Session; PM Receptor Modeling
Chemical Mass Balance Model: EPA-CMB8.2
C. Thomas Coulter,* Robert A. Wagoner,** and Charles W. Lewis*
* National Exposure Research Laboratory, U.S. Environmental Protection Agency, Research
Triangle Park NC 27711
** Pacific Environmental Services, Research Triangle Park, NC 27709
ABSTRACT
The Chemical Mass Balance (CMB) method has been a popular approach for receptor
modeling of ambient air pollutants for over two decades. For the past few years the U.S.
Environmental Protection Agency's Office of Research and Development (ORD) and Office of
Air Quality Planning and Standards (OAQPS) have collaborated to develop a new generation of
CMB software, CMB8. Developmental work was initiated under EPA contract by the Desert
Research Institute and is being continued by Pacific Environmental Services, The current
version, EPA-CMB8.2, incorporates the upgrade features that CMB8 has over CMB7, but also
addresses problems identified with CMB8 and adds enhancements for a more robust and user-
friendly system. EPA-CMB8.2 features include: (1) full use of Windows (32-bit) capability for
file access/management, (2) a tabbed-puge interface that facilitates the sequence of steps for
doing a CMB calculation, (3) multiple indexed arrays for selecting fitting species and sources,
(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 CMB7 and CMB8 flaws, (8) more options for input and output data formats,
(9) addition of a more accurate least squares computational algorithm, (10) a new treatment of
source collinearity, and (11) choice of criteria for determining best fit.
INTRODUCTION
The Chemical Mass Balance (CMB) air quality model is one of several receptor models
that have been applied to air resources management for over two decades. 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 dispersion and transport, and chemical transformation mechanisms
to estimate the contribution of each source to receptor concentrations. CMB helps explain
observations that have been made; it does not predict ambient impacts from sources as do
dispersion models. The two types of models are thus complementary, with each type having
strengths that compensate for the weaknesses of the other.
The U.S. Environmental Protection Agency (EPA) has tacitly approved CMB as a
regulatory planning tool through the Agency's approval of numerous State Implementation Plans
(SIPs) which have had a CMB component, CMB has proven to be a useful tool in applications
where steady-state Gaussian plume models are inappropriate (e.g., stagnant wind regimes), as
well as for confirming and or adjusting emission inventories. Since about 1990 EPA has
supported and freely distributed the current version of the CMB software (CMB7). CMB8 was
originally developed in 1998 by Desert Research Institute, Reno, NV under contract with EPA's
1
-------
ORD and OAQPS, In September 1999, Pacific Environmental Services began work under
contract with ORD/OAQPS to repair, reconfigure and enhance CMB8. This joint contractual
venture resulted in a beta test version of this new generation of CMB model: EPA-CMB8.2,
which is the subject of this paper. Associated documentation and guidance will be released when
a fully functional and tested model is complete.
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.1 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.
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.
The effective variance weighted solution2 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 precision of both the ambient
concentrations and source profiles; and 4) gives greater influence to chemical species with higher
precision in both the source and receptor measurements than to species with lower precision. The
effective variance is a simplification of a more exact, but less practical, generalized least squares
solution proposed by Britt and Luecke.3
The CMB model can be written as:
where C, is the ambient concentration of specie i, at. is the fractional concentration of specie i in
the emissions from sourcej, St is the total mass concentration contributed by source j,p is the
number of sources, and n is the number of species, with n > p. The C, and are known and the
Sj are found by a least squares solution of the overdetermined system of equations (1). While the
system is simple in appearance its solution is not, because of the need to take account of
uncertainties in both the C/s and the fly's. These uncertainties are introduced through "effective
variance" weighting of the terms in Eq. (1).
EPA-CMB8.2 is intended to replace CMB74 as a more convenient method of estimating
contributions from different sources to ambient chemical concentrations. EPA-CMB8.2 returns
the same results as CMB7, but it operates in a Windows-based environment and accepts inputs
THE CMB APPROACH
(1)
EPA-CMB8.2 FEATURES
2
-------
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 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.eoa. gov/ttn/chief/).
Multiple Arrays for Fitting Species and Fitting Sources
Up to ten indexed arrays of fitting species and fitting source profiles may be specified in
input data selection files. Different arrays can be selected during EPA-CMB8.2 operation. Upon
exit from any selection windows for fitting species or fitting sources, 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 approximation5 to the Britt-Luecke iteration scheme3 was used in CMB7, exercise of
Britt-Luecke algorithm option in EPA-CMB8.2 allows solution without approximation.
Improved Collinearity Diagnostics
Collinearity in a CMB context refers to a situation in which the compositions of two or
more source profiles are so similar that it is difficult to infer the separate impact of each.
Collinearity is a form of mathematical ill conditioning, and its presence is associated with
additional serious problems beyond the one just noted. The basis for detecting and handling
collinearity in EPA-CMB8.2 is the same as that used in CMB7: the analysis and formulation
given by Henry.6,7 However the "packaging" is very different versus CMB7: the
"uncertainty/similarity clusters" description used in CMB7 has been replaced in EPA-CMB8.2
by "maximum source uncertainty", " minimum source projection", and "estimable sources"
terminology. User experience will be needed however before the new freedom to select
collinearity parameter values can be exercised confidently.
The approach in EPA-CMB8.2 more directly mirrors Henry's treatment of the collinearity
problem and exploits its potential more quantitatively. It gives the user access to 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 the CMB volatile organic
compounds (VOC) applications which have become of increasing interest.
Search for Best Fit
3
-------
Using a user-selected best fit criteria, EPA-CMB8.2 can systematically check up to 10
possible paired combinations of fitting species and sources arrays. The best fit pair of arrays are
then indicated in their respective windows.
User-set 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.
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 computer, not by
pre set memory limitations.
Flexible Input and Output Formats
Comma-separated value (CSV), xBASE (DBF), and worksheet (WKS) formats are
supported as input and output files, in addition to the blank-delimited ASCII text files (TXT)
supported by CMB7.
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. Text output can
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.
Miscellaneous Changes
The following items are differences between CMB7 and EPA-CMB8.2 that are not as
fundamental as the "featured" ones described above, but which are useful to list.
• An indexing error present in the CMB7 "Autofit" option has been corrected. The error
caused a mismatch between sample label and computed source contributions, when VOC data
were analyzed.
• The user can define a customized 'Fit Measure' more inclusive than either or r2 alone.
4
-------
RUNNING EPA-CMB8.2
EPA-CMB8.2 employs an event-driven 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. This type of presentation is clear
and eliminates the confusion associated with multiple buttons for the same functions. A tabbed
interface provides users with visual clues regarding the logical progression of steps necessary to
run the model. EPA-CMB8.2 also features robust, context-sensitive Windows-style help,
including keyword search. These improvements provide a Windows look-and-feel to the CMB
software that is familiar to most users.
Input Files
Figure 1 shows the screen that first appears when EPA-CMB8.2 is launched. Most users
will have prepared a Control File for a particular CMB application. Control Files are commonly
used in air quality models to specify input files that will be invoked during runtime. In EPA-
CMB8.2, the Control File would contain the names of the required ambient data and source
profile data input files, as well as three optional selection files for ambient data samples, fitting
species, and fitting source profiles. 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 associated) input files, the other mode should be selected. If Use
Control File was selected, a browse box appears as shown in Figure 2,
Figure 1. Launching EPA-CMB8.2
ChtmFcal Mm* Balanc*
c Use Control File
f Enter Individual input Files
]i—ok—a i
Figure 2. Browse Box for Selecting a
Control File
CMB Control File
look in: L_J Demo
^jlNdemol
£*} INdemo2
Open {
Flwofiyper | CM9 Canted Fife finMnQJ
Cancel
Once a Control File is chosen, the Input Files window appears, as depicted in Figure 3.
In the Input Files window, the name of the Control File 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 task bar at the top. Even though a Control File has been
selected at this point, another one can easily be chosen via its browse box. This is also true for
any of the input files (they may 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. If at start-up
(Fig. 1) a Control File is not used, an Input Files screen appears as illustrated in Figure 3, except
that the Control File name is absent. Files are selected via their respective browse boxes.
5
-------
EPA-CMB8.2 Options
Several options are available in EPA-CMB8.2 that are selected from the Options tab
(Figure 4), where various values and selections may be changed from their default values.
Figure 3. Input Files Screen
Figure 4. Options for Current Run
Iteration Delta
This paiameter 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 (default = 20) is adjusted via the
scrolling arrows.
Maximum Source Uncertainty / Minium Source Projection
These parameters allow the eligible space collincarity evaluation method of Henry7 to be
implemented with each CMB calculation. The eligible space method uses: 1) maximum source
uncertainty; and 2) minimum source projection on the eligible space. The maximum source
uncertainty is expressed as a percentage of the total measured mass and is adjustable via the
scrolling arrows (default = 20% ). The minimum source projection is set to a default value of
0.95, but can be changed in the display field.
Decimal Places Displayed
This parameter sets the number of decimal places displayed in the output window and
output files, and depends on the units used in the input data files. This parameter may be
adjusted by using the scrolling arrows. 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).
Britt and Luecke
Checking this box applies the Britt and Luecke3 linear least squares solution that is
explained by Watson et al.5 when applied to CMB calculations. This option enables a general
6
-------
solution to the least squares estimation that includes uncertainty in all the variables (i.e., the
source compositions as well as the ambient concentrations), whereas the default 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 %' 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 %2 will
be superior in terms of its physical meaning. Invocation of this option does affect 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.
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.
Best Fit
Checking this box causes the program to cycle through the corresponding pairs 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, FPA-
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. After a Best Fit has been
made the fitting species and fitting sources arrays will be tagged (highlighted) in their respective
windows.
Fit Measure Weight
These are the weights (coefficients) applied to each of the performance measures r,
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 10,000 may be entered by typing into the appropriate
display fields. Defaults are 1.0 for each performance measure weight.
Selecting Ambient Data Samples
The screen for selecting a subset of samples on which source apportionment will be
performed is shown in Figure 5, If an (optional) DS*.SEL file is being directed from the Control
File, one or more samples may appear selected initially. 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 be also be used to help establish the desired list of
samples. A counter displays on the top task bar the number of samples selected at any given
time. As indicated, for each sample, the collection date, duration, start hour, and size fraction
(particles) are displayed. 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 all
selected (tagged) samples only, or for all samples. Use of the "VCR" control buttons on the top
task bar can help navigate down a long list of samples.
7
-------
Figure 5. Ambient Data Selection
^ *** •
"!"l "! "I 1 I ^ 1 :1 • 'cwwffciwd^jat-;' >. vsp^m: sc^s ; - £' v-^ : '
hvUPlea | Opttona SMtmte* | Specfe* ] Source? J Rosut* j
AMBIENT DATA SAMPLES
SOJECTBOlsiTE
ffMTE
jaw
tISTART
I SIZE
jTMAC
TMMftt
K3IC •
torn I
sac
sm
M*TG
IWTB L
CROWS
02/15©9
24
0
Fre
23.999700
1.533600
13.991400
0.726403
2186900
0.141100
5 081100
0^72300
CROAS
02£*1«9
24
0
F*E
19,228500
1.078400
9.552600
0.516000
1.7S4600
O.I 25100
3.071000
0.21 S900
CRCWS
03j23j89
24
0
fihe
7 961100
0.B3S4OO
2.181500
0.21 oeoa
0.394700
0.095900
0.919700
0.101400
CROWS
0e
27 767900
1 473Q00
S.085100
0.3S300G
S.3567«i
0 326200
4.S110C0
0 242709
FELLOW
01)10 JS9
24
0
Fth£
26,374400
1,412000
8.36C6CO
0.454700
3.974»M
0.214900
4.253700
0 230800
FELLOW
01A6J33
24
0
FJNE
39.239400
2.019100
13.694300
D.7OBO00
5.066 4tXI
0 265230
5.823200
C,303700
FELLOW
~1/22J03
24
0
Fsrt
21 6344U0
1 187300
4 385300
0.281800
1.&1G2C0
0 253730
3.167500
0.101600
T3I
Select AH Samples
De-select All Samples
View Selected
For any sample displayed, clicking View Graph will provide a bar chart such as shown in
Figure 6. This graph is useful to verify that input data files have been properly read. View Grid
returns to the ambient data selection screen.
Figure 6, Speciation Graph for Ambient Sample
Eto U*Ip
m} «| ~ | m| Eg j | j ConaolFte iNdemol.WJ Sampteac 1 Specie* 20 Source* 6
hptiFtoa | Cplera Sample* | Sp«des j Sowces | FltadLt }
AMBIENT DATA SAMPLE - FELLOW07/26/88
21
20
19
10
17
16
15
14
13
12
! 11
l1Q
9
a
7
6
5
4
3
2
1
0
o %j o
S ^ -
«-> u o O *¦> y
-------
Selecting Fitting Species
Fitting species are used in the calculation of source contribution estimates. Species not
included in this calculation are termed floating species. 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 7. This
screen is initiated by data read from the (optional) species selection file (PO*,SEL). Species
contained in the selection file are listed down the left-hand side and a field of up to 10 arrays is
provided. At start-up, 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,
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. Use of the "VCR" control buttons can help
navigate down a long list of species, and a counter displays on the top task bar 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.
Figure 7. Fitting Species Arrays
£3b H*P
r-" | | ~ | fri [ Eft ( | j Certid Fte SNdemolJnS Sampte* 1 Spec&t; 20 Soure©r 6
Fie* | CtjU"* f Swrvtos Specie* j Smcw j Rwlt ]
SPECIES FITTING ARRAY
specie
[SP€CiENAMe[ -f
2
3
4
5 | i
i | 7
8 1
| | 19 j _•*
N03 *
' NKretfe Oy '<
S4tC
* a by C iugj»n3)
N4TC
NH4
•
* Ammcnkm by AC (uofrr>3)
KPAC
K-S
*
* Sdubie PrtaisMm by AA (u£j/fT>:
NAAC
MA ¦
* Soluble adiunt by AA [ug*n3)
ECTC
OCTC
EC
QC
* fctertwrtaiCwbcr? Dy TOR (uqjiw
' Organic Carton by TOR
Select All Array 4 |
ALXC
SIXC
AL *
s> "
*
• AJurwwar by XKF (ug£n3)
* by KRF
Cloni Al! Array 4 |
suxc
s
*
* Siifur t»y XRF (tjgsn.J)
CLXC
C1_
*
4 CNorde by XRF (ugif.3)
KPXC
CAXC
K *
CA •
,
' PotosaBjm by XRF (ugJ>n3i
CaicljT by XRF (uaMWTl
View Sal^cled J
TIXC
Tl
-
«
*
-
T t*Y KRF Cuj^n3)
VAXC
V ¦
*
VanaOxjn &y Cug#n3)
C*XC
Cfi
*
.
ChromMn ty XRF (u^MO)
MNXC
l»»N *
*
-
Manganese by XRF
FEXC
FE
iron by XRF (ugAfD)
WIXC
fj •
»
teekeH by XRF 33
znxc
ZN
Zinc try XRF (ug>r>3)
BfiXC
m
"
erpmino by XRf
P9XC
p©
Lead by >£RF
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.
Selecting Fitting Sources
For fitting sources the user should select profiles that represent the emissions most likely
to influence receptor concentrations. The screen for selecting fitting sources is shown in Figure
8. This screen is initiated with data read from the (optional) species selection file (SO*.SEL).
9
-------
Source profiles contained in the selection file are listed down the left-hand side 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). Select/Clear All Array X, "VCR" control buttons, and View Selected/View All
buttons function as described above for fitting species. Note that a counter displays on the top
task bar the number of source profiles tagged for any selected array.
Figure 8. Fitting Sources Arrays
£*>
EB I a
CootroiFl® iNdemol.hS Samples 1
Specwr. 20
HaJ FVss ] opttorts | SairetM } Spades Source* J Raauis }
SOURCES FITTING ARRAY
PNO |SC>
|SiZE j
1
2
3
4
5
s
8
9
10 j COMMENT
N3C
W3UJ
S4IC
jjJBSSsofLo1
FIM
STOCKTCN AGRJCUL URAL SC
0OO27QO
Q.00470C
o.ooo
SJV002 SOLOS
FINE
FF3ESNO PAVED ROAD
o.ooocoq
0 003&00
0 005
SJV003 SOIL 04
FINE
VTSALIA AORtOJLTURAL SCHL
o.ooocoo
0 003000
coco
SJVOOfi SOLOS
FINE
VfSALiA AGRICULTURAL SCL
0.000300
0.002600
o.ooo
SJV005 SCIL06
rwe
visalia sand ako ©ravel
o.doooqo
o.osseoo
0 000
SJVGOB SOLO?
FINE
VISALiA URBAN UNPAVEt?
0 00 JSC©
0 00290G
0 000
Select AH Array S
SJVOG7 SGtLOS
FtNS
VfSALiA PAVED ROAD
o.ooteoo
0.0051 OQ
0.003
SJVOOB SOtLGS
FS>«
BAKEffSFlELO AGRICULTURAL
0.000000
0J3023C0
0.004
~ear Alt Array 5
SJV009 SOIL10
FINE
8AKERSFIELD AGRICULTURAL
~ ~01100
0.009600
0 000
SJV010SCl.il
FINE
EAKERSFtELD UWAVED ffOAC
0 000500
O.DC30OQ
0 002
sjvmt sct.12
FINE
•
BAKERSTIELD PAVED ROAD
9 001100
o.ocsaco
0 011
SJV012SOB.13
FINE
SAKER3FIELD WANDBLOWN UR
0 001300
0 002300
0.003
Vlaw Salectad
SJV013 SOft.14
FINE
,
BAKERSFiELD AGRICULTURAL
0.000000
O.OC3SOO
0.001
SJV014 SOIL.15
FINE
9AKERSFC1.D AGRICULTURAL
0001400
0.016300
e.001
Hide Data
SJVQ1S SQH-15
FINE
BAKERSFSELD UfsPAVED RCAL
a 002*30
0OC1800
0.001
SJV016 SOL17
FINE
•
TAFT LtNPAVED ROAD
0 000000
0 002500
0 007
SJVQ17 BAMAJC
FINE
'
•
BAKERSFCD CORCWOOD. MA,
0.004600
0.001200
001 i
SJV018MAMAJC
FINE
MAMMOTH LAKES CORCfcNOOC
0,001700
0-C00400
0-006
,
SJV019 MaFISC
FINE
BAKERSFELD CORtWOOD, F1S
0.000300
0.00 (WOO
0.001
View Giaph j
SJVQ20 MAOEC
FtNE
MAMMOTH LAKES OtESEL TOLA
0.000600
0.000100
0005
SJV021 DAAOBC
FINE
BAKEflSFBLD AGRt. BURN
-------
the upper right-hand corner. Viewing any sample result in a series is quite easy with the use of
the VCR buttons on the task bar. Using appropriate buttons, the current or all results may be
deleted from the output window. A report may be printed for any/all sample results via the print
button on the task bar.
Figure 9. Calculation Results - Main Report
£*>
SOURCE
EST CODE
FITTING STATISTICS:
0,9?
0,24111
0,4013?
STD ERR
% MASS
DECREES FREEDOM
T5TAT
.lit S'fi;
In the Main report are several information blocks which present performance measures.
First are fitting statistics: r2, %2, 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, standard
errors, and values for T-stat. The field 'EST5 under 'SOURCE' indicates (YES or NO) whether
a source's contribution was estimable by EPA-CMB8.2. The next block is the eligible space
display: an echo of the measured concentration and error for the sample, eligible space
dimension for the chosen maximum uncertainty, inverse singular values, the number of estimable
sources for the chosen minimum source projection, and estimable linear combinations of
inestimable sources. Finally (not shown), there is a block listing 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
residual (calculated - measured mass) to the uncertainty.
Contributions by Species
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
10). This report 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. 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 task bar.
11
-------
Figure 10. Calculation Results - Contribution by Species
1-1-I
Control Hoc INdemol.HS '
I options | Swrcfe* ] mparts* ] Sources Re*ufla j
SAKEUs
SiteFEtLCW
Dale ormx® Duration 24
Stmt Hour O
OPTIONS herMlcm Delta 20
Maximum Source (Jncerfakity P*> 20
Minimum Source Pro}«rtton 0.9S
Decimal Places Displayed s
Units pg*n3
P |BrmXui?CKe]
P Stmj c» Etvninatlon
Sttecles fftt Array 4
Sources FN Array 5
? <5 hr1i;>p.,in 1
. M,iss
Conlrlbuilona by Spetlea
MPtM M«trhi
SOURCE
NAME
SFECIES
CALCULATED
SOIL12
BAMAJC
SFCROC
M0VES2
AKSUL
AMHIT
THAU
19.4 B17 3
0 2C6
0 . 066
O. 209
C . 124
0 . 288
0 Oil
N3IU
0 25155
0 023
0 . 031
0 . 000
0 250
0 . 000
a 878
54IU
S.57843
0 . 009
O . 004
0.178
C.Oli
0 .876
0 . 000
H4TV
1 . 7S067
0 . 000
0.001
0 . 000
0, coo
0.094
0 029
KPAtf
0.032 BO
0 . 201
0.495
0.023
0 00c
0 . 000
Q . 000
NAAU
C.07398
0 2S4
0.013
0. 230
0 coc
0.000
0.000
ECTU
1.73805
0 065
0.273
0. 000
1 764
0 000
0 000
OCTU
2.79661
0 213
0 . 1S 4
0 001
0.341
Q QOQ
0 00 Q
&LXU
C.32674
1.1O0
C .000
0 . 000
0 , 007
0-000
0 . 00D
SIXU
0 . 88718
0 .913
0 . 000
D.OCO
0 , 027
a. 000
0.000
sexv
1 a087S
a. 011
0.004
0.122
0 014
Q 747
Q . DOO
CLXU
0.04495
0 581
0.992
a.033
0 O?0
0.000
0 . QOQ
KPXTT
0.14563
0 531
C .349
0 . Oil
0. 001
0.000
0 00c
CAXD
0.21216
1 159
0 .006
0.015
0 011
0. aoo
n 000
TIKU
C.02135
0 701
c .000
0 . 015
0,001
0.000
o.coo
7AX0
C.03829
0 . 036
0 . QOQ
1009
0.001
0 . ODD
C . DOO
CRJCU
0.03178
0 202
~ .000
0 ose
0 coo
0 Q00
0 coc
MHXU
0.00564
0 766
o.coo
0 . 078
0 128
0. oco
0 000
FEXV
Q.2S3SS
0 . 925
0.000
0 . 034
0.000
0. oco
0 , COO
NIXtf
0.03602
0 012
0 .000
0 . 939
0 GOG
0 000
0 000
CUXU
0.00102
0 . 006
0 . 000
0 . 000
0 .oot
0.000
0. coo
ZNXU
0.02462
0.110
0. 014
0.12&
0.01s
0. oco
c 000
SRXU
0.00761
C 041
0.013
0. 000
0 . 651
0 000
c 000
PBXU
0 01,063,
0 385
O.OOO
0 . 000
t . 430
0 000
0 , 000
"3
Delete Current
Modified Pseudo-Inverse Normalized (MPIN) Matrix
Another report that may be of interest is the Modified Pseudo-Inverse Normalized
(MPIN) matrix (not shown). The MPIN matrix identifies which fitting species have the largest
influence on the source contribution estimates from each profile. 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 printout of the MPIN matrix may be obtained.
CONCLUSION
Features of the new version of CMB software (EPA-CMB8.2) have been described. The
model requires additional beta testing to reveal and resolve problems before a final version can
be released by EPA. It is anticipated that EPA CMB8.2 will significantly increase the
convenience, efficiency and accuracy of performing receptor modeling by the CMB method.
DISCLAIMER
The information in this document has been funded wholly or in part by the United States
Environmental Protection Agency under Purchase Order No. 9D-1844-NTSA to Pacific
Environmental Services. It has been subjected to Agency review and approved for publication.
Mention of trade names or commercial products does not constitute endorsement or
recommendation for use.
REFERENCES
1. Friedlander, S.K. Environ. Sei, TechnoL 1973, 7, 235-240.
2. Watson, J.G. JAPCA 1984, 34, 619-623.
12
-------
3. Britt, H.I.; Luecke, R.H. Technometrics 1973, 15(2), 233-247.
4. Environmental Protection Agency, 1990. Receptor Model Technical Series, Volume III
(1989 Revision). CMB7 User's Manual. EPA-450/4-90-004. Office of Air Quality
Planning and Standards, Research Triangle Park, NC.
5. Watson, J.G.; Cooper, J.A.; Huntzicker, J.J. Atmos. Environment 1984, 18(7), 1347-
1355.
6. Henry, R.C. "Stability analysis of receptor models that use least squares fitting," in
Proceedings of the AWMA Specialty Conference on Receptor Models Applied to
Contemporary Pollution Problems, SIMS, Air & Waste Management Association:
Pittsburgh, 1982; pp. 141-157.
7. Henry, R.C. Atmos. Environment 1992, 26A, 933-938.
13
-------
nerl—rtp-heasd-oo—197 TECHNICAL REPORT DATA
1. Report Noepa/600/A-00/099 2.
5. Report Date
6. Performing Organization Code
4. Title and Subtitle
Chemical Mass Balance Model; EPA-CMB8.2
7. Author(s)
C. Thomas Coulter, Robert A. Wagoner, Charles W. Lewis
8. Performing Organization
Report No.
9.Performing Organization Name and Address
National Exposure Research Laboratory
Office of Research and Development
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
10. Program Element No.
11. Contract/Grant No.
9D-1844-NTSA
12.Sponsoring Agency Name and Address
National Exposure Research Laboratory
Office of Research and Development
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
13. Type of Report and Period
Covered
Symposium Presentation
14.Sponsoring Agency Code
EPA/600/09
15. Supplementary Notes
16. Abstract
The Chemical Mass Balance (CMB) method has been a popular approach for receptor modeling of
ambient air pollutants for over two decades. For the past few years the U.S. Environmental Protection
Agency's Office of Research and Development (ORD) and Office of Air Quality Planning and
Standards (OAQPS) have collaborated to develop a new generation of CMB software, CMB8.
Developmental work was initiated under EPA contract by the Desert Research Institute and is being
continued by Pacific Environmental Services. The current version, EPA-CMB8.2, incorporates the
upgrade features that CMB8 has over CMB7, but also addresses problems identified with CMB8 and
has been reconfigured and enhanced for a more robust and user-friendly system.
17. KEY WORDS AND DOCUMENT ANALYSIS
A. Descriptors: receptor modeling, chemical mass
balance, ambient sample, source profile, source
apportionment, speciation
B. Identifiers / Open Ended
Terms
C. COSATI
18. Distribution Statement
RELEASE TO PUBLIC
19. Security Class (This
Report)
UNCLASSIFIED
21. No. of Pages
20. Security Class (This
Page)
UNCLASSIFIED
22. Price
File: S:\HEASD\Forms\Tech-Form-2220-1
------- |