. followed by an examination of the spatial distribu-
The predictive power of the model as measured tion of the ozone response among all 39 urban
by the R2 statistic, ranges from 0.56 (Tampa) to areas. Next, the impact on ozone trends is exa-
0.80. Fig. 2 shows an interpolated surface of the R2 mined with a discussion of how inter-annual
statistic for each urban area computed by using all meteorological variations affect the adjustment
of the predicted and observed daily maximum 8-h process.
ozone values from the 9-year period.
4.1. Individual meteorological effects on ozone

Each variable plays a unique role in explaining


Table 2 variations in ozone through its own particular
Meteorological parameters used in the model and frequency of response, or effect. For example, increasing tem-
significance perature is usually associated with increasing ozone,
while increasing wind speed is usually associated
Meteorological parameters Number of cities
(out of 39)
with decreasing ozone (i.e. dilution effect). Because
many of the meteorological effects are non-linear, a
Daily maximum temperature ( 1C) 36 partial response plot offers an intuitive way to
Mid day average (10 am–4 pm average) 37 reveal the relationship between ozone and meteor-
relative humidity (%)
Morning (7–10 am) average wind speed 16
ological variables. A partial response curve shows
(m s1) the effect of a particular meteorological variable on
Afternoon (1–4 pm) average wind speed 27 ozone after accounting for the effects of all the other
(m s1) variables. Thus, the partial response curve accounts
Morning surface temperature difference 30 for any inter-correlation that may exist among the
(1200 UTC) (temperature at
925 mb–temperature at surface) ( 1C)
explanatory variables. Fig. 3 displays the partial
Deviation in 1200 UTC temperature of 850 17 response of ozone to each of four meteorological
mb surface from 10-year monthly average variables using data for the Cleveland urban area.
( 1C) Because a log-link function is used in the fitting
Transport direction (degrees clockwise from 37 process, the y-axis represents the log of ozone
North)
Transport distance (km) 35
concentrations after adjusting for the overall
average.
As visible in the upper-left quadrant in Fig. 3,
the effect of temperature is relatively small for
temperatures below the threshold of 20 1C, but
is very pronounced above that point. The individual
response curves for relative humidity and transport
distance are decreasing and approximately log-
linear over the range of data. Since transport
distance measures the distance traveled by the air
mass within the 24-h period, larger transport
distances are associated with higher transport wind
speeds, which act in the dilution of ozone and
precursor gases. As higher humidity levels are
Fig. 2. Spatial Interpolation of the R2 statistic for predicted vs. usually associated with greater cloud abundance
observed daily maximum 8-h ozone. and atmospheric instability, the photochemical
ARTICLE IN PRESS
L. Camalier et al. / Atmospheric Environment 41 (2007) 7127–7137 7131

Fig. 3. Partial response of ozone to selected meteorological parameters—Cleveland, OH. Dashed lines are 95% confidence bounds for the
response.

process is slowed and ground level ozone is


depleted. The ozone response to transport direction
in Cleveland is nearly uni-modal. The mode
indicates the direction from where the highest
incremental impact on ozone occurs. In this case,
the highest incremental impact on ozone is asso-
ciated with transport winds originating in the south
and southwest, while the lowest impact occurs from
winds originating from the north. The 95%
confidence bounds on each partial effect plot are
very narrow, indicating greater certainty in the
predicted response in the more dense portions of
the data.
It is interesting to compare the sensitivity of
ozone obtained from this particular study with
Fig. 4. Percent change in ozone 1 1C1 increase in daily
similar sensitivity results obtained from studies maximum 1-h temperature.
using a numerical air quality model (Dawson
et al., 2007). As noted, an example given in the
Dawson article based on a July 2001 episode 4.2. Spatial distribution of meteorological effects on
indicates that daily maximum 8-h ozone concentra- ozone
tion in Atlanta increases by approximately 4 ppb
2.5 K1 (1.6 ppb K1). This result compares Although each urban area is modeled indepen-
favorably with the sensitivity to temperature found dently, there are distinct spatial patterns in the
from this study. For example, the ozone response to response of each of the most important meteorolo-
temperature in Atlanta is approximately 3.4% K1 gical predictors. Since the ozone response for most
(Fig. 4). Since the seasonal average ozone in Atlanta of the variables has at least some degree of non-
for 2001 is approximately 60 ppb, the ozone linearity, presenting a spatial picture of the entire
sensitivity to temperature equates to 2 ppb in- non-linear response is difficult. As an alternative, a
crease in ozone K1. linear approximation of the effect of a particular
ARTICLE IN PRESS
7132 L. Camalier et al. / Atmospheric Environment 41 (2007) 7127–7137

meteorological variable on ozone is calculated using


the central 50% of the data. The approximate linear
effect (i.e. ozone response) is defined as the
difference between the ozone predicted by the model
at the 75%-tile and the 25%-tile of the meteorolo-
gical variable, divided by the difference between
these two percentiles. The approximate linear effect
of maximum temperature on ozone is illustrated as
a slope in Fig. 5. The effect, or slope, is thus an
approximate rate of change in ozone that is
predicted to occur as the maximum temperature
varies from a typical low to typical high value. The
effect can also be expressed as percent change in
ozone per a 1 1C increase in temperature (%
change ¼ slope  100). Fig. 6. Percent change in ozone per 1% increase in mid-day
Fig. 4 shows the spatial distribution of the relative humidity.
approximate linear effect of temperature on ozone,
expressed as percent change in ozone 1 1C1
increase in temperature. Generally, the temperature
effect is positive and clearly has the largest impact in
the North and Northeast portions of the domain. In
the vicinity of the Great Lakes, a 1 1C increase in
temperature is associated with about a 4% increase
in ozone. Urban areas along the eastern seaboard
have the largest ozone response to temperature
(approximately 5%). The magnitude of the tem-
perature effect gradually decreases southward and
diminishes to below 1% 1C1 in the Gulf Coast
regions. Fig. 6 shows the spatial distribution of the
approximate linear effect of relative humidity on
ozone. The relative humidity effect is largely
negative with greatest impact in the southern urban
Fig. 7. Effect of 24-h transport wind on ozone concentration.
areas and less pronounced for the more northern
The arrows show the transport wind direction associated with the
largest incremental increase in ozone; the length of the arrow is
proportional to the increase in ozone concentration accompany-
ing the indicated direction.

urban areas. Differing spatial dependency of me-


teorological effects could be due to the fact that the
maximum temperature varies more in the northern
areas as they experience summertime cold fronts,
which typically do not reach further south.
The direction of transport can play an important
role in the formation and movement of ozone. To
characterize the effects of the transport direction on
ozone, Fig. 7 is used to show the spatial pattern of
the directional effects among the urban areas. The
arrows on the map in Fig. 7 indicate the direction of
transport wind associated with the largest incre-
mental increase in ozone concentration; transport
Fig. 5. Schematic for the calculation of an approximate linear winds indicate the direction from which the wind is
effect of maximum temperature on ozone. coming, as they are based on backward trajectories.
ARTICLE IN PRESS
L. Camalier et al. / Atmospheric Environment 41 (2007) 7127–7137 7133

The length of each arrow is proportional to the


magnitude of the incremental increase (or effect).
Referring back to the bottom-left quadrant of Fig. 3
for Cleveland, the highest incremental ozone is
associated with transport winds originating from
the southwest (200–2501, clockwise from North).
Overall, the spatial pattern shows that urban
areas with the largest response to transport winds
are located along the periphery of the domain, while
urban areas in the central portion of the domain are
associated with the smallest directional impact
(Tennessee, Kentucky, Ohio and West Virginia). A
general explanation for this pattern is that the
largest ozone producing sources of NOx emissions
(US Environmental Protection Agency, 2005) are
Fig. 8. Partial response of ozone to transport wind direction for
located near the central portion of the eastern US,
Gulf Coast urban areas demonstrates the potential for multiple
where air masses are more stagnant and individual directions of high impact.
transport directions are less important. The peri-
meter urban areas, including the Northeastern
US, upper Midwest, and Gulf Coast, appear to (bands of color). The information conveyed in
indicate at least some impact of transport from Fig. 9 confirms the existence of spatially consistent
the central portion of the domain (CAIR Modeling patterns in the aggregate effects of the meteorolo-
Analysis, http://www.epa.gov/cair/pdfs/finaltech02. gical drivers on ozone. As previously seen by the
pdf); an exception may be the southeast Texas and spatial effects maps, day-to-day variations in
Louisiana Gulf Coast. temperature play the most significant impact on
As noted previously, only the transport direction ozone in the north while day-to-day variations in
associated with the largest incremental impact is relative humidity have a more dominant effect on
indicated in Fig. 7. While the figure is convenient for ozone in the south. Both relative humidity and
conveying the overall pattern, the one arrow temperature play comparable roles in the mid-
approach can mask variations in directional impact portion of the eastern US. The outskirts/coasts of
especially in urban areas that indicate impact from the extreme north and south are driven heavily by
more than one principal direction. For example, transport parameters.
urban areas in the Gulf coast region, including Other meteorological parameters are included in
Beaumont, TX and Baton Rouge, LA, show a bi- the model, as well as a term to account for day of
directional impact of transport direction (Fig. 8). week effects. The spatial effects of these additional
One possible reason for this is the existence and parameters have been explored, and some interest-
location of large sources of VOC emissions (e.g., ing geographic patterns have been found. For
petrochemical processes around Houston and example, some geographic patterns of day of week
southern Louisiana). Finally, it is important to note effects have been observed. In the northeastern US
that individual ozone episodes may be driven by and the Great Lakes area, such as Detroit, week-
source–receptor linkages that are not reflected in the ends seem to be associated with higher ozone
average transport wind effect. concentrations while weekdays seem to be asso-
All variables selected to be in the model have at ciated with lower ozone values. Interestingly, in the
least some impact on ozone. However, because the southeastern US area, the day of week effect on
model is fit separately for each urban area, each city ozone appears to be reversed, with weekdays being
can have different prevailing meteorological para- more associated with higher ozone concentrations.
meters which drive the ozone response. The two
prevailing meteorological parameters for each city 4.3. Aggregate meteorological effects on ozone
displayed in Fig. 9 are those with the largest trends
F-statistics. The map synthesizes the dominating
meteorological parameters in each area, illustrating The original motivation for improving the statis-
the geographic zones of meteorological influence tical model described in this paper is to better
ARTICLE IN PRESS
7134 L. Camalier et al. / Atmospheric Environment 41 (2007) 7127–7137

Fig. 9. Geographic zones of dominating meteorological influence, based on the F statistics of the meteorological parameters.

understand ambient ozone trends, especially in


those urban areas affected by the implementation
of control programs (US Environmental Protection
Agency, 2005). Fig. 10 illustrates the effect of
meteorological adjustment on ozone concentrations
using the Cleveland urban area as an example. The
vertical axis is the daily maximum 8-h ozone value
for each ozone season between 1997 and 2005. The
dotted line represents the actual, observed concen-
tration data, while the solid line connects the
meteorologically adjusted seasonal ozone averages
(i.e. the term, Yk, in Eq. (1)). The standard error for
each of the meteorologically adjusted averages is
2–3%.
The meteorologically adjusted ozone trend shown Fig. 10. Observed vs. meteorologically adjusted ozone trends for
in Fig. 10 is generally smoother than the observed Cleveland, OH.
ozone trend, showing less of the inter-annual
variability that is mainly caused by inter-annual
fluctuations in meteorology. The direction and Anomalies during the ozone season for relative
magnitude of the adjustments in Fig. 10 can be humidity in 2004 and 2005 are also noticeably
largely explained by the direction and magnitude of different. Average humidity values in 2004 are much
the relative humidity and temperature anomalies higher than the 9-year average, especially in the
from year to year. To illustrate, Fig. 11 shows the Midwest and Northeast areas. The relative humidity
relative humidity and temperature anomaly for 2004 anomaly in 2005 is in general, opposite and more
and 2005 for each urban area, where the anomaly is extreme than the 2004 anomaly, where average
calculated as the difference between the mean for a humidity in 2005 is much lower than the composite
given year’s ozone season and the 9-year ozone 9-year average. The combined effect of higher-than-
season average. The average temperature in 2004 is average temperature and lower-than-average
generally lower than the 9-year average, especially humidity in 2005 contributes to the downward
in the Midwest and Great Lakes regions. Tempera- adjustment in ozone in 2005. Similarly, the com-
ture anomalies in the following year contrast with bined effect of lower-than-average temperature and
2004, with 2005 temperatures warmer than the 9- higher-than-average humidity in 2004 contributes to
year average. the upward adjustment in ozone.
ARTICLE IN PRESS
L. Camalier et al. / Atmospheric Environment 41 (2007) 7127–7137 7135

Fig. 11. Temperature and relative humidity anomalies for the ozone seasons of 2004 and 2005. Each anomaly is measured as the difference
between the average for the given year and the 9-year average. Negative values are lower than average and positive values are higher than
average.

Fig. 12. Meteorological adjustment to ozone (%). Positive values indicate an upward adjustment and negative values indicate a downward
adjustment in ozone.

Because meteorological effects on ozone generally ozone. Positive values indicate an upward adjust-
occur on a regional scale, adjustments within the ment while negative values indicate a downward
same general geographic area are expected to be adjustment. The adjustment values for 2004 were
similar. To examine this issue, a yearly adjustment predominately positive, which means that 2004
percentage is calculated for each urban area and seasonal ozone averages were adjusted upward in
displayed geographically (Fig. 12). The adjustment most urban areas. Overall, the adjustments in 2004
percentage is calculated via range from 4 to 8% throughout the central and
north central portions of the domain. In contrast,
Oadj  Oraw
adjustment% ¼  100, adjustments in 2005 were mostly negative with the
Oraw
largest adjustments in the Midwest and Great Lakes
where Oadj is the meteorologically adjusted ozone region. The smallest downward adjustments are
(i.e. the year effect) and Oraw is the raw average generally confined to the southeast.
ARTICLE IN PRESS
7136 L. Camalier et al. / Atmospheric Environment 41 (2007) 7127–7137

Fig. 13. Comparison of average ozone between 2004 and 2005 (top two quadrants). The bottom two quadrants show the percent
difference in ozone between 2005 and 2004, before and after meteorological adjustment. Since the standard error, obtained from
bootstrapping, is 2–3%, the resulting margin of error for percent change is 5 percentage points.

Fig. 13 illustrates the importance of using from 2004 to 2005 is driven mainly by meteorolo-
meteorological adjustment methods when trying to gical differences between the two years and not due
interpret differences in air quality levels between 2 to a fundamental shift in air quality.
years. The top two quadrants show the seasonal
average of daily maximum 8-h ozone for 2004 and 5. Concluding remarks and future applications
2005. Ozone concentrations in 2005 are much higher
than ozone concentrations in 2004. In 2004, ozone In 2004, the National Research Council issued a
values higher than 60 ppb are confined to a small report entitled Air Quality Management in the
region; while in 2005, ozone values greater than 60 United States (NRC, 2004), providing several
ppb dominate most of the domain. The bottom two recommendations regarding improved decision-
quadrants show the percent difference between 2005 making in the context of environmental health and
and 2004 in average ozone, before and after air quality. One of the major recommendations is to
meteorological adjustment. Percent differences in better track air quality progress in order to enable
observed ozone are most pronounced in the western more informed evaluations of past and present air
portion of the domain, where differences peak policy decisions. In order to track progress toward
above 20%. After adjusting for meteorology, air quality goals, routine air quality trend analyses
percent differences between the 2 years are insignif- are needed to confirm if emission controls are
icant. To estimate the significance of the percent indeed reducing pollutant concentrations.
changes shown in Fig. 13, a 3-day, non-overlapping, Meteorological adjustment of ozone concentra-
blocked bootstrap (Davison and Hinkley, 1999; tions provides a method to examine the underlying
Hastie and Tibshirani, 1990) is used to estimate the effects of control programs apart from the random
standard error (2–3 percentage points) of the inter-annual effects of meteorology. The benefits of
percent change in ozone. Only a few urban areas accounting for meteorological conditions when
have adjusted changes which exceed twice the examining air quality trends have been demon-
bootstrap estimated standard error. It can be strated in numerous publications over the past
concluded that the increase in ozone concentrations decade. This paper expands on previous analyses
ARTICLE IN PRESS
L. Camalier et al. / Atmospheric Environment 41 (2007) 7127–7137 7137

through an improved regression method as well as Davis, J.M., Eder, B.K., Nychka, D., Yang, Q., 1998. Modeling
the inclusion of several new variables provided by the effects of meteorology on ozone in houston using cluster
the HYSPLIT trajectory model. analysis and generalized additive models. Atmospheric
Environment 32 (14/15), 2505–2520.
The authors are beginning to explore the applica- Davison, A.C., Hinkley, D.V., 1999. Bootstrap Methods and
tion of similar methods for quantifying the effects of their Application. Cambridge University Press, Cambridge.
meteorological conditions on PM2.5 and PM2.5 Dawson, J.P., Adams, P.J., Pandis, S.N., 2007. Sensitivity of
components, specifically sulfates, nitrates and or- ozone to summertime climate in the eastern USA: a modeling
ganic carbon. Such pollutants are much more case study. Atmospheric Environment 41, 1494–1511.
Draxler, R.R., Hess, G.D., 1997. Description of the HYSPLIT_4
complex than ozone and thus present unique Modeling System. NOAA Technical Memorandum ERL
challenges for statistical modeling and for gaining ARL-224. Air Resources Laboratory, Silver Spring, Mary-
a more complete understanding of the effect of land.
meteorology on PM2.5 concentrations. Hastie, T.J., Tibshirani, R.J., 1990. Generalized Additive Models.
There is a growing interest in understanding and Chapman & Hall, New York.
Harrell, F.H., 2001. Regression Modeling Strategies. Springer,
quantifying the global effects of climate change on New York, Inc.
the environment. Advances in statistical modeling of McCullagh, P., Nelder, J.A., 1989. Generalized additive models.
the relationship between air quality and meteorology Chapman & Hall/CRC, London.
should help corroborate the results being obtained NRC, 2004. Air Quality Management in the United States.
National Academy Press, Washington, DC.
from numerical simulations which predict the re-
R Development Core Team, 2006. R: A Language and
sponse of air quality to changes in climate conditions. Environment for Statistical Computing. R Foundation for
The knowledge gained by contrasting the simulated Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0,
and observed response to meteorology should URL /http://www.R-project.orgS.
provide additional insight into the likely long-term Thompson, M.L., Reynolds, J., Cox, L.H., Guttorp, P.,
effects of climate change on air quality levels. Sampson, P.D., 2001. A review of statistical methods for
the meteorological adjustment of tropospheric ozone. Atmo-
spheric Environment 35, 617–630.
US Environmental Protection Agency, 2004. The Ozone Report:
References Measuring Progress through 2003, EPA 454/K-04-001,
Washington, DC.
Bloomfield, P.J., Royle, J.A., Steinberg, L.J., Yang, Q., 1996. US Environmental Protection Agency, 2005. Evaluating Ozone
Accounting for meteorological effects in measuring urban ozone Control Programs in the Eastern United States, EPA 454-K-
levels and trends. Atmospheric Environment 30 (17), 3067–3077. 05-001, Washington, DC.
Cox, W.M., Chu, S., 1993. Meteorologically adjusted ozone Venables, W.N., Ripley, B.D., 2002. Modern Applied Statistics
trends in urban areas: a probabilistic approach. Atmospheric with S. Spring, New York, Inc.
Environment 27B, 425–434. Zheng, J., Swall, J., Cox, W.M., Davis, J., 2006. Interannual
Cox, W.M., Chu, S., 1996. Assessment of Interannual ozone variation in meteorologically adjusted ozone levels in the
variation in urban areas from a climatological perspective. Eastern United States: A comparison of two approaches.
Atmospheric Environment 30 (14), 2615–2625. Atmospheric Environment.

You might also like