Sonoma Technology; Inc.
1360 Redwood Way, Suite C
Petaluma, CA 94954-1169
707/665-9900
FAX 707/665-9800
www.sonomatech.com
A TECHNICAL APPROACH USING AMBIENT DATA
TO TRACK AND EVALUATE AIR QUALITY
PROGRAMS
Draft Final Report
STI-905213.02-2918-DFR
By:
Hilary R. Hafner
Paul T. Roberts
Sonoma Technology, Inc.
1360 Redwood Way, Suite C
Petaluma, CA 94954-1169
Prepared for:
Dennis Doll
Work Assignment Manager
Emissions, Monitoring and Analysis Division
Ambient Air Monitoring Group
Office of Air Quality Planning and Standards
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
March 13, 2006
-------
TABLE OF CONTENTS
Section Page
LIST OF FIGURES v
LIST OF TABLES vii
1. INTRODUCTION 1-1
1.1 Background and Obj ective 1-1
1.2 Overview of the Technical Approach 1-1
2. BUILDING A CONCEPTUAL MODEL 2-1
2.1 Decision Trees 2-1
2.2 Identify Spatial Scale and Timing of Emissions 2-3
2.3 Identify Spatial Scale and Timing of Control Implementation 2-6
2.4 Identify Pollutant-Specific Temporal and Spatial Scales 2-8
2.5 Identify Availability of Ambient Data 2-9
3. PERFORM STATISTICAL AND GRAPHICAL ANALYSES 3-1
3.1 Basic Data Analyses 3-1
3.2 Assessing Trends 3-3
3.2.1 Metrics for Consideration: Ozone 3-3
3.2.2 Metrics for Consideration: PM2 5 Mass 3-4
3.2.3 Trends Analysis 3-5
3.3 Adjusting for Meteorological Impacts 3-6
3.3.1 Mathematical Techniques for the Meteorological Adjustment of Air
Quality Trends 3-7
3.4 Corroborative Analyses 3-8
3.4.1 Transport and Trajectories 3-8
3.4.2 Source Apportionment 3-11
3.4.3 Estimating Relative Emission Contributions 3-13
3.4.4 Other Corroborative Evidence 3-16
4. SELECTING EXAMPLES FOR CASE STUDIES 4-1
4.1 Overview 4-1
4.2 NOx and SO2 Reductions 4-1
4.3 Multiple Pollutant Reductions 4-2
4.4 Diesel Emissions Reductions 4-4
4.5 Air Toxics Reductions 4-5
4.6 Summary 4-7
5. REFERENCES 5-1
iii
-------
-------
LIST OF FIGURES
Figure Page
2-1 Decision tree for assessing trends starting with known emission changes or control
strategy implementation 2-2
2-2 Decision tree for assessing trends starting with an observed trend in an ambient
pollutant concentration over time 2-3
3-1 Annual average PM2.5 mass concentrations at sites in the Detroit area 3-2
3-2 Annual average reconstructed PM2.5 mass components at sites in the Detroit area 3-2
3-3 Preliminary map of annual average benzene concentration trends from 1990-2005
by EPA region 3-6
3-4 Spatial probability density plots for all samples and for samples with high
contributions from a mining factor identified in source apportionment of PM2.5
data collected in Phoenix, Arizona, during 2001-2003 3-10
3-5 Average VOC mass apportioned to mobile and non-mobile sources at Azusa in
Los Angeles, California, in 1987 using CMB and in 2001-2003 using PMF 3-12
3-6 Spatial probability density plots for Fresno, California, for the summers of 2002
and 2004 3-14
3-7 Transport pattern differences between the summers of 2002 and 2004 for Fresno,
California 3-15
3-8 Geographic distributions of SO2 and (b) NOx EIP for the 20%-worst visibility
days and 20%-best visibility days observed at Hercules-Glades, Missouri 3-16
3-9 Corroborative information including vehicle miles traveled, gasoline composition,
and controls to investigate changes in PM2.5 components in Washington, DC 3-17
4-1 Clinton Drive site near the Houston Ship Channel 4-4
4-2 Annual average perc concentrations at three sites in Los Angeles, California, and
the NEI estimates of perc emissions 4-6
v
-------
-------
LIST OF TABLES
Table Page
2-1 Example summary of conceptual model information for S02 2-4
2-2 Example summary of conceptual model information for NOx 2-5
2-3 Summary of control information for RFG 2-7
2-4 Summary of information to consider for ozone 2-8
2-5 Summary of information to consider for PM2.5 mass 2-9
2-6 Important issues about data availability 2-10
vii
-------
-------
1. INTRODUCTION
1.1 BACKGROUND AND OBJECTIVE
The U.S. Environmental Protection Agency (EPA) has promulgated and begun
implementing several regional-scale programs to reduce concentrations of ozone and particulate
matter less than 2.5 microns (PM2.5). The oxides of nitrogen (NOx) multi-state implementation
plan (SIP) call has resulted in significant reductions in NOx emissions from large utilities and
boilers over the past several years in eastern parts of the United States. The EPA documented
improvements in summertime ozone air quality in rural and urban areas, especially between 2002
and 2005, and linked these improvements to the emissions reductions (U.S. Environmental
Protection Agency, 2005). While regional-scale improvements in ozone have been noted, the
EPA wants to develop refined methodologies to correlate emissions controls to changes in
ambient air quality that can be applied not only to other regional-scale pollutants, such as PM2.5,
but also to urban (local) scales where local controls should noticeably impact air quality.
The goal of this document is to provide a technical approach to assess the link between
emissions controls and air quality changes with a focus on local and urban scales. The focus of
this work is primarily ozone and PM2.5; however, techniques developed may be applicable to
other pollutants such as air toxics.
1.2 OVERVIEW OF THE TECHNICAL APPROACH
This technical approach was developed to guide analysts in their evaluation of the
impacts of regional control programs (e.g., those that affect multiple states) and local control
programs (e.g., those that affect an urban area) on air quality. This evaluation is complicated and
will be stepwise and site- and pollutant-specific. A major challenge is the scale of influence of a
control and of the impact of that control on air quality. Previous investigations of ambient air
quality changes encountered the confounding influences of multiple controls applied within
similar time frames and at different spatial scales. One goal of this technical approach is to
provide a more detailed process that will improve confidence in identifying and quantifying the
cause and effect of changes in ambient pollutant concentrations (i.e., accountability).
While the focus of the technical approach is retrospective, the thought process, and some
of the results, could be used to assess future controls and their potential impact on air pollutants.
For example, if a future control scenario resembles a past control application, the technical
approach in this document could be used to estimate future changes in ambient air quality based
on the changes observed in the past.
The technical approach in this document provides a structure or road map for analysts. A
key consideration is that consensus among several analyses gives the analyst more confidence in
the cause and effect of a selected control on ambient data changes. The technical approach
comprises the development of a conceptual model for the analysis (Section 2), discussion of
statistical and graphical analysis tools or techniques and their applicability to different pollutants
(Section 3), and a straw plan for an example application (Section 4). The example application
will be addressed in the next task of this work assignment.
1-1
-------
-------
2. BUILDING A CONCEPTUAL MODEL
The first step in the analysis is to develop a conceptual model to help the analyst define
the parameters of interest in the analysis, set up expectations, and identify the potential
limitations and confounding issues for the analysis. The conceptual model aids the analyst in
developing hypotheses to test, which include null hypotheses (e.g., comparing air quality trends
in areas with and without the selected control). This section discusses the issues in assessing
pollutant trends over time with respect to emissions controls and changes.
2.1 DECISION TREES
A decision matrix or tree provides an overview of required steps for evaluating various
control scenarios. A decision matrix guides an analyst through the thought process of an
analysis. Two approaches to a trends analysis can be taken depending on the availability of
information: an emission control approach and an ambient data approach. Starting with the
control measure, Figure 2-1 illustrates a possible approach to an analysis:
Select a control measure.
Identify the air pollutants expected to be affected and the available data, other controls
that might have affected the pollutants, and other pollutants that may have been affected.
Consider the spatial scale, or zone of influence (ZOI), of the control measure. Was the
control applied at a single facility (monitor-specific or fence line), at an urban scale
(Metropolitan Statistical Area [MSA]-wide), regional scale (e.g., multi-state NOx SIP
call), national scale (e.g., 49-state automobile emission rules), or global scale (e.g.,
Montreal protocol)?
Determine the timing and magnitude of the changes. Was the control phased in over a
period of time, applied to specific emitters?
Consider the magnitude of the expected air quality changes relative to the variability in
the ambient data. If the inherent variability in the ambient data is very large, a small
change in emissions may not be observable.
Select the appropriate statistical metrics or approach for the analysis. Data treatments
may help reduce the variability in the data so that trends can be observed.
Develop hypotheses of expected changes, identify supporting evidence of changes, and
investigate corroborative evidence of the changes. It is often helpful to test for changes
in data sets or pollutants in which changes were not expected (i.e., check the null
hypothesis).
An example of this approach is the selection of reformulated gasoline (RFG)
introduction. The EPA Phase I RFG was implemented in 1995 with targeted benzene reductions
of up to 43% in gasoline. Models predicted significant benzene reductions in gasoline vehicle
exhaust caused by the use of RFG. Model predictions also indicated reductions in total aromatic
hydrocarbons and 1,3-butadiene and increases in formaldehyde. Investigations of ambient
benzene concentrations, weight fractions, and ratios showed that statistically significant
2-1
-------
reductions were observed at Northeastern and Mid-Atlantic cities and that these reductions could
be tied to gasoline composition changes (Main, 1998). Benzene reductions were not observed in
ambient data in cities that did not introduce RFG. Similarly, hydrocarbons not expected to be
affected by RFG, such as isoprene, did not change.
Figure 2-1. Decision tree for assessing trends starting with known emission
changes or control strategy implementation.
Starting with an observed trend in ambient pollutant concentrations, Figure 2-2 illustrates
a possible approach to the analysis:
Quantify the change observed in the ambient data. This approach could also be applied
to a pollutant in which a change was not observed but expected.
Apply meteorological adjustments to the pollutant trend. The goal is to reduce the effect
of meteorology on the ambient concentrations so that the underlying trend in emissions
can be more readily observed.
Identify and assess other data sets and sites that may have also been affected by a similar
control measure or emission change to understand the spatial scale of the ambient change.
If the control was applied across a broad area, changes at additional sites might be
expected.
Identify potential emissions changes or control measures that could have contributed to
the ambient trends. Local knowledge is often a key component of this part of the
analysis.
Compare the control measure implementation schedule with the ambient trends. Do the
timing of the control implementation and the change in ambient concentrations coincide?
2-2
-------
Investigate corroborative evidence of the change and test for changes in pollutants in
which a change was not expected. It is important not to over-interpret changes in
ambient data.
Identify Trend in
Pollutant
A
Quantify Change Observed
Evaluate Corroborative Identify Scale -
Evidence and Hypotheses Zone of Influence
Apply Meteorological
Adjustment
1
Identify Possible
and Implementation
Period
Control Measures Compare Control
Implementation with
Ambient Trends
Figure 2-2. Decision tree for assessing trends starting with an observed trend in
an ambient pollutant concentration over time.
An example of starting with the data was illustrated in recent air toxics work in which a
site with significant declines in ambient lead concentrations were observed in Philadelphia,
Pennsylvania. Local contacts confirmed that an industrial facility near the monitoring site had
implemented PMio controls that also resulted in reduced lead emissions (Hafner et al., 2004).
The following sections provide a guide to develop the conceptual model. The conceptual
model helps the analyst form expectations, develop hypotheses to build the case for
accountability, and identify an analytical procedure. Example tables are used to illustrate
emissions, controls, ambient data, and monitoring site considerations.
2.2 IDENTIFY SPATIAL SCALE AND TIMING OF EMISSIONS
Issues to consider about emissions include spatial distribution of emissions and typical
surrogates for emissions density (e.g., population density), data sources (is it important to
understand the projections used over time periods?), spatial maps (determine resolution needed),
source areas (e.g., Ohio River Valley), expected changes from emission to receptor and time
scale of these changes (e.g., conversion of sulfur dioxide [SO2] to sulfate), and biogenic versus
anthropogenic sources. Primary emissions for consideration include PM2.5 mass, carbon
monoxide (CO), organic carbon (OC), elemental carbon (EC)/ black carbon (BC), nitrogen oxide
(NO), volatile organic compounds (VOCs), S02, ammonia (NH3), metals, and some toxics.
Examples are provided for SO2 (Table 2-1) and NOx (Table 2-2).
2-3
-------
Table 2-1. Example summary of conceptual model information for SO2.
Geographic distribution
2001 SO2 emissions by county. Higher SO2 emission rates
occur to the east of the Mississippi River and in urban
locations.
Tons PerYodri'Squnre Ml*
I ao.0.12
I tua-aao
I I 0.30-0.60
~ 0.81 -4.3
I ll 4.4-11O0
Atmospheric residence time
One to one-and-one-half days. It takes about a day to form
sulfate PM.
Major emission source
categories
Fuel combustion, principally point sources
Emission inventory for the
locations of concern
Compare a given location to the United States as a whole
(2001):
Transportation
S%
Miscellaneous
<1%
Seasonal variability
SO2 -to-sulfate transformation rates peak in the summer
because of enhanced summertime photochemical oxidation
and SO2 oxidation in clouds. Sulfate concentrations vary
between seasons by at least a factor of two.
Biogenic sources?
No.
2-4
-------
Table 2-2. Example summary of conceptual model information for NOx.
Geographic distribution
2001 NOx emissions by county. Emissions are typically higher in
urban areas.
Tons Pw Year/Square Mile
I no-1 .
IH 1.6-3J0
I I at -5js
I I 5.7-13
I I U-2200
Atmospheric residence time
NOx can contribute to PM nitrate formation at night and during the
day; daytime photochemistry also forms ozone.
Major emission source categories
Mobile sources, power generation
Emission inventory for the
locations of concern
Compare a given location to the United States as a whole (2001):
Transportation
56%
Miscellaneous
2%
Seasonal variablity
NOx -to-nitrate transformation rates are favored in winter. It is
estimated that about one-third of anthropogenic NOx emissions in
the United States are removed by wet deposition. Nitrate
concentrations vary between seasons by at least a factor of two.
Biogenic sources?
Yes, natural sources include lightning, biological and abiological
processes in soil, and stratospheric intrusion. Ammonia and other
nitrogen compounds produced naturally are important in the cycling
of nitrogen through the ecosystem.
These tables help illustrate the spatial scales encountered for emissions. SO2 emissions
are dominated on a national scale by power generation, while NOx emissions at a national scale
are a combination of transportation and power generation. Thus, for NOx, it is difficult to
demonstrate the effect of regional trends in power plant NOx controls on regional ozone changes
because ozone downwind of an urban area has been influenced more by urban NOx than by NOx
from a regional power plant.
2-5
-------
2.3 IDENTIFY SPATIAL SCALE AND TIMING OF CONTROL
IMPLEMENTATION
For changes in emissions and emission control implementation, the analyst needs to
know as much as possible about the spatial scale and timing of control implementation in order
to determine effective analyses of the ambient data. Information useful to the analyst includes
where and when changes were implemented, model predictions of affected pollutants and
magnitude of change in ambient air, the relative magnitude of change in emissions over space
and time, what ambient pollutants or precursors were targeted, what other species could be
affected, the availability of corroborative evidence (e.g., gasoline information), potential
conflicting issues, the need for, and availability of, local knowledge, method of control
enforcement (i.e., voluntary versus fine system), and the uniformity in space and time of the
control implementation.
An example summary of control implementation information is provided in Table 2-3 for
the introduction of RFG at a national level. The table summarizes the types of information
useful in developing a conceptual model of emissions controls:
Spatial scale. The graphic provided in Table 2-3 illustrates the inherent spatial
complexity in some of the control programs. A wide range of gasoline regulations, which
would have a range of emissions changes, is shown.
Implementation period. Some areas have separate winter and summer gasoline
requirements. Changes in gasoline formation often occurred relatively abruptly (short
phase-in periods), and some areas did not have to implement RFGor opted to
implement RFG even though they were not required to do so. Changes in gasoline
formulation also occurred over time. RFG was implemented with set seasons indicating
that subsets of annual data, rather than annual averages, would need to be investigated.
Targeted pollutants. While specific pollutants were targeted (i.e., reduce NOx, ozone,
benzene), changes in other pollutants were likely to occur, such as reductions in
1,3-butadiene and increases in carbonyl compounds. Other considerations for the
introduction of RFG include the changes in total aromatic hydrocarbon content (i.e.,
could secondary organic aerosol formation be affected?), VOC and NOx emissions, and
sulfur content (i.e., would sulfate concentrations be affected?).
Magnitude of control. The estimated changes in emissions because of RFG were large
(i.e., more than 40% reduction in gasoline benzene content), which indicated that changes
could likely be detected in the ambient data.
Corroborative evidence and hypothesis testing. Gasoline composition data were
available for most areas implementing RFG (although gasoline composition data were
not readily available prior to the change) to help explain potential year-to-year
differences in ambient benzene concentrations. In addition, some areas of the country did
not implement RFG so that the benzene trends in those areas could be compared to the
areas with RFG.
2-6
-------
Table 2-3. Summary of control information for RFG.
Implementation geographically3
U.S. Go toil no Requirements
^
*4 W. J B-M T
M . ft w
!
BU.* v - .Vr-'rf
- Jr -i. xC*
i r «
* ~ * -,,
f' V J
f1 ^
HI ***** I tin
CACOO ¦¦ ¦
OF&CAGOG ; J 1M ,i i i i
v cma WM ' * ***
Oy Rr**1 ¦¦ 7CM ^^
.W mm
0hy|"H*70lW HI S ' ¦ i . - t m ¦
Q*VRm4*PlVP [ ~ . ¦bCpprn ft
Major emission source categories
Gasoline-fueled mobile sources
Implementation schedule
(incomplete)
Phase 1 simple model introduced in 1995
California RFG (California only) introduced in 1996
Phase 1 complex model introduced in 1998
Phase 2 complex model introduced in 2000
Targeted pollutants
Phase 1 (1995): lower Reid Vapor Pressure (RVP) (e.g., butanes), and
benzene; add oxygen
California RFG (1996): lower RVP, benzene, aromatics. olefins, and
sulfur
Phase 1 (1998): lower VOC, toxics; add oxygen
Phase 2 (2000): lower VOC, toxics, and NOx; add oxygen
Magnitude of change expected
> 40% reduction in gasoline benzene content
Phase 1 (1998): 15% VOC reduction
Phase 2 (2000): 25% VOC reduction; 6.8% NOx reduction
Model predictions
20% reduction in 1,3-butadiene emissions
20% increase in formaldehyde emissions
Increase in i-butene (if methyl tert-butyl either [MTBE] used as
oxygenate)
Seasonality
A summer-only program was implemented for RFG.
A winter oxygenate program was also implemented in some areas.
Scope of program (market
penetration)
The California program is statewide; the program is federally mandated for
areas with severe ozone problems; some other areas opted in.
Other considerations
Could reduction in aromatics content lead to change in OC?
More information is needed about parallel sulfur reduction.
Corroborative data
Gasoline composition information is available for RFG cities from 1995+;
pre-RFG gasoline content data are less available.
Other considerations
While vehicle emissions might be reduced by the control, increases in
population or vehicle miles traveled could potentially offset the reductions.
a Graphic courtesy of http://www.eia.doe. gov/oiaf/servicerpt/fuel/ozone.html.
2-7
-------
2.4
IDENTIFY POLLUTANT-SPECIFIC TEMPORAL AND SPATIAL SCALES
Different data analysis methods will be required for primary and secondary pollutants
because of differences in formation and removal rates and in spatial and temporal variation in
concentrations and emissions. These differences complicate the scale (local, regional) of the
cause and effect relationship between emissions and the receptor. The pollutant list includes
Primary: PM2.5 mass, CO, OC, EC/BC, NO, VOCs, SO2, NH3, metals, and some toxics.
Secondary: ozone, nitrogen dioxide (NO2), some portion of PM2.5 mass, ammonium
sulfate, nitric acid (HNO3), ammonium nitrate, OC, VOCs, and some toxics.
For the conceptual model, the analyst must identify the key temporal and spatial scales so that
appropriate analyses can be selected later. Tables 2-4 and 2-5 summarize information for ozone
and PM2.5 mass as examples. Understanding the formation, transport, seasonality, variability,
and regional background concentrations of the pollutant helps explain the correlation between
control implementation and ambient changes in the pollutant. In the two tables, note that ozone
formation is complex, but seasonal and diurnal variability is generally well-understood. PM2.5
mass also has complex formation processes, but in contrast to ozone, it also exhibits complex
seasonal and diurnal variability, which are not well-understood. Both pollutants have large
spatial scales because of the importance of transport.
Table 2-4. Summary of information to consider for ozone.
Formation
A complex series of reactions including sunlight, NOx, and
VOCs. Photochemical modeling indicates ozone formation is
sensitive to VOC/NOx ratios.
Atmospheric residence time
Hours to days
Seasonal variability
Highest concentrations are expected during the warm season.
Production of concentrations is increased with extended
sunlight.
Diurnal variability
Photochemical production is highest during the day. In urban
areas, titration by NO may occur in the early morning and
overnight further suppressing ozone concentrations.
Less diurnal variability is expected at rural sites.
Day-of-week variability
Weekday-weekend differences are possible for areas in which
the VOC/NOx ratio changes significantly.
Regional background
concentrations
Thirty to 40 ppb remote background; regional background
concentrations can be significantly higher.
Spatial scale
Local, subregional, and regional. Transport on a regional
scale (hundreds of kilometers) can contribute significantly to
urban ozone concentrations. Measurements of ozone above
ground, but within an altitude that could be part of the mixed
layer the next day, have contributed significantly to ground-
level ozone.
2-8
-------
Table 2-5. Summary of information to consider for PM2.5 mass.
Formation
Chemical reaction, nucleation, condensation, coagulation, and
cloud/fog processing. Secondary components of PM2.5 mass
include ammonium sulfate, ammonium nitrate, and secondary
organics.
Atmospheric residence time
Days to weeks
Seasonal variability
PM2.5 exhibits a seasonal pattern. In areas with sulfate-
dominated PM, warm season peaks may be expected (e.g.,
eastern United States). In areas with high nitrate
concentrations, cool season peaks may be expected (e.g.,
California). High PM2.5 concentrations can occur during any
season.
Diurnal variability
Diurnal peaks in PM2.5 mass concentrations vary by location
(e.g., proximity to sources) and by season (e.g., meteorology,
source type).
Day-of-week variability
Weekday-weekend differences are not usually observed in
PM2.5 mass but have been observed for PM components such
as EC.
Regional background
concentrations
Visually protected area measurements (i.e., IMPROVE21)
provide typical concentration ranges. Regional contributions
of PM2.5 can be significant but vary by PM component and
region. For example, local contributions of sulfate are very
small compared to local contributions of OC.
Spatial scale
Local, subregional, regional, and global. Transport distances
range from hundreds to thousands of kilometers. Visible dust
and smoke plumes can be tracked by satellite.
a IMPROVE Interagency Monitoring of Protected Visual Environments monitoring network.
2.5 IDENTIFY AVAILABILITY OF AMBIENT DATA
Data sets available include routinely collected data such as national and state criteria
pollutant monitors (available in the EPA's Air Quality System [AQS]), ozone precursor network
(Photochemical Assessment Monitoring Stations [PAMS]), particulate matter networks
(Speciation Trends Network [STN], IMPROVE monitoring network), other networks including
the Clean Air Status Network (CASTNET) and Southeast Aerosol Research Characterization
Study (SEARCH), special studies data (such as PM2.5 Supersites), and large-scale field studies,
such as the California Regional PMi0/PM2.5 Air Quality Study (CRPAQS), Central California
Ozone Study (CCOS), Texas Air Quality Study (TxAQS), and NARSTO-Northeast.
For trends analysis, Table 2-6 provides a checklist of important issues about data
availability. For an effective analysis of trends, the analyst needs to know the proximity of the
monitor relative to emissions, have data of sufficient length of record (preferably several years
before and after an emission change), have stable data (i.e., no analytical changes, or if a change
2-9
-------
did occur, have knowledge of the effect on the data), and have data of known validity with few
gaps in the record. For an emission control that likely affected a broad spatial area, it will be
important to have data from nearby sites and urban/rural pairs to understand spatial
representativeness and urban excess contributions. Meteorological data are needed for adjusting
trends and to improve the understanding of source-receptor relationships. Finally, the
availability of previous data analysis results for an area or a monitor can save time and provide
additional insight into the investigation.
Table 2-6. Important issues about data availability.
Data Issue
Importance/Relevance
Proximity of emissions to monitor
For some analyses, the analyst may want data
collected near a type of emission source (local
impact of primary emissions, for example), while
for other analyses (regional impact for secondary
pollutant, for example), the analyst will want data
far removed from the emission source.
Length of data record
Data collected throughout the complete period
of interest at a long-term site are vital.
A longer data record is preferable.
Stability, validity, and completeness of
record
Few analytical changes, valid data, and
sufficient data above the detection limit are
desirable.
Complete data for all periods of interest (i.e.,
hours, days, months, and seasons) are vital to
properly represent each period and to enable
valid comparisons between periods.
Consistent measurement methods are needed to
enable comparisons across years.
Data from nearby sites
Several sites in an area may be needed to
understand spatial variation and to document the
local impact of a control.
Urban/rural pairs
Urban/rural pairs are desirable for analyses;
however, the analyst needs to understand data
collection and chemical analysis differences prior to
data analysis.
Spatial representativeness of a site (and
how it may have varied over time)
Understand site characteristics and possible changes
over time (e.g., a suburban site is now considered
urban due to growth of the city).
Meta data availability
More details about the monitoring location, sample
collection, and analysis are preferable.
Collocated measurements including
meteorological data
Additional supporting data are preferable.
Previous data analysis results
Previous characterization of data at a site, including
source apportionment and trends analyses, is
desirable.
2-10
-------
3. PERFORM STATISTICAL AND GRAPHICAL ANALYSES
This section is intended to set up the types of analyses to perform and provide the
relationship of the analysis to the conceptual model. A few examples are given in this section to
illustrate some of the steps, issues, and pitfalls of performing the types of analyses that are
identified in Section 2.
3.1 BASIC DATA ANALYSES
Prior to any analysis, data validation must be performed to ensure the integrity of the
data. Data validation steps for ozone precursor data, ozone data, and PM25 data are provided
elsewhere (e.g., Main and Roberts, 2000, 2001; Hafner, 2003). Once the data are validated,
basic data analyses should be performed to select the appropriate next steps in the analysis.
Basic data analyses include
Reviewing summary statistics by site, month, and year for similarity among sites, such as
local versus regional patterns, urban versus rural differences, and other spatial gradients.
Inspecting the time series of all data and monthly and annual metrics to identify potential
trends, breaks in the data, and diurnal, seasonal, or annual patterns. For example, ozone
concentrations peak during the warm season (May through October) throughout the
country. PM25, in contrast, exhibits high concentrations in either winter or summer
seasons (and in some places, both seasons) depending on a monitor's location relative to
source types. NOx is a precursor both to ozone and to PM25; thus, the impact of warm-
season reductions in regional NOx versus reductions in cool-season regional NOx needs to
be considered in the technical approach.
Inspecting day-of-week variations to identify potential influences from emissions
sources.
Investigating species relationships (e.g., scatter plots, correlation matrices) to identify
which species co-vary.
Performing monitor-to-monitor correlations to understand how measured concentrations
at one monitor compare to concentrations at other monitors. Monitors with
concentrations that correlate well (e.g. ,r2> 0.75) with concentrations from a different
monitor indicate similar influences. Conversely, a monitor with concentrations that do
not correlate with other nearby monitored concentrations may be affected by a local
source.
One type of analysis that may arise is the need to identify the impact that a control of a
local PM2.5 source has on ambient concentrations. A basic analysis should be performed to
determine whether the available data prior to the control being implemented shows a spatial
gradientwith higher concentrations closer to the source. For example, in Detroit, Michigan, a
"bubble" of PM2.5 mass in the city, with gradients away from the industrial sites (i.e., the
Dearborn site, Figure 3-1), indicates influences of local sources on PM2 5 mass (and source
apportionment results of speciated PM2 5 indicate that local influences exist). A similar plot
showing the PM2 5 composition by site (Figure 3-2) also indicates that a possible local influence
3-1
-------
from sulfate, OC, EC, and metals (incorporated in the computation of "soil") at the Dearborn
site.
Ccric.,
ug/m3
"254
Figure 3-1. Annual average PM2.5 mass (2000-2004) concentrations (,ug/m ) at
sites in the Detroit area (Kenski, 2006). Sites are ordered roughly south to north.
The line indicates the PM2.5 National Ambient Air Quality Standards (NAAQS).
8
7
6
5
4
3
2
1
0
Sulfate I I Nitrate i I OC 1 I EC 1 I Soil
All available data used for avaraging, using compete years
Figure 3-2. Annual average reconstructed PM2 5 mass (2000-2004) components
(|jg/m3) at sites in the Detroit area (Kenski, 2006). Sites are ordered roughly
south to north.
3-2
-------
A spatial gradient with high concentrations at a monitor close to a source and a decrease
in concentrations at monitors further away gives the analyst confidence that the data set can be
used to look for trends related to local or regional controls, provided the data record is sufficient
and controls were implemented during the data record time frame. Similar analyses
characterizing the spatial gradient of a selected pollutant will be necessary to identify locations
and pollutants where the influence of local controls may be observed.
Another consideration in the basic analysis stage is the use of derived parameters. For
example, ratios (e.g., EC/OC), differences (e.g., urban versus rural), and other derived values
(e.g., non-soil iron [Fe]) may be vital to understand how (and how widespread) a control
program affected a specific area. It can be useful to understand the regional-level concentrations,
as observed at rural sites, compared to the concentrations at urban sites, where the difference is
the urban excess. Further analysis of trends in the regional level and urban excess may be
important to explain whether overall trends/changes over time are caused by changes in regional
level and urban excess levels. An example of an urban excess computation is provided in a
trends report in which urban and rural pairs were used to quantify the PM2.5 components and the
urban excess (Frank, 2005).
Basic analyses help the analyst understand the data set characteristics and limitations and
build a conceptual model and hypotheses.
3.2 ASSESSING TRENDS
In a trend analysis, the general approach is to begin with validated data; select a pollutant
and/or derived parameter; select indicators (metrics), such as mean, median, maximum,
minimum, selected percentiles, etc.; select appropriate time periods to investigate (e.g., season,
episode, annual, etc.); apply statistical procedures for detecting trends; evaluate the trend for
direction, rate of change, statistical significance, etc.; and compare trends among indicators. A
consensus of indicators for trends analyses will reduce the uncertainty of any conclusion. This
approach can be applied to raw indicators, or the indicators can be adjusted to remove
meteorological influences (Section 3.3).
3.2.1 Metrics for Consideration: Ozone
Statistical indicators to consider for ozone include
Design values for ozone by year at each site. The design value is the average of the
fourth highest value recorded at a site for three consecutive years. A recent study showed
that the EPA ozone design value method provides a reasonable estimate of the "true" air
quality design value for ozone nonattainment areas and of peak ozone levels within those
nonattainment areas for the initial three-year compliance period.
The average maximum 1-hr and 8-hr ozone concentrations by year at each site. This
statistic is another measure of the peak concentrations.
3-3
-------
The average of various percentile bin values (e.g., average of the 1st, 5th, and 10th
percentile concentrations) by year at each site. This metric also helps illustrate the
distribution of ozone and precursor concentrations.
The average exceedance concentration for each year at each site. This statistic is used to
indicate the average air quality on days when the air is considered poor according to the
1-hr ozone NAAQS.
The number of exceedances above selected thresholds. This statistic can also be used to
gain insight into the nature of the average exceedance concentration, and lends a different
perspective than the ranking of the top three exceedance concentrations when a site
experiences more than three exceedances within a single year. Similar to the ranking
statistic, the identification of exceedance thresholds gives the analyst a sense of whether
the highest exceedance concentrations are anomalies or are representative of commonly
experienced elevated ozone concentrations.
A ratio of early morning VOC-to-NOy measurements. This ratio can be used to assess
whether VOCs or NOx (or both) in the ambient air limit the overall formation of ozone.
This insight can be gained despite the complexity of the photochemistry because certain
observed ratios of early morning VOC and NOx concentrations show reductions or
increases in the amount of ozone formed over the course of a day (National Research
Council, 1991). However, attention to detail is needed in defining and consistently
computing the VOC portion of the ratio. VOC definitions and concentrations vary by
measurement and analytical method.
Ozone formation potential using maximum incremental reactivity (MIR) applied to the
VOC data. While VOC concentrations may not have changed significantly, the VOC
composition and ozone formation potential may have changed over time in response to a
control measure.
Other species concentrations, such as morning CO or individual hydrocarbon species.
CO trends may be an indicator of benzene trends (both pollutants are present in motor
vehicle exhaust) at some sites, for example. However, this relationship is likely site- and
year-specific and may be difficult to use in a trend analysis.
3.2.2 Metrics for Consideration: PM2.5 Mass
In addition to design values, average of various percentile bin values, and number of
exceedances above selected thresholds, statistical indicators to consider for PM2.5 mass include
Annual PM? s average concentrations by site and year.
Composition metrics, such as mass and percent of pollutant contribution to total mass of
the major PM components (ammonium sulfate, ammonium nitrate, OC, EC, and soil) by
site and year. Overall mass may not have changed over time, but specific components of
the PM2.5 may have changed.
Days in a selected air quality index (API) range by site and year.
3-4
-------
3.2.3 Trends Analysis
Multiple approaches to long-term trends should be considered because individual trend
analyses are often limited by uncertainties in the measurements and/or the unavailability of
critical measurements. Because the relationship between emissions controls, population growth,
meteorology, and air quality is complex, true trends are difficult to establish using any single
method. When measurement issues such as missing years of measurements or differences in
reporting units are also considered, trends can be less conclusive or misleading. However, the
use of multiple trend analyses involving meteorological adjustment techniques suggests when
one method is not robust enough or applicable for a particular site.
Of the several approaches to trend analyses, the approaches differ in their measurement
requirements and numerical intensity. Even simpler approaches require considerable amounts of
air quality measurements (e.g., at least three continuous years of valid ozone exceedance
concentrations). More complex approaches have greater measurement requirements (e.g., at
least three continuous years of valid daily maximum ozone concentrations, concurrent precursor
measurements, and/or concurrent meteorological measurements). To enable more complete
analyses, gaps in meteorological measurements must be supplemented with nearby
measurements.
The following are examples of trends analysis methods:
Linear model. Simple linear regression applied to annual summary statistics can be used
to quantify trends. The significance of the difference between two years could be
confirmed using a t-test or a comparison of confidence intervals around the means of the
metric under consideration. The years before and after a change can be grouped to
partially remove meteorological changes among years.
Nonparametric methods. These approaches test for and estimate a trend without making
distributional assumptions using such techniques as the Mann-Whitney U test,
Spearman's rho test of trend, and Kendall's tau test of trend. The Mann-Whitney U test
is analogous to the t-test; however, the Mann-Whitney U test is based on the ranks of the
values within each year, not the actual values. Thus, it essentially is a test based on the
median values that does not assume any underlying distribution for the data.
Nonparametric tests provide the advantage of not assuming a linear trend and are
reasonably robust against outliers.
Time series models. Statistical modeling of ozone concentrations using time series
models takes into account the serial dependence of the concentrations. An example of
time series model is the auto-regressive integrated moving averageARJMA.
Extreme-value theory. This method estimates the distributions of annual maximum
hourly concentrations and estimates the distributions of the number of days exceeding the
NAAQS. Examples include the Chi-square test and Poisson process approximation.
In order to guide the analyst in selecting the appropriate trends analyses methods, data
completeness and the metrics under consideration need to be assessed. It is useful to initially,
and more qualitatively, inspect trends in the data by generating maps that show the trends of
selected metrics at all sites (or combinations of sites), such as Figure 3-3.
3-5
-------
-38% N - 10
-4.9%
-17%
Figure 3-3. Preliminary map of annual average benzene concentration (jig/m3)
trends from 1990-2005 by EPA region. N is the number of sites included in the
trend; the percent change shown is based on the linear regression line. Site data in
75% of the trend period years needed to be included in the trend.
Figure 3-3 also helps illustrate another consideration when investigating trends. The
trend for Region 2 shows the only upward trend among the regions over the entire trend period.
However, if the first year or two of data were not included, a downward trend would be
observed. This example points out that trend assessments can be misleading because different
groupings of years can lead to different trends. Consistent sets of sites and years should be used
in trend assessments.
3.3 ADJUSTING FOR METEOROLOGICAL IMPACTS
Day-to-day variations in pollutant concentrations may be caused by meteorology alone.
When changes in emissions over decades are considered, the influence of emissions on pollutant
concentrations becomes more significant. The variations in pollutant concentrations caused by
meteorological influences can conceal or mask the influence of emissions reductions. Therefore,
when looking at trends in ambient pollutant concentrations, it is desirable to remove the
meteorological influences from the trends.
It has long been known that weather conditions have a strong influence on ozone
concentrations, for example. Because of this, multi-year cycles in weather can influence multi-
year ozone and ozone precursor trends. Therefore, when trying to explain ozone and ozone
3-6
-------
precursor (or other pollutant) trends that are caused by changes in emissions, the analyst needs to
account for the influence of weather on the trends. To properly perform this "meteorological
adjustment" to the trend requires (1) the use of mathematical techniques that account for the
influence of weather on long-term air quality trends and (2) the use of data that capture the full
range of meteorological processes that influence the pollutant.
3.3.1 Mathematical Techniques for the Meteorological Adjustment of Air Quality Trends
Mathematical techniques to statistically adjust ozone concentrations to account for
meteorological variables have been developed (Thompson et al., 2001; Porter et al., 2001).
Different approaches have various advantages and disadvantages depending on the type of ozone
metric and monitoring site being investigated. Therefore, it is best to investigate various
techniques to achieve the most accurate meteorologically adjusted trends. These techniques have
been less broadly applied to PM2.5 data and even less work has been done with air toxics. An
overview of available approaches follows:
Regression-based Models. Linear regression-based models are the simplest method for
the meteorological adjustment of trends. These models use linear and additive relationships
between the pollutant and meteorological variables to adjust pollutant trends. While relatively
simple, these models are open to criticism because ozone and PM2.5 formation and some
meteorological variables do not have a linear and/or additive relationship.
Tree-based and Stratified Models. Tree-based and stratified models adjust ozone
trends in two steps based on differences in the association between ozone and meteorology under
different meteorological regimes. Typically, these models use Classification and Regression
Trees (CART) or cluster analyses (Larsen, 1999) to identify the different meteorological regimes
under which ozone formation occurs. Regression analyses are then applied separately for each
regime to quantify the trends. This two-part analysis provides the added benefits of better
identifying meteorological regimes under each cluster (which may lead to more sensitive models
of extreme values) and accommodating seasonal effects.
Extreme-value Models. The inherent averaging in regression models makes them poor
predictors of extreme values that can be important for attainment designations. Extreme-value
models avoid this problem. One example is the approach used by Cox and Chu (1993; 1996). In
this technique, concentrations are adjusted by fitting ozone measurements to a Weibull
probability distribution as a function of multiple meteorological parameters. This process is
somewhat labor-intensive and requires specialized software, but is preferable when modeling the
highest ranges of concentrations rather than using regression-based models, which can be unduly
influenced by extreme values.
Filter-based Models. Filter-based models adjust ozone and its precursor concentrations
by smoothing out the non-trend variation in the data (Rao et al., 1995; Flaum et al., 1996;
Milanchus et al., 1997; Porter et al., 2001). For example, ozone varies on diurnal, synoptic,
seasonal, annual, and interannual frequencies. Moving averages or band-pass filters can be used
to remove the higher-frequency variations in ozone concentrations to identify underlying trends.
These models assume that meteorology affects ozone concentrations at short- and long-term
3-7
-------
scales and that these components are separable. Applying a filter to the data removes shorter-
frequency variability while leaving the long-term underlying trend. A disadvantage of this type
of model is that it requires a data set that is at least three to four times longer than the longest-
frequency meteorological variation in order to identify and remove that variation. This technique
requires at least three to four years of historical data for development.
3.4 CORROBORATIVE ANALYSES
Once the conceptual model has been prepared, trends in ambient data identified and
quantified (with and without meteorological adjustments), and possible control effects identified,
the analyst can improve the confidence in the results by finding corroborative evidence for
correlation between the trends and the controls. In addition, the null hypothesis must be tested.
For example, Main et al., (1998) verified that ambient benzene did not decline at a site without
RFG. Multiple methods for detecting and quantifying trends in air quality caused by changes in
emissions (i.e., consensus) are needed to provide more confidence in the conclusions.
Corroborative analyses include
Trajectory analyses to separate local and transport days or identify source areas. These
results can then be used to further stratify and investigate the ambient data and verify the
source areas versus the spatial coverage of the control implementation. Other
investigations of transport include flux calculations to show the relative magnitude of a
pollutant or precursor concentration across a boundary or ventilation/recirculation
measures to show the relative importance of transport versus stagnation.
Source apportionment (positive matrix factorization [PMF], chemical mass balance
[CMB], principal component analysis [PCA], and others) applied to data collected before
and after a change, for example, to investigate changes in source profiles and source
categories.
Modeling or application of methods for estimating relative emission contributions
(Transported Emissions Assessment Kit [TEAK], Section 3.4.3) to identify source areas
or quantify expected changes.
Other measures of emissions, such as population and vehicle miles traveled (VMT), to
illustrate, in the absence of a decline in ambient concentrations, what changes may have
occurred but were obscured by increases in emissions caused by more emitters.
3.4.1 Transport and Trajectories
The following issues should be considered when assessing pollutant transport:
Definition of boundaries (and scale). Over which boundary is transport being assessed?
Where do pollutants and precursors transported into an area originate? What route(s) do
they take?
3-8
-------
Pollutants and continuing reactions. The pollutant and its precursors need to be
considered. How much of the high pollutant concentrations can be attributed to
transport?
Transport versus carryover (recirculation). Are pollutants transported into the area of
interest; are the pollutant and precursors from the area carried over from day to day; or is
the area experiencing a combination of transport and recirculation? How frequently does
transport occur?
Aloft and surface air quality and winds. What data are available and where are the sites
located? Are special studies data available?
Transport time. How long does it take for air parcels from upwind sources to reach the
receptor? When do pollutants and precursors arrive in the area?
Characteristics along the transport path. Are important pollutant or precursor emissions
sources situated along the transport path?
Methods for quantifying transport include computing the relative emissions in the upwind
and downwind areas using ratios of precursor emissions in the upwind airshed to those in the
downwind airshed; comparing the ratio of upwind to downwind emissions with emissions
accumulated along a typical trajectory path; and computing the ratio of upwind to downwind
emissions using meteorological and photochemical models.
Many tools are available for preparing trajectories and exploring the differences between
dates. A common approach is to prepare 72-hr backward trajectories for all sample dates using
the National Oceanic and Atmospheric Administration (NOAA) Hybrid Single-Particle
Lagrangian Integrated Trajectory (HYSPLIT) model (Draxler and Hess, 1997). For 24-hr
sampling periods, trajectories are prepared for every three, four, or six hours at two to three
heights. These trajectories can then be mapped as a spatial probability density (SPD°):
^ Count of hourly trajectory endpoints within search radius
Count of trajectories run (3 1)
The largest SPD values are in areas where the backward trajectories have spent the most time.
A conditional probability function (CPF) can then be applied to help interpret the SPD
results (Kim and Hopke, 2004; Kim et al., 2003, 2004; Ashbaugh et al., 1985). In CPF, the
transport patterns of the 20%-highest concentration days for a source contribution are compared
to the climatological transport patterns. After finding SPD0, backward trajectories for the
20%-highest concentration days are run and mapped (SPD'). This density is then compared to
the SPD for all days (i.e., the climatology), so that the differences in transport and source areas
on high concentration days of a given pollutant or source contribution are highlighted:
CoPIA' = SPD' - SPD0 (3-2)
This Conditional Probability Integrative Analysis (CoPIA) is similar to the CPF analyses
employed in other studies (Kim and Hopke, 2004; Kim et al., 2003, 2004; Ashbaugh et al.,
3-9
-------
1985); however, CoPIA is adapted to take advantage of tools available in a geographic
information system (GIS) framework. An example is shown in Figure 3-4.
Figure 3-4. Spatial probability density plots for all samples and for samples with
high contributions from a mining factor identified in source apportionment of
PM2.5 data collected in Phoenix, Arizona, during 2001-2003 (Brown et aL, 2005).
Also shown is the conditional probably plot overlaid with mining emissions.
In the example (Figure 3-4), source apportionment results for Phoenix, Arizona, PM2.5
data were investigated (Brown et al., 2004). Composition of one factor identified in the analysis
was consistent with copper mining emissions. On days with higher contributions from this
factor, CoPIA and the emission inventory indicated a higher probability that air parcels were
transported from an area with significant copper mining operations.
A similar approach was used by Grego et al., (2006) to investigate the impact of \(\
emissions reductions on ozone in the eastern United States. One of the approaches used in this
work was to compare pollutant concentrations on days when air parcels were transported from
the Ohio River Valley to days on which no transport occurred.
Limitations to trajectory analyses exist, such as spatial scale, that need to be considered,
especially in investigating local influences. The commonly used HYSPLIT model grid-spacing
is too large (40 km) to differentiate between "local" and transport influences. Local-scale
Conditional
Probability
Factor 4
Mining
Legend
CwKHMDJMm
¦ oixCI
A Supersite
High mining factor samples
Trajectory
Density
2000 - 2003
All Sc
Spatial
Probability
Density
Factor 4
Mining
3-10
-------
meteorology may not be modeled wellsuch as onshore/offshore flow near large bodies of
water. It is clearly important to ensure that a trajectory analysis is appropriate and representative
of the scale of interest.
3.4.2 Source Apportionment
Source apportionment analyses using receptor models are useful in quantifying the
impact of specific sources and source types on ambient concentrations and identifying how these
sources affect specific pollutants over time. The two receptor models currently in wide use are
CMB (Watson et al., 1990) and PMF (Paatero, 1997; Paatero and Tapper, 1994). CMB can be
applied to a single sample and quantifies sources by correlating known source profiles to ambient
data; thus, correct and area-appropriate source profiles are vital. PMF quantifies sources by
extracting the source profiles based on the internal variability and uncertainty of the ambient data
and, therefore, requires larger data sets with well-characterized uncertainties. Both models can
provide a wealth of information, but results need to be put in context; thus, development of a
conceptual model and use of trajectory analysis are needed to fully understand source
apportionment results.
CMB and PMF have been applied to routine and special study speciated PM2.5 data at
many rural and urban locations across the United States. Routine speciated PM2.5 data are
available at a number of rural locations as part of the IMPROVE monitoring network since the
early to mid-1990s, while data collection started in urban areas as part of the Speciation Trends
Network (STN) in 2001. Special study data, including molecular analysis of OC, have also been
collected and analyzed. Thus, a fairly large amount of data are available for source
apportionment of PM2 5, especially after 2001. VOC data, mainly collected as part of
Photochemical Assessment Monitoring Stations (PAMS) and special studies, have been used in
source apportionment applications, with most analyses in the 1990s using CMB; more recent
applications of PMF to VOC data have been made. Long-term speciated VOC data exist only at
a handful of sites. Lastly, receptor models have rarely been applied to air toxics, continuous PM,
and criteria pollutant data. Limitations to source apportionment include lack of site-specific
source profiles (CMB), lack of detection limits low enough so that sufficient data are available
(air toxics, trace metals), and lack of continuous data from source-specific markers (both CMB
and PMF).
Source apportionment analyses can be applied to ambient data to understand changes due
to known regulations by quantifying changes in
1. Concentrations of total mass (i.e., PM2.5 or total VOC) or of specific components (i.e.,
benzene or lead) attributed to a specific source or source type using an "average" profile
over the time period. Either model can be applied to a data set that spans the time frame
of the regulatory change, while either using (CMB) or determining (PMF) an "average"
profile over the time period. This method may have limitations because of the
use/determination of an "average" profile.
2. Profiles (i.e., fingerprints) of specific sources or source types (PMF only). CMB requires
a known profile; this analysis using PMF could then be used to provide "before" and
"after" profiles for CMB.
3-11
-------
3. Both profiles and contributions to ambient concentrations. For CMB application, the
changes in mass attributed to a specific source before and after a regulatory change will
rely solely on accurate source profiles of all sources before and after the changethis is a
challenge. For PMF application, sufficient data before and after the change are needed;
typically, at least 100 samples are desirable, although more than 300 samples before and
after the change would yield more certain results. Thus, if accurate profiles are available
before and after a known regulatory change but data are limited, CMB application is
likely the best method. PMF should be applied by itself, or concurrent with CMB if
sufficient data are available.
To ensure quality results, analytical and collection methods need to be the same between
source profiles and ambient data, or the differences in data between methods need to be known
and quantified.
Source apportionment analyses have been applied to data sets over a long period of time
in few locations. In Los Angeles, California, CMB and PMF analyses have been performed on
VOC data; an example summarizing the average results in 1987 (CMB) and 2001-2003 (PMF) is
shown in Figure 3-5. It shows that total VOC concentrations have decreased over time, mostly
because of a decrease in mobile source influence. While mobile sources have decreased, non-
mobile sources have not changed.
500
450
O 350
CL
§ 300
ro
250
0
§ 200
O
§ 150
100
50
0
Azusa, Summer-Fall 1987 Azusa, Summer 2001-2003
B Other
~ Mobile
Figure 3-5. Average VOC mass (ppbC) apportioned to mobile and non-mobile
sources at Azusa in Los Angeles, California, in 1987 (Harley et al., 1992) using
CMB and in 2001-2003 (Brown et al., 2005) using PMF.
3-12
-------
3.4.3 Estimating Relative Emission Contributions
Performing comparisons between trends in emission inventories and trends in ambient
monitoring data requires an understanding of which emissions sources are likely impacting a
given monitor during a specified time period. A suite of desktop tools has been developed called
the Transported Emissions Assessment Kit (TEAK) for examining the probability of transport,
source locations, and emissions with respect to air quality issues. TEAK employs SPD to
explore where air masses originate and travel under certain conditions (e.g., on days with poor
visibility or days with high ozone concentrations). SPD uses GIS technology to convert a large
ensemble of modeled trajectories into a quantitative surface density plot that is easy to visualize.
In addition, SPD output can be combined with emission inventory information to assess the
emission impact potential (EIP) of a given source or geographic region. The result is an estimate
of the relative potential for emissions from different facilities, counties, or grid cells to impact
the receptor site.
The SPD is used to weight the emissions from individual counties and estimate the
potential for specific upwind areas to impact the receptor. The EIP of any county is calculated as
Ev *D0
EIP = ¦ p 0
/(distance) (3_3)
where:
Ep = county total emissions of pollutantp
Do = spatial probability density at the county centroid
/ = function of distance between county and receptor
Using a tool like TEAK can help ensure that the emission trends analyses are comparing
emissions most likely impacting a given sitesomething that can change over time. For
example, overall transport patterns for the Fresno, California, summer ozone seasons of 2002 and
2004 were analyzed and SPD plots produced for each ozone season. Figure 3-6 shows that the
SPD plots are very similar for these two years. However, Figure 3-7 shows the difference
between the two plots and reveals that in 2002, transport from the north of the San Joaquin
Valley (SJV) was more likely (up to 10%) than in 2004. In summer of 2004, Fresno experienced
about 25% more transport from the Pacific Coast. In terms of ozone exceedances, 2004 was a
cleaner year than 2002.
3-13
-------
(a)
(b)
June* September 2002
Spatial Probability Density
~ 0 CO-004
H 005-01?
| 0U-0.27
I aa-0.49
¦ 030 ¦ 0 71
¦ 072-100
June - September 2004
Spatial I Probability Denalty
^ 000 - 0.08
~ ooe-o ie
L| 0-17 0 33
¦ 0 34 - 0 52
¦ 0 53 - 073
B 074-100
Figure 3-6. Spatial probability density plots for Fresno, California, for the
summers of (a) 2002 and (b) 2004.
3-14
-------
Transport Pattern Difference
Backward Trajectory Prevalence
| More in 2002
] Slightly more in 2002
Similar or none
"HI Sightly more in 2004
¦ More in 2004
Figure 3-7. Transport pattern differences between the summers of 2002 and 2004
for Fresno, California.
The EIP may be divided by a distance function to roughly account for dilution and
increased uncertainty in model outputs far from the receptor site. However, for the example
shown in Figure 3-8,/equals 1. Figure 3-8 shows the SOx and NOx EIP values by county for
the 20%-worst and 20%-best visibility days for measurements made at Ftercules-Glades,
Missouri. These analyses were used to help the central states assess important clean-air
corridors1 and the importance of emissions sources within and outside the central states that
contribute to poor visibility.
1 In this example, clean-air corridor is defined as the transport pathway predominantly associated with 20%-best
visibility days.
3-15
-------
Figure 3-8. Geographic distributions of (a) S02 and (b) NOx EIP for the
20%-worst visibility days (red bars) and 20%-best visibility days (blue bars)
observed at Hercules-Glades, Missouri.
3.4.4 Other Corroborative Evidence
Other corroborative evidence or hypothesis testing is needed to add confidence to the
results. Figure 3-9 shows an example exploration of the ambient changes in PM2.5 organic
carbon mass (OM = 1.8*OC) concentrations potentially tied to gasoline regulations, including
the introduction of RFG in Washington, DC. This example combines several of the
considerations discussed in the technical approach.
The premise behind the example was that RFG targeted reductions in the aromatic
content of gasoline, and aromatic hydrocarbons have the potential to form secondary organic
aerosol (SOA), which is a component of OM in PM2.5. The question was asked whether future
changes in gasoline formulation might result in OM reductions, and the historical data record
was explored to identify evidence of such a link. Figure 3-9 shows the gasoline aromatics
content in the top portion; aromatics content of pre-1995 gasoline was about 32%. The lower
half of the figure shows annual average concentrations of OM, other PM2.5 components not
expected to be affected by the change in gasoline formulation (nitrate, metal oxides [soil]), and
VMT estimates. The controls that may affect OM concentrations are summarized in the
accompanying table.
»SW«lVwb«Tt>»*i
8*tr Oar*
3-16
-------
c 28
o
tA 24
CO
20
c
16
8
2 -
% Aromatics
a. b
d e fg h
S
/-*
ir
\
r ii m r
OMC (+ STD)
-B-
EC (i STD)
±
S04 (± STD)
VMT
' Projected VMT values
i-rsjonmcoi'-cooio
000otJ><7)0
o>o)0)o>o)©(j>(no)o
*-
-------
analysis of the annual average concentrations does indicate that OM and EC concentrations
decreased coincidentally with the introduction of RFG in 1995. However, the decrease in
concentrations of other PM mass components like nitrate and sulfate makes attribution of this
decrease problematic because it could simply be a systematic analytical or network difference.
An additional analysis of major components of PM mass at the nearby rural Shenandoah
monitoring site indicated that OM and EC concentrations were unchanged from 1994-1995.
Personnel at the University of California, Davis, Crocker Nuclear Laboratory, which operates the
IMPROVE monitoring network, confirmed no network changes during this time (White, 2005).
Thus, it is unlikely that the decrease in OM or EC concentrations in Washington, DC, was
caused by a systematic analytical or sampling change.
Other considerations in this example were not fully explored during this analysis but
would be useful to improve the understanding of the relationship between the introduction of
RFG and ambient OC changes. For example, meteorological adjustments were not made to the
data to account for differences from year to year. However, statistically significant differences
between three-year averages before and after the 1995 RFG introduction indicate that
meteorology may not be the sole source of the changes. Other indicators to test include trends in
the sulfate, nitrate, OM, and EC fractions in addition to the concentrations. Other hypotheses
might be tested:
Did SO2 emissions decline over the same time period?
Did ozone concentrations decline during this period, and, thus, change the photochemical
production of OC (and possibly sulfate and nitrate)?
Were similar changes observed in other cities with (or without) RFG introduction during
this time frame?
This example illustrates the need to critically think through the analysis and identify the
issues and considerations. Once the questions have been asked, the analyst can develop an
approach to answer them.
3-18
-------
4. SELECTING EXAMPLES FOR CASE STUDIES
4.1 OVERVIEW
We begin the process of selecting application examples for the technical approach
discussed in this document. A spectrum of cause-and-effect types needs to be considered. For
example, a local control/local effect case study could be one of the simpler analyses in which a
regional effect is not expected, the emission control influenced a primary pollutant, and
sufficient data existed before and after the change (e.g., Philadelphia, Pennsylvania, or St. Louis,
Missouri, lead concentrations). In contrast, a combination of applied regional and local
emissions controls affect a range of primary and secondary pollutants over varying spatial scales,
thus complicating an individual analysis of one control and the resulting air quality.
Candidates for potential controls investigation include NOx reductions (e.g., NOx SIP
call), S02 reductions (Acid Rain program), RFG, diesel retrofits, and maximum achievable
control technology (MACT) (for air toxics). Earlier discussions with the EPA indicate that
several cities are of interest for this analysis as well. The following sections provide ideas for
consideration.
4.2 NOx AND S02 REDUCTIONS
The objective of this control scenario is to assess the impact of NOx and SO2 emissions
reductions on ambient air quality in a selected city. A likely candidate city is Detroit, Michigan,
because of its wealth of data and mix of local and transported emissions. Using the thought
process described in Sections 2 and 3, the following items will need to be addressed:
Air pollutants expected to be affected. NOx, SO2, ozone, PM2.5 mass, PM sulfate, and PM
nitrate are pollutants expected to be affected. As shown in Figures 3-1 and 3-2, speciated
PM2.5 data are available for 2000-2004. Because ozone is a consideration due to the NOx
reductions, VOCs are also important.
Spatial scale of pollutants. This assessment would likely focus on the secondary
components (sulfate, nitrate, ozone) and the PM2.5 mass, which is a mix of primary and
secondary components. A mix of regional (sulfate, nitrate, ozone) and urban-level
(nitrate) scales needs to be addressed. Regional-level concentrations are important, so a
mix of urban and rural monitoring data is needed.
Spatial scale of the control measure. The NOx SIP call and Acid Rain programs affected
power generation and large industrial facilities. One area experiencing the most
reductions was the Ohio River Valley. Other controls of NOx may be important. As with
the pollutants, a mix of regional- and urban-level scales needs to be addressed.
Knowledge of local controls in Detroit is needed, such as whether other controls were
implemented for PM2.5 mass.
Timing and magnitude of changes. A summary of NOx and VOC emissions controls is
summarized in the EPA (2005) evaluation of the NOx SIP call. A summary of the SO2
4-1
-------
reductions needs to be prepared, including information about the potential magnitude of
changes in air quality.
Magnitude of expected ambient changes. The EPA prepared model predictions at a
regional level, at least for ozone. A summary of model results needs to be compiled.
Statistical metrics to consider. Focusing on the PM2.5 mass in Detroit, seasonal and
annual averages of PM2 5 mass and composition need to be investigated. Meteorological
adjustments are likely important to remove meteorological influences.
Basic analyses. Section 3.1 analyses need to be performed (or previous analysis results
reviewed). The spatial patterns in PM2.5 and its components by season and year,
including urban-rural comparisons, are of particular interest for analysis. The spatial
pattern in the PM2.5 mass data indicates that if a control on industrial PM or precursor
emissions were implemented, the effects would likely be observed at the Dearborn site,
for example.
Corroborative analyses. Separating the days dominated by transport from more local
days is important for this assessment, but hampered by the model resolution. High
sulfate or nitrate concentration days should be examined with SPD plots, for example.
Extensive source apportionment analyses have been performed with data from Detroit;
these analyses can be revisited to look for trends over time.
Other considerations. Less emission inventory information is available from the
Canadian sides of the Detroit area, which complicates an assessment of trajectories
coupled with the magnitude of emissions.
4.3 MULTIPLE POLLUTANT REDUCTIONS
The objective of this control scenario is to assess the impact of RFG introduction on
multiple pollutants in a selected city. A likely candidate city is Houston, Texas, with other
choices including Phoenix, Arizona, and Washington, DC (initial investigation is shown in
Figure 3-8); previous analysis of VOCs includes Main and O'Brien (2001) and Main et al.
(1998). Using the thought process described in Sections 2 and 3, the following items need to be
addressed:
Air pollutants expected to be affected. For RFG, NOx, ozone, total VOC, benzene, total
aromatic hydrocarbons, 1,3-butadiene, formaldehyde, and possibly PM2.5 mass, nitrate,
OC, and EC could be affected. Key to this analysis is the availability of speciated VOC
data before and after the change in gasoline formulation. Highly time-resolved data (i.e.,
a 1-hr to 3-hr duration) are needed because of complicated meteorological and
monitoring site issues. A focus on sampling during the morning rush hour should be
considered because motor vehicle emissions are highest, mixing heights are low (thus,
concentrations are high), and reactivity is minimized.
Spatial scale of pollutant. The focus of this assessment is primary and secondary
pollutants, with an emphasis on the urban scale. Local influences near the monitors need
to be considered as well. Transport at this scale is important; thus, meteorological data
collocated with the air quality data will be required.
4-2
-------
Spatial scale of the control measure. RFG was applied to a selected urban area. If
Houston, Texas, was selected for study, the production of gasoline in the area would
mean that the industrial emissions might be affected by RFG introduction as well. Other
VOC or NOx controls in the area need to be identified. For Houston, offshore VOC and
NOx emissions should also be considered.
Timing and magnitude of changes. RFG changes to benzene, aromatics, VOC, and NOx
emissions have been documented and should be available for the Houston area.
Magnitude of expected ambient changes. Model predictions have been made for ozone
changes caused by the introduction of RFG. Early model work indicated formaldehyde
increases (if MTBE was used as the oxygenate) and 1,3-butadiene decreases could be
expected with RFG. A summary of model results needs to be compiled. Less is known
about the potential impacts on PM2.5 mass and components from RFG.
Statistical metrics to consider. Focusing on the VOCs, statistical summaries of
concentrations, weight percents, and ratios would be needed. For ozone, the typical suite
of metrics should be explored. Meteorological adjustments are likely important for
ozone; less work has been performed on the VOCs. The VOCs of interest are influenced
by temperature and photochemical production and destruction, so it is likely that
meteorological adjustments need to be explored.
Basic analyses. Section 3.1 analyses need to be performed (or previous analysis results
reviewed). Of particular interest in this example is wind direction relative to mobile and
industrial sources. Figure 4-1 shows a satellite photo of the Clinton Drive site in the
Houston Ship Channel. The U.S. 1-610 freeway is located west of the site, a residential
area located to the north, and industrial sources located to the south and east. To isolate
the mobile source emissions, samples when winds are from the west are needed. Data
from other sites outside the industrial section of town are also available for corroborative
analyses.
Corroborative analyses. Analysts need to look for changes in pollutants not expected
with the introduction of RFG. For example, in the PM2.5 data, changes in the soil
component would not be expected with the introduction of RFG. For the VOCs,
acetylene or the biogenic isoprene should not have been affected the introduction of by
RFG. Extensive source apportionment analyses have been performed with data from
Houston; these analyses can be revisited to investigate trends over time. Some emission
factor analyses using the tunnel in the Houston Ship Channel have been performed.
Other considerations. The onshore-offshore flow reversals in Houston and the
complexity and abundance of industrial emissions are factors that make analysis of data
in Houston particularly challenging.
4-3
-------
AutoGC
THC/Nox
Figure 4-1. Clinton Drive site near the Houston Ship Channel (green dot in the
center of the photo).
4.4 DIESEL EMISSIONS REDUCTIONS
The objective of this control scenario is to assess the impact of diesel vehicle retrofits to
reduce PM emissions in a selected city. A candidate city is Los Angeles, California, but other
cities may have cleaned up their bus fleets. Using the thought process described in Sections 2
and 3, the following items will need to be addressed:
Air pollutants expected to be affected. The focus is likely EC, but other pollutants that
could be affected include NOx, SO2, PM2.5 mass, OC, PM sulfate, and PM nitrate. Highly
time-resolved data, such as hourly (or sub-hourly) Aethalometer BC data, are likely
required to isolate the time periods and days of week with the most activity.
Spatial scale of pollutants. The focus of this assessment is on primary EC emissions at
an urban scale or smaller.
Spatial scale of the control measure. Diesel particulate matter (DPM) emissions are
typically a relatively small portion of the PM;.? mass in most urban areas (using EC as a
marker, although not a unique one, for DPM). The effect of a diesel retrofit program is
likely measurable only near areas with a high density of retrofitted equipment. An ideal
monitoring site would be located near a bus yard or transit center, for example. Local
knowledge of program timing and number of buses, etc. is needed.
4-4
-------
Timing and magnitude of changes. A summary of the expected emissions reductions
needs to be prepared.
Magnitude of expected ambient changes. Based on the expected emissions reductions, an
estimate of expected ambient changes needs to be made. Are model results available for
these types of scenarios?
Statistical metrics to consider. Summary statistics of concentrations, ratios of EC-to-PM
components, and the fraction of EC in the PM2.5 will be needed. If dual channel
Aethalometer data are available, an investigation of the channels may provide
additional information about DPM. The signal produced by the near-ultraviolet (UV)
channel (BC370) may be useful for detecting UV-absorbing organic compounds. Jeong et
al. (2004) reported an increase in BC370 relative to BCgso when a Pennsylvania site was
impacted by smoke from a large forest fire in Quebec (Jeong et al., 2004). It has also
been suggested that BC370 is an indicator of fresh diesel emissions (Hansen, 1998).
Meteorological adjustments are not likely very important for primary pollutants measured
so close to a source.
Basic analyses. Section 3.1 analyses need to be performed (or previous analysis results
reviewed). Assessment of weekday-weekend differences in PM and its components is of
particular interest when investigating DPM from mobile sources.
Corroborative analyses. Source apportionment may be useful in this analysis. Maps and
spatial distribution of sources of DPM are key supporting information.
Other considerations. The relative importance of bus emissions to the total PM (or EC)
mass and the influence from other diesel emissions (shipping, construction equipment,
rail) need to be considered.
4.5 AIR TOXICS REDUCTIONS
The objective of this control scenario is to assess the reduction of air toxics as a result of
the introduction of MACT regulations. A candidate city is Los Angeles, California, but other
cities may have applied a range of MACT regulations focusing on specific pollutants or source
types. Using the thought process described in Sections 2 and 3, the following items will need to
be addressed:
Air pollutants expected to be affected. The air toxics affected by control programs are
MACT-specific. As an example, consider tetrachloroethylene (i.e., perchloroethylene, or
perc) reductions in Los Angeles. Perc is primarily emitted by dry cleaners.
Spatial scale of pollutants. The focus of this assessment is on primary emissions at an
urban scale or smaller.
Spatial scale of the control measure. The spatial scale of the control is also MACT-
specific. This control will depend on the size and number of facilities affected and their
proximity to the monitors. As with other examples, local knowledge of program timing
and affected facilities is needed.
4-5
-------
Timing and magnitude of changes. A summary of the expected emissions reductions
needs to be prepared.
Magnitude of expected ambient changes. Based on the expected emissions reductions, an
estimate of expected ambient changes needs to be made.
Statistical metrics to consider. Summary statistics of concentrations and possibly ratios
of the affected pollutant to other air toxics will be needed.
Basic analyses. Section 3.1 analyses need to be performed (or previous analysis results
reviewed).
Corroborative analyses. Emissions estimates, promulgation and compliance dates, and
maps and spatial distribution of pollutant sources are key supporting information.
Figure 4-2 shows an example from the Los Angeles basin for perc emissions reductions
from dry cleaners. The figure shows MACT implementation and compliance dates,
National Emission Inventory (NEI) emissions data for dry cleaners in Los Angeles
County, and annual average concentrations of perc at three sites in the Los Angeles air
basin. Note that two of the sites indicate a clear decline that seems to correspond to the
MACT, but one site does not.
Figure 4-2. Annual average perc concentrations at three sites in Los Angeles,
California, and the NEI estimates of perc emissions.
Other considerations. One consideration is to understand how the air toxic pollutant was
reducedthrough controls or through substitution of a different chemical. If monitoring
is performed for that new chemical, the analyst could look for increases. For area source
4-6
-------
emissions, more may need to be known about where the emissions sources are actually
located rather than relying on typical surrogates, such as population density or land use.
4.6 SUMMARY
The examples in this section are meant to provide indications of the thought process
necessary to select examples for investigation. Other cities and controls of interest are available
and a similar approach can be taken to aid in the selection process. Once the selection of the
control measures and cities for investigation is complete, the technical approach discussed in this
report will be applied to the case studies.
4-7
-------
-------
5. REFERENCES
Ashbaugh L.L., Malm W.C., and Sader W.Z. (1985) A residence time probability analysis of
sulfur concentrations at Grand Canyon National Park. Atmos. Environ. 19 (8), 1263-1270.
Brown S.G., Frankel A., Raffuse S.M., Hafner H.R., and Anderson D.J. (2004) Source
apportionment of PM2 5 in Phoenix, Arizona, using positive matrix factorization. J. Air &
Waste Manage. Assoc. (submitted for publication).
Brown S.G., Frankel A., and Hafner H.R. (2005) Source apportionment of VOCs in the Los
Angeles area using positive matrix factorization, (submitted) (STI-2725).
Cox W.M. and Chu S.H. (1993) Meteorologically adjusted ozone trends in urban areas: a
probabilistic approach. Atmos. Environ. 27B, 425-434.
Cox W.M. and Chu S.H. (1996) Assessment of interannual ozone variation in urban areas from a
climatological perspective. Atmos. Environ. 30, 2615-2625.
Draxler R.R. and Hess G.D. (1997) Description of the HYSPLIT 4 modeling system. Technical
memorandum by NOAA, ERL ARL-224, December 24.
Flaum J.B., Rao S.T., and Zurbenko I.G. (1996) Moderating the influence of meteorological
conditions on ambient ozone concentrations. J. Air & Waste Manag. Assoc. 46, 35-46.
Frank N.H. (2005) Retained nitrate, hydrated sulfates, and carbonaceous mass in federal
reference method PM2.5 for six eastern U.S. cities. J Air Waste Manag Assoc (accepted).
Grego E., Porter P.S., Gilliland A., and Rao S.T. (2006) An examination of the impact of
nitrogen oxides emission reductions on ozone air quality over the eastern United States.
J. Appl. Meteorol. (submitted).
Hafner H.R. (2003) Data validation workshop. Prepared for the Lake Michigan Air Directors
Consortium and U.S. Environmental Protection Agency, Region V, Chicago, IL, by
Sonoma Technology, Inc., Petaluma, CA, STI-903510-2424, October.
Hafner H.R., McCarthy M.C., Brown S.G., Raffuse S.M., Shields E., Stiefer P.S., and Wheeler
N.J. (2004) Workshop on air toxics data analysis. Preparedfor Lake Michigan Air
Directors Consortium; presented in Rosemont, IL, June 2-3 (STI-903555).
Hafner H.R., Brown S.G., McCarthy M.C., Frankel A.P., Miller D.S., Raffuse S.M., and Roberts
P.T. (2006) Midwest Urban Organics Study. Presented at the Lake Michigan Air
Directors Consortium Data Analysis Workgroup Meeting, Des Plaines, IL, January 26-
27 by Sonoma Technology, Inc., Petaluma, CA.
Hansen A.D.A. (1998) The Aethalometer. Instument guide prepared by the Magee Scientific
Company, Berkeley, CA.
5-1
-------
Harley R.A., Hannigan M.P., and Cass G.R. (1992) Respeciation of organic gas emissions and
the detection of excess unburned gasoline in the atmosphere. Environ. Sci. Technol. 26,
2395-2408.
Jeong C.H., Hopke P.K., Kim E., and Lee D.W. (2004) The comparison between thermal-optical
transmittance elemental carbon and Aethalometer black carbon measured at multiple
monitoring sites. Atmos. Environ. 38 (31), 5193-5204.
Kenski D. (2006) Lake Michigan Air Directors Air Consortium, Des Plaines, IL. Personal
communication to Hilary Hafner. Personal communication, February.
Kim E., Hopke P.K., and Edgerton E.S. (2003) Source identification of Atlanta aerosol by
positive matrix factorization. J. Air & Waste Manage. Assoc. 53, 731-739.
Kim E. and Hopke P.K. (2004) Improving source identification of fine particles in a rural
northeastern U.S. area utilizing temperature-resolved carbon fractions. J. Geophys. Res.
109 (D9).
Kim E., Hopke P.K., and Edgerton E.S. (2004) Improving source identification of Atlanta
aerosol using temperature resolved carbon fractions in positive matrix factorization.
Atmos. Environ. 38, 3349-3362.
Larsen L. (1999) A semi-empirical model relating meteorology and ozone in the San Francisco
Bay Area. Air Pollution VII, WIT Press, Southampton, UK.
Main H.H., Roberts P.T., and Reiss R. (1998) Analysis of photochemical assessment monitoring
station (PAMS) data to evaluate a reformulated gasoline (RFG) effect. Report prepared
for the U.S. Environmental Protection Agency, Office of Mobile Sources, Fuels and
Energy Division, Washington, DC, by Sonoma Technology, Inc., Santa Rosa, CA, STI-
997350-1774-FR2, April. Available on the Internet at
-------
Milanchus M.L., Rao S.T., and Zurbenko I.G. (1997) Discerning changes in anthropogenic
ozone in the presence of meteorological fluctuations. J. Environ. Econ. Manage., 17-21.
National Research Council (1991) Rethinking the Ozone Problem in Urban and Regional Air
Pollution, National Academies Press, Washington, DC.
Paatero P. and Tapper U. (1994) Positive matrix factorization: a non-negative factor model with
optimal utilization of error estimates of data values. Environmetries 5, 111-126.
Paatero P. (1997) Least squares formulation of robust non-negative factor analysis.
Chemometrics and Intelligent Laboratory Systems 37, 23-35.
Porter P.S., Rao S.T., Zurbenko I.G., Dunker A.M., and Wolff G.T. (2001) Ozone air quality
over North America: part IIan analysis of trend detection and attribution techniques.
Atmos. Environ. 51, 283-306.
Rao S.T., Zalewsky E., and Zurbenko I.G. (1995) Determining spatial and temporal variations in
ozone air quality. J. Air & Waste Manag. Assoc. 45, 57.
Thompson M.L., Reynolds J., Lawrence H.C., Guttorp P., and Sampson P.D. (2001) A review of
statistical methods for the meteorological adjustment of tropospheric ozone. Atmos.
Environ. 35, 617-630.
U.S. Environmental Protection Agency (2005) Evaluating ozone control programs in the eastern
United States: focus on the NOx budget trading program, 2004. Report prepared by the
U.S. Environmental Protection Agency, Office of Air and Radiation, Office of Air
Quality Planning and Standards, Office of Atmospheric Programs, Washington, DC,
August. Available on the Internet at
last accessed March 9, 2006.
Watson J.G., Robinson J.F., Chow J.C., Henry R.C., Kim B.M., Pace T.G., Meyer E.L., and
Nguyen Q. (1990) The USEPA/DRI chemical mass balance receptor model, CMB 7.0.
Environ. Soft. 5, 38-49.
White W. (2005) IMPROVE network at the Crocker Nuclear Laboratory at U.C. Davis, Davis,
CA. Telephone conversation with Steven Brown, Sonoma Technology, Inc., Petaluma,
CA. November 22.
5-3
------- |