EPA-CMB8.2 Users Manual
-------
-------
EPA-452/R-04-011
December 2004
EPA-CMB8.2 Users Manual
By:
C. Thomas Coulter
Air Quality Modeling Group
Emissions, Monitoring & Analysis Division
Office of Air Quality Planning & Standards
Research Triangle Park, NC 27711
US. Environmental Protection Agency
Office of Air Quality Planning & Standards
Emissions, Monitoring & Analysis Division
Air Quality Modeling Group
-------
ACKNOWLEDGMENTS
CMB8 was originally developed as part of Contracts
5D1607NAEX and 5D2724NAEX between EPA's National
Exposure Research Laboratory (NERL) and the Desert Research
Institute (DRI) of the University and Community College System
of Nevada. The manual was originally drafted as part of Contract
5D1808NAEX with EPA's Office of Air Quality Planning &
Standards (OAQPS). The Project Officers were C. Thomas
Coulter1 and Charles W. Lewis2. Substantial contributions to
initial development of the CMB8 model were made by DRI staff
members John G. Watson, Norman F. Robinson, Judith C. Chow,
Eric M. Fujita, and Douglas H. Lowenthal. Teri L. Conner2 and
Robert D. Willis3 also made important contributions to CMBS's
initial development.
Tom Coulter and Donna Kenski (then at EPA Region 5) spent
considerable time troubleshooting CMB8 in May 1999. An effort
was initiated by EPA in 1999 to repair and redesign CMB8. In
September 1999, Pacific Environmental Services (PES) began
work to repair and enhance CMB8 under contract 9D-1844-NTSA,
an effort that led to the redesigned CMB8.2. CMB8.2 was
developed principally by Vickie Kriegsman and Robert Wagoner
PES (now MACTEC), Research Triangle Park, NC. Additional
issues and operational problems were discovered and documented
in early 2000. Subsequently, Ideal Software, Inc. (Apex, NC) was
retained under contracts 2D-6228-NTSA (July 2002) and
4D-6389-NTSA (July 2004). Ideal Software engineers
successfully compiled the model's C++/Fortran-based DLL (the
first since DRI to do so), and resolved many behavioral problems
with the model, now called EPA-CMB8.2. Ideal's John Scalco
created many of the runtime features EPA-CMB8.2 now exhibits
through its (Delphi) enhanced User Interface. Tom Coulter
reorganized and refactored C++ and Fortran code for the DLL,
reworked error handling, and removed obsolete variables and
subroutines. With careful assistance from Ed Anderson of Science
Applications International Corp., Tom updated the LINPACK
library to LAPACK v3.0 and modified the Fortran code to
accommodate it. Tom also served as Contracting Officer
Representative for these contracts, assisted with the Delphi
Office of Air Quality Flaming & Standards, Research Triangle Park, NC 27711
National Exposure Research Laboratory, Office of Research & Development, Research Triangle Park, NC 27711
ManTech Environmental Technology, Inc., Research Triangle Park, NC 27709
ii
-------
reprogramming, and rewrote the manual. Tom's diligent work
resulted in a level of documentation and disclosure unknown in the
history of the CMB model.
EPA-CMB8.2 has its foundation in the previous versions that span
two decades. OAQPS' involvement in supporting CMB
development, dating back to the mid-80s, was initiated under the
guidance of Tom Pace. The present author is indebted to the many
contributors to the earlier work.
During September - December 2004, EPA-CMB8.2 and its
documentation (this Users Manual and its companion Protocol for
Applying and Validating the CMB Model for PM2.5 and VOC) were
subjected to scientific peer review under EPA Contract 4D-6097-
NTSX. EPA is grateful for the Peer Review panel - Jamie
Schauer, Donna Kenski, Robert Willis - and its diligent work and
helpful suggestions for both the model and its documentation.
Further suggestions from users are welcome, and may be directed
to Tom Coulter at
Coulter.Tom@epa.gov.
DISCLAIMER
This manual was reviewed by EPA for publication. The
information presented here does not necessarily express the views
or policies of EPA. Any mention of trade names or commercial
hardware and software in this document does not constitute
endorsement of these products. No explicit or implied warranties
are given for the software and data sets described in this document.
in
-------
Abstract
The Chemical Mass Balance (CMB) air quality model is one of
several receptor models that have been applied to air resources
management. EPA-CMB8.2 incorporates the upgrade features that
CMB8 has over CMB7, but also corrects errors/problems
identified with CMB8 and adds enhancements for a more robust
and user-friendly system. EPA-CMB8.2 is a 32-bit (Windows® 9x
and higher) version of CMB modeling software that substantially
facilitates the estimation of source contributions to speciated PMio
(particles with aerodynamic diameters nominally less than lOjim),
PM2.5 (particles with aerodynamic diameters nominally less than
2.5|im), and Volatile Organic Compounds (VOC) data sets.
EPA-CMB8.2 features: (1) full use of Windows® (32-bit) for file
access/management, (2) a tabbed page interface that eases the
necessary progression for doing a CMB calculation, (3) multiple,
indexed arrays for selecting fitting sources and species, (4)
versatile display capability for ambient data and source profiles,
(5) mouse-overs and on-line help screens, (6) increased attention to
volatile organic compounds (VOC) applications, (7) correction of
some flaws in the previous version (CMB7), (8) flexible options
for input/output data formats, (9) addition of a more accurate least
squares computational algorithm, (10) upgraded linear algebra
library, (11) a new treatment of source collinearity, and (12) choice
of criteria for determining best fit.
This manual introduces EPA-CMB8.2 and its development history.
It describes hardware and software requirements and shows how to
install EPA-CMB8.2 on a personal computer. It explains EPA-
CMB8.2 menu options and input and output file formats. The
manual provides a step-by step tutorial of EPA-CMB8.2 operations
using an example data set provided with the model. Performance
measures are briefly described, though their use in practical
applications is deferred to a separate application and validation
protocol. A comprehensive list of references is included for those
desiring more information about CMB, its utility and applications.
IV
-------
Table of Contents
List of Figures vi
List of Tables vii
1. Introduction 1-1
1.1 EPA-CMB8.2 Features 1-2
1.2 Chemical Mass Balance Overview 1-4
1.3 CMB Software History 1-6
1.4 Organization of the Users Manual 1-7
2. Software Installation 2-1
2.1 Hardware and Operating System 2-1
2.2 EPA-CMB8.2 Software and Related Files 2-1
2.3 Installing EPA-CMB8.2 Software 2-3
3. EPA-CMB8.2 Operation 3-1
3.1 Input Files 3-1
3.2 Options 3-5
3.3 Select Ambient Data Samples 3-7
3.4 Select Fitting Species 3-8
3.5 Select Fitting Sources 3-10
3.6 Calculation Results 3-11
3.6.1 Main Report 3-12
3.6.2 Contributions by Species 3-14
3.6.3 Modified Pseudo-Inverse Normalized (MPIN) Matrix 3-15
4. Input and Output Files 4-1
4.1 File Naming Conventions 4-1
4.2 Input File Relationship 4-2
4.2.1 Control File: IN*.in8 4-2
4.2.2 Ambient Data Input File (AD*.csv, AD*.dbf, AD*.txt, AD*.wks) 4-4
4.2.3 Source Profile Input File (PR*.csv, PR*.dbf, PR*.txt, PR*.wks) 4-6
4.2.4 Source (PR*.sel), Species (SP*.sel), and
Sample Selection (AD*.sel) Input Files 4-8
4.3 Output Files 4-11
4.3.1 Report Output File 4-11
4.3.2 Data Base Output File 4-11
4.4 Creating Data Input Files 4-12
4.5 Reading Output Files 4-13
5. Using EPA-CMB8.2: Test Cases 5-1
5.1 Starting EPA-CMB8.2 5-1
5.1.1 Control File Operation 5-1
5.1.2 Individual File Operation 5-10
5.2 Best Fit Option 5-11
6. EPA-CMB8.2 Performance Measures 6-1
6.1 Main report 6-1
6.1.1 Source Contribution Estimates and Fitting Statistics 6-1
6.1.2 Eligible Space Collinearity Display 6-3
6.1.3 Species Concentration Display 6-5
6.2 Additional Performance Measures 6-7
6.3 Best Fit Measure 6-9
7. References 7-1
-------
Appendix A Theory of the Chemical Mass Balance Receptor Model A-l
Appendix B Source Code for EPA-CMB8.2: Fortran & C++ B-l
Appendix C The Linear Algebra Library for EPA-CMB8.2: LAPACK C-l
Appendix D The User Interface for EPA-CMB8.2: Delphi D-l
Appendix E Error Messages from EPA-CMB8.2 E-l
Appendix F Data Input Issues for EPA-CMB8.2 F-l
Appendix G Notes on EPA-CMB8.2 Diagnostics G-l
List of Figures
Figure 3.1 Launching EPA-CMB8.2 3-1
Figure 3.2 Browse Box for Selecting a Control File 3-2
Figure 3.3 Input Files Screen 3-3
Figure 3.4 Banner Page for EPA-CMB8.2 3-4
Figure 3.5 Options for Current Run 3-5
Figure 3.6 Ambient Data Selection 3-7
Figure 3.7 Fitting Species Arrays 3-8
Figure 3.8 Fitting Sources Arrays 3-10
Figure 3.9 Calculation Results 3-11
Figure 3.10 Calculation Results - Main Report 3-12
Figure 3.11 Calculation Results - Contributions by Species 3-14
Figure 3.12 Calculation Results - MPIN Matrix 3-15
Figure 4.1 EPA-CMB8.2 Input Files 4-3
Figure 5.1 Input Files for Test Case - Control File Mode 5-1
Figure 5.2 Ambient Samples for Test Case 5-2
Figure 5.3 Speciation Graph for Ambient Sample (Test Case) 5-3
Figure 5.4 Species Selection for Test Case 5-4
VI
-------
Figure 5.5 Source Profile Selection for Test Case 5-5
Figure 5.6 Speciation Graph of Selected Source Profile (Test Case) 5-5
Figure 5.7 Main Report - Test Case 5-6
Figure 5.8 Contributions by Species - Test Case 5-7
Figure 5.9 MPIN Matrix - Test Case 5-8
Figure 5.10 Input Files for Test Case - Individual Files Mode 5-10
Figure 5.11 Input Files for Testing the Best Fit Option 5-11
Figure 5.12 Results for Test Case Using Best Fit Option 5-12
vn
-------
-------
1. INTRODUCTION
The Chemical Mass Balance (CMB) air quality model is one of several receptor models that
have been applied to air resources management. Receptor models use the chemical and physical
characteristics of gases and particles measured at source and receptor to both identify the
presence of and to quantify source contributions to receptor concentrations. Receptor models are
generally contrasted with dispersion models that use pollutant emissions rate estimates,
meteorological transport, and chemical transformation mechanisms to estimate the contribution
of each source to receptor concentrations. The two types of models are complementary, with
each type having strengths that compensate for the weaknesses of the other.
This manual describes how to operate EPA-CMB8.2 modeling software to calculate source
contributions to ambient PMio (particles with aerodynamic diameters nominally less than lOjim),
PM2.5 (particles with aerodynamic diameters nominally less than 2.5|im), and volatile organic
compounds (VOC).
A separate applications and validation protocol (EPA, 2004) describes how to apply EPA-
CMB8.2 to specific situations and how to evaluate its outputs. Several review articles, books,
and conference proceedings provide additional information about the CMB and other receptor
models (Chow etal., 1993; Gordon, 1980, 1988; Hopke and Dattner, 1982 ; Hopke, 1985, 1991;
Pace, 1986, 1991; Stevens and Pace, 1984; Watson, 1979, 1984; Watson et al, 1989, 1990,
1991).
1 - 1
-------
1.1 EPA-CMB8.2 Features
EPA-CMB8.2 replaces CMB7 (EPA, 1990; Watson etal, 1990) as a more convenient
method of estimating contributions from different sources to ambient chemical concentrations
(Coulter and Scalco, 2005). EPA-CMB8.2 returns the same results as CMB7, but it operates in a
Windows®-based environment and accepts inputs and creates outputs in a wider variety of
formats than CMB7. The major EPA-CMB8.2 enhancements are:
Windows®-based, event-driven operations: EPA-CMB8.2 makes full use of Windows® (32-
bit) features, including a tabbed interface that facilitates the necessary progression for doing
a CMB calculation. Commands may be executed with hot-keys or toolbar buttons, and
features are described via mouse-overs and context sensitive on-line help screens. EPA-
CMB8.2 also offers flexible options for input/output data formats. Input formats are
compatible with output files from EPA's source profile library SPECIATE
(www. epa.gov/ttn/chief/}.
Multiple arrays for fitting species and fitting sources: Up to ten indexed arrays of fitting
source profiles and fitting species may be specified in input data selection files. Different
arrays can be selected during EPA-CMB8.2 operation. Upon session exit, an option is
provided to conveniently save (update) or rename selection files to reflect arrays that are
added, modified or deleted during the session.
Britt-Luecke algorithm: A general solution to the least squares estimation problem that
includes uncertainty in all the variables (i.e., the source compositions as well as the ambient
concentrations) is available. While an approximation to the Britt-Luecke iteration scheme
(Britt and Luecke, 1973) was used in CMB7, exercise of Britt-Luecke algorithm option in
EPA-CMB8.2 allows solution without approximation.
Improved collinearity diagnostics: The uncertainty/similarity clusters have been replaced with
a singular value decomposition eligible space treatment that allows the user to define an
acceptable error and an acceptable collinearity among weighted source profiles.
Better handling of VOC applications: EPA-CMB8.2 gives the user control in adjusting
collinearity parameters which in CMB7 are "hard-wired" and not necessarily optimum for
every application. Values for these parameters were chosen in CMB7 to be compatible with
characteristics of particulate mass measurements, but they may not be as well suited to CMB
solutions involving VOC.
Search for best fit: Using a user-selected weighted optimization of performance measures,
EPA-CMB8.2 can systematically check up to 10 possible paired combinations of fitting
species and sources arrays as it searches for a maximum of an empirical composite measure.
The best fit arrays are then indicated in their respective windows.
User-selected preferences: In EPA-CMB8.2, the user may set options for maximum iterations
for convergence, eligible space tolerances, positions of decimal points in output, receptor
concentration units, special calculation alternatives, and performance measure weights for
use in Best Fit mode.
1 -2
-------
Negative source contributions: EPA-CMB8.2 calculations can be set to eliminate negative
contributions.
Improved memory management: EPA-CMB8.2 memory is limited only by the available
RAM on the host computer, not by pre-set memory limitations.
Upgraded linear algebra library: The linear algebra library that EPA-CMB8.2 uses to perform
its effective variance, least-squares regressions has been updated with LAPACK v3.0.
Versatile graphic display capability: For ambient samples and source profiles, bar charts for
species concentrations can be displayed within EPA-CMB8.2, which is useful for visual
inspection. These can be cut from their windows and pasted into other Windows®
documents.
Context-sensitive on-line help: Context-sensitive on-line help is accessible directly from the
User Interface.
Flexible input and output formats: comma-separated value (CSV), xBASE (DBF), and
worksheet (WKS) formats are supported as input files, in addition to the formatted, blank-
delimited ASCII text files (TXT) supported by CMB7. Output files formats are ASCII and
CSV (which ports nicely to Microsoft Excel®).
File handling: EPA-CMB8.2 differs from CMB7 in several ways with regard to the files used
by each. EPA-CMB8.2 does not support CMB6 style ambient data and source profile data
files. Control File (formerly filenames file), source profile, ambient data, and sample
selection file formats differ slightly from CMB7. CMB7 source profile and ambient
concentration data files can be read directly by EPA-CMB8.2, however, so backward
compatibility is assured. Different Control Files can be loaded during the same session,
obviating the need to terminate the application. In EPA-CMB8.2 graphical output is not
provided as HPGL text files. Instead output can be printed through Windows® or copied to
the clipboard for insertion into documents. Text output can also be directed to the printer,
the clipboard, or a report file. A feature of EPA-CMB8.2 is that the computational
machinery files (*.exe, *.dll, etc.) need not reside in the same folder as either the input or
output files. This facilitates file management.
The naming structure for (optional) selection files has been changed to one that is more
logical and intuitive - a convention that meshes with the one used for naming the required
input data files:
Source PRofile selection file:
SPecies selection file:
Ambient Data (sample) selection file:
Previously
SO*.sel
PO*.sel
DS*.sel
EPA-CMB8.2
PR*sel
SP*.sel
AD*.sel
1 -
-------
1.2 Chemical Mass Balance Overview
The CMB receptor model (Friedlander, 1973; Cooper and Watson, 1980; Gordon, 1980,
1988; Watson, 1984; Watson etal, 1984; 1990; 1991; Hidy and Venkataraman, 1996) consists
of a solution to linear equations that express each receptor chemical concentration as a linear
sum of products of source profile abundances and source contributions. For each run of CMB,
the model fits speciated data from a specified group of sources to corresponding data from a
particular receptor (sample). The source profile abundances (i.e., the mass fraction of a
chemical or other property in the emissions from each source type) and the receptor
concentrations, with appropriate uncertainty estimates, serve as input data to CMB. The output
consists of the amount contributed by each source type represented by a profile to the total mass,
as well as to each chemical species. CMB calculates values for the contributions from each
source and the uncertainties of those values. CMB is applicable to multi-species data sets, the
most common of which are chemically-characterized PMio, PM2.5, and Volatile Organic
Compounds (VOC). The theory of CMB is described in Appendix A.
The CMB modeling procedure requires: 1) identification of the contributing source types;
2) selection of chemical species or other properties to be included in the calculation; 3)
knowledge of the fraction of each of the chemical species which is contained in each source type
(source profiles); 4) estimation of the uncertainty in both ambient concentrations and source
profiles; and 5) solution of the chemical mass balance equations. The CMB approach is implicit
in all factor analysis and multiple linear regression models that intend to quantitatively estimate
source contributions (Watson, 1984). These models attempt to derive source profiles from the
covariation in space and/or time of many different samples of atmospheric constituents that
originate in different sources. These profiles are then used in a CMB solution to quantify source
contributions to each ambient sample.
Several solution methods have been proposed for the CMB equations: 1) single unique
species to represent each source (tracer solution) (Miller etal., 1972); 2) linear programming
solution (Hougland, 1983); 3) ordinary weighted least squares, weighting only by uncertainty of
ambient measurements (Friedlander, 1973; Gartrell and Friedlander, 1975); 4) ridge regression
weighted least squares (Williamson and DuBose, 1983); 5) partial least squares (Larson and
Vong, 1989; Vong et a/., 1988); 6) neural networks (Song and Hopke, 1996); and 7) effective
variance weighted least squares (Watson etal.., 1984).
The effective variance weighted solution is generally applied because it: 1) theoretically
yields the most likely solutions to the CMB equations, providing model assumptions are met; 2)
uses all available chemical measurements, not just so-called "tracer" species; 3) analytically
estimates the uncertainty of the source contributions based on uncertainty of both the ambient
concentrations and source profiles; and 4) gives greater influence to chemical species with lower
uncertainty in both the source and receptor measurements than to species with higher
uncertainty. The effective variance is a simplification of a more exact, but less practical,
generalized least squares solution proposed by Britt and Luecke (1973).
CMB model assumptions are: 1) compositions of source emissions are constant over the
period of ambient and source sampling; 2) chemical species do not react with each other (i.e.,
they add linearly); 3) all sources with a potential for contributing to the receptor have been
1 -4
-------
identified and have had their emissions characterized; 4) the number of sources or source
categories is less than or equal to the number of species; 5) the source profiles are linearly
independent of each other; and 6) measurement uncertainties are random, uncorrelated, and
normally distributed.
The degree to which these assumptions are met in applications depends to a large extent on
the particle and gas properties measured at source and receptor. CMB performance is examined
genetically by applying analytical and randomized testing methods, and specifically for each
application by following an applications and validation protocol. The six assumptions are fairly
restrictive and they will never be totally complied with in actual practice. Fortunately, CMB can
tolerate reasonable deviations from these assumptions, though these deviations increase the
stated uncertainties of the source contribution estimates (Cheng and Hopke, 1989; Currie etal.,
1984; deCesar etal, 1985, 1986; Dzubay etal, 1984; Gordon etal, 1981; Henry, 1982, 1992;
Javitz and Watson, 1986; Javitz etal., 1988a, 1988b; Kim and Henry, 1989; Lowenthal etal,
1987, 1988a, 1988b, 1988c, 1992, 1994; Watson, 1979).
The formalized protocol for CMB application and validation (EPA, 2004 is applicable to the
apportionment of gaseous organic compounds and particles (Watson etal., 1994a; Fujita etal.,
1994). This seven-step protocol: 1) determines model applicability; 2) selects a variety of
profiles to represent identified contributors; 3) evaluates model outputs and performance
measures; 4) identifies and evaluates deviations from model assumptions; 5) identifies and
corrects model input deficiencies; 6) verifies consistency and stability of source contribution
estimates; and 7) evaluates CMB results with respect to other data analysis and source
assessment methods.
CMB is intended to complement rather than replace other data analysis and modeling
methods. CMB helps explain observations that have been made; it does not predict ambient
impacts from sources as do dispersion models. When source contributions are proportional to
emissions, as they often are for PM and VOC, then a source-specific proportional rollback
(Earth, 1970; Cass and McCrae, 1981; Chang and Weinstock, 1975; deNevers and Morris, 1975)
is used to estimate the effects of emissions reductions. Similarly, when a secondary compound
apportioned by CMB is known to be limited by a certain precursor, a proportional rollback is
used on the controlling precursor. The most widespread use of CMB over the past decade has
been to justify emissions reduction measures in PMio non-attainment areas. More recently,
CMB has been coupled with extinction efficiency receptor models (Lowenthal etal., 1994;
Watson and Chow, 1994) to estimate source contributions to light extinction and with aerosol
equilibrium models (Watson etal, 1994b) to estimate the effects of ammonia and oxides of
nitrogen emissions reductions on secondary nitrates.
CMB does not explicitly treat profiles that change between source and receptor (assumption
#2 above).1 Most applications use source profiles measured at the source, with at most dilution
to ambient temperatures and <1 minute of aging prior to collection to allow for condensation and
rapid transformation. Profiles have been "aged" prior to submission to CMB using aerosol and
gas chemistry models to simulate changes between source and receptor (Friedlander, 1981; Lin
and Milford, 1994; Venkatraman and Friedlander, 1994). These models are often overly
For a discussion of special approaches for treating secondary formation with EPA-CMB8.2, refer to EPA, 2004
1 -5
-------
simplified, and require additional assumptions regarding chemical mechanisms, relative
transformation and deposition rates, mixing volumes, and transport times.
CMB requires species with different abundances in different source types. The consistency
of a species abundance is more important than the uniqueness for source quantification. The
uniqueness is useful to identify which sources to include in a CMB analysis. Combining particle
and gas properties for source emissions, normalized to NMHC (non-methane hydrocarbon) or
PM2.5 mass emissions, could assist the apportionment of both VOC and PM2.5.
New analytical methods, however, such as isotopic abundances, specific organic
compounds, and single particle morphology may be used in CMB when they have been applied
to source and receptor samples to more precisely differentiate among contributions from
different sub-types. CMB performs tests on ambient data and source profiles that tell how well
source-type contributions can be resolved from each other for different combinations of source
profiles and chemical measurements.
CMB quantifies contributions from chemically distinct source-types rather than
contributions from individual emitters. Sources with similar chemical and physical properties
cannot be distinguished from each other by CMB. CMB model calculates source contribution
estimates for each individual ambient sample. The combination of source profiles that best
explains the ambient measurements may differ from one sample to the next owing to differences
in emission rates (e.g., some days may have wood-stove burning bans in effect and others will
not), wind directions (e.g., a downwind point source would not be expected to be contributing at
an upwind sampling site), and changes in emissions compositions (e.g., different gasoline
characteristics and engine performance in winter and summer may result in different profiles).
1.3 CMB Software History
The CMB receptor model was first applied by Winchester and Nifong (1971), Hidy and
Friedlander (1972), and Kneip et al. (1973). The original applications used unique chemical
species associated with each source-type, the so-called "tracer" solution. Friedlander (1973)
introduced the ordinary weighted least-squares solution to the CMB equations, and this had the
advantages of relaxing the constraint of a unique species in each source-type and of providing
estimates of uncertainties associated with the source contributions. The ordinary weighted least
squares solution was limited in that only the uncertainties of the receptor concentrations were
considered; the uncertainties of the source profiles, which are typically much higher than the
uncertainties of the receptor concentrations, were neglected.
The first interactive, user-oriented software for CMB was programmed in 1978 at the
Oregon Graduate Center in Fortran IV on a PRIME 300 minicomputer (Watson, 1979). The
PRIME 300 was limited to 3 megabytes of storage and 64 kilobytes of random access memory.
CMB Versions 1 through 6 updated this original version and were subject to many of the
limitations dictated by the original computing system. CMB7 was completely rewritten in a
1 -6
-------
combination of the C and Fortran languages for DOS-based PCs with floating-point
coprocessors, hard disk systems with tens of megabytes storage, and available memory of 640
kilobytes. CMB8 was developed but not officially released by EPA. CMB8 created a user
interface for CMB7 calculations using the Borland Delphi object oriented language.
EPA-CMB8.2 incorporates the upgrade features that CMB8 has over CMB7, but also
corrects errors/problems identified with CMB8 and adds enhancements for a more robust and
user-friendly system. The source code, executable, and test cases are available from EPA's
website (www.epa.gov/scram001).
1.4 Organization of the Users Manual
Section 1 introduces EPA-CMB8.2 and the scope of this manual. Section 2 describes
hardware requirements and related files, and describes how to install EPA-CMB8.2 on a
personal computer. Section 3 describes EPA-CMB8.2 key model features while Section 4
documents input and output file formats. Section 5 provides a step-by step tutorial of EPA-
CMB8.2 operations using a test case from example data sets provided with the model.
Performance measures are briefly described in Section 6, though their use in practical
applications is deferred to the application/validation protocol (EPA, 2004). Section 7 includes a
bibliography of CMB-related literature, including references cited throughout this manual.
1 -7
-------
-------
2. SOFTWARE INSTALLATION
This section describes the hardware requirements, computer programs, and installation
procedures for EPA-CMB8.2.
2.1 Hardware and Operating System
The minimum requirements for running EPA-CMB8.2 software are:
• IBM® PC compatible desktop, portable, or laptop computer with 386 processor and
16MB RAM
• Hard disk drive with 4 megabytes of storage
• Windows® 9x or higher operating system
The recommended hardware configuration is:
• IBM® compatible Intel Pentium® microcomputer with 64MB of RAM and 100MB
storage.
• Super VGA video graphics adapter and monitor.
• Graphics capable printer.
• Windows® XP or NT 4.0 operating system.
2.2 EPA-CMB8.2 Software and Related Files
The EPA-CMB8.2 software, as well as this manual, can be retrieved from the EPA's Support
Center For Regulatory Air Models (SCRAM) website:
www. epa.gov/scram001
The following files are available and can be downloaded as needed:
• EPA-CMB82.zip: A ZIPped file that contains the EPA-CMB8.2 executable, its
companion DLL file, and a help file (Section 2.3). This installation is compatible for all
applications using Windows® 9x or higher.
• EPA-CMB82 testzip: A ZIPped file that contains all files needed for the test case
described in Section 5 of this manual. Included are PM2.5 data (ambient and source
profile) from several sites in California's San Joaquin Valley Air Quality Study
(SJVAQS; Chow etal, 1990; 1992)
2- 1
-------
• EPA-CMB82 Manualpdf: An Adobe Acrobat® version of this users manual. A color
printer supporting PostScript fonts is recommended. Use this manual to learn EPA-
CMB8.2 features and operating methods.
• CMB Protocolpdf: An Adobe Acrobat® version of the Protocol for Applying and
Validating the CMB Model for PM2.s and VOC (EPA, 2004). A color printer supporting
PostScript fonts is recommended. This protocol is an important companion document
that provides useful guidance on interpreting CMB's diagnostic statistics and on
assessing the integrity of its apportionments.
• Source82.zip: A compressed file of EPA-CMB8.2 source code. This file preserves the
source code for further updates and allows it to be inspected for scientific verification.
Most users do not need this file.
EPA-CMB8.2 software is written in the Fortran, C++, and Delphi (Pascal) computer
languages. The Fortran and C++ code (Appendix B) are compiled into a main DLL
which is called at run time by the Delphi client (executable). This Delphi client handles
the user interface, and is produced using the Delphi 7 compiler from Borland Software
Corporation (Appendix D).
Note: The examples given in this manual are specific to the Windows® 95 (and higher)
installation and use of the 32-bit software. The compressed files for test data sets and any
documentation can be obtained by unZIPping the files listed above into a suitable folder. It is
recommended that you also create a folder \test case\ and extract EPA-CMB82 test.zip. This
test case is used in Section 5 of this manual.
2-2
-------
2.3 Installing EPA-CMB8.2 Software
Extract the compressed EPA-CMB82.zip into a suitable folder (e.g., \CMB8.2\). When
successfully installed, the following files will appear:
EPACMB82.exe Executable file.
EPACME82.hip2 The context-sensitive help file.
CMB82.dll Dynamically Linked Library file; called at runtime by the executable.
EPACMB82.ini3 Initializes file access information for use in the next session.
Using the right mouse button, a .Shortcut can created for the executable (EPACMB82.exe) and
moved to your desktop.
In addition, several "scratch" files will be created at run time and are stored temporarily in
this working directory. When the user advances off the Select Input Files screen, the following
proient (direct access) files are created in the executable directory:
AMBdirect.dat (binary file read by Fortran code; location of ambient data directory)
PROdirect.dat (binary file read by Fortran code; location of source profile data directory)
Once a Run is made, the following files are also created in this directory:
SumDirect.dat (binary file read by Fortran code; created only when a fit calculation is made)
$TEMPOUT. txt (ASCII buffer that stores results used by the Delphi User Interface)
These 4 scratch files are destroyed when the model is terminated normally.
This help file, accessed by the Delphi client at run time, is generated by a help compiler from the help project file
EPACMB82.hpj, which contains EPACMB82.rtf, etc. (provided with the source code).
This initialization file will appear once EPA-CMB8.2 is executed.
20
- J
-------
-------
3. EPA-CMB8.2 OPERATION
This section describes the EPA-CMB8.2 model commands. Although written using an
object oriented programming language, the previous (i.e., CMB8) interface used an old-style,
menu-driven approach. This design presented several buttons that each launched a separate
form, and some buttons were redundant on different forms. Unfortunately, this approach did not
take advantage of the underlying object oriented programming language of Delphi to create an
event-driven system.
EPA-CMB8.2 uses an approach that features a logically organized menu, a system tool bar
containing buttons for frequently used functions such as opening files and printing reports, a
tabbed page interface, and a status bar at the top of each screen. This type of presentation is
clear and eliminates the confusion associated with multiple buttons for the same functions. The
tabbed page interface consolidates the numerous forms dictated by the EPA-CMB8.2 source
code. A tabbed interface also provides users with visual clues regarding the logical progression
of steps necessary to run the model. These improvements provide a Windows® look-and-feel to
the CMB software that is familiar to most users.
Another benefit to this design is that the source code is much easier to maintain. Instead of
maintaining separate programming modules (called units in Delphi) for each form, the source
code is contained in a single unit. A list of run-time error messages has been compiled
(Appendix E).
3.1 Input files
Figure 3.1 shows the screen that first appears when EPA-CMB8.2 is launched.
EPA CMB 8.2 Start up
Chemical Mass Balance
Startup Options
(• jUse Control Fije!
f~" Select Individual Input Files
X Abort
Figure 3.1 Launching EPA-CMB8.2
Most users will have prepared a Control File (Section 4.2.1) for a particular CMB
application. Control Files are commonly used in air quality models to specify input files that
will be invoked during runtime. Input files listed in this file are described in Section 4.
Selection of this mode brings up a Windows® browse box for selecting a Control File. If a
Control File has not been prepared and the user simply wants to run CMB with particular (freely
3 - 1
-------
associated) input files, the other mode should be selected. Clicking on Cancel returns you to the
(previous) Startup dialog.
If Use Control File was selected, a browse dialog appears as shown in Figure 3.2. This
dialog allows the user to find a particular Control File. Once a Control File is chosen, the Select
Input Files window appears, as depicted in Figure 3.3.
Filename: ||Ndemo1.in8
Files of type: EPA-CMB8.2 Control File (INK.in8)
Figure 3.2 Browse Dialog for Selecting a Control File
In the Select Input Files window, the name of the Control File4 is prominently displayed
across the top, and the various input files it directs appear below. The Control File in use during
any session also appears on all screens in the status bar at the top. Note that even though a
Control File has been selected at this point, another one can easily be chosen via its browse
dialog. As mentioned earlier, different Control Files can be loaded repeatedly during the same
session, obviating the need to terminate the application. For any particular Control File, any of
the input files may also be changed or removed via respective browse functions. EPA-CMB8.2
also gives the user the option from this screen to create a new Control File (with the new input
file(s)) by using the File/Save function in the upper left-hand corner. Note, however, that once
the initial Control File has been modified, its name will no longer appear on the top of the screen
and must be reselected (via browse), assuming that a new file was created (saved).
See Section 4.2.1 for more details on the structure and function of the Control File.
3 -2
-------
H EPA-CMBB.2
File Help
Control File: INdemol.inS Samples:
Species:
Output Format: ASCII (.txt) -> Default
Input Files! Options | Samples | Species | Sources | Results
SELECT INPUT FILES
CONTROL FILE
|lNderncil.in3
REQUIRED INPUT FILES
Ambient Data
Source Profiles
OPTIONAL SELECTION FILES
Ambient Data
Species
Source Profiles
ADdemo.dbf
|pRdemo.dbf
&• Bro
JADdemolei
& Browse..
jSPdemo.sel
|pRdemo,sel
Figure 3.3 Select Input Files Screen
If at startup (Fig. 3.1) the user opts not to use a Control File, a null Select Input Files screen
appears as illustrated in Figure 3.3, except that all the input file names are absent (including the
Control File name). Files are selected via their respective browse dialogs and can be located in
any directory. Note that EPA-CMB8.2 will accept *.csv, *.dbf, and *.wks data formats for
ambient sample (AD*) and source profile (PR*) files, which are the only input files required by
the model. The selection files are optional.
Even if a particular Control File has been loaded, different input files can be selected via
their respective browse dialogs. If this is done, the user can create a new Control File by
clicking the File icon at the top and following the prompts. This same functionality applies to
the optional selection files (Sections 3.3 - 3.5).
Note that the 'Help' button on the top toolbar presents the option 'Contents'. Clicking this
loads and presents the on-line help for EPA-CMB8.2. You can navigate through this system to
find help on a variety of operational topics. Note that pressing Fl on any screen presents help
for that screen (on-line help is in this sense context-sensitive).
3 -
-------
Note that clicking on the Help button also presents the option "About". This button brings
down a banner page as shown in Figure 3.5. It is good practice to verify this against the postings
on EPA's modeling web site (Section 2.2) to assure that the most recent revision is being used.
EPA-CMB8.2 is being continually improved as users respond with recommendations or
difficulties. There is also information on the model developers, as well as EPA project officers.
£1 About EPA-CMBB.2 _ "D *X
/5^€DSr^v Chemical Mass Balance
/! £ ft \\ Receptor Model
lo j)
^^** jf/ EPA-CMB8.2
U
eweiupcrs
Desert Research Institute
John G, Watson
Norman F. Robinson
Pacific Environmental Services
Vicky J, Kriegsman
Robert A. Wagoner
Ideal Software
Richard F. Albury http://www.idealsw.com John V, Scalco
E
CH i eiimiLai representatives
C. Thomas Coulter
Office of Air Quality Planning and Standards
Research Triangle Park, NC 27711
(919)541-0332
Coulter , Tom@epa , aov
( y
Office of Research and Development
Research Triangle Park, NC 27711
(919)541-3154
Lewis , CharlesW@epa , go v
'OK ||
Figure 3.4 Banner Page for EPAOCMB8.2
3 -4
-------
3.2 Options
Several options are available in EPA-CMB8.2 that are selected from the Options tab (Fig.
3.4), where various values and selections may be changed from their default values. Note that
the Control File name is clearly displayed on the status bar (top).
HEPA-CMB8.2
File Help
-I-IH'
1
Control File: INdemol.inS
Output Format; A5CII (.txt) -> Default
Input Files Options | Samples ] Species | Sources | Results |
SET OPTIONS FOR CURRENT SESSION
Iteration Delta
Maximum Source Uncertainty (*^&)
Minimum Source Projection
Decimal Places Displayed
Measurement Units
Output File Format
Britt and Luecke
Source Elimination
Fit Measure Weights: Chi Square
R Square
Percent Mass
Fraction Estimate
Reset to Default
Figure 3.5 Options for Current Session
Iteration Delta. This parameter sets the maximum number of iterations EPA-CMB8.2 will
attempt to arrive at a solution. If no convergence can be achieved, there is probably excessive
collinearity for this sample and combination of fitting sources. Its value is adjusted via the
spinners. (Must be >0; no theoretical upper limit; default = 20)
Maximum Source Uncertainty / Minium Source Projection. These parameters allow the
eligible space collinearity evaluation method of Henry (1992) to be implemented with each CMB
calculation (Section 6.2). The eligible space method uses: 1) maximum source uncertainty; and
2) minimum source projection on the eligible space. The maximum source uncertainty is a
threshold expressed as a percentage of the total measured mass and is adjustable via the spinners
(default = 20% ; acceptable range 0 - 100). The minimum source projection is set to a default
value of 0.95 (acceptable range 0.0 - 1.0), but can be changed in the display field. See Section
6.1.2 for more discussion of eligible space.
Decimal Places Displayed. This parameter sets the number of decimal places displayed in the
output window and output files. This depends on the units used in the input data files. For
example, data reported in ng/m3 require fewer decimal places than values expressed in |ig/m3.
Reducing this value from 5 to 4 accommodates most PM2.5 mass and chemical concentrations
expressed in |ig/m3. A value of 1 or 2 is best for concentrations expressed in ng/m3 or for VOC
3 -5
-------
expressed in ppbC or |ig/m3. This setting affects the display columns for source contributions
estimates, measured species concentrations (ambient samples and source profiles), calculated
contributions by species, as well as for inverse singular values. This parameter may be adjusted
by using the spinners. The default value is 5 and the maximum value is 6.
Units. The units used for reporting results may be changed via a pull-down menu. Other typical
units are available, or one may be created (the number of characters is limited to 5 or less).
Output File Format. The file format for spreadsheet-type output is selected in the pull-down
box. As discussed in Section 4.5, the default is ASCII (txt); comma-separated value (CSV) is
also available (which ports nicely to Microsoft Excel®). This selection is echoed on the status
bar at the top of the screen.
Britt and Luecke. Checking this box applies the Britt and Luecke (1973) linear least squares
solution that is explained by Watson et al. (1984) when applied to CMB calculations. This
option allows the source profiles used in the fit calculation to vary, and enables a general
solution to the least squares estimation that includes uncertainty in all the variables (i.e., the
source compositions as well as the ambient concentrations). The default (option disabled) is the
same approximation to the Britt-Luecke algorithm used in CMB7. Note that while the exact
Britt-Luecke algorithm must generate a fit whose x2 value is equal to or better (i.e., smaller) than
that from the approximation algorithm, there is no guarantee that the solution with the better x2
will be superior in terms of its physical meaning. Invocation of this option affects the fit
obtained and, as in the case of EPA-CMB8.2's new treatment of collinearity, user experience is
necessary to judge the utility in exercising the new Britt-Luecke algorithm. The Britt-Luecke
algorithm, as implemented in EPA-CMB8.2. has not undergone comprehensive testing, and is
not recommended for inexperienced users. Its inclusion as an option is mainly intended to
provide the opportunity for interested advanced users to perform research investigations needed
to establish its future viability. Note also that species concentrations that will appear in the Main
Report reflect this algorithm's modification to the source profile matrix. The individual species
concentrations for each source which appear in the (spreadsheet-type) output file are calculated
using the UNmodified source profile matrix (and therefore will be different). This was done to
maintain continuity with CMB 6.
Source Elimination. Checking this box eliminates negative source contributions from the
calculation, one at a time. After each fit attempt, if any sources have negative contributions, the
source with the largest negative contribution is eliminated and another fit is attempted. This
process is repeated until EPA-CMB8.2 finds no sources with negative contributions. Invocation
of this option affects the fit obtained by effectively removing collinear sources (Section 6.1.2).
Best Fit. Checking this box causes the program to cycle through the corresponding pairs (same
array index) of fitting species and source profile arrays specified in the source and species
selection input windows until the best composite Fit Measure has been achieved. When Best Fit
is invoked, EPA-CMB8.2 ignores any arrays of species and sources that may have been
selected. The first fitting species array is paired with the first fitting sources array, and so on.
EPA-CMB8.2 only attempts a search for a best fit among available corresponding pairs. Any
arrays without a corresponding array to make a pair are ignored. The fit with the largest Fit
Measure is then displayed and becomes the current fit. After a Best Fit has been made, the
fitting species and fitting sources arrays will be tagged (highlighted) in their respective windows.
The Fit Measure algorithm is described in Section 6.3.
Fit Measure Weight. These are the weights (coefficients) applied to each of the performance
measures chi square, r-square, percent mass, and fraction of eligible sources (number in eligible
space divided by number of fitting sources). Adjustment of these weights is not enabled in EPA-
CMB8.2 unless Best Fit is invoked. Positive values between 0 and 1 may be entered by typing
into the appropriate display fields. Defaults are 1.0 for each performance measure weight. See
Section 6.1 for more discussion.
3 -6
-------
3.3 Select Ambient Data Samples
The screen for selecting a subset of samples on which source apportionment will be
performed is shown in Figure 3.6. If an (optional) AD*.sel file is being directed from the
HEPA CMB8.2
File Hel
M|«
3
>•
Input Files
M
i|
i
SIM
Options Samples
Control File: INdemol.inS Samples:! Species: 20 Sources: 6 Output Format: Comma-separated Value (.CSV)
| Species ] Sources | Results |
AMBIENT DATA SAMPLES
SELECTED
SITE
BAKERS
BAKERS
BAKERS
BAKERS
BAKERS
BAKERS
BAKERS
BAKERS
BAKERS
BAKERS
BAKERS
BAKERS
BAKERS
BAKERS
BAKERS
BAKERS
BAKERS
BAKERS
BAKERS
BAKERS
BAKERS
BAKERS
BAKERS
BAKERS
mUHl BAKERS
BAKERS
BAKERS
BAKERS
BAKERS
DATE
06/20/6:3
07/02/88
07/26/88
08/07/88
08/19/88
08/25/88
08/31/88
09/06/88
09/12/88
10/18/88
11/11/88
11/17/88
11/23/88
11/29/88
12/05/88
12/11/88
12/17/88
12/23/88
12/29/88
01/04/89
01/10/89
01/16/89
01/22/89
01/28/89
02/03/89
02/09/89
02/15/89
02/21/89
02/27/89
DUR
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
START
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
SIZE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
TMAC
17.2738
23.5425
26.7742
21.9185
22.6664
30.1214
25.3387
30.8824
27.6091
47.1986
22.6296
21.5811
14.0767
44.4116
136.4439
164.4591
43.7105
32.7483
78.0441
58.5364
49.9733
108.6817
87.4695
109.5098
10.5159
38.0024
57.2517
35.7633
53.5523
TMALI
0,992
1,2744
1.425
1.2008
1.2339
1.5825
1.3569
1.6189
1.4641
2.4114
1.2314
1.1833
0.8563
2.2761
6.8398
8.2375
2.2394
1.7062
3.9326
2.9669
2.5453
5.4554
4.4002
5.4966
0.714
1.9605
2,9033
1.8521
2.7206
N3IC
0,2816
0.8306
0.2054
0.4096
0.5093
0.613
0.4207
0.667
0.6924
N3IU
0.1715
0,1761
0.1715
0.1732
0.1729
0.173
0.1716
0.1736
0.1745
4.2687 0.276
3.47 0,2431
3.174| 0.2328
2.6161
13.3413
48.8061
61.2912
10.4595
5.7176
15.7947
13.8542
14.5694
27.4688
25.1026
38.5835
1.3636
11.2574
20.8357
8.3371
18.0102
0.2153
0.6887
2.532
3.138
0.5503
0.3328
0.808
0.7134
0.7481
1.5312
1.4266
2.0441
0.1839
0.5882
1,0557
0.4504
0.9166
S4IC
2.8204
3.2224
3.4881
3.1228
3.7134
3.5371
3.4671
S4IU
N4TC |l A
0.1612 1.1415
0,1791 1.1149
0.1911
0.1748
0.2014
0.1932
0.19
2.7646 0.1586
3.2503 0.1803
5,4811 0,2853
1.6402 0.113
1.1997 0.0982
1.4961
3.9645
8.0547
10.0051
4.5647
3.4971
7.7026
7.2325
3.8032
6.6259
4.5252
6.7365
0.1081
0.2131
0.4102
0.5063
0.2412
0.1914
0.393
0.3699
0.2054
0.3404
0.2394
0.3457
1.358
1.2689
1.4265
1.4201
1.4299
1.1528
1.363
3.5247
1,7894
1.5184
1.397
5.7479
18.0959
21.8674
5.2344
3.4681
7,5614
7.3213
5.8227
12.0771
9.6044
14.9511
0.9696 0.0918 0.799
2.9Z91 0.1659 4.6234
3,7161 0,2015 7.4596
2.9483 0.1667 3.6295
3.9818 0.2138 7.614 V
•
Select All Samples
De-select All Samples
View Selected
Hide Data
View Graph
Figure 3.6 Ambient Data Selection
Control File, one or more samples may appear selected initially (see Section 4.2.4). Otherwise,
individual samples are "tagged" by clicking in the respective fields under "SELECTED".
Clicking again deselects the sample. Use of Select /Clear All Samples may also be used to help
establish the desired list of samples. Note that a counter displays on the status bar (top) the
number of samples selected at any given time. As indicated, the collection date, duration, start
hour, and size fraction (particles) are displayed for each sample. Show/Hide Data toggle
between modes in which speciated data (alternating concentration and uncertainty) for the
samples are either shown or masked. Toggling between View Selected/View All determines
whether data will be displayed for selected (tagged) samples only, or for all samples in the list.
Clicking View Graph will provide a bar chart for any sample (selected or not) for which any
field is filled with blue (note that because of the physical constraints of the graph, some
distortion will occur if the number of species exceeds -25). This graph is useful to verify that
input data files have been properly read. Note that use of the "VCR" control buttons on the top
toolbar can help navigate down a long list of samples.
3 -7
-------
There will be times when you data will include (dichotomous) measurements for Fine and
Coarse size fractions. When this is the case and you want to apportion Fine and Coarse samples
in a batch run, you must take care that compatible sample pairs (matched by site ID, date,
sampling duration, and start hour) are selected. These pairs may be Fine/Coarse ... or
Coarse/Fine ...
Upon exit, EPA-CMB8.2 will detect changes that may have occurred in arrays initialized by
the optional input files during the session:
1) If an initial array of selected samples (as directed by AD*.sel) has been modified, the user will
be prompted and asked if AD*.sel should be updated (overwritten). The file may also be
conveniently renamed.
2) If no selection file was in use but an array of tagged samples has been created during the
session, the user will be prompted and asked whether a samples selection file should be saved. If
so, an appropriate name (e.g., AD*.sel) should be entered; the extension .sel will be appended
automatically.
3.4 Select Fitting Species
Fitting species are used in the calculation of source contribution estimates. Species not
included in this calculation are termed floating species (Section 4.3.2). The comparison of
calculated and measured values for floating species is part of the model validation process.
Fitting species should be selected that are major or unique components of the source types
influencing the receptor concentrations. The screen for selecting fitting species is shown in
Figure 3.7. This screen is initiated by data read from the (optional) species selection file
H EPA-CMBB.2 EDflDS
File
Help
' 1 -
Input Files ]
5PECIE
IffiBM
S4IC
N4TC
KPAC
NAAC
ECTC
OCTC
ALXC
5IXC
5UXC
CLXC
KPXC
CAXC
TIXC
VAXC
CRXC
MNXC
FEXC
NIXC
cuxc
ZNXC
BRXC
PBXC
n | ig
Options
Ltd
Samples Species
FF
5PECIENAMEJ 1
NO3
SO4
MH4
K-S
NA
EC
OC
AL
SI
5
CL
K
CA
TI
V
CR
MN
FE
NI
CU
ZN
BR
PB
*
2
3
4
Control File: INdemol.inS Samples: 1 Species: 20 Sources: 6 Output Format; Comma-separated Value (.csv)
Sources Resuts |
FTING SPECIES ARRAYS
5
6
7
8
9 | 10 [COMMENT
I 1 |*|*|*|*|*|*|
»
*
*
*
*
*
*
»
*
«
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
Nitrate by 1C (ug/m3)
Sulfate by 1C (ugjfmS)
Ammonium by AC (ug/m3)
Soluble Potassium by AA (ug/m3)
Soluble Sodium by AA (ug/m3)
Elemental Carbon by TOR (ug/m3)
Organic Carbon by TOR (ug/rn3)
Aluminum by XRF (ug/m3)
SulFur by XRF (ug|Tm3)
Chloride by XRF (ug/m3)
Calcium by XRF (ug/m3)
Titanium by XRF (ug/m3)
Vanadium by XRF (ug/m3)
Chromium by XRF (ug/m3)
Manganese by XRF (ug/m3)
Iron by XRF (ug/m3)
Nickel by XRF (ug/m3)
Copper by XRF (ug/m3)
Zinc by XRF (ug/m3)
Bromine by XRF (ug/m3)
Lead by XRF (ug/m3)
Select All Array 1
Clear All Array 1
View Selected
Figure 3.7 Fitting Species Arrays
3 -
-------
(SP*.sel; see Section 4.2.4). Species contained in the selection file are listed down the left-
handside and a field of up to 10 arrays is provided. At startup, the first array (in a series) will
always be initially selected as a default. Other arrays may be selected by clicking on the array
index (1 - 10). Within a given array that is first activated by clicking its index number, species
may be added or removed by clicking in the appropriate field. Select/Clear All Array X may
also help in configuring a selection array. Note that use of the "VCR" control buttons can help
navigate down a long list of species, and that a counter displays on the status bar (top) the
number of species tagged for any selected array. For a selected array, toggling between View
Selected/View All determines which species will be displayed. This can be handy for a long list
of species that would be impossible to display on the screen. If comments are provided in the
selection file, they will be displayed on the right-hand side.
Multiple arrays for fitting species are useful when CMB calculations are performed on
samples from several locations or during different times of the year that have different
contributors. They are also used by the Best Fit option to cycle through different source
combinations until the weighted Fit Measure is optimized (Section 3.2).
Upon exit, EPA-CMB8.2 will detect changes that may have occurred in arrays initialized by
the optional input files during the session:
1) If an initial array of selected species (as directed by SP*.sel) has been modified, the user will
be prompted and asked if SP*.sel should be updated (overwritten). The file may also be
conveniently renamed.
2) If no selection file was in use but an array of tagged species has been created during the
session, the user will be prompted and asked whether a samples selection file should be saved. If
so, an appropriate name (e.g., SP*.sel) should be entered; the extension .sel will be appended
automatically.
3 -9
-------
3.5 Select Fitting Sources
Fitting source profiles are included in the CMB calculation. The user should select profiles
that represent the emissions most likely to influence receptor concentrations. Several profiles
may be available that represent the same source type, but only one of these is usually used as a
fitting source. Profiles of similar chemical composition are often found to be collinear when two
or more are selected as fitting sources. The screen for selecting fitting sources is shown in
Figure 3.8. This screen is initiated with data read from the (optional) species selection file
(PR*.sel; see Section 4.2.4). Source profiles contained in the selection file are listed down the
left-hand side
H EPA-CMBB.2 QUO®
File Hel
1
3
Input Files |
JPNO
KB
5JV002
JSJV003
SJV004
3JV005
SJV006
SJV007
SJV008
S.1V009
SJV010
SJV011
SJV012
SJV013
SJV014
SJV015
SJV016
SJV017
SJV018
SJV019
5JVOJJO
SW021
SJV022
SJV023
S:V024
SJV025
SJV026
SJV027
3JV026
<
SID
M
«| A I
Options j Samples ]
SOIL01
50IL03
50IL04
50IL05
50IL06
SOIL07
SOIL08
SOIL09
SOIL 10
SOIL 11
SOIL 12
50IL13
5OIL14
SOIL 15
SOIL 16
5OIL17
BAMAJC
MAMAJC
MAFISC
MADIEC
BAAGBC
ELAGBC
FRCONC
STAGBC
VIAGBC
VIDAIC
5FCRUC
CHCRUC
SIZE
1
2
Spe
f
Control F
cies Sources
ZITTING SOL
3
4
5
e: INdemol
Results
JRCES A
6
7
n8 Samples:! Species: 20 Sources: 6 Output Format: Comma-separated Value (.csv
RRAYS
8
9
10
FINE
FINE ~|~~»
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
*
*
*
*
*
*
*
*
*
*
*
*
COMMENT
STOCKTON AGRICULTURAL 50
FRESNO PAVED ROAD
VISALIA AGRICULTURAL SOIL
VISALIA AGRICULTURAL SOIL
VISALIA SAND AND GRAVEL
VISALIA URBAN UNPAVED
VISALIA PAVED ROAD
BAKERSFIELD AGRICULTURAL
BAKERSFIELD AGRICULTURAL
BAKERSFIELD UNPAVED ROAD
BAKERSFIELD PAVED ROAD
BAKERSFIELD WINDBLOWN UR
BAKERSFIELD AGRICULTURAL
BAKERSFIELD AGRICULTURAL
BAKERSFIELD UNPAVED ROAD
TAFT UNPAVED ROAD
BAKERSFIED CORDWOOD, MAJ
MAMMOTH LAKES CORDWOOD
BAKERSFIELD CORDWOOD, FI5
MAMMOTH LAKES DIE5EL TOUR
BAKERSFIELD AGRI. BURN (W
EL CENTRO AGRI. BURN (WHE
FRESNO HIGHWAY 40 CONSTR
STOCKTON AGRI. BURN (WHEA
VISALIA AGRI BURN (WHEAT)
VISALIA DAIRY/FEEDLOT DUS
SANTA FE CRUDE BOILER (WE
CHEVRON RACETRACK CRUDE
N3IC
N3IU
0.0027 0.0047
0 0.0036
0
0.0003
0
0.0015
0.0018
0
0.0011
0.0005
0.0011
0.0013
0
0.0014
0.0024
0
0.003
0.0026
0.0658
0.0029
0.0051
0.0023
0.0096
0.003
0.0058
0.0023
0.0039
0.0163
0.0018
0.0025
0.0046 0.0012
0.0017 0.0004
0.0003 0
0.0006 0.0001
0.0065 0.0016
0.0038 0.001
0.0176 0.0028
0.0042 0.0025
0,0035 0.0007
0.0926
0
o
0.039
0
0
*
v
Select All Array 1
Clear All Array 1
View Selected
Hide Data
View Graph
Figure 3.8 Fitting Sources Arrays
and a field of up to 10 arrays is provided. As with fitting species, arrays are selected by clicking
on the array index (1 - 10). Within a given array, source profiles may be added or removed by
clicking in the appropriate field. Select/Clear All Array X may also help in configuring a
selection array. Note that use of the "VCR" control buttons can help navigate down a long list of
sources, and that a counter displays on the status bar (top) the number of source profiles tagged
for any selected array. For a selected array, toggling between View Selected/View All
determines which source profiles will be displayed. This can be handy for a long list of sources
that would be impossible to display on the screen. If comments are provided in the selection file,
they will be displayed on the right-hand side.
3- 10
-------
As for ambient samples, clicking View Graph will provide a bar chart for any source profile
(selected or not) for which any field is filled with blue (note that because of the physical
constraints of the graph, some distortion will occur if the number of species exceeds -25). This
graph is useful for visual inspection; it helps to verify that input data files have been properly
read and to identify abundant components in each profile. View Grid returns to the array screen.
Upon exit, EPA-CMB8.2 will detect changes that may have occurred in arrays initialized by
the optional input files during the session:
1) If an initial array of selected source profiles (as directed by PR*.sel) has been modified, the
user will be prompted and asked if PR*.sel should be updated (overwritten). The file may also
be conveniently renamed.
2) If no selection file was in use but an array of tagged source profiles has been created during
the session, the user will be prompted and asked whether a samples selection file should be
saved. If so, an appropriate name (e.g., PR*.sel) should be entered; the extension .sel will be
appended automatically.
3.6 Calculation Results
Once options have been set, one or more samples selected, as well as a suitable array of
fitting species and source profiles, EPA-CMB8.2 is ready to do a calculation. The Results screen
shown in Figure 3.9 is where fitting results and statistics are reported for examination. When the
screen is first viewed, the user is prompted to click on Run in order to initiate a calculation.
Information
Press RUN to generate results.
OK
Figure 3.9 Calculation Results Tab
When Run is invoked, EPA-CMB8.2 performs the least-squares estimation of source
contribution estimates and performance measures on the selected sample data using the
designated fitting species and source profiles. Note also that if more fitting sources than fitting
species have been selected, a warning appears and the user is forced to reconfigure.
3- 11
-------
3.6.1 Main Report
When Run in invoked, EPA-CMB8.2 attempts fits of all samples selected. If more than one
sample is selected, the model runs in a batch mode5 As each sample is apportioned, results are
successively written to the output window as part of the Main Report (Figure 3.10). For any
H EPA-CMBB.2
File Help
•• M ijj H Control File : INdemol.
Input Files Options Samples | Species Sources Results
SAMPLE Site: BAKERS Date: 02/03/89 Duration: 24
OPTIONS Iteration Delta: 20 l~~ Britt-Luecke
Maximum Source Uncertainty (9 b): 20 f~ Source Elimin
Minimum Source Projection: 0.95
Decimal Places Displayed: 5 Species Fit Arra;
Units: Mg/m5 Sources Fit Array
Main Report \ Contributions by Species
FITTING STATISTICS:
R SQUARE 0.95 % MASS
CHI SQUARE 0.98 DEGREES FREEDOM
SOURCE CONTRIBUTION ESTIMATES:
SOURCE
EST CODE NAME SCE([jg/m = ) Std Err Tstat
YES SJV011 SOIL12 0.50003 0.11297 4.42634
YES SJV017 BAMAJC 1.84022 0.42100 4.37107
YES SJV027 SFCRUC 1.86010 0.22181 8.38614
YES SJV036 MOVES2 3.02327 0.77463 3.90287
YES SJV051 AMSUL 0.81674 0.21987 3.71471
YES SJV054 AMNIT 1.92641 0.27654 6.96623
9.96677
MEASURED CONCENTRATION FOR SIZE: FINE
10. 5+- 0.7
Eligible Space Display
BBS
n8 Samples: 1 Species: 20 Sources: 6 Output Format: Comma-separated Value (.csv
Start Hour: 0 Size: FINE
F Best Fit Fit Measure Weights
ition
R Square i
: 4
. 5 Fraction Estimate i
MPIN Matrix
*
94.8
14
ELIGIBLE SPACE DIM. = 6 FOR MAX. UNC. = 2.10313 (20.* OF TOTAL MEAS. MASS)
1 / Si ngul ar Value
0.11100 0.18607 0.22939 0.29361 0.41206 0.78006
NUMBER ESTIMABLE SOURCES = 6 FOR MIN. PROJ. = 0.95
PROJ. SOURCE PROJ. SOURCE PROJ. SOURCE PROJ. SOURCE PROJ. SOURCE
1.0000 SJV011 1.0000 SJV017 1.0000 SJV027 1.0000 SJV036 1.0000 SJV051
1.0000 SJV054
ESTIMABLE LINEAR COMBINATIONS OF INESTIMABLE SOURCES
COEFF. SOURCE COEFF. SOURCE COEFF. SOURCE COEFF. SOURCE SCE Std Err
___ __ __ __ __ __
~
Result 1 of 1
Delete Current
Delete All
Run
Figure 3.10 Calculation Results - Main Report
(current) result displayed in the output window, the header at the top reflects basic identifying
information pertaining to the sample that has been apportioned and an echo of the options
settings. The species and source profile fitting array indices are also indicated. If more than one
sample was apportioned, the number will be reflected in the box in the upper right-hand corner.
Viewing any sample result in a series is quite easy with the use of the VCR buttons on the
toolbar. Using appropriate buttons, the current or all results may be deleted from the output
window. The report may be printed for the current or all sample results via the print button on
Note that in batch mode, certain warning /error messages that require user intervention are suppressed.
3- 12
-------
the toolbar; it is also possible to print to Adobe Acrobat® (PDF). A file may be created via the
Print To File option. The header for this file is embellished with an echo of all input files used in
the calculation - another feature of EPA-CMB 8.2.
In analyzing data that may have been collected in sampling networks that employ
dichotomous samplers, it is common to input speciated data for two complementary size
fractions, fine and coarse (Section 4.2.2). If EPA-CMB8.2 detects that it has such a sample pair
for a given period, it will assume them to be complementary. When a fit is performed, an
additional report is created in which results are summed for the fine and coarse fractions to give
the Total (frequently = PMio). In this case, the fitting array indices are the same as for the fine
and coarse component samples, as is the value for Degrees of Freedom. For % Mass explained,
the value is determined as follows:
% Mass = 7 — ~ ~'cak'
+
2>L
where SF = size fraction
For a given sampling site and period, the samples tagged for analysis must be presented as
alternating dichotomous pairs, i.e., Fine/Coarse; Fine/Coarse; Fine/Coarse; ... or Coarse/Fine;
Coarse, Fine; Coarse/Fine ... The 'Total' report is always appended to that for the 2nd in the
dichotomous pair.
In the Main report are several information blocks which present performance measures
(Section 6.1.1). First are fitting statistics: r2, x2, percent mass explained, and degrees of freedom.
Next is a block that presents the most basic results from EPA-CMB8.2: source contribution
estimates. For each source selected is presented a source contribution estimate (SCE) in user-
chosen units (Section 3.2), standard errors, and values for T-stat. The series of SCEs are
summed to provide a convenient check on the % mass explained value (EPA-CMB8.2 feature):
% Mass =
SCEs
total measured cone.
XWO
In making this check, be aware that the full uncertainty of the measured concentration value is
not displayed in the Main Report. The field 'EST' under ' SOURCE' indicates (YES or NO)
whether a source's contribution was estimable in EPA-CMB8.2's attempt at a fit using the
settings in Options. The next block is the Eligible Space Collinearity Display: an echo of the
measured concentration and error for the sample, eligible space dimension for the chosen
maximum uncertainty (Section 3.2), inverse singular values, the number of estimable sources for
the chosen minimum source projection (Section 3.2), and estimable linear combinations of
3- 13
-------
inestimable sources. The concepts of estimable sources and estimable space are discussed in
Section 6.1.2. Finally, there is a block detailing species concentrations. Shown for each species
(fitting species are tagged with asterisks) is its measured and calculated mass and uncertainty,
the ratio of calculated/measured mass (± uncertainty factor), and the ratio of the signed residual
(calculated - measured mass) to the uncertainty of that residual. See Section 6.1.3 for more
details.
Of note is the way EPA-CMB8.2 handles missing values in source and receptor files
(designated by -99. in place of the value). When a fitting species value is missing from either an
ambient sample or source profile, that species is automatically removed from the calculation and
the species selection flag (ordinarily an asterisk) is set to "M" in the report output file. See
Section 4.2.2 & 4.2.3; Appendix F.
For any of the information blocks in the main report, the presence of a series of asterisks for
a numerical value field represents an overflow condition. Reducing the number of decimal
places displayed (Section 3.2) should correct the problem.
3.6.2 Contributions by Species
Beyond the traditional apportionment results generated by EPA-CMB8.2, it is also of
interest to analyze the way in which pollutant mass is distributed among sources by species.
This distribution is presented in the report Contributions by Species (Figure 3.11). This report
HEP* CHBB.2
File Help
| *• | w | El | & \ Control File: INdemol.inB
BUS
Samples: 1 Species: 20 Sources: 6 Output Format: Comma-separated Value (.csv)
Input Files | Options | Samples | Species | Sources Results
SAMPLE Site: BAKERS Date: 02/03/89 Duration: 24 Start Hour: 0 Size: FINE
OPTIOM5 Iteration Delta: 20 F Britt-Luecke T Best Fit
Maximum Source Uncertainty (ti): 20 |~~ Source Elimination Cni Square i
Minimum Source Projection: 0.95
Decimal Places Displayed: 5 Species Fit Array: 4
Units: ug/m^ Sources Fit Array: 5
Main Report i Contributions by Species i MPIN Matrix ^
SOURCE NAME
SPECIES CALCULATED MEASURED SOIL12 BflMflJC SFCRUC MOVES!
TMAC 9.966 10.5159 0.048 0.175 0.177 0.287
N3IC 1.562 1.3636 0.000 0.006 0.000 0.044
S4IC 1.097 0.9696 0.006 0.027 0.390 0.097
N4TC 0.659 0.7990 0.000 0.002 0.000 0.000
KPAC 0.077 0.0687 0.038 1.069 0.016 0.000
NAAC 0.020 0.0829 0.051 0.031 0.171 0.000
ECTC 1.937 2.4625 0.003 0.119 0.000 0.665
OCTC 2.421 1.9178 0.049 0.428 0.001 0.785
ALXC 0.038 0.0121 3.021 0.000 0.000 0.192
SIXC 0.126 0.0634 1.529 0.000 0.003 0.456
SUXC 0.343 0.3690 0.007 0.026 0.275 0.085
CLXC 0.038 0.0377 0.048 0.932 0.010 0.023
KPXC 0.084 0.0907 0.108 0.810 0.008 0.003
CAXC 0.027 0.0266 0.874 0.048 0.042 0.082
TIXC 0.002 0.0048 0.490 0.000 0.039 0.006
VAXC 0.015 0.0113 0.013 0.000 1.350 0.003
CRXC 0.000 0.0000 -9.999 -9.999 -9.999 -9.999
MNXC 0.001 0.0020 0.250 0.000 0.093 0.423
FEXC 0.032 0.0482 0.593 0.000 0.081 0.001
NIXC 0.014 0.0168 0.003 0.000 0.875 0.000
CUXC 0.000 0.0226 0.004 0.000 0.000 0.007
ZNXC 0.009 0.0172 0.067 0.096 0.281 0.093
BRXC 0.008 0.0039 0.013 0.047 0.000 2.047
PBXC 0.012 0.0219 0.046 0.000 0.000 0.515
< |
x
AMSUL AMNIT
0.078 0.183
0.000 1.095
0.612 0.000
0.279 0.544
0.000 0.000
0 . 000 0 . 000
0 . 000 0 . 000
0 . 000 0 . 000
0 . 000 0 . 000
0 . 000 0 . 000
0.537 0.000
0.000 0.000
0.000 0.000
0.000 0.000
0.000 0.000
0.000 0.000
-9.999 -9.999
0.000 0.000
0.000 0.000
0.000 0.000
0.000 0.000
0 . 000 0 . 000
0 . 000 0 . 000
0.000 0.000
>l
Result 1 of 1
Delete Current
Delete All
Run
Figure 3.11 Calculation Results - Contributions by Species
3- 14
-------
provides one more dimension to the Species Concentrations block in the Main Report. These
results are useful when source contributions to species other than total mass are of interest. The
report also indicates which sources are the major and minor contributors to each species. Since
values in this report are ratios of calculated species concentrations to the measured total species
concentration, multiplying the values by their respective measured value and summing will
confirm the values listed in the sum of calculated species contributions column (left-hand side).
For convenience, both the calculated and measured columns are from the Main Report are
reproduced here - another feature of EPA-CMB8.2. As with the Main Report, a print-out of
Contributions by Species may be obtained for the current sample result via the print button on
the toolbar, and a file may be created via the Print To File option. For more information see
Section 6.2.
3.6.3 Modified Pseudo-Inverse Normalized (MPIN) Matrix
Another report that may be of interest is the Modified Pseudo-Inverse Normalized (MPIN)
matrix (Figure 3.12). The MPIN matrix identifies which fitting species have the largest
influence on the source contribution estimates from each profile (Section 6.2). Examining these
weights suggests sensitivity tests to determine the extent to which source contributions vary with
changes in profile abundances or the selection of fitting species. As with the Main Report and
Contributions by Species, a print-out of the MPIN matrix may be obtained for the current sample
result via the print button on the toolbar, and a file may be created via the Print To File option.
g|EPA-CMB8.2 SOS
File
Help
*• I
H m
Input Files | Options
a
Samples
Species
SAMPLE Site: BAKERS Date: 02/03/89
OPTIONS
Iteration Delta: 20
Maximum Source Uncertainty (<%): 20
Minimum Source Projection: 0.95
Decimal Places Displayed: 5
Units: ug/ms
Main Report
SP
!CIE5
N3IC
s
N
I:"
N
lie
»TC
'AC
.ai-
ECTC
0
A
S
c
1
c
T
V
c
M
F
N
B
UC
.xc
E: :c
.: :c
.vr
c
EXC
SXC
:. c
JXC
:: :c
E. C
.,:
PBXC
SOIL12
-0.01
-0.01
0.02
-0.03
0.02
-0.13
-0.08
0.40
0.54
-0.04
0.04
1.00
0.04
-0.02
0.02
0.11
0.78
-0.05
-0.08
-0.03
BAMAJC
0.00
0.00
0.00
0.56
0.01
-0.02
0.21
-0 . 04
-0.08
0.84
1 . 00
-0.05
0.00
0 . 00
0.00
-0.07
-0.07
0.00
-0.10
-0.09
s
Control File: INdemol.inS Samples: 1 Species: 20 Sources: 6 Output Format: Comma-separated Value (, csv)
Sources Results
Duration: 24 Start Hour: 0 Size: FINE
F Britt-Luecke F Best Fit
T Source Elimination Chi Square i
R Square i
Species Fit Array: 4
Sources Fit Array: 5
Contributions by Species ! MPIN Matrix i|
FCROC MOV
A
SOURCE NAME
:S2 AMSUL AMNIT
0.07 0.07 -0.29 1.00
0.06 0
-0 . 10 -0
0.00 -0
0.03 -0
06 1.00 -0.28
10 0.45 0.47
11 0.00 0.00
01 -O.01 0.00
-0.01 1.00 -0.08 -0.02
-0.01 0
-0.02 -0
-0.02 0
0.00 -0
0.00 -0
-0.02 -0
0.00 -0
0.67 -0
0.01 0
0.02 0
0.02 -0
1.00 -0
-0.01 0
-0.01 0
68 -0.06 -0.01
03 0.00 0.00
13 -0.01 0.00
14 -0.01 0.00
21 -0.01 0.00
04 0.00 0.00
01 0.00 0.00
01 -0.15 0.04
00 0.00 0.00
29 -0.03 0.00
11 0.00 0.00
01 -0.22 0.06
69 -0.05 -0.01
55 -0.04 -0.01 v
Result 1 of 1
Delete Current
Delete All
Run
Figure 3.12 Calculation Results - MPIN Matrix
3- 15
-------
-------
4. INPUT AND OUTPUT FILES
This section describes the structure of EPA-CMB8.2 input and output files and methods of
generating these files. Each type of input file structure is illustrated with one of the test data sets
packaged with EPA-CMB8.2.
4.1 File Naming Conventions
EPA-CMB8.2 input and output files can have any file name with a three-character extension
that indicates the file type. A suggested naming convention is PP*.ext, where:
• PP: Type of file. Common definitions are:
IN Control file identifying specific input data files.
AD Ambient Data. Selection file initiates sample selection from the ambient data file
for apportionment during an CMB session; data file contains the measured
ambient concentrations and their uncertainty values.
PR Source PRofile. Selection file identifies initial fitting profiles and source profile
descriptions; data file contains mass-fraction chemical abundances and their
uncertainties.
SP SPecies selection file identifies initial fitting species for the CMB session.
• * Study identifier. This code allows separate studies to be distinguished from one
another. EPA-CMB8.2 allows Windows® flexibility for this name string (i.e., it is
not character-limited).
• ext Extension that also identifies file type or format. The following file extensions
are recognized by EPA-CMB8.2:
in8 Input control (ASCII text) file. EPA-CMB8.2 lists files with this extension in the
Control File browse window at startup.
sel Fitting profiles, fitting species, and sample selection (ASCII text) files. EPA-
CMB8.2 recognizes files with this extension as containing initial selections that
can be entered external to the program. This extension applies only to the PR. SP.
and AD file types.
csv Ambient data or source profile comma-separated value ASCII text file. Each
field is separated by a comma. Comma-delimited ASCII data base output files are
written with this extension.
dbf Data base file generated by dBase or FoxPro compatible data management
software. Most commonly used spreadsheets offer this as an output option.
dBase or FoxPro output files are written with this extension.
4- 1
-------
txt Ambient data or source profile data blank-delimited ASCII text file. Formatted,
blank-delimited ASCII data base output files may be created with these
extensions.
wks Lotus 1-2-3 version 1 spreadsheet format. Most commonly used spreadsheets
offer this as an output option. This is the most useful output format for the data
base output file when source contribution estimates will be analyzed using a
spreadsheet.
Note that if neither input file (ambient data or source profile data) is supplied in ASCII
format (*.txt), EPA-CMB8.2 converts any of the csv, dbf, and wks input data files to the blank-
delimited (txt) files which are actually used by the program. These txt files are created "on the
fly" as soon as the user moves off of the Input Files Screen (Figure 3.3), and will appear in the
same subdirectory that stores any of the csv, dbf, and/or wks files. Such txt files created from
dbf files are nicely formatted and easy to read. If any txt files are supplied in the subdirectory
but not directed for use as input files in the Control File, they will be overwritten (replaced) by
new versions created by EPA-CMB8.2. If, however, any txt files are supplied and directed for
use as input data by the Control File, they will be retained (not modified).
4.2 Input File Relationship
Six data files are normally used for input to EPA-CMB8.2, the first of which is a control file
that directs EPA-CMB8.2 to five specific files. Three of the files are optional selection files,
which provide substantial user convenience by establishing commonly used arrays and sample
subsets that would otherwise need to be initialized each time the model is run. The remaining
two - the ambient and source profile data files - are required by EPA-CMB8.2. Figure 4.1
presents the relationship of the files whose descriptions appear in the following subsections.
4.2.1 Control File: IN*.in8
This fixed format file contains a list of the names of EPA-CMB8.2 input data files, all
of which must reside in the same directory that stores the Control File itself. This filename (e.g.,
INsjvf.inS, exemplified in Figure 4.1) consists of five lines as shown below. These lines, in
succession, contain the names of the files which are described in the following subsections. If a
selection file is absent, the corresponding line in the Control File should be labeled with one or
more characters, e.g., a series of asterisks ('******'). or any name that doesn't reside in the
Control File directory. Here's an example:
PRsjvf.sel
SPsjvf.sel
*******
ADsjvf.csv
PRsjvf.csv
File name entries should be left justified and in the sequence shown. In EPA-
CMB8.2, the only restriction on file names is that they are acceptable to the operating system.
This means that extended file names may be used. The utility of the Control File is to save the
effort of keying in the input filenames individually. If a Control File is not used at startup, EPA-
CMB8.2 will accept the names of individual data input files on the fly, provided they are
compatible with each other.
4-2
-------
Figure 4.1. EPA-CMB8.2 Input Files
Control File
PRsjvf
SPsjvf
ADsjvf
ADsjvf
PRsjvf
sel
sel
sel
txt
txt
(INsjvf.in8)
Fitting sources profile selection file (PRsjvf.sel)
SJVOOl SOIL01
SJV002 SOIL03
SJV003 SOIL04
SJV004 SOIL05
SJV005 SOIL06
STOCKTON AGRICULTURAL SOIL (PEAT)
FRESNO PAVED ROAD
VISALIA AGRICULTURAL SOIL (COTTON/WALNUT)
VISALIA AGRICULTURAL SOIL (RAISIN)
VISALIA SAND AND GRAVEL
Fitting species selection file (SPsjvf.sel)
TMAC TOT
N3IC N03
S4IC S04
N4TC NH4
KPAC K-S
Mass by gravimetry (ug/m3)
Nitrate by 1C (ug/m3)
Sulfate by 1C (ug/m3)
Ammonium by AC (ug/m3)
Soluble Potassium by AA (ug/m3)
Ambient data selection file (ADsjvf.sel)
BAKERS
CROWS
FELLOW
FRESNO
KERN
STOCKT
07/26/88
07/26/88
07/26/88
07/26/88
07/26/88
07/26/88
24
24
24
24
24
24
0
0
0
0
0
0
FINE *
FINE *
FINE *
FINE *
FINE *
FINE *
Ambient data input file (ADsjvf.txt)
ID DATE DUR STHOUR
BAKERS 06/20/88 24
BAKERS 07/02/88 24
BAKERS 07/26/88 24
BAKERS 08/07/88 24
BAKERS 08/19/88 24
SIZE TMAC TMAU N3IC N3IU S4IC S4IU N4TC N4TU KPAC
0 FINE 17.2788 0.9920 0.2816 0.1715 2.8204 0.1612
0 FINE 23.5425 1.2744 0.8306 0.1761 3.2224 0.1791
0 FINE 26.7742 1.4250 0.2054 0.1715 3.4881 0.1911
0 FINE 21.9185 1.2008 0.4096 0.1732 3.1228 0.1748
0 FINE 22.6664 1.2339 0.5093 0.1729 3.7134 0.2014
Source profile input file (PRsjvf.txt)
PNO SID SIZE N3IC N3IU S4IC
SJVOOl SOIL01 FINE 0.002700
SJV002 SOIL03 FINE 0.000000
SJV003 SOIL04 FINE 0.000000
SJV004 SOIL05 FINE 0.000300
SJV005 SOIL06 FINE 0.000000
S4IU N4TC N4TU KPAC
0.004700 0.000400 0.
0.003600 0.005000 0.
0.003000 0.000000 0.
0.002600 0.000000 0.
0.065800 0.000000 0.
KPAU NAAC NAAU ECTC ECTU
001300 0.000600 0.000500
006600 0.000300 0.000500
001100 0.000000 0.000100
000900 0.000000 0.000000
023800 0.000000 0.001100
-------
4.2.2 Ambient Data Input File (AD*.csv, AD*.dbf, AD*.txt, AD*.wks)
Ambient data files may be formatted as comma-separated values in ASCII text (*.csv),
xBASE (*.dbf), blank-delimited ASCII text (*.txt), or Lotus Worksheet (*.wks). The csv and
dbf formats are preferred, as they are easier to prepare in spreadsheet (e.g., Microsoft Excel®,
Corel QuatroPro®, Lotus 123) and data base (e.g., Microsoft Access®, dBASE) software than the
other formats. The wks format creates large files and requires substantial translation time for
EPA-CMB8.2 input and output, so it is the least desirable of these alternatives. The TXT format
is most consistent with CMB7, so older CMB7 data files can be used for EPA-CMB8.2 input
without modification. NB: if using TXT format, make sure the file's not tab-delimited! Recall
from Section 4.1 that under some circumstances, EPA-CMB8.2 will create an ambient data input
file in txt (ASCII) format "on the fly", and that it is actually this format that the model uses for
calculations. The appropriate file extension must be associated with each format, as EPA-
CMB8.2 recognizes the file type by this extension.
Examples of all supported file types are provided with theEPA-CMB8.2test.ZIP test
data. Following is an example of the ADsjvf.csv file:
ID,DATE,DUR,STHOUR,SIZE,TMAC,TMAU,N3IC,N3IU PBXC,PBXU
BAKERS,06/20/88,24,0,FINE,17.2788,0.9920,0.2816,0.1715 0.0236,0.0052
Of the first 5 field names in the header, note that as currently configured, EPA-CMB8.2 limits
the first 2 field names to 4 characters; the last 3 fields must be named identically as indicated
above. The "total" pair (e.g., TMAC & TMAU) preceding the species list differentiates the
AD*.* header from that for PR*.*; there is no practical limitation for the pair names. All
subsequent species names in the header are restricted to 6 characters.
The delimited forms of this file do not require fixed format spacing, only that a
comma (or a blank character for TXT files) separate each field from prior and subsequent fields.
The 1st line contains the field identifiers, followed by (beginning in field 6) the species codes,
which occur in pairs (concentration, followed by uncertainty). Note that the name (code) for the
concentration component of each pair must correspond identically (including case) with its
counterpart in both the species selection files (SP*.sel) described in Section 4.2.4. and the source
profile data input file (PR*.*) described in Section 4.2.3. Note that it is on this line where EPA-
CMB8.2 gets the labels that appear in the species selection window (not from the species
selection file, SP*.sel). Regardless of how the AD*.sel file is organized in terms of the sequence
of ambient samples listed, in the sample selection screen they will appear according to the
sequence dictated by the ambient data file (e.g., AD*.txt). Note also this line is the source of all
species labels appearing in the Main Report. Beginning with the 2nd line are the actual data, one
line per ambient sample, starting with the Site ID in Field 1. Note also that it is this ambient data
input file that controls the sequence of ambient samples displayed in the samples selection
window (not the samples selection file, AD*.sel). The records for each sample are formatted as
follows:
Field 1: Site ID (up to 12 characters)
Field 2: Sampling date (up to 8 characters)
Field 3: Sample duration (up to 2 characters)
Field 4: Sample start hour (up to 2 characters)
Field 5: Particle size fraction (up to 6 characters)
Field 6: Total Mass concentration (any number of characters in integer, floating point, or
exponential format)
Field 7: Uncertainty of total mass concentration (same format as Field 6)
Field 8+2n: Concentrations of chemical species (same format as Field 6), where n = 0, 1,2, ...
Field 9+2n: Uncertainty of species concentrations (same format as Field 6), where n = 0, 1,2,
4-4
-------
EPA-CMB8.2 always assumes that Field 6 is the total mass concentration, and it does
not use this as a fitting species. Uncertainty values in fields 7 and 9+2n are in the same units as
the measured concentration values. For EPA-CMB8.2, the total number of ambient data records
can reach into the thousands, limited only by computer memory. This makes it especially useful
for examining multi-species hourly data obtained from automated gas chromatographs and time-
of-flight mass spectrometers. For particles, up to four different size fraction identifiers may be
used for the same sampling site and period, and the user can select mnemonics that suit
individual purposes.6 The size fraction names FINE and COARSE are reserved for the PM2.5 and
coarse particle (PMio to PM2.s) size fractions that are commonly measured in PMio source
assessment studies. As mentioned in Section 3.6.1, when these size fraction identifiers (i.e.,
'FINE' & 'COARSE') are used, an additional report is produced that sums the fine and coarse
source contribution estimate(s) (and uncertainties) to provide the estimate(s) for 'Total'
(frequently = PMio). Any other designator can be placed in the size column for non-segregated
samples, such as "PM25" or "VOC". Where semi-volatile materials are being apportioned, the
particle (PART) and gas (GAS) phases are good designations.
Positive uncertainty values (> 0) must be assigned to all (non-missing) chemical
concentrations used as fitting species. EPA-CMB8.2 will return an error message when it
detects (in the input data file for ambient samples) a value for uncertainty that is less than or
equal to zero for any species concentration value not flagged as missing. A species for which the
concentration value is missing (i.e., invalid) cannot be used as a fitting species for that sample.
Missing values for either concentrations or associated uncertainty must be designated by placing
-99. in the appropriate field in the data file. If a species is to be flagged as "missing", the value
-99. must appear in the concentration field; substituting -99. for its uncertainty is optional
because, as mentioned in Section 3.6.1, the species will automatically be removed from the
calculation (though they will appear in the Main Report). For any species, when the fields for
missing values species are substituted in this way, -99.0000 will appear in applicable fields
under "MEASURED" in the Main Report so long as decimal places displayed (Section 3.2) is set
to a value < 5. When the species concentration is flagged with -99., the value for
CALCULATED/MEASURED in the Main Report will always be 0.00. The value for this
diagnostic as well as for RESIDUAL/UNCERTAINTY will in this case be meaningless. See
Appendix F for a discussion of EPA-CMB8.2's interpretation and treatment of input data
conditions.
If more that 4 size ranges are supplied for a given sampling site and period, the following error message from the DLL
will appear (Appendix E):
Number of fitting sources =0 > Number of fitting species = 0, followed by:
The number of fitting sources must be positive and <= number of fitting species.
4-5
-------
4.2.3 Source Profile Input File (PR*.csv, PR*.dbf, PR*.txt, PR*.wks)
Source profile data files may be formatted as comma-separated values in ASCII text
(csv), xBase (dbf), blank-delimited ASCII text (txt), or Lotus Worksheets (wks). The csv and
dbf formats are the most portable and easily prepared. NB: if using TXT format, make sure the
file's not tab-delimited! Recall from Section 4.1 that under some circumstances, EPA-CMB8.2
will create a source profile input file in txt (ASCII) format "on the fly", and that it is actually this
format that the model uses for calculations. The appropriate file extension must be associated
with each format, as EPA-CMB8.2 recognizes the file type by this extension. Examples of all
supported file types are provided with EPA-CMB8.2's test case (EPA-CMB82test.zip).
Following is an example of the PRsjvf.csv file:
PNO,SID,SIZE,N3IC,N3IU PBXC,PBXU
SJV001,SOIL01,FINE,0.002700,0.004700 0.000000,0.000000
As mentioned for the AD*.* file *(Section 4.2.2), all species names in the header are restricted
to 6 characters.
The delimited forms of this file do not require fixed format spacing, only that a
comma (or a blank character for TXT files) separate each field from prior and subsequent fields.
The 1st line contains the field identifiers followed by the species codes (fields > 3), which can be
up to 6 alphameric characters in length. As with the ambient data input file, these codes appear
in abundance / uncertainty pairs, and that the name (code) for the concentration component of
each pair must correspond identically (including case) with its counterpart in both the species
selection files (SP*.sel) described in Section 4.2.4. and the ambient data input file (PR*.*)
described in Section 4.2.3. Beginning with the 2nd line are the actual data, one line per source
profile. The first two fields are the source code and mnemonic, respectively. The source code
must correspond identically with that used in the source profile selection file (PR*.sel) described
in Section 4.2.4 (and note that these two fields are what appear as the left two columns in the
sources selection window). The limitations on each field are:
Field 1: Profile number or source code (up to 6 characters)
Field 2: Source mnemonic (up to 8 characters)
Field 3: Particle size fraction (up to 6 characters)
Field 4+2n: Fraction of species in primary mass of source emissions (floating point or
exponential format), where n = 0, 1,2, ...
Field 5+2n: Uncertainty of fraction of species in primary mass of source emissions (same
format as Field 4), where n = 0, 1,2, ....
If the name (code) used in field 1 does not match identically (including case) its counterpart in
the source profile selection file, the source will not appear on the Sources selection screen when
EPA-CMB8.2 is run, and thus not be available for use in a calculation.
Source profile abundances are expressed in fractions of total mass, not in percent. In
the example input file snippet above for source profile SJV001, the NO3" abundance is 0.27% of
the total mass, and its uncertainty is 0.47%. Unlike the ambient data file, the source profile
input file does not contain a mass concentration field because all species abundances have been
divided by this mass. The abundances and their uncertainties are typically unitless, but in certain
4-6
-------
applications can be represented as concentrations. The user will have to account for this latter
case in interpreting the SCEs that are computed. The total number of records included depends
on the number of species, number of sources, and size of the computer memory.
From one to four different size fraction identifiers may be used, but these must be the
same as those used in the ambient data and sample selection files. Missing values for chemical
species in source profile files can be replaced by a best estimate with a large uncertainty if they
are to be used as fitting species. Missing values must be flagged with -99. if the species is not
intended for use as fitting species. Species with mass fractions so flagged will automatically be
removed from the calculation (though they will appear in the Main Report). While uncertainty
values for species in source profiles are allowed to be £ 0.0, some effort should be made to
supply values > 0.0. Default values of 0.0 for the fraction and 0.0001 to 0.01 for the uncertainty
are often chosen for species that are expected to be present in small abundances. This indicates
that the species is present in source emissions at a concentration less than 0.01% to 1%. A
smaller value may be appropriate for certain source types and species. See Appendix F for a
discussion of EPA-CMB8.2's interpretation and treatment of input data conditions.
In certain cases in which an uncertainty value >0 is applied to a species whose abundance =
0, the uncertainty represents the lower quantifiable limit of the measurement. A zero value for
abundance means that the true value is something between zero and the detection limit. Zeros
are important in some source profiles when they occur in other source profiles. This makes that
species a marker for the source in which it occurs. The uncertainty for the zero adds to the
effective variance weighting. If this uncertainty is high and the source contribution estimate is
high, the influence of that species is reduced by the weighting.
4-7
-------
4.2.4 Source (PR*.sel), Species (SP*.sel), and Sample Selection (AD*.sel) Input Files
The optional source, species and sample selection files provide initial selection arrays
that do not have to be entered from the program each time a EPA-CMB8.2 session is begun.
These files limit the profiles, species, and ambient data records to those listed in the selection file
arrays, even though a larger number may be included in the ambient and source profile data files.
This means that the data files need not be edited when only subsets of variables are desired for a
specific EPA-CMB8.2 modeling session. Variable definitions can also be documented in these
files (comment fields). The selection files also dictate the sequence in which species and sources
appear in the output. The total species list is that common to the species selection file, the
ambient data file, and the source profiles file. The total sources list is that common to the
sources selection file and the source profiles file. The total ambient samples list is that common
to the ambient sample selection file and the ambient data file. As mentioned in Sections 3.3, 3.4
& 3.5, if any of the selection (*.sel) files are modified during a session, EPA-CMB-8.2 will
detect the change and, upon program exit, allow the opportunity to update (overwrite) these files.
The file may also be conveniently renamed.
Following is an example of the source profile selection file PRsjvf.sel. In this
selection file, as well as the two that follow, the first two lines are shown only for field location
(they are not part of the file itself):
01234
1234567890123456789012345678901234567890
SJV001
SJV002
SJV003
SJV004
SJV005
SJV006
SJV007
SJV008
SJV009
SJV010
SJV011
SJV012
SJV013
SJV014
SJV015
SJV016
SJV017
SJV018
SJV019
SJV020
SJV021
SJV022
SJV023
SJV024
SJV025
SJV026
SJV027
SOIL01
SOIL03
SOIL04
SOIL05
SOIL06
SOIL07
SOIL08
SOIL09
SOIL10
SOIL11
SOIL12
SOIL13
SOIL14
SOIL15
SOIL16
SOIL17
BAMAJC
MAMAJC
MAFISC
MADIEC
BAAGBC
ELAGBC
FRCONC
STAGBC
VIAGBC
VIDAIC
SFCRUC
STOCKTON AGRICULTURAL SOIL (PEAT)
FRESNO PAVED ROAD
VISALIA AG SOIL (COTTON/WALNUT)
AGRICULTURAL SOIL (RAISIN)
SAND AND GRAVEL
URBAN UNPAVED
PAVED ROAD
AGRICULTURAL SOIL, ALKALINE
AG SOIL, SANDY LOAM
UNPAVED ROAD (OILDALE)
PAVED ROAD
WINDBLOWN URBAN UNPAVED
AG SOIL, WASCO SANDY LOAM
AG SOIL, CAJON SANDY LOAM
UNPAVED ROAD (RESIDENTIAL)
ROAD
VISALIA
VISALIA
VISALIA
VISALIA
BAKERSFIELD
BAKERSFIELD
BAKERSFIELD
BAKERSFIELD
BAKERSFIELD
BAKERSFIELD
BAKERSFIELD
BAKERSFIELD
TAFT UNPAVED
*****
*****
BAKERSFIED CORDWOOD, MAJESTIC FIREPLACE
MAMMOTH LAKES WOOD, MAJESTIC FIREPLACE
BAKERSFIELD WOOD, FISHER MAMA BEAR STOVE
MAMMOTH LAKES DIESEL TOUR BUSES (IDLING)
BAKERSFIELD AG BURN (WHEAT AND BARLEY)
EL CENTRO AGRI. BURN (WHEAT)
FRESNO HIGHWAY 40 CONSTRUCTION
STOCKTON AGRI. BURN (WHEAT)
VISALIA AGRI BURN (WHEAT)
VISALIA DAIRY/FEEDLOT DUST
SANTA FE CRUDE BOILER
4-8
-------
*****
*****
*****
* * * *
30% UNLEADED
10% UNLEADED
5% UNLEADED
CHEVRON RACETRACK CRUDE BOILER
MODESTO TIRE POWER PLANT
STANISLAUS RESOURCE RECOVERY FACILITY
NBS CEMENT DUST
ROCK CRUSHING 1987 SCAB
SANDBLASTING AND PLASTERING
MARINE
MOVES -SS(NEA-E,WOB,T42,TVMT)
MOVES -SS(NEA-E,WOB, WOT, TVMT)
MOVES - SCAB (ARB-E,WOB, WOT, CM)
MOVES - SCAB (NEA-E,WOB, WOT, CM)
MOVES - SCAB (NEA-E,WB1,T42, CM)
GYPSUM DUST, (TOTAL FROM CAS04)
AMMONIUM SULFATE
AMMONIUM BISULFATE
SULFURIC ACID
AMMONIUM NITRATE
NITRIC ACID
SODIUM NITRATE
50% DIESEL, 20% LEADED,
75% DIESEL, 15% LEADED,
85% DIESEL, 10% LEADED,
PURE ORGANIC CARBON
LIMESTONE
CROWS LANDING AGRI.
CROWS LANDING PAVED ROAD
KERN UNPAVED ROAD
KERN AGRI.
As with the source profile input file, the first fields are the source profile code and
mnemonic, respectively. A source code with up to six characters is located in Columns 1 to 6.
Columns 9 to 16 are available for an eight-character profile name (mnemonic). The names used
in both fields must correspond identically with those used in the source profile input file (Section
4.2.3). Note that these two fields are what appear as the left two columns in the sources
selection window. Asterisks in Column 19, 21, 23, 25, 27, 29, 31, 33, 35 and 37 designate initial
arrays (independent sets) of fitting sources when EPA-CMB8.2 is executed. The maximum
number of sources is essentially unlimited. Comments can be added to this file beginning in
column 39 to document the source profiles.
For each source listed in the selection file, some information must appear in either (1)
the mnemonic field, (2) the selection array field (at least one column must be tagged with '*'), or
(3) the comment field in order for the source to appear on the selection screen in EPA-CMB8.2.
If the name used in field 1 does not match identically (including case} its counterpart in the
source profile input file, or if any of the aforementioned conditions is not met the source will not
appear on the Sources selection screen when EPA-CMB8.2 is run, and thus not be available for
use in a calculation.
SJV028
SJV029
SJV030
SJV031
SJV032
SJV033
SJV034
SJV035
SJV036
SJV038
SJV039
SJV040
SJV041
SJV051
SJV052
SJV053
SJV054
SJV055
SJV056
SJV057
SJV058
SJV059
SJV060
SJV061
SJV062
SJV063
SJV064
SJV065
CHCRUC
MOTIBC
SCRRFC
CDCEMT
CDRKCR
CDSAPL
MARINE
MOVES1
MOVES2
MOVES3
MOVES4
MOVES5
MPGYPU
AMSUL
AMBSUL
H2S04
AMNIT
HN03
NAN03
MVDEN1
MVDEN2
MVDEN3
OC
LIME
SOIL28
SOIL29
SOIL30
SOIL31
4-9
-------
Following is an example of the species selection file SPsjvf.sel:
1234
1234567890123456789012345678901234567890
TMAC TOT Mass by gravimetry (ug/m3)
N3IC N03 * * * Nitrate by 1C (ug/m3)
S4IC S04 * * * Sulfate by 1C (ug/m3)
N4TC NH4 * * * Ammonium by AC (ug/m3)
KPAC K-S * * * Soluble Potassium by AA (ug/m3)
NAAC NA * * * Soluble Sodium by AA (ug/m3)
ECTC EC * * * Elemental Carbon by TOR (ug/m3)
OCTC OC * * * Organic Carbon by TOR (ug/m3)
ALXC AL * * * * Aluminum by XRF (ug/m3)
SIXC SI * * * * Silicon by XRF (ug/m3)
SUXC S Sulfur by XRF (ug/m3)
CLXC CL * * * * Chloride by XRF (ug/m3)
KPXC K * * * * Potassium by XRF (ug/m3)
CAXC CA * * * * Calcium by XRF (ug/m3)
TIXC TI * * * * Titanium by XRF (ug/m3)
VAXC V * * * * Vanadium by XRF (ug/m3)
CRXC CR * * * * Chromium by XRF (ug/m3)
MNXC MN * * * * Manganese by XRF (ug/m3)
FEXC FE * * * * Iron by XRF (ug/m3)
NIXC NI * * * * Nickel by XRF (ug/m3)
CUXC CU * Copper by XRF (ug/m3)
ZNXC ZN * Zinc by XRF (ug/m3)
BRXC BR * * * Bromine by XRF (ug/m3)
PBXC PB * * * * Lead by XRF (ug/m3)
A species code with up to six characters is located in Columns 1 to 6, and must
correspond identically with that used in both the ambient data input file (Section 4.2.2) and
source profile input data file (Section 4.2.3). Columns 9 to 16 are available for an eight-
character species name, which is optional (this is the only place they can be installed). Note that
these two fields are what appear as the left two columns in the species selection window.
Asterisks in Column 19, 21, 23, 25, 27, 29, 31, 33, 35 and 37 designate initial arrays
(independent sets) of fitting species when EPA-CMB8.2 is executed. The maximum number of
species is essentially unlimited. Comments can be added to this file beginning in column 39 to
document the meaning and units of the chemical components; this is the only place they can be
installed.
As mentioned above for each source listed in the source selection file, for each species
listed in the species selection file, some information must appear in either (1) the name field, (2)
the selection array field (at least one column must be tagged with '*'), or (3) the comment field
in order for the species to appear on the selection screen in EPA-CMB8.2. This constraint
notwithstanding, the name field is truly optional. Note that if the species code in field 1 doesn't
match exactly (including case} its counterpart in both the ambient data input file and the source
profile input file, or is missing, the species will not appear in the selection screen when the
model is run regardless of any other condition. If the species does not appear on the selection
screen when EPA-CMB8.2 is run, it will not be available for use in a calculation.
4- 10
-------
For the ambient data records (sample) selection file, columns 1 through 12 are for the
Site ID, columns 14 through 21 are for the date, columns 23 and 24 for the sample duration,
columns 26 and 27 for the sample start hour, and columns 29-33 for the particle size fraction, if
appropriate. Intermediate columns should be blank. An asterisk in column 36 initializes
(selects) a record for processing.
Following is an example of the ambient data records selection file ADsjvf.sel:
1234
12345678980123456789012345678901234567890
BAKERS 07/26/88 24 0 FINE *
CROWS 07/26/88 24 0 FINE *
FELLOW 07/26/88 24 0 FINE *
FRESNO 07/26/88 24 0 FINE *
KERN 07/26/88 24 0 FINE *
STOCK! 07/26/88 24 0 FINE *
Note that the tagging asterisks are placed in column 36 (not 35, as in CMB8.0). This change is
to maintain spacing for the new 6-character field for size fraction (i.e., COARSE). Note also that
regardless of the sequence of ambient samples listed, they will appear in the sample selection
window according to the sequence dictated by the ambient data input file (e.g., AD*.txt); see
Section 4.2.2.
4.3 Output Files
Both general report (a listing) and data base output files are produced by EPA-CMB8.2.
4.3.1 Report Output File
The report output file is generated from the Main Report screen (Section 3.6.1) and
presents the source contribution estimates, standard errors, model performance measures, and
measured and calculated chemical concentrations for each sample (Section 3.6). The report
written to the output file is identical to that which appears in the Output window during an
interactive modeling session. It is in ASCII text format and can be imported into word
processing programs to document the source contributions calculated for each sample. All
information needed to independently repeat the source apportionment is contained in this report,
including an echo of the *.in8 input file that was used in its generation. Examples of the report
are shown in Section 5 & 6.
4.3.2 Data Base Output File
The data base (spreadsheet-type) output file records the contribution of each source-
type to a particular species in a single data record, one record per species. Sample identifiers and
model performance measures are also included in each record. As EPA-CMB8.2 is currently
configured, this file may be generated in blank-delimited (*.txt) or comma-separated value
(*.csv) formats.7 The file structure is:
7Binary-type formats (e.g., DBF & WKS) are disabled in EPA-CMB8.2.
4- 11
-------
Field 1: Species Code
Field 2: Species Name
Field 3: Fitting flag; a '*' indicates a fitting species, while a '_' indicates a floating species
Field 4: Sampling site identifier
Field 5: Sampling date
Field 6: Sample start hour
Field 7: Sample duration
Field 8: Particle size fraction
Field 9: Measured species concentration
Field 10: Uncertainty of measured species concentration
Field 11: Calculated species concentration
Field 12 Uncertainty of calculated species concentration
Field 13: R-square value
Field 14: Chi-square value
Field 15: Percent of measured mass
Field 16+2n: Source contribution estimate, n = 0, 1,2, ....
Field 17+2n: Standard error of source contribution estimate, n = 0, 1,2, ....
Fields 1, 2, and 4 through 8 record the sample information. Fields 3 and 11 through
15 provide information about the EPA-CMB8.2 calculation.8 The remaining fields correspond to
each source profile in the PR*.* data file and contain the source contribution estimates and
standard errors for these sources. A value of -99. is recorded when a profile was not used in the
calculation.
The first record in this output file contains the field identifiers. All subsequent records
contain data. Fields 16+2n and 17+2n are labeled with source codes and source names,
respectively.
4.4 Creating Data Input Files
When using a Control File, note that if several data input files (e.g., AD*.* & PR*.*)
exist for which the only difference is format, EPA-CMB8.2 assumes that the files will be named
identically except for the extension. For blank-delimited and comma-separated value input files,
there are three common methods of creating EPA-CMB8.2 input files: 1) manually entering the
data in the correct format using a text editor or word processing program; 2) editing existing
input files with a text editor or word processing program; or 3) transferring files from
computerized data bases.
A text editor or word processor in text mode can be used to type entire input files. It is
best to bring the example files into the editor, then insert the new values in the same locations as
the existing values by using the editor in TYPEOVER mode. Spaces between fields should be
entered with the space bar; tabs should not be set. Each line should be terminated with the
Fields 11 & 12 (calculated species concentration/uncertainty) are a new feature in EPA-CMB8.2 that make the output
data file more robust.
4- 12
-------
ENTER key rather than using the wraparound feature present in many editors. No blank lines at
the end of the file should be present. Completed files should be saved with an appropriate
filename. A wide variety of very good ASCII text editors are available for creating and
manipulating input files.
When data files have been prepared for other applications (e.g., source profiles may be
common to several different data sets), these files may be cut and pasted to produce the needed
input data files. Owing to differences in individual editing programs, the user is should consult
the manual for the editing program to be used for directions on opening a copy of the existing
file, deleting and adding material, saving the changes, and renaming the file. When using word
processors (e.g., MSWord or WordPerfect), the files must be saved as DOS text with line breaks.
Otherwise, extra information is included in the files that EPA-CMB8.2 cannot read.
Input files are most easily produced with spreadsheet or data base software. Many
source profile and ambient data sets are available in data base management formats. Selections
of data, field names, and data structure can be easily made by the data base software. These can
be saved using the Save As or Export selections from the File menu. The CSV, DBF, and WKS
formats can be selected from the "Save as type" option box that usually appears in the "Save As"
window.
4.5 Reading Output Files
Main Report (ASCII) text files can be read directly into a word processing application
where the detailed output for each sample can usually be displayed on a single page with
columns aligned using a non-proportional font, e.g., a New Courier or Letter Gothic 8-point to
10-point. Such a fixed pitch font in which every character occupies the same space is necessary
for columns to be correctly aligned. As mentioned in Section 4.3.2, data base (spreadsheet-type)
output files from EPA-CMB8.2 are formatted either as ASCII (default) or comma-separated
value (CSV), and this selection must be made on the Options screen before a calculation is made.
The output file will be stored in the directory in which the Control File resides (which can, of
course, be changed in the browse dialog). These output files can be opened directly by data base
or spreadsheet applications, e.g., Microsoft's Excel®, and from there exported into any of several
different formats, e.g., xls. The contents of the EPA-CMB8.2 output screen can also be selected
and copied to the clipboard for pasting into other Windows® applications. Graphs made with
EPA-CMB8.2 (Section 3.3 & 3.5) can be printed directly from the screen or copied to the
Windows® clipboard via copy command, then pasted into a text box or frame in a word
processing application.
4- 13
-------
-------
5. USING EPA-CMB8.2: TEST CASES
This section illustrates EPA-CMB8.2 commands and operations using the San Joaquin
Valley, CA, PM2.5 data set. Other test data sets (Section 2.2) are available on EPA's modeling
website (www.epa.gov/scram001) as examples of additional data base file formats and for
independent practice in EPA-CMB8.2 application and validation. These examples are most
effective when accompanied by actual application of EPA-CMB8.2 on the user's computer.
5.1 Starting EPA-CMB8.2
Start EPA-CMB8.2 by double-clicking on the EPA-CMB8.2 icon on your desktop (or
by selecting it from the Start menu). The first screen that will appear is depicted in Figure 3.1.
As discussed in Section 3.1, at this point EPA-CMB8.2 requires the user to choose between
using a stored Control File or to choose input files (which are otherwise directed by the Control
File) individually.
5.1.1 Control File Operation
Select the Control File mode and use the browse feature to access the test case
folder. A screen like the one shown in Figure 3.2 will then appear. Select INsjvf.inS and open
it. A screen as shown in Figure 5.1 will appear. Select the options screen and use the defaults as
HEPA-CMBS.Z
File Help
Control File: INsjvf.inS Samples:
Species:
Output Format: ASCII (.txt) -> Default
Input Files|| Options | Samples | Species | Sources | Results |
SELECT INPUT FILES
CONTROL FILE |lNsjvf.in8
& Browse..,
REQUIRED INPUT FILES
Ambient Data |ADsjvf.txt
Source Profi
OPTIONAL SELECTION FILES
Ambient Data
Species
Source Profiles
|PRsjvF.txt
& Brows*
& Browse... |
|ADsjvf.st
|SPs]vf.sel
|pRsjvf,sel
& Brows*
& Brows«
& Browse.,
Figure 5.1 Input Files for Test Case - Control File Mode
shown in Figure 3.4. Choose the Select Samples screen and click View Selected to present only
the tagged samples. Note for each sample are the following attributes: selection status, site,
5- 1
-------
sample date, sampling duration (hours), start hour, and size fraction (Figure 5.2). This is
followed by a series of fields for measured concentration and uncertainty for total mass and for
constituent species in the sample. Measured values for species may be viewed by moving to the
right through the field using the horizontal scroll, and these values may also be hidden (Hide
Data).
HEPA-CMBB.Z
File Help
j I Ja I Control File: INsjvf,inS
Input Files | Options Samples I Species j Sources | Results [
Samples: 6
Species: 19
Output Format: ASCII (,txt)-> Default
AMBIENT DATA SAMPLES
BAKERS 07/26/88 24 0 FINE 26.7742 1.425 0.2054 0.1715 3.4881 0.1911 1.358
CROWS 07/26/88 24 0 FINE 21.0764 1.1717 0.9071 0.2027 3.2867 0.1875 1.3026
3H Uo7/26/88 24 0 FINE 21.5363 1.19 0.2128 0.1852 5.1475 0.2709 1.9154
FRESNO 07/26/88 24 0 FINE 26.4104 1.4242 0.4132 0.1938 3.7157 0.2056 1.642S
KERN 07/26/88 24 0 FINE 17.9463] 1.0251 | 0.1292 0.1985 3.6236 0.2026 1.5038
STOCK! ,07/26/88 24 ;0 IpINE 18.7432] 1,0445 1 0.2728 0.1793 2,8282 0.1633 1,0787
„
< >
Select All Samples
De-select All Samples
View All
Hide Data
View Graph
Figure 5.2 Ambient Samples for Test Case
Note that the "VCR" control buttons (red rectangle on toolbar) can be used to facilitate
navigation through the selection field, for purposes of selecting or de-selecting individual
samples.
Highlight any cell for the ambient sample for Fellows, CA (FELLOWS 07/26/88), as shown
in Figure 5.2, and see a graphical representation of the sample (View Graph), as shown in Figure
5.3. This graph may be printed from the print icon on that screen. Note that you may step
through the list of samples by using the "VCR" control buttons to view their graphical
representations. You may then return to the Sample screen (View Grid).
5-2
-------
HEPA-CMBB.2
File Help
I
Control File: INsjvf.inS
Samples: 6
5pecies: 19
Output Format: ASCII (.txt) -> Default
Input Files | Options Samples Species | Sources | Results
AMBIENT DATA SAMPLE - FELLOW 07/26/88
^ Tlnl AC
• S4IC
^ KPAC
^ NAAC
M ECTC
^ ALXC
• 5 IXC
M SUXC
• CLXC
^ KPXC
• CAXC
• TIXC
• VAXC
^ CRXC
• MNXC
• FEXC
^H NIXC
^ CUXC
• ZNXC
M BRXC
M PBXC
21. 536
5.147
0. 115
0.148
0.317
ill .295
0.943
2.016
0. 027
0. 163
0.178
0.029
0. 036
0. 006
0. 005
0.274
0. 136
0. 093
0.010
0. 023
±
± C
± 0
± L
± C
± 0
± 0
± C
± 0
± 0
± 0
± 0
± 0
± 0
± C
± C
± c
± C
± C
± r
1. 19
.270
. 023
. 058
.193
. 03
. 061
.102
. 007
. 010
. Oil
.006
. 003
. 001
. 001
.018
. 069
. 041
.000
. OOE
Species
Select All Samples
De-select All Samples
View AN
Hide Data
View Grid
Figure 5.3 Speciation Graph of Ambient Sample (Test Case)
Finally, note also that the counters on the status bar (top) reflect the number of ambient
samples, fitting species, or source profiles that are selected on their respective screens.9
Go to the Species screen to select an array of fitting species for the calculation (Figure 5.4).
As mentioned in Section 3.4, Array 1 will be always be initiated (highlighted) by default at
startup. Different arrays are selected by clicking on the index (integer) that labels each array.
Click on 2 to select the 20 fitting species "tagged" in Array 2 (yellow shading on screen view).
The number of species tagged in a selected array is indicated by the counter on the status bar
at the top of the screen. If only tagged species are of interest, choose View Selected to show
tagged species for the selected array (which toggles back via View All). Comments will appear
if present in the selection file (SP*.sel).
Note that the "VCR" control buttons operate in the same way as in the ambient samples
screen to facilitate navigation through a series of species in a selected array.
Note that at startup, initial counter values will appear as directed from any selection files (PR*.sel, SP*.sel, AD*.sel)
that have been loaded.
5-
-------
H EPA-CMB8.2
File He
1
P
^
Input Files
M
ll I J|
Options Samples Species
Fi-
SPECIE 1 SPECIE NAME
IB9MN03
S4IC 504
N4TC
KPAC
NAAC
ECTC
OCTC
ALXC
SIXC
SUXC
CLXC
KPXC
CAXC
TIXC
VAXC
CRXC
MNXC
FEXC
NIXC
CUXC
ZNXC
1
2
3
4
Control F
Sources
FTING SF
5
6
e: INsjvf.inS
Results
>ECIES i
7
011®
Samples: 6 Species: 20 Sources: 7 Output Format: ASCII (.txt) -> Default
i\RRAYS
8
* *
NH4
K-S
NA
EC
oc
AL
SI
5
CL
K
CA
TI
V
CR
MN
FE
NI
CU
ZN
BRXC BR
PBXC PB
#
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
9
10
COMMENT
Nitrate by 1C (ug/rn3)
Sulfate by 1C (ug/m3)
Ammonium by AC (ug/m3)
Soluble Potassium by AA (ug/m3)
Soluble Sodium by AA (ug/m3)
Elemental Carbon by TOR (ug/m3)
Organic Carbon by TOR (ug/m3)
Aluminum by XRF (ug/m3)
Silicon by XRF (ug/m3)
Sulfur by XRF (ug/m3)
Chloride by XRF (ug/m3)
Potassium by XRF (ug/m3)
Calcium by XRF (ug/m3)
Titanium by XRF (ug/m3)
Vanadium by XRF (ug/m3)
Chromium by XRF (ug/m3)
Manganese by XRF (ug/m3)
Iron by XRF (ug/m3)
Nickol by XRF (ug/m3)
Copper by XRF (ug/m3)
Zinc by XRF (ug/m3)
Bromine by XRF (ug/m3)
Lead by XRF (ug/m3)
>*
Select All Array 2
Clear All Array 2
View Selected
Figure 5.4 Fitting Species Selection for Test Case
Go to the Sources screen to select an array of fitting sources. As mentioned in Section 3.4,
Array 1 will be always be initiated (highlighted in) by default at startup. For fitting sources, use
the 6 shown in Array 5 on the screen (Figure 5.5) by clicking on its index (label) at the top. Note
that the "VCR" buttons operate in the same way as in the ambient samples screen to facilitate
navigation through a series of species in a given array.
Highlight any cell for any source and see a graphical representation of the profile (View
Graph), as shown in Figure 5.6 for Soil 12. This graph may be printed from the print icon on
that screen. As with the graphs of ambient samples, note that you may step through the list of
sources in a selected array by using the "VCR" control buttons to view their graphical
representations. You may then return to the Source screen (View Grid).
The number of sources "tagged" in a selected array is indicated by the counter on the status
bar at the top of the screen. If only tagged sources are of interest, choose View Selected to show
tagged sources for the selected array (which toggles back via View All). Comments will appear
if present in the selection file (PR*.sel). The comments field is followed by a series of fields for
abundance and uncertainty for all species in the source. The View Selected button shows only
the tagged sources(s) for the array selected, and values for species may be viewed by exploring
the field to the right by using the horizontal scroll. These values may also be hidden (Hide
Data).
5-4
-------
H EPA-CMBB.2
File He
Input Fil
PNO
5JV001
SJV002
SJV003
SJV004
5JV005
SJV006
5JVOQ7
SJV003
SJV009
SJV010
SJV011
5 JVO 12
5 JVO 13
S JVO 14
5JV015
SJV016
SJV017
S JVO 13
5 JVO 19
5JV02U
SJV021
SJV022
SJV023
SJV024
5JV025
SJV026
SJV027
< 1
P
-
W
Opt
SID
5OIL01
SOIL03
SOIL04
SOIL05
SOIL06
SOIL07
SOILOS
5OIL09
SOIL 10
SOIL 11
S87BM
SOIL 13
SOIL 14
SOIL 15
SOIL 16
SOIL 17
BAMAJC
MAMAJC
MAFISC
MADIEC
BAAGBC
ELAGBC
FRCONC
STAGBC
VIAGBC
VIDAIC
5FCRUC
*l i
Control File: IHsjvf.ine
ons 1 Samp es ] Species Sources
FITTING SOL
SIZE | 1
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
FINE
*
*
*
2 1 3 1 •»
lie
*
*
— —
»
»
»
*
m
*
5
#
«
*
Results
JRCES ft
6
? \
Samples: 6 Species: 20 Sources: 6
RRAYS
a | 9
ID | COMMENT
STOCKTON AGRICULTURAL SO
FRESNO PAVED ROAD
VISALIA AGRICULTURAL SOIL
VISALIA AGRICULTURAL SOIL
VISALIA SAND AND GRAVEL
VISALIA URBAN UNPAVED
VISALIA PAVED ROAD
BAKERSFIELD AGRICULTURAL
BAKERSFIELD AGRICULTURAL
BAKERSFIELD UNPAVED ROAD
BAKERSFIELD PAVED ROAD
BAKERSFIELD WINDBLOWN UR
BAKERSFIELD AGRICULTURAL
BAKERSFIELD AGRICULTURAL
BAKERSFIELD UNPAVED ROAD
TAFT UNPAVED ROAD
N3IC
0.0027
0
D
0.0003
0
0.001S
0.0018
0
0.0011
0.0005
0.0011
0.0013
0
0,0014
0.0024
0
BAKERSFIED CGRDWOOD., MAji 0.0046
MAMMOTH LAKES CORDWOOD, 0.0017
BAKERSFIELD CORDWOOD, FIS
MAMMOTH LAKES DIESEL TOUR
BAKERSFIELD AGRI. BURN (W
EL CENTRO AGRI. BURN (WHE
FRESNO HIGHWAY 40 CONSTR
STOCKTON AGRI. BURN (WHEA
VISALIA AGRI BURN (WHEAT)
VISALIA DAIRV/FEEDLOT DUS
SANTA FE CRUDE BOILER (WE
0.0003
0.0006
0.0065
0.0038
0.0176
0.0042
0.0035
0.0926
rrirnifxi
Output Format: A5CII (.txt) -> Default
rj 3iu
0,0047
0.0036
0.003
0.0026
0.0658
0.0029
0.0051
0.0023
0.0096
0.003
0.005S
0.0023
0.0039
0.0163
0.0018
0.0025
0.0012
0.0004
0.0001
0.0016
0.001
0.0028
0.0025
0.0007
0.039
Q| ° ~
m
Select All Array 5
Clear All Array 5
View Selected
Hide Data
View Graph
Figure 5.5 Fitting Source Profile Selection for Test Case
H EPA CMBB.2
File Help
Control File: INsjvf.inS
Output Format: ASCII (.txt) -> Default
Input Files | Options | Samples | Species Sources | Results
PROFILE - S3V011 SOIL 12
^ N3IC
•• S4IC
— N4TC
^m KPAC
^m NAAC
•• ECTC
^ OCTC
^ ALXC
— SIXC
^m suxc
^m CLXC
^m KPXC
^ TIXC
^ VAXC
^m CRXC
^m MNXC
^ FEXC
^ NIXC
— i CUXC
^ ZNXC
^B BRXC
^m PBXC
0. 001
0. Oil
0. 000
Cl. OO5
0 , 0 0 S
O. 015
0. 193
0. 005
0. 003
O. 019
0. 00-!
0. 000
o. ooo
0. 001
0. 057
0. 000
0. 000
0. 002
0. OOO
0.002
± 0.005
± 0.018
± 0.000
± 0.001
± O.008
± O.01O
± 0.022
± O.OOO
± 0.000
± O.002
± 0.000
± 0.000
± o.ooo
± o.ooo
± O.O06
± 0.000
± 0.000
± o.ooo
± o.ooo
± o.ooo
Species
Select All Array 5 |
Clear All Array 5
View Selected
I
Figure 5.6 Speciation Graph of Selected Source Profile (Test Case)
5-5
-------
Go to the Results screen to perform a calculation on the selected samples. (Note on the
status bar that 6 samples have been selected, as well as 20 fitting species and 6 fitting sources.)
As the information window reminds you, the Run button must be clicked in order to execute the
calculation (Figure 3.9). In this example, the model was run in a batch mode (>1 sample
selected; EPA-CMB8.2 calculations will always be performed for all samples selected). When
the calculation is finished, the Main Report appears showing the result for the first sample (i.e.,
Bakers 07/26/88). Using the "VCR" buttons (upper left-hand corner), advance through the series
of results for each sample until the results for Fellows, CA (FELLOW 07/26/88) is displayed, as
shown in Figure 5.7. Note that all the key information for the calculation is displayed in the
header: 6 samples, 20 fitting species and 6 sources.
HEPA-CMB8.2 Q@[H)
File
H
Help
A *• W i§ ^ Control File: INsjvf.inS Samples: 6 Species: 20 Sources: 6 Output Format: ASCII (.txt) -> Default
Input Files | Options | Samples | Species | Sources Results
SAMPLE Site: FELLOW Date: 07126188 Duration: 24 Start Hour: 0 Size: FINE
OPTIONS Iteration Delta: 20 l~~ Britt-Luecke l~ Best Fit Fit Measure Weights
Maximum Source Uncertainty (°b): 20 |~ Source Elimination
Minimum Source Projection: 0,95
Decimal Places Displayed: 5 Species Fit Array: 2
Units: ug/m^ Sources Fit Array: 5
j Main Report 1 Contributions by Species | MPIN Matrix •
FITTING STATISTICS:
R SQUARE 0.97 % MASS 90.5
CHI SQUARE 1.48 DEGREES FREEDOM 14
SOURCE CONTRIBUTION ESTIMATES:
SOURCE
EST CODE NAME SCE(ug/m=) Std Err Tstat
YES SJV011 SOIL12 4.44176 0.28844 15.39941
YES SJV017 BAMAJC 1.42855 0.39207 3.64360
YES SJV027 SFCRUC 4.50317 0.40427 11.13896
YES SJV036 MOVES2 2.66157 0.72729 3.65957
YES SJV051 AMSUL 6.20556 0.57253 10.83875
YES SJV054 AMNIT 0.24111 0.24847 0.97038
19.48172
MEASURED CONCENTRATION FOR SIZE: FINE
21. 5+- 1.2
Eligible Space Display
ELIGIBLE SPACE DIM. = 6 FOR MAX. UNC. = 4.30726 (20.3! OF TOTAL MEAS. MASS)
1 / Si ngul ar Val ue
0
.24291 0.26918 0.39729 0.40137 0.57745 0.73143
NUMBER ESTIMABLE SOURCES = 6 FOR MIN. PROJ. = 0.95
PROJ. SOURCE PROJ. SOURCE PROJ. SOURCE PROJ. SOURCE PROJ. SOURCE
1.
1.
0000 SJV011 1.0000 SJV017 1.0000 SJV027 1.0000 SJV036 1.0000 SJV051
0000 SJV054
ESTIMABLE LINEAR COMBINATIONS OF INESTIMABLE SOURCES
COEFF. SOURCE COEFF. SOURCE COEFF. SOURCE COEFF. SOURCE SCE Std Err
f.
•
<_ _>J
Result 3 of 6
Delete Current
Delete All
Run
Figure 5.7 Main Report - Test Case
5-6
-------
Note that settings for various options are echoed in the header, including the Britt &Luecke
flag, indicating whether or not the Britt and Luecke (1973) solution was selected, as well as
whether or not the fit was obtained using Source Elimination. As mentioned in Section 3.6.1, the
field 'EST' under 'SOURCE' indicates (YES or NO) whether a source's contribution was
estimable in EPA-CMB8.2's attempt at a fit using the settings in Options. The concept of
estimable sources is discussed in Section 6.1.2.
Note that for the Fellows sample, all 6 sources were estimable ("YES"), given their
uncertainty. Acceptable uncertainty is specified in the Options screen (Section 3.2). The %
mass explained can be readily verified by dividing the sum of the source contribution estimates
(19.48 i-igm"3) by the measured mass (21.5 |igm"3). Though Best Fit was not invoked, the default
Fit Measure Weights are displayed.
Examine the species concentrations display (farther down in the Main Report). Particular
attention should be paid to the ratios: calculated/measured and residual/uncertainty. See Section
6.1.3 for a detailed discussion of these and other performance measures.
Print this report in ASCII format by clicking on the Printer icon on the toolbar. An ASCII
(txt) file may also be created by choosing the Print to File option, then OK, which brings up a
dialog box for file disposition. As stated in Section 3.6.1, the header for this file is embellished
with an echo of all input files used in the calculation.
Go to the Contributions by Species screen, which is shown in Figure 5.8. Note that the
header information indicating the calculation parameters is carried through to this screen. Note
^EPA-CMBB.2
File
M
Help
"« *
Input Files |
w El
Options
a
Control File: INsjvf.inS
Samples Species | Sources Results
SAMPLE Site: FELLOW Date: 07/26/88 Duration: 24
OPTIONS
Iteration Delta: 20
MaKimum Source Uncertainty (%): 20
F Britt-Luecke
ELD®
Samples; 6 Species: 20 Sources: 6 Output Format: ASCII (.txt) -> Default
Start Hour: 0 Size: FINE
F Best Fit Fit Measure Weights
F Source Elimination chl Square i
Minimum Source Projection: 0.95
Decimal Places Displayed: 5
R Square i
Species Fit Array: 2
Units: ug/m' Sources Fit Array: 5
Main Report
SP
FCTFS
TMAC
N
S
N
K
N
3 1C
4 1C
4TC
PAC
flflC
ECTC
0
fl
S
S
c
K
C
T
V
C
M
F
N
C
Z
B
CTC
Lxc
IXC
IXC
LXC
PXC
nxc
IXC
AXC
RXC
NXC
EXC
IXC
IIXC
NXC
RXC
PBXC
<_
1
CALCULATE
1 MFA
19.481 21
0.25
5.57
1.75
0.08
0.07
1 0
3 5
J 1
2 0
q n
1.738 0
2.79
0.32
0.88
1.80
0.04
0.14
0.21
0.02
0.03
0.00
0.00
0.26
0.03
0.00
0.02
0.00
5 3
5 0
7 0
3 2
1 0
5 0
2 0
1 0
3 0
1 0
5 0
! 0
5 0
L 0
4 0
7 0
0.018 0
Illl
! Contributions by Species I MPIN Matrix
SURED
.5363
.2128
.1475
.9154
.1151
. 1488
.8172
.8926
.2951
.9433
.0164
.0275
.1632
.1782
.0298
.0366
.0066
.0058
.2746
.0379
.1383
.0932
.0108
.0231
SOURCE NAME
SOIL12 BAMA3C SFCRUC M
0.206 0.066 0.209 0
0.023 0.031 0.000 0
0.009 0.004 0.178 0
0.000 0.001 0.000 0
0.201 0.495 0.023 0
0.254 0.013 0.230 0
0.085 0.278 0.000 1
0.213 0.164 0.001 0
1.100 0.000 0.000 0
0.913 0.000 0.000 0
0.011 0.004 0.122 0
0.581 0.992 0.033 0
0.531 0.349 0.011 0
1.159 0.006 0.015 0
0.701 0.000 0.015 0
0.036 0.000 1.009 0
0.202 0.000 0.068 0
0.766 0.000 0.078 0
0.925 0.000 0.034 0
0.012 0.000 0.939 0
0.006 0.000 0.000 0
0.110 0.014 0.126 0
0.041 0.013 0.000 0
0.385 0.000 0.000 0
A
JVES2 AMSUL AMNIT
124 0.288 0.011
250 0.000 0.878
016 0.876 0.000
000 0.884 0.028
000 0 . 000 0 . 000
000 0 . 000 0 . 000
764 0.000 0.000
341 0.000 0.000
007 0.000 0.000
027 0.000 0.000
014 0.747 0.000
028 0.000 0.000
001 0.000 0.000
Oil 0.000 0.000
001 0.000 0.000
001 0.000 0.000
000 0 . 000 0 . 000
128 0.000 0.000
000 0 . 000 0 . 000
000 0 . 000 0 . 000
001 0.000 0.000
015 0.000 0.000
651 0.000 0.000
430 0.000 0.000 v
>l~
Result 3 of 6
Delete Current
Delete All
Run
Figure 5.8 Contribution by Species - Test Case
5-7
-------
also that the total measured mass for the sample also appears on the top line (TMAC). The
source names that appear are the mnemonics provided in the source profile input file (Section
4.2.3).
For convenience, both the calculated and measured concentration for each species are
repeated from the Main Report. All other values in the matrix are ratios. For any species, the
calculated value may be confirmed (regenerated) by multiplying each ratio by the measured mass
for that species, and then summing all the products.
This report may be printed in ASCII format by clicking on the Printer icon on the toolbar.
Depending on the number of sources selected in the calculation, printing may be impractical (the
right-hand side of the file may be truncated or there may be wrap-around problems).
Alternatively, a formatted ASCII (txt) file may also be created by choosing the Print to File
option, then OK, which brings up a dialog box for file disposition.
Go to the MPIN Matrix screen, which is shown in Figure 5.9. Note that the header
information indicating the calculation parameters is also carried through to this screen. As with
the Contributions by species screen, the source names that appear are the mnemonics provided in
the source profile input file (Section 4.2.3). Print this report by clicking on the Printer icon on
the toolbar. Depending on the number of sources selected in the calculation, printing may be
impractical (the right-hand side of the file may be truncated or there may be wrap-around
problems). Alternatively, a formatted ASCII file may also be created by choosing the Print to
File option, then OK, which brings up a dialog box for file disposition.
glEPA-CMBB.2 EMU*
File Help
M •* *_|w|Gl|i|i| Control File; INsjvf.inS Samples: 6 Species: 20 Sources: 6 Output Format: ASCII (, txt) -> Default
Input Files Options | Samples Species Sources Results
SAMPLE Site: FELLOW Date: 07126188 Duration: 24 Start Hour: 0 Size: FINE
OPTIONS Iteration Delta: 20 l~ Britt-Luecke P Best Fit Fit Measure Weights
Maximum Source Uncertainty (%): 20 F Source Elimination
Minimum Source Projection: 0.95
Decimal Places Displayed: 5 Species Fit Array: 2
Units: ug/m^ Sources Fit Array: 5
Main Report | Contributions by Species | MPIN Matrix ]
SOURCE NAME
SPECIES SOIL12 BAMAJC SFCRUC MOVES2 AMSUL AMNIT
N3IC 0.00 0.00 0.03 0.01 -0.25 1.00
S4IC -0.01 0.00 0.09 0.04 1.00 -0.13
N4TC 0.01 0.00 -0.10 -0.04 0.95 0.13
KPAC -0.10 0.60 0.01 -0.09 0.00 0.00
NAAC 0.06 0.00 0.05 -0.01 -0.01 0.00
ECTC -0.17 0.00 0.00 1.00 -0.01 -0.05
OCTC -0.05 0.21 -0.01 0.61 -0.01 -0.03
ALXC 0.88 -0.16 -0.03 -0.05 0.00 0.00
SIXC 0.90 -0.18 -0.03 0.02 0.00 0.00
CLXC -0.11 0.98 0.00 -0.13 0.00 0.00
KPXC 0.19 1.00 -0.01 -0.19 0.00 0.00
CAXC 1.00 -0.17 -0.02 -0.05 0.00 0.00
TIXC 0.37 -0.07 0.00 -0.03 0.00 0.00
VAXC -0.03 -0.01 1.00 0.00 -0.10 0.01
CRXC 0.08 -0.02 0.03 -0.01 0.00 0.00
MNXC 0.42 -0.12 0.03 0.21 -0.01 -0.01
FEXC 0.95 -0.17 0.00 -0.07 0.00 0.00
NIXC -0.05 0.00 0.95 -0.01 -0.10 0.01
BRXC -0.08 -0.10 0.00 0.69 -0.01 -0.03
PBXC 0.08 -0.11 -0.01 0.51 -0.01 -0.02 v
< INI | >~
Result 3 of 6
Delete Current
Delete All
Run
Figure 5.9 MPIN Matrix - Test Case
5-
-------
Once satisfied with the results, the output from EPA-CMB8.2 may be exported by clicking
on the File icon, then using the Save Results dialog. As discussed in Section 4.5, the output
format will be as selected in Options (default is ASCII).
The output file is a spreadsheet-type format and contains the contribution of each source to
each measured species in each sample (Section 4.3.2). Fitting species are identified by an
asterisk in the third column, and performance measures follow. Source contribution estimates
and their standard errors are presented in subsequent columns, identified by mnemonics in the
first row. Contributions from non-fitting profiles for a sample are identified as -99. Common
spreadsheet data analysis tools can be used to interrogate files and select records for different
chemical species, to group contributions from different profiles representing the same source
type, to calculate average contributions, and to plot results
Close down EPA-CMB8.2 before moving on to the next exercise.
5-9
-------
5.1.2 Individual Input File Operation
Restart EPA-CMB8.2, select the Individual Input Files mode, and a null screen like
the one shown in Figure 5.10 will appear, except that the Control File name is absent and the file
name boxes are blank. As described in Section 3.1, files are selected via their respective browse
dialogs and can be retrieved from any directory. For ambient and source profile data files, select
the ADsjvf.dbf and PRsjvf.dbf files, respectively. For the optional ambient data, species and
source profiles selection files, select the same ones used in the previous exercise, i.e., ADsjvf.sel,
SPsjvf.sel, and PRsjvf.sel, respectively.
HEPA-CMBB.Z
File Help
Samples:
Species:
Output Format: ASCII (.txt) -> Default
Input Files | Options | Samples | Species | Sources | Results
SELECT INPUT FILES
CONTROL FILE
REQUIRED INPUT FILES
Ambient Data
Source Profiles
OPTIONAL SELECTION FILES
Ambient Data
Species
Source Profiles
Figure 5.10 Input Files for Test Case - Individual Files Mode
From here, EPA-CMB8.2 will function exactly as it did in the previous exercise. After
options are set, a sample(s) chosen, and arrays of fitting species and source profiles are selected,
a calculation can be performed. Note that you can also go back and change any of the previous 5
files on the Input Files screen. If a selection file is changed, the following message will appear,
indicating that EPA-CMB8.2 has detected a change in operating status. If the change is as you
want, confirm the change (answer 'Yes') and continue. Note also that if no selection file(s)
is(are) used, nothing will be initially selected and selection arrays will be blank.
•> You are running without a Control File and one or more input files have changed this session.
.^*J Do you want to start a new CMB session?
(This will purge any previous results and reload with new input files; NO = cancel.)?
No
Close down EPA- CMB8.2 before moving on to the next exercise.
5- 10
-------
5.2 Best Fit Option
Restart EPA-CMB8.2 and select the Control File mode. Select INsjvfBF.inS and
Open. A screen as shown in Figure 5.11 will appear. Go to the Options screen and select the
Best Fit option. The default Fit Measure weights will be initialized at 1.000. Go to the Samples
screen and note that 15 are initially selected. Click on View Selected to show only these 15
HEPA-CMBS.Z
File Help
Control File: INsjvfBF.inS
Samples:
Species:
Output Format: ASCII (.txt) -> Default
Input Files;] Options | Samples | Species | Sources | Results
SELECT INPUT FILES
CONTROL FILE |lNsjvfBF.in8
REQUIRED INPUT FILES
Ambient Data |ADsjvf.txt
&Brc
Source Profiles |PRsjvf.btt
& Browse..
OPTIONAL SELECTION FILES
Ambient Data
Species
Source Profiles
|ADsjvfBF.sel
&Brc
|SPsjvfBF.sel
& Browse..
|pRsjvfBF.sel
Figure 5.11 Input Files for Testing the Best Fit Option
samples. Click on De-select All Samples and then click on Select All to reestablish the list of
206 ambient samples. Again click De-select All Samples and, using the "VCR" forward arrow
or the vertical scroll arrow on the right -hand side of the screen, move down the list and select
the FRESNO 02/27/89 sample.
Go to the Results screen and perform a calculation using the Best Fit option selected.
It does not matter which fitting sources and species arrays are selected because, as described in
Section 3.2, EPA-CMB8.2 will first reset the array indices to 1 and then step through
corresponding array pairs looking for a maximum of the Fit Measure, using the weights supplied
on the Options screen (Section 6.3). If nonconvergence is encountered, the message "COLUMN
NUMBER X of AFIT = 0" may appear (see Appendix E). In this case, "COLUMN NUMBER"
refers to the source profile. Choose 'OK' and continue.
5- 11
-------
When EPA-CMB8.2 is finished, Arrays 4 for fitting species and fitting sources will have
been selected as giving the best overall performance measure (value = 0.72), as shown in Figure
5.12. Returning to the species and source profile selection screens will reveal that indeed the 4th
array is selected for each. Note that the Fit Measure Weights are echoed on the Results screen.
These will also be echoed in printed reports.
61EPA-CMBU
File Help
- n-x
Input Files I Options | Samples I Species
Control File: INsjvfBF.inS
Sources
Samples: 1
Species: 20
Sources: 7
Output Format: ASCII (.txt) -> Default
Results j
SAMPLE Site: FRESNO Date: 02/27/89 Duration: 24
Start Hour: 0
OPTIONS Iteration Delta: 20
Maximum Source Uncertainty (%): 20
Minimum Source Projection: 0,95
Decimal Places Displayed: 5
Units: ug/m!
F Britt-Luecke 17 Best Fit
f Source Elimination
Species Fit Array: 4
Sources Fit Array: 4
Size: FINE
Fit Measure Weights
Chi Square i
R Square i
%Mass i
Fraction Estimate i
Main Report i Contributions by Species
WIN Matrix
Result 1 of 1
Delete Current
FITTING STATISTICS:
R SQUARE
CHI SQUARE
FIT MEASURE
0.96
1.06
0.72
% MASS
DEGREES FREEDOM
83.2
13
SOURCE CONTRIBUTION ESTIMATES:
SOURCE
EST CODE NAME SCE(ng/m3) Std Err Tstat
YES SJVOQ2 SOIL03 1.22757 0.15884 7.72S09
YES SJV017 BAMAJC 3.45106 0.77033 4.47998
YES SJVQ27 SFCRUC 0.20426 0.12900 1.58336
YES SJV036 MOVES2 7.38975 1.82640 4.04607
YES SJV051 AMSUL 3.57262 0.55662 6.41843
YES SJVQ54 AMNIT 25.35610 2.11060 12.01368
YES SJVQ56 NAN03 0.67954 0.35110 1.93544
41.88089
MEASURED CONCENTRATION FOR SIZE: FINE
50.3+- 2.6
Eligible Space Display
ELIGIBLE SPACE DIM. = 7 FOR MAX. UNC. = 10.06866 (20.% Of TOTAL MEAS. MASS)
1 / Singular Value
0.12852 0.15620 0.34974 0.54445 0.76252 1.83131 2.11281
NUMBER ESTIMABLE SOURCES = 7 FOR MIN. PRQJ. = 0.95
PROJ. SOURCE PROJ. SOURCE PROJ. SOURCE PROJ. SOURCE PROJ. SOURCE
1.0000 SJV002 1.0000 SJV017 1.0000 SJV027 1.0000 SJV036 1.0000 SJV051
1.0000 SJV054 1.0000 SJV056
ESTIMABLE LINEAR COMBINATIONS OF INESTIMABLE SOURCES
COEFF. SOURCE COEFF. SOURCE COEFF. SOURCE COEFF. SOURCE SCE Std Err
Delete All
Run
•
Figure 5.12 Results for Test Case Using Best Fit Option
5- 12
-------
6. EPA-CMB8.2 PERFORMANCE MEASURES
This section describes the different performance measures that are used to evaluate the
validity of source contribution estimates. Greater detail on the use of the performance measures
is presented in EPA (2004).
6.1 Main Report.
As discussed earlier (Section 3.6.1), the performance measures are presented in four
separate groups in the Main Report produced by EPA-CMB-8.2: 1) fitting statistics; 2) source
contribution estimates, (3) eligible space display; and (4) the species concentration display.
Each of these displays is discussed below.
6.1.1 Source Contribution Estimates and Fitting Statistics
An example of a source contribution table display is shown below for the Fresno, CA
sample (San Joaquin Valley data set):
Chemical Mass Balance Version EPA-CMB8.2
Report Date: 10/25/2004
SAMPLE: OPTIONS: INPUT FILES:
SITE: FRESNO BRITT & LUECKE: No INsjvfBF.inS
SAMPLE DATE: 02/27/89 SOURCE ELIMINATION: No PRsjvfBF.se!
DURATION: 24 BEST FIT: Yes SPsjvfBF.se!
START HOUR: 0 FIT MEASURE WEIGHTS: ADsjvfBF.se!
SIZE: FINE R Square: 1 ADsjvf.txt
Chi Square: 1 PRsjvf.txt
Species Array: 4 % Mass: 1
Sources Array: 4 Fraction Estimate: 1
FITTING STATISTICS:
R SQUARE
CHI SQUARE
FIT MEASURE
SOURCE CONTRIBUTION
SOURCE
EST CODE NAME
YES SJV002 SOIL03
YES SJV017 BAMAJC
YES SJV027 SFCRUC
YES SJV036 MOVES2
YES SJV051 AMSUL
YES SJV054 AMNIT
YES SJV056 NAN03
0.96
1.06
1.228
ESTIMATES:
SCE(|jg/m3)
1.22757
3.45106
0.20426
7.38975
3.57262
25.35610
0.67954
41.88090
MEASURED CONCENTRATION FOR SIZE:
50. 3+- 2.6
% MASS 83.2
DEGREES FREEDOM 13
Std Err Tstat
0.15884 7.72809
0.77033 4.47998
0.12900 1.58336
1.82640 4.04607
0.55662 6.41843
2.11060 12.01368
0.35110 1.93544
FINE
-1
-------
Source contribution estimates are the main output of EPA-CMB8.2. The sum of these
concentrations approximates the total mass concentration. Negative source contribution
estimates are not physically meaningful, but they can occur when a source profile is collinear
with another profile or when the source contribution is close to zero. Collinearity is usually
identified in the eligible sources display. When the absolute value of a positive or negative
source contribution estimate is less than its standard error, the source contribution is
undetectable. Two or three times the standard error may be taken as an upper limit of the source
contribution in this case.
The standard errors reflect the uncertainty of the ambient data, the source profiles, and
the amount of collinearity among different profiles. Standard errors should be reported with
every source contribution estimate. The standard error is a single standard deviation. There is
about a 66% probability that the true source contribution is within one standard error and about a
95% probability that the true contribution is within two standard errors of the source contribution
estimate.
The T-statistic (Tstat) is the ratio of the source contribution estimate to the standard
error. A Tstat value less than 2.0 indicates that the source contribution estimate is at or below a
detection limit. Low Tstat values for several source contributions may be caused by
collinearities among their profiles; this will be indicated in the Eligible Space Collinearity
Display (Section 6.2).
The reduced chi-square, degrees of freedom10, R-square, percent mass, and fit measure
are performance measures for the least squares calculation. The chi-square is the weighted sum
of squares of the differences between the calculated and measured fitting species concentrations.
The weighting is inversely proportional to the squares of the uncertainty in the source profiles
and ambient data for each species. Ideally, there would be no difference between calculated and
measured species concentrations and chi-square would equal zero. A value less than 1 indicates
a very good fit to the data, while values between 1 and 2 are acceptable. Chi-square values
greater than 4 indicate that one or more species concentrations are not well explained by the
source contribution estimates. The degrees of freedom equal the number of fitting species minus
the number of fitting sources. The degrees of freedom is needed when statistical significance
tests are applied to the chi-square value.
The R-square is the fraction of the variance in the measured concentrations that is
explained by the variance in the calculated species concentrations. It is determined by a linear
regression of measured versus model-calculated values for the fitting species. R-square ranges
from 0 to 1.0. The closer the value is to 1.0, the better the source contribution estimates explain
the measured concentrations. When R-square is less than 0.8, the source contribution estimates
do not explain the observations very well with the fitting source profiles and/or species.
The degrees of freedom (DF = no. fitting species - no. fitting sources) value is used to determine the statistical
significance for a given confidence level (e.g., 99%). The test statistic is computed by normalizing (dividing) the %2 value by
DF. With typical degrees of freedom of 13 to 20 using commonly measured ions, carbon and elements, a %2/DF value less than
about 3 is not significant at the 99% confidence level. When many fitting species are used (as when apportioning organic
compounds), the critical value for the test statistic (%2/DF) becomes smaller (e.g., <2).
6-2
-------
Percent mass is the percent ratio of the sum of the model-calculated source
contribution estimates to the measured mass concentration. This ratio should equal 100%,
although values ranging from 80 to 120% are acceptable. If the measured mass is very low (< 5
to 10 |ig/m3), percent mass may be outside of this range because the uncertainty of the mass
measurement is on the order of 1 to 2 |ig/m3.
6.1.2 Eligible Space Collinearity Display
Maximum source uncertainty (Section 3.2) defines the eligible space as that spanned
by eigenvectors with inverse singular values less than or equal to the maximum source
uncertainty. Sources lying within the eligible space may be estimated with an uncertainty less
than the maximum source uncertainty. In practice, this strict criterion of inclusion is relaxed
somewhat so that an estimable source is defined to be one whose projection into the eligible
space is at least the minimum source projection. Inestimable sources have small projections
within the eligible space. Certain linear combinations of inestimable sources may be estimable,
and the program lists any of these that may exist. This may be understood as removing
uncertainty by combining collinear sources. Different values for the maximum source
uncertainty (ranging from 0 to 100%) and minimum source projections (ranging from 0 to 1.0)
may be adjusted on the Options screen and will be retained for subsequent source contribution
calculations during the session.
The eligible space display identifies the potential for collinearity and the potential
reductions in standard errors in the source contribution estimates when source profiles are
combined. An example appears below for the same Fresno, CA sample:
Eligible Space Collinearity Display
ELIGIBLE SPACE DIM. = 7 FOR
1 / Singular Value
0.12852 0.15620 0.34974
NUMBER ESTIMABLE SOURCES = 7
PROJ. SOURCE PROJ. SOURCE
1.0000 SJV002 1.0000 SJV017
1.0000 SJV054 1.0000 SJV056
ESTIMABLE LINEAR COMBINATIONS
COEFF. SOURCE COEFF. SOURCE
MAX. UNC. = 10.
06866 (20.X OF
0.54445 0.76252 1.83131
FOR MIN. PROJ.
PROJ. SOURCE
1.0000 SJV027
OF INESTIMABLE
COEFF. SOURCE
= 0.95
PROJ. SOURCE
1.0000 SJV036
SOURCES
COEFF. SOURCE
TOTAL MEAS. MASS)
2.11281
PROJ. SOURCE
1.0000 SJV051
SCE Std Err
Henry's (1992) eligible space treatment uses the maximum source uncertainty, expressed as a
percentage of the total measured mass, and the minimum source projection. These may be
changed from their default values of 20% and 0.95, respectively, in the Options menu. As stated
earlier, the maximum source uncertainty defines a space, called the eligible space, to be that
spanned by those eigenvectors with inverse singular values less than or equal to the maximum
-------
source uncertainty. The first part of this display gives the eligible space dimension and the
uncertainty used in its calculation. This is followed by a listing of the inverse singular values.
It was mentioned in Section 1.1 that EPA-CMB8.2 gives the user control in adjusting
collinearity parameters which in CMB7 are "hard-wired" and not necessarily optimum for every
application. Historically these parameters were chosen in CMB7 to be compatible with
characteristics of paniculate mass measurements, particularly those made by X-ray fluorescence.
On an absolute basis with different measurement units, uncertainties associated with VOC may
be up to an order of magnitude lower than those for particles. In an effort to resolve this
discrepancy, the maximum source uncertainty (threshold set in Options; Section 3.2) in EPA-
CMB8.2 is expressed as ^percentage ("unit neutral").
The next part of the display gives the number of estimable sources, the minimum
source projection used in the calculation, and the projections of each profile vector into the
eligible space. Inestimable sources are caused by excessive similarity (collinearity) among the
source profiles or by high uncertainties in the individual source profiles. The standard errors
associated with the source contribution estimates of one or more inestimable sources are usually
very large, often too large to allow an adequate separation of these source contributions to be
made. Inestimable sources will not appear if the two above-stated conditions (i.e., collinearity or
large std. error of the SCE) do not occur. An absence of inestimable sources means that the
source contributions can be resolved in the specific application. Since ambient data
uncertainties, and relative levels of source contributions, vary from sample to sample, it is
possible that a given set of profiles may appear in the ineligible (inestimable) space for one set of
ambient data, but not for another set. For this reason, it is impossible to decide a priori that a set
of profiles is collinear or not. The decision must be made for each set of data and each set of
profiles combined with those data.
If collinearity is the cause of these excessive standard errors, then certain linear
combinations of inestimable sources may be estimable, and the final part of the display lists
these, if any exist. This may be understood as removing uncertainty by combining collinear
sources. This linear combination may be more useful than the individual source contribution
estimates if the standard error of the linear combination is substantially lower than the standard
errors of each source contribution estimate. However, the treatment does not allow
differentiation among the contribution estimates of the sources contained in the linear
combination. Also, as stated above for the individual source data, the number of decimal places
used in the presentation of the inverse singular values is that set in the Options menu. For more
discussion of collinearity, see Section 4.1 of EPA, 2004.
-4
-------
6.1.3 Species Concentration Display
An example of the species concentration display is shown below for the same Fresno,
CA sample:
SPECIES CONCENTRATIONS:
SPECIES FIT
TMAC
N3IC
S4IC
N4TC
KPAC
NAAC
ECTC
OCTC
ALXC
SIXC
SUXC
CLXC
KPXC
CAXC
TIXC
VAXC
CRXC
MNXC
FEXC
NIXC
CUXC
ZNXC
BRXC
PBXC
TMAU
N3IU
S4IU
N4TU
KPAU
NAAU
ECTU
OCTU
ALXU
SIXU
SUXU
CLXU
KPXU
CAXU
TIXU
VAXU
CRXU
MNXU
FEXU
NIXU
CUXU
ZNXU
BRXU
PBXU
50
* 19
* 2
* 7
* o
* o
* 4
* 5
* o
* o
1
* o
* o
* o
* o
* o
* o
* o
* o
* o
0
0
* o
* o
MEASURED
.34330+-
.26080+-
.87790+-
.04960+-
.14960+-
.19820+-
.55270+-
.99850+-
.06410+-
.18690+-
.09520+-
.06410+-
.16950+-
.04500+-
.00060<
.00160<
.00200+-
.00490+-
.11250+-
.00170+-
.02140<
.02950<
.01660+-
.03990+-
2
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
.56520
.97930
.16530
.36360
.02350
.05660
.59790
.84490
.02420
.03920
.05650
.00800
.01070
.00710
.01930
.00810
.00170
.00090
.01290
.00100
.06790
.04030
.00100
.00560
CALCULATED
41.88089+-
20.31037+-
2.92384+-
6.69662+-
0.14187+-
0.19290+-
4.57617+-
5.42610+-
0.11887+-
0. 33528+-
0. 97978+-
0.07043+-
0.16243+-
0.04837+-
0.00648<
0.00212<
0.00039+-
0.00356+-
0.07588+-
0.00174+-
0.00062<
0.01050<
0.01998+-
0.03198+-
2
1
0
0
0
0
1
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
.77273
.97128
.36943
.58473
.08541
.07618
.47563
.80669
.01347
.10710
.12324
.02214
.04288
.00768
.00103
.00041
.00015
.00178
.00860
.00024
.00022
.00250
.01123
.01530
CALCULATED
RESIDUAL
MEASURED UNCERTAINTY
0.83+-
1.05+-
1.02+-
0.95+-
0.95+-
0.97+-
1.01+-
0.90+-
1.85+-
1.79+-
0.89+-
1.10+-
0.96+-
1.07+-
10.80< '
1.32<
0.19+-
0.73+-
0.67+-
1.02+-
0.03<
0.36<
1.20+-
0.80+-
0.07
0.12
0.14
0.10
0.59
0.47
0.35
0.33
0.73
0.69
0.12
0.37
0.26
0.24
t****
6.70
0.18
0.39
0.11
0.62
0.09
0.49
0.68
0.40
-2.2
0.5
0.1
-0.5
-0.1
-0.1
0.0
-0.3
2.0
1.3
-0.9
0.3
-0.2
0.3
0.3
0.1
-0.9
-0.7
-2.4
0.0
-0.3
-0.5
0.3
-0.5
This display shows how well the individual ambient concentrations are reproduced by
the source contribution estimates. This display offers clues concerning which sources might be
missing or which ones do not belong in the calculation. Fitting species are marked with an
asterisk in the column labeled ' FIT '. Note that the symbol < flags measured values less than
the uncertainty. The block of asterisks indicate numerical overflow, meaning that the number is
too large to be displayed in the allocated field.
The column labeled CALCULATED / MEASURED displays, for each species, the
ratio of the calculated (C) to measured (M) concentrations ± the standard error of the ratio for
every chemical species with measured data:
std. err =
C_
M'
err,
c
C
err,
M
M
-5
-------
MJ M2
where:
errc = error associated with the calculated concentration, and
errM = error associated with the measured concentration.
The column labeled RESIDUAL / UNCERTAINTY displays, for each species, the
ratio of the signed difference between the calculated and measured concentrations (residual =
calculated - measured) divided by the uncertainty (standard error) of that residual:
std. err = Jerr2 + errf^
Note that the uncertainty is the square root of the sum of the squares of the uncertainty in the
calculated and measured concentrations. The ratio specifies the number of uncertainty intervals
by which the calculated and measured concentrations differ. When the absolute value of this
ratio exceeds 2, the residual is significant. If it is positive, then one or more of the profiles is
contributing too much to that species. If it is negative, then there is an insufficient contribution
to that species and a source may be missing. The sum of the squared ratio for fitting species
divided by the degrees of freedom yields the chi square. The highest ratio values for fitting
species are the cause of high chi square values. Also, as above for the individual source data, the
number of decimal places used in the presentation of the species data is that set in the Options
menu.
As an example, if the calculated/measured ratio for TI was 10.8, this would indicate
that TI is overestimated (over-explained) by EPA-CMB8.2 by an order of magnitude. While this
condition could be within uncertainty limits, this behavior suggests that TI might be a candidate
for removal as a fitting species, or that a source having a source contributing a large amount of
TI might be a candidate for removal as a fitting source.
6-6
-------
6.2 Additional Performance Measures
Another analysis available from the Results screen is the Contributions by Species
report, which is a table showing the fraction of each species' calculated ambient concentration
contributed by each source in the fit. The sources that are major contributors to each species can
be determined by examining this display. An example of this display is shown below for the
Fresno, CA sample, where it appears that the titanium concentration is substantially
overestimated by the SOIL03 profile. While this condition could be within uncertainty limits, it
might be advisable to substitute another profile with a lower titanium abundance, assuming the
alternative profile is not unsuitable in other ways.
Chemical Mass Balance Version EPA-CMB8
Report Date: 10/25/2004
SAMPLE: OPTIONS:
SITE: FRESNO BRITT & LUECKE: No
SAMPLE DATE: 02/27/89 SOURCE ELIMINATION: No
DURATION: 24 BEST FIT: Yes
START HOUR: 0 FIT MEASURE WEIGHTS:
SIZE: FINE R Square: 1
Chi Square: 1
Species Array: 4 % Mass: 1
Sources Array: 4 Fraction Estimate: 1
Contributions by Species:
SPECIES
TMAC
N3IC
S4IC
N4TC
KPAC
NAAC
ECTC
OCTC
ALXC
SIXC
SUXC
CLXC
KPXC
CAXC
TIXC
VAXC
CRXC
MNXC
FEXC
NIXC
CUXC
ZNXC
BRXC
CALCULATED
41
20
2
6
0
0
4
5
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
880
310
923
696
141
192
576
426
118
335
979
070
162
048
006
002
000
003
075
001
000
010
019
MEASURED
50
19
2
7
0
0
4
5
0
0
1
0
0
0
0
0
0
0
0
0
0
0
0
3433
2608
8779
0496
1496
1982
5527
9985
0641
1869
0952
0641
1695
0450
0006
0016
0020
0049
1125
0017
0214
0295
0166
SOIL03
0.024
0.000
0.002
0.000
0.027
0.014
0.006
0.034
1.766
1.415
0.006
0.036
0.142
0.900
10.639
0.230
0.184
0.301
0.670
0.072
0.011
0.100
0.007
BAMAJC
0.069
0.001
0.017
0.000
0.920
0.024
0.120
0.257
0.000
0.000
0.016
1.028
0.812
0.054
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.105
0.021
SFCRUC
0.004
0.000
0.014
0.000
0.001
0.008
0.000
0.000
0.000
0.000
0.010
0.001
0.000
0.003
0.034
1.047
0.010
0.004
0.004
0.949
0.000
0.018
0.000
MOVES2
0.147
0.008
0.080
0.000
0.000
0.000
0.879
0.614
0.089
0.378
0.070
0.033
0.003
0.118
0.123
0.046
0.000
0.422
0.001
0.000
0.017
0.133
1.175
.2
INPUT FILES:
INsjvfBF.inS
PRsjvfBF.se!
SPsjvfBF.sel
ADsjvfBF.sel
ADsjvf.txt
PRsjvf.txt
AMSUL
0.071
0.000
0.902
0.138
0.000
0.000
0.000
0.000
0.000
0.000
0.792
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
AMNIT
0.504
1.020
0.000
0.811
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
NAN03
0.013
0.026
0.000
0.000
0.000
0.927
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
-7
-------
Yet another diagnostic available from the Results screen is the transpose of the
normalized modified pseudo-inverse matrix (MPIN). This matrix indicates the degree of
influence each species concentration has on the contribution and standard error of the
corresponding source category. MPIN is normalized such that it takes on values from -1 to 1.
Species with MPIN absolute values of 0.5 to 1.0 are considered influential species.
Noninfluential species have MPIN absolute values of 0.3 or less. Species with absolute values
between 0.3 and 0.5 are ambiguous but should generally be considered noninfluential. There are
a number of useful references on identifying influential species (e.g., Belsley et a/., 1980; Kim
and Henry, 1999).
An example display of this diagnostic for the Fresno, CA sample is shown below:
SAMPLE:
SITE:
SAMPLE DATE:
DURATION:
START HOUR:
SIZE:
Species Array:
Sources Array:
MPIN Matrix:
SPECIES SOIL03
N3IC
S4IC
N4TC
KPAC
NAAC
ECTC
OCTC
ALXC
SIXC
CLXC
KPXC
CAXC
TIXC
VAXC
CRXC
MNXC
FEXC
NIXC
BRXC
PBXC
-0.01
0.00
0.01
-0.02
0.00
-0.17
-0.11
0.83
0.43
-0.05
0.04
0.76
0.07
0.00
0.04
0.08
1.00
-0.02
-0.11
-0.05
Chemical Mass Balance Version EPA-CMB8.2
Report Date: 10/25/2004
OPTIONS: INPUT FILES:
FRESNO BRITT & LUECKE: No INsjvfBF.inS
02/27/89 SOURCE ELIMINATION: No PRsjvfBF.sel
24 BEST FIT: Yes SPsjvfBF.sel
0 FIT MEASURE WEIGHTS: ADsjvfBF.sel
FINE R Square: 1 ADsjvf.txt
Chi Square: 1 PRsjvf.txt
4 % Mass: 1
4 Fraction Estimate: 1
BAMAJC
0.00
0.00
0.00
0.50
0.00
-0.01
0.16
-0.06
-0.06
0.90
1.00
0.01
0.00
0.00
0.00
-0.06
-0.06
0.00
-0.07
-0.08
SFCRUC
0.01
0.00
-0.01
0.00
0.00
0.00
0.00
-0.02
-0.01
0.00
0.00
-0.01
0.00
0.13
0.01
0.00
0.00
1.00
0.00
0.00
MOVES2
0.04
0.01
-0.04
-0.08
0.00
1.00
0.71
-0.06
0.17
-0.10
-0.17
0.06
-0.01
0.00
-0.01
0.39
-0.17
0.00
0.70
0.68
AMSUL
-0.11
1.00
0.12
0.00
0.01
-0.08
-0.06
0.01
-0.01
-0.01
0.00
-0.01
0.00
-0.01
0.00
-0.03
0.01
-0.06
-0.06
-0.06
AMNIT
1.00
-0.20
0.90
0.00
-0.12
0.01
0.01
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.01
0.00
0.00
NAN03
0.05
0.01
-0.06
0.00
1.00
0.00
0.00
0.00
0.00
-0.01
-0.01
0.00
0.00
0.00
0.00
0.00
0.00
-0.01
0.00
0.00
-------
6.3 Best Fit Measure
As discussed in Section 3.2, when more that one species or sources selection arrays are
provided, EPA-CMB8.2 may be run in a Best Fit mode to have the model iterate among possible
combinations of arrays and determine which results in an overall best fit. To do this, EPA-
CMB8.2 uses a criterion called the Best Fit Measure that is calculated for each possible array
pair combination. The Best Fit Measure is a linear combination of four factors related to
fundamental performance measures that are combined as follows:
(FracEst]
% mass explained < 100
FM = X J _ 0 mass _ % mass explained > 1 00
W1 + W2 + W3 + W4
Where
FM = Fit Measure
X2, R2, and % mass explained are the performance measures calculated by EPA-CMB8.2
FracEst is the ratio of the number of estimable fitting sources to the total number of fitting
sources.
Wj = x2 weight
W2 = R2 weight
W3 = % mass weight
W4 = fraction estimated weight
As mentioned in Section 3.2, these weights are set to 1.0 by default, and may be changed in the
Options screen. And as mentioned in Section 5.1.3, for a given set of (non-negative) weight
values, a larger value for FM corresponds to a better fit by EPA-CMB8.2.
6-9
-------
-------
7. REFERENCES
Anderson, E, Z. Bai, C. Bischof, S. Blackford, J. Demmel, J. Dongarra, J. Du Croz, A.
Greenbaum, S. Hammarling, A. McKenney and D. Sorensen, 1999. LAPACK Users'
Guide, Third Edition. Society for Industrial and Applied Mathematics (SIAM),
Philadelphia, PA; ISBN 0-89871-447-8 (paperback).
Earth, D., 1970. Federal motor vehicle emissions goals for CO, HC, and NOX based on
desired air quality levels. JAPCA 20:519.
Belsley, D.A., E. Kuh and R.E. Welsch, 1980. Regression Diagnostics: Identifying
Influential Data and Sources of Collinearity. John Wiley & Sons, Inc., New York.
Britt, H.I. and R.H. Luecke, 1973. The estimation of parameters in nonlinear, implicit
models. Technometrics 15: 233.
Cass, G.R. and GJ. McRae, 1981. Minimizing the cost of air pollution control. Environ.
Sci. Technol. 15: 748-57.
Chang, T.Y. and B. Weinstock, 1975. Generalized rollback modeling for urban air pollution
control. JAPCA 25: 1033-7.
Cheng, M.D. and P.K. Hopke, 1989. Identification of markers for chemical mass balance
receptor model. Atmos. Environ. 23: 1373-84.
Chow, J.C., J.G. Watson, D.H. Lowenthal, L.C. Pritchett and L.W. Richards, 1990. San
Joaquin Valley Air Quality Study, Phase 2: PM10 modeling and analysis, Volume I:
Receptor modeling source apportionment, Final Report. Report No. DRI 8929. IF,
prepared by Desert Research Institute, Reno, NV.
Chow, J.C., J.G. Watson, D.H. Lowenthal, P.A. Solomon, K.L. Magliano, S.D. Ziman and
L.W. Richards, 1992. PM10 source apportionment in California's San Joaquin Valley.
Atmos. Environ. 26A: 3335-54.
Chow, J.C., J.G. Watson, D.M. Ono and C.V. Mathai, 1993. PM10 standards and
nontraditional particulate source controls: A summary of the A&WMA/EPA international
specialty conference. JAWMA 43: 74-84.
Cooper, J.A. and J.G. Watson, 1980. Receptor oriented methods of air particulate source
apportionment. JAPCA 30: 1116-25.
Coulter, C.T. and J.V. Scalco, 2005. Chemical Mass Balance Software: EPA-CMB8.2.
Proceedings of A&WMA's 98th Conference & Exhibition; Minneapolis, MN, June 21-
24, 2005.
Currie, L.A., R.W. Gerlach, C.W. Lewis, W.D. Balfour, J.A. Cooper, S.L. Dattner, R.T.
deCesar, G.E. Gordon, S.L. Heisler, P.K. Hopke, J.J. Shah, G.D. Thurston andH.J.
Williamson, 1984. Interlaboratory comparison of source apportionment procedures:
Results for simulated data sets. Atmos. Environ. 18: 1517.
deCesar, R.T., S.A. Edgerton, M.A.K. Khalil and R.A. Rasmussen, 1985. Sensitivity
analysis of mass balance receptor modeling: methyl chloride as an indicator of wood
smoke. Chemosphere 14: 1495-501.
7- 1
-------
deCesar, R.T., S.A. Edgerton, M.A.K. Khalil and R.A. Rasmussen, 1986. A tool for
designing receptor model studies to apportion source impacts with specified precision. In
Transactions, Receptor Methods for Source Apportionment: Real World Issues and
Applications, Pace, T.G., editor. Air Pollution Control Association, Pittsburgh, PA. pp.
56-67.
deNevers, N. and J.R. Morris, 1975. Rollback modeling: basic and modified. JAPCA 25:
943.
Dzubay, T.G., R.K. Stevens, W.D. Balfour, HJ. Williamson, J.A. Cooper, I.E. Core, R.T.
deCesar, E.R. Crutcher, S.L. Dattner, B.L. Davis, S.L. Heisler, JJ. Shah, P.K. Hopke and
D.L. Johnson, 1984. Interlaboratory comparison of receptor model results for Houston
aerosol. Atmos. Environ. 18: 1555.
Environmental Protection Agency, 1990. Receptor Model Technical Series, Volume III
(1989 Revision). CMB7 User's Manual. Report Number EPA-450/4-90-004. Office of
Air Quality Planning and Standards, Research Triangle Park, NC.
Environmental Protection Agency, 2004. Protocol for Applying and Validating the CMB
Model for PM2.5 and VOC. Report No. EPA-451/R-04-001. December 2004. Office of
Air Quality Planning & Standards, Research Triangle Park, NC.
Friedlander, S.K., 1973. Chemical element balances and identification of air pollution
sources. Environ. Sci. Technol. 7: 235-40.
Friedlander, S.K., 1981. New developments in receptor modeling theory. In Atmospheric
Aerosol: Source/Air Quality Relationships, Macias, E.S. and Hopke, P.K., editors.
American Chemical Society, Washington, D.C. pp. 1-19.
Fujita, E.M., J.G. Watson, J.C. Chow and Z. Lu, 1994. Validation of the chemical mass
balance receptor model applied to hydrocarbon source apportionment in the Southern
California Air Quality Study. Environ. Sci. Technol. 28: 1633-49.
Gartrell, G. and S.K. Friedlander, 1975. Relating particulate pollution to sources: The 1972
California Aerosol Characterization Study. Atmos. Environ. 9: 279-99.
Gordon, G.E., 1980. Receptor models. Environ. Sci. Technol. 14: 792-800.
Gordon, G.E., 1988. Receptor models. Environ. Sci. Technol. 22: 1132-1142.
Gordon, G.E., W.H. Zoller, G.S. Kowalczyk and S.W. Rheingrover, 1981. Composition of
source components needed for aerosol receptor models. In Atmospheric Aerosol: Source/
Air Quality Relationships, Macias, E.S. and Hopke, P.K., editors. American Chemical
Society, Washington, D.C. pp. 51-74.
Henry, R.C., 1982. Stability analysis of receptor models that use least squares fitting. In
Receptor Models Applied to Contemporary Air Pollution Problems, Hopke, P.K. and
Dattner, S.L., editors. Air Pollution Control Association, Pittsburgh, PA. pp. 141-62.
Henry, R.C., 1992. Dealing with near collinearity in chemical mass balance receptor
models. Atmos. Environ. 26A: 933-8.
7-2
-------
Hidy, G.M. and S.K. Friedlander, 1972. The nature of the Los Angeles aerosol. In Second
International Clean Air Congress, Washington, D.C.
Hidy, G.M. and C. Venkataraman, 1996. The chemical mass balance method for estimating
atmospheric particle sources in Southern California. Chem. Eng. Comm. 151: 187-209.
Hopke, P.K., 1985. Receptor Modeling in Environmental Chemistry. John Wiley & Sons,
New York, NY.
Hopke, P. K., 1991. Receptor Modeling for Air Quality Management. Elsevier Press,
Amsterdam, The Netherlands.
Hopke, P.K. and S.L. Dattner, 1982. Receptor Models Applied to Contemporary Pollution
Problems. Air & Waste Management Association, Pittsburgh, PA.
Hougland, E.S., 1983. Chemical element balance by linear programming. 73rdAnnual
Meeting of the Air Pollution Control Association, Atlanta, GA.
Javitz, H.S. and J.G. Watson, 1986. Methods of receptor model evaluation and validation.
In Transactions, Methods for Source Apportionment: Real World Issues and
Applications, Pace, T.G, editor. Air Pollution Control Association, Pittsburgh, PA.
Javitz, H.S., J.G. Watson, J.P. Guertin and P.K. Mueller, 1988a. Results of a receptor
modeling feasibility study. JAPCA 38:661.
Javitz, H.S., J.G. Watson and N.F. Robinson, 1988b. Performance of the chemical mass
balance model with simulated local-scale aerosols. Atmos. Environ. 22: 2309-22.
Kim, B.M. and R.C. Henry, 1989. Analysis of multicollinearity indicators and influential
species for chemical mass balance receptor model. In Transactions, Receptor Models in
Air Resources Management, Watson, J.G., editor. Air & Waste Management
Association, Pittsburgh, PA. pp. 379-90.
Kim, B.M. and R.C. Henry, 1999. Diagnostics for Determining Influential Species in the
Chemical Mass Balance Receptor Model. JAWMA 49: 1449-1455.
Kneip, T.J., M.T. Kleinman and M. Eisenbud, 1973. Relative contribution of emission
sources to the total airborne particulates in New York City. In Third International Clean
Air Congress, Dusseldorf, FRG.
Larson, T.V. and R.J. Vong, 1989. Partial least squares regression methodology:
Application to source receptor modeling. In Transactions, Receptor Models in Air
Resources Management, Watson, J.G., editor. Air & Waste Management Association,
Pittsburgh, PA. pp. 391-403.
Lin, C. and J.B. Milford, 1994. Decay-adjusted chemical mass balance receptor modeling
for volatile organic compounds. Atmos. Environ. 28: 3261-76.
Lowenthal, D.H., R.C. Hanumara, K.A. Rahn and L.A. Currie, 1987. Effects of systematic
error, estimates and uncertainties in chemical mass balance apportionments: Quail Roost
II revisited. Atmos. Environ. 21:501-10.
7-
-------
Lowenthal, D.H. and K.A. Rahn, 1988a. Reproducibility of regional apportionments of
pollution aerosol in the Northeastern United States. Atmos. Environ. 22: 1829-33.
Lowenthal, D.H. and K.A. Rahn, 1988b. Tests of regional elemental tracers of pollution
aerosols. 2. Sensitivity of signatures and apportionments to variations in operating
parameters. Atmos. Environ. 22: 420-6.
Lowenthal, D.H., K.R. Wunschel and K.A. Rahn, 1988c. Tests of regional elemental tracers
of pollution aerosols. 1. Distinctness of regional signatures, stability during transport, and
empirical validation. Environ. Sci. Technol. 22:413-20.
Lowenthal, D.H., J.C. Chow, J.G. Watson, G.R. Neuroth, R.B. Robbins, B.P. Shafritz and
RJ. Countess, 1992. The effects of collinearity on the ability to determine aerosol
contributions from diesel- and gasoline-powered vehicles using the chemical mass
balance model. Atmos. Environ. 26A: 2341-51.
Lowenthal, D.H., B. Zielinska, J.C. Chow, J.G. Watson, M. Gautam, D.H. Ferguson, G.R.
Neuroth and K.D. Stevens, 1994. Characterization of heavy-duty diesel vehicle
emissions. Atmos. Environ. 28:731-44.
Miller, M.S., S.K. Friedlander and G.M. Hidy, 1972. A chemical element balance for the
Pasadena aerosol. J. Colloid Interface Sci. 39: 165-76.
Pace, T.G., 1986. Transactions, Receptor Methods for Source Apportionment: Real World
Issues and Applications. Air Pollution Control Association, Pittsburgh, PA.
Pace, T.G., 1991. Receptor modeling in the context of ambient air quality standard for
parti culate matter. In Receptor Modeling for Air Quality Management, Hopke, P.K.,
editor. Elsevier, Amsterdam, The Netherlands, pp. 255-97.
Song, X.H. and P.K. Hopke, 1996. Solving the chemical mass balance problem using an
artificial neural network. Environ. Sci. Technol. 30: 531
Stevens, R.K. and T.G. Pace, 1984. Review of the mathematical and empirical receptor
models workshop (Quail Roost II). Atmos. Environ. 18: 1499-506.
Venkataraman, C. and Friedlander, S.K., 1994. Source resolution of fine particulate
polycyclic aromatic hydrocarbons using a receptor model modified for reactivity.
JAWMA 44: 1103-8.
Vong, R.J., P. Geladi, S. Wold and K. Esbensen, 1988. Source contributions to ambient
aerosol calculated by discriminant partial least squares regression (PLS). J.
Chemometrics 2: 281-96.
Watson, J.G., 1979. Chemical element balance receptor model methodology for assessing
the sources of fine and total particulate matter in Portland, Oregon. Ph.D. Dissertation.
Oregon Graduate Center, Beaverton, OR.
Watson, J.G., 1984. Overview of receptor model principles. JAPCA 34:619-23.
Watson, J.G. and J.C. Chow, 1994. Clear sky visibility as a challenge for society. Annual
Rev. Energy Environ. 19: 241 -66.
7-4
-------
Watson, J.G., Cooper, J.A. and JJ. Huntzicker, 1984. The effective variance weighting for
least squares calculations applied to the mass balance receptor model. Atmos. Environ.
18: 1347-55.
Watson, J.G., J.C. Chow and C.V. Mathai, 1989. Receptor models in air resources
management: A summary of the APC A international specialty conference. JAPCA 39:
419-26.
Watson, J.G., N.F. Robinson, J.C. Chow, R.C. Henry, B.M. Kim, T.G. Pace, E.L. Meyer and
Q. Nguyen, 1990. The USEPA/DRI chemical mass balance receptor model, CMB 7.0.
Environ. Software 5: 38-49.
Watson, J.G., J.C. Chow, Z. Lu, E.M. Fujita, D.H. Lowenthal and D.R. Lawson, 1994a.
Chemical mass balance source apportionment of PM10 during the Southern California Air
Quality Study. AerosolSci. Technol. 21: 1-36.
Watson, J.G., J.C. Chow, F.W. Lurmann and S. Misarra, 1994b. Ammonium nitrate, nitric
acid, and ammonia equilibrium in wintertime Phoenix, Arizona. JAWMA 44: 405-12.
Williamson, H.J. and D.A. Dubose, 1983. Receptor model technical series, Volume III:
User's manual for chemical mass balance model. Report No. EPA-450/4-83-014, U.S.
Environmental Protection Agency, Research Triangle Park, NC.
Winchester, J.W. and G.D. Nifong, 1971. Water pollution in Lake Michigan by trace
elements from aerosol fallout. Water Air and Soil Pollution 1: 50-64.
7-5
-------
-------
APPENDIX A
THEORY OF THE CHEMICAL MASS BALANCE RECEPTOR MODEL
A- 1
-------
A.1 INTRODUCTION
Receptor models use the chemical and physical characteristics of gases and particles
measured at source and receptor to both identify the presence of and to quantify source
contributions to the receptor. The particle characteristics must be such that: 1) they are present
in different proportions in different source emissions; 2) these proportions remain relatively
constant for each source type; and 3) changes in these proportions between source and receptor
are negligible or can be approximated.
Common types of receptor models include: (1) chemical mass balance (CMB); (2) principal
component analysis (PCA, otherwise known as factor analysis); and (3) multiple linear
regression (MLR). Descriptions of these models and some of their variations are given in Henry
et al. (1984) and Hopke (1991). Chemical mass balance is the fundamental receptor model, with
all other approaches (including PCA and MLR) based on the use of the mass balance concept.
The chemical mass balance consists of a least squares solution to a set of linear equations
which expresses each receptor concentration of a chemical species as a linear sum of products of
source profile species and source contributions. The source profile species abundances (i.e., the
fractional amount of the species in the emissions from each source-type) and the receptor
concentrations, with appropriate uncertainty estimates, serve as input data to the CMB model.
The output consists of the amount contributed by each source-type to each chemical species. The
model calculates values for the contributions from each source and the uncertainties of those
values. Input data uncertainties are used both to weight the importance of input data values in
the solution and to calculate the uncertainties of the source contributions.
A.2 CMB Mathematics
The source contribution (Sj) present at a receptor during a sampling period of length T due
to a source] with constant emission rate Ej is
SJ = DJ * EJ
where:
T [^ -> 1
= \d u(t),ff(t),xj\dt (A-2)
A L -I
is a dispersion factor depending on wind velocity (u), atmospheric stability (a), and the location
of source j with respect to the receptor (Xj). All parameters in Equation A-2 vary with time, so
the instantaneous dispersion factor, DJ, must be integrated over time period T (Watson, 1979).
Various forms for DJ have been proposed (Pasquill, 1974; Benarie, 1976; Seinfeld and
Pandis, 1998), some including provisions for chemical reactions, removal, and specialized
topography. None are completely adequate to describe the complicated, random nature of
dispersion in the atmosphere. The advantage of receptor models is that an exact knowledge
of DJ is unnecessary.
A-2
-------
If a number of sources, J, exists and there is no interaction between their emissions to cause
mass removal, the total mass measured at the receptor, C, will be a linear sum of the
contributions from the individual sources.
Similarly, the concentration of elemental component i, Q, will be
(A-4)
where: F;j = the fraction of source contribution Sj composed of element i. The number of
chemical species (I) must be greater than or equal to the number of sources (J) for a unique
solution to these equations.
Solutions to the CMB equations consist of: (1) a tracer solution; (2) a linear programming
solution; (3) an ordinary weighted least squares solution with or without an intercept; (4) a ridge
regression weighted least squares solution with or without an intercept; and (5) an effective
variance least squares solution with or without an intercept. An estimate of the uncertainty
associated with the source contributions is an integral part of several of these solution methods.
Weighted linear least squares solutions are preferable to the tracer and linear programming
solutions because: (1) theoretically they yield the most likely solution to the CMB equations,
providing model assumptions are met; (2) they can make use of all available chemical
measurements, not just the so-called tracer species; (3) they are capable of analytically
estimating the uncertainty of the source contributions; and (4) there is, in practice, no such thing
as a "tracer." The effective variance solution developed and tested by Watson et al. (1984): (1)
provides realistic estimates of the uncertainties of the source contributions (owing to its
incorporation of both source profile and receptor data uncertainties); and (2) gives greater
influence to chemical species with smaller values for uncertainty in both the source and receptor
measurements than to species with higher values for uncertainty. The effective variance solution
is derived by minimizing the weighted sums of the squares of the differences between the
measured and calculated values of Q and F;j (Britt and Luecke, 1973; Watson etal., 1984). The
solution algorithm is an iterative procedure which calculates a new set of Sj based on the Sj
estimated from the previous iteration. It is carried out by the following steps expressed in matrix
notation. A superscript k is used to designate the value of a variable at the kth iteration.
A-3
-------
1. Set initial estimate of the source contributions equal to zero.
= o
(A-5)
2. Calculate the diagonal components of the effective variance matrix, Ve. All off-diagonal
components of this matrix are equal to zero.
F«
(A-6)
3. Calculate the k+1 value of Sj.
(A-7)
4. Test the (k+l)th iteration of the Sj against the kth iteration. If any one differs by more than
1%, then perform the next iteration. If all differ by less than 1%, then terminate the algorithm.
if (s**1 -
>f(
> 0.01, go to step 2
< 0.01, go to step 5
(A-8)
5. Assign the (k+1 )th iteration to Sj and
-------
S = (Sj ... Sj)T, a column vector with Sj as the jth component
F = an I x J matrix of Fy, the source composition matrix
-------
M—
Percent Mass = 100 ^ S, Q , where Ct is the total measured mass
-7=1
(A-ll)
R square = l-[(l-J)%2]/
(A-12)
Modified Pseudo-Inverse Matrix =
(A-13)
The Singular Value Decomposition of the weighted F matrix is given by (Henry, 1992)
-1/2
(A-14)
where U and Fare I XI and J X J orthogonal matrices, respectively, and where D is a diagonal
matrix with J nonzero and positive elements called the singular values of the decomposition.
The columns of Fare called the eigenvectors of the composition and their components are
associated with the source types.
A-6
-------
References
Benarie, M.M., 1976. Urban Air Pollution Modeling Without Computers. U.S. Environmental
Protection Agency, Research Triangle Park, NC.
Britt, H.I. and R.H. Luecke, 1973. The estimation of parameters in nonlinear, implicit models.
Technometrics 15,233-233.
Henry, R.C., C.W. Lewis, P.K. Hopke, and HJ. Williamson, 1984. Review of receptor model
fundamentals. Atmos. Environ., 18, 1507-1515.
Henry, R.C., 1992. Dealing with near collinearity in chemical mass balance receptor models.
Atmos. Environ. 26A, 933-938.
Hopke, P.K. (Ed.), 1991. Receptor Modeling for Air Quality Management. Elsevier Science
Publishing Company, Inc. New York, NY. 329pp.
Pasquill, F., 1974. Atmospheric Diffusion. Ellis Horwood, Chichester, England.
Seinfeld, J.H. and S.N. Pandis, 1998. Atmospheric Chemistry and Physics: From Air Pollution
to Climate Change. John Wiley & Sons, New York, NY.
Watson, J.G., 1979. Chemical Element Balance Receptor Model Methodology for Assessing the
Sources of Fine and Total Particulate Matter. Ph.D. Dissertation, Oregon Graduate Center,
Beaverton, OR. University Microfilms International, Ann Arbor, MI.
Watson, J.G., Cooper, J.A., and JJ. Huntzicker, 1984. The effective variance weighting for least
squares calculations applied to the mass balance receptor model. Atmos. Environ. 18, 1347-
1355.
Williamson, HJ. and D.A. Dubose, 1983. Receptor model technical series, Volume III: User's
manual for chemical mass balance model. Report No. EPA-450/4-83-014, U.S.
Environmental Protection Agency, Research Triangle Park, NC.
A-7
-------
-------
APPENDIX B
Source Code for EPA-CMB8.2: Fortran & C++
B- 1
-------
The following is an inventory and brief description of the source code used to build the main
DLL (Dynamic Link Library) for EPA-CMB8.2:
Anatomy of the DLL Source Code
Files:
CMB82.def
CMB82a.for
f. fRun
s. FitAlloc
s. CMBSVD
s. ChecksFit
s. FracEst
s. PData
s. SScont
s. PIN
CMB82b.for
s. AmbData
s. GetAmbDirec
f. Initialize
s. LoadATOT
s. MATMUL
s. MATSIN
s. OpenAmbDir
s. OpenProDir
s. OpenSumDir
f. ProRec
f. SetAmbRec
s. SETiWORK
s. SETWORK
s. SfitHead
s. SfitOut
s. SUBATOL
s. WinMessage
s. Write2ErrBuf
s. Write2OutBuf
s. XLCunc
Function1
Definition file that defines CMB82C EXPORT to CMB82.dll
Contains fRun and all routines called directly by fRun
Defines cases and assigns integer values consistent with those defined in CMB82.c; selects 6
cases:
Performs the bulk of the numerical calculations, including the singular value decomposition
(SVD). Key fit statistics are computed, along with the modified pseudo inverse (mPIN)
matrix.
Checks the fit for sources to eliminate.
Determines the number of estimable sources.
Finishes computations and writes them to the Results screen.
Computes species contributions to each sources SCE (source contribution estimate).
Prepares mPIN matrix for output.
Supplemental; contains all routines not directly called by fRun.
Declares variables, dimensions arrays & defines Fortran unit numbers for direct-access
(binary) scratch files AmbDir.dat; ProDir.dat, SumDir.dat.
Opens ambient data file and specifies record lengths via RECL stmt.
Opens source profile data file and specifies record length via RECL stmt.
Opens summary data file and specifies record lengths via RECL stmt.
Creates header for the source fit display.
Creates DB (spreadsheet-type) output file.
References MessageBoxC_( message ) in CMB82.C.
Computes source uncertainties.
CMB82c.c
CMB82.inc
WKShead.h
Visual C++ file; critical interface for Delphi; the main program file for EPA-CMB8.2
Include file for variables used globally in EPA-CMB8.2
Header file referenced by CMB82.C for (Lotus) WKS output
Historically, the CMB8.2 code has not been well documented. For this reason, even after inspection, it's not always
clear what certain routines are doing. As experience is gained, the functionality notes should become more complete.
B-2
-------
Organization of the source code (Fortran & C).
The 4 Fortran files associated with CMB8.0 (CMB85_A.for, CMB85_B.for, CMB85_C.for & CMBSSF.for) have been reorganized and merged
into 2 files: the main CMB82a.for & the supplemental CMB82b.for. The Fortran portion of the source code consists entirely of 28 modules (24
subroutines and 4 functions). In the Fortran code, there is no "main program unit" (though function fRun in CMB82a.for initially controls calls to
all Fortran routines); the C-code file CMB82.C (not shown) serves as the main program. In an effort to purge legacy code, 28 routines that were
unnecessary in CMB8.0 were removed, as were many variables associated with the remaining routines. All routines called by fRun are in
CMB82a.for; all others are in CMB82b.for. The Fortran routines are listed here, along with other routines they may call subsequently:
CMB82a.for (8 routines, in calling sequence)
CMB82b.for (20 routines, alphabetically)
f. fRun
calls:
s. FitAlloc H>
s. CMBSVD2 ^>
s. ChecksFit =&>
s. FracEst
s. PData3 =£>
s. SScont =&>
s. PIN ^>
SUBATOL, AmbData, SETiWORK, SETWORK
WinMessage, ProRec, MATMUL, MATSIN, Write2ErrBuf,
XLCunc, OpenSumDir
SETiWORK, SETWORK
OpenSumDir, Write2OutBuf, AmbData, GetAmbDirec
Write2OutBuf, AmbData, ProRec
AmbData, Write2OutBuf
s. AmbData
s. GetAmbDirec
f. Initialize4
s. LoadATOT
s. MATMUL
s. MATSIN5
s. OpenAmbDir
s. OpenProDir
s. OpenSumDir
f. ProRec
f. SetAmbRec3
s. SETiWORK
s. SETWORK
s. SFitHead3
s. SFitOut3
s. SUBATOL
s. WinMessage
s. Write2ErrBuf
s. Write2OutBuf
s. XLCunc
^ OpenAmbDir
^> SUBATOL
^> LoadATOT
^> OpenProDir
^> OpenAmbDir, SUBATOL
=£> ProRec
^> OpenSumDir, ProRec
^> MessageBoxC_( message )6
=&> WinMessage
Calls LAPACK's SGESVD
3Calls LAPACK's SGESVD (only if numinestsource > 0)
4Called by CMB82.C
5Calls LAPACK's SPOTRF & SPOTPJ
6This string is in CMB82.C
B-3
-------
Figure B-l shows conceptually the general direction and flow across the main program units.
The Delphi system that controls the GUI interacts with CMB82.C (via CMB82.dll). CMB82.C is
critical for the Delphi interface, and serves as the main program unit that controls actions within
the computation machinery in the Fortran code. The occurrence of specific cases in CMB82.C
triggers fRun to make specific calls in CMB82a.for. While the intricacies of the entire logic
flow aren't shown, you should get a macro sense of how program flow is directed. For
orientation, the appearance of certain screen views in EPA-CMB8.2 are indicated.
Delphi
CMB82.C
CMB82a.for
13 Main Case Modules:
case DO_CMB8init
T
case DO_Send_Codes
T
case DO_SetDisplayDecPl
(Main Form of UI appears)
case DO_SetBatchFlag
case DO_SetDichotFlag
(All routines called by fRun are in CMB82a.for)
case DO CMB
case DO SendFracEst
case DO_SendInfo
case DO_SendSNGVAL
case DO PData
case DO SScont
case DO mPIN
fRun ^> s. FitAlloc
fRun ^> s. CMBSVD
s. ChecksFit
fRun ^> s. FracEst
fRun ^> s. PData
fRun ^> s. SScont
fRun ^> s. PIN
case DO_SFitOut
(Results screen appears, populated with calculated values)
case DO CMB Terminate
Figure B-l. Conceptual flow diagram for program sequence in EPA-CMB8.2.
B-4
-------
Building the DLL for EPA-CMB8.2
The dll is compiled from code written in Fortran and C++ languages. This dll is resident in
the installation subdirectory for EPA-CMB8.2 and is used by the Delphi client (executable) at
run time (Appendix D). The 32 bit dll, CMB82.dll, is created with Compaq® Visual Fortran and
Microsoft Visual C/C++ using the following instructions:
Assumption:
• Visual Fortran 6.6B7 is installed (you may have to download the 6.5 ^> 6.6 and 6.6 ^> 6.6B or
6.6C updates from Compaq's website: http://compaq.com/fortran/visual/updates.htmn
• Visual C/C++ v.6 (http://msdn.microsoft.com/visualc/) is installed.
• Intel® Fortran Compiler for Windows®
(Available @ http://www.intel.com/software/products/compilers/fwin/index.htm) Intel® 7.1
(15 September 2004) is advisable; its installation executable is W_FC_PC_7.1.027.exe.8
Directions:
• Launch Visual Fortran Developer Studio (or Visual C++; they're quite similar for this
purpose)
• Create a new Fortran Dynamic Link Library (File/New), calling the project CMB82 and
locating it somewhere like C:\CMB\Develop\CMB82. This will create a folder with this
name.9
• In the one and only step in the App Wizard, select the option which reads "A DLL
application with exported symbols" (not "An empty DLL application"), click Finish, and then
click OK in the dialog that appears (which echoes the pathway to the project folder).
• Using Windows Explorer®, copy the CMB82 source code to the folder you just created (e.g.,
C :\CMB\Develop\CMB 82).
The 8 source code files are:
CMB82.def CMB82.inc
CMB82a.for WKShead.h
CMB82b.for LAPACK.lib10
CMB82.C BLAS.lib10
It may also be assumed that Visual Studio service pack 4 or 5 is installed. Note that Microsoft Visual C++ v6.0 has
been upgraded to Visual C++.NET. However, please also note that Visual C++.NET reportedly does not integrate with
Compaq® Visual Fortran.
Q
Once this compiler is installed, you will enable it within Visual Fortran by going to Tools Customize and click Add-
ins and Macro Files' check the Intel® Fortran Compiler Build Tool, and Close the dialog.
Note that, while later versions of the Intel® compiler are available, these require .NET versions of MS Visual Studio and MS
Visual C++.
C)
These files will also be added:
CMB82.dsp (project file)
CMB82.dsw (workspace file)
CMB82.f90 (template for Fortran subroutine)
CMB82.opt (project options file; appears after Workspace is closed)
CMB82.ncb (browse information file built whenever a project is loaded; the tree view in the ClassView tab uses this file to populate itself)
CMB82.plg (project log file; added after Build - F7)
See Appendix C for information on LAPACK and for instructions on building the LAPACK static library.
B-5
-------
1) switch back to Developer Studio
2) make sure the Workspace pane is visible, selecting Workspace from the View menu
3) click on FileView tab (not the ClassView tab)
4) expand the tree view until the one and only Fortran file (CMB82.f90) is visible under the
folder Source Files
• select this file in the tree and delete it
• delete the Resource Files folder (it's not needed)
5) right-click on the project (the node which reads "CMB82 files") and select Add Files to
Project
• change the Files of type filter to All Files (*.*)
• select CMB82.f90 and delete it
• select all the source files listed above and click OK11
6) right-click on the project and select .Settings.
7) select All Configurations in the Settings For: combo box
• click on the Fortran tab
• select Compatibility in the Category.: combo box
• check I/O Format and List Directed I/O Spacing in the PowerStation® 4.0
Compatibility Options listbox (Libraries should already be checked)
• select Libraries in the Category.: combo box
•select Multi-threaded in the Use run-time library: combo box
8) select Win32 Debug in the Settings For: combo box
•select Debug Multi-threaded in the Use run-time library: combo box
• click on the C++ tab
• select General from the Category.: combo box
• select Program Database in the Debug info: combo box
• click OK to exit Project Settings12
9) click OK to exit Project Settings
10) press F7 to build13
The compiling and linking functions will proceed and details will be echoed to the output
window at the bottom of the screen. The volume of the product DLL is minimized since CVF
will only link the parts of the LAPACK library needed by EPA-CMB8.2; the debug version of
the DLL should be of order 900KB; the release version is of order 700KB. Since the default
configuration is Win32 Debug, the file CMB82.dll will be stored in the Debug folder within the
project folder of similar name. To confirm/set the DLL destination,
click the Link tab under Project]Settings, select General in the Category, and adjust the path &
name in the Output file name field.
Take care to select only the 8 source files listed.
When the Workspace is closed, *.opt will be added to the CMB82 project folder.
To hold/use these settings for future builds (in Visual Fortran or Visual C++):
• select Options from the Tool menu
• click on the right scroll arrow in the upper right of the dialog until the Workspace tab is visible
• click on the Workspace tab
• check both of the "Reload ..." checkboxes
•File
• Open Workspace
• click on the project folder (e.g., CMB82)
• F7 (The output window at the bottom will echo linking details for any source files that were change since the last build.)
Completion of Build (F7) will add *.plg to the project folder.
B-6
-------
APPENDIX C
The Linear Algebra Library for EPA-CMB8.2: LAPACK
C-l
-------
In order to solve the effective variance, least-squares regression that CMB employs for an
apportionment, notably the Singular Value Decomposition (SVD), the Fortran code accesses a
library of linear algebra routines. CMB8.0 relies on the linear algebra library UNPACK.14
LINPACK was designed for supercomputers in use in the 1970s and early 1980s and is a
collection of Fortran subroutines that analyze and solve linear equations and linear least-squares
problems. The package solves linear systems whose matrices are general, banded, symmetric
indefinite, symmetric positive definite, triangular, and tridiagonal square. In addition, the package
computes the QR and SVDs of rectangular matrices and applies them to least-squares problems.
LINPACK uses column-oriented algorithms to increase efficiency by preserving locality of
reference.
This library was up updated to LAPACK. LAPACK was developed by a consortium of
university and government laboratories, and is the industry-standard subprogram package offering
an extensive set of linear system and eigenproblem solvers. Reference:
Anderson, E, Z. Bai, C. Bischof, S. Blackford, J. Demmel, J. Dongarra, J. Du Croz, A.
Greenbaum, S. Hammarling, A. McKenney and D. Sorensen, 1999. LAPACK Users' Guide,
Third Edition. Society for Industrial and Applied Mathematics (SLAM), Philadelphia, PA;
ISBN 0-89871 -447-8 (paperback) http://www.siam.org/
LAPACK is written in Fortran77 and provides routines for solving systems of simultaneous
linear equations, least-squares solutions of linear systems of equations, eigenvalue problems, and
singular value problems. The associated matrix factorizations (LU, Cholesky, QR, SVD, Schur,
generalized Schur) are also provided, as are related computations such as reordering of the Schur
factorizations and estimating condition numbers. In all areas, similar functionality is provided for
real and complex matrices, in both single and double precision.
The original goal of the LAPACK project was to make the widely used LINPACK libraries
run efficiently on shared-memory vector and parallel processors. On these machines, LINPACK
is inefficient because its memory access patterns disregard the multi-layered memory hierarchies
of the machines, thereby spending too much time moving data instead of doing useful floating-
point operations. LAPACK addresses this problem by reorganizing the algorithms to use block
matrix operations, such as matrix multiplication, in the innermost loops. These block operations
can be optimized for each architecture to account for the memory hierarchy (shared memory), and
so provide a transportable way to achieve high efficiency on diverse modern machines. We use
the term "transportable" instead of "portable" because, for fastest possible performance, LAPACK
requires that highly optimized block matrix operations be already implemented on each machine.
LAPACK routines are written so that as much as possible of the computation is performed by calls
to the Basic Linear Algebra Subprograms (BLAS)15. While LINPACK is based on the vector
operation kernels of the Level 1 BLAS, LAPACK was designed at the outset to exploit the Level 3
BLAS - a set of specifications for Fortran subprograms that do various types of matrix
multiplication and the solution of triangular systems with multiple right-hand sides. Because of
the coarse granularity of the Level 3 BLAS operations, their use promotes high efficiency on
many high-performance computers, particularly if specially coded implementations are provided
14Developed by Jack Dongarra, Jim Bunch, Cleve Moler and Pete Stewart; 1 Feb 1984. Available from NETLIB:
http: //www. netlib. org/linpack/
The BLAS library includes the industry-standard Basic Linear Algebra Subprograms for Level 1 (vector-vector),
Level 2 (matrix-vector), and Level 3 (matrix-matrix) applications.
C-2
-------
by the manufacturer. LAPACK has been thoroughly tested on many types of computers, and a list
of known problems, bugs and compiler errors for LAPACK is maintained on NETLIB:
http://www.netlib.org/lapack/release_notes.html
The LAPACK SVD (s. SGESVD) is better numerically than its counterpart in LINPACK (s.
SSVDC). For this reason, as well as for certain increases in efficiency and "modularity" that lends
itself to accommodating LAPACK updates, EPA-CMB8.2 has been modified to use the LAPACK
library. As an additional consequence of this upgrade, LINPK.for (which contained a series of
LINPACK routines used by CMB8.0) has been removed as a source file; it is replaced by a pair of
companion static libraries: LAPACK.lib and BLAS.lib (Basic Linear Algebra Subprograms).
These libraries can be reconstructed for building EPA-CMB8.2's main DLL (Appendix B) at any
time that a new LAPACK becomes available. Building instructions are provided in this appendix.
The LAPACK routines are freely available and are copyrighted. The entire suite of
(single-precision) routines for LAPACK v3.0 (latest version), including its requisite Level 1, 2 &
3 BLAS, was updated 31 May 2000 and is listed and described here (download lapack.pc.df.zip)
http://www.netlib.org/lapack/single/index.html
At run time, EPA-CMB8.2 directly calls these 3 Fortran routines in LAPACK (see Appendix B):
SGESVD16 (called by s. CMBSVD & s. PData in CMB82a.for)
SPOTRF17 (called by s. MATSIN in CMB82b.for)
SPOTRI18
These 3 routines subsequently call (or at least reference) the following 22 Fortran routines (listed
alphabetically) - some in LAPACK, and others in BLAS (as indicated):
LAPACK BLAS
ILAENV
SBDSQR
SGEBRD
SGELQF
SGEQRF
SLACPY
SLAMCH
SLANGE
SLASCL
SLASET
SLAUUM
SORGBR
SORGLQ
SORGQR
SORMBR
SPOTF2
STRTRI
LSAME
SGEMM
SSYRK
STRSM
XERBLA
Computes the singular value decomposition (SVD) of a real M by N matrix A, optionally computing the left and/or
right singular vectors. The SVD is written A = U * S * VT, where S is an M by N matrix which is zero except for min(m,n)
diagonal elements, U is an M by N orthogonal / unitary matrix, and V is an N by N orthogonal / unitary matrix. The diagonal
elements of S are the singular values of A; they are real and non-negative, and are returned in descending order. The first
min(m,n) columns of U and V are the left and right singular vectors of A.
Computes the Cholesky factorization of a real, positive definite matrix A. The factorization has the form A = UH *
U, where U is an upper triangular matrix.
1 8
Computes the inverse of the matrix A computed by SPOTRF.
C-3
-------
Building the static LAPACK/BLAS Libraries for EPA-CMB8.2
To build the static LAPACK and companion BLAS libraries for EPA-CMB8.2 (see Appendix
B), first obtain the 1291 LAPACK routines (12.2MB), and the 141 BLAS routines (966KB) from
the source listed above.
Assumption:
• Visual Fortran 6.6B (or later)19 is installed (you may have to download the 6.5 ^> 6.6 and 6.6
^> 6.6B or 6.6C updates from Compaq's website:
http://compaq.com/fortran/visual/updates.htmn
• Intel® Fortran Compiler for Windows® (see footnote 8 in Appendix B)
Directions:
• Launch Visual Fortran Developer Studio (or Visual C++; they're quite similar for this purpose)
• Create a new Fortran Static Library (File/New), calling the Workspace (and 1st project)
LAPACK and locating it somewhere like C:\CMB\Develop\CMB82. This will create a folder
with this name.20
Build the LAPACK library
1) Using Windows Explorer®, copy the LAPACK source code (*.f files) to the folder you just
created (e.g., C:\CMB\Develop\LAPACK).
2) switch back to Developer Studio
3) make sure the Workspace pane is visible, selecting Workspace from the View menu
4) click on FileView tab (not the ClassView tab)
5) expand the tree view delete the Header Files folder (it's not needed)
6) right-click on the project (the node which reads "LAPACK files") and select Add Files to
Project
• change the Files of type: filter to Fortran Files
• select all the source files listed in the Look in: window and click OK21
19
It may also be assumed that Visual Studio service pack 4 or 5 is installed. Note that Microsoft Visual C++ v6.0 has
been upgraded to Visual C++.NET. However, please also note that Visual C++.NET reportedly does not integrate with Compaq
Visual Fortran. You will likely have to use the Visual C++.NET environment separately to compile the C code, and then use the
newer linker to link the application.
Eventually, all these files will also be added by CVF:
LAPACK.dsp (project file)
LAPACK.dsw (workspace file)
LAPACK.ncb (browse information file built whenever a project is loaded; the tree view in the ClassView tab uses this file
to populate itself)
LAPACK.plg (project log file)
21This will load the 1291 source files into the LAPACK project.
C-4
-------
Build the companion BLAS library
• right-click on the workspace (the node which reads Workspace 'LAPACK':) and select Add
New Project to Workspace
• select Fortran Static Library, enter "BLAS" for Project name:, and click OK twice.
• expand the tree view delete the Header Files folder (it's not needed)
• Using Windows Explorer®, copy the BLAS source code (*.f files) to the folder you just
created (e.g., C:\CMB\Develop\CMB82\BLAS).
• switch back to Developer Studio
• right click on the BLAS project and select Add Files to Project. Select the \BLAS
folder and change the Files of type: filter to Fortran Files
• select all the source files listed in the Look in: window and click OK22
7) For both static library projects - LAPACK & BLAS - the Library Settings must
generally be set to match those for the main project CCMB82' - see Appendix B). For
both LAPACK & BLAS:
• right-click on the project and select Settings.
• select All Configurations in the Settings For: combo box
• click on the Fortran tab
• select Libraries in the Category: combo box
•select Multi-threaded in the Use run-time library: combo box
• click OK to exit Project Settings23
8) Build the libraries for both - LAPACK & BLAS - projects.
• click Project\Set Active Project «& LAPACK; press F7 to build24
• click Project\Set Active Project "^ BLAS; press F7 to build
As the compiling and linking functions will proceed for each build command, details will be
echoed to the output window at the bottom of the screen. Since the default configuration is Win32
Debug, the file(s) name.lib will be stored in the Debug folder within the respective project folder
of similar name (this can be changed by clicking on the Library tab, then entering an alternate
path/file name in the Output file name: field).
This will load the 141 source files into the BLAS project.
When the Workspace is closed, *.opt will be added to the LAPACK project folder.
TSTote that compilation of the 1291 LAPACK routines will take some time. Occasional informational messages of the
following type may be disregarded:
Info: This directive is not supported in this platform.
CDEC$ NOVECTOR
Completion of Build (F7) will add *.plg to the project folder.
C-5
-------
-------
APPENDIX D
The User Interface for EPA-CMB8.2: Delphi
D- 1
-------
The following lists and describes the source code, on which the Delphi1 client that serves as the
User Interface (UI) for EPA-CMB8.2, is based:
Anatomy of the Delphi Source Code2
EPACMB82.res The project resource file. This is a binary file which normally contains the
version info resource (if required) and the application's main icon. Changes
to the version info and the application icon in Delphi's Project Options
dialog will result in changes to this file.
EPACMB82.dpr The Delphi project file. This is actually source code for the application: the
project file moniker is a misnomer.
EPACMB82.cfg The project configuration file. Stores the project configuration settings,
primarily compiler and linker settings.
EPACMB82.dof The Delphi options file. Stores the project option settings, such as
additional compiler and linker settings, directories, conditional directives,
and version information.
StartupFrm.pas / StartupFrm.frm
AboutFrm.pas / AboutFrm.dfm
The source code and form file for displaying the
startup screen, the first form shown when the
application is run.
The source code and form file for displaying the
banner 'About EPA-CMB8.2'.
BLWarningFrm.pas / BLWarningFrm.frm The source code and form file for displaying the
Britt-Luecke warning message.
WaitFrm.pas / WaitFrm.dfm
MainFrm.pas / MainFrm.dfm
The source code and form file for displaying the
temporary 'Please wait' message, shown when
loading input data.
The source code and form file for displaying the
main CMB model screen (Select Input Files;
Options; Sample; Species; Sources; Results).
About half of the application's logic can be
found in the source code for the main form.
Borland Software Corporation
D-3.
The bulk of the application's code is in MainFrm.pas and CMBInternals.pas. Delphi file types are described on p.
D-2
-------
PrintOptionsFrm.pas / PrintOptionsFrm.dfm The source code and form file for displaying the
Print window enabled on the Results screen
when the results are in the buffer. Most of the
logic for printing reports or saving them to a text
file can be found in the source code for the Print
Options form.
SaveResultsFrm.pas / SaveResultsFrm.dfm
CMBInternals.pas
The source code and form file for displaying the
Save Results screen. This form allows saving of
the current record or all included records
displayed in the Main Report buffer. Most of the
logic for saving result data to a text file can be
found in the source code for the Save Results
form.
Controls all Delphi setup and interaction with
CMB82.C in calls to the C++/Fortran DLL. Also
contains a number of math routines, e.g. Best Fit,
and the code which refines the output report
generated by s. PData in CMB82a.for.
D-
-------
General notes for Delphi file types:
*.pas Delphi source code file, called a unit (source code module), which is written in Pascal.
Each application may contain several unit source files, which typically contain most of
the code for the application.
*.dfm Delphi form file. A project-related file, *.dfm contains the description of the properties
of the form and the components it owns. This description may be present in text form (a
format very suitable for version control) or in a compressed binary format. Each form
file represents a single form, which usually corresponds to a window or dialog box in an
application. The Delphi IDE (Integrated Development Environment) allows you to view
and edit aliform files as text, and to save form files as either text or binary. Each
application has at least one form and, while not all unit files (e.g., a math library of
functions and type information) have a corresponding/0?™ file, each form has a
corresponding unit file (by default, having the same name as Reform file). In the Delphi
IDE, if a unit defines a form (i.e., uses a visual component such as TForm), a form file is
automatically created. N.B.: If a unit file (e.g., OptionsDlg.pas) defines a form and
therefore has a corresponding form file (e.g., OptionsDlg.dfm), both must be resident in
the working directory in order to open either in the IDE or Delphi will complain.
*.dpr Delphi project file. Another source file, *.dpr is explicitly created by association with a
project in the Delphi IDE. The project file (which corresponds to the "main" program
file in traditional Pascal) organizes the unit files into an application. The Delphi IDE
automatically creates and maintains a unique project file for each application. The
project file may be created by selecting a new project from the File menu in Delphi, e.g.,
File New Application. When the gallery dialog appears, double click on the icon
entitled 'Application'. Delphi will then create a new project file for you. It will create a
file called unitl.pas (and unitl.dfm) as a default because Delphi assumes that you want to
have a main form in your application.
*.dof Delphi options file. Contains the current settings for project options, such as compiler
and linker settings, directories, conditional directives, command-line parameters and
version information. These are set using the Project Options dialog box (Project
Options), and Delphi saves them in text form for easy maintenance, version control, and
sharing. Each project has an associated options file with the same name as the project
(*.dpi) file.
*.res Delphi resource file. A project-related file, it contains the version information resource
(if required) and the application's main icon. This file may also contain other resources
used within the application but these are preserved as is. Do not delete this file if your
application contains any references to it.
*.cfg Project configuration file. This file stores project configuration settings and has the same
name as the project file.
D-4
-------
Construction of the Delphi executable for EPA-CMB8.2
The Delphi client (executable) calls CMB82.dll (Appendix B) at runtime. This client controls the
user interface for EPA-CMB8.2 and is created with Delphi 7, a product of Borland Software
Corporation (http://www.borland.com/delphi/). using the following instructions:.
Assumptions:
• Delphi 6 or 7 is installed.
Ordinarily, the compiler should be optimized for a build:
Project | Options Compiler Code generation: check Optimization
Directions:
• Ensure that all of the following 19 Delphi source files are stored in a single working directory
(folder):
EPACMB8.2.res
EPACMB82.dpr
EPACMB82.cfg
EPACMB8.2.dof
StartUpFrm.pas StartUpFrm.dfm
AboutFrm.pas AboutFrm.dfm
BLWarningFrm.pas BLWarningFrm.dfm
WaitFrm.pas WaitFrm.dfm
MainFrm.pas MainFrm.dfm
PrintOptionsFrm.pas PrintOptions.Frm.dfm
SaveResultsFrm.pas SaveResultsFrm.dfm
CMB82Internals.pas3
Launch Delphi.
Select Open Project from the File menu.
Browse to the folder that contains the source files.
Select the file EPACMB82.dpr and click OK.
Select Build EPACMB82 from the Project menu.
The built executable should now appear in the working directory (folder), along with the
following compiled unit files (*.dcu):
3N.B.: The original call in CMB8.0 to CMB8wn32.dll from:
UNIT1.PAS(157): argl 1, arg!2 : Pointer ): Longlnt; stdcall; External 'cmb8wn32.dll' name '_CMB80C@52'
was changed in EPA-CMB8.2 to a call to CMB82.dll from:
CMBInternals.pas(260): arg!2: Pointer): Integer; stdcall; external 'CMB82.dll' name 'CMB82c'
D-5
-------
WaitFrm.dcu
StartUpFrm.dcu
SaveResultsFrm. dcu
PrintOptions.Frm.dcu
MainFrm.dcu
BLWarningFrm. dcu
AboutFrm.dcu
CMB Internals.dcu
These intermediate *.dcu files , which are binary images for each unit file, are necessary to create
the final executable file (EPACMB82.exe). However, after the .exe has been created, you may
delete them. The next time you compile the application, the Delphi IDE will regenerate them
automatically.
D-6
-------
APPENDIX E
Warning and Error Messages from EPA-CMB8.2
E- 1
-------
There are many different warning and error messages that EPA-CMB8.2 may produce to indicate that an input file is not named
properly or does not follow the proper data format, that a model run is not set up correctly, or that a run did not produce satisfactory
results. Most of the error messages that involve user intervention include suggestions for appropriate corrective actions. See Appendix
F for a more detailed discussion of EPA-CMB8.2's behavior that may result in certain error conditions. A [] indicates a numeric value
or alphameric string.
CMBInternals.pas1
Error reading source selection filename. Please refer to the manual for file naming convention.
Error reading species selection filename. Please refer to the manual for file naming convention.
Error reading ambient data selection filename. Please refer to the manual for file naming convention.
Error reading ambient sample data filename. Please refer to the manual for file naming convention.
The ambient sample data file does not exist.
Error reading source profiles data filename. Please refer to the manual for file naming convention.
The source profiles data file does not exist.
* * * * NO CONVERGENCE AFTER [] ITERATIONS * * * *
AKT*VEFFIN*AK MATRIX NEEDS IMPROVEMENT. CHANGE FITTING SOURCES OR FITTING SPECIES.2
The number of fitting sources [] must be positive and <= the number of fitting species [].
Memory allocation error. Try shutting down other applications or increasing RAM.
File Open Error. Check that the file exists or try rebooting.
Species Set Error. Please see documentation regarding input file construction.
Sources Set Error. Please see documentation regarding input file construction.
The number of fitting sources must be positive and <= number of fitting species.
These messages originate from the Delphi client.
If a model run does not converge, the results of that run will not be reliable. This does not have any negative impact on other successful runs within the same EPA-
CMB8.2 session, so the other results may still be used.
E-2
-------
Positive uncertainties are required for all fitting species and sources.
This error should not occur. Contact Tom Coulter - Coulter.Tom@epa.gov
The fitting source profiles matrix (Afit) contained a column of all zeros. Check the source profiles data file or change the fitting sources.
The choices for fitting species and sources have produced a near-singular matrix. There is most likely collinearity between two or more of the
fitting sources.
The fitting algorithm failed to converge. There is most likely collinearity between two or more of the fitting sources.
File Read Error. Please see documentation regarding input file construction.
Unknown error.
File Close Error. If your disk is full, try freeing up some space, or try rebooting.
File Write Error. If your disk is full, try freeing up some space, or try rebooting.
Ambient Data Error. Please see documentation regarding input file construction.
Source Profiles Error. Please see documentation regarding input file construction.
Error reading species selection file. Please see documentation regarding input file construction.
Error reading source selection file. Please see documentation regarding input file construction.
Error reading ambient data selection file. Please see documentation regarding input file construction.
Error reading ambient data file. Please see documentation regarding input file construction.
Error reading source profiles file. Please see documentation regarding input file construction.
Memory free error. Try shutting down other applications or increasing RAM.
MainFrm.pas1
The ambient sample data file is empty!
The ambient sample data file is open!
Got an empty line of data from the ambient sample data file!
The number of data values for this sample is wrong!
Didn"t get any data from the ambient sample data file!
The source profiles data file is empty!
The source profiles data file is open!
The source profiles selection file [] does not exist. Please edit the Control File and try again.
The species selection file [] does not exist. Please edit the Control File and try again.
The ambient data selection file [] does not exist. Please edit the Control File and try again.
The ambient sample data file [] does not exist. Please edit the Control File and try again.
The source profiles data file [] does not exist. Please edit the Control File and try again.
Exactly *ONE* sample must be selected for the Best Fit option.
The number of fitting species selected must be greater than or equal to the number of fitting sources selected.
E-3
-------
The number of size ranges in the selected samples exceeds 4 for the same site and sampling period.
CMB82.C3
Can't open CMB8.2 ambient data file: []
(=> The file doesn't exist, is open and locked (located?) elsewhere, or is read-only.)
Incorrect number of tokens in header line of file []. Should be an odd number > 5
(=> The header line in the ambient sample data file is invalid.).
Total number of species = []
(Total #spp <= 0)
Can't open CMB8.2 source profiles data file: []
(=> The file doesn't exist, is open and locked (located?) elsewhere, or is read-only.)
Can't create CMB8.2 direct access temporary file
(=> The file doesn't exist, is open and locked (located?) elsewhere, is read-only, or the user's machine is out of free disk space)
Memory allocation error
(=> too many tokens)
Number of tokens in line [] of file [] not equal to number of tokens in header line4
Total number of profiles data records = []
(Total #src profiles <=0)
Memory allocation error for [] from line []
Pointer [] not defined in ptinfo, line = []
Unknown pointer [] from line []
Memory overrun for pointer [] from line []; size = []
Unknown action [] from line []
Not able to open output file: []
These messages originate from the C / Fortran DLL.
The file listed may be either an ambient data (AD*.*) file or source profile (PR*.*) file; see Appendix F.
E-4
-------
CMB82a.for2
(all in s.CMBSVD)
NUMBER OF FITTING SOURCES = [] > NUMBER OF FITTING SPECIES = []
FITTING SPECIES [] HAS ZERO AMBIENT DATA UNCERTAINTY.
Replace uncertainty with a positive detection limit or remove the species from the fitting species list..
No record for source profile code [] and size []
Column (source) Number [] of Afit = O5
AKT*VeffIn*AK matrix needs improvement. Change fitting sources or fitting species.6
NO CONVERGENCE AFTER [1 ITERATIONS.
This error condition is caused by a column of the fitting matrix containing all zeros. The column corresponds to one of the fitting sources. A zero column for a fitting
source means that all of the fitting species' fractions are zero for that source. The fitting algorithm can't handle a zero column in the fitting matrix (it's the matrix equivalent of
division by zero).
The AKT*Veffln*AK matrix is used internally by the fitting algorithm. If one of its columns consists of elements close to zero, the algorithm has problems because of
computer round-off error, in which case MATSIN returns info = 1. This condition is usually caused by collinearity among fitting sources. Changing fitting sources and/or fitting
species is intended to remove the collinearity.
E-5
-------
-------
APPENDIX F
Data Input Issues for EPA-CMB8.2
F- 1
-------
In Section 3.6.1 we mentioned that EPA-CMB8.2 detects "missing values" for species
concentrations by their flags (-99.) in the input files, and that when these flags are encountered,
the species is "automatically removed from the calculation" and the selection flag in the Main
Report is changed from an asterisk (*) to 'M'. Some discussion of EPA-CMB8.2 interpretation
and treatment of data values encountered in either ambient sample or source profiles is in order.
The first thing to note is that EPA-CMB8.2's behavior when it encounters certain kinds of data
values in these two input sources is not parallel, and this is so for reasons that will be explained
below.
Ambient (Sample) Data.
We stated in Section 4.2.2 that a species for which the concentration is missing (i.e., invalid)
cannot be used as a fitting species for that sample and that concentration value must be substituted
by -99. in the input data file. We also stated that the species will automatically be removed from
the calculation. But what does that mean? By "removed from the calculation", we mean that they
are removed as fitting species but concentrations are calculated in the total list of source
contributions so long as entries appear in the source profiles. Recall in Section 3.2 and 4.3.2 we
mentioned floating species. These are species that participate in the CMB calculation
(apportionment) but are not used as fitting species. When a species' mass is missing in a
particular ambient sample, CMB will attempt to apportion mass for that species as long as it is
designated as a fitting species (Species Selection screen, Section 3.4) and EPA-CMB8.2 "finds" it
in the source profiles. In this case, the species is treated as a floating species, and mass is
apportioned. While in the Main Report there are appropriate indicators that the species is missing,
a calculated mass appears in the Main Report to tell us what we should expect in the ambient
sample (which allows us to estimate a value for the missing data), but more importantly it tells us
if the apportionment is reasonable.
If, however, a legitimate (^ 0) value for the species concentration is entered, then the
associated value for uncertainty MUST be greater than zero or else the following error message
will be returned :
EPA CMB8.2 Fortran Section
Fitting Species S4IC
has ZERO Ambient Data Uncertainty!
Replace uncertainty with a positive detection limit,
or remove the species from the fitting species list.
OK
and execution will be halted during the first iteration attempt. To implement the effective variance
least squares solution, the ambient uncertainty can never be zero because the first step in the
iteration assumes that all source contributions are zero. That is, in the first iteration, there is no
contribution to the effective variance from any of the source uncertainties (this is equivalent to the
ordinary weighted least squares method described by Friedlander, 1973). Thus, EPA-CMB8.2
starts out by dividing by zero if there is no ambient data uncertainty for a fitting species and the
program terminates.
F-2
-------
Source Profile Data.
For source profile input data files, we stated in Section 4.2.3 that missing mass fraction values
must be substituted by -99. and that the associated species will automatically be removed from
the calculation. By this we mean that they are removed as fitting species and no concentrations
are calculated. If a mass fraction is substituted by -99. in the source profile data file, when EPA-
CMB8.2 is run the value appearing for that species in the Main Report under "CALCULATED"
will be a large negative value (if the value for number of decimal places displayed is set low
enough; Section 3.2). The values listed for the diagnostics CALCULATED/MEASURED and
RESIDUAL/UNCERTAINTY will, of course, be meaningless. The value for the species in
question listed in the Contribution by Species for the affected source will also be meaningless.
As we said in Section 4.2.3, while uncertainty values for species in source profiles are
allowed to be <, 0.0, some effort should be made to supply values > 0.0. However, for appropriate
reasons, the same kinds of behavior (error messages, etc.) do not occur with equivalent input data
conditions as described above for the ambient data files. For example, if for a particular species in
a source profile record the mass fraction is valid (i.e., £ 0.0), EPA-CMB8.2 allows an associated
uncertainty value to be < 0.0. No error message will appear and EPA-CMB8.2 will execute
normally.
Recall from Appendix A the effective variance weighting to which all of the ambient
concentrations are normalized. In Equation A-6, the effective variance is the concentration
uncertainty squared plus the sum of source contribution squared times the corresponding
elemental mass fraction uncertainty squared:
• °F.
H * IJ
Consider a situation for a woodsmoke profile in which the uncertainty for K (potassium) was
input as £ 0.0 for a mass fraction > 0.0.
If the uncertainty value for K < 0.0 (e.g., -99.), this uncertainty in the woodsmoke profile may
overwhelm all of the other profile uncertainties (which are less than one because profiles are
normalized to unity). This would effectively remove K as a fitting species, because of division by
the effective variance resulting in a very low weight for this species. This will also propagate as a
very high uncertainty on the source contribution estimate and on the calculated value for K.
If the uncertainty for K = 0.0, there are other components of the effective variance from the K
uncertainty in the ambient sample as well as from the K uncertainties in the other profiles, so there
is not division by zero just because K uncertainty in the woodsmoke profile is zero.
You may also obtain an unweighted least squares solution by setting all of the source profile
uncertainties to zero and all of the ambient concentration uncertainties to the same value (e.g.,
1.0). This gives the solution that you would get by multiple linear regression in a spreadsheet with
the intercept forced through zero. The solution in this case is dominated by the species with the
highest concentrations. Species such as Se (selenium) and V (vanadium) might get no
representation with an unweighted least squares solution.
F-
-------
For source profiles, in general, you wouldn't enter a value of-99. for a species' uncertainty
unless you also entered -99. for its abundance. A value of zero for a species' uncertainty might be
used for testing of least-squares approaches or for convenience, as illustrated in the example
above. In EPA-CMB8.2 the decision was explicitly made NOT to force ("hardwire") source
profile uncertainties > 0.0 for the following reasons:
1. As noted above, zero uncertainties in the source profiles default to the ordinary weighted least
squares (OWLS) method. This is a solution that some may still use. (Given what is known about
the dominance of source profile variability on the error, it has been shown that the OWLS solution
is biased compared with an effective variance solution.)
2. When data are unavailable for some components that are not likely to be in a source profile,
you should insert zeros for the value and the uncertainty. An example is in the NFRAQS data set
(Section 2.2), where we have lots of organic compounds for wood burning and cooking, and diesel
and gasoline vehicles, but none for the soil or the road salt or the secondary ammonium nitrate and
ammonium sulfate. You could put in a zero entry and a very low uncertainty, e.g., 1x10"7, to
indicate a detection limit. But this is essentially no different from zero when it is squared and
added into the effective variance with the uncertainties from those profiles that have real values.
It is also effectively zero when compared to the ambient data uncertainty.
Source Profile Matrix: Afit
Sometimes a special situation may arise that affects the choice of fitting species (Section 3.4).
This situation is likely to occur when single constituent sources are used as fitting sources. Single
constituent sources are described elsewhere (e.g., EPA, 2004; Appendix G) but essentially involve
the designation of secondary compounds as "sources" to allow EPA-CMB8.2 to account for
secondarily formed materials observed in ambient samples. For such modeling scenarios,
secondary compounds are involved that require selection of the constituent cations or anions. If,
for example, you have sodium nitrate as a fitting source, but haven't selected (included) sodium or
nitrate as fitting species, you will get a zero column in the source profiles (AFIT) matrix. When
this happens, the following error message can be returned when a calculation is attempted:
EPA-CMB8.2 Fortran Section
Column (=source) Number 5 of AFit = 0
OK
When 'OK' is clicked, the screen shown below appears. When EPA-CMB8.2 attempted to
apportion the fitting sources to the ambient sample, it expected certain species for one or more
single constituent sources (secondary compounds) that were excluded in the fitting species array.
This message will commonly appear when EPA-CMB8.2 is run in the Best Fit mode (Section
3.2 & 5.2). As EPA-CMB8.2 advances through various corresponding pairs of fitting species and
source profile arrays, it is apt to encounter inappropriate combinations. The model will advance
through all possible pairs, but will produce invalid results for the pairs that are ill-suited.
F-4
-------
HEPA-CMBB.2
File Help
t- H
Control File: INsjvf.inS
Samples: 1
Species: 15
Output Format: ASCII (.txt) -> Default
Input Files | Options | Samples | Species ] Sources Results
OPTIONS
Site: FELLOW Date: 07/26/88
Iteration Delta: 20
Maximum Source Uncertainty (%): 20
Minimum Source Projection: 0.95
Decimal Places Displayed: 5
Units: ug/m!
Duration: 24 Start Hour: 0
F Britt-Luecke F Best Fit
[~~ Source Elimination
Species Fit Array: 3
Sources Fit Array: 5
Size: FINE
Fit Measure Weights
Chi Square i
R Square i
°/o Mass i
Fraction Estimate i
Main Report
Contributions by Species
The fitting source profiles matrix (Afit) contained a column of all zeros.
Check the source profiles data file or change the fitting sources.
Delete Current
Delete All
Logic suggests that the 2nd line of the message should read: "... or change the fitting species."
Specifically for this example, the error message might have said (or been interpreted to say):
"Select a fitting species that corresponds to a positive entry for that species in the NaNO3 profile."
This error message will also be triggered for situations when there are many organic species and a
soil profile when we eliminate all of the geological elements. We usually have zero entries for the
organics in the soil profile because we don't measure them. It also happens when a biogenic VOC
source profile is included but there's no isoprene, and for many of the single constituent solvent
and coating profiles (e.g., toluene, acetone, etc.).
Using EPA-CMB8.2 with Factors besides Chemical Abundance: e.g.. Wind Direction.
As discussed above, an entry of-99. for species abundance (mass fraction) in a source profile
automatically removes that species from the source contribution calculation. It does not remove it
from the concentration calculation if there is also an entry for that species in the ambient data file.
This allowance of discrepancy, i.e., a valid (>0.0) value for a constituent in the ambient sample
data file and a value of-99. for its abundance in the source profile file(s), can be exploited for
cases when we want to include other information in the ambient data file that is not appropriate for
the source profile. For example, wind direction (WD) can be included to indicate that an upwind
source profile should be used or not used. The -99. for this variable (factor) in the source profile
file does not allow it to be used as a fitting species, but the direction is still easily available for
F-5
-------
consultation when picking sources. This is more efficient (and convenient) than using a separate
data base or look-up table for WD. It also allows the source contributions to be stratified by wind
direction (or other variable such as light scattering, temperature, or wind speed) by including it in
the ambient sample data (receptor) file.
The structure exists to display such external information corresponding to each sample that is
not, and cannot be, used as a fitting species. For example, one of the ways we recommend to
choose a profile for a sample is to determine if a source is upwind or downwind. For the sample
period(s) in question, the period-averaged WD can be imported directly into the ambient sample
data file as a "constituent" of each sample. The data file can thus be configured with a WD
variable into which you put the resultant wind direction, and set up a corresponding variable in the
source profile with a -99. value for "abundance". This information can be easily viewed on-
screen along with the species' concentration information (Section 3.3). If, for example, you see
that the sample data indicate that it is downwind (± 5°) from a coal-fired power plant, and you also
see a lot of selenium in this sample which doesn't appear in previous samples with prevailing
winds from opposite or very different directions, you may decide to include the power plant
profile, or retain a source combination you've already set up with that profile in it.
Highest wind speed during a sampling period is also good consideration for questions such as
"Do I want the road dust or the windblown desert dust profile in this sample?" It's also useful
when you have other data, like CO and NOX, VOC, etc. at the receptor that might help you to
choose (and justify the choice of) a certain profile or source type, but which doesn't have an entry
in any of your source profiles. By including these variables as sample "constituents", it puts all
the information in front of you rather than forcing you to analyze from several different (external)
sources of information.
ABSENT DATA
One condition in which EPA-CMB8.2's behavior is parallel with respect to ambient sample
or source profile data is that of "absent data". This condition is represented not by a flag for
"missing" (invalid) data but rather the complete omission of the value from the input data record.
Ambient Sample Input File (AD*.*):
If a value for a species concentration, for example, is absent (omitted) from the ambient
sample input data file, an error message such as the following will be returned:
Number of tokens in line 71 of file C:\CMB8\Bench Test\Testing\general\ADmiss.txt not equal to number of tokens in header line!
as soon as EPA-CMB8.2 control is moved off of the Input Files screen. As indicated, on line 71
of the ambient data input data file there is a missing (= absent) value. The following message
appears immediately after 'OK' is clicked on the message above:
F-6
-------
Ambient Data Error, Please see Section 4 of the Users Manual regarding input file construction,
OK
The following message then appears immediately after 'OK' is clicked on the message above:
The number of data values for this sample is wrong! (C:\CMB8\Delphi\Build 9\MainFrm,paSj line 2910)
OK
EPA-CMB8.2 must then be closed and an appropriate edition made to the input file after careful
examination.
Source Profile Input File (PR*.*):
If a value for a species concentration, for example, is absent (omitted) from the source profile
input data file, an error message such as the following will be returned:
Number of tokens in line 18 of file C:\CMB8\Bench Test\Testing\general\PRmiss4.txt not equal to number of tokens in header line!
as soon as EPA-CMB8.2 control is moved off of the Input Files screen. As indicated, on line 18
of the source profile input data file there is a missing (= absent) value. The following message
appears immediately after 'OK' is clicked on the message above:
Sources Set Error, Please see Section 4 of the Users Manual regarding input file construction,
I OK
F-7
-------
The following message then appears immediately after 'OK' is clicked on the message above:
The number of data values for this source profile is wrong! (C:\CMB8\Delphi\Build 9\MainFrm.pas., line 3056)
As for the example illustrated above for the ambient data file, EPA-CMB8.2 must be closed and
an appropriate edit be done on the input file.
F-8
-------
APPENDIX G
Notes on EPA-CMB8.2 Diagnostics
G-l
-------
EPA-CMB8.2's Main Report: EST
In the top portion of the Main Report is the diagnostic attribute "EST", which may be either
"YES" or "NO". The manual states in several places (e.g., Sections 3.6.1 & 5.1.1) that "The
field 'EST' under 'SOURCE' indicates (YES or NO) whether a source's contribution was
estimable in EPA-CMB8.2's attempt at a fit, using the settings in Options." This is also stated in
Table 4.2-1 of the Protocol for Applying and Validating the CMB Model for PM-2.5 and VOC.
(EPA, 2004). In particular, it is first order indication of whether or not the profile/source
contribution estimate combination1 meets or does not meet the criteria that have been set up in
Options. For any particular CMB fit, it is possible to set values for Maximum Source Uncertainty
and Minimum Source Projection in Options to indicate all YESs or all NOs (see below).
We currently don't know how those settings translate into useful information, as indicated by
the YESs and NOs. Unless you're researching how CMB responds to different combinations of
source profiles and uncertainties, this parameter may not be very useful. The most important
evaluation parameter is the uncertainty of the source contribution estimate, which is a function of
the input uncertainties and the collinearity (variance inflation). You'll notice that when you have a
negative source contribution, it usually has a very high uncertainty and there is another
compensating source contribution (usually a very high positive value) that also has a high
uncertainty. This is a good indication that these are collinear. This is why the model gives the
uncertainties and why they should be considered in decision-making, as described in the Protocol
for Applying and Validating the CMB Model for PM-2.5 and VOC. How to use this diagnostic in
a practical setting has not been sufficiently studied. For now, it is really there for research
purposes and not for everyday application.
In sensitivity testing for the two Options parameters mentioned above, the following has been
seen:
Maximum Source Uncertainty.
The default is value is 20%. As this value is reduced, the occurrence of NOs increases. Once
this value is set to zero, EST = NO for all sources. Conversely, as this value is increased, the
occurrence of YESs will increase.
Minimum Source Projection.
The default value is 0.95%. As this value is substantially reduced, the occurrence of YESs
increases. As it approaches zero, EST = YES for all sources. If set to 1.00, EST = NO for all
sources.
The array (pattern) of YES/NO under EST is fairly sensitive to settings of Maximum Source
Uncertainty; it is fairly insensitive to settings for Minimum Source Projection. However,
adjustment of these two parameters has an independent (and opposite) effect on the EST variable.
This "combination" is important because a profile may be well separated for a high contribution, but it may fall within
the uncertainties of the other profiles for a low contribution.
G-2
-------
United States Office of Air Quality Planning and Standards Publication No. EPA-452/R-04-011
Environmental Protection Air Quality Strategies and Standards Division December 2004
Agency Research Triangle Park, NC
------- |