United Stales
Environmental Protection
Agency
Office of Air Quality
Planning and Standards
Research Triangle Park NC 27711
EPA-454/R-96-006
October 1996
Air
PHOTOCHEMICAL ASSESSMENT
MONITORING STATIONS
1996 DATA ANALYSIS
RESULTS REPORT
-------
A
EPA-454/R-96-006
Acknowledgements
Revision Number: 0
Date: November 1996
Page: i
cknowledgments
As a compilation of existing analyses and studies, this document would simply not have possible
without the generosity and creative efforts of the PAMS data analysis community. Unfortunately,
this collection of individual air quality analysts, modelers, and statisticians is too numerous to
allow acknowledgment by name or organization. Instead, the report attempts to credit all direct
and indirect contributions to the effort through footnotes within and references following each
chapter. Any omissions are entirely accidental and will be corrected in future editions of this
report.
The authors wish to acknowledge the invaluable contribution of Mr. William F. Hunt, Director of
the Emissions, Monitoring and Analysis Division. In addition to his role as senior manager to all
the report's authors and the creative energy behind the PAMS program, Mr. Hunt conceived of
the idea of this document and provided the opportunity and instruction which allowed its
development. We would also like to thank the following Group Leaders within EMAD, Mr. Joe
Tikvart, Mr. David Mobley and Dr. Dave Guinnup for their encouragement and empowerment to
craft a document true to the subject matter and to the intended audience.
A list of the report's authors (and their phone numbers') is provided below. The reader is
encouraged to contact the authors to convey comments and suggestions, and/or to request
additional information. Each chapter's lead author is italized.
Chapter 1 Chapter 2 Chapter 3
Shao-Hang Chu (5382) Rich Scheffe (4650) Tom Pace (5634)
James Hemby (5459) Ned Meyer (5594) Mark Schmidt (2416)
Bill Cox (5563) Ellen Baldridge (5684) Chet Wayland (4603)
Desmond Bailey (5248)
For all authors, the area code is "919" and the prefix is "541".
-------
EPA-454/R-96-006
Acknowledgements
Revision Number: 0
Date: November 1996
Page: ii
Chapter 4 Chapter 5
Rich Scheffe (4650) Mark Schmidt (2416)
Rhonda Thompson (5538)
All authors may also be reached via e-mail. All e-mail addresses are of the form
"lastname.firstname@epamai\.epa.gov".
-------
EPA-454/R-96-006
Table of Contents
Revision Number: 0
Date: November 1996
Page: 1
ontents
Executive Summary E-l
Introduction 1-1
Purpose I-1
Document Organization I-1
PAMS Brief Description 1-2
Why PAMS? 1-2
Regulatory Requirements 1-2
Status 1-3
Further Information 1-4
Comments 1-4
Request for Additional Copies 1-5
Chapter 1: Characterization of Ambient Air Quality for Ozone and Its Precursors 1-1
Introduction 1 -1
Episode Characterization Using Meteorological Measurements 1-1
Characterizing Episode "Severity" 1-1
Determining Source/Receptor Orientations Corresponding to Ozone
Conducive Conditions 1-2
Identifying Critical Circulations Associated with High Ozone Events 1-2
Identifying Boundary Layer Structures Associated with High Ozone
Events 1-3
Episode Charatenzation Using Air Quality Measurements 1 -3
Indicators of Ozone Episodes 1-4
Distinguishing Among Episode Types 1-4
Characterizing Precursor Species During Episodes 1-4
Assessing Air Mass Aging Using PAMS Precursor Data 1-5
Temporal Variation in PAMS Precursor Data 1-6
Statistical Models of Relationships between Ozone and Precursors 1-8
References 1-13
-------
EPA-454/R-96-006
Table of Contents
Revision Number: 0
Date: November 1996
Pace. 2
Chapter 2: PAMS Data in Support of Ozone Modeling Applications 2-1
Introduction 2-1
Model Overview 2-1
Model Evaluation Using PAMS Data 2-3
Example: Los Angeles, CA 2-3
Example: Houston Ship Channel 2-3
Development and Testing of Model Inputs 2-4
Episode Selection and Domain Specification 2-4
Development of Meteorological Inputs and Meteorological Model
Evaluation 2-4
Mixing Depth 2-5
Wind Fields 2-6
Additional Uses for PAMS Meteorological Data 2- 7
Development and Evaluation of Emissions Inputs 2-8
Discussion of AQSM Performance and Corresponding Uses of PAMS Air
Quality Data 2-8
PAMS and Compensating Errors 2-9
Suggested Uses of PAMS Data for Model Evaluation bu Compound
Class 2-9
Total NMOC and NMHC 2-10
Speciated VOC and Carbonyls (Isoprene and Formaldehyde) 2-11
Nitrogen: NOJNO, NO2), NO, 2-12
Chapter 3: Evaluation of Emission Factors, Models and Inventories with PAMS
Data 3-1
Introduction 3-1
Background 3-1
Potential lnvenlor\ Problems 3-2
Difficulties in Comparing Ambient Data and Emissions Estimates 3-3
PAMS Results 3-4
Examples of Indicator Species or Compounds (Tracers) 3-5
Examples Using NMOC/NO? Directional and Time Series Analyses 3-7
Example of Inventory Evaluation for Lake Michigan Inventory 3-9
Examples Using Multivariate Analyses and Chemical Mass Balance (CMB) 3-10
Example of Inventory Evaluation in Atlanta 3-11
Example of Inventory Evaluation in Southern California 3-11
Case Study-Example of Inventory Evaluation In Houston, Texas 3-12
Conclusions 3-14
References 3-14
-------
EPA-454/R-96-006
Table of Contents
Revision Number: 0
Date: November 1996
Page: 3
Chapter 4: Observational Based Methods for Determining VOC/NO, Effectiveness 4-1
Introduction 4-1
Empirical Techniques 4-1
VOC/NOX Ratios 4-1
Reactive (Oxidized) Nitrogen (NO^ NOZ) and Ozone Correlation
Techniques 4-3
Observational Models 4-3
Smog Production Algorithm - MAPPER Program 4-3
GIT Model 4-5
ChapterS: Quality Assurance 5-1
Introduction 5-1
Data Assessment 5-1
NPAP and Proficiency Studies 5-1
Precision and Accuracy Data 5-2
Data Validation 5-3
Summary Statistics and Historic Precedence (Scatter Plots) 5-4
Frequency Distributions 5-4
Spatial and Temporal Plots 5-5
Inter-Site Comparisons and Inter-Species Comparisons 5-5
References 5-6
-------
EPA-454/R-96-006
Executive Summon
Revision Number: 0
Date: November 1996
Page: i
E
xecutive Summary
Having successfully progressed through its planning, design and implementation stages,
the Photochemical Assessment Monitoring Stations (PAMS) program is currently undergoing a
shift in emphasis to focus more acutely on the analysis and interpretation of the data generated.
Such a change in direction is necessary to fully realize the value of this rich and voluminous data
set in a regulatory and policy context.
The PAMS networks produce a wealth of information invaluable to the development and
evaluation of ozone control strategies and programs. In addition to providing a long-term
perspective on changes in atmospheric concentrations of ozone and its precursors, the PAMS
program will specifically help to improve emissions inventories, assist in evaluating the
performance of photochemical grid models, furnish information to evaluate population exposure,
and provide routine measurements of selected hazardous air pollutants. Data from PAMS will
also allow for the development of the critical feed-back mechanism to evaluate the efficiency and
effectiveness of emission control programs. Most importantly, PAMS will assist in deriving a
more complete understanding of tropospheric ozone formation and transport, so that we may
move toward the best solution to this complex environmental problem.
This document presents example analyses illustrating the utility of PAMS data across the
range of ozone management applications. Where appropriate, limited critical evaluation is also
included to suggest future program refinements. The document is organized in a functional
manner based on mix of methodology and objectives. The Chapters describe general data
characterization approaches, methods for evaluating emissions inventories, the relationship of
PAMS data to emissions and observation based modeling, and quality assurance of the PAMS
data.
CHARACTERIZATION OF AMBIENT AIR QUALITY FOR OZONE AND ITS
PRECURSORS
Accurate characterization of ozone and its precursors is extremely important for
understanding tropospheric ozone formation and accumulation, and crafting effective control
strategies to address this environmental issue. Analyses have demonstrated that PAMS data are
invaluable in characterizing ozone episodes and identifying features that may be linked to
significant pollutant transport. For example, evidence of potentially significant pollutant transport
-------
EPA-454/R-96-006
Executive Summary
Revision Number: 0
Date: November 1996
Page: ii
1M4 Northeast Air Chanty Study
V«ltay.PA July-aim
1W4 NerthMst Air Qtailty Study
Figure E-l. Time series cross-section of winds, mixing depth, and inversion conditions measured on July 12-13,
1994 at Bermudian Valley , PA, indicating jet formation and mixing depths. The thin solid line
denotes the height of the mixed layer and the thick line denotes the subsidence inversion. Each wind
barb indicates direction and speed. (Lindsey et al., 1995).
(e.g. detection of a nocturnal jet as shown in Figure E-l) can be identified by combining PAMS
hourly surface and upper air meteorological data. In addition, the PAMS requirement for routine
measurements of organic species allows improved characterization of precursor conditions
associated with ozone episodes and provides, for the first time, a data set sufficiently detailed to
power statistical investigations of the relationships between ozone and its precursors. For
example, Figure E-2 compares observed ozone levels from Philadelphia, PA in 1994 with
predicted ozone levels produced by two statistical models. The first uses only meteorological
data to predict ozone levels while the second adds hydrocarbon data as an input. The
improvement in the model's ability to "explain" downwind ozone (when hydrocarbon data are
included) clearly demonstrates the value of speciated volatile organic compound (VOC) data in
accurately characterizing and understanding ozone concentrations. A more detailed description of
how PAMS data can be used to better characterize ambient conditions can be found in Chapter 1.
PAMS DATA IN SUPPORT OF OZONE MODELING APPLICATIONS
PAMS data support emissions-based model (EBM) applications by providing additional
information for evaluating (1) model predicted concentrations for ozone and its precursors; and
(2) meteorological and emissions inputs which drive model simulations. The use of PAMS data
-------
EPA-454/R-96-006
Executive Summon
Revision Number: 0
Date: November 1996
Page: in
Model 1 - Meteorological Data
1994 PAMS Data; Philadelphia, PA
0.15
Model 2 - Meteorological and Hydrocarbon Data
1994 PAMS Data. Philadelphia. PA
0.15
O.l
o °'05
005 01
Predicted Ozone Levels
0 15
0.05 0 1
Predicted Ozone Levels
0 15
Figure E-2. Comparisons of Observed Ozone Levels with Those Predicted from Two Statistical Models Using Data
from Philadelphia, PA (Model 1, R2=0.45; Model 2. R2=0.89)
to evaluate model simulated concentrations of ozone and its precursors is an important
incremental contribution to the overall model performance evaluation. The addition of PAMS
precursor measurements reduces the degrees of freedom in the model evaluation process and
(assuming acceptable model performance) increases the probability that the model is correctly
predicting surface ozone for the right reasons rather than as a result of compensating errors. In
addition, the upper air meteorological monitoring requirements of PAMS yield improvements in
the representativeness of simulated wind fields and mixing heights (as shown previously in Figure
E-l) both of which are critically important inputs to the EBMs. Finally, key use of the PAMS
speciated VOC data is the evaluation of emissions inventories (a significant component of EBMs).
Such evaluations with PAMS data provide insight into the number and mix of emissions sources
as well as potential gaps in emissions configurations. An example illustrating the value of PAMS-
like data in model evaluation is taken from a study performed in Texas. On August 19, 1993, a
highly localized ozone peak of 231 ppb was observed in Houston for which the modeling results
did not replicate the timing or the magnitude. A series of across-the-board emissions sensitivity
runs for VOC and oxides of nitrogen (NOJ failed to improve model performance. An analysis of
nearby VOC data indicated an anomalous peak in the total non-methane hydrocarbons (NMHC).
Through further analysis of the ambient speciation, the emissions inputs were adjusted to coincide
-------
EPA-454/R-96-006
Executive Summon
Revision Number: 0
Date: November 1996
Page- iv
with the ambient data. Subsequent model runs resulted in improved representation of the peak
ozone as shown in Figure E-3. A more detailed explanation of this example and a thorough
discussion of the use of PAMS data as a means of improving EBM simulations is provided in
Chapter 2.
i
8 100
Figure E-3. Time series plots of ozone at two sites before and after correction in emissions. Notice that simulated
ozone changed significantly at site TNIO. responding to emissions change, and little change occurred at sue TN2 (top).
EVALUATION OF EMISSIONS FACTORS, MODELS AND INVENTORIES USING
PAMS DATA
As mentioned previously, data from the PAMS program provide an important tool for
evaluating and refining estimates of ozone precursor emissions. The concept of using ambient
measurements to improve emissions models, factors and inventories is not new, having been first
used to evaluate particulate matter inventories in the 1970's. To date, several interpretive
techniques have been used to evaluate emissions inventories with the PAMS data. Types of
screening analyses include comparisons of ambient- and emissions-derived hydrocarbon / NOX
ratios, association of certain compounds with transport direction, time series analyses, and the
detection of chemical species associated with certain events or episodes. Multi-variate models
(e.g., factor analysis, Source Apportionment by Factors with Explicit Restrictions (SAFER),
Chemical Mass Balance (CMB)) can also be used to interpret the data. Such analyses and studies
-------
EPA-454/R-96-006
Executive Summan
Revision Number: 0
Date: November 1996
Page: \
can reveal missing sources and/or suggest improvements to the spatial or temporal resolution of
the emissions inventory. They can also determine the need for better emissions factors or activity
inputs to emissions models or factors.
i
For example, data collected during the 1991 Lake Michigan Ozone Study (LMOS) were
used to compare emissions inventory and ambient concentration ratios of non-methane organic
compound (NMOC), NOX and carbon monoxide (CO) for Chicago, Gary, and Milwaukee.
Comparisons of 7-9 a.m. ratios for two ozone episodes (June 25-28 and July 16-18) showed that
the ambient computed ratios were generally higher than the inventory ratios. The relative
individual NMOC species compositions of the ambient and inventory data were also examined.
As a result of LMOS, the Lake Michigan Air Directors Consortium (LADCO) reevaluated the
emissions inventory and made several significant changes to the point, area, and mobile source
figures. Speciation profiles and background assumptions were also revised. Tables E-l and E-2
below show the computed ambient and emission NMOC/NOX ratios both before and after the
LADCO inventory revision. Chapter 3 provides a more complete discussion of how PAMS data
can be used to help improve emissions estimates.
Table E-l.
Lake Michigan Area - Ambient Versus Original Set of Emissions Inventory NMOC/NO,
Ratios, 1991 (Korc, 1993)
Site
.Gary"1
Chicago
Milwaukee"
Ambient
NMOC/NO,
5.3
48
64
Emissions
NMOC/NO,
4.3
2.6
4.2
Ambient/El
1.2
1.9
1.6
Table E-2.
J Ambient NMOC/NO, ratios correspond to June 26, July 16 and 18, 1991
Lake Michigan Area - Ambient Versus Revised Set of Emissions Inventory NMOC/NO,
Ratios, 1991 (Korc. 1993)
Site
Gary1
Chicago
MilwaukeeJ
Ambient
NMOC/NO,"
4.8
4.7
6.4
Emissions
NMOC/NO,
5.0
3.6
3.8
Ambient/El
1.0
1.3
1.7
J Ambieni NMOC/NO, ratios correspond to June 26, July 16 and 18, 1991
Ambient NMOC/NO, with background correction
-------
EPA-454/R-96-006
Executive Summan
Revision Number: 0
Date: November 1996
Page: vi
OBSERVATIONAL-BASED METHODS FOR DETERMINING VOC/NOX
EFFECTIVENESS
Observational-based methods, which require ambient precursor measurements as inputs,
can provide directional guidance on the relative effectiveness of reducing NOX or VOC in
reducing ozone levels. As such, they serve as an important means of corroborating results
obtained from EBMs. The PAMS program provides ambient air quality data of greater spatial,
temporal and compositional detail than previously available and therefore improves the basis for
exercising observational methods. For example, the Georgia Institute of Technology (GIT)
observational-based model (OEM) requires detailed speciated VOC measurements of the type
provided by PAMS. Figure E-4 illustrates an application of the GIT-OBM for Atlanta covering
several monitoring locations. The model provides a relative assessment of the role of emissions
groups on ozone formation. Descriptions of the uses of PAMS data in several observation-based
methods are presented in Chapter 4.
0.5
«: 0.4
!
Of.
"3 0.3
1
£
V
£, 0.2
Multiple Day Analysis
JS
OJ
0.1
0.0
1
Mare Hill
GaTecfa
ML King
FortMcPh
Tucker
Dekalb
NO
AHC
NHC
CO
Figure E-4. Results from GIT-OBM applied to Atlanta. (Cardehno and
Chamiedes, 1995).
-------
EPA-454/R-96-006
Executive Summan
Revision Number: 0
Date: November 1996
Page: vii
LINKAGES AMONG TECHNIQUES
Although the examples are presented in this report as distinct entities to facilitate
unambiguous descriptions, the significance of the individual analysis areas to the ozone
management process is best understood when they are considered as elements of an integrated
whole. For example, the value of spatial, temporal and speciated displays of ozone and precursor
data is considerably greater when such displays are viewed as techniques to quality assure data
and provide direction for more refined analysis. Similarly, the significance of emissions inventor}'
evaluations is more substantial in the context of their use in improving (or corroborating) the
inputs to photochemical models. The results from the application of OBMs are far more
important when coupled with results from more traditional EBMs.
The step-wise air quality modeling process can be used as a framework to illustrate the
linkages among these analysis categories. The modeling process can be viewed as sequential
four-step process:
1. Selecting model days or episodes and specifying modeling domain;
2. Developing and evaluating meteorological and emissions inputs;
3. Testing and evaluating the model under present conditions; and
4. Applying the model in a control strategy context and evaluating/corroborating
performance.
Basic data characterization efforts for air quality and meteorology form the basis for
almost all model process elements. The selection of appropriate model days historically has been
based on a review of meteorology and ozone data. PAMS provides these elements, and adds
highly resolved upper air meteorological data and precursor data as additional factors in episode
selection. Transport and aging related analyses yield other considerations in characterizing the
days to be modeled. Perhaps the most important contribution of PAMS to the modeling process
is an improved ability to evaluate the basic emissions and meteorological inputs to drive the
simulations. The examples contained in this report describe the unique contributions of PAMS
data in resolving vertical mixing phenomena (mixing heights and winds) and providing a
continuous, speciated record of hydrocarbon compounds to check emissions estimates. The
spatial and temporal data characterizations are also applied in evaluating "current" day EBM
behavior. Finally, greater confidence in emissions-based models' predictive abilities (i.e., the
ability to correctly reflect ozone response to emissions changes) is gained when observational-
based models (which independently assess control strategy preference) produce agreement.
Combined with a longer term trends perspective, the PAMS data can be used to continuously
check the predicted response of the model leading to iterative analyses and refinements when
measurements diverge from original projections. This longer term approach speaks to the
greatest value of PAMS - a long-term record for ongoing evaluations of control programs.
-------
EPA-454/R-96-006
Introduction
Revision Number: 0
Date: November 1996
Page: 1
INTRODUCTION
This document is the first edition of the PAMS Data Analysis "Results" Report, a
summary and compilation of salient examples and illustrations of the uses of data from the
Photochemical Assessment Monitoring Stations (PAMS) program1. As such, the Report utilizes
examples of analyses available at the time of its development and contains no independent or
"new" analyses. However, this Report will be updated annually and will include the results of
future analyses performed by the Office of Air Quality Planning and Standards (OAQPS) and
others. The Report summarizes the current state of PAMS data analysis as a vehicle for
transferring the resultant techniques and insights, and encouraging a dialogue among analysts
using the data.
I.I PURPOSE
This report is intended to capture the best examples of the uses of PAMS data to
understand the tropospheric ozone issue and to motivate regulatory and control program activity.
Although the design is to show "what can be done" with PAMS data, the hope is that these
examples will also serve to catalyze a commitment to making full use of this rich data set, to
instruct those interested in analyzing the data and to, some degree, help guide those endeavors.
In no case, have all existing examples been used. Instead, those of greatest illustrative power or
unique character have been chosen for inclusion.
The primary audience for this Report is comprised of air quality data analysts charged with
and/or interested in using the PAMS data. Obviously, the document also addresses issues and
covers subjects which are of importance to ozone control program and regulatory staff, as well as
photochemical modelers. Finally, through this Report, policy and decision-makers grappling with
the complex issues of the tropospheric ozone will discover the applicability and relevancy of
PAMS data to this environmental challenge.
1.2 DOCUMENT ORGANIZATION
The report is organized into the following five chapters:
Characterization of ambient air quality data for ozone and its precursors;
PAMS data in support of ozone modeling applications;
Evaluating emissions factors, models and inventories with PAMS data;
1 40 CFR 58 Subpart E.
1-1
-------
EPA-454/R-96-006
Introduction
Revision Number: 0
Date: November 1996
Page: 2
Observational based methods for determining VOC/NOX effectiveness; and
Quality assurance.
The order of the subject matter implies neither a suggested sequence for the reader nor a required
process for analyzing data from the PAMS program. Each chapter is constructed as a"stand
alone" description of the associated subject and therefore can be read independently. An effort
has been made through citations and footnotes to "link" all examples which transcend the
individual chapters allowing the reader to reference previous discussions which dealt with the
example of interest. The figures and tables for each chapter have been appended to the end of the
respective text in the order presented in the material.
1.3 PAMS - A BRIEF DESCRIPTION
1.3.1 Why PAMS?
Of the six criteria pollutants, the most pervasive environmental problem continues to be
ozone. The most prevalent photochemical oxidant and an important contributor to "smog",
ozone is unique among the NAAQS pollutants in that it is not emitted directly into the air, but
instead results from complex chemical reactions in the atmosphere between volatile organic
compounds (VOCs) and nitrogen oxide (NOJ emissions in the presence of sunlight. Further,
there are literally thousands of sources of VOCs and NO, across the country. To track and
control ozone we need to understand not only the pollutant itself, but also the chemicals,
reactions, and conditions that go into forming it.
In 1991, the National Academy of Sciences (NAS) released a report entitled, Rethinking
the Ozone Problem in Urban and Regional Air Pollution, criticizing the EPA for failing to
establish monitoring networks to adequately track trends in ozone precursor emissions,
corroborate emission inventories, and support photochemical modeling. In accordance with the
"enhanced monitoring" provisions of Title I of the Clean Air Act Amendments of 1990, EPA
developed the PAMS program to address the concerns raised by the NAS2-3. The PAMS program
reflects the need to complement the Agency's historically based emissions modeling approach with
In addition to the NAS report, the PAMS program is part of the Agency's response to recommendations
contained in the 185b Report to Congress
Another current effort reflecting the Agency's response to the recommendations of the NAS is the North
American Research Strategy for Troposphenc Ozone (NARSTO), a field study and modeling research program with
which the PAMS program has close interaction.
1-2
-------
EPA-454/R-96-006
Introduction
Revision Number: 0
Date: November 1996
Page 3
ambient techniques, consistent with the basic tenets of the NAS report.
1.3.2 Regulatory Requirements
Section 182(c)(l) of the 1990 Clean Air Act Amendments called for improved monitoring
of ozone and its precursors, VOC and NOX, to obtain more comprehensive and representative
data on tropospheric ozone. Responding to this requirement, EPA promulgated regulations to
initiate the PAMS program in February 1993. The PAMS program requires the establishment of
enhanced monitoring networks in all ozone non attainment areas classified as serious, severe or
extreme. The 22 affected ozone areas, shown in Figure 1-1, cover 113 thousand square miles and
have a total population of 79 million people4.
Each PAMS network will consist of as many as five monitoring stations, depending on the
area's population5. Table 1-1 displays the ozone non-attainment areas required to implement the
PAMS program and the number of sites they are expected to implement. The PAMS stations will
be carefully located based on meteorology and other conditions at the site. Figure 1-2 presents a
schematic of a model network for a larger non-attainment area. Generally, each PAMS network
will consist of as many as four different monitoring sites (Types 1, 2, 3, and 4) designed to fulfill
unique data collection objectives. The Type 1 site is located upwind of the metropolitan area to
measure ozone and precursors being transported into the area. The second site, Type 2, is
referred to as the maximum precursor emissions impact site6. As the name implies, it is designed
to collect data on the type and magnitude of ozone precursor emissions emanating from the
metropolitan area and is typically located downwind of central business district. These sites
operate according to a more intensive monitoring schedule than other PAMS stations, are capable
of measuring a greater array of precursors and are suited for the evaluation of urban air toxics
also. The Type 3 stations are intended to measure maximum ozone concentrations, and are sited
downwind of the Type 2 sites and therefore of the urban area as well. The fourth PAMS site is
The text and map both reference the twenty-two areas originally required to participate in the PAMS
program However, Beaumont. Texas was reclassified to a moderate non attainment status effective June 1996 and
therefore no longer affected by PAMS requirements. Hence, there are actually twenty-one non attainment areas subject
to PAMS requirements currently
For more detailed descriptions of PAMS network requirements, see the PAMS Implementation Manual.
EPA -4 54/B -93-051, 1994.
fr A second type 2 site may be required in some PAMS areas and is positioned to capture the precursor
emissions in the second-most predominant morning wind direction. This additional type 2 site constitutes the fifth
PAMS site in the network
1-3
-------
EPA-454/R-96-006
Introduction
Revision Number: 0
Date: November 1996
Page: 4
located downwind of the non attainment area to assess the ozone and precursor levels exiting the
area and potentially contributing to the ozone problem in other areas.
States which experience significant impact from long-range transport of ozone or its
precursors, or are proximate to other nonattainment areas (even in other States) can collectively
submit a network description which contains alternative sites to those that would be required for
an isolated area as shown in Figure I-27. Such coordinated network plans should be based on the
example depicted in Figure 1-3, and must include a demonstration that the alternative design
satisfies the monitoring data uses and fulfills the PAMS objectives8.
1.3.3 Status
Over its first four years, the PAMS program has exhibited steady and successful growth.
Currently, there are approximately seventy PAMS surface air quality and meteorology monitoring
stations established and operating across the nation. This represents at least one monitoring
station in each of the twenty areas involved in the PAMS program9. Table 1-2 lists the established
and operating PAMS monitoring sites by non-attainment area and provides their AIRS site
identification numbers. Table 1-3 summarizes the minimum network requirements by non-
attainment area and sampling frequencies by PAMS site type.
The data collected at the PAMS sites includes measurements of ozone, oxides of nitrogen,
a target list of VOCs including several carbonyls (see Table 1-4) as well as surface and upper air
meteorology. Most PAMS sites measure 56 target hydrocarbons on an hourly basis during the
ozone season. Included in the monitored VOC species are nine compounds classified as
hazardous air pollutants (HAPs). The type 2 sites also collect data on carbonyl compounds every
three hours during the ozone monitoring period. All stations measure ozone, nitrogen oxides and
surface meteorological parameters on an hourly basis.
The PAMS networks produce a wealth of information invaluable to the development and
PAMS Implementation Manual pages 2-4 to 2-6.
Both California (South Coast Air Basin and Southeast Desert Modified AQMD non attainment areas) and
Lake Michigan (Chicago and Milwaukee non attainment areas) have adopted approved plans for "combined" networks.
Although there are twenty-two areas classified as senous, severe or extreme for ozone, the flexibility of the
PAMS program allowed areas (in close proximity to one another) in two regions consolidate their monitoring operations
I see footnote 8] and one original area has been reclassified [see footnote 4]. Therefore, only nineteen PAMS networks
exist
1-4
-------
EPA-454/R-96-006
Introduction
Revision Number: 0
Date: November 1996
Page: 5
evaluation of ozone control strategies and programs. In addition to providing a long term
perspective on changes in atmospheric concentrations in ozone and its precursors, the PAMS
program will specifically help to improve emission inventories, serve as input to photochemical
grid models, provide information to evaluate population exposure, and provide routine
measurements of selected HAPs. Most importantly, PAMS will assist in delivering a more
complete understanding of the complex problem of ozone, so that we may move toward the best
solution10.
1.4 FURTHER INFORMATION
1.4.1 Comments
Please forward your comments, suggestions, etc. to:
James Hemby
MD14
EMAD/OAQPS/U.S. EPA
79 T.W. Alexander Drive, Building 4201
Research Triangle Park, NC 27711
hemby.james@epamail.epa.gov
1.4.2 Request for Additional Copies
Additional copies of this report are available through the Emissions, Monitoring and
Assessment Division. Please contact Linda Ferrell at 919-541-5558 to request a copy.
For a more complete discussion of the intended uses of PAMS data, see Section I of the PAMS
Implementation Manual and its treatment of PAMS' data quality objectives.
1-5
-------
Figure 1-1.
PHOTOCHEMICAL ASSESSMENT
MONITORING STATIONS
AREAS SUBJECT TO
ENHANCED OZONE
MONITORING REQUIREMENTS
NUMBER OF AFFECTED AREAS
TOTAL = 22
-------
Figure 1-2. Basic PAMS Scheme
NETWORK DESIGN
("*)
EXTREME DOWNWMD
MAXIMUM OZONE
PRIMARY AFTERNOON
WMD
rmuan HOHMNO WIND
-------
Figure 1-3. Multi-Area and Transport Area Network Design
ITY Z
U2
-------
ble 1-1. PAMS Requirements for Currently Affected Areas
CURRENTLY-AFFECTED AREA NAME
Beaumont-Port Arthur, TX1
Portsmouth-Dover-Rochester, NH-ME
Southeast Desert Modified AQMA, CA
Baton Rouge, LA
El Paso, TX
Springfield, MA
Ventura County, CA
Milwaukee-Racine, WI
Providence-Pawtucket-Fall River, RI-MA
Sacramento. CA
Atlanta, GA
Baltimore, MD
Boston-Lawrence-Worcester, MA-NH
Chicago-Gary-Lake County (IL), IL-IN-W1
Greater Connecticut, CT
Houston-Gal veston-Brazoria, TX
Los Angeles-South Coast Air Basin. CA
New York-New Jersey-Long Island. NY-NJ-CT
Philadelphia-Wilmmgton-Trenton. PA-NJ-DE-MD
San Diego. CA
San Joaquin Valley, CA
Washington. DC-MD-VA
TotaK
POPULATION
RANGE
Less Than
500,000
500,000 to
1,000,000
1,000,000 to
2.000,000
More Than
2,000,000
CLASSIFICATION
OF
NONATTAINMENT
AREA
Serious
Senous
Severe
Serious
Serious
Serious
Severe
Severe
Serious
Serious
Serious
Severe
Serious
Severe
Serious
Severe
Extreme
Severe
Severe
Severe
Senous
Serious
22 Areas
MINIMUM
NUMBER OF
REQUIRED
SITES
-i
->
n
3
3
3
3
4
4
4
5
5
5
5
5
5
5
5
5
5
5
5
90
'Reclassified on 6/1/96 to moderate nonartamment status therefore not required to implement PAMS program.
-------
TABLE 1-2. PAMS MONITORING SITES
AREA
Boston
Connecticut
Portsmouth
Providence
Springfield
New York
Baltimore
Philadelphia
Washington
Atlanta
Lake Michigan
SITE
Borderland #1
Lynn- #2
Newbury - #3
Cape Eliza.. ME- #4
Sherwood Island #l/#3
E. Hartford #2
Stafford #3
Stratham #1
Kittery,ME#2
W. Greenwich #1
E. Prov #2
(Borderland, MA #3)
Agawam #1
Chicopee #2
Ware #3
Purchase #2
Botanical Gardens #2
(Sherwood Island. CT #3)
FortMeade,MD#l
Essex #2
Morgan State #2
Aldino #3
(Lums Pond. DE #4)
Clifton Pk. or Living Rm #2A
Lums Pond (DE) #1
East Lycoming Lab #2
Rider College (NJ) #3
Corbm. VA #1
(Caroline Co. Met Only #1)
McMillan Reservoir #2
(Fort Meade, MD #3)
(Lums Pond, DE #4)
Yorkville #1
So DeKalb #2
Tucker #2
Conyers #3
Braidwood #1
Milwaukee UWM #2
Chicago NWU #2
Chicago-Jardme #2
Gary, IN #2
Harrington Bch #3
Zion#4
Manitmvnr WT #4
AIRS
NUMBER
25-005-1005
25-009-2006
25-009-4004
23-005-2003
09-001-9003
09-003-1003
09-013-1001
33-xxx-xxxx
23-031-3002
09-001-0017
44-007-1010
25-005-1005
25-013-0003
25-013-0008
25-015-4002
36-119-5003
36-005-0083
09-001-9003
24-003-0019
24-005-3001
24-510-0050
24-025-9001
10-003-1007
xx-xxx-xxxx
10-003-1007
42-101-0004
34-021-0005
51-033-0001
51-033-0002
11-001-0043
24-003-0019
10-003-1007
13-223-0003
13-089-0002
13-223-0003
13-247-0001
17-197-1007
55-079-0041
17-031-0039
17-031-0072
18-089-1016
55-089-0009
17-097-1007
s's-rni-nnn?
-------
Houston
Baton Rouge
El Paso
South Coast/ SEDAB
San Diego
Ventura Co
Sacramento
San Joaquin
AREA
Galveston #1
Galleria#2
Clinton Dr #2
HRM No. Three #2
Deer Park #2
Aldine #3
Pride #l/#3
Capitol #2
Bayou Plaquemine #3/#l
N. Campbell #2
Chamizal #2
UTEP#3
Pico Rivera #2
Azusa #3
Banning #2
Upland #4/1
El Cajon #2
Overland #2A
Alpine #3
El Rio #2
Simi Valley #3
Del Paso #2
Folsom #3
Elk Grove-Bruceville
Golden St Ave #2 (Bkrsfld)
Clovis-Villa #2
Arvin #3
Parher #3
SITE
48-167-0014
48-201-0067
48-201-1035
48-201-0803
48-201-1003
48-201-0024
22-033-0008
22-033-0009
22-047-0009
48-141-0027
48-141-0044
48-141-0037
06-037-1601
06-037-0002
06-065-0002
06-071-1004
06-073-0003
06-073-0006
06-073-1006
06-111-3001
06-111-2002
06-067-0006
06-067-1001
xx-xxx-xxxx
06-029-0010
06-019-5001
06-029-5001
06-019-4001
AIRS
NUMBER
-------
Table 1-3.
PAMS MINIMUM NETWORK REQUIREMENTS
MINIMUM NETWORK REQUIREMENTS
POPULATION OF MSA/CMSA
LESS THAN
500,000
TO
1,000,000
1,000,000
TO
2,000,000
GREATER
THAN
2.000.000
FREQ
TYPE
SITE LOCATION
B/E!
Aorc
AWC
B/E
B/E
A«C
AWC
(2)
(3)
(2)
(2)
(2)
(2)
(3)
(4)
VOC SAMPLING FREQUENCY REQUIREMENTS
Freq
A
B
£
Requirement
8 3-Hour Samples Every Third Day
1 24-Hour Sample Every Sixth Day
g 3-Hour Samples Everyday
1 24-Hour Sample Every Sixth Day (year-round)
8 3-Hr Samp 5 Hi-Event/Prev Days/Every 6th Day
I 24-Hour Sample Every Sixth Day
IcARBi
IONYL SAMPLING FREQUENCY REQUIREMENTS
Freq
D
E
F
Requirement
8 3-Hour Samples Every Third Day
8 3-Hour Samples Everyday
8 3-Hr Samp 5 Hi-Event/Prrv Days/Every 6th Day
MINIMUM PHASE-IN
YEARS AFTER
PROMULGATION
}
NUMBER OF
SOBS OPERAITNG
1
2 2
3
3
OPERATING
STTE LOCATION
RECOMMENDATION
2
2.3
1.2.3
4 4 1.2.3.4
5 ' 5 : 1.2.2.3.4
-------
Table 1-4. PAMS Target Species
VOC COMPOUNDS
Ethylene
Acetylene
Ethane
Propylene
Propane
Isobutane
1-Butene
n-Butane
trans-2-Butene
cis-2-Butene
Isopentane
1-Pentene
n-Pentane
Isoprene
trans-2-Pentene
cis-2-Pentene
2,2-Dimethylbutane
Cyclopentane
Total NMOC
2,3-Dimethyl butane
2-Methylpentane
3-Methylpentane
2-MethyM-Pentene
n-Hexane
Methylcyclopentane
2,4-Dimethylpentane
Benzene
Cyclohexane
2-Methylhexane
2,3-Dimethylpentane
3-Methylhexane
2,2,4-Trimethylpentane
n-Heptane
Methylcyclohexane
2,3,4-Trimethylpentane
Toluene
2-Methylheptane
3-Methylneptane
n-Octane
Etnylbenzene
m/p-Xylene
Styrene
o-Xylene
n-Nonane
Isopropyl benzene
n-Propylbenzene
m-Ethyltoluene
p-Ethyttoluene
1,3,5-Trimethylbenzene
o-Ethyltoluene
1,2,4-Tr imethylbenzene
n-Decane
1,2,3-Trimethylbenzene
m-Diethyl benzene
p-Diethylbenzene
n-Undecane
CARBONYL COMPOUNDS
Acetaldehyde
Acetone
Formaldehyde
Hazardous Air Pollutants (HAPs)
-------
EPA-454/R-96-006
Chapter 1
Revision Number: 0
Date: November 1996
Page 1
CHAPTER 1
CHARACTERIZATION OF AMBIENT AIR QUALITY DATA FOR
OZONE AND ITS PRECURSORS
1.1 INTRODUCTION
Characterization of ozone episodes is extremely important. A key consideration in
devising an effective control strategy is to ensure that the strategy works under a variety of
conditions which have been observed to correspond with high measured ozone. The relative
importance of ozone/precursors transported into an area is likely to be one important
distinguishing characteristic of episodes. Thus, the needs to accurately characterize episodes and
identify the potential role of transport are closely related. In a modeling analysis, performed to
assess adequacy of proposed control strategies, it is important to have chosen a limited number of
episodes which are representative of differing conditions leading to high observed ozone. More
generally, a knowledge of meteorological, precursor characteristics and other ambient conditions
which correspond with high ozone is useful in helping to formulate policies which are likely to
lead to improvements in measured ozone levels. In this chapter, we present examples illustrating
the added value of PAMS data in characterizing ozone episodes and identifying features that may
be linked to significant pollutant transport.
Both Sections 1.2 and 1.3 focus on how PAMS data can be used to enhance the
characterization of local ozone episodic events. The examples presented illustrate the potential
uses for and the value-added of both the meteorological and ambient air quality data collected at
PAMS sites. The discussions include the extent to which the PAMS data may be useful in
assessing the potential role of transport in different episodes.
1.2 EPISODE CHARACTERIZATION USING METEOROLOGICAL
MEASUREMENTS
Ozone conducive meteorological conditions such as high insolation, high temperature,
high stability (as often reflected by low mixing heights), low winds, and low midday relative
hurrudity have been identified by various researchers in the past (Bruntz et al., 1974; Lamb et al.,
1987; Chu, 1987; Chu and Doll, 1991; Cox and Chu, 1993; Robinson, 1952; Hosier, 1961;
Ludwig et al., 1977; Pagnotti, 1990). These ozone conducive meteorological conditions can thus
be used to characterize the local ozone episodes.
-------
EPA-454/R-96-006
Chapter 1
Revision Number: 0
Date: November 1996
Page: 2
1.2.1 Characterizing Episode "Severity'
EPA has developed relatively simple regression based models to estimate the severity of
daily ozone episodes using meteorological data from the National Weather Service. Presently,
this methodology considers only surface meteorological variables to characterize episode severity.,
Since PAMS provides on-site surface as well as upper air meteorological data, the potential for
more accurate classification and characterization of distinct episode types (and eventually
severity) exists. The routine upper air (rawinsonde) data collected by the National Weather
Service (NWS) network are intended to resolve synoptic scale weather systems which have a
length scale of about 2000 - 4000 km. Air quality episodes, however, are usually observed in a
smaller domain of 1000 km or less in which meso-scale dynamics play a significant role. Thus,
the region wide PAMS upper air measurements (in particular, the data collected by RASS and
radar profilers) provide valuable information on local mixing height changes, as well as detecting
nocturnal jets and topographically induced meso-scale circulations. These, in turn, will help better
describe the local characteristics of the episodes and provide inference of possible intra- and/or
inter-regional pollutant transport.
Cox and Chu (1993; 1996) have developed a statistical model to predict ozone producing
potential using local surface and upper air meteorological data. The model has been applied to
minimize meteorological influences on ozone trend analysis, to rank the local severity of ozone
episodes, and to select episodic days for modeling and control strategy designs. The PAMS
hourly surface and upper meteorological air data (e.g., winds, temperature, humidity, and mixing
height data) will certainly increase the power of these models to better define the characteristics of
local episodic ozone events as should PAMS air quality data described in Section 1.3. PAMS
data, particularly from the RASS and radar profiler sites, will provide better estimates of mixing
heights as well as other meteorological measurements aloft. These variables could improve the
skill of the statistical model in ranking meteorological ozone forming potential.
1.2.2 Determining Source/receptor Orientations Corresponding to Ozone Conducive
Conditions
Chu (1995) has shown that the frequency distribution of local predominant wind directions
(PWD) on high ozone days is useful in describing source/receptor orientations which can be
predicted by a set of ozone conducive meteorological variables: daily maximum temperature,
morning (7-10 a.m.) average wind speed, afternoon (1-4 p.m.) average wind speed, and midday
(10 a.m. - 4 p.m.) average relative humidity. By better characterizing local mixing heights, PAMS
data help better define PWDs with accompanying high ozone forming potential. The frequency
distribution of the PWDs (including near calm conditions) on high ozone days may help to identify
which type of ozone episodes (i.e., stagnation or transport or a mixture) is most often observed
(i.e., representative) locally. Further insight into potentially significant pollutant transport may be
-------
EPA-454/R-96-006
Chapter 1
Revision Number: 0
Date: November 1996
Page: 3
possible from flux analysis using PAMS upper air wind measurements (e.g, leading to detection of
a nocturnal jet, such as that shown in Figure 1-1) (Lindsey, 1995). Trajectory analyses, like that
shown in Figure 1-2, are useful in classifying episodes according to source/receptor orientation
(Lindsey, 1995).
1.2.3 Identifying Critical Circulations Associated with High Ozone Events
Very high local ozone concentrations observed in episodes are often heavily influenced by
certain meso-scale circulations embedded in large, stagnant synoptic systems. Most of these
meso-scale systems are topographically induced, as in sea/lake breeze circulations and mountain-
valley flows. Under a stagnant, synoptic-scale high pressure system, these localized meso-scale
systems become a major mechanism for mixing, dispersing, and transporting pollutants. For
example, in coastal areas, observations often suggest that local high ozone concentrations
generally coincide with the sea breeze convergent zone. Although a single PAMS upper air
monitor (either a RASS or a radar profiler) may not be sufficient to resolve these meso-scale
circulations, it could still provide some valuable information in the integrated analysis of the
episode with surface meteorological and air quality data. An example of using the PAMS-like
data in an integrated analysis of the influence of sea breeze recirculation on Houston high ozone
events is illustrated in Figures 1-3 to 1-6 (Systems Applications International).
1.2.4 Identifying Boundary Layer Structures Associated with High Ozone Events
Since meteorological conditions are quite distinctive on high ozone days, it is not
surprising that the structure of the atmospheric boundary layer in which the pollutants are mixing,
reacting, and dispersing would be quite different from those on non-episodic days. The boundary
layer profile of temperatures, winds, humidity, and mixing heights would have a direct impact on
surface ozone concentrations. Due to the advanced technical capability of the instrumentation
employed, the hourly winds, temperature, humidity, and reflectivity (Cn:) data collected by the
PAMS RASS and/or radar profilers add more detailed local information to the routine (and often
remote) NWS rawinsonde observations and thus increase our understanding of the boundary layer
structure on high ozone days. Figure 1-7 is an example showing the diurnal mixing height change
derived from Cn2 (Dye et al., 1995).
1.3 EPISODE CHARACTERIZATION USING AIR QUALITY MEASUREMENTS
PAMS data play a critical role in characterizing air quality episodes used in photochemical
modeling and in the design of cost effective control measures. Since NAMS and SLAMS were
originally deployed, monitor siting technology has improved considerably such that newly located
PAMS stations are meeting various objectives (e.g. maximum ozone concentration levels) with
greater assurance and accuracy than previously possible. More importantly, precursor data allows
-------
EPA-454/R-96-006
Chapter !
Revision Number: 0
Date: November 1996
Page: 4
characterization of episodic events by species mix which may be used to assess the likely impact
of particular strategies on ozone levels. Moreover, measured species data coupled with on-site
meteorological data will lead to improved statistical models used to characterize ozone formation
potential and relative frequency of occurrence for various episode types. The utility of precursor
measurements to help characterize episode severity is expected to increase as the period of record
for the PAMS measurements increases. This follows since the number of episodes which have
precursor data and which are candidates for modeling or policy analysis will increase with passing
time.
1.3.1 Indicators of Ozone Episodes
Ozone levels during the summer of 1995 in much of the US are reported to be somewhat
higher than found in previous years. EPA conducted a short intensive study of the relative
severity of ozone levels in 1995 compared with those in previous years using statistical methods
that factored in measured meteorological conditions and urban specific ozone trends over the past
decade. Based on the preliminary analysis, EPA concluded that while the summer of 1995 was
unusually warm, overall conditions for elevated ozone were not atypical compared with previous
years having high ozone levels. While most of the ozone measurements used in this study were
from NAMS/SLAMS sites, a significant number of sites (- 50) were operating PAMS ozone
sites. One of the findings from the study was the preponderance of PAMS ozone sites (47 of 50)
that reported at least one exceedance of 120 ppb during 1993-1995 suggests that PAMS sites are
well placed for detecting peak ozone levels.
1.3.2 Distinguishing among Episode Types
Relative importance of transport is an important means for distinguishing among ozone
episodes. Some types of ozone episodes produce a distinctive trend in diurnal patterns that may
he suggestive of transport conditions. An example of such an episode for July 20-22 of 1994 in
which a smooth northeasterly time progression of the ozone peak was observed over sites in New
England from Lynn, MA through Jonesport ME is displayed in Figure 1-8 (NESCAUM, 1995).
In a companion plot (Figure 1-9), 8-hour moving average ozone showed a clear broadening of the
plume over time (NESCAUM, 1995).
Although the maximum 8-hour concentrations declined with distance, the broadening of
the plume caused exposures above 70 ppb over a larger portion of the day. Since PAMS sites are
located at upwind and downwind locations, this phenomenon can be more fully investigated in
PAMS urban areas. For example, plots of the observed diurnal pattern for ozone, NO, NO2 and
VOC species should reveal general decay in primary species coupled with an increase in
secondary species if significant transport is occurring.
-------
EPA-454/R-96-006
Chapter 1
Revision Number: 0
Date: November 1996
Page: 5
1.3.3 Characterizing Precursor Species During Episodes
A major strength of the PAMS program is the requirement for speciated hydrocarbon
measurements to be made on a pontinuous (hourly or 3-hour) basis for up to 55 targeted
hydrocarbon (and 3 carbonyl) compounds. Utilizing this attribute, episodes can be better
characterized through analysis of the abundance of precursors experienced during periods of
elevated ozone levels. An analysis of the 1994 data from the Northeast by the NESCAUM
Ambient Monitoring and Assessment Committee is used to illustrate such assessments below
(NESCAUM, 1995).
For the ozone episodes 07/06/94-07/08/94 and 07/20/94-07/22/94, the abundance of
targeted VOCs from the five PAMS sites in the Northeast is shown in Figure 1-10 (NESCAUM,
1995). The study revealed that the most prevalent species are remarkably consistent across the
region for the time period of interest. Seven compounds (i.e., isopentane, toluene, propane,
ethane, n-butane, m&p-xylene and n-pentane) were found to be among the ten most abundant
species at all sites except the Cape Elizabeth site. Even for this site, five of the seven species were
among the ten most abundant. Interestingly, the biogenic isoprene was found at significant
concentrations at four of the sites'. The report concluded that the pattern of similar abundances
throughout the urbanized portion of the region (and to some extent, the remote portions such as
Cape Elizabeth) was perhaps to be expected given the ubiquity of mobile sources.
The results above were further compared to data for five Northeastern cities from the 6:00
to 9:00 AM time period from a previous study (Wixtrom, R.N., et al.). As shown in Tablel-1, six
of the seven compounds of greatest abundance for the 1994 PAMS data were also the most
prevalent for these five cities (NESCAUM, 1995). An extensive study of speciated VOCs in the
Los Angeles area (Lurmann, F.W., et al.) found "the same seven anthropogenic compounds to be
most abundant and, with the exception of propane, in virtually the same rank of occurrence."
Volatile organic compounds react at different rates and with different reaction mechanisms
due to their variations in their chemical composition and structure. As a result, VOCs differ
significantly in their potential to form ozone. The use of incremental reactivities of VOCs
provides a way to avoid an oversimplification of treating abundance estimates of all VOCs as
equivalent. Incremental reactivity allows analysis of the effect of changing the
concentration/abundance of a VOC on ozone formation. In the NESCAUM analysis, the
maximum incremental reactivity (MIR) scale developed by Carter was used to show the relative
ozone forming potential of the various VOC and carbonyl species (Carter, 1994). The hourly
average abundances (calculated from the PAMS data) were scaled by the MIRs and displayed as
'At the East Hartford, CT site the isoprene concentrations were thought to be underestimated and compound
misidentification is suspected of being reason for the lower abundance.
-------
EPA-454/R-96-006
Chapter I
Revision Number: 0
Date: November 1996
Page. 6
(potential) ppbv of ozone. The results are displayed graphically in Figure 1-11 (NESCAUM,
1995). The report noted that, although acetone and isopentane were found to be quite abundant.
"their low reactivities result in rather low ozone forming potential". In addition, the increase in
the relative significance of formaldehyde following the application of the MIR is highlighted.
Although these data are preliminary and cover a short time span, the study concluded that the
results were expected to be "a valid snapshot of upper limit conditions that typically exist in the
Northeast" during periods of high ozone.
1.3.4 Assessing Airmass Aging Using PAMS Precursor Data
The same analysis of 1994 PAMS data from the Northeast by the NESCAUM Committee
effectively used a comparison of ratios of VOC species to illustrate the effects of airmass aging.
The series of graphics, Figures 1-12 through 1-17, from this study are used to describe the results
(NESCAUM, 1995). Figure 1-12 displays the estimated benzene/toluene (B/T) and
xylene/toluene (X/T) ratios based on 1990 Atlanta source profiles and source mix. Figure 1-13
shows the measured hourly ratios of benzene and m/p-xylene to toluene from the urban (type 2)
PAMS site in E. Hartford, CT during the July, 1994 episode periods.
The scatter plot in figure shows that toluene levels at E. Hartford were highly correlated
with both benzene and m/p-xylenes. This is consistent with a hypothesis that all three are
primarily emitted by mobile sources. The B/T and X/T ratios at E. Hartford are also consistent
with the predicted ratios displayed in Figure 1-13 derived from the Atlanta source profile data -
suggesting that the E. Hartford source profiles and source mix are consistent with those in Atlanta
(NESCAUM, 1995). Figure 1-14 shows the hypothetical effect of aging on Atlanta B/T and X/T
ratios, while Figure 1-15 shows the measured B/T and X/T ratios for the rural type 3 PAMS site
in Stafford, CT site (downwind of E. Hartford) during the July, 1994 episodes (NESCAUM).
The scatter plot in Figure 1-15 shows that the B/T ratio at the downwind, rural Stafford
site has increased and the X/T ratio has decreased in comparison to the urban E. Hartford site
shown previously in Figure 1-13 (NESCAUM). This is consistent with the predicted effect of
airmass aging, as the more reactive species are differentially removed during transport. The points
plotted in Figure 1-15 also exhibit greater scatter (B/T and X/T correlations are poorer) than
Figure 1-13. While common (motor vehicle-related) sources are still anticipated to be a
predominant cause of benzene, toluene and m/p-xylenes at the Stafford site, the species inter-
correlations are diminished during transport, as the degree of aging depends on variable factors
such as wind speed, wind direction, solar radiation, NOX, etc.
Figures 1-16 and 1-17 show the relationships between toluene and m/p-xylenes at E.
Hartford and Stafford, with different symbols to distinguish between daytime and nighttime
samples (NESCAUM). At the urban E. Hartford site, there is relatively little difference in the X/T
-------
EPA-454/R-96-006
Chapter 1
Revision Number. 0
Date: November 1996
Page: 7
ratios between nighttime samples (when reactivity is minimal) and daytime samples (when
reactivity is maximal). This is consistent with a predominant, continuous influence of fresh, local,
motor vehicle-related emissions at this site.
At the rural Stafford site (Figure 1-17), there's a more distinct difference between the
daytime and nighttime X/T ratios. Nighttime ratios show a stronger correlation, and a slope
similar to the predicted value of 0.6 for fresh emissions (and East Hartford's). For daytime
samples at Stafford, there's a clear downward shift in the X/T slope (and a much poorer X/T
correlation). This is consistent with a predominant influence of transported, motor vehicle-related
emissions, which are photochemically aged (in a highly variable way) during the day, but which
remain relatively unaged in the absence of sunlight.
1.3.5 Temporal Variation in PAMS Precursor Data
The typical daily patterns of hourly VOC concentration data and the comparison of these
patterns with those of other PAMS areas, with national averages, and with historical data from
this area are useful analyses to undertake to determine the unique source contributions from a
particular area and changes in the composition of the urban area's ambient air quality. Average
diurnal patterns or profiles are calculated by computing the average of all samples collected
during each hour of the day. Most PAMS VOC species exhibit well defined diurnal cycles (or
average values over time) which reflect source activity (e.g., traffic patterns), familiar daily
meteorological patterns, and photochemical activity.
The first example result plotted in Figure 1-18 displays the average diurnal profiles of m,p-
xylene and isoprene for six Northeastern PAMS sites for two ozone episodes in July 1994
(NESCAUM, 1995). The diurnal pattern for m/p xylene is similar to a number of other reactive,
anthropogenic VOCs (toluene, o-xylene, isopentane, etc.), with emissions generally dominated by
automotive-related sources. Isoprene is emitted predominantly by deciduous vegetation, as a
function of solar radiation and temperature. As a biogenic compound, isoprene has a unique
diurnal pattern which is distinct from that of the anthropogenic VOC species. While the morning
(6'00 to 9:00 AM) levels and reactivities of m/p xylene and isoprene are quite similar, the reactive
anthropogenic pollutants are generally depleted rapidly during the day. Although isoprene also
reacts rapidly, its rate of production exceeds its rate of destruction during mid-day.
The next example result. Figure 1-19, introduces a variation on the basic diurnal pattern
technique depicted above by using box plots for each hour rather than displaying simple means.
Note that each panel contains 24 box plots corresponding to each hour of the day. The figure
contains diurnal patterns for 1993 PAMS data on acetylene, olefins, toluene, ethylene, xylene and
isoprene from Baltimore's Site #2 (Cox, 1995). The organic species for this analysis clearly
indicate the typical diurnal trends for anthropogenic VOCs described above. Median values for all
-------
EPA-454/R-96-006
Chapter 1
Revision Number: 0
Date: November 1996
Page: 8
species, except isoprene, show a tendency for higher morning and evening concentrations. These
patterns are interpreted as follows:
the morning maximum is associated with high emissions and limited mixing;
the mid-day minimum is associated with decreased mobile source emissions, increased
mixing due to rapid growth of the daytime boundary layer, and increased reaction rates
due to higher temperatures and maximum solar radiation;
the early evening maximum is associated with gradual build up of emissions in the
boundary layer over the course of the day, increased mobile source emissions during the
afternoon culminating in an early evening commute traffic peak, and decreased mixing.
The diurnal profile for isoprene reflects the fact that biogenic emissions are a strong function of
temperature and solar radiation and are short-lived in the atmosphere.
The final example results for this section on diurnal patterns are from the analysis of 1993
PAMS data from Houston, Texas (Stoeckenius, November 1994). Figures l-20a through l-20c
present separate calculations of the diurnal profiles of species concentrations for weekdays and
weekends; an effort to identify the impact of differences in emissions between these two groups of
days2. Comparisons of profiles between species and between weekdays and weekends can reveal
much about underlying meteorological, chemical, and emission factors. The vertical bars in these
figures indicate the 95 percent confidence limits for the mean value in each hour. Thus, pairs of
hours with non-overlapping bars have significantly different mean concentrations.
At the Galleria site, weekday concentration profiles for all species except isoprene exhibit
a strong morning peak. Morning peaks of acetylene, ethylene, toluene, olefins, and xylenes are
noticeably lower on weekend mornings; these differences are consistent with reduced morning
mobile source emissions on weekends. The lowest concentrations occur between 9:00 AM and
1:00 PM for all species (except isoprene) and the levels rise again to a second peak by early
evening. These patterns are similar to those found in other analyses by other investigators.
"Monday and Friday holidays were grouped in with weekends. Holidays falling in the middle of the week were
lefi out of this analysis since traffic patterns and business activities on these days tend to differ from regular weekdays or
weekends. Mean values were not computed for hours for which data was not available for at least two-thirds of
sampling days. This restriction primarily affected the hours around midnight (hours 11,0, and 1) which were sampled
less frequently than other hours. A cubic spline fit was used to generate the smooth curves in these figures; the dots
along the curves represent the actual hourly averages and are plotted at a position corresponding to the mid-point of the
hour they correspond to. This plotting technique accentuates the diurnal patterns making graphs that are easy to read but
one must be careful not to over interpret the peaks and troughs that may be shown as occurring between the actual
hourly average values.
-------
EPA-454/R-96-006
Chapter I
Revision Number: 0
Date: November 1996
Page: 9
Diurnal profiles (based on weight percents) for the same VOC species above are presented
in Figures l-21a through l-21c (Stoeckenius, November 1994). For the most part, the profiles
are similar to the concentration profiles in Figures l-20a through l-20c. Some differences are
evident, however. Toluene, olefins, and xylene weight percents exhibit smaller diurnal variations
than their concentration counterparts although the pattern is basically the same.
1.3.6 Statistical Models of Relationship Between Ozone and Precursors
In the context of the PAMS program, the use of regression analysis involves developing
empirical models to statistically describe the relationship (or potential relationship) between
independent parameters (ozone precursors and meteorological data) and the dependent parameter
(ozone). This analytic technique takes advantage of one of the unique characteristics of the
PAMS data: the concurrent measurement of VOCs, meteorological parameters, nitrogen species
and ozone. The results of these analyses (explanation in the variability of daily ozone maxima
based on meteorological and precursor data) can lead to invaluable insights about the design of
effective control strategies for the area analyzed.
However, this analytic technique and the example results which follow should only be
considered as "exploratory" given the early stages of the PAMS program and the remaining
further investigations that are required to verify and understand these initial evaluations. The
reader is cautioned not to make inferences as to the nature of cause-effect relationships between
the VOC species included in the regression models and downwind ozone (i.e., it is premature to
predict or conclude how ozone levels would behave if concentrations of the included VOCs were
reduced) based on these analytic results. These analyses should be viewed as another means to
begin using PAMS data to develop a more complete understanding of the ozone phenomenon in
PAMS cities. Finally, it is important to remember that these models have been constructed (to
date) without selecting variables for inclusion in the model based on their hypothesized physical
relationship to ozone.
The examples presented below demonstrate the technique using the 1994 PAMS data
from the Philadelphia, PA\ The daily maximum ozone concentration, the dependent variable, was
calculated from the downwind ozone data (Site#3). Data for the independent variables (i.e.,
VOCs, nitrogen species, meteorological parameters) were taken from the upwind urban site (Site
#2). Several indicators for the independent variables were calculated (averages and maxima for
selected intervals: three, six, twelve and twenty-four hour periods during the day of interest or
lagged for earlier days). The relationships between all of these indicators and the dependent
variable were then analyzed and those with the strongest correlations were then selected for
" The ozone precursor and meteorological data included in the model are from the PAMS Site #2, and the
downwind ozone data are from the PAMS Site #3.
-------
EPA-454/R-96-006
Chapter 1
Revision Number: 0
Date: November 1996
Page: 10
inclusion in the regression model. Figure 1-22 uses a scatter plot to display how well the initial
model, Model 1, estimated the observed daily maximum ozone values. This model utilized only
meteorological parameters: temperature (mean for interval 12:00 PM to midnight) and wind
speed (maxima for interval 6:00 to 9:00 AM). The R-square value for this model is 0.45
reflecting that the model using only meteorological parameters explains approximately 45% of the
variability of downwind ozone levels.
-------
EPA-454/R-96-006
Chapter 1
Revision Number: 0
Date: November 1996
Page: 11
The results presented in Figure 1-23 reveal an improvement in the model's "goodness of
fit" (R-square) with the inclusion of VOCs. The following hydrocarbon species have been
included with the earlier meteorological data in Model 24:
VQC Statistic/Interval
Isopentane Mean/8 hr lag
3-Methyl pentane Mean/9-12
n-Hexane Mean/8 hr lag
n-Octane Mean/19-22
Nitric Acid Max/13-15
2,2-Dimethyl butane Mean/6-9
Nitrogen Dioxide Max/1 -12
2-Methyl heptane Mean/6-9
Ethyl benzene Max/8 hr lag
Propane Mean/8 hr lag
The resulting R-square value (0.84) is an improvement over that of the meteorology-only model.
This example result suggests that the measurement of speciated hydrocarbons is essential for
accurately representing and understanding the character of maximum ozone levels.
Neural networks provide a more flexible alternative to standard regression methods for
relating dependent variables to a set of independent variables. Because neural networks are more
general, they can accommodate both non-linearities and interactions among independent variables
without explicit parameterizations required in non-linear regression models. While neural
networks offer the potential for better prediction of the response variable, results are often
difficult to interpret, mainly due to model complexity and multicollinearity of the process.
Weights (regression coefficients) are typically unstable, usually "inflated" and vary considerably
from true optima if local optima are found m the fitting process. Another major drawback is that
extrapolation is risky: producing significant errors under certain conditions.
Crowe and coworkers have applied neural networks using data collected in the southeast
Texas region at the Galleria Site near Houston. The data consisted of hourly meteorological data
(net radiation, temperature, wind direction and speed, wind variation), nitrogen oxides (NO2 and
NOX) and seven hydrocarbon species based on carbon bond 4 chemistry. Hourly ozone data were
taken from a downwind site located at the Clinton site near east Houston. Three neural network
An additional model (not presented in this report) for the 1994 Philadelphia PAMS data was constructed and
is available upon request. The model incorporated a number of additional VOC species to those already utilized in
Models 1 and 2.
-------
EPA-454/R-96-006
Chapter 1
Revision Number: 0
Date: November 1996
Page: 12
models were developed using proprietary software (Process Insight) developed by Pavilion
Technology Inc. The predictor variables for the first model (MET) consisted of the five
meteorological parameters for the same hour as the ozone measurement and also for 6 time delays
to account for possible effects of transport and chemical reactions. The second model
(METNOX) included the same meteorological variables but added the species NO and NOX. The
third model (METNOXCB4) dropped the time lagged variables (apparently unimportant) but
added seven hydrocarbon species based the carbon bond 4 chemistry.
The models showed progressively better predictive capability (using the entire data set) as
evidenced by increasing R2 values from 0.70 (MET), 0.80 (METNOX) and 0.91 for
METNOXCB4). The authors reported that selected hydrocarbon species are more sensitive
predictors of hourly ozoneincreasing olefins associated with decreasing ozone and increasing
paraffins associated with sharply increasing ozone levels. The authors view this work as very
preliminary and have neither attempted to associate any cause and effect to such relationships nor
attributed any physical and/or chemical significance to their findings to date.
Capone applied neural network technology to predict down wind hourly ozone data in the
Baton Rouge area using a more complicated network in which data from two downwind sites
were used as predictors. The model consisted of hourly meteorological and NOX measurements
(NO2, NO and NOX) at each site and was successful at predicting hourly ozone patterns as
evidenced by graphics comparisons of the diurnal pattern and scatter plots between measure and
predicted values. Interestingly Capone's model did not involve any hydrocarbon species,
apparently due to a lack of relatively complete measurements in this area.
Meteorological and PAMS data for 1994 were used in an exploratory application of
neural networks to predict daily maximum 1-hour ozone levels in the Philadelphia CMS A. The
PAMS data were taken from an upwind Delaware site (10-003-1007) where hourly measurements
are available from June-August. PAMS data included in the analysis were average TNMOC (6-9
am), olefins (6-9 am), NO (6-9 am) and NO2 (6-9 am) and mid-day maximum hour isoprene.
Daily composite meteorological variables were computed using available data from the nearest
National Weather Service station. These variables included maximum surface temperature,
morning and afternoon average wind speed, and mid-day average relative humidity and cloud
cover. Out of 92 candidate days, only 51 days had sufficient data to be included in this analysis.
Using commercially available software, a neural network was used to relate maximum
daily ozone to the combination of five meteorological and five PAMS precursor variables. For
comparison purposes, neural networks of size 1 (logistic regression11 weights) through size 3 (a
total of 37 "weights") were fit using all 10 predictor variables. Figure 1-24 is a scatter plot of the
log of the observed vs predicted ozone using for the smallest sized network. The fit is relatively
good (R2 =0.68) which is not surprising since the five meteorological variables alone are known to
-------
EPA-454/R-96-006
Chapter 1
Revision Number: 0
Date: November 1996
Page: 13
-}
be good predictors of ozone forming potential in many U.S. urban areas. The R~ using just the
five meteorologically variables is nearly as good (0.64) suggesting that the precursor species
provides only a marginal improvement in the fit.
The coefficients (weights) of the independent variables from this single node neural
network may be interpreted in a similar manner as with ordinary regression models. For example,
the coefficients for temperature are positive suggesting that increases in temperature are
associated with higher ozone. Conversely, the coefficients for wind speed and cloud cover are
negative suggesting that increases in winds and cloud cover are associated with lower ozone
levels. Of the species variables, olefms and mid-day isoprene were negatively related to ozone
although the magnitude of the coefficients were small relative to their standard errors.
Figure 1-25 show a similar plot of fitted ozone from a neural network of size=3.
Although the R2 statistic has increased considerably (0.89) the number of fitting parameters is
large relative to the number of observations. Also, the relative complexity of the model makes
interpretation of the coefficients difficult. For example, coefficients for temperature are both
positive and negative in the three linear inputs feeding the hidden layer. Although the net effect of
temperature is positive, it is difficult to interpret the relative role that temperature (and other
variables) have within each layer. As more PAMS data becomes available, the stability of larger
sized neural networks can be better assessed.
Clearly, the process of model building is very much an art at this point and will require
close coordination among dispersion modelers, atmospheric scientists, and data analysts. Neural
networks may help establish a practical upper bound on the predictive ability of statistical models
and provide insight into reasonable model structures that may be more interpretable. Hopefully,
the process will lead to development of physically meaningful input parameters that will help
simplify the structure of these models and lend credibility to their potential applications. For
example, better predictive models should provide better selection of episodes for modeling and
more accurate assessment of the seventy of episodes used in model based attainment
demonstrations. Also, properly structured empirical models may provide supportive information
regarding the potential effectiveness of emission control strategies including the relative benefits
of VOC/NOX emission reductions on ozone levels.
Crowe and coworkers applied neural network methods to predict hourly ozone levels^
using PAMS like data taken at Texas air monitoring sites during 1990-1994. They report R
statistics on the order of 0.75 when meteorological data alone are used injhe model. When
hourly precursor concentrations of selected VOC's, NOX are included, R statistics increase to
approximately 0.85 suggesting that precursor species are important predictors of ozone levels.
-------
EPA-454/R-96-006
Chapter 1
Revision Number: 0
Date: November 1996
Page: 14
1.4 REFERENCES
Bruntz, S.M.; Cleveland, W.S.; Graedel, I.E.; Kleiner, B.; and Warner, J.L. "Ozone
Concentrations in New Jersey and New York: Statistical Association with Related Variables.'1
Science 186, 1974.
Capone, R.L. Presentation to OAQPS, "Predicting Downwind Air Quality with a Neural
Network", March, 1996.
Carter, W.P.L. "Development of Ozone Reactivity Scales for Volatile Organic Compounds."
Journal of Air and Waste Management Association 44. 1994: 881-898.
Chu, S. H. "Coupling High Pressure Systems and Outbreaks of High Surface Ozone
Concentration." Proceedings of the 80th APCA Annual Meeting. 1987: 87-113.5.
Chu, S. H. "Meteorological Considerations in Siting Photochemical Pollutant Monitors."
Atmospheric Environment 29. 1995: 2905-2913.
Chu; S. H. and Doll, D. C. " Summer Blocking Highs and Regional Ozone Episodes." Preprints
of the Seventh Joint Conference on Applications of Air Pollution Meteorology with AWMA.
American Meteorological Society 1991: 274-277.
Cox, W.M. "A Workbook for Exploratory Analysis of PAMS Data." June 1995.
Cox, W. M. and Chu, S.H. "Meteorologically Adjusted Ozone Trends in Urban Areas: A
Probabilistic Approach." Atmospheric Environment 27B. 1993; 425-434.
Cox, W. M. and Chu, S.H. "Assessment of Interannual Ozone Variation in Urban Areas from a
Climatological Perspective." Atmospheric Environment 30. 1996: 2615-2625.
Crowe, W. and DeFries, T.H. Presentation to OAQPS, "Use of Observation Based Models to
Predict Ambient Ozone Levels." February 27, 1996.
DeFries, Timothy H. Neural Network Modeling of Ambient Ozone Using the South East Texas
Regional Planning Commission Ambient Air Monitoring Data Set. Radian Corporation, June 12,
1995.
Dye, T.S.; Lindsey, C.G.; and Anderson, J.A., "Estimates of Mixing Depths from 'Boundary
Layer' Radar Profilers." Preprints of the 9th Symposium on Meteorological Observations and
Instrumentation Charlotte, NC, March 27-31, 1995.
-------
EPA-454/R-96-006
Chapter 1
Revision Number: 0
Date: November 1996
Page: 15
Hosier, C.R. "Low-level Inversion Frequency in the Contiguous United States." Mon. Wea.
Rev. 89. 1961: 319-339.
Lamb, B.; Guenther, A.; Gay, D.; and Westberg, H. "A National Inventory of Biogenic
Hydrocarbon Emissions." Atmospheric Environment 21. 1987: 1695- 1705.
Lindsey, C.G.; Dye, T.S.; Roberts, P.T.; Anderson, J.A.; and Ray, S.E. Meteorological Aspects
of Ozone Episodes in Southeast Texas. Paper No. 95WP96.02 presented at the 88th Air &
Waste Management Association Annual Meeting, San Antonio, TX, June 18-23, 1995.
Lindsey, C.G.; Dye, T.S.; Blumenthal, D.L.; Ray, S.E.; and Arthur, M. Meteorological Aspects
of Summertime Ozone Episodes in the Northeast. Paper FA 5.8 presented at the 9th Joint
Conference on the Applications of Air Pollution Meteorology with AWMA at the 76th American
Meteorological Society Annual Meeting, Atlanta, GA, January 28-February 2, 1996.
Ludwig, F.L.; Reiter, E.; Shelar, E.; and Johnson, W.B. The Relation of Oxidant Levels to
Precursor Emissions and Meteorological Features. Part 1: Analysis and Findings. Final Report.
EPA Contract 68-02-2084. SRI International, Menlo Park, CA. EPA Report No. 450/3-77-
022a, 1977: 153.
Lurmann, F.W. and Main, H.H. Analysis of Ambient VOC Data Collected in the Southern
California Air Quality Study. Report to the California Air Resources Board. Sonoma
Technologies, Inc., Santa Rosa, CA. 1992.
Northeast States for Coordinated Air Use Management (NESCAUM), The Ambient Monitoring
and Assessment Committee. Preview of 1994 Ozone Precursor Concentrations in the
Northeastern U.S. August 1995.
Pagnotti.V. "Seasonal Ozone Levels and Control by Seasonal Meteorology." Journal of the Air
& Waste Management Association 40. 1990.
Robinson, E. "Some Air Pollution Aspects of the Los Angeles Temperature Inversion." Bulletin
of the American Meteorological Society 33, 1952: 247 - 250.
Stoeckenius, T.E.; Ligocki, M.P.; Shepard, S.B.; and Iwamiya, R.K. Analysis of PAMS Data:
Application to Summer 1993 Houston and Baton Rouge Data. Draft Report. U.S. EPA Contract
68D30019, Systems Applications International, SYSAPP-94/115d. November, 1994.
-------
EPA-454/R-96-006
Chapter 1
Revision Number: 0
Date: November 1996
Page: 16
Stoeckenius, I.E.; Ligocki, M.P.; Cohen, J.P.; Rosenbaum, A.S.; and Douglas, S.G.
Recommendations for Analysis of PAMS Data. Final Report. U.S. EPA Contract 68D30019,
Systems Application International, SYSAPP94-94/011rl. February, 1994.
Systems Application International, Sonoma Technology, Inc., Earth Tech, and Alpine
Geophysics. Gulf of Mexico Air Quality Study. Volume 1: Summary of Data Analysis and
Modeling. Draft final report prepared for U.S. Department of the Interior, Minerals Management
Service, Gulf of Mexico OCS Region, New Orleans, LA. DCS Study, MMS 94-0046,
SYSAPP-95/013d, 1995.
Wixtrom, R.N. and Brown, S.L. In: Journal of Exposure Analysis and Environmental
Epidemiology 2, 1992: 51. Edited by Edo Pellizzari (averages recalculated here for five
Northeast cities only).
-------
Figure 1-1.
1M4 NoithMt Air Qnaltty Study
Time senes cross section of winds, mixing depth, and inversion conditions measured by the radar profiler on July 12-13. 1994 at
Bermudian Valley. PA. The thin solid line denotes the height of the mixed layer estimated using C n: and RASS temperature
data The thick line denotes the subsidence inversion. The shaded area indicates the region of the nocturnal low-level wind
maxima (Lindsey et al., 1995b)
-------
Figure 1-2. Back trajectories at the surface and 300m agl calculated from a ferry equipped with an
ozone monitor (operated by the state of Maine) that recorded exceedances of the ozone standard on
Julv 21, 1994
-------
Figure 1-3.
30.5
-«S.O -W.5
W«tt Longitude
-nuo
Plot of surface winds and ozone concentrations in the southeast Texas region
at 1400 CST on September 8, 1993 (SAI 1995). The dashed line indicates
the location of the sea breeze front.
-------
Figure 1-4.
Otmttwb)
150
100
SO
0
NOxtppb)
40
30
20
10
0
1 ,
A/ I
- A //V/\\
: , YVL.r-. ,, //X v^t^ i^-. -/V-j L^^^^,,^, i f . ri i i i
12
,
i i i I i i i I ' i 1 1 1 1 1 1 1 L
J L-
_L
I . I 1 1 1 1-
13 W1> 12 W11 12 Mil 12 Wll 12
Time (CST-60-Begin)
Tune series of ozone, NO, NOZ, wind speed, and wind direction measured at the Gilcrest, TX surfaci
period August 9-14, 1993 (SAI, 1995). When flow was offshore, high NOX and titrated ozone was e
When the flow reversed, high ozone concentrations were observed. However, with continued onsho:
concentrations decreased
-------
Figure 1-5.
SoulhMM Houston
Ventilation Analysis
8/16/93 0600 COT through 8/16/931700 CDT
outlMMt Houston
fetation
Mghtotand
OalwMon
0 < M UIII021S2»atSMO
MflhMmd
net
MI
outhMstHoMton
MghMMid
_
f in.
* taa
1211 :
M7:
a
U 14
Vector-integrated transport distances, resultant wind directions, and recirculadon factors (R), calculated from data collected by the
southeast Houston (SEH), Galveston (GAL) and High Island Platform (HIP) radar profilers for the period 0600-1700 CDT on
August 16, 1993.
-------
Kigure 1-6.
South***! Houston
Ventilation Analysis
a/19/83 OflOO CDT throuah 8/19/931700 CDT
SouttMttt Houston
Orivmton
GalvMton
High Island
« so miMzazToiiiMO
Mgh Island
o « so i* IM 221 rro lit aao
Southeast Houston
High toUnd
*1T
am u
IJB
Vector-integrated transport distances, resultant wind directions, and recirculation factors (R), calculated from data collected by the
southeast Houston (SEH), Galveston (GAL) and High Island Platform (HIP) radar profilers for the period 0600-1700 CDT on
August 19, 1993.
-------
Figure 1-7. Time-height cross-section of CN2 for July 12-13, 1994 at New Brunswick,
NY. Thick line denotes the subsidence inversion; thin line dunng the day
denotes the top of the mixed layer. On July 12, a subsidence inversion is
shown in the profiler data as a region of high reflectivity between 1750
and 2000 m agl. The inversion limited afternoon growth of the CBL to
below 2000 m. A slower growth occurred on July 13 resulting in
reduced vertical mixing of precursors, and an attendant elevated afternoon
ozone level
-------
onesport (86 ppb at 21:00)
arbor»ME <102 PPb at
I&e Au Haut, ME (116 ppb at 17:00)
ort Clyde, ME (124 ppb at 17:00)
ippsburg, ME (148 ppb at 15:00)
Cape Elizabeth, ME (148 ppb at 15:00)
Kennebunkport, ME (141 ppb at 14:00)
Rye, NH (135 ppb at 13:00)
Lynn, MA (105 ppb at 12:00)
Figure 1-8. Maximum Ozone Concentrations and Hours of Occurrence, 07/21/94
-------
120
100
<
oi
3
tr
5
o
x
Jonesport. ME
Bar Harbor, ME
Isle Au Haul. ME
Port Clyde. ME
Phlppsburg. ME
Cape Elizabeth. ME
Kermebunkport, ME
Rye. NH
Figure 1-9. 8-Hour Moving Average Ozone Concentrations on 7/21/94
-------
16 -
14 -
12
10 -
8
Dates: 7/6-7/8 & 7/20-
Isopentane
* Lynn
Chicopee
*- E.Htfd.
Stafford
* Cape Eliz.
*-E.Provid.
Toluene
Acetone
f
Formaldehyde
I. "
Figure 1-10. VOC Abundances for Six Northeastern PAMS Sites, July 1994.
-------
30
25
20
15
10
5
0
Dates: 7/6-7/8 & 7/20-
-Lynn
- Chicopee isoprene
* E.Htfd. |
^-Stafford
" -^- Cape Eliz.
-»- E.Provid. j.
jEthene |SOpentane A
1 A * ' $ '
i
Formaldehyde [
.
s r
) i .
N
C c 1
"
* s 1 M
t= 1 i
o i
">. 1
* X ^
A t i^\
0 » 0 t
2 S o
* *r -r
C
o
\
0
c
^
Figure 1-11. VOC Abundances Adjusted for Reactivity (Using Carter's MIRs) for Six Northeastern PAMS
Sites, July 1994.
-------
0>
X
S=
x.
O
c
N
C
ffl
M/P Xylene
Toluene
8
Toluene
16
20
Figure 1-12. Estimated Urban B/T and X/T Ratios from Atlanta, GA Source Profiles (from
Henry et al., 1994)
-------
u
.Q
Q.
4>
C
X
o.
E
N
C
o>
m
11
9
7
5
3
0
O
0
o oo ^ $
0
o
4 8 12
Toluene (ppbc)
0 Xylene ' Benzene
0
Figure 1-13. Measured Urban B/T and X/T Ratios from E. Hartford, CT PAMS Site
during July 1994 Episodes
-------
0>
c
o
4>
C
N
C
0)
CO
M/P Xvlene
Toluene
Figure 1
Aging)
8 12 16 20
Toluene
14. Predicted Changes in B/T and X/T Ratios at Rural Sites (Resulting from Airmass
-------
u
.Q
Q.
.
X
Q.
E
c
o>
N
C
4)
m
246
Toluene (ppbc)
8
0 Xylene
Benzene
Figure 1-15. Measured Rural B/T and X/T Ratios from Stafford, CT PAMS Site during
July 1994 Episodes
-------
o
.Q
Q.
O.
0)
c
X
d.
8 12
Toluene (ppbc)
0 Day
Night
Figure 1-16. m/p-Xylene versus Toluene at E. Hartford, CT PAMS Site during July 1994
Episodes
-------
o
£t
a.
2.8
2.4
2.0
1.6
1.2
0.8
0.4
0.0
-i r
0.0
0
0
1.5 3.0
Toluene (ppbc)
4.5
0 Day
Night
Figure 1-17. m/p-Xylene versus Toluene at Stafford, CT PAMS Site during July 1994 Episodes
-------
12
10 -
Q.
a
2
* - m/p Xylene
Isoprene
x »
V
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Hour of Day
Figure 1-18 Diurnal Isoprene and m/p Xylene (averaged for 6 NESCAUM PAMS Sites for July, 1994
Episodes)
-------
Figure 1-19.
BALTIMORE PAMS DATA 1993
DIURNAL BOX PLOTS--ORGAN I CS
2-1
A
C
T
Y I
L
0J
HOUR
2-
1
0-
HOUR
T
0
i'
E
2
HOUR
1-
8-
E
T 6
H
Y 4-
L
2i
8
6-
2-1
HOUR
HOUR
HOUR
-------
Figure l-20a.
Diurnal Profiles for
HOUSTON Galleria
Weekday Weekend
Acotytaw
Acftytene
OJftUUItIM OJIItt«1l*1»
Bhytone
Bhyitni
0 3 ( * U 15 11
lev
Etnane
0 3 I I 12 IS 11 11
Ethane
S
o t e u « 11 »i M o i ( i « H u 11 a*
-------
Figure l-20b.
Weekday
taopnm
Diurnal Profiles for
HOUSTON -Gaiieria
Weekend
taoprara
Benzene
Benzene
o 3 e i u u ti r
our
Toluene
i u « 11 v M
Toluene
I!]
i:'.
0 3 I I IJ 1! II M
Hour
0 3 I I 12 U U V J4
-------
Figure l-20c.
Diurnal Profiles for
HOUSTON -Galleria
Weekday
Otefins
Weekend
CMre
e .
-TT*
o 3 e t a is u n it
tar
o i i i 12 u it 21 it
Xylene
Xyiene
£
0 ,
0 3 I I 12 1S U
tar
0 J I I 12 15 It
-------
Figure l-21a.
Diurnal Profiles for
HOUSTON - Galieria
Weekday Weekend
Acetylene
Acetylene
i s-
t I
03IIBUU11C( 01(1 12 IS
Ethyiene
Ethytene
t
«. M
3 C I U 15 II Zi * 0 S I I U w II 21 n
(BUT |_-_
Ethane
Ethane
c 3 « » u is u n
hou
0 3 I U 15 II 21
-------
Figure l-21b.
Diurnal Profiles for
HOUSTON - Galleria
Weekday Weekend
»
toopmnt
o s i t e u 11 *
I-i
Benzene
C 3 I U IS
now
Toluene
Benzene
9 .
i t u is u r M
Toluene
0 3 e I « 15 U Z1 H
09II12UW21W
-------
Figure l-21c.
Diumal Profiles for
HOUSTON-Galleria
Weekday
Otefins
Weekend
Ofefins
. w
I "'
II '
Xytene
Xylene
c 3 e 9 w i: 11 n
lour
0 9 « I tZ 15 II 21 «
-------
g 0.04
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14
PREDICTED OZONE LEVELS
Figure 1-22. Model 1- Meteorological Data Only; 1994 PAMS Data; Phialdelphia, PA; R2=0.45, n=92.
-------
0.15
0.1
o
N
O
Q
W
OL
S 0.05
PQ
O
0
0
0.05 0.1
PREDICTED OZONE LEVELS
0.15
Figure 1-23. Model 2 - Meteorological and Ozone Precursor Data; 1994 PAMS Data; Philadelphia, PA; R2=0.84,
n=71.
-------
5.0
4.8
4.6
o
« 44
4.2
4.0
4.0 4.4 4.8
Log predicted Ozone
Figure 1-24. Philadelphia PAMS -Neural Network Size=l
Met, VOC and NO,
-------
Table 1-1. Most Abundant Anthropogenic VOCs in Selected Measurement Campaigns
1.
2.
3.
4.
5.
6.
7.
Northeast '94
Isopentane
Toluene
Propane
Ethane
n-Butane
m/p-Xylene
n-Pentane
5 City '84
Isopentane
n-Butane
Toluene
n-Pentane
m/p-Xylene
Propane
Isobutane
Los Angeles '87
Propane
Isopentane
Toluene
n-Butane
Ethane
m/p-Xylene
n-Pentane
-------
d>
c
o
N
O
5.0
4.8
4.6
4.2
40
40 44 48
Log predicted Ozone
Figure 1-25. Philadelphia Neural Network Size=3~Met,
VOC, NOX
-------
EPA-454/R-96-006
Chapter 2
Revision Number: 0
Date: November 1996
Page: 1
CHAPTER 2
PAMS DATA IN SUPPORT OF OZONE MODELING APPLICATIONS
2.1 INTRODUCTION
Models are valuable tools that provide enormous spatial, temporal and predictive
capabilities beyond the scope of monitoring networks. The use and interpretation of models are
often criticized due to uncertainties in model inputs and process characterizations, as well for
poor performance. The combined and complementary use of models and observations provides a
more credible basis for analysis than the sum of independent analyses using models and data
alone.
Support for photochemical modeling is one of several objectives targeted by the
Photochemical Assessment Monitoring Station (PAMS) program. The objectives of this
document are to provide example applications and recommendations for the use of PAMS data in
ozone modeling applications. An emphasis is placed on the utility of PAMS type data for
supporting the emissions based models (EBMs) commonly used in ozone assessment studies and
State Implementation Planning (SIPs). This document does not replace existing EPA Guidance
for Urban Airshed Modeling, but should be viewed as support to improve model application
studies by incorporating PAMS data.
This chapter starts with an overview of ozone modeling, followed by examples illustrating
the uses of PAMS data in supporting model evaluations, developing model inputs and performing
weight of evidence analyses in attainment tests. The chapter ends with a discussion on the uses
and caveats on a compound by compound (or compound class) basis of PAMS measurements and
their relation to model support.
2.2 MODEL OVERVIEW
Photochemical air quality simulation models (PAQSMs) used in most applications are the
gridded. fixed-frame (Eulerian) systems including the UAMTV, UAMV, ROM, RADM, SAQM,
CALGRTD, URM and others. All modeling systems invoke many approximations both in the
description of physical and chemical processes as well as in the solution of the system of
mathematical equations embodying the physics and chemistry. The spatial extents, or "domains",
of model applications often are characterized as being of urban (100-500 km), regional (500-
2000 km) or super-regional (> 2000 km) scales. Gridding refers to the horizontal resolution used
to delineate simulated air quality concentrations and provide detail on the emissions distribution
and meteorological variables (e.g., winds and temperatures). Typical horizontal grid resolution
ranges from 2-5 km and 20- 80 km for urban and regional applications, respectively. Most
-------
EPA-454/R-96-006
Chapter 2
Revision Number: 0
Date: November 1996
Page: 2
models produce hourly outputs, which can be aggregated for other averaging times of interest
(e.g., 8-hr, 24-hr, seasonal, annual). Models are resolved vertically (typically 5-15 levels or
more) to account for varying meteorology and emissions and to approximate vertical mixing
phenomena. Some recent systems accommodate nested or variable scale gridding schemes which
allow for detailed spatial treatment in urban areas (2- 8 km) and less dense horizontal resolution in
peripheral/rural portions (20-80 km) of the domain in order to optimize computational resources
and apparent precision. A trend toward increasing regional-scale (or mixed regional/urban),
Eulerian modeling has been apparent recently, in recognition of the interaction between
regional/rural and urban areas and associated "transport" issues.
Models should be viewed as a "system" (Figure 2-1), including the meteorological and
emission preprocessing models and the air quality simulation model (AQSM). Preprocessors
process raw data (i.e., emissions inventories and meteorological measurements) into the spatial
and temporal fields required by the AQSM. The AQSM calculates concentration fields of air
quality species (i.e., compound, element, free radical, or surrogate group), which are determined
by the combined interactive effects of source emissions, mixing processes (advection and
dispersion), deposition and chemical transformations.
So-called "transport" into a modeling domain is quantified through boundary conditions
which are user-specified concentrations surrounding the established modeled domain (planes on
each side and top), which can be brought into the domain through advection and dispersion. The
AQSM includes the chemical mechanism which performs the chemical transformation calculations
through highly condensed approximations (e.g., 20-50 species; 100 reactions) of the thousands of
actual chemical reactions occurring in the atmosphere. Several chemical mechanisms have been
developed. They all use condensation schemes which utilize surrogate "species" for aggregating
organic compounds. The carbon bond approach used in the urban airshed model (UAM) groups
compounds (and parts of compounds) by bond characteristics. Other approaches like the
Statewide Air Pollution Research Center (SAPRC) mechanism use a "lumped" approach which
groups compounds by similarities in reaction mechanism attributes.
These modeling systems are driven by emission inputs and often are referred to as
emissions based models (EBMs). acknowledging the difference between data driven observational
based models, and semi-empirical models like EKMA/OZIPM4.
Typically, a series of steps are required for all model applications:
A. Establishment of model domain, characterization and selection of modeling
episodes.
-------
EPA-454/R-96-006
Chapter 2
Revision Number: 0
Date: November 1996
Page. 3
B. Raw data gathering and processing of model inputs.
C. Model performance and diagnostic testing, including component testing of
emissions and meteorological preprocessors.
D. Development of emission control strategies and model application and
interpretation of results.
E. Corroboration and evaluation of strategy results.
The following examples illustrate methods in which PAMS data aid the support of model
evaluation and development of modeling inputs.
2.3 MODEL EVALUATION USING PAMS DATA
Two examples are presented to illustrate the use of PAMS (or PAMS-type) data in
evaluating ozone modeling applications. The examples are based on field programs from the
Southern California Air Quality Study (SCAQS) and the Texas-Louisiana COAST programs, both
of which collected the speciated and temporally resolved data required in PAMS. The purpose of
these examples are purely illustrative, to provide perspective on how PAMS can support ozone
modeling.
2.3.1 Example: Los Angeles, CA
Figures 2-2 through 2-4 depict time series plots of modeled versus observed values for
ozone, reactive hydrocarbon (RHC) and NO2, respectively, based on a 1985 historical episode in
Los Angeles.-CA (Wallerstein et a]., 1994). The figures are presented for exemplary purposes
only, to show a simple plotting procedure for conducting model/measurement comparisons of
ozone and precursors.
In addition to the example time-series plots, other graphical displays using tiling or
isopleths could be developed to provide comprehensive, two-dimensional views relating measured
and simulated data. Measured data could be superimposed on simulation maps, or side by side
maps of measured and simulated results could be displayed.
2.3.2 Example: Houston Ship Channel
This example is based on a preliminary analysis trying to diagnose the cause of poor model
performance using PAMS-type VOC data in Houston. The example should be viewed as a
hypothetical example of the potential value-added provided by speciated hydrocarbon data.
-------
EPA-454/R-96-006
Chapter 2
Revision Number: 0
Date: November 1996
Page: 4
Several subsequent analyses of this case have suggested significant problems related to both the
modeling and interpretation of results. Therefore, this example should be viewed solely for
illustration, no substantive conclusions are credible.
A highly localized ozone peak of 231 ppb was observed on August 19, 1993 in Houston,,,
Texas (Figure 2-5.) UAM-V modeling was unable to replicate timing, magnitude of the peak. A
series of VOC and NOX across-the-board emission sensitivity runs failed to improve model
performance. An analysis of nearby VOC data at the Clinton site indicated an anomalous peak in
the total NMHC (Figure 2-6). Through further analysis of the ambient speciation, the emission
inputs were adjusted to coincide with the ambient data. Subsequent model runs resulted in
improved capture of the peak ozone (Figure 2-7).
2.4 DEVELOPMENT AND TESTING OF MODEL INPUTS
2.4.1 Episode Selection and Domain Specification
Procedures for evaluating and selecting meteorological episodes for UAM SIP
applications have emphasized frequency and pervasiveness of high ozone concentrations (EPA,
1991). Over the last several years perspectives on episode selection have changed, largely in
recognition of the need for "richer" data bases, increasing trend toward regional analyses, and the
difficulty of translating results from highly stochastic events to the form of the ozone standard.
Arguably, the availability of PAMS data and other more intensive field studies (e.g., NARSTO-
NE, SOS) is reasonable justification (or a prerequisite) for selecting a modeling period. A model
application based on supporting precursor and upper meteorological data is more informative and
credible than an application with less data. Hence, any decision matrix for episode selection is
likely to contain a column indicating relative strength of supporting measurements.
Domain specifications
Almost all gridded model applications utilize regional (or greater) spatial scales to account
for important mixing processes due to transport, recirculation and other processes.
Consequently, any reasonable extension of modeling domains to incorporate any supporting data
is advised. Similar domain adjustments would hold true for urban specific domains which are
modeled with nested systems (ROM/UAM) or variable grid models (UAM-V).
Episode Selection
As discussed in Chapter 1, PAMS data assist in developing various conceptual pictures of
certain attributes of episodes, such as relative influence of transport, source mix, aging of air
masses, and propensity toward NOX or VOC-limiting conditions.
2.4.2 Development of Meteorological Inputs and Meteorological Model Evaluation
-------
EPA-454/R-96-006
Chapter 2
Revision Number: 0
Date: November 1996
Page: 5
The spatial and temporal attributes of meteorological data used in the specification of
mixing heights (or related vertical diffusivity parameterizations) and three dimensional wind fields
have substantial impact on simulating atmospheric mixing processes, and subsequent air quality
and control strategy predictions. Perturbations in either mixing height or wind fields have been
shown to change both the level of precursor control as wellas the preferred direction (i.e., VOC
or NOX) of control needed to demonstrate attainment (Sistla et al., 1994; Sistla et al, 1996).
Although different meteorological modeling approaches can produce disparate results, the
availability of high quality meteorological data should reduce differences in simulation results
among various techniques.
The historical shortage of quality upper meteorological data, often limited to twice daily
FAA soundings in one (or nearby) domain based location, has created concern in several UAM
applications. The PAMS requirement for upper meteorological monitoring is potentially a major
contribution toward improving model applications. PAMS minimum requirements specify 4
soundings per day of winds and temperature. While rawinsondes can be used to meet these
requirements, automated remote sensing techniques such as Radar/Sodar wind profilers and Radio
Acoustic Sounding Systems (RASS) offer enormous spatial (50-100 m increments, vertical Z-
space) and temporal (hourly) resolution, and are used as a basis for subsequent discussions.
Wind and temperature fields with greater temporal resolution should reduce much of the
uncertainty associated with growth of the surface-based mixed layer, particularly during the rapid
growth, late morning transition period. PAMS upper meteorological monitoring requirements
reflect a commitment in the right direction toward enhancing modeling efforts. The PAMS
contribution of 1 upper meteorological monitoring site per PAMS network is a foundation for
developing spatially (horizontal) representative monitoring networks capable of characterizing
regional gradients in mixing height and wind fields.
The following discussion focuses on the ability to improve wind and mixing depth fields
brought about by the PAMS upper meteorological monitoring requirements, with an emphasis on
the use of continuously operating radar profiler and RASS instrumentation. (More detailed
discussions regarding the use of PAMS meteorological data are found in Dye, 1995.)
2.4.3 Mixing depth
Mixing (and associated dilution/concentration) of ozone and precursors throughout an
episode impacts ozone concentrations, and subsequent control strategy calculations.
Characterization of the mixed layer throughout the morning to afternoon growth period may be
just as important as estimating maximum late-afternoon mixing depth. Continuously operating
radar and RASS profilers provide a means to depict this growth with improved temporal and
spatial (vertical) resolution. Particularly promising mixing height applications based on the use of
Radar profiler reflectivity coefficients, C2n, should provide a strong complement to RASS
-------
EPA-454/R-96-006
Chapter 2
Revision Number: 0
Date: November 1996
Page: 6
temperature profiles for characterizing mixing heights.
Advantages of Radar profiler and RASS
Information produced by profilers/sodars and RASS will improve characterization of the
mixed layer depth. Both the temperature profilers produced by RASS and C:n coefficients have
temporal resolution of at least one hour enabling improved description of diurnal mixing depth
development, a vast improvement over the twice daily FAA soundings. The use of C2n is
particularly attractive, as it is a direct indicator of mixing depth. Comparisons between aircraft
measurements and C2n (Figure 2-8) suggest that C2n adequately tracks mixing depth. As profiler
and RASS data from recently installed sites continues to be processed and reviewed, an improved
understanding of the strengths and limitations of remote sensing techniques under different
meteorological conditions will evolve. Currently, these methods suggest outstanding potential for
characterizing mixing and advection processes for modeling applications in the lower troposphere.
The ability of C2n to depict the diurnal mixed layer growth for July 12 and 13 is shown
clearly in Figure 2-9. The slower growth on the second day coincided with higher ozone
concentrations, suggesting that reproducing the temporal growth of the mixed layer is a model
sensitive component.
The potential value of C2n, relative to RASS also is illustrated in Figure 2-10.
Penetration of RASS often is limited to about 1000m, falling short of many typical daytime mixing
depths which often exceed 2000 m. Radar profilers penetrate well beyond the well-mixed layer.
Consequently, reflectivity values provided by profilers extend to high afternoon elevations of the
mixed layer that are beyond the scope of RASS. Some caution regarding the representativeness
of C:n during night is warranted given the relatively high nighttime values.
2.4.4 Wind fields
Wind fields strongly influence the outcome and interpretation modeling results. Given that
most of the air volume under consideration in any application is strongly influenced by upper air
flows, the importance of characterizing vertical wind field gradients should not be understated.
Upper air flow characterization dictates much of the "direcf'source-receptor relationships, as well
as the degree of mixing and attendant effects on atmospheric chemistry. The implementation of
continuous operating sodars and radar profilers is a major improvement, providing more resolved
time and space (vertical) data to better reproduce the most fundamental modeling inputs.
Additional surface stations and a local upper air station will improve the
representativeness of the processed wind fields. Previously, twice-daily soundings from the
nearest or most representative FAA location provided the raw data for vertical wind profiles.
PAMS requirements will insure the operation of at least one upper meteorological site within an
-------
EPA-454/R-96-006
Chapter 2
Revision Number: 0
Date: November 1996
Page: 7
"urban" scale domain.
Advantages of remote sensing instrumentation
Remote sensing instruments such as profilers/sodars and RASS provide a means to
characterize wind flow phenomena, and often are capable of resolving various meteorological
phenomena including land-sea breeze regimes, recirculation patterns and nighttime jet formation.
Figure 2-11 displays an analysis of output from a radar profiler in the Northeast U.S. over a 2-day
period. Various phenomena are superimposed on the diagram, based mostly on interpretation of
the wind data. While the wind data (speed and direction) form the basis for direct input into
meteorological models like the UAM Diagnostic Wind Model, the ability of profiler data to
enable interpretation of low-level jets and mixing depths sets provide phenomena for comparison
with meteorological models.
Graphical comparisons between simulated wind fields and remote sensing data should be
produced for those episodes incorporating upper meteorological data. Modeled wind fields
plotted in a manner compatible with that displayed in Figure 2-11 can be compared visually with
observed wind fields. These comparisons should include 1-dimensional (vertical) site by site
comparisons of several vertical levels and time periods, covering important temporal transition
periods. Two-dimensional (horizontal) comparisons between profiler outputs and simulated wind
fields for surface and elevated layers be produced to evaluate the model's ability to replicate
observed horizontal gradients.
2.4.5 Additional uses for PAMS meteorological data
Upper air temperature profiles
As discussed above, the RASS vertical temperature profiles provide increased vertical and
temporal resolution for estimating mixing depths. Additionally, temperature profiles are required
for atmospheric chemistry calculations performed within the chemical mechanisms
Solar and UV Radiation
PAMS requires measurements of total solar (.10 to 4.0 ^m) and UV (0.10 to 0.40 /urn)
radiation. The photolytic reactions are strongly sensitive to UV, and such measurements could
be used for calculating photolysis rates of key reactions. However, calculation of photolysis rate
constants often is automated within different modeling codes. UAM-IV, for example, assumes
bright sky conditions and sun position (spatial/temporal coordinates) for internal calculation of
photolysis rates. The UV data should be used whenever available they provide a more realistic
representation of the physics driving atmospheric chemistry routines in the AQSMs. Future
modeling guidance will need to address the use of radiation measurements.
-------
EPA-454/R-96-006
Chapter 2
Revision Number: 0
Date: November 1996
Page: 8
Relative Humidity
Water vapor plays an important role in ozone formation, and relative humidity (in
combination with temperature) can be utilized to estimate water concentrations. Relative
humidity is a critically important parameter for estimating visibility impairment, and should prove
valuable in future fine particle and visibility modeling applications.
Development of Boundary and Initial Conditions
The use of air quality data for developing initial and boundary conditions is largely an
iterative process coupled closely with model evaluation concepts discussed above. The spatial
coverage provided by PAMS is not adequate to develop comprehensive sets of air quality model
inputs. Consequently, the available data are best suited to evaluating initial and boundary inputs
used to drive the model simulation. Over the last ten years the spatial and temporal limits have
been expanded for most model applications, partly because extended simulations are less sensitive
to initial conditions. Therefore, this document places much greater emphasis on the use of data
for evaluation purposes rather than input development. Nevertheless, PAMS-type data should be
used for examining air quality inputs, especially boundary conditions which are a critical
component of any simulation.
For most urban scale model applications, one (or possibly two) Type I "upwind" sites will
be available for examination of boundary values. All of the comments regarding compensating
errors, the advantages of reducing degrees of freedom for general model evaluation apply equally
for evaluation of boundary conditions. Although it is important to quantify boundary ozone
(supported by PAMS Type 1 sites), measurements of NMOC, NOX and carbonyls add strong
value to characterizing boundary conditions. Boundary conditions and transport are closely
related, in the sense that transport phenomena are quantified as boundary values in a model
application. Transport involves several factors in addition to the "additive" effect of incoming
ozone. Transport includes movement of precursors or precursor "sinks" such as PAN and N:OS
which under certain conditions can release NOX. The role of ozone transported from an upwind
location and acting as a radical initiator downwind (ozone is the dominant source of hydroxyl
radicals) is a component of transport, in addition to a strictly additive role of imported ozone.
2.5 DEVELOPMENT AND EVALUATION OF EMISSIONS INPUTS
The PAMS speciated NMOC data provide strong support for modeling by enabling
evaluation of the emission inputs driving models. The reader is referred to Chapter 3 for
examples of the use of PAMS data in evaluating emissions.
2.6 DISCUSSION OF PAQSM PERFORMANCE AND CORRESPONDING USES OF
PAMS AIR QUALITY DATA
-------
EPA-454/R-96-006
Chapter 2
Revision Number: 0
Date: November 1996
Page: 9
Traditional ozone model evaluations have relied on operational evaluations comparing
surface ozone observations with modeled ozone predictions. Ozone models often produce
reasonable surface level spatial and temporal reproductions of observed ozone fields. However,
the use of ozone as the sole indicator of model performance/behavior can produce misleading
confidence in the model's ability to correctly predict response to other meteorological episodes,
or more importantly, emission reduction scenarios.
2.6.1 PAMS and Compensating Errors
Ozone is a secondarily formed pollutant with concentrations dependent on many factors
many combinations of which could lead to similar ozone. Various combinations of compensating
errors can produce apparently "correct" ozone fields for wrong reasons. The chance that the
model is reproducing ozone for the "right" reasons may or may not be greater than the chance
that some combination of compensating errors is responsible. For example, underestimates of
emissions can be compensated by restricted mixing through underestimates of mixing heights or
wind speeds. The complexity of the coupling of meteorological, emissions, deposition and
transformation processes introduces numerous opportunities for compensating errors, and renders
a very difficult identification of those processes and factors involved in compensation.
The issue underlying compensating errors is that models are applied in a prospective
predictive mode with an assumed confidence, built on operational evaluation of ozone, that the
physical and chemical processes are adequately characterized and therefore the model will
respond correctly to emission perturbations. However, an operational evaluation only assures
that surface ozone is estimated reasonably well under the existing model input scenario, a
phenomenon resulting from either (1) adequate characterization of physical/chemical processes or
(2) the operation of compensating errors. Because so many non-linearities exist, encountering a
"flaw" in the modeling system may not happen until rather aggressive control scenarios are
imposed which can not be evaluated because the testing of "projected" emission control
experiments is impossible. Consequently, the model could produce misleading, and even
misdirected, conclusions regarding control strategies.
The influence of compensating effects may or may not lead to incorrect conclusions
regarding control strategy analysis. But the assumption that compensating errors are not present
and processes are characterized appropriately because of a successful operational evaluation is not
valid. A prevailing consensus does not exist on the frequency and importance of significant
compensating errors in current applications.
2.6.2 Suggested Uses of PAMS Data for Model Evaluation by Compound Class
The previous discussion on compensating errors is a logical lead-in to describing the value
-------
EPA-454/R-96-006
Chapter 2
Revision Number: 0
Date: November 1996
Page: 10
of PAMS data in support of the model evaluation process. Compensating errors are related to
degrees of freedom within a system. Fewer data categories result in a less constrained evaluation
system and consequently increase the likelihood of a strong compensating error effect. The
availability of additional measurements beyond surface ozone restricts the overall freedom for "
model calibration" and similarly reduces the propensity toward compensation. A simulation
exhibiting good agreement with ozone and precursors would yield far greater confidence that the
model is working properly, than with the use of surface ozone only. Hence, the intrinsic value of
incorporating precursor data in the model evaluation process is the potential for reducing
compensating errors in the modeling system.
The following sections discuss and provide examples of the use of subclasses of PAMS
data in model evaluations. Somewhat more emphasis is placed on caveats associated with the
data in order to forewarn the user to prevent inappropriate model-data comparisons. Any
comparison of measured air quality and simulated air quality reflects the cumulative effects due to
various physical/chemical processes. Insight on the relationship between physical and chemical
processes and associated measurements should be used to guide the types of analyses performed
on the available data. Certain types of data may be subject to greater relative influence from a
specific process (e.g., less reactive NMHC and emissions), as opposed to other measurements
subject to multiple processes (e.g., ozone affected by emissions, meteorology and
photochemistry).
Model-measurement comparisons should^make full use of the spatial and temporal ranges
available in the measured data. In addition to the station by station time-series plots, 2-
dimensional spatial displays during selected time periods using isopleths, tiling, or other displays
should be produced to convey a sense spatial representativeness replicated by the model.
Extreme care should be exerted since the sparse spatial density of PAMS sites could create
misleading depictions of the true atmospheric state.
2.6.3 Total NMOC and NMHC
Comparisons of PAMS data with a corresponding estimate of modeled aggregate non-
methane organic compounds (NMOC) provide a means to determine if the modeling system is
capable of characterizing precursor/VOC levels. The limitations and definitions of both measured
NMOC and modeled NMOC must be understood in order to develop logical comparisons. The
definition of the sum of compounds reported as measured NMOC are likely to never match an
analogous definition for modeled NMOC. This discrepancy is due both to the variety of
"measured" NMOC definitions and the various NMOC condensation schemes incorporated in
chemical mechanisms. Such differences do not preclude model/observation comparisons, but
they must be understood to explain the fraction of disagreement not attributed to model error.
-------
EPA-454/R-96-006
Chapter 2
Revision Number: 0
Date: November 1996
Page: 11
Measurement/reporting Issues.
Measured NMOC is not a clearly defined term, and varies with reporting procedures,
instrumentation and instrument technique. The precise definition of NMOC is all organic
compounds minus methane. The closest measured approximation from a typical PAMS site
would include all compounds from both GC and cartridge techniques (carbonyls are NMOC).
However, the NMOC reported to AIRS typically is the list of 56 or so targeted species based on
Gas Chromatograph (GC) measurements. The targeted species are but a subset of all the species
that elute through a GC column. Sometimes the cumulative areas under all peaks are reported as
NMOC. Different instruments and techniques have substantial impacts on GC derived NMOC,
adding an additional source of inconsistency concerning NMOC data reporting. For example, the
loss of many oxygenated and polar acting species, including biogenic compounds (e.g.,
monoterpenes), associated with sample pretreatment and water management is system dependent.
Generally, as sampling methods and GC instruments and techniques improve, more compounds,
including carbonyl species, will be captured and passed though to the detector. This is desirable
progress, but an accounting of the different techniques, changes in methods, etc. must be available
to provide a basic understanding of what measured NMOC represents. Further complicating this
issue is the uneven detector response to different organic compounds. These issues are especially
important with respect to trends analyses.
Carbonyls can account for 30% or more of the total NMOC, yet a much smaller fraction
is reported as PAMS requires the reporting of only three compounds: acetone, formaldehyde and
acetaldehyde. Numerous issues are associated with carbonyl sampling and analysis as with any
monitoring technique. These measurement related issues are not intended to discourage
model/ambient comparisons. However, clear understanding of what is measured/reported will
improve the interpretation of model/ambient comparisons.
Simulated NMOC and NMHC
Thousands of reactions and hundreds of species are required to explicitly characterize the
atmospheric chemistry phenomena responsible for ozone formation. The inclusion of so many
reactions and species would extract an enormous computational burden; consequently, PAQSMs
utilize chemical mechanisms which condense the number reactions and species down to
manageable levels. Typically, the inorganic species and reactions (e.g., principal NOX and ozone
reactions) are treated explicitly. However, the organic chemistry is highly parameterized with
several "surrogate" species used to represent compound classes, bond types or other functional
relationships. Aggregating all of the modeled VOC groups represents an approximation of
NMOC. However, given the limitations of carbonyl measurements as well as the model's
treatment (or lack of) numerous carbonyl compounds, more meaningful model/data comparisons
would be conducted with estimates of non-methane hydrocarbon compounds (NMHC). In most
chemical mechanisms, hydrocarbon groups are differentiated from oxygen containing organic
groups allowing for an adequate aggregation to represent NMHC. As discussed earlier, an
-------
EPA-454/R-96-006
Chapter 2
Revision Number: 0
Date: November 1996
Page: 12
understanding of what is actually measured and reported is a prerequisite for conducting
comparisons.
2.6.4 Speciated VOC and Carbonyls (Isoprene and Formaldehyde)
Comparisons between modeled and measured species are limited by the model's ability to
depict explicit species. Explicit organic species in the CB4 mechanism are formaldehyde,
isoprene, ethanol and methanol. Comparisons with both formaldehyde1 and isoprene are strongly
recommended, given the potential value both species provide in characterizing important
photochemical and natural emissions processes. Isoprene is the most important biogenic emission
precursor specie. In the Eastern U.S., the suspected levels of biogenic emissions are so large that
any observed corroboration of their predictability is worthwhile. The level of biogenic emissions
influences the relative need for NOX or VOC control strategies. Although isoprene is reactive, the
emission levels are so large that resulting ambient levels (> 1 ppb) should provide a reasonable
basis for comparison between modeled and observed levels.
Formaldehyde is emitted directly as a primary species, formed secondarily, and undergoes
photolysis. Although formaldehyde may not act as a surrogate for other carbonyl species, the
observed levels, chemical reactivity and multiple activities suggest that formaldehyde is an
important indicator of model performance.
2.6.5 Nitrogen: NO, (NO, NO2), NOy
Comparisons of NO, NO: and total oxidized nitrogen (NOy) with corresponding modeled
estimates provide a basis for corroborating the NOX emissions component and atmospheric
chemistry phenomena. Comparisons of the diurnal pattern of NO concentrations are influenced
mainly by proximity of NOX emission sources, particular during the morning hours. The titration
of ozone due to NOX emissions diminishes during the day as NO is oxidized to various products.
Total NOV comparisons, assuming measured NOy are available, complement NO
comparisons. During periods when NOX is most likely to reduce ambient ozone, NO constitutes
the major fraction of NOV However, as oxidation processes proceed, emitted NO eventually
transforms into other more highly oxidized species (NO2, nitric acid and PAN) which are major
components of NOy. In fact, the ratio NO/NOy is a useful metric for comparison as it provides a
relative measure of air mass aging, higher ratios reflecting air laden with "fresh" NO, emissions.
Total NO^ measurements should account for NOX, HNO3, PAN, and organic nitrates, many of
which are semj-volatile and exist in the particle phases; species typically not accounted for in most
1 Certain parts of various olefin compounds are aggregated as Formaldehyde in the CB4 mechanism.
Thus, the modeled formaldehyde will always be somewhat higher than "true" simulated formaldehyde.
-------
EPA-454/R-96-006
Chapter 2
Revision Number: 0
Date: November 1996
Page: 13
chemical mechanisms. Thus, some unknown amount of negative (underestimate) bias due to
organic nitrate species that are not modeled is present. Nevertheless, on balance modeled NO,
(sum of NOX, HNO3 and PAN) probably is subject to similar difficulties associated with using
modeled NMHC as a basis for comparison with ambient data, and should be considered a useful,
albeit rough estimate of total NOy.
Comparisons with NO2 provide a strong test of model's ability to capture the timing and
magnitude of atmospheric chemistry phenomena, given the central role that NO2 plays in oxidant
chemistry. Unfortunately, NO2 measurements rarely are available, limited to a few select research
grade efforts in special field studies. NO2 data reported from most monitoring sites are reported
as the difference between measured NOX and NO. The chemiluminescence technique for
measuring NOX and NO is more reliable for NO. However, the total NOX measurement, which
typically requires conversion of NO2 to NO through a molybdenum catalyst, is subject to strong,
positive interferences from other NOy species (e.g., HNO3, PAN). Thus, the NOX measurement
is believed to be reside somewhere between actual NO. and total NOV.
A y
2.7 USE OF PAMS DATA TO CORROBORATE MODELED STRATEGIES
Modeled attainment demonstrations require one to project emissions and corresponding air
quality estimates several years into the future to a statutory attainment date. Model predictions are
then compared to the national ambient air quality standard (NAAQS) to determine whether the
simulated control strategy is likely to be sufficient to meet the NAAQS. This latter exercise is
referred to as a "modeled attainment test".
There are substantial uncertainties inherent in modeled attainment demonstrations. These
arise from uncertainties in the data bases driving the model, approximations of chemical/physical
processes made in the model's formulation and uncertainties about a number of factors affecting
emission projections. In recognition of these uncertainties, future modeled attainment tests are
likely to incorporate a "weight of evidence" concept to assess adequacy of a proposed emission
control strategy. In a weight of evidence analysis, air quality modeling results serve as one
(perhaps the most important) element. Other elements include a series of corroborative analyses,
many of which will draw on the kinds of data produced by the PAMS network. Table 2-1
illustrates elements which would be considered in a typical weight of evidence analysis. The
middle column in the table identifies factors which affect the credence or "weight" given to a
particular element. The right-hand column describes outcomes which would be consistent with
concluding that the attainment test is passed despite air quality modeling results which do not
quite show attainment.
Looking at the elements in Table 2-1, there are a number of potential uses for PAMS data.
For example, we see that the more extensive the (air quality) data base used to formulate inputs to
and evaluate performance of the photochemical grid model, the greater the weight that can be
-------
EPA-454/R-96-006
Chapter 2
Revision Number: 0
Date: November 1996
Page. 14
assigned to its results. Looking at the trend data element, we see that presence of precursor trends
which are consistent with apparent ozone trends would lend weight to results of the trend analysis.
Other trend analyses may be used as means for assessing model performance, thereby affecting
weight assigned to modeling in subsequent "mid-course reviews" of strategies prior to the
statutory attainment date. For example, as noted in Chapter 4, observed trends in ratios of
indicator species like HCHO/NOy or trends in highly reactive to less reactive VOC species may be
compared to model predictions to help assess model performance. Good performance would
increase the weight given to modeled results in the attainment test. Table 2.1 also shows how
results produced from observational models would be used in a weight of evidence analysis. The
term "observational models" in Table 2-1 includes "receptor models" (Chapter 3, this report) as
well as the observational models described in Chapter 4. The principal value of the observational
models is to provide qualitative indicators of which strategies are likely to be most effective in
reducing future ozone levels. Thus, they may be used to corroborate whether a proposed strategy
simulated with a photochemical grid model is addressing the appropriate classes of sources.
PAMS data are also of potential use for the "selected episodes" element in the weight of evidence
analysis. As described in Chapter 1, the PAMS data base will provide much more extensive
information about meteorological conditions aloft than is currently available. Because conditions
aloft are believed to be important factors affecting observed surface ozone concentrations, this
information might be used to good effect in helping us to identify distinctive meteorological
regimes leading to high ozone. This will provide increased confidence that we are considering the
important regimes corresponding to high ozone, thereby increasing the weight given to the
estimated severity of selected episodes in the weight of evidence analysis.
-------
EPA-454/R-96-006
Chapter 2
Revision Number: 0
Date: November 1996
Page: 15
2.8 REFERENCES
Dye, T.S.; Lindsey, C.G.; and Anderson, J.A. "Estimates of Mixing Depths from "Boundary
Layer" Profilers." Preprints of the 9th Symposium on Meteorological Observations and
Instrumentation Charlotte, NC, March 27-31, 1995.
Roberts, P.T.; Main, H.H.; and Korc, M.E. Comparison of 3-D Air Quality Data
with Model Sensitivity Runs for the South Coast Air Basin. Paper No. 93-WP-69B.05 presented
at the Air & Waste Management Association Regional Photochemical Measurement and
Modeling Studies Conference, San Diego, CA, November 8-12, 1993.
Sistla, G.; Ku, J.Y.; Zhou, N; Hao, W.; and Rao, ST. Sensitivity of the UAM-predicted Oznne
Concentrations to Wind Fields in the New York Metropolitan Area. Presented at the 8th Joint
AMS/AWMA Conference on the Applications of Air Pollution Meteorology, Nashville, TN,
1994.
Sistla, G., Zhou, N.; Hao, W.; and Rao, ST. Sensitivity of the UAM to Boundary Conditions.
Presented at the 9th Joint AMS/AWMA Conference on the Applications of Air Pollution
Meteorology, Atlanta, GA, 1996.
Systems Applications International. Gulf of Mexico Air Quality Study Final Report. Prepared for
the U.S. Department of the Interior. MMS 95-0038, 1995.
U.S. Environmental Protection Agency. Guideline for Regulatory Application of the Urban
Airshed Model. EPA-450/4-91-013. 1991.
Wallerstein. B.R.; Broadbent, J.P.; Hogo, H.: Cassmassi, J.; Mitsutomi, S.; Bassett, M.; Lester,
J.C; and Zhang. X. Ozone Modeling - Performance Evaluation. Draft Technical Report V-R.
Prepared for the South Coast Air Quality Management District, California, June 1994.
-------
Air Quality Modeling System
Air Quality Fields
ICs/BCs
I Meteorological
Model
Meteorological
Fields
( Emissions'
\Model
Emissions Fields
Ambient Concentrations
Deposition levels
Figure 2-1. Schematic of air quality modeling system.
-------
6 12 IB
JUNE 5. 1985
6 12 18
JUNE 6. 1985
6 12 IB
JUNE 7, 1985
Figure 2-2. Companson between predicted and measured ozone concentrations for June 5-7, 1985 in the South Coast Air
Basin (Wallerstein et al.. 1994). The solid line = distance-weighted mean value (average of 4 gnd cells), shaded
area=maximum and minimum value of the 9 cells around the station. Observed values are squares
-------
i
.t 10 -
6 12 IB
JUNE 5, 1985
6 12 18
JUNE 6, 1985
8 12 IB
JUNE 7, 1985
2-4
Figure 2-3. Companson between predicted and measured NO, concentrations for June 5-7, 1985 in the South Coast Air
Basin (Wallerstem et aJ, 1994) The solid line = distance-weighted mean value (average of 4 grid cells), shaded
area=maximum and minimum value of the 9 cells around the station. Observed values are squares.
-------
8 12 IB 24
JUNE 5. 1985
JUNE 6, 1965
JUNE 7. 1985
Figure 2-4. Companson between predicted and measured reactive hydrocarbon (RHC) concentrations for June 5, 6,
and 7, 1985 in the South Coast Air basin of Claifornia (Wallerstein et al., 1994) The solid line = distance weighted mean
value (i.e., average of 4 grid cells).
-------
3380.00-
3360.00-
3340.00-
3320.00-
3300.00-
3280.00-
Figure 2-5. Ambient isopleths of surface level ozone for Houston, TX on August 19, 1993 (SAl, 1995?).
-------
Concentration (ppb)
I
t
§
i
D
-jt
5
O
O
O
CO
m
as
c
re
Ov
>
cr
n
n
c.
c:
n
0
5'
c
X
cr
ft
o1
-n
>
-J
VO
VO
$
L/l
-------
, II. 1MU AOCm M. IW»
0 4 e 12 l« 20 24 28 SZ 36 40
. IB. JM»
4B ss ae oo 64
200
g.
90
TNZ
- ntOILATIDl
- saanjxaa
SOOOMBO
QflH
e 12 IB 20 34
404440529000040872
17. 1OB3
12 16 20
32
10. 1003
38 40 44
40
A»f
-------
Avg. difference = 2.8 m
rms difference = 196. 2 m
std of the differences * 200.2 m
§500 +
Figure 2-8.
500 1000 1500 2000 2500 3000 3500
Mixing depth from aircraft data (m msi)
Scatter plot of CN2 derived mixing depths estimated from aircraft
profiles of pollutant concentrations, turbulence and temperature.
Twenty five comparisons were made using aircraft data collected
in the afternoon near three profilers in southeast Texas (SAI,
1995).
-------
Figure 2.9 Time-height cross-section of CN2 for July 12-13, 1994 at New
Brunswick, NY. Thick line denotes the subsidence inversion; thin
line during the day denotes the top of the mixed layer. On July 12,
a subsidence inversion is shown in the profiler data as a region of
high reflectivity between 1750 and 2000 m agl. The inversion
limited afternoon growth of the CBL to below 2000 m. A slower
growth occurred on July 13 resulting in reduced vertical mixing of
precursors, and an attendant elevated afternoon ozone level.
-------
3000 -
XRASS Tv-based mixing depth
9 12 15 18 21
8/18
6 9 12 15 18 21
Hour(CST)
8/19
6 9 12 15 18 21 0
8/20
Figure 2.10 Time series plot of mixing depths estimated from CN2 and Tv data from a
meteorological model for a radar profiler site in Houston, TX for August 18-20,
1993 (Dye, etal. 1995).
-------
1994 Northeast Air Quality Study
Bwmudlaa Valley. PA
July 12.1904
1994 Northeast Air Quality!
BMmudtenViltoy.PA
C.1- and T.-Derived
Mixing Depth
Figure 2-11. Time-height cross-section of winds on July 12-13, 1994 at Bermudian Valley, PA, indie
mixing depths. Each wind barb indicates direction and speed.
-------
Table 2-1. Factors Affecting Weight of Evidence and Acceptance of Model
Results Nearly Passing the Attainment Test
Tvue of Analysis
Factors Increasing
Weight of Evidence
Factors Supporting
Deviation from Test
Benchmark(s)
Photochemical Grid
Model
-good performance
-extensive data
base
-short projection
period
-confidence in
inventories &
projections
-overpredictions
-maj or improvement
in predicted AQ
using a variety
of indicators
-results come very
close to meeting
the benchmark(s)
-other peer-
reviewed grid
models predict
comparable or
better improvement
in ozone
Trend Data
-extensive
monitoring network
-precursor & ozone
trends avail.
-statistical model
normalizing trend
explains much
variance
-little bias in
statistically
predicted highest
ozone
-short projection
period
-pronounced, stat.
significant
normalized trend
-continued,
comparable
relative
reductions in
emissions
provided for
-pronounced
downward
normalized trend
exceeding that
anticipated with
grid model
Observational Models
-extensive
monitoring network
-QA'd, self-
consistent results
-plausible,
physical
explanations for
findings
-indicates sources
other than those
in modeled
strategies play
significant roles
-------
Selected Episodes
-all met.regimes
corresponding w.
high obs. O3
considered
-met.ozone
potential of
episodes exceeded
I/year
-observed 03 »
design value
-Severity of met.
conditions
expected to be
exceeded « 1/yr
Incremental
Costs/Benefits
-good documentation
for cost estimates
-lack of
alternatives for
reducing emissions
-lack of model
responsiveness for
variety of
strategies as
benchmark is
approached
-lack of model
responsiveness
accompanied by
high incremental
costs
Other (optional)
Analyses
-rationale
documented
-------
EPA-454/R-96-006
Chapter 3
Revision Number: 0
Date: November 1996
Page: 1
CHAPTERS
EVALUATING EMISSIONS FACTORS, MODELS & INVENTORIES
WITH PAMS DATA
3.1 INTRODUCTION
One of the principal benefits of the PAMS monitoring network is the feedback it provides
to the various elements of the ozone regulatory program. Emissions inventories are an important
element of the ozone program. PAMS provides valuable feedback which can be used to evaluate
and improve the inventory. This chapter begins with a brief background description of emissions
inventories and potential inventory problems that PAMS data can elucidate. This introduction is
followed by examples illustrating the use of PAMS data to evaluate and refine emissions
inventories. These examples are separated into two sections: direct comparisons of the inventory
components with measured data and more refined analyses using multivariate and chemical mass
balance analyses. The chapter concludes with a more detailed description of the analytical
techniques.
3.2 BACKGROUND
Emissions inventories have long been a cornerstone of air quality management. Emissions
estimates are important for determining applicability of sources in permitting and control
programs, ascertaining the air quality impact of sources and appropriate mitigation strategies, and
a number of other related applications by an array of users, including federal, state, and local
agencies, consultants, and industry. Data from source-specific emissions tests or continuous
emissions monitors are usually preferred for estimating a source's emissions because those data
provide the best representation of the tested source's emissions. However, test data from
individual sources are not always available and, even if they are, the tests only represent a
snapshot in time and may not reflect the variability of actual emissions over time. Continuous
emissions monitors could resolve this variability concern but are expensive and technologies are
unproven for some pollutants. Thus, emissions factors are frequently the best or only method
available for deriving emissions estimates, in spite of their limitations.
Emissions estimates are the product of emissions factors established for various source
categories of VOC, NOX and CO as well as activity levels established for each source category.
Emissions factors describe the amount of emissions produced per unit of activity at a source.
Examples are "Ibs. NOX produced per megawatt of power produced by a utility" or "grams of
VOC produced per vehicle mile traveled by the current mix of automotive sources". Emissions
factors are often based on measurements made for a limited number of sources within a broad
-------
EPA-454/R-96-006
Chapter 3
Revision Number: 0
Date: November 1996
Page: 2
source category. Further, the measurements may be dated or made at geographical locations far
removed from the area of interest.
The activity level estimate from a source describes the number of units produced by a
source, which result in the by-product of pollutant emissions. In the two examples cited above,
activity levels would be the megawatts of power produced by the utility in question and the
vehicle miles traveled by the automotive fleet over a period of interest. In some cases, such as for
utilities, activity levels can be determined with considerable reliability. In others however, activity
levels must be estimated using surrogate indicators like population, sales figures and employment
statistics.
Emissions inventories are used for a variety of purposes. Some of these, like use in a
photochemical modeling exercise, require a high degree of spatial and temporal resolution, as well
as information about the chemical composition of the emissions. Emissions models are used to
develop this information which, in turn, is used as input to the photochemical models. Emissions
models need to address issues like, "how are emissions distributed within a city?", and "what kind
of temporal or diurnal patterns are reasonable to assume for the emissions from various source
categories?" These estimates are obtained from emissions factors, estimated activity levels,
estimated spatial/temporal patterns of activity using various surrogates, available information on
sensitivity of emissions factors to meteorological conditions and limited information on the
chemical composition of emissions from the various source categories.
The concept of using ambient measurements to improve emissions models, factors and
inventories is not new. In 1985, Air Pollution Control Association Specialty Conference on
Receptor Methods for Source Apportionment, several authors discussed the potential application
of receptor models to identify unknown sources or source categories. In that conference, several
authors demonstrated the use of receptor models to interpret speciated ambient measurements of
paniculate matter, (Pace, 1986). One very early investigation to reconcile atmospheric
hydrocarbons with sources was conducted in 1974, (Mayrsohn). Recently, a critical evaluation of
studies that used ambient data to infer weaknesses in ozone precursor emissions inventories was
completed. (Yarwood).
Identifying possible discrepancies in the inventory compared to the ambient data is only
the first step in the process. This has to be followed up with detailed emissions surveys for
sources around the monitor, careful review of the speciation and temporal profiles used to convert
the inventory to something compatible with the monitored results. This is costly and time-
consuming, but is the only way to truly improve the quality of an emissions inventory. Therefore,
while PAMS data can provide a very good tool for assisting in the overall improvement of
emissions inventories, it does have certain limitations as discussed above. This Chapter is not
designed a "cookbook" on how to use PAMS data; rather, its intent is to illustrate some of the
-------
EPA-454/R-96-006
Chapter 3
Revision Number: 0
Date: November 1996
Page: 3
effective uses of PAMS data for inventory evaluation.
3.2.1 Potential Inventory Problems
c
Potential problems that can be identified (and potentially resolved) through the use of ,
PAMS data are of two general types: emissions factor/model representativeness and proper
application of the factors, including spatial/temporal resolution. These issues are summarized in
Table 3-1.
3.2.2 Difficulties in Comparing Ambient Data and Emissions Estimates
The use of ambient measurements to corroborate emissions estimates are potentially useful
for focusing attention on particular assumptions underlying the emissions inventory. There are
many limitations associated with using ambient data to evaluate emissions inventories and care
should be taken to not over-simplify the results of such comparisons. A discussion of some of the
difficulties and issues follows:
Spatial Representativeness
One issue that has not been completely resolved is determining which emissions area to
compare with the ambient measurement. Several gridding techniques have been proposed [Pace
(1978) and Main]. These basically consist of developing finely gridded micro inventories for
comparison with the ambient data, forming concentric cells or grids representing increasingly an
larger area around the site. Figure 3-1 represents one such gridding system used to analyze
Hartford emissions data (Main). Under transport conditions, it may be necessary to use emissions
from a sector upwind of the monitor for the time just preceding the sample collection. The
representativeness of the monitoring network also needs be considered. To fully evaluate an
inventory, several monitoring sites may be needed depending on the complexity of the sources and
the nature of the area. The mix of emissions sources, the meteorology and the location of the
monitor all influence the analysis, a consistent (or inconsistent) analysis at one urban monitor may
not mean that the emissions estimates in all parts of the area are good (or poor). To get a sense
of the representativeness of the PAMS data for inventory evaluation, it is useful to perform
correlation analyses of total VOC and VOC species to see how they compare with data from
other sites.
Meteorological Issues
The most important influence to consider when comparing emissions estimates with
ambient data is to account for the influence of meteorology. All non-point source emissions in an
inventory are considered to occur at the surface and may be appropriate to evaluate with ambient
data. However, elevated point sources may not be accurately represented at an emissions-
oriented monitor near the ground due to meteorological influences. Also, the relationship
-------
EPA-454/R-96-006
Chapter 3
Revision Number: 0
Date: November 1996
Page: 4
between emissions rates and ambient air concentrations is quite complex, depending on many
meteorological parameters such as wind speed and direction, atmospheric stability, temperature
inversion heights, and horizontal and vertical diffusion rates. Therefore, careful analysis of the
meteorological data at the site needs to accompany the analysis of the ambient data in order to
define the appropriate comparisons to make.
Chemical Reactivity
As stated previously, emissions inventories are based on emissions factors and activity
level estimates. These emissions factors are available for the criteria pollutants (NOX, VOC, CO,
PM, SO2). In reality, VOC emissions are a composite of multiple hydrocarbon compounds
generated through some type of chemical process. These compounds react and form other
compounds once they are released into the atmosphere. Some of these reactions occur very
rapidly and others may take hours to occur. This is important to consider because a monitor is
measuring a point in time and may not reflect the original mix of emitted species that is being
transported into that area at that point in time due in part to these chemical reactions. Thus, along
with the meteorological influences mentioned above, the reactivity of the emissions from sources
near the monitor may lead to concentrations or ratios that are inconsistent with the original profile
of the area's emitted species.
3.3 PAMS RESULTS
PAMS data have already been useful in analyses to evaluate the viability of the emissions
inventory. However, the full potential of these data are just beginning to be realized. A recent
report (Yarwood et al., 1994) evaluated 25 studies related to this topic and provided objective
and critical overview of each. Appendix A tabulates some of the key analyses identified in that
report. When researchers have access to the vast body of data being collected in PAMS, the full
benefit of the PAMS network for evaluating emissions data can be realized.
PAMS data can be useful to the regulatory community in three ways. First, measured
concentrations of certain indicator species can be compared with emissions estimates.
NMOC/NO, ratios, and comparisons of inventory estimates of key VOC species with ambient
measurement of these species are two common types. Second, quantitative analyses, such as
Chemical Mass Balance (CMB) can be used to compare emissions estimates for specific source
categories. CMB is a sophisticated least squares statistical method for identifying the most likely
source categories contributing to a given ambient sample, based on the relative amounts of each
species present in the sample and the relative amounts of those same species present in the
emissions of source categories. The CMB differs from tracer methods in that the indicator
species need not be unique to a given source category. Third, PAMS data provide an opportunity
to compare the PAQSM-estimated species concentrations with ambient measurements of these
species at the time of day that the measurements were taken. The value of such comparison is
-------
EPA-454/R-96-006
Chapter ?
Revision Number: 0
Date: November 1996
Page: 5
obvious, both for inventory evaluation and to verify the transformation algorithms in the model.
Unfortunately, chemical mechanisms commonly used in photochemical models aggregate
emissions species. Thus, our ability to compare predicted and observed precursor species is
limited.
This section begins with a discussion of key tracer species that can be indicative of source
presence and strength at the PAMS site. Then it presents examples of both qualitative and
quantitative methods that can yield valuable information about the inventory. Since the PAMS
network is relatively new, this section will supplement PAMS-based examples with examples from
other databases that have similar data to that being collected by PAMS.
3.3.1 Examples of Indicator Species or Compounds (Tracers)
If a chemical species measured in PAMS is unique to a particular source, considerable
information can be gleaned without employing sophisticated receptor modeling techniques, e.g.,
CMB. These chemical species, unique or nearly unique to one source or source type, are referred
to as "indicator species, compounds or tracers of convenience".1 Table 3-2 identifies some PAMS
target compounds frequently used as tracers. The Chemical Mass Balance model (see Section 3.5
for examples) is a more powerful technique because it is not limited to tracers that are uniquely
associated with only one source type.
Discussion of Specific Tracers
As noted in Table 3-2, various compounds are useful in evaluating the emissions of
industrial manufacturing sources. Usually, the presence of these compounds may be determined
by discussion with the emissions inventory specialist. Benzene, propene, toluene, and ethene have
also been suggested as tracers for motor vehicle exhaust. Benzene was used extensively in
various solvent intensive industrial processes until toxicity concerns and subsequent regulations
greatly curtailed that type of usage. Benzene still falls short of being an ideal tracer for motor
vehicle exhaust since it is also present in evaporative emissions and emissions from various
combustion processes. The introduction of reformulated gasoline (RFG) in select urban areas in
1995 (as mandated by the Clean Air Act Amendments of 1990) will impact the chemical makeup
of motor vehicle exhaust. RFG exhaust and evaporative emissions speciation profiles have
already been developed and included in the 1993 SPECIATE database update, though future
Since very few emissions sources have their own distinct tracer, the scope of tracer of analysis is limited, a few
methods tor overcoming this handicap have been explored, including adjusting the tracer concentrations to account for
other sources (Yarwood 1994). Another alternative utilized (to some limited degrees of success) is the intentional
release of an "artificial" tracer into source emissions. Concentrations of this distinct artificial tracer found at a receptor,
indicate a definite non-zero contribution of emissions from the source. This option has been limited to point sources
since it would be impractical to inject tracer material at area sources. Even with point sources, though, large costly
amounts of the tracer material are required to produce detectable concentrations at the receptor (Yarwood 1994).
-------
EPA-454/R-96-006
Chapter 3
Revision Number: 0
Date: November 1996
Page. 6
refinement of the profiles is anticipated2. Carbon monoxide (CO), a criteria pollutant, is perhaps
the best of all motor vehicle exhaust tracers and should be used to supplement other tracer
information for mobile sources when available.
Butane has been noted as an excellent tracer for gasoline evaporative emissions. Butane
accounts for approximately 35% of those type emissions. Isopentane, one of the largest NMOC
components, is also regarded as a tracer for gasoline vapor. Almost all evaporative emissions are
ascribed to vehicle-related sources (U.S.EPA, 1991a).
Isoprene, a-pinene and p-pinene are the only PAMS compounds predominantly associated
with natural emissions and thus are the only available for use as tracers. All, however, are
extremely reactive. Because all three compounds are very reactive, analysis relating ambient
concentrations to source emissions often result in emissions underestimations. Unfortunately,
there are no biogenic tracer alternatives.
Toluene has been mentioned as a tracer for motor vehicle exhaust. It, like other exhaust
components, is generally present in characteristic ratios. Excess beyond this characteristic
portion, can be used as a tracer of graphic arts and surface coating processes. Both of these types
of operations commonly use solvent-based paints or inks. Propane can be used as an area-source
indicator (liquefied petroleum gas use) or a point-source indicator (oil/gas production and
petroleum refineries). Moderate to high concentrations of ethane can indicate natural gas use or
leakage but, as noted before, measurement difficulties are commonly encountered. Isobutane has
been suggested as a tracer for consumer product emissions; most aerosol products now utilize
isobutane as a propellant since chlorofluorocarbons were banned (Stoeckenius et al.).
Indicator species afford data analysts an opportunity to characterize source types from
ambient concentrations with minimum inputs. Although many PAMS target species are emitted
from multiple sources, some are typical to only one or two, like those in Table 3-2. When using
tracers to ascertain relative source emissions, one must be careful to consider the ramifications of
reactivity. Reactions breaking down tracer compounds can result in incorrect estimations of the
source's impact. Interpretation of tracer data is straightforward if the compounds are unique and
men. As mentioned above, complex software models such as CMB can be used if the tracer is
not unique or if multiple tracers are available for several sources.
Using Tracer Data
To be an effective tracer, a compound should be relatively inert. Otherwise,
photochemical reactions breaking the compound down will result in altered (usually reduced, but
possibly increased) ambient concentrations of the compound thus precipitating a mis-estimation of
- U.S.EPA. "SPECIATE database." URL: http://134.67.104.12/html/chief/spec-dn.htm. October 29,1996.
-------
EPA-454/R-96-006
Chapter 3
Revision Number: 0
Date: November 1996
Page: 7
the source emissions impact. Unfortunately, many PAMS target compounds are extremely
reactive. Thus, one option to utilize these reactive species as tracers is to use ambient
concentrations from hours when emissions are generally high and photochemical reactivity is
relatively low. The 6-9 a.m. time frame is frequently mentioned for mobile related analysis. A
second option is to consider only tracers which are relatively unreactive (i.e., lifetime greater than
eight hours).
Ambient air samples containing inert tracers can be used to estimate the concentration
contribution of all emissions (or an emission category total such as TNMOC) from the tracer
source. To compute the contribution of the total emissions P from source;', simply divide the
observed concentration of the tracer t by R, the relative proportion of the tracer component found
in the emissions at the source: P} = t} I Rr To find the concentration associated with a specific
pollutant (p), substitute r, the relative proportion of the tracer to the specific pollutant (at the
source) for R thus giving: p} = /) / rf If the indicator species is not truly unique, the preceding
formulas can be used to compute upper limits of the emissions impact. Likewise, if a tracer is
somewhat reactive, the formulae can give a lower limit to the emissions impact.
For example, a local gasoline emissions profile shows that butane accounts for 35%
(weight percent) of total non-methane hydrocarbon emissions (NMHC). An ambient monitor
recorded average butane levels of 3 //g/m3 (converted from ppbC). Hence, R = .35 and t=3. We
can therefore estimate the total NMHC associated with gasoline evaporative emissions (P) to be
about 8.6 /^g/m3 (P = 3 / .35). Since butane is also present in vehicle exhaust emissions, the
computed figure should be considered an upper bound. If the gas profile also showed that
cyclopentane (another PAMS target compound) accounted for .5 weight percent of NMHC, we
know that the relative proportion of the butane tracer to it (r) is 7 (35 / .5) and we can apportion
.4 /^g/m3 of cyclopentane to that particular source (p= 3 / 7). The computed Mg/m3 values can be
converted to ppbC using species-specific conversion factors.
3.3.2 Examples Using NMOC/NO,, Directional and Time Series Analyses.
Ratio information is often of great use for making qualitative judgments about whether a
source category or individual source is missing from the inventory, misplaced or otherwise grossly
mischaracterized. The use of ratios or single species which serve as tracer compounds for a
source category is enhanced when combined with several of the other screening tools described in
this Section. This Section also describes time series analyses and the use of wind direction
information to evaluate the direction associated with high concentrations of speciated data.
NMOC/NOX Ratios and Ratios of Other Species
As mentioned previously, problems associated with species reactivity must be addressed
before the ratio approach can be used effectively. One way is to confine the analysis to periods of
-------
EPA-454/R-96-006
Chapter 3
Revision Number: 0
Date: November 1996
Page: 8
the day in which atmospheric chemistry is minimal. This is generally at nighttime or shortly after
sunrise (e.g., 6-9 am). During such periods, one must take care to account for the lack of vertical
mixing in the atmosphere (e.g., by excluding all elevated sources from the inventory) when
comparing emission-derived and measured ambient ratios for a diverse set of sources since the
surface level monitoring site is unlikely to see any contribution from these sources at this time of
day. In general, excluding elevated point sources in calculating the emissions-derived
NMOC/NOX ratio increases the calculated ratio because elevated point sources are far more
prevalent for NOX emissions than for VOC. The second approach for discounting effects of
reactivity is to consider only ratios of pollutants which are relatively unreactive [see Altshuller,
Lewis or Carter for further information on reactivity].
Use of ratio data as a possible means for corroborating inventory estimates is illustrated in
Figures 3-2 and 3-3. Figure 3-2 plots 6-9 am NMOC/NOX ratios measured at two suburban New
Jersey sites during summer 1993 (NESCAUM). After excluding certain sources from
consideration (as described above) the emission-derived ratio can be compared to the range of
observed ratios to check for possible gross errors in the estimates. If a plot like Figure 3-2 is
comprised of data from more than one site, the data can be scrutinized to see whether there are
any apparent systematic differences in the ratios observed at different sites. For example, in
Figure 3-2, ratios at the Plainfield site appear to be somewhat higher than those measured at
Newark. This interpretation can be compared with the emission-derived estimates. If not similar,
other analyses such as looking at the ratio of a reactive to less reactive species could be explored
to see whether the difference in the measured NMOC/NOX ratios is attributable to greater
transport at one of the sites or some alternative explanation. If the findings are not consistent
with alternate explanations, this may imply a potential problem with the inventory near one or
both of the sites.
Figure 3-3 illustrates how observed ratios among species might be used as a means of
identifying presence of emissions from certain source categories (Chameides et al.). For
example, in the figure there is a high correlation observed between trans-2-pentene and cis-2-
hutene. This implies a common source for the two species is impacting the monitoring site. One
can check the inventory of nearby sources to see whether the emissions estimates are consistent
with these observations.
Directional Associations
As the name implies, this screening analysis entails subdividing air quality observations
into bins which correspond with different measured wind directions. Often eight principal wind
directions are considered. Distributions of various air quality indicators can then be constructed
for each of the wind directions. Comparing differences in the air quality distributions among wind
directions can be readily done by graphically displaying the data in a "pollution rose". Figure 3-4
illustrates the procedure using ozone as the air quality indicator (Incecik et al., 1995). The
-------
EPA-454/R-96-006
Chapter ?
Revision Number: 0
Date: November 1996
Page: 9
procedure is amenable to using a variety of air quality indicators, including individual VOC
species or ratios of species.
Use of directional associations between pollutant species or ratios enables those reviewing
the inventory to better focus on portions of the inventory which may need further verification and
refinement. For example, for a given wind direction one can compare observed distributions of
species or ratios of species with emissions estimates which are upwind from the monitoring site.
Presence or dominance of species which appear to be unaccounted for by the upwind emissions
estimates will help direct the analyst's attention to geographical locations where the inventory
may need improvements. In a similar vein, presence of pollutant ratios which seem inconsistent
with proportions of species in upwind emissions estimates would also help direct a review of the
assumptions, methodology and data base underlying the emissions estimates.
Time Series Analyses
Time series analyses of VOC, its species, NO, NO2, NOX and ratios of VOC to NOX are
potentially useful for corroborating whether assumptions in emissions models about diurnal
emissions patterns are supportable. In addition, time series plots can serve as useful guides in
helping the analyst decide which portion of the inventory to compare with ambient concentrations
at different times of day. For example, if a time series plot indicated that a source category
impacted a monitor at midday, but not at 6-9 am, this would serve as additional rationale for
excluding that source from the inventory in the 6-9 am comparison. Third, time series analyses
can, when used with other information (described later), help the analyst corroborate whether
assumptions in the emissions model about sensitivity of emissions to meteorological conditions
are consistent with observations.
Figure 3-5 shows PAMS-based time series plots on weekdays and weekends for three VOC
species (Stoekenius et al.). Recall that acetylene is a tracer for automotive exhaust. One can see
from comparing weekday vs. weekend time series for acetylene that the two time series differ.
This is consistent with a possible need to use differing diurnal activity levels for automotive
sources on weekends vs. weekdays. Figure 3-6 is a time series plot of isoprene data measured at
a suburban site in Houston (Stoekenius et al.). Note the distinctly different diurnal pattern for
isoprene vs. the ones for other species shown in Figures 3-5 and 3-6. The observed pattern is
consistent with our understanding of how the emissions factor for isoprene varies as a function of
temperature. The pattern in Figure 3-6 also shows us that it would be unwise to include biogenic
emissions (for which isoprene is a tracer) in the previously described comparisons of 6-9 am
emission derived NMOC/NOx ratios with monitored ratios.
There are many reasons why it may be useful to perform previously described analyses on
days with high ozone and contrast the results with other days. With respect to evaluating
emissions, this may provide us with insight about whether there is something different about
-------
EPA-454/R-96-006
Chapter 3
Revision Number: 0
Date: November 1996
Page: 10
emissions on days with high ozone vs. other days. Since our concern with emissions often
focuses on high ozone days (e.g., for reasonable further progress calculations and for modeled
attainment demonstrations), priorities may be greater to resolve discrepancies between emissions-
derived estimates and monitored data on these days.
The graph shown in Figure 3-7 contrasts ozone observed on an episode day vs. mean
observations for that month (Bigler-Engler et al.). Similar graphs in which other PAMS species
are plotted could be used to identify possible discrepancies in average emissions estimates (often
used in emissions models) vs. what appears to be consistent with observations on a limited
number of episode days. Figure 3-8 plots an observed relationship between midday isoprene
levels and temperatures, suggesting that inclusion of a strong temperature dependency in the
emissions factor for biogenic emissions of that species is likely to be valid (Fehsenfeld et al.).
Similar comparisons could be undertaken for tracers of other, anthropogenic sources.
3.4 EXAMPLE OF INVENTORY EVALUATION FOR LAKE MICHIGAN
INVENTORY
Data collected during the 1991 Lake Michigan Ozone Study was the basis for comparison
of the emissions inventory and ambient concentration ratios of NMOC/NOX and NMOC/CO for
Chicago, Gary, and Milwaukee (Korc et al., 1993). Comparisons of 7-9 a.m. ratios for two ozone
episodes (June 25-28 and July 16-18) showed that the ambient ratios were generally higher than
the inventory ratios. The relative individual NMOC species compositions of the ambient and
emissions inventory data were also examined. Table 3-3 shows the overall average ambient and
emissions inventory relative compositions for individual species and species groups for the three
cities during the. two ozone episodes. The ambient relative compositions of the major groups of
organic compounds are very similar at all three sites. The paraffin content in the ambient data is
about 48 percent; the olefin content is about 10 percent; the aromatic content is about 22 percent;
and the carbonyl content is about 4 percent.. The relative composition of the inventory is very
similar at all three sites. However, the paraffin composition of the emissions inventory is about 5
to 10 percent lower than the corresponding ambient data at all three sites. Olefins are slightly
higher than the ambient data at Gary and Milwaukee and are consistent with the ambient data at
Chicago. The aromatic composition is significantly higher than ambient composition at all three
sites and ranges from 27 percent at Milwaukee to 33 percent at Chicago. The carbonyl
compound composition is significantly lower than the ambient data at all three locations; and the
other species group composition is slightly higher than the ambient data at Chicago and Gary, and
is consistent with the data at Milwaukee.
In response to this study, the Lake Michigan Air Directors Consortium (LADCO)
reevaluated the emissions inventory and made several significant changes to the point, area, and
mobile source figures. Speciation profiles and background assumptions were also revised.
-------
EPA-454/R-96-006
Chapter 3
Revision Number: 0
Date: November 1996
Page: 11
LADCO then compared the revised emissions estimates to the ambient data and found improved
agreement. Tables 3-4 and 3-5 show the computed ambient and emissions NMOC/NOX ratios
both before and after the LADCO emissions inventory revision (Korc et ah, 1993).
3.5 EXAMPLES USING MULTIVARIATE ANALYSES AND CHEMICAL MASS
BALANCE (CMB)
Thus far, the focus of the discussion has been on various types of screening analyses for
PAMS data which could serve as qualitative indicators for investigating certain assumptions
underlying emissions estimates. This Section focuses on a series of analyses which has as its end
product more quantitative estimates of the contributions that various source categories make to
observed ambient measurements of ozone precursors. Although the end product of these analyses
is quantitative, it is obtained through use of much subjective judgment. Therefore, we do not
recommend using the outcome of these analyses to change emissions estimates unless the
methodology for making the emissions estimates is also reexamined and the uncertainties in these
procedures are consistent with the changes implied by the analyses of the ambient data. In short,
although the results obtained with the techniques summarized in this section are quantitative, they
should be used qualitatively to improve emissions estimates.
The Chemical Mass Balance (CMB) model (U.S. EPA, 1990a) can be used to provide
quantitative source category contribution estimates to monitored data. This procedure uses
distinct chemical species profiles for different source categories and then identifies the relative
combination of contributions from each of the selected source categories which best explains the
combination of species observed at the monitoring site. One key prerequisite for a CMB analysis
is choice of the source species profiles. There are two approaches used for choosing source
chemical profiles. The first is to select source categories and their corresponding species profiles
from available local measurements of a source's emissions or to select profiles from a "library" of
source chemical profiles. The latter approach is the simpler of the two. It may be preferred if one
already has a good idea of the source categories which are likely to make important contributions
and has confidence in the profiles which are used. The EPA's Air Emissions Species Manual
(L'.S. EPA, 1990b) documents default species profiles for many sources of VOC. When
available, species profiles measured in recent local field studies should be used instead.
Although species profiles for many types of sources are included in the Air Emissions
Species Manual, some of the data may be outdated or not applicable to the area in which we are
seeking to corroborate emissions estimates. Further, one could argue that using default species
profiles for the CMB which are the same as those underlying the inventory is not a completely
rigorous corroboration of the inventory. Thus, the second approach for choosing chemical
source profiles for use in the CMB is to try to use the ambient observations themselves to derive
-------
EPA-454/R-96-006
Chapter ?
Revision Number: 0
Date: November 1996
Page: 12
source category profiles. Methods to implement this second approach can be broadly
characterized as "multivariate analyses".
In Sections 3.5.1 and 3.5,.2, we illustrate how PAMS data are useful in applying the
chemical mass balance model and corroborating emissions estimates using examples in Atlanta ,
and Southern California. [Additional information on the CMB technique may be found in U.S.
EPA, 1987 and U.S. EPA, 1990a].
3.5.1 Example of Inventory Evaluation in Atlanta
Atlanta has been the focus of many receptor modeling studies in the past decade. CMB
has made possible a new approach to comparing emissions inventories with ambient data using
CMB calculations (Conner). This new approach make use of ranges of source estimates obtained
from the ambient data, which can be used to deal with some of the inherent difficulties of
comparing inventories with ambient data. Table 3-6 compares the CMB output with emissions
inventory data obtained from the Georgia Department of Natural Resources. The comparison is
qualified "by the fact that the CMB results reflect mostly daytime (hours 8-18) conditions, while
the inventory represents 24-hour average emissions."
As seen in Table 3-6, the biogenic mass was accounted for using several different
methods. The biogenic portion of the CMB estimate was set equal to the biogenic portion of the
emissions inventory; then it was set equal to the isoprene percent; then set equal to the
'unexplained' percent; and then also set equal to zero. The inventory highway mobile source
estimate tends to be smaller than the minimum ambient data-derived highway mobile source
estimate, and the inventory area plus point source estimate tends to be larger than the maximum
ambient data-derived estimate for the data set examined." Such a comparison is useful in
evaluating the inventory, but in this case, only limited confirmation of the inventory can be
gleaned from the CMB-based analysis; other analyses would be needed to support further
evaluation of the inventory.
3.5.2 Example of Inventory Evaluation in Southern California
The Southern California Air Quality Study (SCAQS) was conducted in summer and fall
1987 to gain a better understanding of the causes of excess pollution concentrations in
California's South Coast Air Basin (SoCAB). The SCAQS data was used to validate the use of
CMB for NHMC source apportionment (Fujita et al.). One of the goals of their study was to
reconcile source contribution estimates from the CMB with existing emissions inventory
estimates. Table 3-7 shows the mean source contribution estimates for all sites combined by
season and sampling period for the three motor vehicle source categories versus all other source
categories in comparison to the corresponding SCAQS basin wide day-specific emissions
-------
EPA-454/R-96-006
Chapter 3
Revision Number: 0
Date: November 1996
Paae: 13
inventory data for August 27, 1987 and December 10, 1987 (Fujita et al.).
Fujita et al. noted that "the larger calculated contributions of vehicle exhaust and evaporative
emissions were consistent with recent studies that suggest that the motor vehicle hydrocarbon
emissions inventories for motor vehicles have been substantially underestimated".
Fujita et al. concluded from their evaluation that "the CMB application and validation protocol
developed for PM-10 source apportionment is applicable to the validation of CMB for NMHC
source apportionment". They suggested, however, that additional source profiles be designed
specifically for receptor modeling. They noted that "the attribution of source contributions
among the motor vehicle source categories was found to be highly sensitive to the choice of
fitting species and to the relative abundance of combustion byproducts in the exhaust profile,
which vary with emissions control technology, level of vehicle maintenance, and operating mode."
They recommended the creation of site-specific vehicle exhaust profiles.
3.6 CASE STUDY - EXAMPLE OF INVENTORY EVALUATION IN HOUSTON,
TEXAS
This case study highlights a comprehensive review of the emissions inventory in Houston
TX. This example illustrates how several different techniques can be combined to build a
thorough evaluation of an inventory for ozone precursors. The techniques include: ambient and
emissions-derived NMOC/NOx; relative compositions of hydrocarbon groups using both ambient
and emissions derived data; and CMB analysis.
NMHC/NOX ratios were compared for two Houston sites in summer 1993 (Korc et al.,
1995). Emissions inventory data were delineated to 5 different grid areas surrounding the site
using two different allocation approaches. (One approach allocated emissions inventory data
from upwind grid cells with a weighting function of 1; the other allocated emissions inventory
data from upwind grid cells with a weighting function defined as the inverse of the distance
between the ambient monitoring site and the centroid of each grid cell.) As seen in Tables 3-8 and
3-9. the ambient ratios were always significantly higher than the corresponding emissions ratios
(Korc et al., 1995). At Galleria, the ambient ratios were about 2 to 6 times the emissions ratios;
at Clinton, the ambient ratios were approximately 2.5 to 4 times the emissions ratios. Galleria
represents a major urban source dominated by mobile emissions and Clinton represents a major
industrial location.
In the report, Korc also made comparisons between the relative ambient and emissions
compositions. Figure 3-9 shows comparisons of the August 19 day-specific, August 17-20
episode specific average, and August and September weekdays median 0600 CST ambient- and
August 19 emission-derived relative composition of paraffins, olefins, aromatic compounds and
-------
EPA-454/R-96-006
Chapter 3
Revision Number: 0
Date: November 1996
Page: 14
"other" species groups at Clinton (Korc et al.,, 1995). Note that the ambient-derived and
emissions inventory relative proportions of paraffins and olefins are rather comparable, the
emissions inventory estimates for aromatics is significantly higher than the ambient-derived
estimates, and the emissions inventory "other" species estimate is much lower than ambient-
derived figures, a similar chart comparing the relative individual species compositions (top 35
species) for the same four types of estimates is shown for Galleria in Figure 3-10 (Korc et al.,,
1995).
The data in Figure 3-10 show that the emissions inventory toluene and n-butane moleC
percents of NMHC were significantly higher in the emissions inventory estimate than the ambient
derived percents and the relative isopentane, ethane, and propane moleC percents were
significantly lower in the emissions inventory (Korc et al.,, 1995). Evaluating the same set of
ambient data, Lu and Fujita (1995) compared hourly CMB and emissions inventory source
contribution estimates (as percents of NMHC) for the three source categories: mobile, biogenic,
and miscellaneous. As seen in Figures 3-11 and 3-12, the mobile source emissions inventory
diurnal estimates were always lower than the CMB derived estimates at Clinton; at the Galleria
site the two sets of estimates were closer and the emissions inventory estimates exceeded the
CMB figures for 10 of the 24 hours. The biogenic emissions inventory estimates were always
higher than the CMB figures at both locations. The miscellaneous source emissions inventory
diurnal estimates for Clinton exceeded the CMB estimates except for hours 2-8 p.m. when the
two estimates were either very close or the CMB was slightly higher. At Galleria the situation was
almost reversed, with the CMB estimates exceeding the emissions inventory data all hours but 8-
11 a.m.
Based on their review, Sonoma Technologies, Inc. concluded that significant
discrepancies between the ambient data and the emissions inventory still exist. Thus, they made
the following recommendations to improve the inventory:
NMHC and NOX emissions estimates should be reviewed. Preliminary results indicate
that the ambient-derived NMHC/NOX ratios are significantly higher than the emissions-
derived NMHC/NO, ratios. These discrepancies suggest that the absolute amounts of
NMHC and/or NO, emissions were not estimated accurately. Thus, they recommended
that an investigation of the possible biases in the NMHC and NOX emissions inventories be
performed, a series of bottom-up evaluations of the major components of the inventory
are needed. These should be followed by further top-down evaluations.
Speciation profiles should be reviewed. The discrepancies between ambient and emissions
NMHC compositions and the significant overestimation of toluene in the inventory
suggest that some of the assigned organic compound source composition libraries were
not representative of some source category in the region, improper speciation profiles
-------
EPA-454/R-96-006
Chapter ?
Revision Number: 0
Date: November 1996
Page: 15
were assigned, and/or the absolute amounts of NMHC emissions were not estimated
accurately. Further, the significant overestimation of n-butane and toluene and the
significant underestimation of isopentane in the emissions inventory suggest that the
speciation profiles used for mobile sources may not reflect current gasoline reformulations
and should be reviewed.
Diurnal profiles should be reviewed. In particular, the significant overestimation of n-
butane in the emissions inventory early in the morning suggests that the diurnal profiles
used for motor vehicle evaporative emissions may not be representative and should be
reviewed.
Biogenic emissions estimates should be reviewed. In particular, the significant
overestimation of isoprene in the emissions inventory early in the morning suggest that the
biomass data, land use data, and/or the algorithm used to estimate isoprene emissions may
not be accurate and should be reviewed.
3.7 CONCLUSIONS
PAMS data can provide useful information to evaluate emissions inventories. It can
provide a general idea of the relative importance of certain compounds in the inventory that can
suggest the need for improved speciation of the data. Also, it can provide information related to
the spatial or temporal resolution of the inventory. It has been used to identify missing
components of the inventory and gross over/under calculations of the inventory based on
emissions factors.
Several techniques and analytical tools are available to evaluate the inventory. Simple
techniques such as time series analysis, diurnal patterns and pollution roses can be augmented by
Chemical mass balance and other multivariate techniques. Use of multiple techniques can provide
more useful information than relying on one or two methods. Also, PAMS data provides a
unique opportunity to compare the model-estimated species concentrations with ambient
measurements of these species at the time of day that the measurements were taken. However,
photochemical models have not yet progressed to the point where the results for individual
species are tracked along with the transformation chemistry.
Several areas of the US have used PAMS data to evaluate the emissions inventory.
Examples from the Texas Gulf Coast, Atlanta, Houston, Atlanta, Los Angeles and the Great
Lakes have been shown. Additional analyses have been undertaken in Hartford and other cities.
The real value of such analyses is just now being realized as more PAMS networks develop
validated data sets.
-------
EPA-454/R-96-006
Chapter 3
Revision Number: 0
Date: November 1996
Page: 16
3.8 REFERENCES
Altshuller, A.P. "Chemical Reactions and Transport of Alkanes and Their Products in the
Troposphere." Journal of Atmospheric Chemistry 12, 1991: 19-61.
Atkinson, R. "Kinetics and Mechanisms of the Gas-Phase Reactions of the Hydroxyl Radical with
Organic Compounds under Atmospheric Conditions." Chemical Reviews 85, 1985.
Bigler-Engler, V. "Analysis of an Ozone Episode During the San Diego Air Quality Study: the
Significance of Transport Aloft, Paper FM1-I.8." Regional Photochemical Measurement and
Modeling Studies. Volume 1: Results and Interpretation of Field Measurements VIP-48, Air and
Waste Management Association, 1995.
Bradway, R.; and Pace, T.G. "Application of Polarizing Microscopy to the Characterization of
Ambient Suspended Particulates." Proceedings of Third Annual Meeting - Federation of
Analytical Chemists and Spectroscopy Societies Philadelphia, PA, November 1976.
Carter, W.P.L. "Development of Ozone Reactivity Scales for Volatile Organic Compounds."
Journal of Air and Waste Management Association 44. 1994:881-899.
Chameides, W.; Fehsenfeld, F.; Rodgers, M.; Cardelino, C.; Martinez, J.; Parrish, D.; Lonneman,
W.; Lawson, D.; Rasmussen, R.; Zimmerman, P.; Greenberg, J.; Middleton, P.; and Wang, T.
"Ozone Precursor Relationships in the Ambient Atmosphere." Journal of Geophysical Research
97, 1992.
Conner, T.L.; Collins, J.F.; Lonneman, W.A.; and Seila, R.L. Comparison of Atlanta Emission
Inventory with Ambient Data Using Chemical Mass Balance Receptor Modeling. Presented at the
Emission Inventory: Application and Improvement Conference, Air and Waste Management
Assciation, Raleigh, N.C., 1994.
Cox, W.M. "A Workbook for Exploratory Analysis of PAMS Data." June 1995.
Fehsenfeld, F.; Meagher, J.; and Cowling, E. SOS 1993 Data Analysis Workshop Report. 1994.
Fujita, E.; Watson, J.; Chow, J.; and Lu, Z. "Validation of the Chemical Mass Balance Receptor
Model Applied to Hydrocarbon Source Apportionment in the Southern California Air Quality
Study." Environmental Science & Technology 28. 1994: 1633.
-------
EPA-454/R-96-006
Chapter 3
Revision Number: 0
Date: November 1996
Page: 17
Henry, R.; Lewis, C.; and Collins, J. "Vehicle-Related Hydrocarbon Source Compositions from
Ambient Data: The GRACE/SAFER Method." Environmental Science & Technology 28, 1994:
823.
Incecik, S.; and Thomson, D. "Surface Ozone at a Rural Site in the Nittany Valley, PA." Paper
FM1-II.7, Transactions. Regional Photochemical Measurement and Modeling Studies TR-24, Air
and Waste Management Association, 1995.
Korc, M.E.; Roberts, P.; Chinkin, L.; and Main, H. Comparisons of Emission Inventory and
Ambient Concentration Ratios of CO. NMOC and NOX in the Lake Michigan Air Quality Region.
Final Report for the Lake Michigan Air Directors Consortium, October, 1993.
Korc, M.; Jones, C.; Chinkin, L.; Main, H.; and Roberts, P. Use of PAMS Data to Evaluate the
Texas COAST Emission Inventory. Presented at the COAST Data Analysis Workshop, Austin,
T.X., 1995.
Lu, Z. and Fujita, E. Volatile Organic Compound Source Apportionment for the Coastal Oxidant
Assessment for Southeast Texas. Final report prepared for the Texas Natural Resource
Conservation Commission by the Desert Research Institute, Reno, NV., 1995.
Lewis, C.; Connor, T.; Stevens, R.; Collins, J.; and Henry, ,R. Receptor Modeling of Volatile
Hydrocarbons Measured in the 1990 Atlanta Ozone Precursor Study. Paper 93-TP-58.04,
Presented at the 86th Annual Air and Waste Management Association Conference, Denver, C.O.,
1993.
Main. H.: Roberts, P.; and Korc, M. Analysis of PAMS and NARSTO Northeast Data Supporting
Evaluation and Design of Ozone Control Strategies: A Workshop. U.S. EPA Contract
68D30030, Sonoma Technologies, Inc., July 1996.
Mayrsohn. H.; Crabtree, J.H.; Kuranoto, M.; Sothern, R.D.; and Mano, S.H. "Source
Reconciliation of Atmospheric Hydrocarbons 1974." Atmospheric Environment 11, 1977: :189-
192
Northeast States for Coordinated Air Use Management (NESCAUM), The Ambient Monitoring
and Assessment Committee. Preview of 1994 Ozone Precursor Concentrations in the
Northeastern U.S. August 1995.
Pace. T. G., editor. Receptor Methods for Source Apportionment - Real World Issues and
Applications. Air Pollution Control Association, Pittsburgh, PA, 1986.
-------
EPA-454/R-96-006
Chapter 3
Revision Number: 0
Date: November 1996
Page: 18
Pace, T.G. "Microinventories for TSP." Proceedings. Emission Factors and Inventories Specialty
Conference of the Air Pollution Control Association, Anaheim, CA, November, 1978.
Singh, H.; Jaber, H.; and Davenport, J. Reactivity/Volatility Classification of Selected Organic
Chemicals: Existing Data. U.S. EPA Cooperative Agreement CR810346-01, 1984.
Stoeckenius, I.E.; Ligocki, M.P.; Shepard, S.B.; and Iwamiya, R.K. Analysis of PAMS Data:
Application to Summer 1993 Houston and Baton Rouge Data. Draft Report. U.S. EPA Contract
68D30019, Systems Applications International, SYSAPP-94/115d. November, 1994.
U.S. Environmental Protection Agency. Comparison of Ambient NMHC/NOX Ratios with
NMHC/NOX Ratios Calculated from Emission Inventories. EPA-450/3-78-Q26. 1978.
U.S. Environmental Protection Agency. Protocol for Applying and Validating the CMB Model.
EPA-450/4-87-010, 1987.
U.S. Environmental Protection Agency. Receptor Model Technical Series. Volume HI (1989
Revision^. CMB7 User's Manual. EPA-450/4-90-004, 1990a.
U.S. Environmental Protection Agency. Air Emissions Species Manual Volume I. Volatile
Organic Compound Species Profiles, (second edition! EPA-450/2-90-001a. 1990b.
Yarwood, G.; Gray, H.A.; Ligocki, M.; and Whitten, G. Evaluation of Ambient Species Profiles.
Ambient Versus Modeled NMHC/NOX and CO/NOX Ratios, and Source-Receptor Analysis. U.S.
EPA Contract 68C10059, System Applications International, SYSAPP94-94/081. September,
1994.
-------
Emission Inventory Areas Surrounding a Site
Figure 3-1. Schematic illustration of 1-cell, 9-cell, 25-cell, and 81-cell areas surrounding a site, and the upwind
quadrants of the 81-cell area.
-------
1.0 -
a
a
8
no
Z 0-8 '
s
o>
o
^ n ji
n -
NOx Limited ""
Q
Q
»*'
jfips!
E3
&'&
i £ **wa\* *
P . >' '
,. - fi
15to1
»& Q
D *'*J
.^. '
;..:.-.
, 8to1
^ * '
1 «*
#
* * .--**"
*
*
^
f . . 4 to 1
VOC Limited
Newark, N J
a Plainfield, NJ
0.04
0.08 0.12
6 to 9 AM NOx (ppm)
0.16
0.2
Figure 3-2. Morning NMOC.'NOx Ratios at Urban and Suburban New Jersey Sites, Summer, '93
-------
Qwidore. CA
10
a
6-
4-
2.
0-
C
»
i
4-
3-
2-
1-
C
r - 0.78
_ . * '
'i .
.»'
«i
1123'
eta-2-butWM
CUndaru, O
r- -0.22
" "
!tj -: '"" --.
1234
cfc-2-but«M
1U
8
1
"
i
" 2-
o
<
1
ao
70
60
I »
1 «
30-
20-
10*
0.
OJ
r-0.78
" H " ".-'
\^^' ''
*F:r , ', ,-
112345678
eydohMn.
PrtcU. LA
r--0.05
I
. *
V
C-* . ' .
(£:'
Ktic' "
Gfc£fif*'*.'
o ia 4jo BD ax 10;
BydoncBMM
Figure 3-3. Scatterplots of the concentrations (ppbC) of trans-2-pentene, cis-2-butene, cyclohexene. 2-methyl-2-
pentene and isoprene as observed in Pride. Louisiana and Glendora, California
-------
POLLUTION ROSE
June 27 - July 06
Daytime
Figure 3-4. Daytime pollution (ozone) roses in a summer penod.
-------
Diurnal Profiles for
HOUSTON - Galieria
Weekday Weekend
Acrtyttn*
Acctyteiw
Ollllittlltl*
Bhyltne
Ethytoro
0 3 6 I « 15 II 11 W
Etnane
0 1 I I 12 tS 11 11 II
Ethane
8
I 8
I
* s -I
0 3 I I U IS II II M 0 1 I I 12 » 11 11
Figure 3-5.
-------
Diurnal Profiles for
HOUSTON - Galieria
Weekday ' Weekend
^tsoprane
Isoprane
0 1 (
Benzene
> t intsunw
Benzene
o J e i
ii n w
Toluene
o j i > tt 15 11 r
Toluene
E 8
0)111215 HUM
Figure 3-6.
-------
G ALPN observed
I ALPN mo. mean
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Figure 3-7. Hourly Ozone versus Mean, Alpine 9/26/89.
-------
10.0-
o
en
LO-
Gikft County, TN (90)
A GU»Qmnly/TN(9D
Temperature
-------
u
w
(X
y
"o
Clinton
0600 CST Composition of Species Groups
Paraffins
08/19 Emissions
08/17-20 Avg. Ambient
Olefins Aromatics
Species Groups
Other
I 108/19 Ambient
I Median Ambient (Weekday)
Figure 3-9. Houston (Clinton Site) - Comparisons of 0600 CST Ambient and Total Emissions Derived Species (imup Compostion Estimates tor the
81-Cell Area Surronding the Clinton Site, 1993
-------
Figure 3-10.
Galleria
Ambient and Emission Inventory Composition
0600 GST
Mobile Sources
-------
Figure 3-11. Houston (Clinton Site) - Comparisons Between CMB Results and Emission Inventones. August 19.
1993
Comparisons Between CMB Results md Emission Inventories
far AugM 19,1993 tt Climon
«0
50
§40
20
J.
10
A .. >
-CMB
. inventory
40
Btogenk
30
5
1
CMB
Inventory
MisccUaneoai
CMB (Liq. gu. + O«i. vtp. + Indus. * CNQ)
Invtntory (Ana + Point Miirce)
Source: Lu and Fujita, 1995
-------
Figure 3-12. Houston (Galleria Site) - Comparisons Between CMB Results and Emission Inventones. August
19, 1993
Comparisons Between CMB Results sad Ei
L Inventories
for Auimt 19,1993 « OtlteU
MobOe Soiree
CMB
Inventory
40
Btagenic
30
20
CMB
iDvcntofy
100
Miscellaneous
15
to
60
20
CMB (Uq. pi. » Ot». >«p. * Indui. f CNO)
UiYtnttxy (Amt t- Point Kwrc*)
Source: Lu and Fujita, 1995
-------
Table 3-1. Emission Factor Representativeness and Application Issues
Problem
Discussion
Unrepresentative
Factors and
Models
There are over 9,000 processes defined in FIRE, the Factor Information and Retrieval System.
but many processes are not represented for each and every pollutant. Thus, one might choose
a factor for a "similar" process without knowing whether it will be entirely appropriate for the
task at hand. Similarly, the factor may not be parametized all key process variables that could
affect emissions, e.g.. temperature, pressure, maintenance, emission control measures.
Control
Efficiency
The design criteria of air pollution control equipment affect the resulting emissions. Design
criteria include such items as the type of wet scrubber used, the pressure drop across a
scrubber, the plate area of an electrostatic precipitator, and the alkali feed rate to an acid gas
scrubber. Operation and maintenance of control devices can substantially effect emissions.
Within-source
Variability
Short-term emissions from a single specific source often vary significantly with time (i.e..
within-source variability) because of fluctuations in process operating conditions, control
device operating conditions, raw materials, ambient conditions, and other such factors.
Emission factors generally are developed to represent long-term average emissions, so testing
is usually conducted at normal operating conditions.
Variability
Between Sources
Average emissions differ significantly from source to source and, therefore, emission factors
frequently may not provide adequate estimates of the average emissions for a specific source.
The extent of between-source variability that exists, even among similar individual sources,
can be large, depending on the specific process, control system, and pollutant.
Spatial
Resolution
Often, the activity parameter data that must be input to use on emission factor or model is not
readily available on a finely gridded scale. Most emissions models grid the county-level data
based on an appropriate surrogate indicator such as population, land use, or digitized highway
data. Obviously, the better the surrogate, the more accurate the gndded emissions will be that
can be used for comparison to the ambient data. The acquisition of high-resolution surrogate
data can be very difficult and extremely costly. Ambient measurements can help determine
whether appropriate resolution has been attained.
Temporal
Resolution
Photochemical models are often run for specific dates. However, emission factors are not
intended to be representative of specific timerrames and are inherently less reliable under such
demands. Thus, it is very important to evaluate date-specific emissions data with all available
tools, including ambient data. Resolution of temporal issues is difficult because ambient
measurements are generally collected on an hourly basis and most emission inventory
estimates are defined on a daily or annual basis. Also, it is difficult to resolve emissions from
batch processes with ambient data
Speciation
Problems
A problem particular to VOC emission estimates is that the VOC emission must be speciated
by a preprocessor before it is used in a photochemical model and errors can be made that
offset the photochemistry calculations. Unfortunately, inventories only contain unreacted
species, or Ozone precursors, and current photochemical models track groups of species that
may be associated with more than one source. Thus, these comparisons can currently be
made only for unreacted (or possibly for slowly reacting) species
Missing Sources
The importance of using ambient measurements to reveal missing sources or source categories
became apparent as early as 1976 when the microscopic analysis of ambient paniculate matter
samples revealed much more soil-related panicles than were accounted for by the emissions
invpntnrv
flwav an
H Pa
-------
Table 3-2. Commonly Used Tracers for Ozone Precursors (Stoeckenius et al., 1994 a)
Compound
Numerous VOC's
a- and p-pinene
Benzene
Butane
Ethane
Isobutane
Isoprene
Propane
Toluene
Major Source
Specific industrial processes
Biogenic emissions
Motor vehicle exhausts
Gasoline evaporative emissions
Natural gas use/leakage
Consumer product emissions
Biogenic emissions
Liquefied petroleum gas (LPG) use,
refinery emissions, oil & gas production
Motor vehicle exhaust, surface coating
processes such as those involving solvent-
based paints
Comments
May be identified from emissions of various
industrial facilities
Highly reactive, measurement difficulties, also
present in consumer products such as air
fresheners
Also present in evaporative emissions and
numerous combustion processes. Useful in
developing upper bound estimate of motor
vehicle exhaust
Accounts for roughly 35% of motor vehicle
evaporative emissions
Measurement difficulties as noted for acetylene
Has replaced chlorofluorocarbons in most
consumer aerosol products
Highly reactive but most frequently used for
tracing biogenic emissions
Difficult to use as simple tracer if more than 1
type of source in area
Present in characteristic ratios in motor vehicle
exhaust; excess beyond that attributed to surface
coating processes and printing
-------
Table 3-3. Average NMOC weight percent of the individual organic species in the first LMOS emission inventon
and in ambient air at Gary, Chicago and Milwaukee. 1991.
Species
Paraffins
Acetylene
Ethane
Propane
n-Butane
i-Butane
n-Pentane
Branched C5
Branched C6
Metylcydopentane
n-Hexane
Branched C7
Cyclic C7
n-Heptane
Branched C8
n-Octane
n-Nonane
Olefins
Ethene
Propene
Terminal C4 olefins
Internal C4 olefins
Terminal C5 olefins
Internal C5 olefins
Cyclopentene
Isoprene
Terminal C6 olefins
Internal C6 olefins
Aromancs
Benzene
Toluene
Ethylbenzene
Styrene
Tnmethylbenzenes
Propylbenzenes
Xylenes
Carbon vis
Formaldehyde
Aceialdehvde
Gary
Ambient* Emissions
%NMOC %NMOC
47 39
2.5 0.8
5.7 0.7
3.9 0.2
4.0 8.4
1.9 2.7
2.8 2.3
9.5 5.7
6.2 3.4
0.9 0.5
1.4 1.6
3.1 2.1
0.3 2.3
0.6 2.4
3.0 4.7
0.6 0.3
0.3 0.3
9 11
3.1 2.7
0.8 0.8
0.8 0.7
0.9 0.8
0.3 0.9
1.1 1.6
0.4 05
1.4 1.0
0.1 0.9
0.1 1.1
22 31
2.4 2.6
7.1 146
12 1.5
06 0 1
29 26
0.5 06
73 93
4 1
24 04
16 02
Chicago
Ambient Emissions
%NMOC %NMOC
49 40
2.3 1.0
2.9 1.1
3.1 0.5
5.0 9.0
2.1 1.4
3.6 2.4
9.9 6.0
6.1 3.1
1.1 0.4
2.3 1.0
3.4 2.2
0.5 2.6
1.0 2.8
3.8 4.9
0.5 0.3
04 0.3
10 10
4.7 2.5
0.9 1.0
0.7 0.3
1.0 07
0.5 07
2.0 1.6
0.3 0.2
0.2 0.1
0.1 1.1
0.1 1.5
23 33
2.6 2.9
8.4 19.9
1.3 1 1
04 0.5
3.2 1.6
09 0.3
6.3 6.3
3 1
20 06
OQ 0 T
Milwaukee
Ambient* Emissions
<5f-NMOC "frNMOC
50 45
27 1.2
3.5 09
3.7 0.2
6.3 13.2
2.4 1.0
34 3.5
104 7.0
6.4 40
1.1 0.6
1.7 1.5
3.7 2.5
0.5 1.8
0.9 1.7
4.1 53
04 0.2
04 0.2
9 12
3.8 29
1.0 1.1
0.7 04
1.3 0.8
0.4 0.9
1.3 2.2
0.3 0.3
04 0.3
0.3 1.5
0.1 2.0
20 27
2.7 3.0
7.7 129
1.1 1.3
0.5 0.2
27 2.3
04 04
50 70
3 1
20 04
10 0 ">
% NMOC percents correspond to June 26, July 16, and July 18,1991
-------
Table 3-4.
Lake Michigan Area - Ambient Versus Original Set of El NMOC/NO,, Ratios, 1991
Site
Gary"
Chicago
Milwaukee"
Ambient
NMOC/NO,
5.3
4.8
6.4
Emission
NMOC/NOX
4.3
2.6
4.2
Ambient/El
1.2
1.9
1.6
* Ambient NMOC/NO, ratios correspond to June 26, July 16 and 18.1991
Table 3-5.
Lake Michigan Area - Ambient Versus Revised Set of El NMOC/NO, Ratios, 1 99 1
Site
Gary'
Chicago
Milwaukee*
Ambient
NMOC/NO,6
4.8
4.7
6.4
Emission
NMOC/NOX
5.0
3.6
3.8
Ambient/El
1.0
1.3
1.7
'Ambient NMOC/NO, ratios correspond to June 26, July 16and 18. 1991
Ambient NMOC/NO, with background correction
-------
Table 3-6. CMB Vs Emission Inventory Source Contribution Estimates, 1990
Summary of CMB Results*
Summary of Emission Inventory
Assume biogenic % = % reported in inventory (16.8%):
MIN. MAX.
HIGHWAY MOBILE SOURCES: 61% 75%
POINT + AREA SOURCES: 8% 23%
Assume biogenic min. % = isoprene % (1.9%):
MIN. MAX.
HIGHWAY MOBILE SOURCES: 71 % 89%
POINT + AREA SOURCES: 9% 27%
Assume biogenic max. % = % unexplained (47.4%):
MIN. MAX.
HIGHW AY MOBILE SOURCES: 41% 51%
POINT + AREA SOURCES: 5% 15%
Including Biogenics:
HIGHWAY MOBILE SOURCES:
56.0%
POINT + AREA SOURCES:
27.2%
Assume no biogenics:
MIN. MAX.
HIGHWAY MOBILE SOURCES: 73% 91%
POINT + AREA SOURCES: 9% 27%
Excluding Biogenics:
HIGHWAY MOBILE SOURCES:
67.3%
POINT + AREA SOURCES:
32.7%
* Results presented as percent of total apportioned NMOC (sum of source estimates).
-------
Table 3-7. South Coast Air Basin - CMB Vs Emission
Inventory Source Contribution Estimates, 1987
vet
period exh
CMB
0700-0800
1200-1300
1600-1700
emission inventory
0600-0800
1100-1300
1500-1700
daily total
Fall Study
CMB
0700-0800
1200-1300
1600-1700
emission inventory
0600-0800
1100-1300
1500-1700
daily total
ucle liquid gasoline
laust gasoline vapor
Summer Study
50.5
53.7
48.8
49.5
21.9
30.4
28.3
67.9
53.7
56.2
62.1
26.4
38.1
37.4
16.6
14.0
11.4
9.1
5.5
7.3
6.3
14.5
14.0
14.7
8.8
6.0
7.2
6.8
10.9
11.2
10.3
5.6
3.4
8.0
4.5
6.8
11.2
9.5
2.9
3.1
6.4
3.5
nonMV
22.1
21.0
29.5
35.9
69.2
54.3
59.0
10.8
21.0
19.6
26.2
64.5
48.3
52.3
-------
Table 3-8. August 19,1993 0500-0800 CST ambient-and emissions derived NMHC/NOX ratios for total
inventory emissions at Galleria.
Time
(CST)
5
6
7
Emission NMHC/NO,
1-Cell
1.2
1.2
2.0
9-Cell
1.4
1.4
2.4
25-Cell
1.5
1.5
2.7
81-Cell
1.7
1.5
2.6
Quadn"
2.2
1.7
2.8
Quadwb
2.0
1.7
2.7
Ambient
NMHC/
NOX
12.1
9.2
5.7
Linear combination of emission inventory data from upwind gnd cells with a weighing function of 1.
Linear combination of emission inventory data from upwind gnd cells with a weighing function defined as the inverse of the distance between the
ambient monitoring site and the centroid of each grid cell
Table 3-9. August 20, 1993 0100-0800 CST ambient-and August 19, 1993 0100-0800 CST emissions-derived
NMHC/NO., ratios for total inventory emissions at Clinton.
Time
(CST)
1
2
3
4
5
6
7
Emission NMHC/NOX
1-Cell
1.6
1.6
1.6
1.6
1.7
1.5
1.7
9-Cell
2.6
2.6
2.5
2.4
2.3
2.1
2.5
25-Cell
2.0
2.0
2.0
1.9
1.9
1.7
2.2
81-Cell
2.7
2.7
2.7
2.4
2.2
1.9
2.6
Quadn"
1.9
1.5
1.5
1.3
1.5
1.3
2.1
Quadwb
2.0
1.4
1.4
1.3
1.5
1.3
2.0
Ambient
NMHC/
NOX
10.5
11.4
8.6
7.1
5.3
4.9
4.8
Linear combination of emission inventory data from upwind gnd cells with a weighing function of 1
Linear combination of emission inventory data from upwind gnd cells with a weighing function defined as the inverse of the distance between the
ambiem monnonng site and Ihe cent/Old of each gnd cell
-------
AITKNDIX A. VOCSOIIRCF AITOR IIONMI N l/RI-CI:lyl()R MODELING STUDIES
No. / Study Name
Researchers,
Affiliation, Date
Location
1 iine
Sponsor
Approach
No. of
profiles
No. of
fitting
species
Conclusions
Comments
A Chemical Mass Balance for Volatile Organics in Chicago
O'Shea and Scheff, III
Insl of Tech. 1988
Chicago
(1 site)
1985
weekdays
12-1 pm
CMB/trajectory
3
9
MV 61% of 9 VOC species sum
no comparison to El
Wintertime Source-Reconciliation of Ambient Organics
Aronian et al , Univ III ,
1989
Chicago (3 sues)
winter
8 am - noon
EPA
CMB
8
23
MV exh 35% of NMOC; gas
vapor 5%; refineries 1 1 %
good agreement with El
except refineries higher than
El
Source Reconciliation of Ambient Volatile Organic Compounds Measured in the 1990 Atlanta Summer Study: The Mobile Source Component
Lewis and Conner, EPA
AREAL, 1992
Atlanta (1 site)
1990
30-min
diumal
EPA
CMB
1
10
M V exh 40- 1 00% of NMOC
source profile from local
tunnel study
Toxic Volatile Organic Compounds in Urban Air in Illinois
Sweet and Vermette, III.
St. Water Surv., 1992
Chicago,
E. St. Louis
1986-1990
various seasons
III. DENR
factor analysis/
CMB
6
12
MV accounts for most toxics on
average day, less on polluted day
no NMOC analysis
Receptor Modeling of VOCs in Atlanta Georgia
Kenski, Wadden, and
Scheff, Univ. III.,
Lonneman, EPA, 1992
Atlanta (2 sites)
1984-86
summer
6-9 a m.
EPA
CMB
4
29
MV exh 53% of NMOC; gas
vapor 16%
MV exh lower than El
Al
-------
AI'PKNDIX A. VOC source appnrtummcnl/icirpldi modeling studies continued
No. /Study Name
Researchers,
Affiliation, Dale
Location
lime
Sponsor
Approach
No. of
profiles
No. of
fitting
species
Conclusions
Comments
Respeciation of Organic Gas Emissions and the Detection of Excess Unburned Gasoline in ihe Atmosphere
llarley etal , Caltcch,
1992
Los Angeles (9
sites)
1986. 1987
summer
4-hr avg and
diurnal
EPRI
CMB
6
15
MV exhaust 35% of NMOC;
unburnt gas major factor missing
from El
restricted fitting species to
non-reactive VOC
Receptor Modeling of SCAQS Volatile Organic Compounds
Gertler el al., DRI, 1993
Los Angeles (2
sites)
1987 SCAQS
hourly
SCAQS
CMB
4
15-17
MV exh 60-70%
evap and refinery source
could not be resolved
A Receptor Modeling Approach to VOC Emission Inventory Evaluation
Kenski et al., Univ. Ill ,
1993
Detroit
Chicago
Beaumont
Atlanta
Washington
1984-88
summer
6-9 a m. and
diurnal
EPA
CMB
7
29
MV exh: 14% BMT; 28% Del;
41% Chi; 53% All; 56% Wash
agreement with El generally
good
Receptor Modeling of Volatile Hydrocarbons Measured in the 1 990 Atlanta Ozone Precursor Study
Lewis et al., EPA, 1993
Atlanta (1 site)
1990
30-min
diumal
EPA
CMB, source
profiles from
GRACE/SAFER
8
36
Total MV
63-80%, depending on def. of
NMOC
Receptor Modeling of Volatile Organic Compound 1 Emission Inventory and Validation
Scheffand Wadden,
Univ. III., 1993
Chicago
(3 sites)
1987
summer
8-12
EPA, NSF
CMB, trajectory
8
23
M V 2 1 % of 23-species sum;
refinery 7%; gas vapor 7%
59% unidentified; MV good
agreement with Rl
A2
-------
APPENDIX A. VOC source apportionment/receptor modeling studies - continued.
No. / Study Name
Researchers,
Affiliation, Date
Location
lime
Sponsor
Approach
No. of
profiles
No. of
fitting
species
Conclusions
Comments
Source Attribution of Toxic and Other VOC's in Columbus, Ohio
Mukund et al , Battellc,
1994
Columbus (6
sites)
1989 summer 6-9
a in
EPA
CMB
5
16
MV exh 34% of 16-species sum;
gas vapor 19%
primary focus on toxics
Validation of the Chemical Mass Balance Receptor Model Applied to Hydrocarbon Source Apportionment in the Souther California Air Quality Study
Fujila, Watson, Chow,
Lu. 1994.
CA South Coast
Air Basin (9 sites)
1987 fall
CMB
23
Comparison of Atlanta Emission Inventory with Ambient Data Using Chemical Mass Balance Receptor Modeling
Conner, Collins,
Lonneman, Seila EPA,
1994
Atlanta (6 sites)
1990 summer
EPA
CMB
3
18
El undrestimates mobile sources;
El overestimates combined area
& point
Ambient-derived source
estimates expressed as
ranges.
Volatile Organic Compound Source Apportionment for the Coastal Oxidant Assessment for southeast Texas Study (Draft)
Lee, Fjuita
Houston (2 sites)
August 1993
TNRCC
CMB
A3
-------
AITRND1X B. VOCsourie profile siudii-s
No. /Study Name
Researchers,
Affiliation, Date
Objecli>es
Location
Data
Sponsor
Approach
Conclusions
Comments
Source Fingerprints for Receptor Modeling of Volatile Organics
Scheffelal.lll Inst of
Tech , 1 989
develop VOC source
fingerprints for CMB
source profiles
from literature
EPA, NSF
lit. review
1 0 source profiles developed
Volatile Organic Compound (VOQ/parliculate Matter (PM) Speciation Data System, Version 1.4
VOC/PM Speciation
Data System, EPA, 1991
library of source
profiles
VOC and PM
source profiles
EPA
lit. review and eval.;
some new data
generated
over 700 profiles
Source Fingerprints for Volatile Non-Methane Hydrocarbons
Doskey el at., Argonne,
1992
VOC source
fingerprints
Chicago
source profiles
generated
111. DENR
ambient VOC
measurements in
impacted areas
evap profiles are sensitive to
season and grade of gasoline
The Observation of a C5 Alcohol Emission in a North American Pine Forest
Goldan et al., NOAA,
1993
characterize biogenic
emissions
Niwot Ridge,
Colorado
ambient VOC in
remote forested
area
NOAA
ambient meas.
new biogenic species,
emissions = isoprene
MV exhaust profile from
Sigsby; gas vapor profile
from winter blend
MV profiles have been
updated since 1991
used parking garage to
obtain cold start and hot
soak profiles
large meas. uncertainty
Improvement of the Specialion Profiles Used in the Development of the 1991 LMOS Emission Inventory
Korc and Chinkin, STI,
1993
review of VOC source
profiles used in El
VOC source
profiles from
EPA, CARB, and
CIT
LADCO
lit review
replace EPA profiles for gas.
evap. and surf, coating
used ambient data to
identify possible problems
with the profiles
Bl
-------
APPENDIX B. VOT source profile studies c
No. / Study Name
Researchers,
Affiliation, Date
Objectives
Location
Data
Sponsor
Approach
Vehicle-Related Hydrocarbon Source Composiiions from Ambient Data: The GRACE/SAFER Melhod
Henry, USC, Lewis,
EPA, and Conner, USC,
1994
development of VOC
source profiles for
CMB
Atlanta (1 site)
ambient NMOC
EPA
develop
GRACE/SAFER
method to extract
source profiles from
ambient data
Conclusions
develop city-specific profiles
for "roadway", whole
gasoline, and gasoline vapor
Comments
assumes acetylene is a
tracer for MV exhaust
B2
-------
APPENDIX C. AMBIENT RATIO STUDIES
Study Name
Researchers,
Affiliation, Date
Objectives
Location
Time
Data
Sponsor
Approach
Conclusions
Comments
A Review of NMOC, NOx, and NMOC/NOx Ratios Measured in 1984 and 1985
Baugues,
EPA,
1986
develop EKMA
inputs, assess MV
component
30 cities
1984-85
summer avg
6-9 am
ambient NMOC
and NO,
EPA
acetylene as MV tracer;
used NMOC:acetylene =
27 for MV
MV 18-88% of
NMOC; lowest in
Houston; higher
than El
El not adjusted for
time of day or
species meas.
Speciated Hydrocarbon and NOx Comparisons at SCAQS and Receptor Sites
Lonneman el al.,
EPA, 1989
develop EKMA
inputs
Ixis Angeles,
Long Beach,
Claremont
1987 summer (5
days)
6-9, 12-3, and 3-
6
ambient NMOC
and NO,
(SCAQS)
EPA
ambient NMOC:NOy at
source and receptor sites
NMOC:NOy lower
at receptor site
no comparison to El
NMOCNO,
Reconciling Differences Between Ambient and Emission Inventory Derived NMOC/NOx Ratios: Implications for Emission Inventories
Baugues, EPA,
1991
^
El validation
I6ci!ies
1985
ambient NMOC
and NO,
EPA
adjust El to provide best
comparison with ambient
NMOC:NO,
El ratios average
23% lower than
ambient
rule effectiveness
assumptions have
large effect
Comparison of Emission Inventory and Ambient Concentration Ratios of CO, NMOG, and NOx in California's South Coast Air Basin
Fujita et al.,
CARB, 1992
El validation
Los Angeles (8
sites)
1987
summer, fall avg.
7-8 am.
SCAQS
ambient NMOC,
CO, NO,
CARB
comparison of ambient
and El NMOC:NO, and
CO:NO, ratios
ambient ratios higher
than El; MV emis.
underest.
specialed VOC data
used qualitatively
Comparison of Emission Inventory and Ambient Concentration Ratios of NMOC, NOx, and CO in the Lake Michigan Air Quality Region
Korc et al., STI,
1993
El validation
Chicago, Gary,
Milwaukee
1991
summer
7-9 a.m.
LMOS ambient
NMOC, CO,
NO,
LADCO
ambient NMOC:NO, and
CO NO, ratios and VOC
mass fractions
ambient ratios higher
than El; MV emis
underest.
speciated VOC
compared to
speciated El
Cl
-------
APPENDIX C. Amhicnl rjlm
Study Name
Researchers,
Affiliation, Date
Object! \es
Locution
Time
Data
Sponsor
Approach
Conclusions
Comments
Use ofPAMS Data to Evaluate the Texas Coast Emission Inventory
Korc et al. STI.
1995
/./ validation
Southeast
leias
1993 summer
l-8am
COAST
ambient NMOC,
NO,
EPA
comparison of ambient
and El NMHC:NO,
ratios
C2
-------
EPA-454/R-96-006
Chapter 4
Revision Number. 0
Date: November 1996
Page: 1
CHAPTER 4
OBSERVATIONAL BASED METHODS FOR DETERMINING VOC/NOX
EFFECTIVENESS
4.1 INTRODUCTION
Observational based models (OBMs) represent a broad group of data analysis techniques,
subject to many descriptions regarding their formulations and applications. Perhaps the
distinguishing feature of OBMs is that they are driven principally by observed (or ambient) data as
opposed to grid models which are driven by emissions estimates. Observational analysis methods
include receptor models, regression techniques, ambient ratios (e.g., VOC/NOJ, indicator species
and the more semi-empirical based OBMs. This chapter is restricted to those methods which are
capable of inferring control strategy effectiveness (e.g., NOX or VOC control preference). Source
attribution and related methods for assessing emissions inventories are covered in Chapter 3.
This chapter follows a method (or method class) by method approach, starting with highly
empirical1 ratio techniques and proceeding to the more semi-empirical OBMs which incorporate
some degree of mechanistic-based chemistry formulations. All of the techniques discussed require
"routine" data sources comparable to that available from a PAMS network, with certain
exceptions in NOX measurements. Each methodology section provides case examples taken from
the literature or developed from available data sources.
4.2 EMPIRICAL TECHNIQUES
4.2.1 VOC/NOX Ratios
Ratios based on the concentrations of VOC to NOX have been used for relating emission
inventories to ambient data, as well as delineating NOX and VOC control requirements.
Generally, the propensity of an area to be VOC- or NOx-limited is based on the measured (or
predicted) ratio. Higher and lower ratios imply NOX- and VOC-limiting conditions, respectively.
No specific cutoff ratios exist, since whether ozone formation is limited by VOC or NOX depends
on other factors in addition to the VOC/NOX ratio. However, for the conditions often assumed in
EKMA/OZIPM4 ozone isopleth diagrams ratios greater than 16 often were considered NOX-
limited and ratios less than 6 were considered VOC-limited (U.S. EPA, 1989). The VOC/NOX
ratio technique is based on well understood atmospheric chemistry processes, and represented in
All approaches have logical foundations in, or their behavior can be explained by,
atmospheric chemistry principles.
-------
EPA-454/R-96-006
Chapter 4
Revision Number: 0
Date: November 1996
Page. 2
ozone isopleth diagrams (Figure 4-1) derived from photochemical "box-type" models such as
EPA's EKMA/OZIPM4 (U.S. EPA, 1989).
As part of the evaluations of 1993 PAMS data from three sites in Houston, Texas and
Baton Rouge, Louisiana, ratios were calculated from three hour averages of TNMHC and NO,,
and TNMHC and CO for the three PAMS sites2. The summary statistics for the ratios are
presented in Table 4-1 (Stoeckenius et al., 1994). For Houston, the TNMHC/NOX ratios were
found to be higher at the Clinton Drive than at the Galleria site, as might be expected given the
dominance of petrochemical industry sources at Clinton Drive. The ratios at the Baton Rouge site
are lower than at either Houston site. In contrast, morning (6-9 a.m.) ratios are very similar at all
three sites. The Clinton site appears subject to higher evaporative emissions and fewer NOX
sources during the day as compared to Galleria or Capitol.
Relationships between morning TNMHC/NOX ratios and daily maximum ozone
concentrations may provide insight into the potential tradeoffs between VOC and NOX control
strategies. Scatter plots of 6-9 a.m. TNMHC/NO, ratios with daily maximum ozone
concentrations observed within the Houston non-attainment region and daily maximum
temperature (as measured at the PAMS site) are presented in Figures 4-2 and 4-3 (Stoeckenius et
al., 1994). At these sites, ozone concentrations above 0.10 ppm are associated exclusively with
TNMHC/NO. ratios less than about 10. There is some hint in these data that the TNMHC/NO
* x
ratio decreases with increasing ozone concentrations above 0.10 ppm. If this is confirmed by
further analysis, it would suggest that Houston is in a NOX limited regime.
Nearly all of the TNMHC/NOX ratios above 10 at the Houston sites are associated with
maximum ozone values substantially below 0.10 ppm. TNMHC/NOX ratios substantially above 10
may be associated with older air masses with very low NOX concentrations and a lower proportion
of reactive NMHCs\ This could be confirmed by further analyses comparing the reactivity
weighted TNMHC on days with different TNMHC/NOX ratios and a more detailed examination of
meteorological conditions. Figures 4-2 and 4-3 suggest that the meteorological conditions that
place these aged air masses over the Houston site on certain mornings are not conducive to the
formation of high ozone concentrations in the Houston area (Stoeckenius et al., 1994). As is
typical of most urban areas, the daily maximum ozone concentrations in Houston are related to
the daily maximum temperature with the highest ozone found on the warmest days. There are no
obvious relationships between maximum temperature and 6 to 9 a.m. TNMHC/NOX ratios.
"Ratios were set to missing for 3-hour blocks with two or more missing TNMHC or NO, concentrations.
High TNMHC/NOX ratios may also be associated with very low NO, concentrations that are not accurately
measured. Investigation of this possibility was beyond the scope of the present study but must be conducted before any
final conclusions can be drawn regarding the conditions leading to high TNMHC/NO,, ratios.
-------
EPA-454/R-96-006
Chapter 4
Revision Number: 0
Date: November 1996
Page: 3
While the VOC/NOX method is theoretically sound, application of the technique has
several limitations:
1. Historically, applications have relied upon morning, center-city VOC and NOX
measurements, yet the ratio varies widely in time and space. PAMS improves the
spatial and temporal coverage of data, and therefore tempering this particular
concern.
2. Assuming only limited measurement-related difficulties, the ratios delineating NOX
and VOC-limited regimes vary with time and location, and are affected by vertical
mixing processes that often are not accounted for in surface measurements.
Additionally, the prevailing atmospheric chemistry (e.g., composition and age of
air mass) can impart different control responses at the same VOC/NOX ratios.
3. Inconsistent and uncertain measurement techniques affect the ratio. These include
various interpretations of total NMOC, measurement uncertainties and artifacts in
NOX and NMOC, and the representativeness of observations (this latter issue is
more problematic for emission inventory evaluation).
By themselves, VOC/NOX ratios probably cannot be used unambiguously to infer NOX or
VOC control strategy effectiveness. However, in combination with other observational (and
gridded models) techniques, the VOC/NOX method adds corroborative value.
4.2.2 Reactive (oxidized) Nitrogen (NOV, NOJ and Ozone Correlation Techniques
Several correlations relating ozone production to total oxidized nitrogen (NOy) or NOZ
(NO, - NO,) have been suggested as tools capable of implying VOC/NOX control effectiveness, as
well as relative air mass aging. These correlations (e.g., Figure 4-4, Olszyna et al., 1994) are
generally based on large quantities of averaged data covering multiple days, and limited to the
midafternoon time period (1-5 pm).
NO7 typically correlates better with ozone than NOy. This may reflect ozone scavenging
by one of the NO% species (NO). Progressively higher slopes, AO3/ANOZ, reflect NOX limited
regimes and negative or near zero slopes reflect VOC limiting conditions. The existing studies do
not provide sufficient guidance to delineate clearly the range of slopes depicting NOX or VOC
limiting regimes, although slopes greater than 8 appear to suggest NOx-limiting conditions. By
themselves, the NOZ-O, correlation does not provide a reliable NOx-VOC-limiting indicator.
However, in combination with other observational approaches and photochemical modeling, the
correlation can provide corroborative support for evaluating control effectiveness. Future
applications with emerging data bases and cross analysis-method comparisons should provide
-------
EPA-454/R-96-006
Chapter 4
Revision Number: 0
Date: November 1996
Page: 4
greater understanding of the utility of these correlation methods. Additional discussion on these
techniques is found in Appendix A.
4.3 OBSERVATIONAL MODELS
The two semi-empirical OEMs discussed here, the Mapper-Smog Production Algorithm
(SPA) and the Georgia Institute of Technology (GIT) model, rely on some level of mechanistic
description of atmospheric chemistry (relative to correlation techniques) or produce explicit VOC
and NOX control effectiveness results and therefore are distinguished from other indicator and
correlation techniques.
4.3.1 Smog Production Algorithm - MAPPER program
Figure 4-5 provides an example of MAPPER-SPA results. The map provides a visually-
oriented time and space perspective of NOX- (or VOC-) limitation. The circular elements are
clocks reflecting hourly data at the monitoring site, the size of the diameter is in proportion to the
maximum daily ozone observed at the site. Shading gradations are used to reflect extent, E,
which is a indicator for degree of NOx-limitation. Filled black reflects full extent of 1, situations
clearly NOx-limited; white reflects VOC limiting conditions with E < 0.5; and cross-hatched areas
could be considered borderline NOx-sensitive. The example is only illustrative, the shading forced
to illustrate a full range of NOX- and VOC-limiting conditions. Displays such as these can be used
to corroborate control strategy results from gridded ozone models. Due to the potential for
compensating errors, typical model evaluations limited to comparisons between observed and
simulated concentrations do not necessarily provide confidence in the model's response to
emissions perturbations. The use of OEMs which rely on observations provide a unique
corroborative check on the directional ability of EBMs to respond to emissions changes.
Description
The Smog Production Algorithms (SPA) and associated MAPPER software were derived
from the Integrated Empirical Rate (IER) model of Johnson (1984) and later developed by
Blanchard et al. (1994). The original IER was based on smog chamber experiments in Australia.
Johnson (1984) defined smog produced (SP) as:
SP(t) = O3(t) - O3(0) + NO(0) - NO(t)
In simple terms, SP represents the cumulative oxidation of initial NO into ozone and other
oxidized "smog" products (i.e., NOy species), presumably accounted for in the change in NO over
time. With sufficient NO, SP exhibits a linear relationship with cumulative light flux until a
maximum, SP^, is reached. Over much of this linear period, SP can be thought of as being
-------
EPA-454/R-96-006
Chapter 4
Revision Number: 0
Date: November 1996
Page: 5
VOC-limited. This maximum is related to initial NOX in the system by,
SP^ = P[NOX(0)],
where the parameter, P, was assumed to be a constant of 4.1 based on the chamber experiments
and reflects the potential maximum amount of smog produced per unit of NOX input. An extent
of Reaction, E, is defined as:
E(t) = SP(t) / SP
where the extent, E, represents the fractional movement toward maximum smog production.
When the extent =1, virtually all the NOX in the system has been transformed to ozone and
oxidized NOX products (NOZ) and the system is NOX limited. Recalling that the linearized
region of SP(t) is associated with VOC-limiting conditions, the value of E can be interpreted as
rough indicator of VOC or NOx-limiting conditions. That is, E with values very near or equal to
1 clearly reflect NOx-limiting conditions, and values of E far less than 1 (say less than .5) suggest
VOC limiting conditions. Thus, the derivation of equations and associated algorithms which
utilize measurements as independent variables directed toward the calculation of E forms the basis
for MAPPER in delineating NOX and VOC limiting conditions. Details on the derivations and
algorithms for E are found in Blanchard et al. (1994). More recent forms for SP include:
SP(t) = O3(t) + DO3(t) - O3(0) + NO(0) - NO(t),
and
SP = P[NO,(0)]C
where the term D0,(t) accounts for the cumulative ozone lost to deposition from time zero to t
(i.e., ozone that has been produced but deposited), and is parameterized as a function of observed
ozone. Default values for P (19) and a (.67) are empirical coefficients derived from numerous
smog chamber studies.
Measurement Requirements/Relation to PAMS
MAPPER algorithms require hourly measurements of ozone, NO and either NOX or NOy,
depending on the form of the MAPPER algorithm selected. Where available, true NOX or true
NO,, data should be used in MAPPER applications. The scarcity of such data will drive many
applications toward routine NOX data. Since MAPPER operates in true NOX and true NOy
modes and many of the available routine NOX overestimate afternoon NOX, the technique can
provide a "bounding" associated with the positive NOX measurement biases. In addition, PAMS
locations generally are weighted toward urban core areas and the results from PAMS alone are
likely to be skewed toward the urban perspective, which in many cases will infer a VOC limiting
-------
EPA-454/R-96-006
Chapter 4
Revision Number: 0
Date: November 1996
Page: 6
case.
4.3.2 GIT Model
The Georgia Institute of Technology (GIT) OEM (Cardelino and Chameides, 1995)
utilizes most of the data (speciated NMOC and NOX) generated from a PAMS. The GIT-OBM
quantifies the relative roles of various emission groups (e.g., natural and anthropogenic VOC,
NOJ on ozone production. Figure 4-6 illustrates a GIT-OBM application for Atlanta covering
several monitoring locations.
Description4
Cardelino and Chameides (1995) describe the use of a box model for calculating the
sensitivity of ozone to VOC or NOX reductions, which they name the "Observation-Based Model"
and is referred to here as the "GIT-OBM". The model is an OBM because it uses ambient
concentrations rather than emissions estimates to drive the calculations. The calculation is carried
out separately for each monitoring location. Unlike a trajectory model, each box is fixed at the
location of its monitor. The OBM utilizes some features of the OZIPM4 model to account for
dilution and employs a modification of the CBM-4 (Cardelino and Chameides, 1995).
Cardelino and Chameides (1995) define a quantity, PS03.NO, which is the net ozone formed
plus the net NO consumed over a 12-hour period (this quantity is similar to SP of the EER model
but is computed as an integral over time rather than an instantaneous concentration). The
fractional change in PS03-NO divided by the fractional change in the "source strengths" of
precursors, are used to define relative incremental reactivities (RIRs). For each measured species,
instantaneous source strengths are calculated from the measurements and from production and
loss terms. The RIRs for each site are (or can be) averaged to generate area-averaged RIRs.
Cardelino and Chameides (1995) sum RIR terms so as to yield RIRs for NO, anthropogenic
hydrocarbons (AHC), and natural hydrocarbons (NHC). The split between RIR-AHC and RIR-
NHC is accomplished by summing RIRs for species arising from anthropogenic and biogenic
emissions, respectively.
Limitation/Caveats
Numerous assumptions are made in the procedure. Some of the assumptions that appear
to have a potentially substantial effect on the calculations are provided in Dmerjian et al. (1995)
who suggest a potential bias toward overestimating the benefits of anthropogenic NOX reductions.
1. Nature and levels of uncertainty associated with the method. Analyses of uncertainty have not
This description as well as the section discussing limitations and caveats of the GIT-OBM are taken from a
review of OBMs conducted by Demerjian et al. (1995).
-------
EPA-454/R-96-006
Chapter 4
Revision Number: 0
Date: November 1996
Page: 7
been carried out for the OEM. Taking into account the uncertainties deriving from factors such
as those listed in the preceding section, as well as the inaccuracies and unavailability of data for
typical nonattainment cities, there is a clear need to quantitatively characterize the resulting
overall uncertainties in the model output. At present, some of the assumptions appear to generate
biases that would enhance the apparent benefits of controlling anthropogenic NOX.
2. The type of output of the method and the degree of consistency between this output and the
types of information needed for regulatory applications. The output consists of RIRs for NO,
AHC, and NHC. RIRs can also be output for anthropogenic area and point sources, as well as for
other disaggregations that would be of use. The form of the output, as RIRs, can be directly
translated into qualitative control preferences. However, RIRs derived from 12-hour Ps03.No
terms may not be appropriate given the present form of the ozone standard, which requires
compliance with a one-hour ozone average.
3. Availability of data needed as input to the method. The method requires measurements of NO
that are accurate at sub-ppbv concentrations (Cardelino and Chameides, 1995), which appears to
exceed the capabilities of most instrumentation that has been, or will be, deployed in routine
monitoring networks. The OEM also appears to require continuous gas chromatograph (GC)
measurements of hydrocarbon species.
-------
EPA-454/R-96-006
Chapter 4
Revision Number: 0
Date: November 1996
Page: 8
4.4 REFERENCES
Blanchard, C; Lurmann, F.; Korc, M.; and Roth, P. The Use of Ambient Data to Corroborate
Analyses of Ozone Control Strategies. U.S. EPA Contract number 68D30020. 1994.
Cardelino, C.; and Chameides, W. "An Observation-based Model for Analyzing Ozone Precursor
Relationships in the Urban Atmosphere." Journal of Air Waste Management Association 45,
1995: 161-180.
Demerjian, K.; and Roth, P. A New Approach for Demonstrating Attainment of the Ambient
Ozone Standard: Modeling. Analysis, and Monitoring Considerations. U.S. EPA Purchase Orders
EFA035 and EFO036, ENVAIR, San Anselmo, CA. July, 1995.
Johnson, G. "A Simple Model for Predicting the Ozone Concentration of Ambient Air."
Proceedings of the 8th International Clean Air Conference Melbourne, Australia, May 2, 1984:
715-731.
Milford, J.; Gao, D.; Sillman, S.; Blossey, P.; and Russell, A. "Total Reactive Nitrogen (NOy) as
an Indicator of the Sensitivity of Ozone to Reductions in Hydrocarbon and NOX Emissions."
Journal of Geophysical Research 99. 1994: 3533-3542.
National Research Council. Rethinking the Ozone Problem in Urban and Regional Air Pollution.
National Academy Press, Washington, D.C., 1991.
Olszyna, K.; Bailey, E.; Simonaitis, R.; and Meagher, J. "O3 and NOy Relationships at a Rural
Site." Journal of Geophysical Research 99. 1994: 14,557-14,563.
Sillman, S. "The Use of NOY, H2O2, and HNO3 as Indicators for Ozone-NOx-Hydrocarbon
Sensitivity in Urban Locations." Journal of Geophysical Research 100, 1995: 14175-14188.
Sillman, S.; Logan, J.; and Wofsy, S. "The Sensitivity of Ozone to Nitrogen oxides and
Hydrocarbons in Regional Ozone Episodes." Journal of Geophysical Research 95, 1990: 1837-
1851.
Sillman, S.; and He, D. The Use of Photochemical Indicators to Evaluate Oxidant Models: Case
Studies from Atlanta and Los Angeles. Presented at the 9th Joint Conference on the Applications
of Air Pollution Meteorology with the Air and Waste Management Association, 1996.
-------
EPA-454/R-96-006
Chapter 4
Revision Number: 0
Date: November 1996
Page: 9
Stoeckenius, T.E.; Ligocki, M.P.; Shepard, S.B.; and Iwamiya, R.K. Analysis of PAMS Data:
Application to Summer 1993 Houston and Baton Rouge Data, Draft Report. U.S. EPA Contract
68D30019, Systems Applications International, SYSAPP-94/115d. November, 1994.
Trainer, M.; Parrish, D.; Buhr, M.; Norton, R.; Fehsenfeld, F.; Anlauf, K.; Bottenheim, J.; Tang,
Y.; Wiebe, H.; Roberts, J.; Tanner, R.; Newman, L.; Bowersox, V.; Meagher, J.: Olszyna, K.;
Rodgers, M.; Wang, T.; Berresheim, H.; Demerjian, K.; and Roychowdhury, U. "Correlation of
Ozone with NOY in Photochemically Aged Air."Journal of Geophysical Research 98, 1993:
2917-2925.
U.S. Environmental Protection Agency. Procedures for Applying City Specific EKMA. EPA-
450/4-89-012, 1989.
-------
EPA-454/R-96-006
Chapter 4
Revision Number: 0
Date: November 1996
Page: 10
Appendix 4A: DISCUSSION OF NITROGEN-BASED CORRELATION
TECHNIQUES
Definitions:
NOX = NO + NO2
NOy= NOX + PAN + HNO3 + ORGANIC-N plus AEROSOL-N
NOZ = N0y - NOX
Many of the observational based analysis methods applied over the last decade have relied
on high quality NOX and NOV measurements. Several recent studies, designed to characterize
airmass aging and the relationship of in-situ ozone production to NOX or NOy in rural
environments, have produced strong correlations between ozone and various nitrogen groupings
(Trainer et al., 1993; Olszyna et al., 1994). These correlations (e.g., Figure 1) generally are
based on large quantities of averaged data covering multiple days, and limited to the mid
afternoon time period (1-5 pm), to filter out the overwhelming negative impact of fresh NOX
titration of ozone. For the same reasons, NOz-Oq regressions typically correlate better than NOV.
Ozone should correlate positively with NOZ under NOX and VOC-limiting conditions, since the
radical processes associated with VOC reactions during VOC-limiting conditions coincidentally
produce ozone and oxidized nitrogen (NOJ products such as nitric acid. Collectively, the
various studies as well as independent analysis of photochemical model results (Sillman, 1996)
suggest that the slope of the regression line can be used as a qualitative indicator for delimiting
NOX and VOC limiting regimes. Progressively higher slopes, AO?/ANOZ, reflect NOX limited
regimes and lower slopes reflect a move towards VOC limiting conditions. The existing studies
do not provide sufficient guidance to delineate clearly the range of slopes depicting NOX or VOC
limiting regimes, although slopes greater than 8 appear to suggest NOx-limiting conditions. By
themselves, the NO?-O3 correlation does not provide a reliable NOx-VOC-limiting indicator.
However, in combination with other observational approaches and photochemical models, the
correlation can provide corroborative support for evaluating control effectiveness. Future
applications with emerging data bases and cross analysis-method comparisons should provide
greater understanding of the utility of these correlation methods.
In addition to providing insight on control strategy effectiveness, the interpretation of
NOX, NO, and NOZ data can be interpreted to provide insights into air mass aging, ozone
production efficiency, and other analyses. Air mass aging can be defined in many ways, but
empirical indicators using NO, and NOZ data are founded on basic, well-understood atmospheric
chemistry principles. In simple terms, aging is reflected in observations by the relative amount of
NOX that is oxidized (aged) to various NOX oxidation products (i.e., NOZ). The ratio, NO/NOy, is
one indicator of air mass aging, and typically is normalized on a range from 0 to 1, with .6 used as
a suggested indicator of "photochemically aged" air (Trainer et al., 1993):
-------
EPA-454/R-96-006
Chapter 4
Revision Number: 0
Date: November 1996
Page: 11
AGE = 1 - (NO,/NOV)
Ozone production decreases with aged air masses, and the NOx-limited conditions almost always
associate with aged air masses. Aging analyses can provide guidance as to when to expect good
NOz/NOy/O, correlations; poor correlations should be expected with "fresh" air masses. In turn,
one might expect "good" correlations between VOC (or selected groups, species) and ozone
during periods which are not NOx-limited, and poor (or coincidentally good) correlations during
NOx-limited conditions. Again, the sum value of the collective use of several analysis techniques
across multiple species is greater than the sum of individual analyses.
Data limitations/Relation to PAMS. PAMS will result in an enormous expansion of the VOC data
base, and significant, but inadequate, improvements in nitrogen measurements. PAMS introduces
additional spadal and temporal coverage of traditional NOX measurements. However, many of the
NOX instruments are not capable measuring sub-ppb levels. More troubling is the lack of true
NO2 or NOy measurements. Many of the PAMS NOX instruments overestimate NO2 (and
therefore NOX as well), due to PAN and nitric acid interferences. This interference is stronger
during daytime conditions, further compromising the utility of nitrogen-based data analysis
techniques. NOy measurements are not required in the PAMS program. Finally, the majority of
currently operating PAMS sites (mostly Type II) reflect urbanized conditions with relatively fresh
air masses. One would expect these locations to produce poor correlations and infer VOC
limiting conditions.
Model simulation derived methods
Various indicator techniques, which utilize absolute values of selected species or species
ratio values, have been developed by analyzing photochemical model simulations. These include
NON (Milford et al., 1994; Sillman, 1995) along with nitric acid, HNO,, hydrogen peroxide, H2O,
and formaldehyde, HCHO. Results from two different photochemical models suggested that
ozone would be limited by VOC at ambient NOy levels exceeding a threshold level ranging from
10 to 25 ppb, and NOX controls would not result in increased ozone unless ambient NOy levels
exceeded 20-30 ppb (Milford et al., 1994). Sillman (1995) reported similar results for NOy, but
concluded that the ratios, HCHO/NO, < 0.28 and H2O2/NOy <0.4 were effective indicators for
VOC limited regimes.
-------
EPA-454/R-96-006
Chapter 4
Revision Number: 0
Date: November 1996
Pase: 12
Relation to PAMS. Although these studies are based on model simulations, they provide
direction toward the types of measurements useful for inferring control strategy directions. The
potential value of NOy measurements is reinforced by these studies. As discussed above, NO,
measurements are not required by PAMS but are considered an incremental improvement that
gradually could be incorporated in the program. In combination with HCHO, which is measured
in PAMS, another possible corroborative indicator, HCHO/NOy, could be available. Hydrogen
peroxide is still considered a "research" level measurement. However, peroxide concentrations
gradually are becoming standard components of field campaigns such as the SOS. In addition to
an indicator application described here, peroxide measurement are extremely useful for model
diagnosis and evaluation and should be given strong consideration as a PAMS species when and if
monitoring technology allows for more routine H2O2 measurements.
-------
0.2B r
0.24
0.20
I 0.16
o.
Q" 0.12
0.08
0.04
-VOC
LIMITED,
O, (ppm) = 0.08 0.16 0.24 /0.34
0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
VOC (ppmC)
Figure 4-1. Isopleths developed from EPA's EKMA to illustrate VOC/NOx ratios. ( NRC, 1991).
-------
Galleria (3hr Avgs.)
1 _i I
TNMHC/NOx
0.05 0.15
i i i
in
CM
o
CM
. O
in
d
8-1
6
*
?
Max. O3
(ppm)
. %
* *
*
i
Max. Temp.(C)
5 10 15 20 25 30
CO
CO
00
CM
CM
26 28 30 32 34 36
Figure 4-2. Scatter plot matrix for 6-9 a.m TNMHC/NO, raio, daily maximum temperature at Gallena, TX site
and regional daily maximum ozone
-------
Clinton (3hr Avgs.)
0.05 0.15
I I
I t _ t
TNMHC/NOx
.
* r
* .:
M *** *
* *
*%**.
.
* t
I
*
*
'HI:
s
. o
o
«
-s
_ O
in
d
in
o.
*
* *
*** .
;:- ""-.
Max. O3
(ppm)
*
. ' i
..'!"!*
. - . : i
*
V
*
*
0
"
. .. .
*
*
M .
m ^
1
Max. Temp.(C)
-S
-5
-S
-8
.to
CM
CD
' eg
10 20 30 40 50
26 28 30 32 34 36
Fipure 4-3. Scatter plot matnx tor 6-9 a.m TNMHC/NO, raio, daily maximum temperature at Clinton Dnve, TX
site and regional daily maximum ozone
-------
.Q
Q.
CD
c
o
N
O
0
1
NOz (ppbv)
Figure 4-4. Plot of ozone versus NOz for Giles County, TN (Olszyna et al., 1994).
-------
NORTHEASTERN U.S.
O1 Jul 95
SCALE
181 9km
EQUATION Original G Johnson ( alpha-1 00. twta-4 10. O3(0)-40 0 no NOx correction )
PEAK OZONE
ISO ppb'1 i 100 ppb , ! 150 ppb I ; 200 ppb I | 250 ppb i
9am ,
noon
EXTENT SHADING
Insufficient Data Blank
O - .50 White
51 - BO Light Gray
time of peak ozone .81 - 95 Medium Gray
6Pm 96 - 1.0 Black
, 3pm
Figure 4-5. Example MAPPER output for the Northeast U.S.
-------
Multiple Day Analysis
Mars Hill
GaTecfa
ML King
FortMcPh
Tucker
DekaJb
0.0
Figure 4-6. Results from GIT-OBM applied to Atlanta. The model provides a relative assessment of the role of
emissions groups on ozone formation. ( Cardelmo and Chameides, 1995).
-------
TABLE 4-1. Summary statistics for TNMHC/NO., ratios calculated from 3-hour average TNMHC and NOX
concentrations.
Min.
1st Qu.
Median
Mean 3rd Qu.
Max.
S.D.
NA's
N
All 3-Hour Values
Galleria
Clinton
Capitol
Galleria
Clinton
Capitol
4.457
3.857
2.475
4.457
3.857
3.738
9.531
11.2
6.927
7.367
6.751
7.558
13.72
19.22
9.562
9.524
9.522
10.2
16.94
22.76
12.19
6-9 a.m.
11.78
12.91
12.51
19.99
29.86
14.48
Averages
14.4
16.47
14.66
94.25
141,2
82.6
29.22
50.26
45.28
11.165
15.648
9.536
6.166
9.187
7.798
218
297
181
28
29
22
736
736
739
92
92
92
-------
EPA-454/R-96-006
Chapter 5
Revision Number: 0
Date: November 1996
Page: 1
CHAPTERS
QUALITY ASSURANCE
5.1 INTRODUCTION
The quality and applicability of PAMS data analysis results are directly dependent on the
inherent quality of the raw data itself. Data assessment information, such as that obtained from
precision and accuracy (P&A) checks and performance audits, provides valuable measures of the
general quality of PAMS data submitted to AIRS. Although reporting organizations and EPA are
employing increasingly rigorous validation measures to insure optimum data quality, errors still
get through the system. Because of the serious implications PAMS analytical results convey,
PAMS data users are advised to critically examine all data before undertaking analysis in earnest.
In this chapter we will highlight some recent PAMS data quality assessment information and also
illustrate some useful screening procedures being used to identify potential errors that could bias
results.
5.2 DATA ASSESSMENT
The assessment function of PAMS quality assurance involves two key required
components: the National Performance Audit Program (NPAP) and precision and accuracy (P &
A) data. EPA's Quality Assurance guidance mandates that all data collected for regulatory or
research purposes be of known and documented quality. The EPA uses the National Performance
Audit Program to independently quality assure the PAMS monitoring data it is receiving and
permanently storing on AIRS. Audits for the PAMS compounds were added to the NPAP in
1995. Proficiency studies undertaken prior to the NPAP audits provided input to the program.
Precision and accuracy checks are required for all types of PAMS monitors (meteorological,
ozone, nitrogen oxides, VOC, and carbonyl).
5.2.1 NPAP and Proficiency Studies
In 1993 and 1994, the U.S. EPA National Exposure Research Laboratory (NERL),
formerly, the Atmospheric Research and Exposure Assessment Laboratory, conducted
cooperative efforts with the 22 State and local agencies monitoring for the PAMS compounds.
This cooperative effort involved intercomparison studies (proficiency tests) in which these
agencies analyzed samples for the PAMS volatile organic (VOC) and carbonyl compounds and
reported their results to NERL for comparison to the NERL certified concentrations. Over the
two year period, a total of twelve proficiency studies were conducted, 6 for VOCs and 6 for
carbonyls. NERL compiled the results from each agency, compared the results to those from the
referee laboratory and reported the results of the comparison to all agencies. The mean, median,
-------
EPA-454/R-96-006
Chapter ?
Revision Number: 0
Date: November 1996
Page: 2
variance, and the difference from the referee laboratory's results were reported for each analyte.
One of the goals of this cooperative effort was to develop performance limits for a nationwide
audit program for PAMS measurement systems being initiated in 1995 (NPAP) which would be
modeled after these proficiency studies. The intent was to set performance limits which were
reasonable, i.e., limits encompassing at least 90 percent of the audit results. Due to reported
instability problems, proficiency tests were not required for 2-methyl-l-pentene, alpha and beta
pinenes, and isoprene. Table 5-1 shows the 90% probability limits computed from the composite
data of these audits. A column displaying the computed average bias for each parameter is also
included. Bias values outside the range -90% to -f-900% were excluded from the analysis.
As shown in the table, eleven of the fifty-one compounds had average biases exceeding ten
percent. Note also that the upper and lower limits vary considerably from the allowable ±15%
employed under NPAP guidance for the criteria pollutants (NO2, O3, etc). In fact, the "allowable"
range (upper limit minus lower limit) exceeds 30% (the criteria pollutant "allowable" range) for
every one of the parameters. Only one compound, Toluene, had 90% of its bias values within ±
20%. Therefore, based on these factors, EPA decided to use compound-specific limits in the
1995 NPAP PAMS VOC and carbonyl audits at the 90% probability limits shown in Table 5-1.
The NPAP's goal is to provide audit materials and devices that will enable EPA to
assessed the proficiency of agencies that are operating ambient monitors. All agencies operating
designated PAMS VOC and carbonyl sites were required to participate in the 1995 NPAP. The
first two VOC audits proceeded as planned; however, in the third audit, over half of the
compounds were found to be unstable. Sporadic stability problems were noted in all three of the
carbonyl audits performed in 1995. Since the data from the affected audits are questionable, no
summary report for 1995 was issued. Any future reports summarizing the PAMS audit data will
not include individual audits because the NPAP policy requires that individual data results remain
confidential. The 1996 NPAP PAMS audits were temporarily suspended pending outcome of
research into the 1995 problems. As a precautionary measure, EPA arranged for another
performance audit for the 1996 ozone season providing VOCs in canisters. The NPAP was able
to provide two carbonyl audits in 1996 (June and September). A more reliable method of spiking
the cartridges was designed and audit results so far have been excellent. Lengthy study of the
VOC stability problem has not revealed a definitive answer, although it is believed that the
passification procedure used by the manufacturer may have caused the problem. The NPAP
offered one audit for VOCs in 1996 (September) by using only the cylinders that remained stable
over a 4-month period.,
5.2.2 Precision and Accuracy Data
Although precision and accuracy checks are required for all types of PAMS monitors,
EPA has not yet issued guidance for conducting and reporting VOC P&A checks due to the
-------
EPA-454/R-96-006
Chapter 5
Revision Number: 0
Date: November 1996
Page: 3
significant number of target compounds and the non-trivial expense associated with either dual
analysis of cylinder gas or operation of a collocated continuous GC/MD. EPA realizes the
importance of data assessment information and is expending substantial resources researching the
issue. Since guidance is pending, no VOC P&A data have yet been reported to AIRS. Many
reporting organizations, however, utilize some form of P&A audits in their data validation
protocol.
Because ozone and nitrogen dioxide (two parameters of interest in PAMS) are also
criteria pollutants, a P&A policy has already been promulgated for them. EPA has established
95% probability limits (precision) of ± 15% for these two pollutants. Tables 5-2 and 5-3 show
1995 monitor summary P&A data for PAMS that monitor ozone and nitrogen dioxide. Table 5-
2 shows the 95% probability limits of precision bias for PAMS ozone monitors and Table 5-3
shows the same for PAMS nitrogen dioxide monitors.
5.3 DATA VALIDATION
Within 6 months of the end of each quarterly reporting period, data from VOC
measurement systems must be submitted to AIRS. Although data may be collected automatically,
it is interpreted and entered into AIRS manually. Some common human errors observed in this
data processing effort include incorrect units, misread formats, etc. Even after substantial pre-
AIRS QA\QC procedures have been performed, a double check of the data is always good
practice. QA is necessary to identify data errors before they are analyzed and possibly used for
such policy decisions as determination of nonattainment of the standards, control technology,
modeling, or trends.
5.3.1 Summary Statistics and Historic Precedence (Scatter Plots)
Time series plots are useful for locating unusually high changes in the data from one value
to the next or long periods of constant or no change. Figure 5-1 is a 1994 time series plot of four
species groups at Stafford, CT (Main et al., 1995). There is an easily identifiable drastic change
along the paraffin plot around SAM. An unidentified peak was misidentified as a paraffin.
Univariate statistics such as the mean, median, maximum and minimum values, are good
starting points for detection of potential data problems. Figure 5-2 plots summary statistics for
June, 1995 at the E. Hartford, CT and Stafford, CT PAMS sites including the 90th, 75th and 25th
percentiles (Main et al., 1995).
Summary statistics can be turned into box plots to hint at the distribution and variability of
the data. The dark line at the center of each box is the median, the 25th and 75th percentiles are
at the ends of the box. The box plots in Figure 5-3 show that the NMOC concentrations varied
-------
EPA-454/R-96-006
Chapter 5
Revision Number: 0
Date: November 1996
Page: 4
widely at E. Hartford, CT and Stafford, CT sites during June, 1995 (Main et al., 1995).
5.3.2 Frequency Distributions
The cost of monitoring and calibration measurements are just two of the causes of missing
data. Missing data will occur and must be considered. Each analysis should have minimum data
completeness requirements. For mean values, a 75% completeness requirement is common.
Missing data are simply ignored in most statistical analysis software. For examining trends or
time series analysis, missing values should be estimated. See Appendix H of part 50 40 CFR for
time series modeling techniques to fill in the missing values.
Side by side box plots display the number of hours each day reported for every species in
Figure 5-4 (Cox, 1995). The dark line at the center of each box is the median number of hours
reported each day. The 25th and 75th percentiles are at the ends of the shaded box. Values
outside of the whiskers (the narrow areas extending from the box) are isolated dots. Except for
the few measurements of relative humidity, the meteorological data is relatively complete. O3,
NO, NO2, NOX and CO are relatively complete compared to the other continuous pollutant
measurements.
Figure 5-5 displays a matrix for the reported ethylene data by hour for each day during
July, 1993 at the Maryland PAMS site (Cox, 1995). After the 4th, the data is relatively complete
except for morning hours.
Figure 5-6 shows frequency distributions for total NMHC concentrations measured at the
E. Hartford, CT PAMS site during June, 1995 (Main et al., 1995). Data collected from 1600-
2000 were almost exclusively above 50 ppbC. Afternoons make up the majority of the 50-200
pphC groups while morning hours are a large part of the > 250 ppbC groups. Different hour and
ppbC groupings would yield different distributions. Note the left to right skewness of the "all
data" frequency, yet the morning or early afternoon histograms would appear to be more bimodal.
5.3.3 Spatial & Temporal Plots
Spatial and temporal plots are graphical data views which readily show easily identifiable
outliers such as in the next example. The ozone exceedance at 4:00 AM on May 26, 1992 at
Cape Elizabeth, ME of 139 ppb appears erroneous when viewed in spatial and temporal context.
As the top map in Figure 5-7 shows, no other site in the vicinity reported concentrations as high
as 50 ppb much less near 139 (NESCAUM, 1992). Because the plot is by day (x-axis), the strong
diurnal (cyclic) pattern of the temporal plot is apparent. There is an obvious detectable jump on
the 26th even through the flowing pattern.
-------
EPA-454/R-96-006
Chapter 5
Revision Number: 0
Date: November 1996
Page: 5
The top graph in Figure 5-8 show suspect values which practically jump out from the page
as well as away from the rest of the data when displayed temporally (NESCAUM, 1992).
Probably because these values (for the bottom picture) are well below the standard they went
undetected. It appears they were from a misplaced decimal point as shown in the table as "#
Before", "# In", and "# After".
Something as simple as the way points are labeled or marked can facilitate visual
identification of an outlier. In Figure 5-9, the suspect calm wind is clearly marked differently
than the stronger wind vectors (Main et al., 1995). This plot of surface winds on June 27, 1991 at
1900 shows that the calm wind in Bloomington is suspect when compared to the areas around it.
Had there been other calm winds nearby, the wind velocity would be neither easily identifiable nor
suspect.
5.3.4 Inter-Site Comparisons & Inter-Species Comparisons
Intersite and inter-species comparisons are important to identify similarities and
differences. When site similarities are apparent (e.g., similar precursors are detected) similar
control measures and predictors can be used. The example depicted in Figure 5-10 compares the
Stafford, CT and E. Hartford, CT PAMS sites (Main et al., 1995). It contrasts the VOC
composition at the two sites at 8:00 AM on June 3, 1994. While most other species behave
similarly at the two sites, species #11 differs greatly. One possible explanation could be the
influence of local sources. Alternatively, if the two sites were known to have previously had
similar species #11 concentrations, then this reading might be suspect.
When two species are highly correlated, they might be dependent on one another. If so,
neither species would be suitable for a prediction model that required linearly independent
variables as predictors. The scatter plot matrix in Figure 5-11 illustrates the strong correlations
between some VOC species for Stafford, CT during in June, 1995 (Main et al., 1995).
Figure 5-12 simultaneously compares two sites and two species: isoprene and m&p-
xylenes for the PAMS sites at Stafford. CT and E. Hartford, CT (Main et al., 1995). Xylene
concentrations act differently while isoprene behaved similarly at the two sites. Again, one
explanation could be local emissions released at those times when the patterns are different.
-------
EPA-454/R-96-006
Chapter 5
Revision Number: 0
Date: November 1996
Page: 6
5.4 REFERENCES
Cox, W.M. "A Workbook for Exploratory Analysis of PAMS Data." June 1995.
Main, H.; Roberts, P.; and Korc, M. Analysis of PAMS and NARSTO Northeast Data Supponing
Evaluation and Design of Ozone Control Strategies: A Workshop. U.S. EPA Contract
68D30030, Sonoma Technologies, Inc., July 1996.
Northeast States for Coordinated Air Use Management (NESCAUM), The Ambient Monitoring
and Assessment Committee. Preview of 1994 Ozone Precursor Concentrations in the
Northeastern U.S. August 1995.
Northeast States for Coordinated Air Use Management (NESCAUM), The Ambient Monitoring
and Assessment Committee. 1992 Regional Ozone Concentrations in the Northeastern United
States. November 1993.
U.S. Environmental Protection Agency. Quality Assurance Handbook for Air Pollution
Measurement Systems. Volume II: Ambient Air Specific Methods (Interim Edition').
EPA/600/R-94/038b, 1994.
-------
F-ipnrc
LJvoc
JxV Efe
ppbC
100
75
50
25
0
i
Dal - ITMno Sonet Giaph «1|
Edit graph V^mdow Help
| >| En»iie Record | Valid Fslimateri Suspcrl Invalid |
Time(EST)
I
1
-
\
^- L /"\
^ - -/ 'L_^ \, /
r-^~ - ,-^ -J/ '"N ,->^ -/x \ ' ' x"-
~~~' -^' ""^ " v--^ \ "^-v
L- 1 L-I--1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 I I I l I I 1 I I I I I I I ill 1 1 I \ I I I 1 I I I I I I i i t ll 1 .
6/21 6/22 6/23
-lulxl
atornat
olefin
parafn
uidvoc
Time series plot of several species groups at Stafford, CT in 1994. Example of misidentification of a paraffin
for an unidentified peak. (Level 0, preliminary data, CT DEP)
-------
Figure 5-2.
§
I
I
o
600
500
400
300
200
100
E. Hartford, CT
June, 1995
NMHC
0 '1 1 I I I I I I I I I I 1 I I I I I I I I I I I
0 2 4 6 8 10 12 14 16 18 20 22
Hour, EST
Average
Median
Minimum
Maximum
25th
90th
Auf 13. 1995
Stafford, CT
250
200
0
CL
o.
c- 150
03
"E
8 100
o
O
50
n
June, 1995
NMHC
^z****-**.^ - -^~:
Average
V
Median
Minimum
Maximum
25th
75th
a
90th
0 2 1 6 8 10 12 14 16 18 20 22
Hour, EST
Summary statistics plotted for East Hartford and Stafford, CT for June 1995.
(Level 0, preliminary data, CT DEP)
-------
Figure 5-3.
U
I
n
z
fc
508
U <..
o
ha
£ 200
IBB -
1 I I I I 1 J i 1 I I I I I 1 I J 1 1 ! I 1 I
0 2 4 E B IB 12 14 IE 18 2B 22
1 3 5 7 f 11 13 15 17 1* 21 23
Hour, EST
250 -
U
x>
c.
o.
CJ 208
O
150
£ 100 -
0 2 4 £ 6 IB 12 14 IE IB 20 22
1 3 5 7 ? 11 13 15 17 1? 21 23
Hour, EST
Box plots of NMOC by time of day during June 1995 at East Hartford, CT(top) and
Stafford, CT (bottom). Concentrations varied widely at East Hartford
-------
Figure 5-4.
BALIiyORE PAVS DAIA 1993
DAIA COyPLETENESS
NUyBER.OF HOURS PER DAY AVAILABLE
KHUN
SOLRA&
TEIP
fDIR
ISPW
fiOX
N02-
n
CO
OZONE-
ISOPR-
BEN2E-
HLEN-
IOLUE-
OLEfl-
ACiYl-
Numbef Hoois per Day
-------
Figure 5-5.
DISTRIBUTION OF MISSING VALUES
FOR ETHYLENE AT BALTIMORE PAMS SITE
DAY AND HOUR OF DAY FOR JULY 1993
I
HOURS
01 03 05 07 09 11 13 15 17 19 21 23
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
K3fr^^
-------
Frequency of Occurrence
o o
Frequency of Occurrence
-------
Figure 5-7.
Example of identification of suspect data values from the Northeast (NESCAUM 1993). The ozone concentration of
139 ppb reported at Cape Elizabeth, ME on May 26,1992 at 4:00 AM appears erroneous when viewed in a spatial and
temporal context.
Reported Regional Ozone Concentrations at 4A.M. on May 26.1992
(Cape Elizabeth. ME = 139 ppb; No Other Sites Are as High as 50 ppb)
O :<5°PP
> 125 ppb
(Cape Elizabeth. ME|
Gardiner. ME
-------
Figure 5-8.
BA.WMC Waittauk tU HE
(Hourly NE oxen dale »» 1187-92)
This Weitoeok to lo MeirMy emalele imn In NE etote
azone data Netcd In the EM AIRS dateene fa* ot4/1 ZfJ3)
Thta PMC MMM Z ( 11 eatped 1 SSI date vahiet kaei erfe
1MOaiOOIZ r*»»eiaLll Mta. NY)- Tkc
higher dim MHM! M Mi life. >iir>i»d by period! of
mlislig date, and IncamMBM «M Ihe retl «4 *B region.
fTiy awmlng *»»y* »« view la «ad alker tlmllar p*Ma|.
data *MI ttito Mid after atei art lifted page X
m* WMttaokvM knt (pdatad n 5^4/91. ba*ad gn dm
The OZ8792QA.WKB workbook is dynamic - subject to continual revision as addidonal suspect
data points are identified. Ultimately, these points are reported back to the original data generating
state agencies - which can flag alter or eliminate the suspect data points from AIRS.
A claae look at the hourly «ata rran 3 Vermont iltea
akawm a aamaer el laolattd kaurty vatae* wfclch are
aaomaleatry tawer ^y a« lead ZO ppb) kan the reported
nkevM* In the pncndiag aid t*H**tne haw*.
In nearly every cat*, feme appear ta rewH tree)
mtiplcccd deriiMl potaL Far
Lacathm Barlfaglan Beealngte* UnderhW
Date t/4/W
Time 17:00
» Betere 3S ppb
* to t 4 ppb
f After 42 ppb
16:10
12:80
4 ppb
44 apt
6 ppb
«Zppb
Tlreic "»Uoped dedriMl pgkitc" may have gene
«*deteaed elnce aM art a* (he tav end t conaequeairy
ynlmpertaat trftti r»aa«cl to aflahieienl etetui
The»e aad Mher auipect data val«e« arc IMcd en Hie
toHewtng pege. CHb« auepco data values w«l be added
to ftlt worckeek at «hey are Identffled.
-------
Figure 5-9.
LAKE MICHIGAN OZONE STUDY
DATE 06/27/91 10-° ms
TIME It; UU i Lvvtl 1.0 Valid «nd EiUmtUd d«lt
Hot of surface winds on June 27,1991 at 1900 CUT. The
calm wind at Bloomington, Illinois was identified as suspect
(SUS) during the data validation process. (Roberts et al.,
1993)
-------
Figure 5-10.
QVOCOat - Fingerprint Graph «1|
til Ete E* Qi«ph i/ndow Help
JLJ J
ppbC
>| Entire Hucoid } Valid | Estimated) Suspect | Invalid |
15
10
5
0
C
' ' ' ' I ' ' ' ' I '
-
-
" A . I
J\ ;
- 1\i\\ i\
,1 Ui \ f,\ I
' * \J\- 1
) 5 10
i i | i i i i | i t i i | i i i i | i i i i | r 'i i i j i i n-| r r i i | i i i i | i i i i [ i i i i | i i i i |
I
i
<$>
\
i f.
^.::,^,An-AJ1^.7SA.Z ,. J
15 20 25 30 35 40 45 50 55 60 65 70
CT01
CT02
Species
East Hartford CT McCauWfe Park
03Jun35 08:00
»,pna-11.44ppbC
Fingerprint plot of June 3, 1994 at 0800 EST for Stafford and East Hartford, CT. This plot illustrates some
of the differences between composition at the two sites. (Level 0, preliminary data, CT DEP)
-------
Table 5-1. 90% Probability Limits for PAMS Target VOCs and Carbonyls Established by the 1993/1994
Proficiency Studies.
COM POUND
E th y le n e
Acetylene
E th a n e
Propylene
P r o p a n e
Isobutane
1 - B u te n e
n - B u ta n e
trans-2-Butene
cis-2-B u te n e
3 -M e th y I- 1 -B u te n e
Isope ntane
1 -P e n te n e
n-Pentane
trans-2-Pentene
cis-2-Pentene
2-Methyl-2-Butene
2,2-D im e th y !b u ta n e
C yclopentene
4 - M ethyl-1-Pentene
C yclopentane
2, 3-D imethylbutane
2 - M e th y Ip e n ta n e
3 - M e th y Ip e n ta n e
n - H e x a n e
trans-2-H exene
cis-2-H exene
M ethylcyclopentane
2,4-D imethylpentane
Benzene
C > clone xane
2 - M ethylhexane
2 ^ - D imethylpentane
1 - M ethylhexane
2 . 2 .4 - T rimethylpentane
n - H e p ta n e
M eth\ Icy clone xane
2 1 4 - T rimeihylpenlane
T o 1-u e n e
- - M e t h s Iheptane
^ - M cth\lheptane
n - O c i a n e
hlhvlhenzcne
m p - X \lene
S t \ re n e
o - X \iene
n N o n a n e
lvoprop\lhenzene
n-Prop\lbenzene
1.3.5-T rimethylbenzene
1.2 4-Trimeihylbenzene
Formaldehyde -low
Formaldehyde -high
Aceialdehyde -low
AceiaJdehvde -high
Acetone - low
Acelone -high
AVERAGE
BIAS
- 1 3
-9
- 1 2
4
-9
-9
-7
- 1 0
-6
-7
-6
1
-6
-7
4
-9
9
- 1 0
-3
-9
- 1 1
-5
- 1
-6
-9
- 1 1
-4
-5
-8
- 1 3
- 5
-3
_ 7
-7
. 5
-8
_ 3
-6
- 1
. 3
T
- 1 0
4
! 2
- 4
- 1
-6
4
- 1 1
6
- 1 1
0
-2
1
-2
1
-6
LOW E R
LIMIT
-4 0
-3 5
-2 7
-4 0
-3 4
-4 4
-3 4
-3 1
-2 9
-4 3
-2 9
-3 6
-2 9
-2 3
-3 5
-2 5
-3 3
-2 4
-4 2
-3 5
-3 4
-2 8
-3 1
-3 0
-4 2
-3 0
-3 2
-3 3
-3 7
-2 9
-3 8
-3 1
-3 4
-3 1
-4 0
-2 5
-3 3
- 1 9
- 3 2
-2 1
-4 7
-3 5
-4 2
-6 7
-5 1
-2 6
-4 6
-6 7
-5 3
-5 9
-2 2
-2 4
-2 1
-2 5
-2 2
-2 9
1 P P E R
L 1 M IT
T 1
1 1
3 s
-> >
1 6
3-0
1 4
1 9
1 5
3 1
3 1
2 4
1 5
3 I
1 7
4 3
1 3
1 8
2 4
1 3
2 4
2 6
1 9
1 2
2 0
2 2
2 2
1 7
1 1
1 9
3 2
1 7
2 0
2 1
2 4
1 9
2 1
1 7
2 6
2 5
2 7
4 3
6 6
5 9
4 9
1 4
5 4
4 5
6 5
3 7
2 3
2 0
2 4
2 0
2 3
I 6
-------
Table 5-2.
95% Probability Limits of Bias (Precision) for PAMS Ozone Monitors. 1995.
Boston
Connecticut
3ortsmouth
Providence
Springfield
New York
Baltimore
Philadelphia
W ashmgton
Atlanta
Lake Michigan
Houston
Baton Rouge
Beaumont
El Paso
South Coast/
S E D A B
San Diego
Ventura Co.
Sacramento
San Joaqum
Site
1
2
3
4
2
3
2
1
2
3
1
2
3
2
1
2
2
3
4
1
3
1
3
4
2
3
2
2
2
3
4
4
2
2
3
1
2
3
2
2
2
3
2
3
2
4
2
2
3
2
3
2
3
2
2
3
3
UR S ID
250051005
250092006
250094004
230052003
090031003
090131001
230313002
090010017
440071010
250051005
250130003
250130008
250154002
360050083
240030019
240053001
245100050
240259001
100031007
100031007
340210005
510330001
240030019
100031007
130890002
132470001
550790041
170310072
180891016
550890009
170971007
550710007
48201 1C35
482011003
482010024
220330008
220330009
220470009
48245001 1
481410027
48141 0044
481410037
060371601
060370002
060650002
06071 1004
060730003
060730006
060731006
061113001
061112002
060670006
060671001
060290010
060195001
060295001
060194001
Ye
-3
-10
-6
-5
-9
-7
-1 1
-4
-5
-3
-10
-4
-4
-6
-3
-4
-4
-3
-10
-10
-4
-8
-3
-10
-6
-1
-6
-8
-3
-8
-7
-10
-8
-7
-7
-3
-9
-5
-6
-8
-12
-6
-1 1
-7
-16
0
-6
-8
-6
-4
-11
-5
-12
-8
-4
-10
3
ar
UD r orb
4
8
4
2
6
3
8
6
5
4
7
4
3
6
2
2
4
3
11
1 1
4
1
2
11
4
11
3
13
2
8
9
4
4
1 1
4
10
1
7
10
2
13
3
2
4
25
7
6
4
5
9
8
8
0
4
3
4
5
o
-M
-9
-e
-7
-5
-2
-3
-2
2
-21
-11
-2
-5
-3
-6
-4
9
-7
-4
-11
-4
5
-2
-a
-e
-7
-16
-3
-5
-4
-9
1
7
4
3
7
8
1
^
0
1
0
0
17
2
11
0
0
9
3
19
4
6
6
14
10
0
Q
-9
4
-5
-11
-8
-9
-5
-2
-10
-5
-4
-9
-3
-2
-3
-2
-2
-5
-9
-2
-7
-2
-3
-13
-4
-8
-3
-3
-5
-5
-3
-8
-2
-6
-9
-4
-11
-5
-6
2
-5
-7
.3
.5
-8
-1
2
2
1
1
4
2
11
4
2
8
4
8
7
1
3
6
5
5
4
-1
5
6
8
0
15
2
8
8
4
4
6
2
7
-3
8
12
6
3
0
6
23
8
7
4
4
7
0
Q
-3
-5
-6
-3
-9
-3
-11
-1
-S
-3
-11
-3
-3
-2
-5
-3
-6
-2
-M
-14
-5
-6
-5
-M
-6
1
-6
-6
-2
-9
-4
-8
-4
-8
-11
-9
-3
-7
-9
-3
-7
-8
-2
2
-5
10
-8
-2
-13
-8
-4
_l
3
3
4
1
9
3
7
6
3
3
4
3
3
5
3
1
5
1
13
13
7
1
3
13
10
6
12
2
9
8
-3
5
6
8
4
7
4
16
0
_ t
2
13
c
7
6
7
11
Q
-3
-4
-4
-2
-12
0
-7
-2
-3
-2
3
-3
-3
-5
-2
-3
4
-4
-4
-9
-3
-4
0
8
0
-9
-4
-5
-4
-8
-6
-4
-11
-4
10
-3
-7
-30
.3
-12
-6
.4
-10
A
4
13
6
1
2
5
8
11
4
8
3
0
2
3
3
3
2
11
11
4
3
11
2
10
8
3
-2
0
5
3
0
5
7
5
12
4
t
6
11
e
8
0
0
-------
Table 5-3. 95% Probability Limits of Bias (Precision) for PAMS NO: Monitors, 1995.
Area
Boston
Connecticut
Portsmouth
Providence
Springfield
New York
Baltimore
Philadelphia
Washington
Atlanta
Lake Michigan
Houston
Baton Rouge
El Paso
South Coast/
SEDAB
San Diego
Ventura Co
Sacramento
San Joaqum
Site
Type
2
3
2
3
2
2
1
2
3
2
1
2
2
3
4
2
3
1
3
4
2
3
2
2
2
3
4
4
2
3
1
2
3
2
3
2
3
4
2
2
3
2
3
2
3
2
2
2
AIRS ID
250092006
250094004
090031003
090131001
230313002
440071010
250130003
2501 30008
250154002
360050083
240030019
240053001
245100050
240259001
100031007
421010004
340210005
510330001
240030019
1 00031 007
130890002
132470001
550790041
170310072
180891016
550890009
170971007
550710007
482011035
482010024
220330008
220330009
220470009
481410027
481410037
060371601
060370002
060711004
060730003
060730006
060731006
061113001
061112002
060670006
060671001
060290010
060195001
060295001
060194001
Year
wrprb
-11
-6
-9
-10
-3
-9
-13
-16
-17
-10
-3
-4
-3
-4
-9
-6
-6
-10
-3
-9
-15
-7
-8
-17
-6
-10
-7
-7
-2
-8
-8
-2
-2
-13
-12
-11
-12
-5
-14
-4
-4
-9
-9
-4
_c
-6
2
-19
4
uprprb
14
11
9
2
10
3
5
13
6
9
3
7
2
5
17
5
7
5
3
17
9
6
7
9
11
12
15
9
3
1
11
5
8
7
8
21
8
18
20
17
13
4
10
14
10
19
14
8
15
Q1
wrprb
-2
-2
-3
-9
-7
-15
-4
-6
-3
-4
-7
-6
-17
-B
-7
-2
-3
-4
-11
-2
0
12
11
-21
-8
-5
0
-6
.4
C
-1
-6
t)
-16
4
uprprb
T2
14
11
3
7
2
7
10
4
5
6
10
2
2
2
6
4
0
t)
3
6
3
18
-3
10
15
18
7
;
9
16
15
02
wrprb
-14
-2
-13
-11
-9
-t)
-14
-T3
-4
-5
-5
-4
-T2
-7
-4
-14
-4
-T2
-X)
-6
-6
-7
-6
-13
-2
-9
-1
-4
1
-2
.3
.9
-8
-1
-11
-t:
-14
2
.3
-1
-8
.7
6
7
upr prb
t3
7
13
2
2
13
4
5
3
7
3
6
26
6
3
7
3
26
0
6
8
8
7
T3
4
c
2
0
6
;
11
0
3
16
I
14
26
18
18
3
12
17
03
wrprb
-11
-4
-3
-8
-28
-18
-2
-2
-3
-4
-8
-5
-9
-7
-2
-8
-12
-7
-9
-B
-7
-t)
-6
-5
-3
-11
0
2
-13
-6
-12
-2
-1
-3
.7
-9
c
4
i
pr prb
14
8
3
10
3
18
15
9
2
2
4
13
6
14
2
2
13
6
9
7
7
7
12
20
9
4
1
4
8
B
2
X)
11
24
2
7
1
2
T3
15
I
T3
Q4
wr prb
-T3
-^
-8
-11
-a
-14
-14
-8
-3
0
-6
-5
-11
0
1
-1
-8
-1
-2
-8
-2
-2
-13
-18
3
.9
-9
.4
.3
_C
13
2
-13
1
prorb
4
9
3
4
5
7
-5
4
3
XI
6
3
4
X)
T
1
r
6
3
-2
B
5
C
19
24
15
8
6
X)
14
X)
15
X)
12
«
------- |