[NOTE: Figures are available from WALCOFF
(703) 267-3594]


Final Report

CLASSIFICATION AND REGRESSION TREE ANALYSIS TO SUPPORT THE SELECTION OF EPISODES FOR REGIONAL-SCALE PHOTOCHEMICAL MODELING OF THE SOUTHEASTERN U.S.

SYSAPP-95/084

December 1995

Prepared for Mr. John J. Jansen
Southern Company Services, Inc.
600 North 18th Street, 14N-8195
Birmingham, AL 35202-2625

Prepared by Sharon G. Douglas, Hans P. Deuel, and Jay L. Haney
Systems Applications International
101 Lucas Valley Road
San Rafael, CA 94903
415-507-7100


ACKNOWLEDGMENTS
The following SAI technical staff are acknowledged for their contributions to the project (in alphabetical order): Zitian Guo, A. Belle Hudischewskyj, and Nina K. Lolk.

Copyright 1997 by Systems Applications International, Inc.


Contents


Figures

[Note by Webmaster: All figures are available in hardcopy from WOLCOFF.]


1 Wind direction bins used for CART analysis
2 CART classification tree for the Atlanta urban area
3 Average maximum ozone for each year in node D1 for the Atlanta classification
4 Average maximum ozone for each year in node C5 for the Atlanta classification
5 Average maximum ozone for each year in nodes B2, B4, and B6 for the Atlanta classification
6 Average maximum ozone for each year in nodes A2 and A5 for the Atlanta classification
7 CART classification tree for the Charlotte urban area
8 CART classification tree for the Birmingham urban area
9 CART classification tree for the Nashville urban area
10 Daily maximum ozone concentrations for each of the four urban areas during the July 1993 episode .
11 Observed maximum ozone concentrations for 20–30 July 1993

Tables


1 Air quality variables included in the CART analysis
2 Surface meteorological variables included in the CART analysis
3 Upper-air meteorological variables included in the CART analysis
4 Meteorological and air quality monitoring sites included in the CART analysis
5 CART input/controlling parameters
6 Misclassification costs for Atlanta, Birmingham, and Nashville CART applications
7 Misclassification costs for Charlotte CART application
8 Meteorological and air quality conditions associated with ozone maxima ³14 pphm in the Atlanta area
9 Meteorological and air quality conditions associated with ozone maxima between [11, 14) pphm in the Atlanta area
10 Meteorological and air quality conditions in the Charlotte area giving rise to ozone maxima ³14 pphm
11 Meteorological and air quality conditions in the Charlotte area giving rise to ozone maxima in the [11, 14) pphm range
12 Meteorological and air quality conditions in the Birmingham area giving rise to ozone maxima ³14 pphm (A1) and [11, 14) pphm (B1)
13 Meteorological and air quality conditions in the Nashville area giving rise to ozone maxima >14 pphm (A1) and in the [11,14) pphm range (B1)
14 Candidate modeling episode days corresponding to those categories or nodes characterized by meteorological conditions most frequently associated with ozone exceedances

INTRODUCTION
This report documents the episode selection analysis that was performed to support the identification of a regional-scale ozone episode (with emphasis on the southeastern U.S.) to be modeled as part of the Ozone Transport Assessment Group (OTAG) super-regional modeling effort. The objectives of the episode selection analysis were to identify episodes:

 

That are characterized by representative (i.e., occurring with some frequency) episodic air quality and meteorological conditions

That are likely to provide useful quantitative information regarding the effects of future-year emissions changes (i.e., given the known limitations of current state-of-the-science meteorological and photochemical models it is expected that reasonable model performance will be achieved)

 

That are within the context of OTAG, representative of the range of air quality and meteorological conditions under which different attainment/control strategies will be indicated.

 

For this analysis, we applied the Classification and Regression Tree (CART) analysis technique to examine the characteristics of ozone episodes within the southeastern U.S. and to identify possible candidate episodes for regional-scale photochemical modeling. Air quality and meteorological data corresponding to the nine-year period 1985–1993 were included in the analysis.

The CART analysis technique is a binary splitting method which partitions a data set into discrete subgroups based on the value of a user-defined classification variable. The remaining variables in the database are selected as to whether or not they provide a predictive segregation of the remaining data for different values of the classification variable. The final result is a set of series of splits on values of the variables, each series of which leads to a classification value of the dependent variable.

For this application, the CART analysis technique was applied to the analysis of data for four nonattainment areas in the Southeast (Atlanta, Georgia; Charlotte, North Carolina; Birmingham, Alabama; and Nashville, Tennessee). In addition to the local meteorological and air quality variables for each of these areas, regional-scale variables reflecting the maximum ozone concentrations observed within the other urban areas on consecutive days were also included in the analysis.

The results of the CART analyses for the four urban areas were reviewed with respect to (1) whether certain multi-day episode periods would constitute appropriate modeling episodes for one or more of the areas of interest and (2) the potential (as indicated by the observed ozone concentrations, the upper-air airflow patterns, and the CART transport indicators) for intra- and/or inter-regional pollutant transport. Based on this assessment, several multi-day periods were identified as potential regional-scale modeling episode candidates. The attributes and deficiencies of each were considered, and a single episode (20–30 July 1993) was selected.

The remainder of the report provides an overview of the CART analysis technique, a summary of the data and procedures used for this exercise, the CART analysis results for the Atlanta, Charlotte, Birmingham, and Nashville area applications, the results of the episode selection process, and a description of the 20–30 July 1993 ozone episode.


OVERVIEW OF THE CART METHODOLOGY

CART (Classification and Regression Tree) analysis is a binary splitting method which partitions a data set into discrete subgroups based on the value of a user-defined classification variable. The remaining variables in the database are selected as to whether or not they provide a predictive segregation of the data between different values of the classification variable. This restriction presupposes that there is a causal relationship between the independent variables and the dependent variables. Consequently, it is necessary to construct a database of independent variables upon which the dependent variable is likely to depend.

The dependent variable in this case is a representation of daily ozone concentration. The classification technique depends upon the dependent variable being discrete, although the independent variables may be continuous. The classification variable employed in the present analysis was based on the value of the maximum observed ozone concentration for a particular urban area of interest. Specifically, this variable was assigned a value of 1 to 4, corresponding to a maximum ozone concentration of <8, [8,11), [11,14), or 14 pphm. It was assigned a value for each day, where the maximum was chosen from observations from a network of air quality stations (specified below).

A database consisting of meteorological variables and other air quality parameters was constructed. The data constituted the independent variables. The meteorological data were taken from surface and upper-air meteorological monitoring stations. To enable CART to distinguish same-day local ozone formation from recirculation or transport of ozone, meteorological variables corresponding to the day prior to that under consideration were also included in the analysis.

The maximum ozone concentration observed in other urban areas throughout the Southeast was also included in the analysis as an indicator of possible intra-regional ozone transport. Concentrations for up to three days prior to the day under consideration were utilized.

Databases were constructed for four urban areas: Atlanta, Georgia; Charlotte, North Carolina; Birmingham, Alabama; and Nashville, Tennessee. Each of these databases consisted of meteorological variables characterizing the prevailing conditions in the local area for each day, variables characterizing the meteorological conditions in the local area on the previous day, observed ozone concentrations for the local area on the previous day, and variables characterizing the observed ozone concentrations in the vicinity of the other urban areas on the previous three days. The variables constructed for each of the four urban areas were drawn from data collected from air quality monitoring stations, surface meteorological monitoring stations, and upper-air meteorological monitoring stations. The exact variables and the monitoring stations employed for each database are specified below.

Air Quality Variables

The air quality variables used in the CART analysis for each urban area are defined in Table 1.

TABLE 1. Air quality variables included in the CART analysis.

 Variable Namea

 

Description

Amax1

Maximum ozone concentration observed within the urban area the previous day.

Ansta121

Number of air quality monitoring sites in the urban area for which ozone concentrations ³12.0 pphm were observed the previous day.

Amx1(city)

Maximum ozone concentration observed within the urban area the previous day.

Amx2(city)

Maximum ozone concentration observed within the urban area 2 days ago.

Amx3(city)

Maximum ozone concentration observed within the urban area 3 days ago.

AmaxbinE

The classification variable: a value of 1,2,3, or 4 depending on whether the maximum ozone concentration over all sites in the urban area on the present day was <8, [8,11), [11,14), or ³14 pphm, respectively.

a "(city)" corresponds to Atlanta, Charlotte, Birmingham, or Nashville, where the three possibilities chosen are distinct from the urban area under consideration.

 

 

Meteorological Variables

The surface meteorological variables used in the CART analysis for each urban area are defined in Table 2, and the upper-air meteorological variables used in the CART analysis are defined in Table 3. The wind direction bins are illustrated in Figure 1.

Summary of Monitoring Sites

Extensive data sets were created for the Atlanta, Charlotte, Birmingham, and Nashville urban areas. The data set for each of these cities consisted of data from either four (Atlanta) or three (Charlotte, Birmingham, Nashville) air quality monitoring stations, one upper-air meteorological station, and one surface meteorological monitoring site. Each of the above variables were tabulated for each day from April through October for the years 1985 through 1993, inclusive. The station identifiers and locations are summarized in Table 4. These sites were selected based on an examination of data completeness for the nine- year period.

CART Parameters

The method employed by the CART software for minimizing misclassification and optimizing the robustness of the derived classification scheme depends on how

TABLE 2. Surface meteorological variables included in the CART analysis.

Variable Name

Description

Swsp69

Average surface wind speed (m/s) from 0600 to 0900 LST, inclusive.

Swsp1013

Average surface wind speed (m/s) from 1000 to 1300 LST, inclusive.

Swsp1417

Average surface wind speed (m/s) from 1400 to 1700 LST, inclusive.

Swbn69

Wind direction bin value of 1 through 12, indicating that the average surface wind direction from 0600 to 0900 LST, inclusive, was predominantly from 0±15°, 30±15°,...,330±15°, respectively.

Swbn1013

As above, but from 1000 to 1300 LST, inclusive.

Swbn1417

As above, but from 1400 to 1700 LST, inclusive.

SmxT0

Maximum surface temperature (°C) for the present day.

SmxT1

Maximum surface temperature (°C) for the previous day.

Srhum

Surface relative humidity at noon.

Sdewpt

Surface dew point temperature (°C) at noon.

Scloud

A value of 0 or 1, indicating that the opaque cloud cover was £0.5 or >0.5, respectively.

Spres0

Maximum surface pressure on the present day.

 

TABLE 3. Upper-air meteorological variables included in the CART analysis.

 Variable Namea

Description

U(pmb)wbnam

Wind direction bin value of 1 through 12, indicating that the upper-air (pmb) wind direction corresponding to the morning sounding was predominantly from 0±15°, 30±15°,..., 330±15°, respectively.

U(pmb)wbnpm

Identical to above, but for the afternoon sounding.

U(pmb)wspam

Upper-air (pmb) wind speed corresponding to the morning sounding.

U(pmb)wsppm

Upper-air (pmb) wind speed corresponding to the afternoon sounding.

U(pmb)Tam1

Upper-air (pmb) temperature corresponding to the morning sounding on the previous day.

U(pmb)Tpm1

Upper-air (pmb) temperature corresponding to the afternoon sounding on the previous day.

U(pmb)Tam0

Upper-air (pmb) temperature corresponding to the morning sounding on the current day.

U(pmb)Tpm0

Upper-air (pmb) temperature corresponding to the afternoon sounding on the current day.

U(pmb)rhumam

Upper-air (pmb) relative humidity corresponding to the morning sounding on the current day.

U(pmb)rhumpm

Upper-air (pmb) relative humidity corresponding to the afternoon sounding on the current day.

U(pmb)hgtam

Height above sea level of the (pmb) surface as indicated by the morning sounding.

U(pmb)hgtpm

Height above sea level of the (pmb) surface as indicated by the afternoon sounding.

U9Tamdif

Difference between the surface and the 900 mb temperature using the morning temperature sounding data.

a (pmb) is either 700 mb or 850 mb.

FIGURE 1. Wind direction bins (inner circle of numbers) used for

CART analysis. Outer circle shows wind directions.

TABLE 4. Meteorological and air quality monitoring sites included in the CART analysis.

Primary City

Station Type

Mnemonic ID

Station IDa

Atlanta, GA

Air Quality

South DeKalb

130890002

 

Air Quality

Conyers-Monastery

132470001

 

Air Quality

MLK Marta Station

131210053

 

Air Quality

Sweetwater

130970002

 

Surface Met.

Atlanta, GA NWS

13874

 

Upper Met.

Athens, GA NWS

13873

Charlotte, NC

Air Quality

Plaza Road and Lakedell

371190034

 

Air Quality

Westinghouse Blvd.

371191005

 

Air Quality

Mecklenburg Cab Co.

371191009

 

Surface Met.

Charlotte/Douglas Airport

13881

 

Upper Met.

Greensboro, NC NWS

13723

Birmingham, AL

Air Quality

Fairfield Fire Dept.

0107310037

 

Air Quality

Pinson High School

010735002

 

Air Quality

Tarrant Elementary School

010736002

 

Surface Met.

Birmingham Municipal Airport

13876

 

Upper Met.

Centerville, AL NWS

03881

Nashville, TN

Air Quality

Trinity Lane

470370011

 

Air Quality

Percy Priest Visitor’s Center

470370026

 

Air Quality

Rockland Rec. Area

471650007

 

Surface Met.

Nashville Metro Airport

13897

 

Upper Met.

Nashville Metro Airport

13897

a For air quality stations, the AIRS identifier is given, while for surface and upper air meteorological stations, the WBAN identifier is given.

Misclassification and robustness are measured. These measures are a function of a series of user-definable parameters, the settings for which are listed in Table 5. A complete description of these parameters can be found in the on-line documentation for the software, as well as the text by Brieman et al. (1984). The misclassification costs for the Atlanta, Birmingham and Nashville datasets are shown in Table 6.

Good classification of the exceedance days was not obtained for Charlotte using the misclassification costs given in Table 6, possibly due to the distribution of days within each category. To accommodate this, the misclassification costs were modified for the Charlotte dataset and are shown in Table 7.

TABLE 5. CART input/controlling parameters.

 Parameter

Setting

Construction rule

Ordered twoing

Estimation method

Four-fold cross validation

Class priors

Equal

Tree selection

Either 0.1 or 0.01 standard error rule. In all cases, the "best" tree was obtained.

Linear combinations

Not allowed

Maximum number of surrogates

20

Misclassification costs

In general, a linearly increasing series giving greater cost for greater misclassification. In addition, there is an increased cost across the non-exceedance/exceedance transition.

 

 

TABLE 6. Misclassification costs for Atlanta, Birmingham, and Nashville CART applications.

 

 

True Class

Cost If Classified As

 

1 : <8

2 : [8,11)

3 : [11,14)

4 : ³14

1 : <8

0.0

0.5

1.5

3.0

2 : [8,11)

1.5

0.0

1.5

3.0

3 : [11,14)

3.0

1.5

0.0

1.5

4 : ³14

3.0

1.5

0.5

0.0


CART ANALYSIS RESULTS

The CART analysis results for each of the four urban areas are presented and described in this section. For each area, certain meteorological and air quality variables are identified as being important indicators of high ozone concentrations within the area. Such indicators include high surface temperature (which enhances the rate of ozone

 

TABLE 7. Misclassification costs for Charlotte CART application.

True Class

Cost If Classified As

 

1 : <8

2 : [8,11)

3 : [11,14)

4 : ³14

1 : <8

0.0

1.0

1.0

3.0

2 : [8,11)

1.0

0.0

1.0

3.0

3 : [11,14)

3.0

1.0

0.0

1.0

4 : ³14

3.0

1.0

1.0

0.0

production and is reflective of the amount of incoming solar radiation), clear skies or light cloud cover conditions (which optimize the amount of solar insolation and thus enhance the photochemical processes), low relative humidity (which tends to limit afternoon convective activity), low surface wind speeds (which are associated with stagnation or limited ventilation conditions), high 850 mb temperatures (which are indicative of stable, or limited vertical mixing, conditions), and high pressure, as indicated by the height of the 850 or 700 mb surface (which results in subsidence, limited vertical mixing, high temperatures, low wind speeds, and low relative humidity). Other indicators refer to the previous days' conditions and include high surface temperatures on the previous day (which may extend the lifetime of certain precursor pollutants), high ozone concentrations in the urban area of interest on the previous day (which is suggestive of day-to-day carryover or recirculation of ozone), and high ozone concentrations in another urban area on preceding days (which may be indicative of inter-urban ozone transport).

Application for Atlanta

Of the variables listed above, CART found 14 upon which the data set depended most strongly. These were the maximum ozone concentration in the Atlanta area on the previous day (Amax1), maximum observed ozone concentration in Birmingham on the previous day (Amx1birm), maximum surface temperature on the current day and the previous day (SmxT0 and SmxT1), 850 mb evening temperature (U8Tpm0), height of the 850 mb evening sounding (U8hgtpm), average surface wind speed from 1000 to 1300 LST and from 1400 to 1700 LST (Swsp1013 and Swsp1417), average surface wind direction from 1000 to 1300 LST (Swbn1013), 700 and 850 mb morning wind directions (U7wbnam and U8wbnam), cloud coverage (Scloud), relative humidity (Srhum), and month (month). The resulting classification based on these variables is illustrated in Figure 2.

The tree is characterized by five different regimes (labeled Ai, i=1,..,5 in Figure 2) which are associated with daily maximum ozone concentrations of 14 pphm or greater. The meteorological and air quality conditions characterizing each of these regimes are summarized in Table 8.

FIGURE 2. CART classification tree for the Atlanta urban area. Nodes identified as Ai correspond to regimes conducive to formation of maximum ozone concentrations ³14 pphm. Similarly, Bi, Ci, and Di indicate conduciveness to peak ozone formation in the 11 to 14 pphm range, the 8 to 11 pphm range, and less than 8 pphm, respectively. The Ai and Bi regimes are described in Tables 8 and 9.

There are nine distinct regimes (labeled Bi, i=1,...,9 in Figure 2) which are found to give rise to maximum ozone concentrations in the 11 to 14 pphm range. The meteorological and air quality conditions characterizing each of these regimes are summarized in Table 9.

Discussion
Of the total number of days included in the analysis (1926), exceedances of the National Ambient Air Quality Standard (NAAQS) for ozone (observed values greater than 12.0 pphm) were recorded in the Atlanta area on 125 of these days. The exceedance days are listed in the Appendix. Note that only data for those sites listed in Table 4 were considered in the determination of an exceedance day. By considering the classification of the exceedance days, the CART analysis results can be used to evaluate the representativeness of specific ozone episodes based on the frequency of occurrence of the associated episodic meteorological and air quality conditions. Those categories or terminal nodes that contain the greatest number of exceedance days are Node B2 (11 days), Node B4 (20 days), Node A2 (12 days), Node B6 (25 days), and Node A5 (20 days). Of the 125 exceedance days, 41 of these were characterized by maximum ozone concentrations greater than or equal to 14 pphm. Those categories or terminal nodes that contain the greatest number of exceedance days with maximum ozone concentrations greater than or equal to 14 pphm are Node A2 (10 days) and Node A5 (18 days). Given that the Atlanta area is classified as a serious ozone nonattainment area, only those episodes with concentrations greater than or equal to 14 pphm are considered to be appropriate modeling episodes (for control-strategy evaluation purposes). Thus, episode selection for the Atlanta area focused on episodes within Nodes A2 and A5.

TABLE 8. Meteorological and air quality conditions associated with ozone maxima ³ 14 pphm in the Atlanta area.

Node

Meteorological Conditions

A1

Previous day’s maximum ozone £9.28 pphm

 

Maximum surface temperature >25.2 °C

 

Surface relative humidity £59.5 %

 

Average surface wind speed from 1000 to 1300 LST £3.3 m/s

 

Previous day’s maximum surface temperature >29.1 °C

 

Month is April or June

 

Height of the evening sounding 850 mb surface £1540 m

A2

Previous day’s maximum ozone >9.28 pphm

 

Previous day’s maximum ozone £13.6 pphm

 

Average surface wind speed from 1000 to 1300 LST £3.68 m/s

 

700 mb morning wind direction from 1,2,3,5,9,10,11

 

Surface relative humidity £39.5%

 

Month is June, July, or August

A3

Previous day’s maximum ozone >9.28 pphm

 

Previous day’s maximum ozone £13.6 pphm

 

Average surface wind speed from 1000 to 1300 LST £3.68 m/s

 

700 mb morning wind direction from 1,2,3,5,9,10,11

 

Surface relative humidity >39.5%

 

Surface maximum temperature £36.9 °C

 

Previous day’s maximum ozone at Birmingham >12.4 pphm

A4

Previous day’s maximum ozone >9.28 pphm

 

Previous day’s maximum ozone £13.6 pphm

 

Average surface wind speed from 1000 to 1300 LST £3.68 m/s

 

700 mb morning wind direction from 1,2,3,5,9,10,11

 

Surface relative humidity >39.5%

 

Maximum surface temperature >36.9 °C

A5

Previous day’s maximum ozone >13.6 pphm

 

Average surface wind speed from 1400 to 1700 LST £3.87 m/s

 

850 mb evening temperature >19.4 °C

As indicated in Table 8, days classified as belonging to Node A2 are characterized by moderate (9.28 to 13.6 pphm) peak ozone concentrations in the Atlanta area on the previous day, low to moderate (<3.68 m/s) midday (1000 to 1300 LST) surface wind speeds, very low (<39.5%) surface relative humidity at midday, and generally north to northeasterly (occasionally westerly to northwesterly) winds aloft (700 mb) during the morning hours. Days identified as belonging to this category only occurred during the months of June, July, and August. Based on the previous day's ozone concentrations, days within this category are not isolated ozone episode days (i.e., they are part of a

TABLE 9. Meteorological and air quality conditions associated with ozone maxima between [11,14) pphm in the Atlanta area.

Node

Meteorological conditions

B1

Previous day’s maximum ozone £9.28 pphm

 

Maximum surface temperature >25.2 °C

 

Surface relative humidity £59.5%

 

Average surface wind speed for 1000 to 1300 LST £3.30 m/s

 

Previous day’s maximum surface temperature £29.1 °C

 

850 mb morning wind direction from 1,3,4,5,7

 

Surface wind direction from 1000 to 1300 LST from 1,3,10

B2

Previous day’s maximum ozone £9.28 pphm

 

Maximum surface temperature >25.2 °C

 

Surface relative humidity £59.5%

 

Average surface wind speed for 1000 to 1300 LST £3.30 m/s

 

Previous day’s maximum surface temperature >29.1 °C

 

Month is April or July through October

 

Surface cloud cover is >50%

 

850 mb morning wind direction from 1,2,3,6,7,8,9,11,12

B3

Previous day’s maximum ozone £9.28 pphm

 

Maximum surface temperature >25.2 °C

 

Surface relative humidity £59.5%

 

Average surface wind speed for 1000 to 1300 LST £3.30 m/s

 

Previous day’s maximum surface temperature >29.1 °C

 

Month is April or June

 

Height of the 850 evening sounding >1540 m

B4

Previous day’s maximum ozone >9.28 pphm

 

Previous day’s maximum ozone £13.6 pphm

 

Average surface wind speed from 1000 to 1300 LST £3.68 m/s

 

700 mb morning wind direction from 4,6,7,8,12

B5

Previous day’s maximum ozone >9.28 pphm

 

Previous day’s maximum ozone £13.6 pphm

 

Average surface wind speed from 1000 to 1300 LST £3.68 m/s

 

700 mb morning wind direction from 1,2,3,5,9,10,11

 

Surface relative humidity £39.5%

 

Month is April, May, September, or October

B6

Previous day’s maximum ozone >9.28 pphm

 

Previous day’s maximum ozone £13.6 pphm

 

Average surface wind speed from 1000 to 1300 LST £3.68 m/s

 

700 mb morning wind direction from 1,2,3,5,9,10,11

Continued

TABLE 9. Continued.

Node

Meteorological Conditions

 

Surface relative humidity >39.5%

 

Maximum surface temperature £36.9 °C

 

Previous day’s maximum ozone in Birmingham £12.4 pphm

 

Height of the evening sounding 850 mb surface >1540 m

B7

Previous day’s maximum ozone >9.28 pphm

 

Previous day’s maximum ozone £13.6 pphm

 

Average surface wind speed from 1000 to 1300 LST >3.68 m/s

 

850 mb evening temperature >20.0 °C

B8

Previous day’s maximum ozone >9.28 pphm

 

Previous day’s maximum ozone £13.6 pphm

 

Average surface wind speed from 1400 to 1700 LST £3.87 m/s

 

850 mb evening temperature £19.4 °C

B9

Previous day’s maximum ozone >9.28 pphm

 

Previous day’s maximum ozone £13.6 pphm

 

Average surface wind speed from 1400 to 1700 LST >3.87 m/s

multi-day ozone episode). The meteorological characteristics indicate high pressure and subsidence over the Atlanta area with the upper-level high-pressure system centered to the west (typically northwest) of the area. The importance of the morning 700 mb wind direction in defining this category suggests that transport of ozone and/or precursor pollutants into the Atlanta urban area is a distinguishing feature of days within this terminal node. Fourteen days are classified as belonging to Node A2; the expected frequency of occurrence of the meteorological and air quality conditions characteristic of this node is approximately 1.6 times per year. Note that 8 July 1988, one of the episode days being modeled as part of the Atlanta State Implementation Plan (SIP) and OTAG modeling efforts, is contained within this category. The maximum observed ozone concentration for 8 July 1988 (18.6 pphm), however, is significantly higher than the average peak concentration for properly classified (maximum ozone >14 pphm) days within this category (15.7 pphm).

As indicated in Table 8, days within Node A5 are characterized by high (>13.6 pphm) peak ozone concentrations in the Atlanta area on the previous day, low to moderate (<3.87 m/s) afternoon (1400 to 1700 LST) surface wind speeds, and relatively low 850 mb temperatures during the afternoon hours. Given the previous days' maximum concentrations, days within this category are part of a multi-day ozone episode. The light afternoon wind speeds coupled with the low 850 mb temperatures indicate stagnation but with vertical mixing. Day-to-day carryover or recirculation of ozone within the Atlanta area is suggested. Twenty-two days are classified as belonging to Node A5; the expected frequency of occurrence of the meteorological and air quality conditions characteristic of this node is 2.4 times per year. Note that 31 July 1987, one of the episode days being modeled as part of the Atlanta State Implementation Plan (SIP) modeling effort, is contained within this category. The maximum observed ozone concentration for 31 July 1987 (20.1 pphm), however, is significantly higher than the average peak concentration for properly classified (maximum ozone >14 pphm) days within this category (15.9 pphm).

Assessment of Accuracy of the Atlanta Classification

In a typical regression application, there exist standard simple performance measures, usually associated with minimization of a sum of residuals. These measures are not applicable to a CART classification analysis. CART predicts that each distinct regime should give rise to a particular value of the dependent variable. One can thus compare the expected and observed peak ozone concentrations for the days within each category. The primary concern in the present study is an accurate handling of the days on which high ozone concentrations were observed.

Of the days with observed peak ozone concentrations ³14.0 pphm, 90 percent were classified as having conditions that are conducive to ozone formation ³ 14.0 pphm. Similarly, 86 percent of the days with peak observed ozone concentrations in the 11 to 14 pphm range, 72 percent of the days within the 8 to 11 pphm range, and 43 percent of the days with peak concentrations <8 pphm were classified as having an expected maximum concentration in the respective range. Misclassification of a day on which the observed maximum concentration was high always involved an adjacent bin (e.g., if the observed value was >14, it was misclassified as having as peak concentration within the 11–14 pphm range). Very few of the higher observed ozone concentrations were misclassified. Note that there were 1739 days with peak ozone observations <11 pphm, out of a total database of 1926 days.

Ozone Trends in the Atlanta Region

The classification of exceedance and non-exceedance days with respect to meteorological conditions allows for an examination of the trends in peak ozone concentration for days which are meteorologically similar for the nine years from 1985 to 1993. This was investigated by choosing a particular regime or node and examining the observed ozone concentrations corresponding to days within this node for the nine year period. For each of these days, an average daily maximum ozone concentration for the Atlanta area monitoring sites was calculated. The calculated average daily maximum ozone concentration for all the days in a particular year within this meteorological regime were then averaged, and this value taken as the annual average ozone concentration for that particular year and category or regime. This was done for each regime defined in Figure 2.

The results clearly demonstrate that for meteorological conditions considered conducive to all but the highest ozone concentrations (i.e., all but the cases identified in Table 8), there is no discernible trend in ozone concentrations for the Atlanta region for the years examined. Figure 3 shows the annual trend associated with a meteorological regime (node D1 in Figure 2) considered conducive to formation of low peak ozone concentrations (<8 pphm). Figure 4 shows similar information for a typical meteorological regime (node C5 in Figure 2) conducive to peak ozone formation in the 8–11 pphm range.

FIGURE 3. Average maximum ozone for each year in node D1 for the Atlanta classification.

FIGURE 4. Average maximum ozone for each year in node C5 for the Atlanta classification.

Figure 5 shows the ozone concentration trends for three representative meteorological regimes conducive to peak ozone formation in the 11 to 14 pphm range. Figure 5a, 5b, and 5c correspond to node B2, B4, and B6, respectively from Table 9. While there is somewhat more variability in year-to-year values shown within these regimes, there is clearly no unambiguous increase or decrease throughout the years.

Figures 6a and 6b give similar information for meteorological regimes A2 and A5 of Table 8. In this case, the data corresponding to regime A2 indicate a downward trend, while those for regime A5 indicate an increase. However, due to the relatively small number of days within these regimes the yearly averages may not be representative.

Application for Charlotte
Classification of the different ozone-forming regimes for Charlotte was found to depend on 11 of the meteorological variables outlined above. These were maximum surface temperature (SmxT0), cloud coverage (Scloud), maximum observed ozone in Charlotte on the previous day (Amax1), maximum observed ozone in Atlanta and Birmingham on the previous day (Amx1atln and Amx1birm, respectively), maximum observed ozone three days ago in Nashville (Amx3nash), average surface wind speed from 1000 to 1300 LST (Swsp1013), average surface wind direction from 1400 to 1700 LST (Swbn1417), 700 mb morning wind direction (U7wbnam), and the height and humidity of the evening sounding 850 mb surface (U8hgtpm and U8rhumpm, respectively).

Four different meteorological regimes were found which are conducive to ozone concentrations >14 pphm. These are labeled Ai, i=1,..,4 in Figure 7, and are described in Table 10.

There are six distinct meteorological regimes (labeled Bi, i=1,..,6 in Figure 7) which are found to give rise to ozone maximum concentrations in the 11 to 14 pphm range. The meteorological conditions characterizing each of these scenarios are summarized in Table 11.

Discussion

Of the total number of days included in the analysis (1926), exceedances were recorded in the Charlotte area on 53 of these days. The exceedance days are listed in the Appendix. Note that only data for those sites listed in Table 4 were considered in the determination of an exceedance day. Those categories or terminal nodes (Tables 10 and 11) that contain the greatest number of exceedance days are Node B2 (12 days), Node A3 (12 days), and Node B6 (9 days). Of the 53 exceedance days, 12 were characterized by maximum ozone concentrations greater than 14 pphm. Nine of these days were classified as belonging to Node A3. Episode days corresponding to all three of these categories were considered in the episode selection process.

As indicated in Table 11, days classified as belonging to Node B2 (corresponding to maximum ozone concentrations between 11 and 14 pphm) are characterized by relatively low (<10.5 pphm) maximum ozone concentrations in the Charlotte area on the previous

FIGURE 5a. Average maximum ozone for each year in node B2 for the Atlanta classification.

FIGURE 5b. Average maximum ozone for each year in node B4 for the Atlanta classification.

FIGURE 5c. Average maximum ozone for each year in node B6 for the Atlanta classification.

FIGURE 6a. Average maximum ozone for each year in node A2 for the Atlanta classification.

FIGURE 6b. Average maximum ozone for each year in node A5 for the Atlanta classification.

FIGURE 7. CART classification tree for the Charlotte urban area. Nodes identified as Ai correspond to regimes conducive to formation of maximum ozone concentrations ³ 14 pphm. Similarly, Bi, Ci, and Di indicate conduciveness to peak ozone formation in the 11 to 14 pphm range, the 8 to 11 pphm range, and less than 8 pphm, respectively. The Ai and Bi regimes are described in Tables 8 and 9.

TABLE 10. Meteorological and air quality conditions in the Charlotte area giving rise to ozone maxima ³14 pphm.

Node

Meteorological Conditions

A1

Maximum surface temperature £31.9 °C

 

Maximum surface temperature >25.8 °C

 

Height of the evening sounding 850 mb surface >1620 m

 

Previous day’s maximum ozone concentration in Birmingham >9.40 pphm

A2

Maximum surface temperature >31.9 °C

 

Previous day’s maximum ozone concentration £10.5 pphm

 

700 mb morning wind direction from 4

A3

Maximum surface temperature >31.9 °C

 

Previous day’s maximum ozone concentration >10.5 pphm

 

Maximum ozone concentration 3 days ago in Nashville >7.25 pphm

 

Average surface wind speed from 1000 to 1300 LST £2.31 m/s

 

Average surface wind direction from 1400 to 1700 LST from 1,3,7,8,9,10,12

A4

Maximum surface temperature >31.9 °C

 

Previous day’s maximum ozone concentration >10.5 pphm

 

Maximum ozone concentration 3 days ago in Nashville >7.25 pphm

 

Average surface wind speed from 1400 to 1700 LST >2.31 m/s

 

Average surface wind direction from 1400 to 1700 LST from 5

day, high (>31.9 °C) midday surface temperatures, low (<2.19 m/s) midday (1000 to 1300 LST) surface wind speeds, and 700 mb wind directions during the morning hours that are not from the east (in the case of easterly winds aloft, observed ozone concentrations were lower than for this category). The high surface temperatures and low wind speeds suggest that exceedance days within this category are stagnation events, but the dependence on upper-level wind direction suggests some contribution from transport. Eighty days are classified as belonging to Node B2; meteorological conditions characteristic of this node are expected to occur approximately 8.9 times per year.

As indicated in Table 10, days classified as belonging to Node A3 (corresponding to maximum ozone concentrations >14 pphm) are characterized by relatively high (>10.5 pphm) maximum ozone concentrations in the Charlotte area on the previous day, high (>31.9°C) midday surface temperatures, low (<2.31 m/s) midday (1000 to 1300 LST) surface wind speeds, and afternoon (1400 to 1700 LST) surface wind directions that are from the south, southwest, west, or northwest. The previous day's ozone concentrations indicate that days within this category are part of a multi-day ozone buildup within the Charlotte area. While the surface wind speeds indicate stagnation, the surface wind direction is an important distinguishing feature of this node (days with identical characteristics except for surface wind direction are in Node B5, indicating a maximum ozone concentration less than 14 pphm). Twenty days are classified as belonging to Node A3; the expected frequency of occurrence of the meteorological and air quality conditions characteristic of this node is approximately 2.2 times per year.

TABLE 11. Meteorological and air quality conditions in the Charlotte area giving rise to ozone maxima in the [11, 14) pphm range.

Node

Meteorological Conditions

B1

Maximum surface temperature £31.9 °C

 

Maximum surface temperature >25.8 °C

 

Height of the evening sounding 850 mb surface >1620 m

 

Previous day’s maximum ozone concentration in Birmingham £9.40 pphm

B2

Maximum surface temperature >31.9 °C

 

Previous day’s maximum ozone concentration £10.5 pphm

 

700 mb morning wind direction from 1,2,3,5,6,7,8,9,10,11,12

 

Average surface wind speed from 1000 to 1300 LST £2.19 m/s

B3

Maximum surface temperature >31.9 °C

 

Previous day’s maximum ozone concentration £10.5 pphm

 

700 mb morning wind direction from 1,2,3,5,6,7,8,9,10,11,12

 

Average surface wind speed from 1000 to 1300 LST >2.19 m/s

 

850 mb evening relative humidity £59.5 %

B4

Maximum surface temperature >31.9 °C

 

Previous day’s maximum ozone concentration >10.5 pphm

 

Maximum ozone concentration 3 days ago in Nashville £7.25 pphm

B5

Maximum surface temperature >31.9 °C

 

Previous day’s maximum ozone concentration >10.5 pphm

 

Maximum ozone concentration 3 days ago in Nashville >7.25 pphm

 

Average surface wind speed from 1000 to 1300 LST £2.31 m/s

 

Average surface wind direction from 1400 to 1700 LST from 2,4,5,6,11

B6

Maximum surface temperature >31.9 °C

 

Previous day’s maximum ozone concentration >10.5 pphm

 

Maximum ozone concentration 3 days ago in Nashville >7.25 pphm

 

Average surface wind speed from 1000 to 1300 LST >2.31 m/s

 

Average surface wind direction from 1400 to 1700 LST from 1,2,3,4,6,7,8,9,10,11,12

As indicated in Table 11, days classified as belonging to Node B6 (corresponding to maximum ozone concentrations between 11 and 14 pphm) are characterized by relatively high (>10.5 pphm) maximum ozone concentrations in the Charlotte area on the previous day, high (>31.9°C) midday surface temperatures, and somewhat higher (>2.3 m/s) midday (1000 to 1300 LST) surface wind speeds (compared to those days within Nodes B2 and B6). The higher morning surface winds speeds distinguish days within this node from those within Node A3. Forty-three days are classified as belonging to Node B6; meteorological conditions characteristic of this node are expected to occur approximately 4.8 times per year.

Assessment of Accuracy of the Charlotte Classification

Of the days with observed peak ozone concentrations ³14.0 pphm, 100 percent were classified as having conditions conducive to ozone formation ³14.0 pphm. Similarly, 77 percent of the days with peak observed ozone concentrations in the 11 to 14 pphm range, 62 percent of the 8–11 pphm days, and 78 percent of the <8 pphm days were classified as having an expected maximum concentration in the respective range. Misclassification of a day on which the observed maximum concentration was high always involved an adjacent bin (e.g., if the observed value was >14, it was misclassified as having a peak concentration within the 11–1 pphm range). Very few of the higher observed ozone concentrations were misclassified. Note that there were 1818 days with peak ozone observations <11 pphm, out of a total database of 1926 days.

Application for Birmingham
The Birmingham ozone exceedances were found to be primarily dependent on six variables: 700 mb evening wind direction (U7wbnpm), maximum surface temperature (SmxT0), average surface wind speed and direction between 1000 and 1300 LST (Swsp1013 and Swbn1013, respectively), morning temperature difference between the surface and the 900 mb upper air sounding (U9Tamdif), and maximum observed ozone concentration in Birmingham on the previous day (Amax1). The binary decision tree for the Birmingham area is shown in Figure 8.

FIGURE 8. CART classification tree for the Birmingham urban area. Nodes identified as Ai correspond to regimes conducive to formation of maximum ozone concentrations ³14 pphm. Similarly, Bi, Ci, and Di indicate conduciveness to peak ozone formation in the 11 to 14 pphm range, the 8 to 11 pphm range, and less than 8 pphm, respectively. The Ai and Bi regimes are described in Tables 8 and 9.

The tree is characterized by one meteorological regime which is likely to give rise to ozone concentrations ³14 pphm (node A1 in Figure 8) and one meteorological regime likely to give rise to ozone concentrations between 11 and 14 pphm (node B1 in Figure 8). The characteristics of these two scenarios are summarized in Table 12.

Discussion
Of the total number of days included in the analysis (1926), exceedances were recorded at the indicated Birmingham-area monitoring sites on 21 of these days (see the Appendix). Fourteen of the exceedance days are contained in Node B1, making this the

TABLE 12. Meteorological and air quality conditions in the Birmingham area giving rise to ozone maxima ³14 pphm (A1) and [11,14) pphm (B1).

Node

Meteorological Conditions

A1

700 mb evening wind direction from 1 or 3

 

Average surface wind direction between 1000 and 1300 LST from 8 or 9

B1

700 mb evening wind direction from 2,4,5,6,7,8,9,10,11,12

 

Maximum surface temperature >27.5 °C

 

Average surface wind speed from 1000 to 1300 LST £2.99 m/s

 

Morning temperature difference between the surface and 900 mb £0.750 °C

 

Previous day’s maximum ozone concentration >6.25 pphm

most populated of the exceedance nodes. Only 3 of the 21 exceedance days were characterized by maximum ozone concentrations greater than 14 pphm; all three of these days are classified as belonging to Node A1. Episode days corresponding to these two categories or nodes were considered in the episode selection process.

As indicated in Table 12, days classified as belonging to Node B1 (corresponding to maximum ozone concentrations between 11 and 14 pphm) are characterized by previous-day Birmingham-area maximum ozone concentrations greater than 6.25 pphm, relatively high (>31.9 °C) midday surface temperatures, low (<2.99 m/s) midday (1000 to 1300 LST) surface wind speeds, 700 mb wind directions during the afternoon hours that are not from the north or northeast (the case of northerly or northeasterly winds aloft is discussed next), and a morning surface to 900 mb temperature difference (< 0.75 °C) that is indicative of relatively stable conditions. It follows that days within this category are stagnation events and can be single day events. Two hundred seventeen days (including a majority of nonexceedance days) are classified as belonging to Node B1; meteorological conditions characteristic of this node are expected to occur approximately 24.1 times per year.

Days classified as belonging to Node A1 are characterized by midday (1000 to 1300 LST) surface wind directions from the southwest and 700 mb wind directions during the afternoon hours from the north or northeast. All three of the days on which ozone concentrations greater than 14 pphm were observed in the Birmingham area are within this category, and the parameters that CART has identified as corresponding to this node indicate some possible influence of transport. However, the information is too incomplete to draw inferences. Thirty-five days are classified as belonging to Node A1 and thus the expected frequency of occurrence of the meteorological and air quality conditions characteristic of this node is approximately 3.9 times per year. However, only three of these days exhibited maximum ozone concentrations greater than 14 pphm.

Assessment of Accuracy of the Birmingham Classification

Of the days with observed peak ozone concentrations ³ 14.0 pphm, 100 percent were classified as having conditions conducive to ozone formation ³14.0 pphm. Similarly, 80 percent of the days with peak observed ozone concentrations in the 11 to 14 pphm range, 63 percent of the 8–11 pphm days, and 42 percent of the <8 pphm days were classified as having an expected maximum concentration in the respective range. Misclassification of a day on which the observed maximum concentration was high always involved an adjacent bin (e.g., if the observed value was >14, it was misclassified as having a peak concentration within the 11–14 pphm range). Very few of the higher observed ozone concentrations were misclassified. Note that there were 1869 days with peak ozone observations <11 pphm, out of a total database of 1926 days.

Application for Nashville

Classification of the meteorological and air quality conditions giving rise to distinct ozone-forming regimes in the Nashville area depended upon six primary variables. These were maximum surface temperature (SmxT0), maximum ozone concentration in the Nashville area on the previous day (Amax1), average surface wind direction from 0600 to 0900 LST (Swbn69), average surface wind speed from 1000 to 1300 LST (Swsp1013), height of the 700 mb surface based on the morning upper air sounding (U7hgtam), and the wind direction at 850 mb for the evening upper air sounding (U8wbnpm). The partitioning of the meteorological and air quality conditions according to these variables is illustrated in Figure 9.

FIGURE 9. CART classification tree for the Nashville urban area. Nodes identified as Ai correspond to regimes conducive to formation of maximum ozone concentrations ³14 pphm. Similarly, Bi, Ci, and Di indicate conduciveness to peak ozone formation in the 11 to 14 pphm range, the 8 to 11 pphm range, and less than 8 pphm, respectively. The Ai and Bi regimes are described in Tables 8 and 9.

These variables lead to classification of eight distinct meteorological regimes. One of these regimes (node A1 in Figure 9) was conducive to formation of peak ozone concentrations >14 pphm, while three of these regimes (labeled Bi, i=1,...,3 in Figure 9) were conducive to peak ozone formation in the 11 to 14 pphm range. A summary of the meteorological and air quality conditions of these four regimes is given in Table 13.

Discussion

Of the total number of days included in the analysis (1926), exceedances were recorded in the Nashville area on 39 days. These are listed in the Appendix. Note that only data for those sites listed in Table 4 were considered in the determination of an exceedance day. Those categories or terminal nodes (Table 13) that contain the greatest number of

TABLE 13. Meteorological and air quality conditions in the Nashville area giving rise to ozone maxima >14 pphm (A1) and in the [11,14) pphm range (Bi).

Node

Meteorological Conditions

A1

Maximum surface temperature >32.5 °C

 

850 mb evening wind direction from 6,9,11,12

 

Average surface wind direction from 0600 to 0900 LST from 5,6,8,9

 

Height of the morning sounding 700mb surface £3210 m

 

Average surface wind speed from 1000 to 1300 LST £3.18 m/s

B1

Maximum surface temperature >32.5 °C

 

850 mb evening wind direction from 6,9,11,12

 

Average surface wind direction from 0600 to 0900 LST from 1,2,3,4,7,10,11,12

B2

Maximum surface temperature >32.5 °C

 

850 mb evening wind direction from 6,9,11,12

 

Average surface wind direction from 0600 to 0900 LST from 5,6,8,9

 

Height of the morning sounding 700 mb surface >3210 m

B3

Maximum surface temperature >32.5 °C

 

850 mb evening wind direction from 1,2,3,4,5,7,8,10

 

Previous day’s maximum ozone concentration >8.75 pphm

exceedance days are Node B1 (8 days), Node A1 (9 days), and Node B3 (13 days). Maximum ozone concentrations greater than 14 pphm were recorded on seven days within the nine-year period. All of these were classified similarly by CART (Node A1). Episode selection for the Nashville area focused on episodes within Nodes B1, A1, and B3.

As indicated in Table 13, days classified as belonging to Node B1 are characterized by high midday surface temperatures (>32.5°) and 850 mb winds from the southwest or northwest during the afternoon. This node is distinguished from Node A1 in that the early morning surface wind direction (0600 to 0900 LST) is not southwesterly. Sixty-five days are assigned to this category; expected frequency of occurrence of the meteorological and air quality conditions characteristic of this category is approximately 7.2 times per year.

Days within Node A1 are characterized by high midday surface temperatures (>32.5°), 850 mb winds from the southwest or northwest during the afternoon, early morning surface wind directions (0600 to 0900 LST) from the southwest, 700 mb heights (based on the morning temperature sounding) of less than 3210 m, and low midday (1000 to 1300 LST) surface wind speeds (<3.18 m/s). While it is difficult to interpret certain of these indicators, the low midday wind speeds distinguish days within this category from other days that share many of the same features. Thirty-two days (including all seven days on which ozone concentrations greater than 14 pphm were observed) are contained within this node; the expected frequency of occurrence of the meteorological and air quality conditions characteristic of this category is 3.6 times per year.

Days classified as belonging to Node B3 are characterized by high midday surface temperatures (>32.5>°C), 850 mb winds that are not from the southwest or northwest during the afternoon, and moderate to high (>8.75 pphm) maximum ozone concentrations in the Nashville area on the previous day. More than half of the days with surface temperatures greater than 32.5°C meet the second criterion. Among these days, the previous day's ozone concentration is a distinguishing factor. Eighty-eight days are assigned to the category; the expected frequency of occurrence of the meteorological and air quality conditions characteristic of this category is approximately 9.8 times per year.

Assessment of Accuracy of the Nashville Classification

Of the days with observed peak ozone concentrations ³ 14.0 pphm, 100 percent were classified as having conditions conducive to ozone formation ³14.0 pphm. Similarly, 60 percent of the days with peak observed ozone concentrations in the 11 to 14 pphm range, 71 percent of the 8–11 pphm days, and 51 percent of the <8 pphm days were classified as having an expected maximum concentration in the respective range. Misclassification of a day on which the observed maximum concentration was high always involved an adjacent bin (e.g., if the observed value was >14, it was misclassified as having a peak concentration within the 11–14 pphm range). Very few of the higher observed ozone concentrations were misclassified. Note that there were 1839 days with peak ozone observations <11 pphm, out of a total database of 1926 days. The misclassification of days with peak observed ozone concentrations in the 11–14 pphm range is somewhat worse for this case than for the other urban areas.


EPISODE SELECTION

To identify candidate regional-scale modeling episode periods, the results of the CART analyses for the four urban areas were reviewed with respect to the appropriateness of each episode day as a modeling day for each urban area and the potential for intra- and inter-regional pollutant (primarily ozone) transport. This task consisted of the following steps: (1) selection of representative categories or nodes for the Atlanta area and other urban areas, (2) identification of all days (dates) within each appropriate node, (3) identification of regional-scale (Southeast) multi-day episode periods with high ozone concentrations in one or more of the urban areas considered in this analysis, and (4) analysis of the meteorological and air quality characteristics of the candidate episodes (including the potential for intra- and inter-regional transport).

As indicated earlier in this report, exceedance days corresponding to Categories (Nodes) A2 and A5 were considered in selecting episodes for the Atlanta area, those corresponding to Categories B2, A3, and B6 were considered for the Charlotte area, those corresponding to Nodes A1 and B1 were included for the Birmingham area, and those contained in Nodes B1, A1, and B3 were included for the Nashville area. A list of the Atlanta exceedance days (corresponding to Nodes A2 and A5 is provided in Table 14.

TABLE 14. Candidate modeling episode days (yymmdd) corresponding to those categories or nodes characterized by meteorological conditions most frequently associated with ozone exceedances. Atlanta area exceedance days are listed. For the Charlotte, Birmingham, and Nashville urban areas, m indicates an exceedance on the same day as for Atlanta; + and ++ indicate one and two days later, respectively; and – and = indicate one and two days prior.

Atlanta

Exceedance

Days

Charlotte

Exceedance

Days

Birmingham

Exceedance

Days

Nashville

Exceedance

Days

Atlanta Node A2

860626

=

   

860721

 

+

 

860731

+,=

   

870723

–,=,++

   

870821

m

   

880613

 

–,+

+

880615

+

880621

+,++

+

880708

+,++

+

m,+

920712

     

 

Atlanta Node A5

860627

=

   

860723

 

+

 

870610

     

870724

=

   

870725

     

870731

m,+

   

870801

–,m,+

   

870802

–,m

 

–,m

870803

   

m,+

870804

+

   

880616

+

=

880622

m,+

m

–,m

880623

–,m

 

+

880624

   

m

880625

   

900907

 

 

900908

     

930722

m

m

 

930723

 

Those days that also correspond to exceedance days from "appropriate" nodes for the other urban areas were identified and are indicated in Table 14. To accommodate the multi-day aspect of regional-scale (intra-regional transport) episodes, exceedance days for the Charlotte, Birmingham, and Nashville area that are within two days of the Atlanta-area exceedance days were also identified. These relationships, as indicated in the table, were examined to determine whether certain multi-day episode periods would constitute appropriate modeling episodes for one or more of the areas of interest. A conceptual model for intra-regional transport was also developed, based on the succession (timing) of the high ozone concentrations within the four urban areas.

Based on this analysis, five episodes were identified such that (according to the above criteria):

Days preceding and following those identified using the above criteria were also considered if high ozone concentrations were observed. The candidate episodes include 18 July – 5 August 1987, 12–17 June 1988, 22–29 June 1988, 3–10 September 1990, and 20–30 July 1993.

The July–August 1987 episode was characterized by high ozone concentrations throughout the Southeast; a maximum ozone concentration of 20.1 pphm was recorded in Atlanta (the highest ever recorded in that area). Maximum concentrations reached 14 pphm in the Birmingham area and 13 pphm in the Charlotte area. High temperatures and light winds (stagnation) contributed to the high observed ozone concentrations throughout the Southeast; some intra- and inter-regional pollutant transport is indicated later in the episode. This episode was not recommended for inclusion in the OTAG modeling process because it represents an extreme event for the Atlanta area.

Both the 12–17 and 22–29 June 1988 episodes were also characterized by high ozone concentrations throughout the Southeast. Maximum ozone concentrations of 16 and 17.2 pphm, respectively, were recorded in the Atlanta area during the two periods. Meteorologically, both episodes are similar to the 1–15 July 1988 OTAG episode in that a synoptic-scale high-pressure system dominated weather conditions in the eastern part of the U.S. Although several of the days within both episode periods meet the

representativeness criteria for Atlanta and the other urban areas, and intra- and inter-regional scale transport is indicated, these episodes were not recommended/selected. These episodes occurred just prior to the July 1988 episode and were characterized by similar (but not as extreme) synoptic-scale meteorological conditions. Thus, modeling of either of these episode periods would not help to achieve, within the context of OTAG, representation of the range of air quality and meteorological conditions under which different attainment/control strategies will be effective.

The September 1990 episode represents a relatively moderate ozone event for the Southeast. A maximum ozone concentration of 14.6 pphm was observed in the Atlanta area and an exceedance (12.1 pphm) was recorded in Birmingham. An upper-level high-pressure system was located over the Southeast during much of the period, and the airflow patterns associated with the high-pressure system may have resulted in intra- and/or inter-regional (between the Midwest or the Northeast and the Southeast) pollutant transport aloft. Clouds and precipitation, however, contributed to relatively low ozone concentrations in the northern portion of the eastern U.S. during 5–10 September. Due to the modest ozone levels, especially within the Southeast, this episode was not recommended/selected for inclusion in the OTAG modeling analysis.

Maximum ozone concentrations within the Southeast peaked twice during the 20–30 July 1993 episode period (during 21–23 and 27–28 July). Daily maximum ozone concentrations for each of the four urban areas examined as part of this analysis are plotted in Figure 10. A maximum ozone concentration of 18.0 pphm was observed in the Atlanta area (on 23 July) and exceedances were recorded in Charlotte and Birmingham (early during the episode) and in Nashville (during the latter part of the episode). An upper-level high-pressure system persisted over the Southeast throughout the 10-day period. The episode was also characterized by high temperatures. The timing of the surface ozone concentrations (high levels occur on different days in the various cities) and the upper-air airflow patterns (anticyclonic flow associated with the upper-level high pressure system) suggest that intra-regional pollutant transport occurred during the episode.

FIGURE 10. Daily maximum ozone concentrations for each of the four urban areas during the July 1993 episode.

The July 1993 episode period was further characterized by somewhat complex synoptic-scale meteorological conditions over the northern portion of the eastern U.S. A weak low-pressure system produced some precipitation as it moved across the northern tier of states during 25–30 July. Nevertheless, high temperatures were reported and exceedances were recorded at monitoring sites within Ohio, Kentucky, Maryland, Pennsylvania, New Jersey, Indiana, Illinois, Virginia, New York, and Connecticut (most of these during the latter part of the period). Observed daily maximum ozone concentrations for monitoring sites located within the OTAG modeling domain are plotted for each day of the episode in Figure 11. Although a more detailed analysis of the airflow patterns is required, the synoptic-scale airflow patterns suggest that inter-regional pollutant transport may have contributed to the exceedances in both the Northeast and the Southeast during the latter part of the episode period.

In summary, the 20–30 July 1993 episode has been selected to be modeled as part of the OTAG super-regional scale modeling effort. As described above, this episode contains exceedance days that are characterized by representative (occurring with some frequency) air quality and meteorological conditions for the Atlanta, Charlotte, and Birmingham area, is characterized by high ozone concentrations in all four of the urban areas, and is meteorologically different from the other OTAG episodes. Based on inspection of the upper-air airflow patterns, intra-regional transport (within the Southeast) likely occurred during the episode and inter-regional transport was also possible during the latter part of the episode period. An enhanced monitoring network for this episode compared to the other candidates (e.g., PAMS, SCION) as well as supplementary meteorological and air quality data collected during the episode (e.g., SOS) will provide an improved database for input preparation and model evaluation.


REFERENCES
Brieman, L., J. H. Friedman, R. A. Olshen, and C. J. Stone. 1984. Classification and Regression Trees. Belmont, CA: Wadsworth.

FIGURE 11a. Observed maximum ozone concentrations (pphm) for 20 July 1993.
FIGURE 11b. Observed maximum ozone concentrations (pphm) for 21 July 1993.
FIGURE 11c. Observed maximum ozone concentrations (pphm) for 22 July 1993.
FIGURE 11d. Observed maximum ozone concentrations (pphm) for 23 July 1993.
FIGURE 11e. Observed maximum ozone concentrations (pphm) for 24 July 1993.
FIGURE 11f. Observed maximum ozone concentrations (pphm) for 25 July 1993.
FIGURE 11g. Observed maximum ozone concentrations (pphm) for 26 July 1993.
FIGURE 11h. Observed maximum ozone concentrations (pphm) for 27 July 1993.
FIGURE 11i. Observed maximum ozone concentrations (pphm) for 28 July 1993.
FIGURE 11j. Observed maximum ozone concentrations (pphm) for 29 July 1993.
FIGURE 11k. Observed maximum ozone concentrations (pphm) for 30 July 1993.


APPENDIX: CART-BASED CLASSIFICATION OF EXCEEDANCE DAYS

TABLE A-1. List of Atlanta area exceedance days within each CART terminal node.

Node year-month-day

Node year-month-day

Node year-month-day

B1 91 9 10

B2 88 8 18

B2 85 5 13

B2 88 9 14

B2 89 7 9

B2 90 8 14

B2 89 8 3

B2 86 8 16

B2 88 7 22

B2 93 5 7

B2 85 7 20

B2 93 8 25

A1 86 4 26

A1 90 6 5

A1 90 6 18

A1 87 6 24

A1 85 6 26

B3 87 6 2

B3 85 6 4

B3 90 6 11

B3 86 6 22

C5 93 9 22

C5 87 7 29

C6 87 8 7

B4 87 7 22

B4 87 8 20

B4 88 7 25

B4 88 6 20

B4 88 6 8

B4 87 6 3

B4 93 8 20

B4 93 8 1

B4 92 8 10

B4 92 5 11

B4 91 9 16

B4 86 7 8

B4 91 9 15

B4 91 7 24

B4 91 7 2

B4 91 7 1

B4 90 8 27

B4 90 6 29

B4 90 6 28

B4 87 8 5

B5 93 9 23

B5 88 5 30

B5 87 9 2

A2 88 6 21

A2 87 6 9

A2 88 7 8

A2 86 7 31

A2 87 7 23

A2 88 6 15

A2 88 6 13

A2 88 6 6

A2 92 7 12

A2 86 6 26

A2 87 8 21

A2 86 7 21

C7 88 6 27

C7 86 8 2

B6 90 6 20

B6 86 8 4

B6 92 8 5

B6 87 7 19

B6 89 6 1

B6 91 8 16

B6 90 6 21

B6 92 8 31

B6 90 8 15

B6 86 7 26

B6 86 7 22

B6 86 6 16

B6 90 9 5

B6 90 8 28

B6 93 6 11

B6 87 7 30

B6 93 8 12

B6 85 7 11

B6 85 6 6

B6 93 7 9

B6 93 7 8

B6 87 9 3

B6 93 8 28

B6 88 8 1

B6 85 6 5

A3 85 7 12

A3 90 7 9

A4 86 7 18

A4 93 7 27

A4 93 7 21

C8 88 8 2

B7 87 8 23

B7 85 8 1

B7 93 7 29

B7 90 9 6

B8 88 6 14

B8 87 6 11

B8 88 6 17

A5 93 7 23

A5 86 6 27

A5 93 7 22

A5 88 6 25

A5 88 6 22

A5 87 7 31

A5 88 6 23

A5 88 6 24

A5 87 8 1

A5 87 8 2

A5 87 7 26

A5 87 7 25

A5 87 7 24

A5 87 6 10

A5 87 8 4

A5 87 8 3

A5 90 9 7

A5 90 9 8

A5 86 7 23

A5 88 6 16

B9 87 7 27

B9 93 7 28

B9 86 7 19

B9 88 7 9

B9 90 8 29

B9 86 8 1

 

 

TABLE A-2. List of Charlotte area exceedance days within each CART terminal node.

Node year-month-day

Node year-month-day

C1 90 8 16

C2 91 6 10

C2 88 5 30

C2 88 9 14

A1 88 6 13

B2 88 7 15

B2 87 7 31

B2 88 8 17

B2 87 7 20

B2 88 8 9

B2 88 6 16

B2 89 6 1

B2 90 8 28

B2 86 7 14

B2 86 7 7

B2 93 8 17

B2 86 8 1

B3 93 6 10

B3 88 6 29

B3 93 8 31

B3 87 8 27

B3 86 7 21

B3 87 8 20

C3 91 7 24

A2 88 7 7

B4 85 7 19

B4 86 8 2

B4 86 7 16

B4 86 7 15

A3 88 6 17

A3 87 7 21

A3 88 6 1

A3 86 7 29

A3 89 8 4

A3 88 7 10

A3 90 7 10

A3 86 7 28

A3 93 7 22

A3 88 6 8

A3 87 8 5

A3 88 8 18

B5 87 7 24

B5 88 6 21

B6 87 7 22

B6 86 6 25

B6 88 6 2

B6 88 6 22

B6 88 6 23

B6 87 7 25

B6 88 7 9

B6 88 8 2

B6 87 8 21

A4 88 7 8

 

 

TABLE A-3. List of Birmingham area exceedance days within each CART terminal node.

Node year-month-day

Node year-month-day

B1 85 5 26

B1 88 7 9

B1 88 6 22

B1 88 8 3

B1 90 6 12

B1 90 7 8

B1 90 9 6

B1 92 6 23

B1 93 7 22

B1 88 6 1

B1 88 5 27

B1 86 7 24

B1 86 6 16

B1 85 7 11

C2 87 8 11

C3 85 6 6

C4 87 8 21

C4 90 8 17

A1 88 6 7

A1 88 8 26

A1 88 6 15

 

 

 

 

TABLE A-4. List of Nashville area exceedance days within each CART terminal node.

Node year-month-day

Node year-month-day

C1 88 7 26

C1 88 5 27

C1 90 7 17

C1 88 5 29

C1 92 5 23

B1 88 6 24

B1 88 6 28

B1 88 7 8

B1 85 6 4

B1 90 6 28

B1 90 7 29

B1 88 6 8

B1 88 6 7

A1 85 7 12

A1 86 6 22

A1 89 7 8

A1 89 6 26

A1 93 8 23

A1 86 8 25

A1 87 8 3

A1 88 6 20

A1 87 8 25

B2 86 7 24

C3 85 7 18

C3 87 6 24

C3 88 8 1

B3 88 8 3

B3 93 7 28

B3 88 6 14

B3 91 8 6

B3 90 8 29

B3 90 8 28

B3 88 7 9

B3 90 7 27

B3 90 7 8

B3 88 8 2

B3 88 6 21

B3 88 6 22

B3 88 8 18


Submit your comments, feedback, questions, and ideas pertaining this page. Your input will be automatically added to the existing annotations. In order to add a new comment, you must be registered with the OTAG/AQA Peoples Page.