EPA-450/4-84-021
Estimating PM10 And FP Background
Concentrations From TSP And Other
Measurements
By
Pedco Environmental, Inc.
Golden, CO 80401
Project Officer: Thompson G. Pace
U.S. ENVIRONMENTAL PROTECTION AGENCY
Office of Air and Radiation
Office of Air Quality Planning and Standards
Research Triangle Park, NC 27711
August 1984
-------
This report has been reviewd by the Office Of Air Quality Planning And Standards, U.S.
Environmental Protection Agency, and approved for publication as received from the contractor.
Approval does not signify that the contents necessarily reflect the views and policies of the
Agency, neither does mention of trade names or commercial products constitute endorsement
or recommendation for use.
EPA-450/4-84-021
-------
CONTENTS
Figures
Tables
Introduction
Background
Purpose
Scope
Organization
Methodology and Data Base
Evaluation Procedure
Qualifying Predictive Models
Adjustments to Ranking
IP Network .
Other Data Sets
Description of Estimating Methods
Method 1Dichotomous sampler data
Method 2TSP times constant ratio
Method 3Nearby suburban residential site
data
Method 4Nearby suburban value times TSP , ,
TSp rural/
suburban
Method 5IP sulfates and nitrates at suburban
residential site times constant ratio
Method 6TSP sulfates and nitrates at back-
ground site times constant ratio
Method 7-Size selective inlet (SSI) times
constant ratio
Method 8Max 24-h background value as constant
ratio of corresponding annual average
concentration
Method 9Max 24-h IP times constant ratio
Method 10Linear function of visibility
(extinction)
Method 11-Linear function of (visibility + TSP)
Method 12Default values for FP in the west
Method 13Average of values obtained from
two or more of the above methods
v
v
1
2
2
4
5
6
7
10
12
16
16
17
19
19
19
20
21
21
22
22
21
24
24
26
ill
-------
CONTENTS (continued)
4.
Results
General observations
Observations by method
5. Conclusions and Recommendations
Ranking of predictive methods
Procedure for adjusting background concen-
trations
Other considerations for determination
of PM1Q background
Recommendations for additional work
References
Appendix A Calculations for selected background
estimating methods
Page
27
28
32
36
36
38
40
41
43
45
-------
Number
FIGURES
Median Annual Visual Range at Suburban/Nonurban
Locations in the United States
Page
23
Number
2
3
4
5
6
7
TABLES
Methods of Estimating PM-_ and FP Background 3
Available IP and FP Data Sets 9
Summary of EPA's IP Network Data 11
Summary Data from Nine SURE Sites 13
Summary Data from Western Particulate
Characterization Study 14
Summary Data from PEDCo's Compilation 15
Predictive Equations for Estimating Background
IP and FP 28
Standard Errors of Estimate for Predictive
Methods 30
Methods for Estimating Size-Fractionated
Particulate Background Concentrations 37
v
-------
-------
SECTION 1
INTRODUCTION
BACKGROUND
This document has been developed in response to the recog-
nition that there may be a number of States revising their State
Implementation Plan (SIP) for particulate matter (PM) in light of
the recently proposed revision to the National Ambient Air Qual-
ity Standards (NAAQS) for PM1Q. State and local agencies need a
method of estimating background for size-fractionated particulate
as part of SIP development for PM1Q and fine particulate (FP).
The method for estimating background concentration that is
specified in Federal regulations (Section 51.13c) is to use a
concentration measured at a nonurban site in or near the analysis
area that is unaffected by nearby emission sources.
Most regional monitoring networks contain one or more desig-
nated background hivol sites. However, there are relatively few
size-fractionated particulate samplers measuring PM.n and FP
compared with those measuring TSP; and almost all of the former
are in urban locations. Therefore, states may not have direct
measurements of PM10 or FP background available when they prepare
their SIP's. .
This document provides information on how to estimate a PM
background value from other measurements when PM.. _ monitoring
data are not available. Two time frames are considered, annual
and 24-hour (denoted herein as high 24-hour). The discussion
relating to the 24-hour situation should be thought of as a means
of adjusting a 24-hour Total Suspended Particulate (TSP) or
Inhalable Particulate (IP) background observation to PM.. _ for use
with a 24-hour design concentration; and likewise for the annual
situation for adjusting an annual TSP or IP background to PM.. .
for use with an annual design concentration.
10
-------
PURPOSE
The purpose of this study is to evaluate several indirect
methods of estimating PH,Q and FP concentrations at background
locations from other available measurements, and to rank these
alternative methods according to their ability to accurately
estimate background levels in diverse geographic areas. FP is
included in this document because it may be helpful in estimating
the long range transport component of measured data or in
determining the causes of visibility impairment.
The methods evaluated are based on relatively simple rela-
tionships that might be shown to exist (e.g., PMTA annual average
= constant percentage of TSP annual average). The methods to be
evaluated are listed in Table 1. Further descriptions of them
can be found in Section 3.
Several of these methods have been examined previously for
predicting PM.. n and/or FP concentrations at any location, not
1-3
just at remote nonurban locations. The present effort has
attempted to use or build on these past analyses, rather than
arbitrarily selecting possible predictive models.
SCOPE
As indicated in Table 1, several different background esti-
mates are desiredannual average (ann av) PM--, high 24-h PM,0,
ann av FP, and high 24-h FP. In this analysis, the observed max
24-h data are used to develop the high 24-h methodology. In
practice the 24-h background may not be based on the max concen-
trations. In addition, a further breakdown of these previously
mentioned four background concentrations into regionally-gener-
ated and long-range transport components would be of value in the
SIP development process. However, with the resources available
to the present study, this additional breakdown of estimated
background concentrations was not attempted.
The 13 methods in Table 1 are evaluated separately for their
ability to predict each of the four background values, or as many
-------
TABLE 1. METHODS OF ESTIMATING
AND FP BACKGROUND
Estimating method
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
Use dichotomous IP data, adjust total
IP to PM1Q
TSP at background site times constant
ratio
Use IP data from nearby suburban resi-
dential site
IP at nearby suburban residential site
times ratio of IP to TSP values at
surburban background sites
IP sul fates and nitrates at suburban
residential site times constant ratio
TSP sul fates and nitrates at background
site times constant ratio
Size-Selective Inlet concentration at
background site times constant ratio
Prediction of high 24-h value as con-
stant ratio of corresponding ann av
concentration
High 24-h IP times constant ratio
Linear function of visibility (extinc-
tion)
Linear function of (visibility + TSP)
Default values for FP in the west
Average of values obtained from two or
more of the above methods
Background levels
for which applicable
Ann av
PM10
X
X
X
X
X
X
X
X
X
X
High
24-h
PM10
X
X
X
X
c
X
X
c
c
Ann av
FP
b
X
X
X
X
X
X
X
X
High
24-h
FP
b
X
X
c
X
X
c
c
X
The "high 24-h" methods are applicable when exceedances of the short term
. NAAQS are of interest.
Direct measurement obtained from IP dichotomous sampler.
c.
Inadequate data available to evaluate.
-------
of the four to which the method is applicable and for which ade-
quate data are available. It is possible that different methods
may be recommended for estimating the different background
values.
ORGANIZATION
Evaluation and ranking of the 13 methods were performed
primarily on the basis of how closely the background values
predicted by each method match a set of measured background
values from EPA's IP Network. "Closeness" of the predictions is
to be determined by the standard error of estimate (the standard
deviation of differences between measured and predicted values).
The evaluation/ranking system is described in detail in Section
2.
The lack of PM and FP background data is a problem not
only for the states in preparing their SIP's, but also for use as
a comprehensive data base for the present analysis. The small
data base, low concentrations at background sites (near the labo-
ratory analytical thresholds), differences in sample handling and
analysis procedures inherent in a national data set, and signifi-
cant seasonal variations found in fine particulate all contribute
to increased standard errors for the estimating methods.
Because of these deficiencies in the available data base,
some qualitative factors are also considered in evaluating and
ranking the methods. The calculated standard errors and judgmen-
tal adjustments are presented in Section 4, Results.
In the final section, Conclusions and Recommendations, the
acceptable predictive methods are ranked for each type of back-
ground concentration.
-------
SECTION 2
METHODOLOGY AND DATA BASE
METHODOLOGY
Evaluation Procedure
The procedure used to evaluate alternative background predic-
tion models must be capable of ranking them in some sort of
systematic manner. Consequently, the procedure should be equally
applicable to all methods and capable of being employed consist-
ently to different types of data sets (e.g., TSP data, SO, data,
visibility data)
Previous work has been done in evaluating methods of esti-
mating inhalable particulate (IP) and FP concentrations at any
location based on other available particulate air quality
data. ' ' One means of comparing different methods in this
previous work was by their standard errors of estimate, S
y,x
The standard error of estimate is simply the standard devia-
tion of the differences between observed (measured) and predicted
values in a paired data set, or the average difference adjusted
for degrees of freedom. In this case, the observed values are
either annual average or 24-hour IP or FP values at background
sites and the predicted values are from the various estimating
methods. In equation form,
V
z(Ym-y
n-k
= Measured concentrations,Mg/m
= Concentrations predicted by estimating method, yg/irT
n = Number of data pairs
k = Degrees of freedom lost (1 if method is a ratio of
values, 2 if method is linear regression equation)
5
-------
To facilitate comparisons between methods, the standard
error of estimate can be expressed as a percent by dividing by
the mean measured value of IP or FP: S /Y . Two separate
tests of the variation in S x with IP and FP levels have both
shown that the errors are proportional to IP and FP values rather
than a constant error for all levels of IP and FP.2'3 Therefore,
the percent standard error values are as representative a measure
of performance as the absolute S x values and are more directly
comparable.
The lower the percent standard error for an estimating meth-
od, the more precise it is. This evaluation procedure appears to
satisfy all the requirements specified aboveit is systematic,
can be used to rank the alternatives, can be used with all meth-
ods, and can be employed with any type of independent data sets.
There is always the danger of "overprediction" when the same
data set (group of background sites) is used to derive the stan-
dard error that was used to establish the predictive relation-
ship. Therefore, other data sets besides the EPA IP network have
been located and used where possible to get an unbiased measure
of standard error for each method. These independent data sets
are described later in this section. They are particularly
important for statistically-derived models to demonstrate that
the models are applicable to different situations (i.e., loca-
tions) than the ones from which they were developed.
Quantifying Predictive Models
As indicated in Section 3, all the predictive models are
relatively simple. For methods which require a constant ratio to
predict the PM1Q or FP background value, the constant can be
obtained either from the slope of -a linear regression analysis or
from the average ratio of individual data pairs. If the con-
stants produced by linear regression and the average ratio have
approximately equal predictive ability (as measured by percent
Sy/x)/ the average ratio is preferable because it is simpler to
calculate and does not contain the y-intercept term.
-------
The IP and FP estimation models evaluated in previous stud-
ies all had equal or better predictive abilities with average
1 °
ratio constants,. ' In the present study, the average ratio was
used almost exclusively to quantify constants for predictive
models.
Stratification of data, usually by geographic area, with
different constants for each subset, has also been used in previ-
ous work to improve a model's overall predictive power. Con-
stants for each subset are obtained in the same manner as above
and the model or method is still easily evaluated by the percent
S procedure. Because of the small size of the data bases for
y ,x
background sites, the only stratification that was performed in
the present study was dividing sites into eastern (E) and western
(W) subsets. This was routinely done in evaluating each estimat-
ing method. The division between eastern and western sites was
generally along the Mississippi River. Stratification by E/W is
based primarily on the distinctly different sulfate and visibil-
ity levels observed in these two areas.
If the percent standard error is not substantially reduced
by stratification or if the constant ratios for the E/W .subsets
are not significantly different, there is no reason to use strat-
ified data. The appropriate statistical test to determine wheth-
er the constants for the two subsets are significantly different
is the t-test for difference between means (two-tailed), in which
the difference in constants must be greater than the appropriate
tabled t-value times the standard deviation of the difference in
means in order to be significant:
RE -
{t
>
w
Adjustments to Ranking
The standard error of estimate provides a readily computa-
ble, systematic procedure for ranking all the background esti-
mating methods. As with most strictly numerical rating systems,
-------
it cannot be easily adapted to accept nonquantitative input.
Therefore, some manual adjustments to the rankings according to
S have been made based on such other considerations as:
y/x
1. Maintaining a consistent order of methods for the four
background values being estimated (ann av PM.n, high
24-h PM1Q, ann av FP, high 24-h FP). 1U
2. The potential for measurement error is large for the
substitute air quality measurement.
3. Predictions depend heavily on the site designation
(e.g., suburban residential) assigned to the data set
used in the method.
4. The method is only applicable in the east or west.
Whenever an adjustment of the rankings reported in Section 4 is
made, it is noted and reasons for the change are given.
DATA BASE
Four different data sets that have at least one year's IP
and FP data from sites with a wide geographic distribution were
identified:
EPA's IP network (5/1979 - 12/1982)
Sulfate Regional Experiment (SURE) sites
Western Particulate Characterization Study
Previously published particulate data compiled in a
1982 PEDCo report
Sampling methods, tine periods, site locations, and other infor-
mation on these four data sets are summarized in Table 2.
Samplers in these networks were originally designed to
produce sample fractions <2.5 pm (FP) and 2.5 to 15 um (coarse
fraction of IP), although subsequent evaluation of the SURE
samplers indicated a wind speed-sensitive upper cut point that
averaged closer to 10 um than 15 pm. However, the point to be
emphasized is that there are no large-scale long-term data sets
with direct measurements of PM1Q. Analyses of the PM,0 to IP
ratio performed to date show a fairly consistent relationship, as
discussed in Section 3. Variations in the results for these
earlier studies can be attributed to the different data bases
-------
oo
1
UJ
oo
^C
1
<£
a
a.
u_
Q
2:
00
re
O
Cu
*_<
^
a.
UJ
I/)
o
»r
4->
OO
r
i-
QJ
4->
CJ
re
s-
re
jr
0
s-
QJ
c^
-M
O
»»
ift
P
3
o
«J I*"*""
r^*.
-t->
^
E
0
QJ
i_
O
QJ
re
4->
to
E O
S- «53-
QJ
4->
CO
QJ
^j
00
00
*""*
E
s- cr>
QJ
4-2
to
re
UJ
QJ
o
r- LO
S <£>
E t 1
O
»^-
4->
03
z.
a££
s-
0
MSX Jj_^
S- QJ
O E
5;
4J E
QJ r-
c~
to
4- QJ
O 4^
^"
E 00
O
r- l«-
4-> O
re
u
0 0
"" .^
re
CM <:
r^ ^
^^
i ^ oo re
E S. -D QJ
r- « E S-
re o. re re
0 E r-
*J r1* i (
* re QJ
QJ E O 00
*-> O *r- in
o -r- > re
E 4-> S_ i
QJ re QJ o
o^ ^y (/) ^^-^
Q
1 E
QJ .r- -SS-
T3 M- O
00
E 4-^ f~~
re re
O"» <^ ^3 E oo
S- QJ O i
3 E "^ QJ
E CD CD >
O !- QJ QJ
Q.
4-J QJ E
3 JT QJ re
JD rsl
o -r- s_ re
«4-J S- O QJ
VO E QJ <4- S_
, 1 fO >,4-> ro
JO I O tO
s_ r re r E
s re s- QJ re
E 3 re > «">
O OO P** QJ SM
2: 3 U i 3
H-
00 O
QJ
4-> E
r- O
OO !-
4->
"& re to
E O QJ
3 O 4->
O i>r-'
S- CO
CD S-
,NI^ ^^ "C3
C-J H C
*t3 n
.Q (Q O
f" t-
l«- S- CD
O QJ ^>£
4-3 CJ
« !- re
o s- jp
^ °
E
QJ
4-> QJ
0- - S
U +J QJ
i <- o ja
O TD J=
> o v> en *
r- * -r- QJ P-~ CM
IXI t""! O E CTi
OO -i^-l
o
3 re
o
r^ IX"5
t- r*^-
re cri
> T 1
« 1
w
o
QJ < I CM
E U- U_ !>»
O CO OO O
^^ 4^
CT>
*^^
j*^.
i r---
QJ t^
D-^^*. i~"
CO O
J= > U.
4-> E -I- CO
'aa o s- co s-
>u_u_ Et-t^ s-o
r- CO CO 1 ~v> O <4-
3: t i to O t-
O i 1 00
X O <=t
i- -i- O CM
4J CM
O CO
-E 4-J O CT>
1 00 (-> < <
O **~ g"
> *O O CT> s- «3-
i- !- r^ QJ CM
' ~T" * Ci CTi ^^
^-H '«-!.£.'
OO QJ
00 >> t_>
re QJ
2: o
a '
o
SM.
QJ .
a.
4->
E CD
QJ . - E J=
E -.,_:.
D. 4-> -
r- re JE .
3 S- 4->
O" QJ CD
QJ a. E
O QJ
CD i
E .*:
i- S- QJ
i a. o t
Q. co a. Q. 3: ex
E t i u. 4-> E
re QJ re
co 2: co
to
3
o
r
S.
re
>
_%x
CD
QJ
^g
^^^
CM
et
2:
>,
re
t
4->
vo
>>
S-?
>
UJ
QJ
^^ Y^
u re
E <
QJ -
cr >
QJ re
s-
M- re
4->
CD re
r
r S-
CL QJ
E ^~
ro 4-»
O O to O
2: 2: QJ z:
>-
o o o o
00 O O O
QJ 2: 2: 2:
>-
to to 01 to
QJ QJ QJ OJ
>->->->-
to
QJ I/I
3 QJ
i/i re !
E > 00
O QJ
r- CO E 4-J
4-> 4-> re re
ro ro JD S-
4-5 3 Br
E i JQ E
QJ re 3
o 3 in >
E "O QJ
O >>4-=
u > jo re
^ S- 4_
x -o re r
re E a> 3
S-
QJ QJ
r (->
JD f
re -r-
re "o
> QJ
re -^
o
4-» re
o 4->
E to
II II
co
ro
-------
10
concentra-
used. There appears to be support in the literature for use of a
corrected IP value as an acceptable measurement of PM
tion.6'7
IP Network
EPA's IP Network data were selected as the primary data set
for several reasons. These data were collected at sites nation-
wide with uniform equipment and procedures, and have been sub-
jected to fairly rigorous validation both internally and by
external users. The single characteristic of the data base that.
sets it apart from the others available is that it contains all
the other simultaneous sampling data needed to test the proposed
estimating methods, i.e., suburban sites in the same regions as
nonurban sites, sulfate and nitrate data, and hivol and SSI data
at the same sites as dichotomous sampler IP data.
Sites in the IP Network were selected as reasonable back-
ground locations based on firsthand knowledge of the sites'
surroundings. Most were classified as rural or remote in their
site descriptions, but a few were listed as suburban residential
because they were located in developed areas. With a few excep-
tions, the 16 sites identified as background sites were not
influenced by local sources but were impacted slightly by their
proximity to an urban area.
The IP Network data extracted for use in the analyses were
days on which both TSP and dichotomous sampler IP and FP concen-
trations were reported. Available data extended from mid 1979
through 1982, but sites and equipment were generally not active
for that entire period. Because of the intermittent nature of
the simultaneous TSP-IP data, a single summary rather than annual
summaries was developed for each site. These total sample sizes
were generally in the range of 30 to 70 samples, consistent with
the number of samples in an annual data set for a continually-
operated site sampling every sixth day. Summary data for the 16
IP Network background sites are presented in Table 3.
These data had already been subjected to an EPA validation
process. The few outliers found are already removed from the
10
-------
TABLE 3. SUMMARY OF ERA'S IP NETWORK DATA9
Site
Mountain Brook, AL
Carefree, AZ
Litchfield, CT
Braidwood, IL
Marshall town, IA
Acadia National
Park, ME
Winnemucca, NV
Angola, NY
Research Triangle
Park 101, NC
Research Triangle
Park 102, NC
Medina, OH
Sauvie Is!., OR
Clint, TX
Seabrook, TX
Reston, VA
Sauvie 2, OR
Arithmetic mean, yq/m
No. of
samples
73
61
36
42
56
30
67
50
37
63
24
65
87
37
94
51
TSP
53.4
45.7
35.5
61.6
47.8
19.7
55.4
36.5
43.6
37.1
48.2
35.3
77.9
54.3
44.4
40.5
Total IP
28.0
23.3
16.2
27.1
31.6
12.7
29.7
26.5
23.7
23.6
31.3
27.6
49.6
33.7
27.5
27.0
FP
19.2
7.0
10.1
16.4
14.5
8.3
8.7
18.0
19.0
15.1
19.9
13.5
13.3
15.6
18.8
14.1
Max 24-h ,
concentration, yq/m
TSP
91.8
98.2
69.7
100.5
123.6
44.6
267.2
84.6
82.0
86.4
80.3
167.7
259.3
87.8
79.3
Total IP
55.0
68.0
67.3
68.6
80.2
38.0
158.5
59.8
40.8
60.3
49.8
76.5
181.1.
73.6
66.0
Not used
FP
37.3
12.1
55.8
38.6
23.0
33.1
21.3
48.1
31.3
32.4
40.6
43.0
31.0
43.1
49.5
Data are representative of samples taken during the time period May 1979-
December 1982. Actual data for each site are a subset of this period.
11
-------
values shown in Table 3. These deletions are the same as those
removed by other users and are based on the criteria IP/TSP >1.5
or TSP/IP >4 while FP/IP >0.6.2
Other researchers have analyzed the IP Network data for re-
lationships between TSP and IP or FP. ' ' The earlier analyses
did not include 1982 data, which only recently were released.
Their data selection processes were also slightly different from
those used in the present study. However, the important observa-
tion in this case is that the conclusions of these
size-fractionated particulate studies may all be generally con-
sistent largely because they are derived from essentially the
same data base. In this analysis, a new subgrouping, rural
background sites, was considered for the first time.
Other Data Sets
The three other IP/FP data sets identified are important as
independent data sets for checking the standard errors of predic-
tive models developed with the IP Network data base. All three
were used for that purpose in testing some of the methods.
Because the other data sets did not have much additional simul-
taneous particulate sampling data such as TSP or SSI values, they
could not be utilized to evaluate all methods.
Summary data for SURE, Western Particulate Characterization
Study, and PEDCo's compilation are presented in Tables 4, 5, and
6, respectively.
12
-------
TABLE 4. SUMMARY DATA FROM NINE SURE SITES
Site
Montague, MA
Scranton, PA
Indian River, DE
Duncan Falls, OH
Rockport, IN
Giles Co. , TN
Ft. Wayne, IN
Research Triangle
Park, NC
Lewisburg, WV
15-month _
arithmetic mean, yg/m
TSP
22
58
35
42
55
43
40
36
30
IP
23
29
29
31
37
29
28
28
22
FP
17
18
19
20
23
19
20
20
16
Max 24-h -
concentration, yg/m
TSP
82
217
92
163
197
93a
101
99
66
IP
92
102
90
76
120
65
84
69
47
FP
68
81
63
47
86
51
61
50
53
a Removed value of 254 yg/m as outlier according to criteria TSP/IP >4/l
while FP-/IP >0.6.
Source: Reference 1, pp. 6-8 and 6-9.
13
-------
1
cn
3.
c_
» u_
-C C
4-> re
T a
CM 4-> C-
CU i i
>- J
F^ s-
o
i
frwH
ce:
UJ t!
H-
f_3 (.
C£ '.
ec
2: > ro
0 1 £
UL> *r
re
«C I 1 O
Q r-^ -r
CV 4-i C-
>- ". CU >-<
2p '' p
s; " 4J
2: f !-
=3 * i-
co j re
Lf5 j
LU
CO
a:
fO
** Q^
1 ->
I !
1/5
«=T t--.
CM CM
IO CM
co r--
1
. Q-
C-5
C£ "
3 QC
^" ^^
z.
**
to -i£
cn s-
c re
i O_
s_
Q. to
CO -
c
-s= 3
tO O
r- S-
U. CQ
CO CO
«3- CO
>- «3-
CT> CM
11
1
:>
h^
21 "
4_9
" S-
cu o
^^ ^^
fO S-
_J -t-
cC
^t
-c re
a. 4->
3 re
2: 2:
vo
CM
t (
VD
O
O
a.
z.
c
r~
re
4->
c
3
O
^'
U
0
ct:
-
a-
LD
CM
T-H
1
2:
*
cu
re
_j
CU
c
f-»
0
o
CU
2:
cn r*^
CM CO
co cn
r-- co
» <
c
CJ
1 #
l_ w»
ID 4J
S-
- C
= C.
P- S_
re -i
4-> CC
c
3
O O
s: o
s_ re
ro 4->
"O i
cu cu
£. ^r
«sr CM
r~- co
r~- Ln
t t
o
~*
-"^
" f
G£ ^*£
3
^r^ ^
t/J
tn DC
s» .^
3 CU
C CU
CO S^
c«?
^.
CU 4->
CX «"
Q. CU.
=2 CC
r~t &"t
f CM
cn r-<
11 r<
r 1
r
«\
Q_
Z.
C E
0 0
5^
" C
cu re
to o
re
ca cu
o
CO CC
un » i
rr «a-
co cn
CO CM
CM ^-1
o
^9»
d^
r
S "
0-
- 2:
4-J 2:
c
O 4->
Q.
S- CU
r- >
CC CU
to .
c o
re o
o on
s_
o
"3 I
r- «a-
CM CM
CO 0
r~. co
O
CJ
w
,_ i
to
~ cu
Q- C
z: 3
i/i
o -o
a £n
re re
< CO
CI
O 4->
>, re
c: cu
re i-
CO CM
<3- CM
Lf2 CO
r-v. LO
11
r^
r^
^"^
s:
>*^
<- cu
B r-
re p
o re
^>
r
i CU
r r
*^~ CD
*"^
o
cn
re <-
CQ 02
LO CO
CM CM
=i- O
U3 CO
r-J 21
a. z.
9* ^
0- 2.
c c
o o
>^ ^*}
j^ j^;
re re
o o
C 0
re ro
t3 U
cn cn
CM CM
O O
cn «s-
<
(
2:
«\
1 JZ
s: o
c
- re
s_ o:
cu
4-J tO
re -
5 r
CU r
3 CU
i T3
CD O
cn «a-
CM «d"
CM CM
1 CO
I 1
r-4
^^
^
2!
2:
z cu
« 4-:
S to
z. re
CJ
M
c re
0 E
c: NI
^3 CU
c
4-' 0
U. 2.
co cn
CO T-(
r*^ co
CM «a-
p 4
CO >-
*g
OT
T5 -
O C_
O Z.
i 0,
cu c
o o
r 1/1
rn s
cu
re cu
i >-
CM CM
CO CO
CO «=9-
r^ C7^
NJ
,^
" S
^
i ^ «^
to S
CU
i- ro
O S-
u_ cu
^
"u *f~
CU 3
p- O-
*!* "^
S- £=
4-> re
cu s-
O_ CJ3
o cn
CO CM
O CO
CM r»-
i t
3
** £*"*
-M OO
t_
O "
Q. CU
- O
«C E
O to
p 3
re C£
M-
3 4J
CQ s:
O CO
tn «a-
CM r-H
LO CM
< 1 r 1
IXJ
M
21
to
3
O
ro
sC
CU
Q. "
r- 2:
a. 2:
C O
O 4->
cn c
i~ O
O i
CM CM
LO «a-
O Lf5
cn co
0 >-
co 3
V »*
f^-^ J_)
3 i-
2: o
Q.
to S-
cu
-a -=c
c
cc s-
cu
CU t3
ro re
_j _i
i
CO
o
CO
2:
_r
^^
«
^
( %
i+-
^:
f ^
re
S
r^.
CO
Ln
CM
^.
^
cu
c
ro
4-
re
4->
cn r~
CO «W
»-l VO
CM CSJ
i i r i
s:
Ct
a.
"Z.
to
J_
CU K
ro
CD
-0 -r-
JD O
V) CQ
ro 4-J
in co
CO CM
O cn
cn r*"^
^
^
2:
z.
cu >-
[ ^
3 "
CQ re
cn
i O
to re
to s-
o ro
U_ CO
CU
^: 3
i- c
re o
Q. 2:
ro ro
E C
O O
I! II
Q- 2:
QJ
cn
3
u-
cu
c£.
CU
re
a.
ro
.( 9pv
p S_
o c o
r- O E
- -t- CU
S 4-> S
ro
i 4-J i
re oo re
sz sz
o s- o
^ CU -r-
4J CD4-)
re c re
z ro z.
az.
u n
u
C£ O-
s co s:
-
14
-------
TABLE 6. SUMMARY DATA FROM PEDCo'S COMPILATION
Site
Anaconda, MT
Dodge Rd., OR
Carus, OR
Lebanon, OR
Corvallis, OR
St. Louis 1, MO
St. Louis 2, MO
St. Louis 3, MO
St. Louis 4, MO
St. Louis 5, MO
St. Louis 6, MO
St. Louis 7, MO
Angola, NY
Sterling Forest, NY
No. of
samples
>200a
NS
NSb
>200a
>200a
>200a
>200a
>200a
60
60
60
60
NS
NS
Arithmetic mean,, yg/m
TSP
52.2
27.2
32.3
43.3
32.3
54
55
53
60
68 '
60
55
45.9
22.5
IP
39.3
14.7
NDC
ND
ND
31
33
29
40
35
34
28
33.0
15.9
FP
15.5
ND
16.2
13.1
15.5
16
17
16
ND
ND
ND
ND
ND
ND
The actual number of samples for this site was not available in the ref-
erence. This value is a conservative estimate based on data in the
reference and other project files.
NS = Not specified,,
c ND = No data. ,
Source: Reference 4,, pp. 42-46.
15
-------
SECTION 3
DESCRIPTION OF ESTIMATING METHODS
There is a theoretical or logical basis for each of the
methods of estimating background described below. None of these
methods were originally proposed as a result of searching a data
set for any possible correlation between predicted and observed
background PM10- Instead, all were originally proposed on the
belief that they have some logical underlying physical basis.
Several methods in addition to the 13 described here were pro-
posed, but judged to be too complicated or no data were available
with which to evaluate them. An example of a method which failed
on both counts is synthesizing a background data set from two or
more sites located on the perimeter of an urban area by using the
concentration from the site that was generally upwind of the
o
urban area on each sampling day.
For each of the methods, a similar format has been employed.
Each subsection describes the basis for the method, how it would
be applied, data requirements, and problems that might be encoun-
tered.
METHOD 1DICHOTOMOUS SAMPLER DATA
Short of measurement with a PM1Q dichotomous sampler, the IP
dichotomous sampler is the most direct measurement of PM,0 and
FP. The main difference in the two samplers is inlet construc-
tion. Limited side-by-side testing of the IP dichot and PM.Q
dichot samplers in the EPA IP Network has shown that the FP
fraction measured by the two samplers is the same. Several
researchers have analyzed these limited data; they have concluded
that there is a fairly consistent relationship between PM,0 and
IP- ' ' ' Values cited in these 1981 references for PM10 range
from about 70 to 90 percent of IP.
16
-------
The same da1;a set has been used in this study to evaluate
this predictive method, which is obviously a shortcoming. These
annual data are sshown in Appendix Table A-1.
If IP dichot data are available at a background site in the
region, they undoubtedly provide a better estimate of background
than any of the other methods described in this section, espe-
cially with the direct measurement of FP.
The basic relationship to be evaluated for this method is of
the form:
PM10 = K(IP)
where PM,Q concentrations can be approximated by a constant
times the measured IP concentrations.
An alternative measurement of IP can be obtained with a high
volume sampler equipped with a size selective inlet (SSI) and
utilizing quartz filters. This device was used in EPA'-s. IP.
Network and the inlet can be modified to measure PM- directly as
with the dichotomous sampler.
METHOD 2TSP TIMES CONSTANT RATIO
Because of the large number of background sites with TSP
data, it would be desirable to estimate PM-n as a constant
10
fraction of TSP. [i.e. with an equation of the form PM
K(TSP) ] . This method has been investigated more than any of the
others. FP and PM. are both size fractions that are included
within the TSP sample, so there should be some correlation be-
tween TSP and either PM or FP data at a given site. However,
particles larger and smaller than about 2 pm are generated by
completely different mechanisms (sources), so there would not
necessarily be a strong correlation at different sites between FP
and TSP levels. .
The constant ratio between FP or PM-0 and TSP must be eval-
uated for both 24-h concentrations and annual averages. In
testing for a relationship between maximum measured 24-h concen-
trations of a size fraction and TSP, several combinations of data
could be used :
17
-------
Max TSP value in data set - Max PM,0 or FP value in data set
Max TSP value in data set - PM., n or FP value on day of max
TSP
10
Max PM1Q or FP value - TSP value on day of max size fraction
In general, the concept from the first of these combinations has
been used in testing. Since the basic thrust of this method is
to calculate the high PM.J. 24-hour value from measured TSP data,
this is the only combination that is practical. Only 6 of the 16
sites evaluated for this method had both maximum measured values
(TSP and PM10 or FP) occurring on the same day.- Consequently,
evaluation of the second combination (max TSP and measured PM,n
or FP) would generally underestimate the high PM-0 or FP value.
Alternatively, with only TSP data it would be impossible for the
user to select the TSP value that would be associated with the
high PM1Q or FP value (third combination). With the first combina-
tion the analyst would merely search the measured TSP data for
the highest measured value and apply the conversion constant to
calculate the high PM-- or FP value.
For PM1Q, the ratio of IP/TSP was calculated for the five
highest measured IP concentrations at each site. In all cases,
the highest measured TSP concentration occurred on one of these
five days. Next, the average IP/TSP ratio was calculated for
each of the five highest days. The mean ratio (K) for a subgroup
of sites was then calculated from the individual site means.
For FP 24-hour concentrations, the ratios between the highest
measured FP and TSP concentrations were calculated for each site.
Then the average ratio (K) was calculated for each subgroup (E
and W) .
The corresponding data sets for FP, IP and TSP were pre-
sented in Tables 3 through 6. Analysis results are shown in
Appendix Table A-2 through A-4.
The major problem with evaluating this and all subsequent
predictive methods is the lack of PM-Q data. Therefore, based on
the results of PM1Q-IP analyses, IP has been used as a surrogate
18
-------
for PM.- in all the evaluations of estimating methods (except
Method 1).
METHOD 3NEARBY SUBURBAN RESIDENTIAL SITE DATA
It may be that the types of localized sources in suburban .
residential areas primarily increase concentrations of large
particle-size particulates, but do not elevate IP or FP substan-
tially above background levels. Thus, this method uses the sub-
urban residential IP and FP concentrations as estimates of back-
ground (i.e. IP background IP suburban). It is evaluated by
comparing the rural background measurements to suburban residen-
tial sites in the same analysis area.
A problem with this method is the dependence of resulting
estimates on proper site designation and on proximity of the site
to large sources of fine or coarse IP-size particulate.
METHOD 4NEARBY SUBURBAN RESIDENTIAL VALUE TIMES TSP ,/
TSP ruraj.
suburban
Although the; activity-related fugitive sources that are dom-
inant in suburban residential areas might not raise FP concentra-
tions much above background, they probably have some impact on
IP. One method to compensate for this impact, using available
data, is to reduce IP concentrations at a suburban site by the
ratio of TSP concentrations at the background and suburban site
[IP background = IP suburban(TSP rurai/TSP suburban^* Obviously,
this method would only be applicable for estimating IP background.
A potential problem is that the method requires three separ-
ate concurrent peirticulate measurements to generate this esti-
mate. Thus, the chance of obtaining an anomalous prediction is
tripled compared with a method th'at relies on only one particu-
late value.
METHOD 5IP SULFATES AND NITRATES AT SUBURBAN RESIDENTIAL SITE
TIMES CONSTANT RATIO
The hypothesis underlying this method is that sulfate and
nitrate concentrations at a suburban residential site are
19 '" .-."'.
-------
probably less likely to be elevated over background levels than
are any of the other particulate measurements. The unknown
factor is whether a consistent ratio of sulfates plus nitrates to
either IP or FP exists at sites in different geographic areas.
This evaluation should determine whether sulfates and nitrates
make up a fairly constant fraction of IP or FP. The basic rela-
tionships to be evaluated are of the form: K = (suburban IP SO. +
NO3)/background IP and K = (suburban FP SO4 + NO,)/background FP.
Insufficient sulfate/nitrate data for specific days that
corresponded with IP and FP data at background sites were availa-
ble in the IP Network to perform this evaluation for max 24-h IP
or FP concentrations. This also indicates a lack of correspon-
dence in the annual data, which injects another source of error
into the estimates.
METHOD 6TSP SULFATES AND NITRATES AT BACKGROUND SITE TIMES
CONSTANT RATIO
Hivols collect large particles and do not fractionate them.
However, most sulfates and nitrates are fine particles and totals
for these two constituents should not be much greater in TSP than
in IP. The major advantage of starting with TSP sulfates and
nitrates is that these samples are available at background sites,
negating the need for assumptions of uniform concentrations over
large parts of the analysis area. The question of whether a
consistent ratio of sulfates plus nitrates (in TSP) to IP or FP
exists at sites in different geographic areas is the same as in
the preceding method. The equations to be evaluated are: K =
(TSP SO4 + NO3 background)/IP background, and K = (TSP SO. +
NO3)/FP background. Also, the lack of sulfate/nitrate data for
specific days precludes the evaluation of this method for max
24-h concentrations.
An additional source of error in this method is sulfate
artifact formation (due to the adsorption of SO and its oxida-
tion to SO ~) on the hivol glass fiber filters.
20
-------
METHOD 7SIZE SELECTIVE INLET (SSI) TIMES CONSTANT RATIO
The hivol with a size-selective inlet (SSI) collects sub
15 um particles without providing an FP size cut. Because of
apparent problems with artifact formation on the glass fiber
hivol filters, IP measured with an SSI is almost always larger
than IP measured with a dichot. To maintain consistency, it is
recommended that SSI samples be taken on quartz filters. For
samples taken on quartz filters the data can be treated the same
as for the dichot.
If IP measurements taken with an SSI on glass fiber filters
are the only available particulate background samples for an
analysis area (an unlikely situation), they could be converted to
equivalent dichot IP measurements by multiplying by a constant
reduction factor [i.e. IP background = K(SSI background)]. This
then becomes an estimating method for IP; it has not been eval-
uated for predicting FP.
The annual factor (K) was calculated for each of 10 sites
based on the average ratio of dichot IP/SSI concentrations for
simultaneous measurements. Then the overall average factor was
calculated as the mean of the 10 individual factors.
The 24-hour factor was calculated as the average ratio of
high dichot IP~. , /SSI_. . for measured values. The same
24h 24n
rationale outlined for Method 2 was followed in the selection of
maximum measured values for this method.
METHOD 8MAX 24-H BACKGROUND VALUE AS CONSTANT RATIO OF COR-
RESPONDING ANNUAL AVERAGE CONCENTRATION . ;
If background concentrations are in fact primarily due to
long-range transport and local natural sources, the lognormal
distributions .of daily values at different sites should have
similar slopes and the ratio between a 98-percentile (maximum
measured) and annual mean value should be fairly consistent from
site to site (i.e. K = high 24-h IP/mean IP). This relationship
has been found for TSP and may be applicable to IP . The average
ratio for use in this model can easily be determined from the 16
background sites in the. IP Network. : ; .''-',.- . '
- ^ " : ' . ' ' ' * ' 21 " , . ''- '" ". "- ' .
-------
This method is only applicable to the prediction of high
24-h background values. Also, since this prediction is based on
a concentration that is already an estimate, it will contain the
error in estimating the underlying ann av value as well as the
error associated with the ratio of max to mean concentration.
METHOD 924-H IP TIMES CONSTANT RATIO
Carrying the potential interrelationships among the back-
ground values to be estimated one step further, the high FP
concentration may well be closely related to the high IP concen-
tration. If the high 24-h IP is predicted by a method not ap-
plicable to FP values (e.g., from SSI data), the ratio between FP
and IP could provide a simple method for obtaining the high 24-h
FP value. The basic relationship to be evaluated is of the form
24-h FP ^ K(24-h IP).
As with the previous method, the average ratio to use in the
model can be obtained from IP Network data. Also, the predicted
FP value would be based on a concentration that is already esti-
mated. This is not likely to be a precise estimating method.
METHOD 10LINEAR FUNCTION OF VISIBILITY (EXTINCTION)
Trijonis has recently published reports which have indicated
high correlation between IP or FP concentrations and concurrent
airport visibility-relative humidity readings. ' Originally,
the relationship could only be used to predict visibility from IP
or FP data. With recent modifications, these predictive equa-
tions now have reasonable capability of estimating IP and FP con-
centrations, both annual average and daily values, from visibili-
ty data. Much of Trijonis1 success to date has been with data
for the eastern United States. However, one publication presents
a map with median annual visual range at suburban/nonurban loca-
tions across the United States for use in estimating background
levels of fine particles. This map is reproduced in Figure 1.
The predictive models (annual average or max 24-h) for the
eastern United States are:
22
-------
to
o-
c
c
TO
J2
i-
3
X)
CO
oi re
is
o
c o>
i- +» IT)
en
o>
na
S_
fO -r-
Q.
00
> O en
t- 3
i *-> ,ct
^3 ^J
Z5 O **
c o
CO
T3
CU
c
o
OJ
o
3
O
GO
-------
IP = 23.8 + 8.7 B
FP= 8.4+7.5B
where B = Extinction coefficient, in 104 m
= 18.75/V, where V is visual range in mi
The standard error values that Trijonis obtained in developing
these equations are presented in Section 4.
METHOD 11LINEAR FUNCTION OF (VISIBILITY + TSP)
In his most recent report, Trijonis has proposed a predic-
tive model based on visibility and TSP data.3 This was found to
produce lower percentage standard errors than either of the two
variables tested alone. The specific equations (for annual
average or high 24-h) are:
IP = -0.7 + 0.55
FP = -4.1 + 0.68
(17 B + 0.6 TSP)
(10 B + 0.3 TSP)
As with the preceding method, these equations were limited by
Trijonis to use in eastern states, but have been evaluated on
both the E and W data sets in Section 4. The equations above
were applied to the 16 IP Network sites using visual range values
from Figure 1.
While testing the W data sets, it was noted that the two
linear regression equations above consistently under-predicted
both IP and FP measured values. Therefore, alternate equations
still of the form IP = a + b (17 B + 0.6 TSP) and FP = a + b (10
B + 0.3 TSP)were derived from the relatively small W data sets
(primary plus independent data). These alternate equations,
presented below, were then evaluated with the W data sets:
IP.. - -2.6 + 0.86 (17 B + 0.6 TSP)
0 + 0.61 (10 B + 0.3 TSP)
3W
'w
The results of evaluating this predictive methods are shown
in Appendix Table A-5 through A-7.
METHOD 12DEFAULT VALUES FOR FP IN THE WEST
After reviewing the FP data in Tables 3, 5, and 6, it became
apparent that background FP concentrations throughout the west
are uniformly low and vary over relatively narrow ranges in large
24
-------
geographic areas. One possible method of estimating background
that has often been suggested is for researchers to provide
single-valued concentrations that would be appropriate for desig-
nated geographic areas. By averaging the arithmetic mean FP
values reported in the above-mentioned three tables and rounding,
the following "default" values were obtained:
6 Great Plains states - 10.0 yg/m3
8 Rocky Mountain states - 4.0 yg/m3
3 Pacific Coast states.' - 13.0 yg/m3
The resulting predictive method is to simply select the
appropriate default FP background value based on the state in
which the analysis area occurs. The method is not expected to
provide as precise an estimate as some of the other methods, but
the systematic evaluation procedure being used in this study
permits it to be ranked with the others for comparative purposes.
Should it prove to be reasonably precise, default values could
also be developed for IP and for high 24-h background concentra-
tions. This method is expected to be less precise than other
methods because the resulting values are constants rather than a
linear regression equation from a line of best fit.
The large difference in background concentrations between
the Rocky.Mountain states and other western states should be
noted. The difference in concentrations is even more striking
when it is between the Western Particulate Characterization Study
(WPCS), as shown in Table 4, and western sites in the IP Network
or other studies. One explanation for this distinct disparity is
that the sampling equipment in the two networks sample different
size ranges or with different efficiencies. The WPCS employed
stacked filter units and the IP Network and other studies used
dichots. Upon further investigation of extensive unpublished
data, differences in samplers were rejected as the primary reason
for the large deviation in ranges of concentrations. Instead,
they are attributed almost entirely to siting criteria: the WPCS
located its samplers in very remote forested areas such as
25
-------
National Wildlife Refuges, while the IP Network and other studies
generally placed background sites in nonurban locations at the
edge of urban areas.
METHOD 13AVERAGE OF VALUES OBTAINED FROM TWO OR MORE OF THE
ABOVE METHODS
In statistics, as more samples are taken from a population
and more sample means become available to estimate the true
population mean, the average of the sample means approaches the
population mean and the standard deviation of the sample means
decreases. Analogously, if two or more independent estimates of
background are averaged, the resulting value should approach the
"true" background for an area.
Some preselection of the methods to be used in averaging has
been done so that a) the averaging procedure is practical in
terms of input requirements and b) methods which have low stan-
dard errors are employed. Based on preliminary analyses results
from the best 2 or 3 methods identified (lowest S x) were merely
added together and averaged. The method results to be added are
summarized as follows:
ann av
IP =1/3 [Method 2 + Method 6 + Method 11]
max 24-h IP = no equation
ann av FP =1/2 [Method 6 + Method 11]
max
24-h FP = 1/3 [Method 2 + Method 8 + Method 9]
Each of the component methods must be completed prior to the
averaging. The analysis of the averaging method is presented in
Tables A-8 through A-10
east.
This method is only applicable in the
26
-------
SECTION 4
RESULTS
The predictive equations for each estimating method were de-
termined as described in Section 2. The resulting equations for
predicting the four background values of interest are presented
in Table 7. In cases where the equations for eastern and western
locations were significantly different (at the 0.05 level), the
two stratified equations are shown. Also, if a similar predic-
tive equation was previously reported in the literature, it is
shown in the table for purposes of comparison.
It is important to note that, with the exception of Method 1
{use of IP dichot data) , all the equations or models estimate IP
rather than FM
ln-
This situation occurs because measured data
are not available to check the predictive abilities of a model :
for PM To convert a predicted annual IP value to PM , it is
recommended that the ratios 0.88 and 0.77 be used in the east and
west/ respectively. Corresponding values for the high 24-h case
are 0.90 (E) and 0.78 (W) . These four ratios, split by time
period and geographic area, agree better with data collected in
1982 in the IP Network than the previously-proposed IP times 0.90
or FP plus 0.80 IP.
The data used to generate most of the equations in Table 7
are not shown in this report. For example, no SSI data, sulfate
and nitrate concentrations, or suburban residential site data are
presented. The primary reason for not including all these data
sets is that different IP and FP data subsets from those shown in
Table 3 were employed in several of these analyses to obtain
correspondence and the required data presentation would have been
unwieldy and confusing. The data and results for Methods 1,2,
11, and 13 have been summarized in Appendix A.
The purpose of this study is to evaluate the Alternative
models in Table 7. This evaluation has been done through the use
' . ' " '''....' "27 ' '.-':.'..''' '',--' ; ' ' ' .. :
-------
0.
u.
o
ry
a:
o.
1 1
a
§
§
8
C3
CD
*"«
H-
i
1/5
LU
C£
O
LL.
in
0
h- 4
Sc
s
LU
LU
1
O
LU
Q-
*
LU
co
I
o.
u"
JZ
O TT
CM
| f-
C SC
g
u
ii
§
s
en
.AC
u
fa
JS Q-
1 ! u.
1? >
s*s:
ii-
t;
i.
o
w o.
c
o .^
0 CM
1 3
a "en
o HE
s
.p-
- ^
c_
1
O-
: 5
^
S
a
|
Predictive i
CU
U.
a
£
3
re
£
o.
u.
ra
Measure
re re
a. o.
0 CO
en r-
O O
UJ3
re re
c. c.
C3 I--
co r-
d o
UJ 3
re
o
0.
o
c.
in
3
re
re
re
a
o
u
a
CM
^
O «
U- ib-
CX. U £>
~~~
o^."o-
*"""""
er> co
cu «
u. O O
~CM*
in «
i-
0
cu _i> "S
o re
o a. a.
»»
crv co
a. -
u. o o
J= -C
1 1
C" *3"
CM CM
o. a.
CO ro
in CM
O O
UJ 3
<4_
at
Q. CuS-
on on
o co £
«T CM
O
O O f>
UJ 2 O
f~
1
CM
O.
1
O
^^ CM
U- (4-
SS.
c. o. a.
on on on
vo vo vo
0 O O
0
£
TO
VI
c
0
u
tn
i
a.
i
CM
C
C
re
tn
re i
on
EC
in i
o.
u.
> >
re re
c c
c c
V) 10
re re
v; aj
on on
__^
or
a.
H"
CC
C.
1
sc ""or
on on
a. a.
01 ^-,
" C
re
tn jo
3
re .a
H 5
c a.
a; on
o I
tn <
s- s.
c s-
re a.
.a on
!- H-
3
J3 C
3 re
v» ja
i_
>> 3
"£ "§
re wi
at a.
" »
'
(*0
p^
tf
LU
o
tn
O
i
cr
m
o
+
of
c-
U3
m
i
c
o o
"O <"
t/i re
QJ Si.
!.
jburban
:onstant
V)
tn
rm-»
O
Z OJ
^
o
on re
i»
in
re
c
re
1 O.
1 U.
o
CM
.»
oo
cJfe*
on on
o-n.
on on
( H-
O 0
CM CO
> 1
« 1
UJ 2
>
t
tr c
CM C
i on o.
on «-.
o co
CTl U?
O CM
^^i,
O O
z: z:
*- *
on on
a. c. i
on on
on
i to on
co to
in
UJ 3 O
o
*> c
i. 2
+* c
c o
re u
tn O i/t
S 5|
en
*- re
O o "*
z u n
+ in j=
aj i
^* E ^~
O * CM O
on 4J >-
Q- §> re
on on --. i_
h- on sr
10 f*. CO
«r «r c
CM CM C
a. a. re
r*. r«* re
do i
uj x on
.
m
(41.
U
ce
co
in
I CO*
UJ
^
re
_
S
i tn
TO
i
on
_^
n
14-
QJ
CO
1
co
t-
CO
. m
UJ
o
"re
^
c ^-
** J3
O ^»
U >
I °
a. Z>
u
c
J= 3
1 ^
CM S-
re
en c
:n _i
en o"
re
c
re
in
re
OJ
on
*
m
n
o
-m o
vo a> co
a.
t IO
T pno"oT
on
LU o 3 )
^
re
_
c
re
re
oo
+ ^*
CO CO
f^ f) r~.
in cj o
in ce co
+ a. -t- o.
on on
i t
O O
UJ 2
^^
_
IX)
O
O
u
c
01 Cu
c on
_j
r^
i
V)
i/) »- in
c re re
o. o u
re >,
i- u re
O o a.
t- C 0
O "~ 0
0 °. m'
i
i
OJ
*
V)
QJ
( re
co t cr
in ^- c
CM re
O o. o.
~«.r» O
-N 10 VO
UJO CM
*
<^
on »^
co ^~-
Cu IO Cu
on .on
i O s
t- C5
O
CM < O
CM CO
~"o"5
UJ Z s-^
1
* 1 1
co on
O H-
us
mo
CU O
on z -t-
t
O on ^->
rn Cu m
w^0'
o
cu
3
r a;
re >
> o
o
in
01 -o
en o
£5
CU CJ
> =
2
o.
o
u
o
o
u
o
01
i.
cu
o
3J
QJ C t/1 to
en (B TO GJ
OJ JD LU ^C
2B
-------
of standard errors of estimate calculated for each model. The
standard errors of estimate, expressed as fractions of the mean
measured value (S /Y), are presented in Table 8. For each
y, x m
applicable method, S v/Ym is reported first for the data set
y ,x m
from which it was developed (usually the IP Network), then for
any independent data set on which it was tested, and finally for
any urban data sets reported in the literature. Also, S /Y is
y,x m
stratified by E and W subsets if separate equations are given for
E and W.
With the exception of Method 1, the standard error values in
Columns 2 through 7 of Table 8 represent only the prediction of
an IP background concentration and do not include the additional
error involved in converting the estimated IP value to PM ..
Similarly, the standard error values for Methods 7 and 8 do not
include a combined error term even though they require one esti-
mated background concentration as input in order to predict
another background value.
GENERAL OBSERVATIONS
As anticipated, the models for predicting annual averages
generally had lower standard errors of estimate than those for
predicting maximum 24-h values. This result follows from the
statistical observation that measures of central tendency can be
estimated more closely than extreme values or single data points
in a distribution.
Also, IP values were estimated more closely than correspond-
ing FP values. It is possible that much of this added error is
associated with measurement problems with FP, especially at
background sites. For many individual samples, the weight of FP
material collected on. the Teflon filter in 24 hours is only a few
times the error level of the gravimetric analysis procedure. The
hydroscopic nature of the FP samples and lack of bonding of the
collected particles to the filter surface further increase the
measurement error. The level of measurement error is indicated
in Table 8 by the standard errors of 0.21 0.17 in side-by-side
29
-------
sampling for FP with IP dicbotomous and PM' dichotomous samplers
(Method 1 Columns 8 and 11). The annual data from which the 0.21
value was calculated can be found in Appendix Table A-l. Some
portion of the 0.21 value can be attributed to the different
sample sizes used in the analysis for the two samplers, as evi-
denced by the lower value obtained for the high 24-h IP analysis
using paired data (0.17).
Although these two samplers have the same particle size
outpoint for FP, significant sampling imprecision does exist for
these small particles and related low concentrations. It is
probable that collocated samplers of either type would show
equivalent imprecision for the FP measurements. This is not an
indictment of either sampler or the basic sampling mechanism.
Rather, the combination of potential handling problems and the
small amount of collected mass for the FP fraction (near the
laboratory measurement threshold) require that analyses of FP
data also consider the precision of the measurements.
In contrast, the standard errors for PM._ are only 0.12, for
annual data and 0.07 for the high 24-h value, including the error
due to variability in
...
to IP ratios from site to site (also
Method 1) .' A very important observation here is that several of
the predictive methods for IP have a S /Y in the range of 0.10
y ,x m
to 0.15, or about the same as the measurement error.
The standard errors of estimate reported in the literature
appear to be about the same as the standard errors determined in
this study. This result is partly due to the fact that IP Net-
work data have been the basis for all the analyses. However, the
prior studies used mostly urban data, while the present study
employed primarily nonurban data. ' '
The models that had significantly different equations for
eastern and western sites all had lower standard errors at the
eastern sites. This result is hypothesized to be due to the
different major contributing sources to background particulate
concentrations in the east and west. Eastern background is
presumed to be related to uniform, predictable variables such as
30
-------
co
o
o
3C
O
UJ
o:
UJ
I
II
I
UJ
u_
o
oo
ce:
o
a:
UJ
o
a:
o
eC
00
CO
UJ
CO
X
c
QJ
t-
QJ
CL
f estimate
l-
O
1-
fU
o
L.
C
4-»
V)
U-
CM
JZ
cn
O.
u_
C
C
CM
X
O-
ro
c
c
jr 3
O vi
1 C
QJ CD ro
"O -O 4->
C c ro
QJ XJ
CL
ro ro
^ "^
I- "O
CL.
t- >>
QJ X)
4-> 4->
O *n
M
1 C
QJ QJ ro
c n ro
r-t QJ ~O
CL
ro ro
E +>
i- "O
Cu
QJ-01
JT 3
4-* 4->
O *n
4-J
i C
c c ro
cu -D
Cv
E -5
- ro
i- -o
QJ -0
o
1 C
T3 X> **
t QJ "O
Gi-
ro *->
*» XI
i-
O-
TD
O
i
QJ
U
QJ
in ' *5" *3"
i ^r . i i i i i i i .... i i
. M- O H- O **-
°CC UJC? UJ<£
ro co CO
«- CM -*
I .{ 1 1 1 1 .11 III
o o o
UJ
l-s, O"v O CO **^ ""* ^* ""* ^~
f-(C\jcrtf^ rocsj*£> c\im
. . . i t i i . . i iii >
o oo o o oo oo
UJ 3: UJ 3c UJ H
o to ** *
icocvi I i i * i i * * ! ' *
* O **- O *-
O O QJ CD
uj oc uj ce
vo f*^ oo co
1 .1 1 1 1 1 1 1 1 *>ll
o ooo
LU DC UJ
JO
C\jCMf-~ ^" «-Hfn r-iCMCM^-
. . . | . ..| | | I .... I
OOO O OO OOO O
uj 3 UJ3: 3: 3
11 o^ CM ro o^ ro
ro CM ^3- CM
t . . 1 1 1 1 1 1 1 .... 1 1
o o . o **- o n- t
. O CM
O CM in cr> CM ro
i i > i i ii.i i
0000 00
io oo m fo CM co
I . l^T , | | | I . <4- I I .... | II
OQJ OQJ ^»*-OM~
cr cc h QJ o>
f*^. . , O fO
- i .1 i r i it i .11 >.
0 0 0
UJ LU
CM iO O ro CO O ^^ f*-- CO f^-' O
T t r-< 10 m "^ *~< to < <" ro CNJ
. . . t i i ... i t
O'OOO.O OOO OOO
UJ 3 uj 3:^c
c: '
ro ic
O i C I o*n*J t i- T-
^-r'ro-O i-E u ro-r- O-^- > * O> ' o
in at ro 3 Qj4->> in w- s- QJ ro >
3 c t- t- inoi EC ro cu o o >*no
1-3 ro 3 4j.i ro .E »- cuja
o *-> c t- *J -i- o" 4-» *-> c o »- c cro**- 3ro
ro *n eo D- * - toin-f- me «-*-> o 'O -~» a> O »
c .Q't/ic -M ro c ro*-1 *-l**-o_c rowi
o i- l ro roi roOOO rod. *> 4-*tot-cu >t3
ro U 3 * J3 O ro i- SE U lit.* O (Jt^-Cn O,
4-) j2QJCt-Z«r- *J ! C C J C - **- JC
rom in 34-» «D3 4->4-» -fra in j=*Jj=:33-4-Sro O4-»
T3OQJ *n »- JSJD^CC l_QJ ICI H- V-QjI- QJ
*->E *- >> 3 c>*-* *r- CM+J CMO t-4-» i-*J J Cn
j^ u'ro 3Q. to vt c re jrc J=*J QJ^QJ.^-^ t-in i~E
U' o O. (D « QJ O O- *-> t * . Oi o CT) ro C *i C *f~ "^ i- QJ QJ O
O ( 3T<-« H- - «/J 3! . 3! -J _JZ 4J
O t- 10 QJ
J- ID UJ 2
I- C
UJ «- H II
(O J3 UJ 2
3T
-------
sulfates, transport, and visibility. Western background is
possibly more a function of erratic, specific-event variables
such as wind storms, biological debris, large-scale burning, and
agricultural activity.
OBSERVATIONS BY METHOD
Method 1Dichotomous Sampler Data
The standard errors for predicting PM . and FP from IP
dichotomous data are about the lowest of any method. It is
therefore the preferred method for both size fractions.
Method 2--TSP Times Constant Ratio
TSP has been shown to be a highly variable estimator of IP
1 2
for an individual site or a single measurement. This analysis
and two earlier studies have defined consistent relationships
between mean annual TSP and IP data for large data sets.1'2 In
fact, this study has found an identical annual relationship for
nonurban sites as was found for urban sites. However, as the
data in Table 8 shows, the standard errors of estimate for 24-
hour data are greater than those for annual data. Further, the
standard errors of estimate for 24-hour data from these nonurban
sites are lower than those found for other groups of sites. Both
of these observations demonstrate the variability inherent in
estimating IP concentrations from TSP data.
Methods 3, 4, and 5Using Suburban Residential Sites
All three of the methods that utilize data from suburban
residential sites have poor standard errors. Apparently, there
is no close relationship between particulate concentrations
measured at suburban and nonurban sites in the same region.
Method 6TSP Sulfates and Nitrates at Background Sites Times
Constant Ratio
At eastern sites where sulfate concentrations are higher
than in the west, the sulfate plus nitrate concentrations mea-
sured at a background site are a very good indicator of either IP
or FP concentrations at that site. Because sulfates and nitrates
are not determined for every sample, this method is not applica-
ble for estimating maximum 24-h background concentrations. Also,
32
-------
the IP Network samples had consistent handling and analysis
procedures, which is very important to the reproducibility of
sulfate and nitrate measurements. If this consistency is not
present in a local data set or different analytical procedures
than those of the IP Network are used, the predictive ability of
this method would be expected to decline.
Due to the small size of most data sets for TSP sulfates and
nitrates, this method is an excellent candidate for averaging
with the results of other predictive methods for determining
background values (see Method 13).
Method 7Size Selective Inlet Times Constant Ratio
Although IP measured with a dichot and an SSI should be the
same concentration, analysis of side-by-side measurements has
shown that the average ratio is only 0.75 when glass fiber fil-
ters are used in the SSI. However, that ratio is consistent
enough that it can be used to predict IP from the SSI concen-
tration. The S x/Ym values of 0.17 and 0.20 for ann av and max
24-h estimates make it a mid-level ranked method. Qualitatively,
the method is ranked lower because recent measurements and future
measurements taken with an SSI should utilize quartz filters,
making this conversion unnecessary. No attempt was made to test
SSI concentrations as a predictor of FP.
Method 824-h Background Value as Constant Ratio of Corres-
ponding Annual Average Concentration
The standard error values for estimating the high 24-h con-
centration as a constant ratio of the corresponding ann av con-
centration are marginal when compared with the other values in
Table 8. These values do not include any error in deriving the
estimate of ann av background values, so the combined error would
result in unacceptably large total standard errors.
Method 924-h IP Times Constant Ratio
Similarly to Method 8, this method of predicting high 24-h
FP from high 24-h IP relies on one estimated value to derive
another. For the same reason as above, the standard error values
are unacceptably large.
33
-------
Method 10Linear Function of Visibility (Extinction)
Using the relationship between IP or FP and visibility
developed by Trijonis, the method was reported to have fairly
good predictive ability for ann av IP and FP (S /Y =0.25 and
y ,x m _
0.19), but not good ability for high 24-h values (S /Y = 0.42
y r x in
and 0.41). One advantage of this method for "backup" use is that
background concentrations can be calculated for any desired
location based on the visibility data in Figure 1.
Method 11Linear Function of (Visibility + TSP)
This recently-developed model , a combination of Methods 2
and 10, produces lower standard errors for both IP and FP back-
ground estimates than either of the component methods. TSP
values for background are available for almost every region of
the country and an average annual nonurban visibility estimate
can be obtained for any location from Figure 1; therefore, this
method can be employed virtually anywhere. If first order NWS
airport visibility data at a nearby nonurban location are avail-
able for specific dates of interest, this method could also be
used to estimate max 24-h IP and FP values.
Because of its better predictive ability, this method should
supersede Method 10 except in rare instances. However, it should
probably be ranked below Method 2 for IP.
Separate equations were used to improve this method's pre-
dictive capability at western sites. The amount of data avail-
able to derive these equations was relatively small. Therefore,
if additional data become available in the near future, the
coefficients in the equations should be requantified. This
modification should be a high priority item because only one
other method (dichot data) appears to be acceptable for predict-
ing ann av FP in the west.
Method 12Default Values for FP in the West
This preliminary attempt to develop "default" background
values almost comparable in precision to direct measurements did
not prove successful. However, further work in this area is also
encouraged. For example, stratification of background sites into
34
-------
subsets of remote and edge of urban area with separate values for
each would probably greatly decrease resulting standard error
values.
Method 13Average of Values Obtained from Two or More of the
Above Methods
This concept of averaging values from some of the marginal
methods above to obtain a better estimate of background is not as
simple as the others proposed. Even though the standard error
values indicate that it may be the most precise method other than
direct dichot measurements, questions still exist as to its
applicability and to the analysis results. The primary question
concerns the use of sulfate and nitrate data in the method,
because of the problems of artifact formation discussed previous-
ly. For this reason, the method loses some of its high ranking
on a qualitative basis. Also, it is only applicable in the east.
35
-------
SECTION 5
CONCLUSIONS AND RECOMMENDATIONS
RANKING OF PREDICTIVE METHODS
Based on the standard error values reported in Table 8 and
nonguantitative information discussed in Section 4, the various
methods for predicting background concentrations were ranked
relative to one another. The methods with the lowest standard
errors of estimates are shown in Table 9.
The methods listed in Table 9 would not all be described as
acceptable. For example, neither the size-selective inlet using
glass fiber filters nor TSP sulfate and nitrate methods would be
recommended because of continuing concerns about their sampling
techniques. The visibility relationship with ann av FP would, in
most cases, be superseded by the better-fitting visibility-plus-
TSP relationship.
A few other methods that ranked high enough to be included
in Table 9 probably need further verification. In particular,
the method of averaging the results of other methods may be
considered spurious by some users. No data were available to
test the visibility-plus-TSP method for high 24-h predictive
ability, so this ranking is based solely on a published refer-
ence.
At this time, only three of the methods evaluated can be
recommended without reservation;
1. IP or PM sampler measurements at an acceptable
background site
2
3,
TSP data times a constant ratio to estimate PM
Linear function of visibility-plus-TSP to estimate ann
10
av
.
and FP.
36
-------
to
o
r-
a;
D.
U.
>
TO
K
^c
Q.
t-H
JK
t
"^1"
CM
j;:
Cyj
r-
n:
a.
^H
5*
(W
c:
c:
"fl»
T3
O
-e o
QJ
to
ro
QJ
-M
o
cu
S-
r~
O
0)
s_
3
to
ro
o
E
+j
U
cu
r-
o
CL.
I"«H
f^
-2
-o
«3
* 4M
ro
"f*
.
o
>
ro
c
c
ro
CL.
to
H-
t i
VD
*
CD
CM
J=
1
^^
CM
Q.
(
CO
in
o
CO
on
o'
CO
vo
CD
+
t 1
^t"
1
4
1^%.
« 1
v^^*
LT>
LO
O
4
r^»
*
1
4-
CQ
^>.
? H
^H^r
uo
M^
o
4
r^.
O
1
I 1
o
VD
CM
+
f~
1
CM
Cu.
r>-
VD
O
4-
CL.
GO
h-
ro
4-i
(/)
ro
a>
c
^~
-~- t
ro
£Z
E
ro
f^
u.
i i
CO Cu
GO *
4-> ( 1
4 in >
O) *-
to
4 ro 4
QJ
co «a- 1->
c o to
O T- GO TO
* CQ 3^ CD fO ** QJ
-f-
ca
,o
kHV*
t/5
TO
(U
C
r~
^-^b
_^~
1
f^-
CM
CL.
GO
^.
VD
4->
to
re
QJ
C
r
^-~v
CL.
GO
(
VD
*
CM O
4-> CD C * s^
tO r-H Br » 1
ro »-< L~J -F
OD --^
i^-* D- «tf-
»I » J»w * £/*)
0 0
4-to
O)
CQ , 3S
t>» C
rrH r"'
-^ ^.
^MB^.
VO c~
00 1
O CM
4 a.
GO
VD ^~
CM VD
« < r^.
CO
4- 0
T 1 1
CO
4* ^""^
t- 4J CO
!< Q. O
a> to 2:
VO ^ '*"'
03 C r-l "*"
* "F~ LJ[^l ^^"
O O
4- oT i i °°
GO . CL.
VO 1 PO GO
CO I
CM VD »
1 0
ro
'.. i i
1 i O_
> GO C
00 CL. 1 !-
VD GO *
. | --^
O O CO
CO CM O
t t
_g~
1
^tf"
CM
^*4
GO
GO
o
- ".- o
vo "co
* ^^
o z.
4- 4-
ca to «*
TO O
f^ QJ GO
r f
~-^ C Cu
t- l i GO 4->
Lf> GO 1 CO
Lf5 1 l GO > TO
OJ.
O CL. LT3 r-l
GO r~ co c
4 1 -
o <
r^ VD
37
-------
No method other than direct measurement was found that could
be recommended for estimating FP high 24-h. To provide some
means of obtaining an estimate, the averaging method (Method 13)
is the best available for eastern areas. FP measurements are
primarily helpful in receptor modeling and for estimating the
long range transport component of measured data.
The methods shown in Table 9 other than those recommended
should only be used under special circumstances, and should be
accompanied by a detailed justification.
PROCEDURE FOR ADJUSTING BACKGROUND CONCENTRATIONS
The few number of acceptable methods simplifies the adjust-
ment process, which is outlined below. This process assumes an
appropriate TSP or IP surrogate background value has been obtained
and that value must be used in the equations. Determining which
TSP or IP data to use was beyond the scope of this report.
1. If IP or PM-. data are available for a background site
in or near tne analysis area, they are unquestionably
the first choice. As indicated in the introduction to
this report, there are very few IP or
located at background sites.
..-
samplers
PM
10
ann av =0.88 measured -IP ann av in east
= 0.77 measured IP ann av in west
high 24-h = 0.90 measured IP high 24-h in east
- 0.78 measured IP high 24-h in west
= measured FP ann av
= measured FP high 24-h
FP ann av
FP high 24-h
For estimating PM.Q background, TSP background site.
data are the second choice. TSP data (alone) are not
recommended for estimating FP.
PM-0 ann av
= 0.88 (0.61) TSP ann av in east
= 0.77 (0.61) TSP ann av in west
PM n high 24-h = 0.90 (0.75) TSP high 24-h in east
1U = 0.78 (0.75) TSP high 24-h in west
The second choice for FP ann av and third for PM-_ ^^
linear equation with visibility and TSP terms (Method
11). This method can only be used to estimate high
is a
38
-------
4.
24-h PM-_ if appropriate nonurban visibility data for
days corresponding to high TSP measurements can be
obtained for the analysis area.
This method requires TSP input but has larger
errors than just using TSP to estimate PM.-. There-
fore, the only situation in which it is anticipated
that visibility-plus-TSP would be utilized to estimate
PM.» background is if the TSP background value for an
area is outside the normally-accepted range of TSP
levels for that region of the U.S., and some means of
adjusting this questionable value is desired.
PM,Q ann av
-0.7 +0.55 (17 B + 0.6 TSP ann av)
in east
-2.6 + 0.86 (17 B + 0.6 TSP ann av)
in west
PM n high 24-h = -0.7 + 0.55 (17 B + 0.6 TSP high
1U 24-h) in east
= -2.6 +0.86 (17 B + 0.6 TSP high
24-h) in west
FP arm av
= -4.1 + 0.68 (10 B + 0.3 TSP ann av)
in east
= 0 + 0.61 (10 B + 0.3 TSP ann av) in
west
Can only be used if first order NWS airport visibility
data at a nearby nonurban location are available for
same dates as TSP high 24-h values.
In these equations B = 18.75/V, where V = visual
range in miles, obtained from Figure 1 or appropriate
airport visibility records.
The final choice is only recommended provisionally, to
provide some method other than direct measurement to
estimate FP high 24-h in eastern areas. It should not
be used for western locations.
FP high 24-h = 0.33 (0.58 TSP high 24-h +0.67 IP high
24-h + 2.60 FP ann av) in east
It should be noted that additional PM-_0 data that could be
used to refine the preceding analyses are rapidly becoming avail-
able. The EPA document entitled "Procedures for Estimating
Probability of Nonattainment of a PM NAAQS Using Total Suspended
J-« i ^ . ~ - '
Particulates or Inhalable Particulate Data
,,12
will be revised in
39
-------
early 1985 based on this additional PM-. data.
is currently based on very limited
superseded by the revised document.
Method 1, which
.- data, may be modified or
OTHER CONSIDERATIONS FOR DETERMINATION OF PM. BACKGROUND
This report has presented several means whereby PM back-
ground concentrations can be determined from other surrogate
measured data. The implication of these techniques is that the
estimated PM10 concentrations will then be treated as if they
were measured data. However, no guidance has been included with
which to address other considerations for background determina-
tions. As the scope of this study was to provide ways to obtain
these surrogate data, such guidance is beyond the scope of the
present discussion. Therefore, the following discussion is
restricted to identification of rather than resolution of the
other key considerations that need to be addressed in the SIP
process.
Background air quality in an area includes pollutant con-
centrations from a variety of sources including 1) natural
sources; 2) nearby anthropogenic sources; and 3) unidentified
sources . Additional' subgroups of these would include: primary
vs. secondary emissions; and distinctions between localized and
long range sources.
One of the major difficulties in any air quality analysis is
to define representative background data for an area in which
existing measured data includes the impacts from these diversi-
fied sources. These analyses are generally restricted to well-
defined areas and require that background concentrations be
established for several time periods: annual, seasonal, and
24-hour. In all cases selection of representative background
data may be a complicated process. In remote areas there may be
no monitoring data to choose from, while near urban areas a
multitude of data may be available. The analyst must define a
representative site in the study area or create a set of compos-
ite data from several sites located within the study area.
40
-------
Historical data could be evaluated to find a similar time
period during which no source impact occurred. These data could
then be used for background. In the case of 24-hour analyses
certain meteorological criteria could be defined and air quality
measurements taken during these conditions used as background.
A number of source apportionment techniques are available
that can be used alone or in conjunction with the analyses out-
lined above. These techniques are used to resolve measured data
into its various components. They include: chemical and micro-
scopic analyses, use of FP data trajectory analyses, and mass
balance and multivariate analyses.
The methods described herein allow the analyst to obtain
estimates of PM.- data; however, the other considerations de-
scribed above must be resolved in order to make adequate use of
the results.
RECOMMENDATIONS FOR ADDITIONAL WORK
The major deficiency in the procedure for estimating PM n
and FP background concentrations is a method for estimating FP
high 24-h values. The most promising prospect is a linear equa-
tion with visibility and TSP data, but no individual day visi-
bility data were available during the present study with which to
derive such an equation.
An omnipresent problem in performing this study was the
small number of background sites nationwide for .which size-frac-
tionated particulate data could be obtained. As more such data
become available, it may be desirable to redo the analyses and
develop revised equations and rankings.
Other recommendations for additional work that would be
logical extensions of the present study include: " . .
1. Develop a relatively easy procedure for estimating the
split of any background value into components of long-
range transport and regionally-generated particulate.
41
-------
Prepare default maps of the United States that provide
background values for any location. It appears that
the best basis for these maps would be visibility-
plus-TSP-relationship that has shown good predictive
capability in its testing to date.
Develop for presentation with the estimating methods
some information on the se
FP background in the east.
some information on the seasonal variations of PJML - and
42
-------
REFERENCES
1. Watson, J. G., et al. Analysis of Inhalable and Fine Par-
ticulate Matter Measurements. EPA-450/4-81-035, December
1981.
2. Trijonis, J., and M. Davis. Development and Application of
Methods for Estimating Inhalable and Fine Particle Concen-
trations from Routine Hi-Vol Data. Prepared for California
Air Resources Board Under Contract AO-076-32. December
1981.
3. Trijonis, J. Analysis of Particulate Matter Concentrations
and Visibility in the Eastern U.S. (prepublication draft).
Prepared for U.S. Environmental Protection Agency, Research
Triangle Park, North Carolina. August 1983.
4. PEDCo Environmental, Inc. Compilation of Ambient Particu-
late Matter Size and Composition Data. Prepared for U.S.
Environmental Protection Agency under Contract 68-02-3512,
Work Assignment 10. January 1982.
5. Pace, Thompson G. Memorandum to Henry Thomas entitled Re-
view of the Relationships of IPin/ lpic' and TSP. U.S.
Environmental Protection Agency, Research Triangle Park,
North Carolina. July 6, 1981.
6. Rodes, C.E., and E. Gardner Evans "Preliminary Assessment of
the PM1Q Data from Eight Locations in the U.S." Draft
Report. U.S. Environmental Protection Agency, Research
Triangle Park, NC 27711. 1984
7. A. D. Thrall and A. B. Hudischewskyj, "An Update on the Use
of Particulate Ratios to Assess Likely PM.- Attainment
Status," technical memorandum, Systems Applications, Incor-
porated, January 13,1984.
8. Anderson, Michael K., et al. Receptor Model Technical
Series, Volume V, Methods for Combining the Various Source
Apportionment Approaches. Prepared for U.S. Environmental
Protection Agency under Contract 68-02-3514, Work Assign-
ments 21 and 34.
9. Lioy, P. J. Letter to Thompson Pace dated May 27, 1981.
43
-------
10. R. I. Larsen. A New Mathematical Model of Air Pollution
Concentration Averaging Time and Frequency. Journal of the
Air Pollution Control Association. 19:24-30. January 1969,
11. Trijonis, J. Existing and Natural Background Levels of Vis-
ibility and Fine Particles in the Rural East. EPA-450/4-
81-036, August 1981.
12. T. G. Pace, N. H. Frank. Procedures for Estimating Proba-
bility of Nonattainment of a PM-, A NAAQS Using Total Sus-
pended Particulate or Inhalable Particulate Data. Draft.
U.S. Environmental Protection Agency, Research Triangle
Park, NC. 1984.
13. Guideline on Air Quality Models (Revised) Draft. U.S.
Environmental Protection Agency, Research Triangle Park, NC
1984.
44
-------
APPENDIX A
CALCULATIONS FOR SELECTED
BACKGROUND ESTIMATING METHODS
45
-------
This appendix presents the results of the analyses for the
four highest rated methods (lowest S )Method 1, Method 2,
y/x
Method 11, and Method 13. The primary reason for showing the
results for these four methods only is that different IP and FP
data subsets were employed in the other analyses to obtain cor-
respondence and the required data presentation would have been
unwieldy and confusing to the reader. Consequently, it was
decided that only data for the best methods identified would be
included.
Table A-l presents the results of the annual analyses for
Method 1. The first step of the analysis used measured PM . and
IP data to determine the relationship between the two measures.
Using one year of monitoring data for each of the six sites, the
ratio, PM.n/IP, was calculated for the annual means. Then the 6
sites were split into East (Birmingham, Buffalo, Philadelphia)
and West (Phoenix, Rubidoux, Houston). The three ratios deter-
mined for each subgroup were averaged to obtain 0.88 for the
three eastern sites and 0.77 for the western sites.
Because no independent data set was available to evaluate
the developed relationships, the average for the first two quar-
ters of data was obtained for each site for both the PM.n and IP
samplers. Both the total concentration and the FP concentration
were obtained for each sampler at the sites. Then the 0.88 and
0.77 ratios were applied to measured IP at the sites to calculate
an estimated PM..- value.
The differences between the estimated PM.n values and the
measured IP values were calculated as shown in the last column of
Table A-l. In a similar manner the differences between measure-
ments of FP were calculated as shown in the fifth column of the
table. Finally, the standard errors of estimate were calculated
for the two measurements of FP and the two PM, values at each
10
sites (measured and estimated).
46
-------
to
oo
rs
o
o
o
OO
at.
o
i<
Q-
U_
C5
OO
II
oo
^ ^£
C
cu
o
o
r_5
CU
o
cu
s-
O)
4
4-
O
Q_
U_
.
>
«=C
to
it- QJ
O r-'
Q.
: E
O 13
= UO
S-
QJ
ex
E
ro
OO
^
to. -
CO ro
s- cu
ro S-
Oi ro
c
** * C
CO
o; -o
4-> S-
" ' - *r* Z5
°°
^"
^J"
1
LO
»-H
UD
O^ O^
LO CTi
vo ^o
o
VD
1
^ «a-
LT> CTl
CO CM
VD CTi
< 1 CM
O
} [
s: Q.
Q. I i
E
ro
.e
CT>
(~
r
gr
s^
t
ca
VD CM vo r- 1 tn
^i" CM O T ( CM
+ «-< + 4- +
I
co «a- vo i~- «a-
-
r 1 CO Lf) VD LO
«=t- r^ «i- co «*
f^-VD VDCO OOO VDf~ CTiCTi CO
VDCO LDLO irsr^H in r-> CMCO i i
CO Uf> CO CTV *S" Lf> CO "51" ^" LT5 LO
II
0
a.
«=1- CM CTi CO CO
LO CO 0 0 0
+ 1 1 1 i-H
H- .
ror^. CTII-. COCTI OCM CDCOO
COCO LOCM COCM LO«*COCO CTl
t < i 1 LOLO CMCM CMCM « 1 CM CM
II
- ex.
u_
COCO OCM VDCO CMvo CTiVD
CMCM CMCM i-HCM r-Hi I
O O O O O
2: o- : s: a. sTn. sTo. sTcv
Q_ i i CX. > i O- < O- * < D- i <
ro
^*
Q.
X i
X . =3 0 QJ C
C X3 ro ro 4->
QJ T- <4- i U)
O JD <4- -i-3
-C -3 3 -C O
Q- C£ CQ D- 3: .
'
CO
CU
C
r
Q-
i i
O
<*.
to
ro
O)
C
CX
» 1
CO
CO
o
II
'C
1
2:
n_
ro
47
-------
1
o
1 t
s:
c.
s_
o
u_
00
1 t
f*^
cy>
1 (
00
T <
p**.
cri
t i
CO CO
in CM
< <
O IT)
^fc.^^^
^^^
II II II II
MX O
W ^> «£Z
C/} Q.
CM
CM
r 1
O
II
X
t/15*1
o
t <
s:
Q_
I
O)
4J
d;
4-
O
in
s_
o
s_
i-
QJ
T3
1
1
Q.
U-
S-
o
u_
^3-
cy>
*
CM
CO
«
er>
CM
CO
r t
Lf>
un o
vo
^~
II II II
N X
o -
w >>
cr>
0 0
CM
CTl
CM O
II II
Q. X
U_ *
._>
(D
o
c
48
-------
For the high 24-h PM10 concentrations the ratio of m.n
was calculated for the five highest measured PM10 concentrations
at each of the six sites listed in Table A-l.. In all cases, the
highest measured IP concentration occurred on only one of these
five days. Next, the average PM _/IP ratio was calculated for
the E and W sites. The average ratios were calculated to be PM. »
=0.90 (IP) for the east and 0.78 (IP) for the west. The corres-
ponding ratios were then applied to the highest measured IP value
at each site for comparison to the highest measured PM value at
each site. Finally, the standard error of estimate, 0.07, was
calculated from the differences between highest measured values
and the mean of the -highest measured IP values.
The difference between the highest measurements of FP at
each site (IP dichot and PM.. dichot) and the mean of the highest
measured values from the IP dichot were used to calculate a
standard error of 0.17 for the two measurements of FP.
Table A-2 presents the results of the Method 2 analysis for
annual data and the primary data set (IP network data). In the
first step of the analysis the average IP/TSP ratio was calcu-
lated as the mean of the daily ratios for each site. For the 16
sites the average ratios varied from 0.46 to 0.78. Next, the
mean ratio and.standard deviation were calculated for the E and W
sites. Application of the t-test showed that there was no sign-
ificant difference between the average ratio for the two groups
(0.594 E and 0.640 W) . Consequently, the average ratio (0.61)
calculated for all sites was used as the multiplier for TSP to
obtain estimated IP concentrations. The mean TSP values for the
IP network data were multiplied by 0.61 to obtain the values in
Column 4. Finally, the standard error of estimate was calculated
for the sites as shown in the table.
In a similar manner the average FP/TSP ratio was calculated
for each site. In this case, the resulting constants (0.40 E and
0.28 W) were shown to be significantly different by the t-test.
Consequently, the-different constants applied to the separate E
and W TSP means to obtain the estimated FP concentrations shown
49
-------
1
0
14
cc
I
00
c
to
LU
1
oo
l t
LU
_J OO
§2
u. >-
o rr
oof
to C£
>- Q-
ZL
1
1
CM
O
O
H-
LU
s:
CM
1
LU
_J
ca
1
n
CD
a.
c
*->
re
c
cu
u
o
o
«=c
0 QJ M-'
J2 3 T-
CU LL.
II
0-0-0.
u_ oo to
co -sr CM
LU
o o
cu
3 t-'
0 !-
V) t3
*
CO > '
II
0.
1 i r 1
to
*J O 1
CO
LU
LO
M CU
O
Q.
E
O re
3
i.
O
LU
r"
^0-HLOLOvoror^^ r^ ro vo r^ ro to o,
CM«3-OOOCO< IOO<-H LOr-l«£)COCOOCM
CMi <'3-oO'-LOcnco r-. «-. co ro m LO «s- *
OO ^^ LQ CO ^^ ^^ CO ^^ ^^ f^^ C^ PO CO ^O ^t*^ ^^
^^ *5j" c^ r**»» ^tf* r**> ^tf C3*t f*^» CM oo tr^ cr^ ^^^ uo ^~*
10 LO un r>. CM w 0 c^ r**^ cr^ oo P*** r*^.
cvji io\j» tojcvjcMoo CM c\jooc\ioj«d-ooc\j csj
ys f**^ t-f^ i^'^ oo m^ ^D "^5* <"""( o^ CM co i^^ LO T"^ r*1^
COCMCO' ICMCMCMCMCM CMCMC"OCM«a-fOCM
r^* co *2f oo LIO co *.o CM cy> ^o LO vo ^o co co LO
LULULULULULULULULU 3333333
-^T3 * re V
O P T3 +-> (J re
OCUO CU . U ^JiCMS-
i- *r O » * CM CU ^~" 3 O CU
-C"O'i-i i IrHCO, ^-JZCU-f--^!-'!- «C
o-' T3O -i 4-» cuinc:>c:jD>
c: +-> re re cno_ O.-DOO s-s-c3T-re3
4-> -r- S- O C 1 1 CUCU ItS IB -t- (O i CUre
51 ICQCCcCQiQiSQ: «_5S:30OC_50OOO
50
-------
I
I
Q-
S-
o
<£>
en
CM
ii
CM
IT)
CT>
Csl
II
I O-
CT)
O
II
CM
+J
ro
CO
OJ
4-
O
to
s_
o
s_
1
1
Q.
i t
s_
o
en
OO
CO
C\l
1 ~~-
ro
CO
oo
LO
i 1
.
^"
LO
CO
"3"
«d-
f^».
OJ
QJ
T3
03
o
c
fO
4-J
oo
cy
o
X
*\
>1
51
-------
in Column 7. Finally, the standard error of estimate was calcu-
lated as indicated.
In Table A-3 data from the SURE network and from PEDCo's
compilation (Tables 4 and 5) were used to verify the relation-
ships developed for Method 2. Measured mean TSP values for each
site were multiplied by the respective constants to obtain IP and
FP estimates. Then, the standard errors of estimate were calcu-
lated. As can be seen/ these results are comparable to the
results in Table A-2.
Table A-4 presents the Method 2 results for the high 24-h
TSP concentrations. For this analysis the ratio of IP/TSP was
calculated for the five highest measured IP concentrations at
each site. In all cases, the highest measured TSP concentration
occurred on one of these five days. Next the average IP/TSP
ratio was calculated for the five highest days. The mean ratios
for the E and W sites were then calculated from the individual
site means. For both subgroups the average ratio was calculated
to be IP = 0.75 (TSP). Then, the 0.75 factor was applied to the
highest measured TSP value to obtain the results in Column 3.
Finally, the standard error of estimate was calculated.
For FP 24-h concentrations the ratios between the highest
measured FP and TSP concentrations were calculated for each site.
Then, the average ratios for the E and W subgroups were deter-
mined to be 0.577 E and 0.230 W. Application of the t-test
showed that the difference in constants was significant. The
highest measured TSP values were multiplied by the appropriate E
or W factor to obtain the results in Column 6. Then, the dif-
ferences between FP estimates and highest measured FP value were
calculated. Finally, the standard error of estimate shown in
Table A-4 was calculated.
Also shown in Table A-4 is a verification of the procedure
using SURE network data. The appropriate constants for high 24-h
IP and FP were applied to the E and W sites to obtain the data in
the table. The results of the two analyses compare favorably for
FP but not for IP.
52
-------
I
t
o
is
co
z
o
o
a:
co
Q. H-
00 UJ
I CO
_J- Q
«=C
CM
O
CO
I
CO
(O
E
CT>
a.
sr
o
re
S-
t
OJ
c
o
o
1
O 0> M-
to *-> 14-
-Q 3 -i-
^C *""" ^3
to
re o_
O) LA-
IE:
1 1 1 ^g
n.
U- 12. Q.
CO CO
t i
1 p^
to o co
UJ '<* CM
0 0
O)
p
3 <+-
, _1I . - t> n
O «l-
to tD
.D
T 1
t 1 CO CM
UJ UJ UJ
QJ
^
3 O
re c re
M re T-
C &~ "O
O CJ C
,£_ CO > i
co »< en
CO t 1 t-H
O CO CTl
CM CM t 1
r-^ en i i
vo « i r»-
»-l CM « 1
«^- .53- CO
UO CO CM
*""H i^*. cn
CO CO CM
VO VO CM
LD CO VO
CM CO CM
UJ UJ UJ
in
r
re
S- CJ>
a o
re Q. to
o ^s QJ
C O i^~
3 0 -t-
t i r-» i i
* tn «si-
0 o vo
CM CM t <
cr> co en
LD if <-t
i< i ( r 1
VO O t
CO VO CO
CO CO CM
CM CM CM
«3- O CO
«* CM CO
CM CM t 1
UJ UJ UJ
O) CD
CC^
>
>> 3
re JD
3 v>
~
*-> H- OJ
u_ o: _i
o o
«-H T 1
tn vo
11 . i
ID 0
*i- un
I 1 T-H
cr> UD en
i i^r~- T-H
r- co CD
«d- cn r i
<) co co
vo co en
VD T 1 CM
T 1 CO CO
333
_
i i fm* _^_
^J to »T
a: -o 3
c o
cu o *
en u
^^ re ,*
o c +->
a <; co
CO CO
T-H T-H
r-. vo
T-H T-H
CM f^
Lf> «3-
I-H i-H
VO CO «3-
O CO CO
O CD O
co en O
CO CM «!3-
VO CO VO
CO CM VO
OO CO CO
333
~^
to to to
333
O 0 0
1 1 1
-l-> 4J -l->
CO CO CO
uo
VO
o
LO
co
LT>
T-H
*
3
to
3
O
t
4->
CO
vo vo o
CM LO I4D
O O CD
«* CO CO
CO CM CO
VO VO O
VO CO CO
CO CO CM
3 3 UJ
_
to to
3 3
o o re
i i r
0
+J *-> C
CO CO
T-H
l-~-
CO
i-H
UJ 33
O
t .
H '
n^i
c c
r- O
T to c:
s- 3 re
CD ^ J2
4-? re a>
CO C_5 _J
vo
vo*
ID
iD
T-H
en
CO
3
10
r
,M.
fC
.>
c^
*
o
o
VD
f^.
T-H
en
CM
0)
CJ>
re
s-
>
=c
53
-------
I
I
Q.
CM
Uf)
C\J
r-.
r-H O
II
jo.
oo
CO
I
W _>i
CV
cc
o
LT>
II
ro
o
c
ro
54
-------
H- 1
^
o:
t
i
CO
^y
O
<->
et
co
UJ
E:
l-rl
1
o.
CO
1
0
UJ
ce:
-""^
co
s:
» i
X
s
u_
o
co
1
CM
1=1
CO
:r
UJ
s:
4-
»/i i ^ t)
4/1 n *i
,^} Z3 !""
«=C r T3
1
re CM
o>
^" f^
tf^ *-^-
U-
£S
II 0_ 0-
CL. CO CO
U_ p- 1
oo en
4-> LO CM
to
UJ O O
cu
3 4-
01
T^
to "O
-Q
r^
1
to «d~
re CM
o>
S D.
i t
II JZ
0. 1
r-^ CM
4- * f"> Qt
to tO
UJ r
»
0
UJ
Ol
4-5
p
CO
r*^ VO ^*t" it' CO CO ^^ LO !"** LO *s^" T-H 0^ \«O CTi
to LO CTi r^ r-H **o ^^ en rn co LO co vo oo CM
1 1 r ( r-H i 1 r-H r-H «^- r 1 CM CM
CnCOVOr- ICOCOLOODLO r-t O CO O O r i
f^* LO OO CO CFi r-H LO CTi O*i CM CO r * CO r-H CO
CO LO CO CO LO CO ^" ^" ^" r-H CM CM «5d~ CO ^~
OCMOI ^COCOCTicnOO lO«d"«3-r-iVDCM
CO r ^ OD LO OO ^» O^ VO LO CM 00 r-H tfi Q^ f*>
LO *^" UO CM ^" Si" "S3- "C^- ^* CM CM UD CM LO CM
OCOLOOCMCMhvCMO OOOLOLOCyvtD
TLOLO'^''^-! ICOOr~- VDCMr-ILOCMI
r Ir-H CMr-H r l«^-r-Hr-H
COCOLOOOOODCOCOCO OCMLOtOr-4tO
Loi^cricocnoocnvo coooovor-ico
Li^ ^O ^O CO UO ^J1 ^^5 v^1 ^D VD 00 LO f*^». CO f"^«
r-H r-H
CT> CM LO *3" <^~ CM '^.O CTi . ^" CO O CM «d~ VO
vo LO r^^. co vo vo vo vo LO r^* 0*1 co co o*> vo
CM r-(
UJUJUJUJUJUJUJUJUJ 333333
E
-ii -o o re
O i "O 4-J O
O QJ O CO r O NX
S- -I O r-H CM QJ.i 3 O
rv*i 4 3- re re co co re E i- re E QJ o
J=-OT-r r-lr-HEO 4- J= ' Ol -r- 4-> i-
O -i- -O O T-4-> O)tOE>EJ3
E »-> re re cno. cu-oto s_ s- E 3 re
4J-r i_oEi i coo) rere-i rei o>
21 ICQealeCQiQiSlCc: CJSISCOCJCO
CO
oo
CO
CO
vo
o>
en
re
cu
f^
o
cu
3
E
E
O
(J
55
-------
*"~v
a
QJ
3
C
J3
c!
o
u
^-
1
t~£
CO
f
2
n
E
cn
^
C
o
ro
S-
4-'
E
CO
o
u
X
(U
^*
1
O CD 4
to +J <+-
-D 3 !-
«=C r -0
J=
1
OO «3"
ro CM
OJ
E O.
u.
II C. O-
Q. to tn
Ll_ 1 (
co ro
+-» LO CM
Ol
LU O O
a>
3 <*-"
^ f]
O !-
to -o
e£
f~
1
in *3~
ro CM
OJ
2: a.
II JZ
Q. 1
i LO «a-
f^» CM
4J CD C.
GO 1/5
LU t
3
i-
o
LU
CO
4_3
r-
to
[ ^_ ~ p^_ ^. -f~ H~ -s^^ .-»
CM «^ T 1 «3" CM r 1
CM
/yt f«H CO f*^ VO ^^^ **""' f^ 0*5 CvJ
I^» Lf5 CO ^^ ^J" ^" CO !** CO
^j- CM LO CTt «^ LO LO LO CO
* < i^H
CO 1^3 CM i cn
i ro « QJ cn ro
a; c o; u_ -t-> o c i- s-
3 O S_ O >> 3 OJ
cn+j EEC ro ja >
re C ro re C- LO 3 oo eC
+J ro "- O -^ QJ -r~
o o =: 3 o -i- +j i a>
I
I CO
fi. r-l
u_ r^.
u
-o
CO
UD
CO
CO
CTi
CO
X
IB
x
rt!
ro
ro
S-
o
0)
4->
to
on
QJ
O
s-
s-
01
C
ro
i
a.
UD
cn
co
it
o
IO
*
VD
CD
«=3-
ro
CO
LT;
O
CM
O
II
x
re
CO
X
ro
E
56
-------
1
1 CsJ |CM CO C
D- sD ko
u. r~- Ir-.- co sO c
tn kp CsJ v
4t- o ~^^"
i u_ u if u
1
^a " oo
ro i c
i . i i
i 1 1
S-
£
ro
o
c
o>
E .
OJ
Q- ' - '
QJ
S-
o
<4-
O)
I-J
ro
-t->
t/i
0)
**-!.-.
o i LTJ kn co c
Q- CO CO
in i i o O CO r-i C
s- co Ico ro c
o i- - i ____
i- 0 ^^~
i- 1-1- II II II
o>
CM X
o -a
*- P-l >>
ro oo
c It
i <
NJ CO
* "*}"
VJ
£> O
II II
X X
ro x ro
E : - E
>-J
*^__ t/^ f}._
1 1 1
1~ LJ_
X) CO
CO
sj
X) 0
II II
X X
ro x ro
B - E
a. bo a_
57
-------
Table A-5 presents .an evaluation of the relationship deve-
loped by Trijonis (Method 11). For each site, the median annual
visual range (V) was estimated from Figure 1. Then the extinc-
tion coefficient (B) was calculated. Using the measured TSP
annual arithmetic mean at each site, estimates of IP and FP mean
concentrations were obtained using the referenced relationships.
Finally the standard errors'of estimate were calculated for the E
and W sites for both IP and FP. As indicated in the table, the
method is most appropriate for the east. For the western sites,
FP estimates are better than IP estimates.
Table A-6 presents another analysis of Method 11, using data
from the SURE network and for 6 selected western sites. Similar-
ly to what was described for Table A-5, estimates of IP and FP
concentrations were obtained from measured TSP concentrations.
Again, the results for eastern sites are better than for the
western sites.
Table A-7 presents the analysis results of the alternative
equations developed for Method 11. The coefficients in the
alternate equations were derived from the small W data sets using
linear regression. The calculated value (17B + 0.6TSP) or (10B +
0.3TSP) was used for the x-value. The measured IP or FP value
was used as the y-value. New coefficients for the equations were
obtained from the regression equations. As is evident in the
results, the alternate equations yield a better prediction for IP
and FP for western sites than the original equations.
Also included in Table A-7 are equivalent ranges of esti-
mated IP and FP values for the various areas of the west.
Table A-8 presents the results of Method 13 which involves
taking an average of results obtained from three different
methods. In this method the estimated IP annual average concen-
trations are calculated separately for Methods 1, 6, and 11 for 9
eastern sites. For each site the average of the three estimated
values is obtained. Finally, the standard error of estimate is
calculated. As indicated by the result, this averaging of three
results yields a relatively low standard error of estimate.
58
-------
D-
OO
OO
HI
5>
-I
UJ
Ll_ OO
o
oo h-
OO O
i ti
i o;
t-H O-
Q
O
UJ
I
O CD 4 CO VO CP) CM O ^" C\J *ci- CO LT> ^- CO CO «* IO
JO 3'r- ii O t< «d- «3- »-H tif-1 IT> O <' O
to CMi)'3-cooo«icftco
ro O-
oiu- cr>oi£>cococr>ir>cr>co VD r^. «
CO
C UJ <* O
o
43
<4-
O> JO 3 -i- CM
O «=C r -0
o
*
> tO O CSJ it I Lf5l
ro a.
cu ** CO VD ?^» CM vo fo ro <* r^«. cj~ co »* cr> r*-» cri co CM
SI CM iICMtCCMCMCMCOCM CM CvJCOCMCM«^-CO CO
a.
II ir> oo
ir> co ti'
vo
O
-l->|--.? CO^COr-lCsJCOCMCOCM ^-tCvJCM^-ICMCM
to CD
O i
I
cr>
CO
CO 'I O ' O r( rI <H rI
r- CMC3COOCMOOOtf) IT> t
r-ICMtICO»I'-H'<<liI IDi-Hr^-CMtn
CO
s-
o
I I I I 11 1 t,.l 11 I 11 I 11 I I I 1 11 t I I I
CJ
4->
00
-*: -o
O < T3
O O) O
i- -i- O «I CVJ
co <+- 3: fO to o o ro c
_E X3 'i- i «I << C O
O ! "O O !-»->
c -i-J re ns end. ex. -o 01
4J -r- s- uclI a> cu
2: _j co «c «c oi o: s o;.
OJ
CD
o to
s- cu i o j«;
O) O) i 3 O
> i- (O E CD O
cC 4- .E tt) -i- 4-> i.
OJ to C > C J3
S- S- C = -i- H3
c_i 2: 3 oo O oo
O!
en
a
>
59
-------
l/l
to
O!
C
I
I
4->
in
QJ
r i
CO 00
ex. *
LL. , CO
[Q.
|u-
CTl
CO
< 1
o
II
+J
X 10
« (O
>,
(S
in
cu
i
*->
in
ns
CL.
S_
O
O!
o
ro
C
C
4J
oo
in
ID
UD
CM
o
CO
o
in
ro
CU
in
re
QJ
in
OJ
r- LO
C_ OJ
s_
o
cv
ro
o>
Q-
ro
O
in
O>
60
-------
r
1
1
CO
1
1
» t
CD
* t
«-< f
> LU
Lu c£
a:
CO Q
to t
1 LU
cC O
a; uj
1 LU
r-t O
r-H "Z.
0
o
;c
LU
*.
^D
i
UJ
to
E
en
c
i
TO
C
0)
C
0
CJ
el
1
O O) tt_
JD 3 -i-
el r -o
ro D-
s:
Gi-
ll CO OO
D- tD H-
U_ CO
+ -f O
CO 'CO
i . i «^- o
1 « 1
1
O QJ t>_
CO 4-> 4_
a: -o
TO
O O-
s; <
O-
LO tO
l! LO t
O-
O O
(-> 4-
l/> ^* PQ
uj B r^.
t i
- E
to en
H- s.
ID
CO
r 1
t
3
O
LU
QJ
to
r-ILOLOr^iICMCOOLO
LOCOr».CMCOLOLO1 o o to o1*
CTlLOLOCOCTlCOt^OLO
t^ *sj" TH f^ CT* CO *^^ to CZT^
LocoLOcooocricO'-*
O»-ltOOCMCOOO< I
CO
co cf» o^ i i r~- cri co co CM to
CMCMCMCOCOCMCMCMCM CM
LDr^LOt--oOr-icocri
CMI~-.CMC>LOtOr^COC3
CMCMCMCOCOCMCMCMCM
CMCOLOCVJLOCOOtOO
CM LO CO ^" LO !!}" t«J" CO CO
ocniDcoco«Ni-tocoLo
r-^C^CMCOCOCOLOCOCM
i t CT> LD O O i TO
r- re CD en s-
0) C C£ U- 4-> O C i. QJ
=50 J- 0 >, 3 >
cn-4-> ceo re JQ ei
roCTOTOQ-t03 CO
+J fO 'f~ U J^ QJ «r
CS--C5COI .D-5
OOC3O-r-4->i CU
StOH-lODie3LUD;_J
ID CM t 1 CO O O1
CM CM LO h». CM LO
CM CT> LO CM i 1 LO
CO CM LO tD CO LO
< 1 r 1 r 1 t 1
r-. r^. "sj- cr> r i to
O O O"> CO < <
ro
CL>
-D
re
61
-------
CO
QJ X
S *
Q- CO
U.
+J
CO
QJ
D-
U.
I
«=c
o
i-
o
C
(B
CO
ra
a;
O cTi
Cu
CO >< tTi
t. LTJ
C i-
t £
Q)
CM
a -o
CC
CO
C\J
<->
CO
re
O)
«3-
O
x
>,
CO
01
C
f
Q-
S-
0
fO
o
CO
ro
O)
62
-------
i
i
oo
oo
>- UJ
I I
I* t 1
J 00
*
no y
I-H DC
oo uj
1 I (
> OO
S .- t I 1
u_S
0 OH
o
oo u_
H i
"ANALYS
QUATIONS
r-l UJ
UJ
O 1
1 OH
UJ UJ
O 0 <+-
CO H-> 4-
JD 3 T-
r-H OO
UJ O
0 -t-
O QJ <4
to 4-> If-
^o
CO
CD i <
SI
Q-
u to oo
Q_ CO 1
» i « (»Q
o
+ o
4-> «D +
to co
uj CM r-^
1 r-H
r\ cr
oo -^
H- 0)
3.
UO
u r--
co
CO
I 1
r
> E
CO
oo
^- en r-, uo oo r-. r-. < ,3- uo vo CM
00 0 00 CM 00 0 * r-, 00 « 0 CO
ouor~-uooovocMcy>uocMt-HUO in OCMUOUOUO
1 ' >-Ht Ir 1 r 1 II i 1 r 1 r 1 OOUOOO
( 1 i-H
01
^^ ^O f*""^ f^^ *
» IjQ
*sr ^" .00 O t»o co co co * < r*-*. r-^ co co o^ co ^" «~*
O^ O"» «"
^ VO CM 0 * 0 VO CM CO
CM 10 0 rH VD ^ r^ 0 CO
o uo uo i
oovot^vo«?r^ooooo 01 ouooooooo
OO r i CT* r^ CTl OO r^* CT> O"i ^D TU "^ f*^. I t UO UO UO
CMOOCMCM«a-OO OOCCCZ CM «3" t*-. < 1 CM CM
oo *3~ o^ *sj~
r^.
._. oo r- 1 oo co
ocMcriiocMr^tTicouo OCMOOOOOO
ID OO Crt VD 00 O CO CO C5 °V r^t r^ 00 «d-
CMOOCMCM'5d- t 1 CM CM OO
uo c> O uo uo
r-co<=j-ootT>oor^.cMCMoooooo oo«a-^«xr«a-
UO r-» UO UO t~» «3" «sj~ OOCMCMOOCM UOUOOUOO
5r«a-uoooi~^uot-Hi luooo^h oo > t CM oo oo «*
uo
CO CO * -d- CM
OO OO CTi CTi
» r-H
o o o uo o
uoc->e-> ' f^fO«d-CM
UO « 1 r-~. CMUOt If-^VD-Cd-CMCMi-H OCDOOUO
UO UO CM CM i 1
n.
Q.
^ QJ to co co
^ ' C^ fO E CO
O ro co (/) O *r- >
-VJU U-U- to s- C-l-»CJi2:^tO -i-O) 4->S_ r CO
o^1 =s o -a cr->s:cuuci.-f-s-
s-fOEajoa>s_c o> cc to'-.- so
'f-^c^'r-HWt^raJOcocco ^cu1*--*-5 >
^* ^« C 13 *r- ft3 H3 CO fO S ^^ S = O ' O CU ft3 rv^
63
-------
CO
*->
c
QJ
0
O>
O
u
C
10
tO
CD
r- cr.
rC
C-
U- LD
CO
O
U- II
CM
CTl
CM
CM
O
II
X
o
-C
01
(O
O
O)
-o
O
cu
=
CD
O!
o;
I
O!
Q
O
ro
=3
cr
cu
£1
O
CU
CO
cu
64
-------
oo
o
o
1
2:
UJ
ac
o
o
C£.
u_
oo
UJ
_J
Si > >-
co
to a. o
oo 2:
T3 1
.C i-H
(-> CO «*
cu o
s: i oo
t 1 Q-
o t
J= <-n
CU
s: o
cu
"r
oo
CM ^" LO f^ CM CM f^* < H CO
CD CM t*H f^«. LO r^* ^O CO l^5 t~*<
CMi ICM< IOJCMCMCOCM CM
roCMCOi ICMCMCMCOCM
tnco»-ivocyico« 1 ' ILT5O
CSJ CM i 1 CM CM CM CO CO
CO CM CO 1 CM CM CM CM CM
O T ~O to
O O) O t-
S- -I- O *l CM QJ
jc "a -i i i it i c o re re end- Q. TD to
-rJ -i S- O C 1 1 CU CU
CU
4->
re
1/5
O)
c
i-
i.
01
i.
to
re
-M
oo
I
t
a.
o
o
o
o
II LO
i.
o
un
oo
CO
o
*
CO
CM
II
CM
t(
'
CD
II
X
oo
65
-------
Table A-9 presents the results of the Method 13 analysis for
annual average FP concentrations. In a manner similar to that
described for Table A-8 the results calculated for Methods 6 and
11 are obtained and averaged. Once again the standard error of
estimate is quite low.
Table A-10 presents the Method 13 results for 24-hour FP
concentrations. The results of Methods 2, 8 and 9 were averaged
for each site. There are three groups of site data in the table.
The first group and the third group represent IP network data for
E and W sites respectively. The second group includes data from
the 9 SURE network sites. As seen in the calculated standard
errors of estimate this analysis does not seem to be appropriate
for the west and only marginally appropriate for the east.
66
-------
1
CO
Q
O
LU
s
rv
LU
3=
O
O
cc
lo-
co
LU
1
O-
CD ' '
ti LU
O CD
«=C «=t
LU LU
-
_J
et
1
1
O
0
LU
s:
m '
i
«=c
LU
_J
CO
t_
.11
II
n
B
a.
«
o
t->
s_
^ >
cr
QJ
O
0
o>
C71
tD
QJ
J| r*1^ ^^
T H J|X.
r 1
f> l~> J-
f~ Bf..
-M CO >*>
CU *f~ *-*
2: > -r-
11)
i-
CO
ovoor^crio^-it ir-n
C\JOC\li-HC\JO->O« ovocococnir>j v« ^ n*»
r-CTiODVDLOtOLOr-HVO
O LOC\JC\J«5d-O^DOsi
s* u^ ro -H o in
r^chco-si-'xi-i^yDcoir)
OJ
_ . O^
O i -0 £
O CD O o>
S- -i O i I OJ >
j=: -o -i- . « i f-H c o
U !- -O O -1- 4-> ,
+->- S- O C ( f O) O)
_ " -
O> :
(O
^
4->
cyi
CM OJ
1- 1
.0 1 «* «d-
Q- ro ro
trt Ll_ L ._
'S-'. ' . t
OS- II ||
i- 0
S- U.
CL> cj >
- ^ . >
*- CO
03
c
-------
I
CO
o
LU
o:
UJ
o
o
co
UJ
CD
z:
tt
CD
f^_ i
UJ
u. ac
o i
CO CM
4
CO
cc
I
CO
o
o
UJ
o
t-H
I
UJ
CO
OJ
C
3 CD
O &>
.Q M-
«=C -r-
en
re
0)
(O
CO
C
C
o re
o
j= a.
CM
i
CM CM
O OO
01 CO
*5^ tjO
o
jz
CM
-O
o a.
f^ ^-^
2: vo
o
o;
CTiCOOCO'S-OCNJ^OrO
COCOVD
un
00- i
< *«
c +-1 re re CDQ. Q- T3 to
OJ
cr>
re
QJ
s- to
ai i
> »
3 O
s- o >,
O)4-> CCO
re c re re Q. 10
a> ^~
CT -a
re ai
s- 3
O) !=
t-> re -r- o
o>
68
-------
QJ
o
u
QJ
jQ
re
m
E
*O>
3.
C
o
r*
re
s-
+j
C
0)
U
C
Q
QJ U
2 OJ
i S-
o m
ts> ^-
js M-
^C *i~
XI
1
CM
Q_
U_
OJ
CD
re
QJ
>
>
re
CO C
G
XJ re
JC D-
4-> u.
QJ
»
CM
1
CM CM
X> 0-
Off)
V i
-C t
4«>
QJ CO
SI LT>
»
O
0*1 "a"
CM
x>
Or\
i^.
J= I I
+->
s vo
o
QJ
4-3
r
oo
vo oo co r»» ?- o
r>- vo i1*-. LO vo f-»
T4 f-H t»H r~t
» 1 O CO O O r-i CTl
CM CO t-t CO r-4 CO CO
< I CM CM «3- CO «* CM
r^ ro « < ro i i i
CTi CP> Cfi t»». r«. «;J- O
fH CO CM CO PO «*
1'
Nfc^ ^*i ^^^ V4J WXJ |
CM co < < «3 cy> o
CM CM IO CM LO CM
«!* rv. ro t^- ci »*
co «~H ro o co r~-
r-l CM CO CM «;!- »-H
c
3 QJ
o re cr
4-* CJ 03
QJ ^~ O vx ^
oi i 3 o o>
t- re E OJ O >
<+- x: QJ -i- 4-J s- <
aj (/> c > cr -a
s- s_ c 3 T- re
r3 --C - -3 r- 2J
«_> s: 3 oo o oo
i
i
4^
i/>
QJ
5
..r
s_
re
E
r-
^
o.
r r^
vo *o
CO CO
CM CM
CM CM
CO
LD UD CT>
to co
r-l CVJ
~-i '
II II II II
X X
M « re
x> >, E
^ ^ |Q-
|u_
CM
a-
LO
O
II
X
x re
- E
>> i
00 CU
fu-
o.
QJ
re
QJ
s-
(U
1
1
LO
re
QJ
*
4_3
£= '
QJ
X)
C
QJ
Q.
QJ
"C3
i t
<3-
»
O 0
» 1 . <
0 0
^^>
CO CM CM
.
T 1 CM
r 1 ID
»^^^^^
^^^**
II II II II
X X
CM > (C
X3 >> E
w °° lo-
|u_
r t
CO
O
II
X
x re
. E
oo la.
ILL.
c/i
re
QJ
s_
re
o i-
re
x>
re
oo
s_
a.
CM
co
oo
cr>
CM
XJ
CM
CM
CO
CO
00
CO
x
re
o
ur>
CM
O
II
oo
re
69
-------
-------
TECHNICAL REPORT DATA
(Please read Instructions on the reverse before completing)
NO.
» 'EPA-450/4-84-021
3. RECIPIENT'S ACCESSION NO.
TITLE AND SUBTITLE
Estimating PM And FP Background Concentrations
From TSP And Other Measurements
5. REPORT DATE
August 1984
6. PERFORMING ORGANIZATION CODE
8. PERFORMING ORGANIZATION REPORT NO
Pedco Environmental, Inc.
3. PERFORMING ORGANIZATION NAME AND ADDRESS
Pedco Environmental, Inc.
Golden, CO 80401
10. PROGRAM ELEMENT NO.
11. CONTRACT/GRANT NO.
68-02-3512
2. SPONSORING AGENCY NAME AND ADDRESS
U. S. Environmental Protection Agency
Office Of Air Quality Planning And Standards
Monitoring And Data Analysis Division (MD 14)
-- . 77711 .
13. TYPE OF REPORT AND PERIOD COVERED
14. SPONSORING AGENCY CODE
MENTARY NOTES
EPA Project Officer: Thompson G. Pace
This document has been developed in response to the recognition that there may
be a number of States revising their State Implementation Plans (SIP) for ^articulate
Matter (PM) following the recently proposed revision to the National Ambient Air
Quality Standards (NAAQS) for PM1(r State and local agencies need a method of
estimating background for size-fractioned particulate as part of SIP development for
PM and Fine Particulate (FP). The method for estimating background concentration
that is specified in Section 51.13c of Federal regulations uses a concentration
measured at a nonurban site in or near the analysis area that is unaffected by
near emission sources. This document provides information on how to estimate a PM.
background value from other measurements when PM monitoring data are not
available. . , ,
KEY WORDS AND DOCUMENT ANALYSIS
DESCRIPTORS
b.lDENTIFIERS/OPEN ENDED TERMS C. COSATI Field/Group
19. SECURITY CLASS (ThisReport)
21. NO. OF PAGES
75
>0. SECURITY CLASS (This page)
22. PRICE
'A Form 2220-1 (Rev. 4-77) PREVIOUS EDITION is OBSOLETE
-------
------- |