United States
Environmental Protection
Agency
Office of Air Quality
Planning and Standards
Research Triangle Park, NC 27711
EPA-453/R-96-013b
October 1996
Air
C CDA Study of Hazardous Air Pollutant
Emissions from Electric Utility Steam
Generating Units -- Interim Final Report
Volume 2. Appendices A - G
-------
U.S. Environmental Protection Agency
Region 5, Library (PL-12J)
77 West Jackson Boulevard, 12th Floor
Chicago, II 60604-3590
-------
TABLE OF CONTENTS
Appendix A
Appendix B
Appendix C
Appendix D
Appendix E
Appendix F
Appendix G
Median emission factors, determined from test report data, and
total 1990 and total 2010 emissions, projected with the computer
emissin progam A-l
Matrix of Electric Utility Steam-Generating Units and Emission Test Sites B-l
Listing of emission modification factors for trace elements used in the
individual boiler analysis
Discussion of the Methodology Used to Develop Nationwide Emission Totals
Health Effects Summaries ,
Documentation of the Inhalation Human Exposure Modeling for the Utility
Study
Priliminary Uncertainty Analysis for the Characterization of
Human Health Risks from Direct Inhalation Exposures to Electric
Utility HAP EmissionsSunmary of Speciation, Environmental Chemistry,
and Fate of Eight HAPs Emitted from Utility Boiler Stacks
C-l
D-l
E-l
F-l
G-l
-U
ii
-------
Appendix A - Median emission factors, determined from test report
data, and total 1990 and total 2010 emissions, projected with the
computer emission program
As explained in Chapter 3, inorganic hazardous air pollutant
(HAP) emission totals were estimated through the use of emission
modification factors (EMFs) in the emission factor program (EFP).
The stack factors mentioned in Tables A-l thru A-3 are medians of
mean emissions for each test site and were not used in the
computing estimated total emissions for inorganic HAPs. Medians
were used due to a limited amount of data. These factors were
extracted from the test reports and are presented for
informational purposes only.
-------
This page is intentionally blank.
-------
-. to
« c
o o
5l-»
sis
m o> O
2 **
.S ° ""*
IA Q
iu S
II
1!!
p-
ili
£ 0 (0
£ ° c
o rt* «2
S w 1
.1 i =
'ii
s*
5 c
!x °
S o
! c
S ID
"§ S
= °-
.. c
0-2
U '-
JH ,*^
_^ £ „
JS 3 3
« * CD
c £
.2 8
TO _
i i
y
2
i-.
.x o
Si;
e
H-
u °-
1
3
— rf
i X
II
Jl
«
^
CO
6
CM
CD
O
§
CM
in
co
S
o
.c
u
H Hydrogen
o
CM
CO
CM
.. 3
O CD
0 C
a o
c *£
a j-;
"o ^
a> —
5 oo
CM
in
(A
Q
O
"^
o
0}
_o
E
D
•z.
£
0)
(fl "O
2.1
2 o
0 °
=
0)
•o
'c
ID
U
li
0
o
0
in
CM
o
o
in
of
(O
o
*~
10
o
o
CM
^r
*
0)
5
1
|| Hydrogen
CO
CM
O)
T-
o
O3
CM
d
CO
|| Cadmium
CO
00
o
o
^
o
^
CO
CM
CM
|| Chromium
CM
CM
o
,_;
to
jv.
CM
O
"5
J2
8
CO
00
CM
CO
in
o
CO
CM
•o
CD
5
CM
CO
CM
O
00
r-
in
ir-
05
in
*~
CM
CD
|| Manganesi
in
tO
in
^-
CO
o
en
CO
O
CM
1
in
00
CM
in
0
CO
CO
CM
1
o
o
in
CO
o
fv
CM
CO _
3
O *
t^. IE
c £
(D ^
^ £
i
O
vi
O
o
£
'o
h_
Q)
.O
£
D
Z
CD
ffi
5 -D
« 0
If
5 I
Q. °
<
*
£
in
H Phosphoru
o
CO
CM
1
0
CO
o>
CM
(O
en
*—
|| Selenium
u
•rl
JJ
cO
g
O
rl
(0
U
•H
rH
U
u
rH
O
a
"d
(0
1-1
3
"*v.
d
•H
X
o
•rl
T3
Ti
d
(0
*
U
• H
d
10
Oi
rl
o
..
u
-H
C
id
Oi
rl
O
C
• H
tt)
0
d
0)
rj
cr
Oi
ra
Ol
d
• r4
5
O
rH
rH
O
UH
0)
J=
JJ
d
•H
73
01
JJ
tfl
•H
rH
Ol
s^
id
to
T3
d
3
o
o
o
a
•-H
O
rl
Jj
O
O
h
X
o
(0
JJ
to
01
X
JJ
JJ
(0
to
d
o
• H
JJ
10
IH
jj
d
HI
U
d
O
u
dl
s
u
•H
d
id
Ol
SH
O
E
o
rl
*4H
'O
01
d
'E
Q)
JJ
0)
T3
0)
s^
Ol
3
to
SH
O
JJ
u
(0
UH
d
o
-H
to
to
•H
E
(U
d
10
•H
•0
0)
s
.
,— ^
W
JC
rij
ON|
"*— '
CO
c
o
^Q
^
(0
u
o
^4
T3
j:
T3
u
jj
u
0)
JJ
0)
T3
to
(0
3
to
d
0
• H
JJ
(0
rl
JJ
d
01
u
d
o
u
to
(0
Oi
V
3
rH
UH
•O
01
S-l
3
to
10
01
E
0>
Oi
S-l
£
4J
^>i
rH
rH
id
u
•rl
Qj
^
JJ
MH
O
0)
d
o
jj
to
(0
0)
i-H
JJ
10
d
D
3
JJ
0)
rH
JJ
3
0
s^
0)
rH
•H
0
.Q
k
O
«
JJ
0)
rH
JJ
3
O
01
U
•H
^
0)
Tl
ra
id
Ol
0)
3
rH
M-l
Tl
Ol
SH
3
to
(0
o
E
Ol
01
rl
,d
^
rH
rH
Ifl
O
•H
Qj
JX
Jj
UH
O
0)
d
0
4J
to
(0
01
rH
JJ
(0
d
Oi
•C
3
to
jj
rl
0
a
0)
rl
JJ
to
0)
JJ
E
o
SH
<4H
d
Oi
A;
id
jj
0)
rl
01
3
CO
a,
§
0
•H
d
(0
rl
0
d
•rl
rl
0
14-1
0)
J^
o
JJ
o
id
MH
y
O
(0
JJ
w
A
to
d
0
•H
M
to
•H
E
D
•a
01
AJ
to
g
-H
JJ
to
0)
0)
JJ
a
o
rH
Oi
Ol
T3
O
jj
Tl
Oi
to
3
jj
O
d
0)
SH
0)
3
to
^
0
JJ
0
(0
UH
(D
to
ID
£H
T3
0)
JJ
U
Oi
JJ
0)
T3
to
<0
3
to
d
o
• H
JJ
10
M
JJ
d
0)
u
d
0
o
M
&
£
u
•H
d
(0
Oi
^
o
d
•H
UH
O
to
d
o
•rl
to
to
•H
E
0)
rH
id
o
..
to
o
JJ
u
id
UH
^
u
(0
JJ
to
E
o
rl
<4H
TJ
iJ
td
g
•H
JJ
to
0)
J>,
rH
JJ
U
01
IH
• H
•0
JJ
O
d
Ol
SH
0)
3
to
d
o
• H
to
to
•rl
E
Oi
u
-H
d
id
Ol
SH
o
d
-iH
0)
JJ
Ol
to
3
id
u
m
u
^,
^
id
rH
rH
•H
3
to
S-l
o
JJ
o
(0
UH
^
0
<8
to
c
id
•iH
Ti
E
g
0
UH
T>
C
id
E
}_(
Oi
o
a
SH
0)
3
a.
E
o
u
Oi
£
JJ
JS
-H
3
TJ
(1)
JJ
u
Oi
•r-t
o
IH
a
to
fc
s
w
E
0
UH
JJ
0
d
Tl
d
id
to
o
JJ
0
(d
d
o
•H
to
to
•H
E
O)
V
0
to
JJ
to
E
o
MH
Ti
01
d
•H
E
rl
Ol
JJ
01
Oi
rl
0>
3
to
d
o
•H
to
M
-rl
§
Ol
Tl
• H
d
to
o
d
Oi
Ol
o
rl
T3
rj
0>
•H
3
O
•H
JJ
id
Z
•a
w
g
w
e
0
UJ
Jj
O
d
T)
d
td
w
o
o
U-I
d
o
•rl
to
to
• rl
&
0)
A:
u
id
w
£
o
UH
T)
0)
d
•rH
e
01
Jj
0)
T3
Oi
IH
3
en
d
0
• rl
03
in
•rl
£
Ol
to
O
O
,c
a
to
0
a
Oi
t?
•rl
J
d
o
•H
JJ
(0
z
tl
A-l
-------
C
to w
-i d)
O -U
JJ O
U d)
(0 T~l
fc O
M
C PH
O
•H -
W W
w C
•H O
P* - j-J ...
WWW
W JJ
C -H -H
to e c
•HMD
Q) CU '"O
S
CN
I
rH I
O rH
i< CN -H
O
d) rH
rH (0 6
Q j_i n
(0 O rJ
EH EH M-l
— V)
S c
o .2
*• <0
•o .2 UJ
w 'c c
13 5 S
.1 ° ~"
W 0
UJ CM
— in
S c
|.2
V * ^
^11
p rt """"
•c: co
« CO
UJ r-
O
"« 1 3
** o "o *"
1 o ° 1|
S (Q _Q
**^ =
JS
0 o t
5 o c
•S 5 °
E je 5
3 Q ^
Z (0
"I
«S 3
«S Q. -= 2
ls!l
•000 =
ma Z
JCJ *• "^-
.. j,
s Si
5li
E * 8
= 0 Ef
z « S
w
is <
'e I
3 y
O .£
in
CM
in
CO
IT)
00
CS
CO
d
CN
Arsenic
CO
CM
0
in
d
,_
CM
d
*
CM.
O
CO
CM
CO
0)
O
^
d
1 Hydrogen
CO
in
(0
d
q
CO
CO
(O
CM
CO
"8
V
CO
*
If)
en
CO
r—
CO
in
co
o>
| Manganes
CO
d
in
CM
d
CO
O
JO
c
ta
'•6
CD
5
o n
03 V)
X) fc-
E *;
Z *H
tn
o -g
o o> c
£ ® 5
n 5 £
_ o
<
•o
£?
~
W
3
0
a
a.
S
d
fx
^
CO
CO
CO
*t
T—
T—
|| Selenium
c aromatic
•H
rH
U
^
U
^
rH
O
Ol
"c
1C
iH
3
1H
\
C
•H
X
o
-H
•0
T3
c
(C
~
o
•H
c
en
rl
o
x
u
•H
c
*
iH
o
c
•H
01
0
c
0)
3
O1
ill
to
en
c
•rl
3
0
rH
rH
O
«H
01
r]
jj
C
•H
•0
01
JJ
to
•H
i-H
01
iH
1C
IA
TJ
C
3
O
U
a
at the stack, control
to
c
o
• rl
JJ
(C
t-l
c
01
o
c
o
u
a.
£
u
• r4
C
(C
Cn
iH
o
e
o
iH
U-J
T3
0)
C
•H
e
01
JJ
01
13
ID
iH
0)
3
CO
iH
O
JJ
u
M-J
C
o
• H
10
D)
•H
e
0)
c
(C
• H
T3
0)
S3
•
«_*
CO
JE
2
&
«— ' •
(Q
C
O
J3
V4
(C
o
^
•o
^
X
rations was detected.
jj
c
01
u
c
o
u
to
(C
Cn
0)
3
HH
Tl
0)
3
to
(C
01
£
w
01
SH
4J
j>,
rH
rH
rC
U
-H
§
U-J
O
01
c
o
JJ
to
rt
0)
rH
JJ
10
c
01
3
JJ
01
rH
JJ
3
0
iH
01
rH
•H
O
Q
^4
O
..
jj
01
rH
JJ
3
O
0)
u
•H
^
01
T3
hree measured flue gas
jj
>
rH
rH
1C
U
•H
p.
^
JJ
M-l
O
01
c
o
JJ
to
(C
01
rH
JJ
*
£
0)
.c
3
to
jj
^i
o
to
C
o
•H
to
to
•H
E
0>
•c
0)
JJ
(C
E
•H
JJ
w
0)
01
XI
o,
o
rH
0)
0)
TJ
O
JJ
0>'O
01
JJ
tn
jj
£
0
iH
*4— 1
c
01
«
JJ
0)
i-l
01
CO
CU
rt"
kC
u
•rl
c
(0
01
^1
o
c
• H
iH
o
4-1
to
^4
o
JJ
0
1C
MH
,x
u
(C
jj_)
M
^
01
to
3
JJ
O
c
0)
iH
0)
^
w
iH
O
JJ
0
(C
U-f
0)
to
01
f~1
EH
T3
01
JJ
U
0)
jj
01
•n
to
<0
3
to
c
o
•H
JJ
1C
^.4
JJ
C
01
u
o
u
of inorganic HAPS
w
0
•H
to
to
•H
E
0)
rH
(C
JJ
0
JJ
to
^
o
JJ
u
*
MH
^
U
(C
to
E
o
^4
MH
T3
0)
JJ
1C
E
•rl
JJ
to
0)
£»,
rH
JJ
U
01
^]
,r|
T3
JJ
O
C
0)
41
3
to
C
O
•H
to
to
•H
E
0)
o
•iH
c
(C
iH
o
c
•rl
0)
JS
JJ
01
u
c
-rl
W
u
^,
M
(0
£>
rH
rH
•H
3
to
i-i
o
JJ
u
(C
UH
.X
u
(C
4J
W
C
(0
•H
T!
01
E
E
o
in
UH
TJ
C
*
E
(8
Cn
O
a
iH
jj
3
a
E
0
u
0)
JJ
XI
JJ
•rl
^
*o
0)
jj
u
0)
•n
0
a
to
£
u
E
o
iH
M-l
JJ
O
C
T)
C
(0
tn
iH
0
jj
u
U-l
C
0
•H
tn
w
•H
E
01
•V
u
(0
jj
10
e
o
)H
rrj
0)
C
•rl
E
Ol
JJ
CD
T3
0)
in
01
to
c
o
•rl
to
to
•H
E
S
to
3
0
IH
O
J3
a
to
o
r]
a
o>
Ti
•H
^
c
o
•H
JJ
(0
z
•o
A-2
-------
Table A-3. Median Emission Factors, Determined from Test Report
Data, and Total 1990 and Total 2010 HAP Emissions, Projected with.
the Emission Factor Program for Inorganic HAPs from Gas-fired
Units6
Gas-fired units:
inorganic HAPs
Arsenic
Cadmium
Chromium
Cobalt
Lead
Manganese
Mercury
Nickel
Phosphorus
Number of stack
factors: no PM
control
2
1
2
1
2
2
2
2
1
Median stack factor:
no PM control
(Ib/trillion Btu)
0.14
0.044
0.96
0.12
0.37
0.30
< 0.38
2.3
2.2
Estimated total
1990 emissions
(tons)
0.16
0.054
1.2
0.14
0.44
0.37
0.0016
2.3
1.3
Estimated total
2010 emissions
(tons)
0.25
0.086
1.9
0.23
0.68
0.59
0.0024
3.5
2.0
Table A-4. Median Emission Factors, Determined from Test Report
Data, and Total 1990 and Total 2010 HAP Emissions, Projected with
the Emission Factor Program for Organic HAPs from Coal-fired
Units
Coal-fired units: organic HAPs
1 ,1 ,2-trichloroethane
2-chloroacetophenone
2,4-dinitrotoulene
Acetaldehyde
Acetophenone
Acrolein
Benzene
Benzylchloride
Bis(2-ethylhexyl( phthalate
Bromoform
Number of
emission
factors
1
3
3
12
7
6
20
1
9
1
Median emission
factor: Ib/trillion
Btu
4.7
0.29
0.015
6.8
0.68
3.3
2.5
0.0056
4.1
6.6
Computer
program: 1990
total tons
40
2.4
0.13
58
5.8
28
21
0.048
35
57
Computer
program: 2010
total tons
53
3.2
0.17
76
7.7
37
28
0.063
46
75
Compounds are listed in the following sequence: inorganic, organic, and
dioxin/furan/polycyclic aromatic hydrocarbons (PAHs). Median emission factors were
determined from organic HAP concentrations at the stack, control device outlet, or
boiler outlet when at least one of typically three measured flue gas concentrations
was detected.
A-3
-------
Table A-4. (continued)
Coal-fired units: organic HAPs
Carbondisulfide
Carbontetrachloride
Chlorobenzene
Chloroform
Cumene
Dibutylphthalate
Ethylbenzene
Ethylchloride
Methylchloroform
Ethylenedichloride
Formaldehyde
Hexane
Hexachlorobenzene
Isophorone
Methylbro'mide
Methylchloride
Wlethylethylketone
Methyliodide
Methylisobutylketone
Methylmethacrylate
Methyltertbutylether
Methylenechloride
n-nitrosodimethylamine
Naphthalene
m,p-cresol
o-cresol
p-cresol
Perylene
Pentachlorophenol
Phenol
Phthalicanhydride
Propionaldehyde
Quinoline
Number of
emission
factors
8
2
2
2
1
5
5
1
4
3
15
2
1
2
6
3
6
1
3
1
1
5
1
11
2
3
1
1
1
10
1
4
1
Median emission
factor: Ib/trillion
Btu
4.3
3.3
3.2
3.2
0.29
2.8
0.40
2.4
3.4
3.1
4.0
0.82
0.079
24
0.88
5.9
8.0
0.40
4.9
1.1
1.4
13
0.68
0.77
0.68
1.7
0.95
0.075
0.0082
6.1
4.9
10
0.053
Computer
program: 1990
total tons
37
28
27
28
2.5
24
3.5
20
29
27
35
6.9
0.68
200
7.7
51
69
3.4
42
9.3
12
110
5.9
6.6
5.8
14
8.2
0.65
0.070
52
42
89
0.46
Computer
program: 2010
total tons
48
37
36
36
3.2
32
4.6
27
38
35
45
9.1
0.89
270
10
67
90
4.5
53
12
16
150
7.7
8.7
7.6
19
11
0.85
0.093
69
56
120
0.61
A-4
-------
Table A-4. (continued)
Coal-fired units: organic HAPs
Styrene
Tetrachloroethylene
Toluene
Transl ,3-dichloropropene
Trichloroethylene
Vinylacetate
Vinylidenechloride
Xylenes
o-xylenes
m,p-xylenes
Total TEQ for 2,3,7,8-tetra-cMorodibenzo-p-d/oxin
2,3,7,8-tetrachlorodi-benzo-p-dioxin
1,2,3,7,8-pentachlorodi-benzo-p-dioxin
1,2,3,4,7,8-hexachlorodi-benzo-p-dioxm
1,2,3,6,7,8-hexachlorodi-benzo-p-dioxin
1,2,3,7,8,9-hexachlorodi-benzo-p-dioxin
1,2,3,4,6,7,8-heptachlorodi-benzo-p-dioxin
Heptachlorodi-benzo-p-dioxin
Hexachlorodi-benzo-p-dioxin
Octachlorodi-benzo-p-dioxin
Pentachlorodi-benzo-p-dioxin
Tetrachlorodi-benzo-p-dioxin
2,3,7,8-tetrachlorodi-benzofuran
1 ,2,3,7,8-pentachlorodi-benzoturan
2,3,4,7,8-pentachlorodi-benzofuran
1,2,3,4,7,8-hexachlorodi-benzofuran
1,2,3,6,7,8-hexachlorodi-benzofuran
1,2,3,7,8,9-hexachlorodi-benzofuran
2,3,4,6,7,8-hexachlorodi-benzofuran
1,2,3,4,6,7, 8-hept achlorodi-benzof uran
1,2,3,4,7,8,9-heptachlorodi-benzofuran
Heptachlorodi-benzofuran
Hexachlorodi-benzofuran
Number of
emission
factors
7
5
17
1
1
1
2
2
5
8
--
4
3
4
4
4
9
6
8
9
6
9
8
5
5
6
5
4
5
8
4
8
8
Median emission
factor: Ib/trillion
Btu
3.1
3.1
3.6
4.7
3.1
0.42
9.7
4.7
0.82
1.5
--
1.6 x 10'6
4.3x 10-6
9.7 x 10'6
5.8x 10'6
7.3x 10'6
5.7 x 10'6
1.1 x 10'4
2.4 x 10"5
5.8 x 10"5
9.8 x 10'6
7.1 x 10'6
3.9x TO'6
2.4 x 10'6
1.0 x 10'5
1.3x 10'B
4.0 x 10'6
8.5 x 10'6
1.6x1 0'5
2.0 x 10's
1.7 x 10"4
2.4 x 10'5
1.9 x 10'5
Computer
program: 1990
total tons
27
27
31
40
27
3.5
84
40
6.9
13
1.5 x 10"*
1.4x 10'5
3.7 x 10'5
8.3 x 10-s
5.0 x 10'5
6.3 x 10-s
4.9 x 10'5
9.2 x 10^
2.1 x 10"4
5.0 x 10-*
8.5 x 10 5
6.1 x 10'5
3.4 x 10'5
2.1 x 10'5
9.0 x 10'5
1.1 x 10**
3.4 x 10"5
7.3 x 10's
1.4x ID'*
1.7x10''
1.5x 10'3
2.1 x 10"*
1.6x 10"*
Computer
program: 2010
total tons
35
35
41
53
35
4.6
110
53
9.1
17
2.0 x 10"*
1.9 x 10'5
4.8 x 10'5
1.1 X10"4
6.6 x 10s
8.3 x 10-s
6.5 x 10'5
1.2x10-3
2.7x 10"*
6.6 x 10"4
1.1 x 1Q-4
8.0 x 10'5
4.5 x 10'5
2.8 x 10'5
1.2x 10"*
1.5 xlO"4
4.5 x 10'5
9.6 x 10'5
1.8 x 10"*
2.2 x 10"4
2.0 x10'3
2.7 x 10"*
2.1 x lO1*
A-5
-------
Table A-4. (continued)
Coal-fired units: organic HAPs
Octachlorodi-benzofuran
Pentachlorodi-benzofuran
Tetrachlorodi-benzofuran
1 -methylnaphthalene
2-chloronaphthalene
2-methylnaphthalene
Acenaphthene
Acenaphthylene
Anthracene
Benz(a)anthracene
Benzo(a)pyrene
Benzo(e)pyrene
Benzo(b)fluoranthene
Benzolb + k)f luoranthene
Benzo(k)fluoranthene
Benzo(g,h,i)perylene
Biphenyl
Chrysene
Dibenzo(a,h)anthracene
Fluoranthene
Fluorene
Indenod ,2,3-c,d)pyrene
Phenanthrene
Pyrene
Number of
emission
factors
10
9
10
2
2
6
6
5
4
4
6
1
1
1
1
2
1
4
1
6
5
2
7
4
Median emission
factor: Ib/trillion
Btu
1.7 x 10'5
1.8 x 10'5
1.2x 10'5
0.0085
0.040
0.024
0.008
0.0042
0.0042
0.0021
0.0010
0.0012
0.0081
0.0016
0.0036
0.0032
0.34
0.0026
0.0003
0.007
0.013
0.0064
0.032
0.009
Computer
program: 1990
total tons
1.4x 10"*
1.6x 10-*
1.0x 10^
0.076
0.35
0.20
0.07
0.036
0.036
0.018
0.0088
0.010
0.070
0.014
0.031
0.028
3.1
0.022
0.003
0.064
0.11
0.054
0.31
0.081
Computer
program: 2010
total tons
1.9x W*
2.1 x 10"*
1.3 x 10"*
0.10
0.46
0.26
0.09
0.047
0.047
0.0024
0.012
0.014
0.092
0.018
0.040
0.036
4.0
0.030
0.004
0.082
0.15
0.072
0.36
0.103
A-6
-------
Table A-5. Median Emission Factors, Determined from Test Report
Data, and Total 1990 and Total 2010 HAP Emissions, Projected with
the Emission Factor Program for Organic HAPs from Oil-fired Units
Oil-fired units:organic HAPs
Acetaldehyde
Benzene
Ethylbenzene
Formaldehyde
Methylchloroform
Methylenechloride
Naphthalene
Phenol
Tetrachloroethylene
Toluene
Vinylacetate
o-xylenes
m,p-xylenes
Total TEQ for 2,3,7,8-tetra-chlorodibenzo-p-dioxin
2,3,7,8-tetrachlorodi-benzo-p-dioxin
1,2,3,7,8-pentachlorodi-benzo-p-dioxin
1,2,3,4,7,8-hexachlorodi-benzo-p-dioxin
1 ,2,3,6,7,8-hexachlorodi-benzo-p-dioxin
1,2,3,7,8,9-hexachlorodi-benzo-p-dioxin
1,2,3,4,6,7,8-heptachlorodi-benzo-p-dioxin
Heptachlorodi-benzo-p-dioxin
Hexachlorodi-benzo-p-dioxin
Octachlorodi-benzo-p-dioxin
Pentachlorodi-benzo-p-dioxin
Tetrachlorodi-benzo-p-dioxin
2,3,7,8-tetrachlorodi-benzofuran
1,2,3,7,8-pentachlorodi-benzofuran
2,3,4,7,8-pentachlorodi-benzofuran
1,2,3,4,7,8-hexachlorodi-benzofuran
1,2,3,6,7,8-hexachlorodi-benzofuran
1,2,3,7,8,9-hexachlorodi-benzofuran
2,3,4,6,7,8-hexachlorodi-benzofuran
Number of
emission
factors
1
6
2
9
3
2
4
2
1
6
2
1
2
--
1
2
1
2
2
2
2
2
1
2
2
2
2
2
2
2
2
1
Median emission
factor: Ib/trillion
Btu
8.2
1.4
0.49
30
7.6
32
0.33
24
0.55
8.0
5.2
0.84
1.4
-
6.5 x 10'6
5.8 x 10'6
1.2 x 10'5
5.4 x 10~6
8.3x 10'6
2.0 x 10'5
2.0 x 1Q-5
8.1 x 10'6
2.3 x 10'5
5.8 x 10'6
5.7 x 10'6
4.6 x 10-6
4.3 x 10'6
4.8 x 10'6
6.1 x 10'6
3.8 x 10'6
5.8 x 10'6
4.8 x 1 0"6
Computer
program: 1990
total tons
5.0
0.88
0.29
19
4.6
20
0.21
15
0.34
4.9
3.2
0.51
0.82
1.1 x 10's
4.0 x 10'6
3.5 x 10'6
7.6 x 10'6
3.3 x 10'6
5.1 x 1Q-6
1.2x 1Q-5
1.2 x 10'5
5.0 x 10'6
1.4 x 10'5
3.5 x 10'6
3.4 x 10"6
2.9 x 10'6
2.6 x 10*
3.0 x ID'6
3.7 x 10"*
2.3 x 10"6
3.5 x 10"6
3.0 x 10'6
Computer
program: 2010
total tons
2.6
0.45
0.15
9.5
2.4
10
0.10
7.5
0.17
2.5
1.6
0.26
0.42
5.4 x 10'6
2.0 x 10'6
1.8 x 10'6
3.9 x 10-$
1.7 x 10'6
2.6 x 10-6
6.2 x 10'6
6.2 x 10'6
2.5 x 10'6
7.3 x 10'6
1.8 x 1Q-*
1.8 x 10'6
1.4 x 10'6
1.3x 10"6
1.5x 10'6
1.9x10-*
1.2X10"*
1.8x10'6
1.4x 10"6
A-7
-------
Table A-5. (continued)
Oil-fired units:organic HAPs
1,2, 3,4,6, 7,8-heptachlorodi-benzofuran
1,2,3,4,7,8,9-heptachlorodi-benzofuran
Heptachlorodi-benzofuran
Hexachlorodi-benzofuran
Octachlorodi-benzofuran
Pentachlorodi-benzofuran
Tetrachlorodi-benzofuran
2-methylnaphthalene
Acenaphthene
Acenaphthylene
Anthracene
Benz(a)anthracene
Benzo(b + k)fluoranthene
Benzo(g,h,i)perylene
Chrysene
Dibenzo(a,h)anthracene
Fluoranthene
Fluorene
Indenod ,2,3-c,d)pyrene
Nitrobenzofluoranthene
Nitrochrysene/benzanthracene
Phenanthrene
Pyrene
Number of
emission
factors
1
1
1
2
1
2
2
4
2
1
2
3
2
2
3
2
6
5
2
1
1
9
6
Median emission
factor: Ib/trillion
Btu
9.4 x 10'6
1.0x1 0"5
1.5 x 10'6
9.6 x 10'6
1.0x 10'5
7.3x 10'6
5.0 x 10'6
0.017
0.36
0.017
0.015
0.030
0.033
0.021
0.021
0.0081
0.016
0.021
0.024
0.015
0.016
0.025
0.037
Computer
program: 1990
total tons
5.7 x 1(T6
6.2 x 10-*
8.8 x 10'7
5.8 x 1CTS
6.2 x 1
-------
Table A-6. Median Emission Factors, Determined from Test Report
Data, and Total 1990 and Total 2010 HAP Emissions, Projected with
the Emission Factor Program for Organic HAPs from Gas-fired Units
Gas-fired units:
organic HAPs
Benzene
Formaldehyde
Naphthalene
Toluene
2-methylnaphthalene
Fluoranthene
Fluorene
1-phenanthrene
Pyrene
Number of
emission
factors
1
8
2
2
2
1
1
2
1
Median emission
factor: Ib/ trillion Btu *
1.4
35.5
0.70
10
0.026
0.0028
0.0026
0.013
0.0049
Computer
program: 1990
total tons
1.8
55
0.66
13
0.025
0.0034
0.0034
0.016
0.0061
Computer
program: 2010
total tons
2.7
83
1.0
19
0.038
0.0055
0.0051
0.024
0.0094
The geometric mean of kg/109 cubic foot of natural gas factors were used in
estimating organic emissions from gas-fired units. These median emission factors
are in this table for comparison with previous tables and were not used directly to
estimate organic HAPs.
A-9
-------
Appendix B - Matrix of Electric Utility Steam-Generating Units
and Emission Test Sites
Table B-l is a matrix of utility boiler types and
configurations showing each configuration's percentage of the
total fossil-fuel-fired electric utility industry and the number
of emission test sites analyzed in this report that fit into that
category's configuration. Table B-2 shows the emission test
sites whose data were used to develop this Report to Congress.
Some sites are known only by their provider number because of
nondisclosure agreements.
-------
This page is intentionally blank.
-------
S
•d
CD
N
•H
d
10
01
M
o
n
o>
ij
«d
13
(D
4J
•H
§
S
W
•*-• »
= "S
13 a
OS
1
#
'«
&
s
6
to
4-1
'c
«£
. M
n
CJ
0
W °m
• >
•o £
g 0
.2 £
'I
= £
3
cn
®
1%
.4*
r o
(0 U
Q.
* 2~
>t
*-* C
E°
g §
S^
CQ
#
4-1
C
0>
4-^
i
L»
3
*4—
3
CD
"o>
LL
"5
LL
CM
r>
CD
r>
in
in
00
co
co
in
CD
CO
T-
£
CM
£
V)
z
o
Q.
co
"5
"5.
o
U
co
LU
0
4-»
4-*
O
43
X
O
in
V
H Bituminous coal
CO
T~
in
s
00
CO
in
i^
TJ-
in
en
^-
00
00
CO
CO
Q.
CO
o
Q.
CO
co
"5.
o
U
0}
c
o
1 Natural gas
r>
in
00
t"
8
o
in
CM
•*
O
00
CO
r—
*t
r^
CM
£
CO
o
0.
CO
0)
3
"5.
O
U
CO
LU
o
4-*
4^
JB
>.
Q
in
CO
VI
T3
ID
in
AJ
I Bituminous coal
CM
CO
00
o
r-»
00
CO
CD
O
00
in
Tt
^
0)
(O
Q.
w
z
o
Q.
CO
«
"a
o
U
CO
LU
Dry bottom
in
V
H Subbituminous coal
co
o>
O)
CO
o
Tj-
00
•*
0)
O)
00
O)
'fr
o>
05
CO
Q.
CO
o
Q.
CO
ffi
ll
0
o
0
1
in
V
'5
"3
£
I
O
^"
CO
CO
0
Tt
CM
r»
0
00
Tt
in
in
0
<
Z
0)
o
II Natural gas
CM
't
0)
CM
§
O)
O)
CO
>*
co
O)
*~
o>
CO
Q
o
LL
CO
LU
o
4^
o
^J
>
Q
in
CO
VI
•g
(D
in
A]
U Bituminous coal
CM
r—
r*
CM
O
CO
CO
CO
in
CO
CO
CO
*"
r^
CN
Q
U
LL
CO
LU
O
4^
o
J3
Q
ID
V
y Subbituminous coal
CM
CM
CM
CM
0
t^
00
l->
O
t—
r^
CM
CM
in
>t
CO
Q.
CO
Z
o
Q.
CO
"®
"5.
o
O
CO
LU
Wet bottom
in
co
VI
•o
1C
in
Al
1 Bituminous coal
^~
0)
o
CM
o
CM
in
^t
O
r-»
CM
CM
in
<*
CO
Q.
CO
o
Q.
CO
"5
"5.
o
O
CO
LU
in
V
'6
1
LL
C
00
0
CM
in
in
CO
o
rt
o
en
*~
CO
CO
CO
Q.
CO
o
Q.
CO
~v
~CL
O
O
CD
0
+3
3
S
in
V
'5
1
-
in
r^
O
p%
^
t
rx
00
in
.
Q
in
V
1 Bituminous coal
CM
CO
t
O
1^1
CO
t~
1^
T—
O)
o
00
o
a
LL
CO
LU
O
4-*
4-*
£
>
D
in
V
| Bituminous coal
CM
O
O
CM
O
§
O
CD
in
in
o
-
o>
.b
in
fa
"5
O
CO
LU
2
4-*
O
J3
>•
O
in
CO
VI
•o
(D
in
Al
| Bituminous coal
-
in
o
rj-
r^
CO
p^
in
CO
in
*~
CO
in
o.
C/3
O
O.
CO
"3
"5.
o
O
CO
o
Z
in
CO
VI
•°
a
in
^
Al
'5
"Z
LL
CO
in
^-
o
r»
CO
p>.
in
CD
o
00
CO
Q.
CO
z
o
P-
co
"5
"5.
o
U
CO
LU
E
o
4-f
4^
0
.a
4-1
CD
in
V
Subbituminous coal
-
0
r-
O
r-»
o
o
in
in
0
in
o
o
Q
C3
LL
i.
01
4-*
O
.O
(D
U.
2
o
£>
>•
O
It)
V
| Subbituminous coal
0
in
o
*t
in
i»-
CM
in
in
in
O
-
O
(D
LL
CO
LU
E
o
s
J3
s-
Q
in
co
Al
"5
8
§
+*
m
1
en
4i£
'E
"5
en
«
a
5
6
CM
in
q
V
in
in
O
V
w
V
4->
O
i
(C
LL
O
4-1
O
.a
>
Q
in
V
D Bituminous coal
-
in
q
i—
•V
in
in
O
V
CO
LU
U
m
LL
<
in
V
I! Subbituminous coal
B-l
-------
•0
I
•ri
I
O
n
H
•3
IA
11
CO
o 2
z
Q)
5
*
d
z.
in
'c
#
. CO
O •—
Z c
*• 3
P*
""" X
® >
« *•*
11
T3 **
" C
"5
CO
CD
^n *—
Particuli
contro
• a.
i-~ "E
o "5
15
m ~-
*
S
c
o
u
3
2^
3
CO
CD
3
LL
_
3
LL
CN
in
q
V
in
m
d
V
«t^
Q
CO
CD
4V
o
Ja
CO
LL
E
o
Q
in
^
V
^
O
u
to
o
c
1
I Subbitu
^
m
q
V
in
in
d
V
CD
4V
O
J5
CO
LL
LL
m
»~
V
1
CO
m
*"*
in
q
V
in
in
d
V
-^
Q
CO
CD
4V
.0
J5
to
LL
E
O
1
o
in
^
V
To
o
u
en
— O
CO C
i E
1 Bitumin
Subbitu
in
q
V
tn
in
d
V
tr
u
CO
in
^
V
'6
1
LL
in
q
V
in
in
d
V
LL
LL
in
f-
V
'6
"5
LL
in
q
V
in
in
d
V
^
0
CO
LL
LL
E
o
4V
O
JO
Q
in
CO
VI
"O
C
CO
in
^
Al
CD
8
in
O
I Bitumin
in
q
V
in
in
d
V
o
0-
SP/COH
iii
E
o
s
JO
Q
in
^
V
(D
8
CO
1
.•e
m
in
q
V
in
in
d
V
.0
CJ
'E
15
T3 at
Combine
cycle gai
turbine
CO
CO
CD
To
3
a
in
q
V
tn
in
d
V
CD
4V
JO
CO
LL
O
CO
LL
in
T-
V
"5
o
u
o
.1
in
q
V
in
in
d
V
cc
O
CO
CO
5
CD
4V
.0
co
LL
E
o
4V
vO
1
in
co
VI
c
CO
in
^
Al
^
Cl
C
1 Bitumin
(O
1
3
Q)
(D
E
3
1
£
i
0
o
o
c
3
CD
JC
a>
c
a
o
10
c
4V
"5
Q.
CD
CA
"S
CD
E
o
o
II
E
o .
° i
jo 2
4^ tn
CD >
$ <"
^ To
s §
2 i
en CD
si
CD
i I
r~ m
Jk^ l/J
en _.
co £
r- JO
S CO
CD Q.
> CO
li
If
11
3 U
CD CD
WSA = wet sulfuric acid (system)
Coal washing: coal is washed befor
Compi fuel/SIP or NSPS = fuel burn
requirements.
o
CO
"5
c
o
o
3
CD
TO
C
'5
I
CO
D.
CO
Z
Q
a.
CO
CD
CO
£
1
O
en
'c
3
CD
en
CD
«v
4V
0.
0)
u
X
CD
CO
Q.
•z
C3
CL
CO
_^=
I
CD
U
C
CO
"Q.
O
in
CO
TO
"5
3
CD
C
CD
£
"5 in
t:
8|
U
SCR = selective catalytic reduction
Note 1 : These units are essentially
Jo
2
3
8
\J
I
3
CD
O
2
a
—
5~
4^
Q
M
O
VI
a
en
CO
TO
_c
CA
CO
5
"3
8
e
0
Q.
CO
o
CD
in
o
5
u
8
.£
3
CD
in
£
1-
13
CD
in
CO
5
CA
~B
O
O
in
g
E
3
"5
In
O
"co
•o
emissions.
Note 2: From the research performe
collapsed into other unit types.
B-2
-------
CN
ffl
O
H
•8
S
10
r
Q
i
cc
i
1
£
0
CC
X
fi
0
i
"
g
w
^
1
o
U
E £
1*
5
"5
CO
TB
LL
"CD
LL
•
35
J
et
_
CL
CO
CO
CM
,_
C
o
4-*
in
1
LL
o
DC
CL
CO
LU
4—*
0
§
E
3
1
V
15
o
0
inous >
E
m
S
CO
00
LU
§
—.
~
CO
CM
CM
C
o
4-1
1
LL
o
cc
LL
LL
^
Q
3
o
J
15
o
u
V)
minou
3
15
3
CO
"0
CO
O
m
LU
O
0
_
CL
m
CM
co
15 ft
113
ill
0 .£ co
<5 c £
LU ,£
CL
CO
LU
^
Q
E
3
'•is
Q)
15
o
u
in
O
1
m
"co
CO
O
UJ
0
O
_
CL
CO
CN
,,,
_0
"5
i
CD
O
ESP/FG
^
Q
3
o
~J
15
o
u
£
C
O)
12
1
o
1
o
LU
0
Q
„
Cu
CO
ro
CN
LO
j>
"0
4^
CO
m
a.
co
LU
0
^
E
3
1
W
1
u
inous
E
m
CO
j£
Z
LU
O
a
—
CL
CO
g>
^
(O
_0
"0
(D
00
CO
CO
LL
LL
0
^
E
_3
^
^
15
o
0
CO
o
1
m
d
1
Z
LU
O
Q
sr*
CL
CO
ro
™
^
1
CO
0
in 0
c '£
|-
g
CO
) t
LL
Q
CO
^*
Q
g
O
~"
"fa
o
u
in
o
3
15
Si
CO
pringerville
CO
LU
O
O
„
t
CO
O)
^
00
.2
CD
CC
cc
00
CL
CO
LU
^
a
E
.3
"S
5
15
o
u
inous
E
m
ca
2
CO
LU
O
0
_
CL
i
^
CO
"o
a
£
CL
CO
LU
4^
0
^
g
0
15
o
u
CO
minou
3
2
JD
3
CO
CO
<
o
CL
CO
a.
CM
CD
i^
o
"5
9-
c
CL
LU
^
O
5
o
15
o
u
en
minou
3
15
JO
3
CO
^
co
a>
o
Q
00
O
CL
CO
CL
CM
LO
«-
"5
9-
£
CL
CO
LU
O
00
LL
S
o
15
8
m
minou
3
2
JD
3
CO
CM
C?
O
u
CO
m
o
0-
co
a.
CM
ro
^
CM
"5
9-
0
_c
Q.
CO
LU
^*
o
3
o
15
o
u
CA
minou
3
2
£>
3
CO
co
LO
^r
co
1
±
U
OL
CO
CL
CM
5
CM
CO
"o
9-
0
c
LL
LL
^
O
$
O
15
8
CO
minou
3
£
3
CO
r-.
iverside 6,'
cc
CO
CL
CM
CO
«35
•*
"5
9-
o
_c
CL
CO
LU
4^
a>
§
3
o
15
0
u
in
minou
3
^
"§
CO
00
«
1
'£
a
CO
CL
CO
0
r-
o
0)
^
r-
CD
LO
r-
"5
9-
0
c
o
LL
^
D
5
0
15
0
u
CO
minou
3
•_5
•§
CO
CM
herburne 1
CO
O
CL
CO
CL
ro
0
o
CO
(0
CO
"5
0
_c
LL
LL
Q
CO
^*
Q
5
o
15
8
CO
minou
3
2
JD
CO
herburne 3
CO
U
CL
CO
_^
Q.
CM
ro
O
0)
.2
CC
LL
LL
O
m
LL
3
o
15
o
u
inous
E
S
o
2
CO
E
CL
LU
a.
CO
CM
O
.2
CO
cc
CL
CO
UJ
0
^
3
o
15
o
u
CO
minou
3
15
3
CO
CN
O
a>
4-'
CO
E
CL
LU
£
CO
en
0
»—
CM"
O)
o
<-
.2
cc
o
o
CL
CO
LU
«^
O
g
o
15
o
0
in
minou
3
15
J3
3
CO
2
CO
E
a.
LU
—
Q.
CO
ro
O
CM
1
CO
0
w m
0 C
J2 —
o
CO
Q. °-
UJW
± 6
^"
a
E
.3
"S
2
15
o
o
V)
O
1
m
o
2
c/5
E
52
cc
CL
LU
B-3
-------
2
^
r
&
8
e
£
5
CD
CC
o
I
i
»
.2
-2_
2
I
o
It
0 5"
ea
S
"3
10
"3
u.
^
LL
2
55
|
I
a
CO
o
CM
CM
U
w
CO
CD
%£
cc 2
E'-S
o c
£ —
o
to
i g.
LU LU
I 0
c:
o
E
^
CD
15
8
co
3
O
E
3
m
o"
Z
^
g
o
|
2
CO
E
52
cc
CL
LU
CL^
1
CO
CM
.2
CD
cc
LL
U.
a
to
t_
a
5
5
"5
0
^ o
CO CO
0 0
c ?
*— ^
E 1
2 3
co 2
3
CO
^
2
*J
to
£
CL
LU
O.
0)
CO
CO
tn
CM
5}-
CM
C3
CD
O
OL
CO
LU
k.
O
§
5
^_
6
CM
2
55
cc
CL
UJ
67
i
LO
in
CM
Z
OL
CO
LU
%
5
E
3
'•s
"co
o
u
Cfl
3
O
1
3
s
^
2
55
oc
CL
LU
Z
CO
CM
Carnot
u.
LL
C?
o
g
o
"CD
O
u
in
3
O
1
3
m
in
CD
CO
LU
CL
I
^
CM
O
CO
O
CC
o
to
>
Q
g
O
~«
O
CD
CO
CC
Q.
LU
g
05
CO
CM
O
i
o
0.
to
LU
^
Q
g
3
o
00
CD
to
OC
OL
LU
g
o>
O3
CM
1
CD
O
0.
to
LU
k.
o
g
o
^_
o
O)
e
to
oc
OL
LU
CO
en
o
CM
0)
0
CO
.2
CD
cc
0
u
LL
CL
to
LU
w
Q
E
_3
1
15
o
u
CO
3
O
1
2
m
CM
CD
to
CC
CL
LU
Z
r_
CO
Z
CD
O
^
a
co
CD
a
o
CM
CD
to
OC
CL
LU
Z
CM
CO
z
CD
O
^
O
en
CD
O
CM
2
55
oc
Q.
LU
a
co
o>
CM
CO
CO
Radian
LL
LL
2
>
Q
«_
6
CO
CD
X
CO
cc
0.
LU
t
CM
*~
^
CO
Radian
LL
ii
SDA/PJI
>
Q
E
3
1
8
0
CO
3
O
E
3
m
^
2
to
oc
0.
LU
£
CM
o
in
CO
Radian
CL
to
LU
^
Q
E
_3
i
"co
O
u
CA
3
O
"E
3
m
in
2
CO
cc
0.
LU
i
CO
05
1—
CO
CO
.2
•5
CD
CC
0.
to
LU
>_
Q
E
3
Tg
"5
o
u
CO
3
O
1
3
CO
<
O
co
2
(75
cc
OL
LU
a.
CO
g>
T-
CD
CO
.2
CO
cc
a.
to
uu
w.
0
E
.3
V
"co
0
u
CO
3
O
1
3
CD
X
o
z
5
o
J
LL
O
CO
2
to
cc
CL
LU
E:
CO
O)
1^
CO
.2
a
g
o
"5
o
o
in
3
O
1
3
CO
01
2
to
cc
0.
LU
OL_
CO
en
LO
o>
CO
CO
cc
Q
ESP/FG
k.
O
£
_3
1
"to
O
U
en
3
O
'E
3
m
^
CM
2
55
oc
OL
LU
(X
t
tN
CO
O)
CO
O
^
ian/Carm
•o
CD
CC
0.
CO
LU
l_
o
g
o
"co
O
u
CO
3
O
1
3
m
CM
CM
2
55
oc
CL
LU
CL.
CO
0)
co
,—
^
1
b.
CD
O
CD
O
>
Q
g
0
^
6
CO
0
2
CO
oc
0.
LU
B-4
-------
I
•H
I
U
es
i
(Q
(J)
•o
r
o
a
to
ec
i
£
£
|
«-*
to
H
Control"
I I
OQ
*•>
5
LU
u.
CD
CO
Provider*
Q.
CO
CD
CO
O
O
CD
O
>
O
g
°,
o
S
CD
CO
E
CL
UJ
CO
0)
CO
o
CO
o
ID
o
Q
5
o
O
in
O
CD
*^
CO
E
a.
UJ
CO
o>
CO
o
CO
O
o
>
0
5
o
0
(D
0
CD
CO
E
a.
LU
CO
o>
CO
o
CO
O
CD
O
g
O
m^
O
o
03
CO
1
UJ
CO
o>
CO
o
CO
o
e
o
0
EE
3
"8
2
O
CO
O
«
.•E
CO
E
0.
in
CO
0)
CO
I
1
CD
CD
O
Z
Q
5
o
Ml
O
o>
o
CD
+^
CO
£
UJ
CO
o>
en
CN
a
o
UJ
Water
injection
(A
CO
CO
c
.0
E _
U C
c l_
o .3
ti
j: co
§0
I "5
. x
1- 0
0.
LU
CO
O)
en
CO
D.
O
UJ
Q.
CO
LU
Q
5
o
O
o
a
t
o
Z
a.
LU
CO
0)
o
a
o
LU
LL
LL
O
OQ
LL
5
0
__
CO
o
2
D)
_l
^^^
0)
o
a.
7
1-
^-^
•o
o
g
a
I
a.
Ul
CO
0)
T"
If)
a
S
LU
O
x
Q
(4
co
3
O
>-
S
in
1
w
O
LU
CO
O>
CM
CD
a
o
4-*
UJ
ESP/FGD
x
O
E
3
•o
Q)
;g
"<5
o
u
CO
3
0
E
3
m
£
•4?
C
^
LU
O Z UJ CO
D. LL
B-5
-------
Appendix B: References
1. Preliminary draft emissions report for Baldwin Power Station
- Unit 2 (Illinois Power Company) for the Comprehensive
Assessment of Toxic Emissions from Coal-fired Power Plants,
prepared by Roy F. Weston, Inc., for the Department of
Energy/Pittsburgh Energy Technology Center (DOE/PETC), DOE
contract # DE-AC22-93PC93255, Weston project # 10016-011,
Weston report # DOE018G.RP1. December 1993.
2. Preliminary draft emissions report for Boswell Energy Center
- Unit 2 (Minnesota Power Company) for the Comprehensive
Assessment of Toxic Emissions from Coal-fired Power Plants,
prepared by Roy F. Weston, Inc., for the Department of
Energy/Pittsburgh Energy Technology Center (DOE/PETC), DOE
contract # DE-AC22-93PC93255, Weston project # 10016-011,
Weston report # DOE017G.RP1. December 1993.
3. Preliminary draft emissions report for Cardinal
Station - Unit 1 (American Electric Power) for the
Comprehensive Assessment of Toxic Emissions from Coal-fired
Power Plants, prepared by Energy and Environmental Research
Corp. for the Department of Energy/Pittsburgh Energy
Technology Center (DOE/PETC), DOE contract
# DE-AC22-93PC93252. December 1993.
4. Preliminary draft emissions report for Coal Creek
Station - Unit 2 (Cooperative Power Association) for the
Comprehensive Assessment of Toxic Emissions from Coal-fired
Power Plants, prepared by Battelle for the Department of
Energy/Pittsburgh Energy Technology Center (DOE/PETC), DOE
contract # DE-AC22-93PC93251. December 1993.
5. Preliminary draft emissions report for Niles Station Boiler
No. 2 (Ohio Edison) for the Comprehensive Assessment of
Toxic Emissions from Coal-fired Power Plants, prepared by
Battelle for the Department of Energy/Pittsburgh Energy
Technology Center (DOE/PETC), DOE contract
# DE-AC22-93PC93251. December 1993.
6. Preliminary draft emissions report for Niles Station Boiler
No. 2 with NOX control (Ohio Edison) for the Comprehensive
Assessment of Toxic Emissions from Coal-fired Power Plants,
prepared by Battelle for the Department of Energy/Pittsburgh
Energy Technology Center (DOE/PETC), DOE contract
# DE-AC22-93PC93251. December 1993.
B-6
-------
7. Preliminary draft emissions report for Springerville
Generating Station Unit No. 2 (Tucson Electric Power
Company) for the Comprehensive Assessment of Toxic Emissions
from Coal-fired Power Plants, prepared by Southern Research
Institute for the Department of Energy/Pittsburgh Energy
Technology Center (DOE/PETC), DOE contract
# DE-AC22-93PC93254, SRI report No. SRI-ENV-93-1049-7960.
December 1993.
8. Preliminary draft emissions report for Plant Yates Unit
No. 1 (Georgia Power Company) for the Comprehensive
Assessment of Toxic Emissions from Coal-fired Power Plants,
prepared by Electric Power Research Institute for the
Department of Energy/Pittsburgh Energy Technology Center
(DOE/PETC), EPRI report No. DCN 93-643-004-03. December
1993.
9. Results of the Air Toxic Emission Study on the No. 1 Boiler
at the NSPC A.S. King Plant, prepared by Interpoll
Laboratories, Inc., for NSPC, report No. 1-3304. November
1991.
10. Results of the Air Toxic Emission Study on the Nos. 1, 3, &
4 Boilers at the NSPC Black Dog Plant, prepared by Interpoll
Laboratories, Inc., for NSPC, report No. 1-3451. January
1992.
11. Results of the Air Toxic Emission Study on the No. 2 Boiler
at the NSPC Black Dog Plant, prepared by Interpoll
Laboratories, Inc., for NSPC, report No. 2-3496. May 1992.
12. Results of the Air Toxic Emission Study on the Nos. 3, 4, 5,
& 6 Boilers at the NSPC High Bridge Plant, prepared by
Interpoll Laboratories, Inc., for NSPC, report No. 1-3453.
January 1992.
13. Results of the Air Toxic Emission Study on the Nos. 6 & 7
Boilers at the NSPC Riverside Plant, prepared by Interpoll
Laboratories, Inc., for NSPC, report No. 1-3468A. February
1992.
14. Results of the Air Toxic Emission Study on the No. 8 Boiler
at the NSPC Riverside Plant, prepared by Interpoll
Laboratories, Inc., for NSPC, report No. 2-3590. September
1992.
B-7
-------
15. Results of the Air Toxic Emission Study on the No. 1 & 2
Boilers at the NSPC Sherburne Plant, prepared by Interpoll
Laboratories, Inc., for NSPC, report No. 0-3053. July
1990/October 1991.
16. Results of the Mercury Removal Tests on Units 1 & 2, and the
Unit 3 Scrubber System at the NSPC Sherburne Plant, prepared
by Interpoll Laboratories, Inc., for NSPC, report No.
1-3409. October 1991.
17. Results of the May 1, 1990, Trace Metal Characterization
Study on Units 1 & 2 at the NSPC Sherburne Plant, prepared
by Interpoll Laboratories, Inc., for NSPC, report No.
0-3033E. July 1990.
18. Results of the Air Toxic Emission Study on the No. 3 Boiler
at the NSPC Sherburne Plant, prepared by Interpoll
Laboratories, Inc., for NSPC, report No. 0-3005. June
1990/October 1991.
19. Preliminary draft emissions report for EPRI Site 10, Field
Chemical Emissions Monitoring Project, prepared by Radian
Corp. for EPRI. EPRI report No. DCN 92-213-152-35. October
1992.
20. Preliminary draft emissions report for EPRI Site 102, Field
Chemical Emissions Monitoring Project, prepared by Radian
Corp. for EPRI. EPRI report No. DCN 92-213-152-35.
February 1993.
21. Preliminary draft emissions report (and mercury retest) for
EPRI Site 11, Field Chemical Emissions Monitoring Project,
prepared by Radian Corp. for EPRI. EPRI report Nos. DCN
92-213-152-24 and DCN 92-213-152-48. November 1992/October
1993.
22. Preliminary draft emissions report for EPRI Site 110
(baseline and with NOX control) for the EPRI PISCES Study,
prepared by Southern Research Institute, SRI report No.
SRI-ENV-92-796-7496. October 1993.
23. Preliminary draft emissions report for EPRI Site 111, Field
Chemical Emissions Monitoring Project, prepared by Radian
Corporation for EPRI. EPRI report No. DCN 93-213-152-42.
January 1994.
B-8
-------
24. Preliminary draft emissions report for EPRI Site 112, Field
Chemical Emissions Monitoring Project, prepared by Carnot
for EPRI. Carnot report No. EPRIE-10106/R016C374.T. March
1994.
25. Preliminary draft emissions report for EPRI Site 114, Field
Chemical Emissions Monitoring Project, prepared by Radian
Corporation for EPRI. EPRI report No. DCN 92-213-152-51.
May 1994.
26. Preliminary draft emissions report for EPRI Site 115, Field
Chemical Emissions Monitoring Project, prepared by Carnot
for EPRI. Carnot report No. EPRIE-10106/R022C855.T.
27. Preliminary draft emissions report for EPRI Site 117, Field
Chemical Emissions Monitoring Project, prepared by Carnot
for EPRI. Carnot report No. EPRIE-10106/R120C844.T.
January 1994.
28. Preliminary draft emissions report for EPRI Site 118, Field
Chemical Emissions Monitoring Project, prepared by Carnot
for EPRI. Carnot report No. EPRIE-10106/R140C928.T.
January 1994.
29. Preliminary draft emissions report for EPRI Site 119, Field
Chemical Emissions Monitoring Project, prepared by Carnot
for EPRI. Carnot report No. EPRIE-10106/R027C882.T.
January 1994.
30. Preliminary draft emissions report (and mercury retest) for
EPRI Site 12, Field Chemical Emissions Monitoring Project,
prepared by Radian Corp. for EPRI. EPRI report Nos. DCN
92-213-152-27 and DCN 93-213-152-49. November 1992/October
1993.
31. Preliminary draft emissions supplement for EPRI Site 120,
Field Chemical Emissions Monitoring Project.
32. Preliminary draft emissions report for EPRI Site 121, Field
Chemical Emissions Monitoring Project, prepared by Carnot
for EPRI. Carnot report No. EPRIE-12102/R120E916.T.
December 1994.
33. Preliminary draft emissions report for EPRI Site 13, Field
Chemical Emissions Monitoring Project, prepared by Radian
Corp. for EPRI. EPRI report No. DCN 93-213-152-36.
February 1993.
B-9
-------
34. Preliminary draft emissions report for EPRI Site 14, Field
Chemical Emissions Monitoring Project, prepared by Radian
Corp. for EPRI. EPRI report No. DCN 93-213-152-28.
November 1992.
35. Preliminary draft emissions report for EPRI Site 15, Field
Chemical Emissions Monitoring Project, prepared by Radian
Corp. for EPRI. EPRI report No. DCN 93-213-152-26. October
1992.
36. Preliminary draft emissions report for EPRI Site 16 (OFA and
OFA/Low NOX) for the Clean Coal Technology Project (CCT),
prepared by Electric Power Research Institute, for the
Department of Energy/Pittsburgh Energy Technology Center
(DOE/PETC), EPRI report No. DCN 93-209-061-01. November
1993.
37. Preliminary draft emissions report for EPRI Site 18, Field
Chemical Emissions Monitoring Project, prepared by Radian
Corp. for EPRI. EPRI report No. DCN 93-213-152-43. April
1993.
38. Preliminary draft emissions report for EPRI Site 19, Field
Chemical Emissions Monitoring Project, prepared by Radian
Corp. for EPRI. EPRI report No. DCN 93-213-152-41. April
1993.
39. Preliminary draft emissions report for EPRI Site 21, Field
Chemical Emissions Monitoring Project, prepared by Radian
Corp. for EPRI. EPRI report No. DCN 93-213-152-39. May
1993.
40. Preliminary draft emissions report for EPRI Site 22, Field
Chemical Emissions Monitoring Project, prepared by Radian
Corp. and Carnot for EPRI. EPRI report No. DCN
93-213-152-53. February 1994.
41. Preliminary draft emissions report for EPRI Sites 103-109,
Field Chemical Emissions Monitoring Project: Emissions
Report for Sites 103 - 109, prepared by Radian Corp. for
EPRI. March 1993.
42. Final electric utility combined cycle gas-fired gas turbine
emission test report for T.H. Wharton Electric Generating
Station (Houston Lighting and Power Company), prepared by
Entropy, Inc., for the U. S. Environmental Protection
Agency, Emissions Measurement Branch (EPA/EMB), EMB report
No. 93-UTL-2. May 1994.
B-10
-------
43. Final electric utility fuel oil-fired electric utility
boiler emission test report for Northport One
powerplant - Unit 1 (Long Island Lighting Corporation),
prepared by Entropy, Inc., for the U.S. Environmental
Protection Agency, Emissions Measurement Branch (EPA/EMB),
EMB report No. 93-UTL-4. April 1994.
44. Final electric utility coal-fired fluidized bed boiler
emission test report for TNP One - Unit 2 (Texas - New
Mexico Power Company), prepared by Entropy, Inc., for the
U.S. Environmental Protection Agency, Emissions Measurement
Branch (EPA/EMB), EMB report No. 93-UTL-l. June 1994.
45. Final electric utility gas-fired boiler emission test report
for Greens Bayou Electric Generating Station - Unit 5
(Houston Lighting and Power Company), prepared by Entropy,
Inc., for the U. S. Environmental Protection Agency,
Emissions Measurement Branch (EPA/EMB), EMB report No.
93-UTL-3. May 1994.
46. Final electric utility coal-fired boiler emission test
report for Kintigh - Unit 1 (New York State Electric and Gas
Company), prepared by Entropy, Inc., for the U.S.
Environmental Protection Agency, Emissions Measurement
Branch (EPA/EMB), EMB report No. 93-UTL-5. June 1994.
B-ll
-------
Appendix C - Listing of Emission Modification Factors for Trace
Elements Used in the Individual Boiler Analysis
Note: The following test reports were not used to develop
emission modification factors (EMFs) for the reasons listed
below. Northern States Power's (NSP) A.S. King unit is the same
test site as the Electric Power Research Institute's (EPRI's)
Site 102, and the EPA chose to use the EPRI test report.
Northern States Power's Sherco unit 1 and 2 were not used to
develop boiler EMFs because no coal composition data were
provided. Northern States Power's Black Dog unit 1 was not used
to develop boiler EMFs because tangentially-fired emissions were
combined with emissions from two front-fired boilers. Finally,
NSPC's High Bridge was not used to develop boiler EMFs because
the test report was missing the coal feed rate during testing.
-------
This page is intentionally blank.
-------
Table C-l. Tested EMFs and Geometric Means used in the Emission
Factor Program for Circulating Fluidized Bed Furnaces
(coal-fired)
Unit Name
ARSENIC
BERYLLIUM
CHROMIUM
COBALT
LEAD
MANGANESE
MERCURY
NICKEL
SELENIUM
EPRI Site 10
1.00
0.77
0.40
1.00
0.49
0.59
1.00
1.00
1.00
NSP - Black Dog #2
0.59
0.41
0.54
0.36
0.68
1.00
0.45
0.71
EMF (Geometric
mean)
0.77
0.56
0.46
1.00
0.42
0.63
1.00
0.67
0.84
Geometric standard
deviation
1.44
1.56
1.25
N/A
1.24
1.11
1.00
1.76
1.27
Table C-2. Tested EMFs and Geometric Means used in the Emission
Factor Program for Tangentially-fired, dry-bottom furnace with
NOX control (coal-fired)
Unit Name
ANTIMONY
ARSENIC
BERYLLIUM
CADMIUM
CHROMIUM
COBALT
LEAD
MANGANESE
MERCURY
NICKEL
SELENIUM
EPRI Site 1 1
0.92
0.79
0.35
0.72
0.92
1.00
0.98
0.29
0.25
1.00
DOE - Coal
Creek
0.01
0.69
0.35
0.11
1.00
0.61
0.29
0.59
0.85
1.00
0.38
DOE-
Springerville
0.03
0.29
0.87
1.00
0.68
0.73
0.19
0.72
1.00
0.70
0.93
EMF (Geometric
mean)
0.02
0.57
0.62
0.34
0.79
0.74
0.38
0.75
0.63
0.56
0.71
Geometric
standard
deviation
2.17
1.83
1.64
3.01
1.23
1.23
2.38
1.30
1.98
2.07
1.70
C-l
-------
Table C-3. Tested EMFs and Geometric Means used in the Emission
Factor Program for Tangentially-fired, dry-bottom furnace without
NOX control (coal-fired)
Unit Name
ANTIMONY
ARSENIC
BERYLLIUM
CADMIUM
CHROMIUM
COBALT
LEAD
MANGANESE
MERCURY
NICKEL
SELENIUM
EPRI Site 1 5
0.60
0.54
0.01
0.58
0.94
1.00
0.81
1.00
0.43
0.70
DOE - Yates
0.67
1.00
1.00
1.00
1.00
0.97
1.00
1.00
1.00
0.84
0.70
EMF (Geometric
mean)
0.67
0.77
0.74
0.11
0.76
0.95
1.00
0.90
1.00
0.60
0.70
Geometric standard
deviation
N/A
1.44
1.54
22.68
1.47
1.02
1.00
1.16
1.00
1.59
1.01
Table C-4. Tested EMFs and Geometric Means used in the Emission
Factor Program for Opposed-fired', dry-bottom furnace with NOX
control (coal-fired)
Unit Name
ANTIMONY
ARSENIC
BERYLLIUM
CADMIUM
CHROMIUM
COBALT
LEAD
MANGANESE
MERCURY
NICKEL
SELENIUM
EPRI Site
12
1.00
1.00
0.14
1.00
1.00
1.00
1.00
0.74
0.29
1.00
EPRI Site
14
0.50
0.92
0.02
0.67
1.00
0.79
0.93
0.74
0.37
0.05
NSP-
Sherburne
#3
0.79
0.58
0.99
0.49
0.49
1.00
0.67
0.21
EPRI
Site 1 1 1
0.11
0.05
0.20
EPRI Site
16 w/OFA
and LNOX
Burners
0.80
1.00
0.82
0.11
0.69
0.66
0.66
0.88
1.00
0.54
0.37
EPRI Site
16
w/OFA
1.00
1.00
1.00
1.00
0.58
0.61
1.00
0.60
0.64
0.33
1.00
EMF
(Geometric
mean)
0.90
0.59
0.85
0.16
0.55
0.80
0.76
0.84
0.81
0.42
0.33
Geometric
standard
deviation
1.17
2.41
1.25
4.68
1.72
1.31
1.35
1.26
1.22
1.42
3.51
C-2
-------
Table C-5. Tested EMFs and Geometric Means used in the Emission
Factor Program for Front-fired, dry-bottom furnace without NOX
control (coal-fired)
Unit Name
ANTIMONY
ARSENIC
BERYLLIUM
CADMIUM
CHROMIUM
COBALT
LEAD
MANGANESE
MERCURY
NICKEL
SELENIUM
NSP - Riverside
#6-7
0.20
0.99
0.40
0.25
1.00
0.19
0.77
1.00
0.78
1.00
DOE - Cardinal
0.08
0.91
0.96
1.00
0.61
0.96
1.00
0.27
0.41
0.76
0.07
DOE - Boswell
0.59
0.23
0.60
1.00
1.00
0.98
0.42
0.57
0.87
1.00
0.14
EMF (Geometric
mean)
0.21
0.59
0.62
0.63
0.85
0.97
0.43
0.49
0.71
0.84
0.21
Geometric
standard
deviation
2.74
2.25
1.54
2.22
1.33
1.02
2.27
1.73
1.62
1.16
4.09
Table C-6. Tested EMFs and Geometric Means used in the Emission
Factor Program for Cyclone-fired, wet-bottom furnace without NOX
control (coal-fired)
Unit Name
ANTIMONY
ARSENIC
BERYLLIUM
CADMIUM
CHROMIUM
COBALT
LEAD
MANGANESE
MERCURY
NICKEL
SELENIUM
EPRI Site
102
0.48
0.04
0.02
0.25
0.21
0.61
0.33
1.00
0.30
0.65
NSP-
Riverside
#8
0.61
0.51
0.08
0.16
0.22
0.38
0.13
1.00
0.12
1.00
EPRI Site
114
0.15
0.15
0.01
0.30
0.50
0.20
0.73
0.72
0.26
EPRI Site
114, 2nd
Test
0.25
0.15
0.01
0.23
0.84
0.18
0.54
0.31
1.00
DOE-
Niles #2
1.00
0.58
0.26
0.11
0.28
0.20
0.56
0.15
1.00
0.29
1.00
DOE-
Niles #2
w/NOx
Control
0.92
0.85
0.30
0.20
0.35
0.33
0.60
0.19
1.00
0.39
0.92
EMF
(Geometric
mean)
0.82
0.41
0.13
0.04
0.27
0.24
0.56
0.19
0.86
0.31
0.74
Geometric
standard
deviation
1.30
1.86
2.09
5.01
1.19
1.31
1.30
1.37
1.30
1.76
1.70
C-3
-------
Table C-7. Tested EMFs and Geometric Means used in the Emission
Factor Program for Vertically-fired, dry-bottom furnace with NOX
control (coal-fired)
Unit Name
ARSENIC
BERYLLIUM
CADMIUM
CHROMIUM
COBALT
LEAD
MANGANESE
MERCURY
NICKEL
SELENIUM
EPRI Site 1 1 5
0.61
0.52
0.58
0.57
1.00
0.38
0.58
0.78
0.64
0.34
EMF (Geometric mean)
0.61
0.52
0.58
0.57
1.00
0.38
0.58
0.78
0.64
0.34
Geometric standard
deviation
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
Table C-8. Tested EMFs and Geometric Means, used in the Emission
Factor Program for Tangentially-fired, dry-bottom furnace without
NOX control (used as the factor on all units without NOX control,
which are controlled by hot-side ESPs) (coal-fired)
Unit Name
ANTIMONY
ARSENIC
BERYLLIUM
CADMIUM
CHROMIUM
COBALT
LEAD
MANGANESE
MERCURY
NICKEL
SELENIUM
EPRI Site 110
0.55
0.89
0.93
1.00
1.00
0.93
1.00
0.71
1.00
0.91
0.58
EMF (Geometric mean)
0.55
0.89
0.93
1.00
1.00
0.93
1.00
0.71
1.00
0.91
0.58
Geometric standard
deviation
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
C-4
-------
Table C-9. Tested EMFs and Geometric Means used in the Emission
Factor Program for Tangentially-fired, dry-bottom furnace with
NOX control (used as the factor on all units without NOX control,
which are controlled by hot-side ESPs) (coal-fired)
Unit Name
ANTIMONY
ARSENIC
BERYLLIUM
CADMIUM
CHROMIUM
COBALT
LEAD
MANGANESE
MERCURY
NICKEL
SELENIUM
EPRI Site 110 w/NOx
control
0.66
0.39
0.43
0.70
1.00
1.00
0.36
0.76
0.75
0.97
0.82
EMF (Geometric mean)
0.66
0.39
0.43
0.70
1.00
1.00
0.36
0.76
0.75
0.97
0.82
Geometric standard
deviation
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
Table C-10. Tested EMFs and Geometric Means used in the Emission
Factor Program for Opposed-fired, dry-bottom furnace without NOX
control (coal-fired)
Unit Name
ANTIMONY
ARSENIC
BERYLLIUM
CADMIUM
CHROMIUM
COBALT
LEAD
MANGANESE
MERCURY
NICKEL
SELENIUM
DOE - Baldwin
0.32
1.00
0.44
1.00
0.42
0.34
0.53
0.20
0.92
0.43
0.04
EMF (Geometric mean)
0.32
1.00
0.44
1.00
0.42
0.34
0.53
0.20
0.92
0.43
0.04
Geometric standard
deviation
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
C-5
-------
Table C-ll. Tested EMFs and Geometric Means used in the Emission
Factor Program for Front-fired, dry-bottom furnace without NOX
control (oil-fired)
Unit Name
ARSENIC
BERYLLIUM
CADMIUM
CHROMIUM
COBALT
LEAD
MANGANESE
MERCURY
NICKEL
SELENIUM
EPRI
Site 13,
no NOx
1.00
0.50
0.25
0.82
0.03
1.00
1.00
0.89
EPRI
Site
103
0.06
0.02
0.19
0.29
0.32
0.06
1.00
0.69
0.005
EPRI
Site
104
0.10
0.02
0.06
0.22
0.04
0.57
0.28
EPRI
Site
105
0.07
0.06
0.02
0.14
0.16
0.40
0.04
EPRI
Site
107
0.23
0.14
0.66
0.45
0.74
0.07
EPRI
Site
108
0.12
0.36
0.05
0.16
1.00
0.90
0.23
EPRI
Site
117
0.40
1.00
0.03
0.56
0.42
0.32
0.21
1.00
0.76
0.37
EMF
(Geometric
mean)
0.17
0.07
0.13
0.18
0.48
0.09
0.51
1.00
0.62
0.06
Geometric
standard
deviation
2.77
10.64
2.72
3.47
1.62
2.47
2.14
1.00
1.57
5.59
Table C-12. Tested EMFs and Geometric Means used in the Emission
Factor Program for Opposed-fired, dry-bottom furnace without NOX
control (oil-fired)
Unit Name
ARSENIC
BERYLLIUM
CADMIUM
CHROMIUM
COBALT
LEAD
MANGANESE
MERCURY
NICKEL
SELENIUM
EPRI Site 106
0.45
0.02
0.10
0.32
1.00
0.46
1.00
1.00
0.10
EPRI Site 109
0.01
0.39
1.00
1.00
0.26
0.80
0.04
0.79
0.02
EMF (Geometric
mean)
0.08
0.02
0.20
0.57
1.00
0.35
0.89
0.04
0.89
0.04
Geometric standard
deviation
11.80
N/A
2.61
2.23
1.00
1.50
1.17
N/A
1.18
3.41
C-6
-------
Table C-13. Tested EMFs and Geometric Means used in the Emission
Factor Program for Front-fired, dry-bottom furnace with NOX
control (oil-fired)
Unit Name
ARSENIC
BERYLLIUM
CADMIUM
CHROMIUM
COBALT
LEAD
MANGANESE
MERCURY
NICKEL
SELENIUM
EPRI Site 13
0.64
0.47
0.19
0.62
0.08
1.00
1.00
0.71
0.58
EPRI Site 1 1 8
0.14
1.00
0.78
0.29
0.57
1.00
1.00
0.64
0.46
EPRI Site 117
1.00
1.00
1.00
1.00
0.98
0.97
1.00
1.00
1.00
1.00
EMF (Geometric
mean)
0.44
1.00
0.69
0.53
0.56
0.35
1.00
1.00
0.77
0.64
Geometric
standard
deviation
2.83
1.00
1.71
2.42
1.83
3.73
1.00
1.00
1.26
1.50
Table C-14. Tested EMFs and Geometric Means used in the Emission
Factor Program for Tangentially-fired, dry-bottom furnace without
NOX control (oil-fired)
Unit Name
ARSENIC
BERYLLIUM
CADMIUM
CHROMIUM
COBALT
LEAD
MANGANESE
MERCURY
NICKEL
SELENIUM
EPRI Site 1 1 2
1.00
0.79
0.67
0.66
0.38
0.26
0.80
1.00
0.53
1.00
EMF {Geometric mean)
1.00
0.79
0.67
0.66
0.38
0.26
0.80
1.00
0.53
1.00
Geometric standard
deviation
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
C-7
-------
Table C-15. Tested EMFs and Geometric Means used in the Emission
Factor Program for Tangentially-fired, dry-bottom furnace with
NOX control (oil-fired)
Unit Name
ARSENIC
BERYLLIUM
CADMIUM
CHROMIUM
COBALT
LEAD
MANGANESE
MERCURY
NICKEL
SELENIUM
EPRI Site 1 1 9
1.00
0.57
See Note 1
0.44
1.00
0.69
0.62
0.56
0.35
1.00
1.00
0.72
0.64
EMF (Geometric
mean)
0.44
1.00
0.69
0.79
0.56
0.35
1.00
1.00
0.64
0.64
Geometric standard
deviation
N/A
N/A
N/A
1.40
N/A
N/A
N/A
N/A
1.17
N/A
Note 1. - Since the only source of data for this type of unit was limited to data on only two metals, it was
decided to take the data from another similar unit (a front-fired, dry-bottom furnace with NOx control) along
with the 2-data-point set to develop a set of geometric means. This set of geometric means is the data set in
the "See Note 1" column. The geometric means of the "See Note 1" set and the 2-data-point set were
derived. These means were used to represent a tangential-fired, dry-bottom furnace with NOX control burning
oil.
C-8
-------
Table C-16. Tested EMFs and Geometric Means used in the Emission Factor
Program for Fabric Filters (baghouses)
Unit Name
ANTIMONY
ARSENIC
BERYLLIUM
CADMIUM
CHROMIUM
COBALT
LEAD
MANGANESE
MERCURY
NICKEL
SELENIUM
HYDROGEN CHLORIDE
HYDROGEN FLUORIDE
EPRI Site 10
0.001
0.004
0.004
0.002
0.02
0.002
EPRI Site 1 3
(controlling
an oil-fired
unit) a
0.01
0.25
0.001
0.005
0.15
1.00
0.001
NSP-
Riverside
#6-7
0.03
0.03
0.06
1.00
0.05
0.03
0.05
1.00
0.20
0.06
EPRI Site
115
0.03
0.003
0.05
0.01
0.01
0.01
0.01
0.27
0.05
0.02
DOE - Niles
#2 w/NOx
Control
0.01
0.004
0.001
0.01
0.003
0.001
0.001
0.01
0.92
0.001
0.79
DOE-
Boswell
0.06
0.01
0.01
0.06
0.00
0.01
0.01
0.00
0.39
0.01
0.31
Note 1
Note 1
EMF
(Geometric
mean)
0.02
0.01
0.01
0.08
0.02
0.003
0.01
0.01
0.63
0.01
0.12
0.56
1.00
Geometric
standard
deviation
3.42
3.52
4.55
6.18
6.29
3.10
4.09
4.58
1.86
10.57
5.66
Note 1 - These EMFs were developed from four specific emission tests that examined HCI and HF emissions through an ESP. a baghouse, an SDA/FF, and a FGD.
Table C-17. Tested EMFs and Geometric Means used in the Emission Factor
Program for Electrostatic Precipitators - hot side (located before the air
preheater, controlling an coal-fired unit)
Unit Name
ANTIMONY
ARSENIC
BERYLLIUM
CADMIUM
CHROMIUM
COBALT
LEAD
MANGANESE
MERCURY
NICKEL
SELENIUM
HYDROGEN CHLORIDE
HYDROGEN FLUORIDE
EPRI Site 110
0.11
0.01
0.01
0.004
0.02
0.04
0.02
0.04
1.00
0.002
1.00
EPRI Site 110 w/NO,
Control
0.02
0.15
0.01
0.01
0.04
0.02
0.03
0.02
1.00
0.01
0.87
Note 2
Note 2
EMF (Geometric mean)
0.04
0.04
0.01
0.01
0.03
0.03
0.02
0.02
1.00
0.004
0.93
1.00
1.00
Geometric standard
deviation
3.87
7.13
1.08
2.39
1.84
1.55
1.59
1.86
1.00
3.24
1.10
Note 2 - Because there were no data on HCI and HF emissions through an ESP attached to an oil-fired unit or a hot-side ESP attached to a coal-fired unit, the EMF was
left as "1" so that all HCI and HF emissions passed through the ESP.
Although Site 13 is an oil-fired unit, the data were used here because they are a
measure of baghouse performance at removing HAPs)
-------
Table C-18. Tested EMFs and Geometric Means used in the Emission Factor
Program for Electrostatic Precipitators - cold side (located after the air
preheater, controlling an oil-fired unit)
Unit Name
ARSENIC
BERYLLIUM
CADMIUM
CHROMIUM
COBALT
LEAD
MANGANESE
MERCURY
NICKEL
SELENIUM
HYDROGEN CHLORIDE
HYDROGEN FLUORIDE
EPRI Site 1 1 2
0.49
0.23
0.69
0.44
0.25
0.47
1.00
0.17
0.27
EPRI Site 1 1 8
0.55
0.10
0.44
0.08
0.43
0.83
0.58
0.07
0.65
Note 2
Note 2
EMF (Geometric mean)
0.52
0.16
0.69
0.44
0.14
0.45
0.91
0.31
0.14
0.65
1.00
1.00
Geometric standard
deviation
1.09
1.76
N/A
1.00
2.27
1.07
1.14
2.39
2.50
N/A
Note 2 - Because there were no data on HCI and HF emissions through an ESP attached to an oil-fired unit or a hot-side ESP attached to a coal-fired
unit, the EMF was left as "1" so that all HCI and HF emissions passed through the ESP.
Table C-19. Tested EMFs and Geometric Means used in the Emission Factor
Program for Particulate Matter Scrubber Unit - (controlling a coal-fired
unit)
Unit Name
ANTIMONY
ARSENIC
BERYLLIUM
CADMIUM
CHROMIUM
COBALT
LEAD
MANGANESE
MERCURY
NICKEL
SELENIUM
HYDROGEN CHLORIDE
HYDROGEN FLUORIDE
EPRI Site 125
0.10
0.10
0.02
0.09
0.03
0.02
0.03
0.01
0.96
0.01
1.00
0.06
0.09
EMF (Geometric mean)
0.10
0.10
0.02
0.09
0.03
0.02
0.03
0.01
0.96
0.01
1.00
0.06
0.09
Geometric standard deviation
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
C-10
-------
Table C-20. Tested EMFs and Geometric Means used in the Emission Factor
Program for Fluidized Gas Desulfurization Unit - (controlling a coal-fired
unit)
Unit Name
ANTIMONY
ARSENIC
BERYLLIUM
CADMIUM
CHROMIUM
COBALT
LEAD
MANGANESE
MERCURY
NICKEL
SELENIUM
HYDROGEN CHLORIDE
HYDROGEN FLUORIDE
EPRI Site 1 1
0.49
0.85
0.51
1.00
1.00
1.00
0.89
0.88
0.17
EPRI Site 1 2
0.23
0.62
0.40
0.23
0.59
0.03
1.00
1.00
0.27
NSP-
Sherburne
#1-2
0.04
0.02
0.16
0.03
0.02
0.77
0.05
0.17
NSP-
Sherburne
#1-2
(mercury
2nd test)
0.41
DOE - Yates
0.21
0.08
0.11
0.75
0.18
0.23
0.04
0.17
0.54
1.00
0.26
DOE - Coal
Creek
0.16
0.76
0.97
1.00
0.91
0.93
1.00
0.88
1.00
0.60
Note 1
Note 1
EMF
(Geometric
mean)
0.18
0.20
0.13
0.50
0.26
0.46
0.22
0.26
0.71
0.53
0.26
0.21
0.73
Geometric
standard
deviation
1.23
3.29
6.96
2.16
3.72
2.30
6.36
5.65
1.42
3.79
1.66
Note 1 • These EMFs were developed from four specific emission tests that examined HCI and HF emissions through an ESP, a baghouse, an SDA/FF, and a FGD.
Table C-21. Tested EMFs and Geometric Means used in the Emission Factor
Program for Spray Dryer Adsorber / Fabric Filter Units - (controlling a
coal-fired unit)
Unit Name
ANTIMONY
ARSENIC
BERYLLIUM
CADMIUM
CHROMIUM
COBALT
LEAD
MANGANESE
MERCURY
NICKEL
SELENIUM
HYDROGEN CHLORIDE
HYDROGEN FLUORIDE
NSP-
Sherburne #3
0.01
0.001
0.000
0.08
0.03
0.01
0.003
0.54
0.04
0.05
NSP-
Sherburne #3
(mercury 2nd
test)
0.46
EPRI Site 1 1 1
0.04
0.13
0.03
0.04
EPRI Site 14
0.002
0.01
0.19
0.005
0.01
0.002
0.003
1.00
0.01
0.13
DOE-
Springerville
0.001
0.0003
0.0001
0.001
0.0004
0.001
0.001
0.98
0.0003
0.0004
Note 1
Note 1
EMF
(Geometric
mean)
0.003
0.002
0.001
0.04
0.02
0.001
0.003
0.002
0.70
0.01
0.01
0.18
0.18
Geometric
standard
deviation
2.53
8.17
8.48
11.55
2.91
6.05
5.14
1.90
1.50
9.51
22.46
Note 1 - These EMFs were developed from four specific emission tests that examined HCI and HF emissions through an ESP, a baghouse, an SDA/FF, and a FGD.
C-ll
-------
in co
«pg
iff
g 2 £
0 » -a
o
si!
" S £
s
il
§1
= 22
£ .0
lu t; -5
W S
J^lil
w - 7 5 =
tn > tt
gss
0 " 0
0 1
Q >
.
O
co
CM
0
o
0
o
CO
o
o
6
s
o
(O
o
d
1
6
s
d
d
if)
o
d
in
O
d
0
d
o
d
o
o
d
§
0
o
o
s
o
d
0
o
| ARSENIC
CO
^
0
d
co
0
o
o
q
d
CO
0
d
CM
O
d
g
d
CM
0
o
g
o
CM
O
d
CD
O
O
CM
0
o
CO
O
o
o
CM
O
o
0
d
8
O
O
d
o
o
d
o
d
| BERYLLIUM
co
CD
CM
§
0
in
O
d
CO
o
d
S
o
1^
d
CD
d
^
o
o
d
CM
0
d
0
d
CO
d
a>
0)
d
••s-
o
o
d
d
^-
d
§
d
co
O
d
1 CADMIUM
0
in
CM
CM
o
0
in
O
d
0
d
<*
o
d
CM
O
d
in
8
d
o
o
0
CO
0
O
in
O
d
r*.
O
O
^r
o
d
o
o
o
o
g
o
o
o
0
d
CO
O
d
| CHROMIUM
CM
in
CO
0
d
r«.
0
d
CM
8
d
S
d
CM
0
d
0
d
CM
o
o
s
o
m
O
o
o
o
o
o
o
d
s
u
CO
q
tf
CO
0
d
g
d
o
d
§
d
S
d
1
d
CM
O
d
s
o
(D
q
d
(D
o
d
s
o
to
CD
O
in
O
0
d
O
5)
d
in
*
d
CD
0
If)
*fr
d
o
^~
GO
O
o
f-.
d
8
CD
r*.
d
CM
O5
O
o
8
fe
CO
d
8
H MERCURY
CO
q
^
CM
o
d
CO
q
d
0
o
tfr
o
o
CM
O
d
tn
o
o
d
0
o
o
o
CO
0
O
in
o
o
CO
O
O
in
O
•<*
O
o
0
o
rv
CO
d
o
o
CO
8
d
o
d
H NICKEL
to
CO
-------
Appendix D - Discusion of the Methodology Used to Develop
Nationwide Emission Totals
-------
This page is intentionally blank.
-------
D.I INTRODUCTION
To estimate emissions of hazardous air pollutants (HAPs)
from fossil-fuel-fired electric utility units (2:25 MWe) , the EPA
developed the emission factor program (EFP). This program
incorporates unit configuration data from individual units as
well as emission testing data to compute estimated emissions. An
explanation of the program and several assumptions about the data
and how they were used are described here.
D.2 PROGRAM OPERATION
Emissions of HAPs considered in this study consist of two
types: trace elements and organic compounds. Trace elements exist
in the fuel when fired, while the organic HAPs are formed during
combustion and postcombustion processes. Different programing
methods are required for handling the two types of HAPs. Program
diagrams for modeling trace element emissions are shown in Figure
D-l for coal and Figure D-2 for oil and gas. The two figures
differ only in treatment of the fuel before the trace elements
reach the boiler. Figure D-3 shows the program diagram for
modeling organic HAP emissions.
D.3 DATA SOURCES
The EFP was built to accept data from two sources. The
first is a data input file containing plant configurations, unit
fuel usage, and stack parameters. This input file was based on
the Utility Data Institute/Edison Electric Institute (UDI/EEI)
Power Statistics database (1991 edition). The UDI/EEI database
is composed of responses from electric utilities to the yearly
updated U.S. Department of Energy (DOE) Energy Information
Administration (EIA) Form EIA-767.
The second data file is the emission modification factor
(EMF) database. This database contains information from
emissions tests conducted by the Electric Power Research
Institute (EPRI), DOE, and the electric utility industry. The
program first searches the input file for the type of fuel burned
and the amount of fuel consumed per year in an individual unit.
If the fuel type is coal, the EFP then looks for. the coal's State
of origin. Origin is important because the trace elements in coal
are addressed by coal type (bituminous, subbituminous, and
lignite) and State of origin in the U.S. Geological Survey (USGS)
database, which analyzed core and channel samples (3,331 samples)
of coal from the top 50 (1990 or later) economically feasible
coal seams in the United States.
D-l
-------
USGS coal data
(by State and coal type)
Apply coal cleaning
factor
UDI/EEI plant configuration
information
USGS coal data
(by State and coal type)
No cleaning factor
»• Trace elements (TE) to boiler «•
Apply boiler TE emission
modification factor
What is
particulate
patter (PM) contra^
type?*
Apply the PM control TE
emission modification factor
What is
the SQ2 control
type?*
Apply SO2control TE emission
modification factor
kg/yr of specific trace element
exiting unit stack
Figure D-1. Trace elements in coal.
*Taken from UDI/EEI data.
D-2
-------
UDI/EEI plant configuration
information
Oil
Natural gas
Used fuel oil No. 6 (residual)
for all oil types
Trace elements in oil taken
from plant testing
I
Trace elements in natural gas
taken from plant testing
(only two sets of data)
Used a denisty of 8.2 Ib/gal for
feed rate calculation
Trace elements (TE) to boiler
Apply boiler TE emission
modification factor
What is
paniculate
^matter (PM) control,
type?*
Apply the PM control TE
emission modification factor
What is
the SO2 control
type?*
Apply 80^ control TE emission
modification factor
I
kg/yr of specific trace element
exiting unit stack
Figure D-2. Trace elements in oil and natural gas.
*Taken from UDI/EEI data.
D-3
-------
Type of fuel
burning?
Gas
Obtain unit's fuel
consumption
Obtain unit's fuel
consumption
Obtain unit's fuel
consumption
For individual HAPs,
find the median
Ib/trillion Btu emission
factor for a specific
HAP'
For individual HAPs,
find the median
Ib/trillion Btu emission
factor for a specific
HAP2
For individual HAPs, find
the geometric mean
kg/10' cu ft emission
factor for a specific
HAP3
Individual fuel
consumption x
emission factor x
heat content of
150,000 Btu/gal
Individual fuel
consumption x
emission factor
Individual fuel
consumption x
emission factor x
higher heating value
for bituminous coal
(12,688 Btu/lb coal)
Individual fuel
consumption x
emission factor x
higher heating value
for subbituminous
coal (9,967 Btu/lb
coal)
Individual fuel
consumption x
emission factor x
higher heating value
for lignite coal (6,800
Btu/lb coal)
Convert into kg/yr
stack emission for
HAP
'Only oil-fired units were used to obtain these emission factors.
*Only coal-fired units were used to obtain these emission factors.
K^nly gas-fired units were used to obtain these emission factors.
Figure D-3. Organic emissions.
D-4
-------
D.4 OPERATIONAL STATUS OF BOILERS
The operational status of units was taken from the
UNIT_90.dbf file of the EEI/UDI Power Statistics database (1991
edition addressing 1990 data). Only units that were listed as
either operational or on standby were used in the EFP. One
hundred fifty-one units were listed as being on standby in the
EEI/UDI Power Statistics database but were actually on indefinite
standby and thus did not emit any HAPs. These units were
excluded from the nationwide emissions totals in Appendix A.
Only coal-fired, oil-fired, and natural gas-fired units were
included in the EFP. This decision was made because units using
these fuels make up an overwhelming majority of the
fossil-fuel-fired electric utility units with a capacity ^25 MWe.
Anthracite was disregarded as a fuel because of the limited
number of units burning this type of coal.1 Four units burning
anthracite coal (in 1990) were assigned to burn bituminous coal
for program computations.
Coal-fired boiler concentrations were modified for different
heating values, depending on the type of coal, before being
converted to a rate basis (kg/yr of individual HAP). This
procedure was necessary because different coal ranks have
different heating values. For example, it would require burning
more lignite to achieve the same heat input to the boiler as
burning bituminous coal. These values were determined as
averages for each type of coal (see Table D-l).2
The EEI/UDI database had a number of gaps in the fuel
consumption data. Some of these gaps were filled by data
supplied voluntarily by the industry. To address the remaining
gaps, RTI plotted the available data and fitted point-slope
equations to estimate fuel consumption.3 These equations
involved plotting nameplate megawatts (modified to take into
account the unit's capacity factor) against fuel usage. If the
fuel usage and the unit capacity factor in 1990 were not given,
1989 fuel consumption data were used. If 1989 data were not
available, the geometric mean of the 1980-1988 EEI fuel
consumption data was used. When all other options had been tried
unsuccessfully, an average fuel consumption of units rated within
±5 MW of the unit with unknown fuel usage was used.
Capacity factors were taken from the UDI/EEI database for as
many units as possible. If the above linear equation or (±5 MWe)
estimating procedure were used, then the capacity factor for the
D-5
-------
Table D-l. Average Higher Heating Values of Coal4
Class and group*
1. Bituminous
1 . Low-volatile bituminous
coal
2. Medium-volatile bituminous
coal
3. High-volatile A bituminous
coal
4. High-volatile B bituminous
coal
5. High-volatile C bituminous
coal
High-volatile C bituminous
coal
Agglomerating
character
Fixed carbon
limits, % (dry,
mineral-matter-
free basis)
Equator
greater
than
commonly
agglomerating0
-
•
«
•
agglomerating
78
69
—
-
—
—
Less
than
86
78
69
—
—
—
Volatile matter limits.
% (dry, mineral-
matter-free basis)
Equal or
greater
than
14
22
31
—
_
-
Less
than
'-•
22
31
—
—
—
—
Calorific value limits,
Btu/lb (moist,b
basis)
Equator
greater
than
—
—
14,000"
13,000a
1 1 ,500
10,500
Less
than
—
—
—
14,000
13,000
1 1 ,500
Average of Averages (Value used in EFP for bituminous coal)
II, Subbituminous
1 . Subbituminous A Coal
2. Subbituminous B Coal
3. Subbituminous C Coal
nonagglomerating
-
-
—
—
—
—
—
—
—
—
—
—
—
10,500
9,500
8,300
1 1 ,500
10,500
9,500
Average of Averages (Value used in EFP for Subbituminous coal)
Average
14,000
1 3,500
12,250
11,000
12.688
1 1 ,000
10,000
8,900
9,967
M.UgnHfe
1 . Lignite A
2. Lignite B
nonagglomerating
-
—
_.
—
—
—
—
—
„
6,300
—
8,300
6,300
Average of Averages (Value used in EFP for lignite coal)
7,300
6,300
6,800
s This classification does not include a few coals, principally nonbanded varieties,
which have unusual physical and chemical properties and which come within the limits
of fixed carbon or calorific value for high-volatile and Subbituminous ranks. These
excluded coals either contain less than 48 percent dry, mineral-matter-free fixed
carbon or have more than 15,500 moist, mineral-matter-free Btu per pound.
b Moist refers to coal containing its natural inherent moisture but not including
visible water on the surface of the coal.
c It is recognized that there may be nonagglomerating varieties in these groups of the
bituminous class, and there are notable exceptions in high-volatile C bituminous
group.
d Coals having 69 percent or more fixed carbon on the dry, mineral-matter-free basis
shall be classified by fixed carbon, regardless of calorific value.
D-6
-------
unit (with unknown fuel consumption) would fit an industry norm
for that size unit and fuel type.
Limestone is used in circulating fluidized bed (CFB)
combustors to control sulfur dioxide (S02) . Early in the
program's development, the EPA sought to address limestone's
contribution to trace metal emissions. Based on the fact that
limited trace metal data were available and that there were only
19 listed CFB units in the country in 1990, limestone's effect
was disregarded.5
Utility units may burn coal that originated from several
States; however, in the EFP each coal-fired unit was assigned a
single State of coal origin.6 The State of origin used in the
EFP was the State that contributed the highest percentage of the
unit's coal.
D.5 BOILER CONFIGURATION
EPA received 51 emissions tests conducted by EPRI, DOE, and
industry in time for inclusion in the EFP. Because of this
limited sample, all boiler configurations, particulate, and S02
control types could not be sampled. To estimate the emissions
from all units in the U.S., the substitution of unknown units
into units with known EMFs was necessary. After studying the
tested EMFs, the following patterns were observed. Coal-fired
unit emissions seemed to be affected by whether the unit had a
dry- or wet-bottom furnace. Oil-fired unit emissions seemed to
be affected by whether or not the unit had nitrogen oxides NOX
control. Since only one type of gas-fired boiler was tested, all
gas-fired units obtained their EMFs from this type of unit.7
One of the emission test reports that analyzed an
oil-burning, tangentially fired (with NOX control) unit contained
information on two trace metals. Because this was the only unit
of its kind to be tested, it was necessary to substitute the
trace metal data of another similar unit (one having NOX control)
for which more data were collected. The EMFs of the oil burning,
front-fired unit (with NOX control) were averaged (by geometric
mean) into the unit along with the two trace metal concentrations
found in the tangentially fired boiler. Because there were
organic HAP concentration numbers available for the
tangentially-fired boiler, these numbers were maintained without
modification.
No conventional emission testing (multimetals, volatile
organic sampling train (VOST), semi-VOST) was done on
combined-cycle gas turbines (CCYC). The Fourier Transform
D-7
-------
Infrared (FTIR) system was used to test a CCYC unit for organic
HAPs, but few HAPs were found. Combined-cycle gas turbines were
categorized as conventional gas-fired units to address their
emissions.
Testing by FTIR was also done on one example each of
pulverized coal-, circulating fluidized bed-, oil-, and
conventional gas-fired boilers and a combined-cycle gas turbine.
However, the EPA decided not to use the data in developing
estimated emissions.
Of the test reports received, three contained data that were
not feasible for use in the EFP because the test contractors did
not or could not test between the boiler and the particulate
control device. The result was a test containing only a fuel
analysis and stack emission numbers.
One EPRI emission test report (identified as EPRI Site 10)
contained only one sample run instead of the normal three runs.
Because only two emission test reports on CFBs (including Site
10) were available, the EPA decided to use these data.
Units were deemed dual-fuel-firing units if they fired more
than 10 percent of at least one other fuel. Dual-fuel-firing
emissions were modeled by splitting the dual-firing units (only
oil- and gas-fired units) into two separate units with emissions
exiting from the same stack. If the unit was listed as an
oil-fired unit, its oil consumption rate and configuration were
used to obtain its HAP emission rates for oil. The unit in
question was then split into a gas-fired portion by using its gas
consumption rate and changing its boiler type to the equivalent
gas-fired type. This method was the most equitable way to
represent dual-fuel-fired emissions, for both trace metals and
organic HAPs created by either oil-fired or gas-fired boilers,
respectively.
Substitution was also performed on particulate control and
S02 control devices. Particulate control was addressed in one of
six ways: electrostatic precipitator, cold-side (ESP,CS);
electrostatic precipitator, hot-side (ESP,HS); electrostatic
precipitator, cold-side, controlling an oil-fired unit
(0-ESP,CS); fabric filter (FF); particulate scrubber; or no
control.
For coal-fired units that had hot-side electrostatic
precipitators (ESPs), the boiler was assigned as either HTANGNOX
or HTANGNONOX depending on whether or not the unit had NOX
control. The HTANGNOX and HTANGNONOX boiler EMFs were determined
D-8
-------
from emissions testing data measured at a site that had a
tangential boiler, a hot-side ESP, and a cold-side ESP; testing
was conducted with and without NOX control. Cold-side ESPs are
placed after the air preheaters, while hot-side ESPs are placed
before the air preheaters. The UDI/EEI database reported several
units with combination HS/CS ESPs. These are units with separate
ESPs before and after their air preheaters. Although one such
unit was tested for HAP emissions, during the majority of its
testing the cold-side ESP was turned off. Therefore, the data
for this unit were used to develop hot-side ESP EMFs for the EFP.
Because more data were available on ESP,CS devices, and because
units controlled by HS/CS ESPs had a cold-side ESP as their last
particulate matter (PM) control device, HS/CS ESPs were projected
to behave like cold-side ESPs in terms of trace metal emissions.
In assigning the boiler type for coal-fired units, when there was
no information on whether the unit had NOX control, it was
assumed that the unit had no NOX control and the unit was
assigned HTANGNONOX boiler factors. The boiler and PM control
device data were assigned in this manner for units that had
hot-side ESPs since the temperature at the inlet to the hot-side
ESP was approximately 700° F, whereas the temperature at the
inlet to cold-side ESPs were typically around 300° F. The
assignment was made to account for any effect that the
approximately 700° F temperature might have on air toxic
emissions.
Emission modification factors for particulate control by
scrubbers were derived from data on controling trace elements by
one venturi scrubber used for combined S02 and PM control.
Particulate matter scrubbers use water only, while flue gas
desulfiriztion units (FGDs) use water and a reagent (lime,
limestone, etc). Although the presence of this reagent could
cause the FGD to affect HAPs differently from the PM scrubbers,
the EPA believes that the small number of PM scrubbers (<5)
should not cause U.S. aggregate emissions to be adversely
effected.
Mechanical collectors (multicyclones) are used either as
precollection devices, before FFs or ESPs, or as primary
collection devices for some oil-fired plants. No HAP emissions
testing was done exclusively on mechanical collectors. Since
mechanical collectors were projected to have little or no effect
on reducing HAPs because of their ineffectiveness at removing
small particles, units with only multicyclones were determined to
have no control effect on HAPs in the program.
In the EFP, devices for controlling S02 emissions were
classified as either WETSCRUB (containing all types of wet FGDs)
D-9
-------
or DRYSCRUB {containing all types of spray dryers/dry scrubbers).
This substitution was necessary due to the lack of test data on a
variety of wet FGD and dry scrubber types. Also, the EMFs
include data from units tested with bypasses operating when using
a bypass is normal operation.
D.6 STACK CHARACTERISTICS
Stack data in the UDI/EEI from some electric utility units
were incomplete. Some of these gaps were due to the database
reporting stack parameters from a shared stack on only one of the
plant's units instead of reporting on both. The shared stack
parameters were completed for these sister units. Next, an
industry contractor made contact with a number of utility plants
to retrieve missing stack data. This information was useful but
still incomplete. The remaining gaps in stack parameter data
were filled by either (1) finding a sister unit of the same
configuration (and site, if possible) in order to duplicate its
stack data, or (2) using the original EEI/UDI stack data to
create a set of equations to estimate the relationships between
stack height and gas flow, stack exit temperature, and exit
velocity from stack diameter, respectively. These linear
equations (point-slope) were specific to coal-, oil-, gas-, and
combined cycle gas turbine-fired units. A spreadsheet procedure
was developed to enter a stack height for a unit and use four
separate equations to estimate the other parameters.
A few stack latitudes or longitudes not addressed in either
the original EEI database or the contractor's research were found
by calling the operators of the utility plants in question.
D.7 TRACE ELEMENT CONCENTRATION IN FUEL
The USGS database contains concentrations of trace elements
that were extracted from coal in the ground but does not include
analyses of coal shipments. The concentrations of trace elements
in coal in the ground and in coal shipments to utilities may
differ because, in the process of preparing a coal shipment, some
of the mineral matter in coal may be removed. Since
approximately 77 percent of the Eastern and Midwestern bituminous
coal shipments are cleaned8 to meet customer specifications on
heat, ash, and sulfur content, a coal cleaning factor was applied
to most bituminous coals in the EFP.9
Arithmetic averages of the concentrations of trace elements
were determined from the USGS database by State of coal origin,10
and the average concentrations were then used in the EFP. (Note:
statewide data were not separated by coal region, and statewide
D-10
-------
averages were not weighted by coal production within the State.)
Two sets of concentration data exist for coal that originated
from Arizona and one set for coal that originated from
Washington.11 The two sets of Arizona data were averaged with
data for Colorado, Utah, and New Mexico coal. The trace element
concentrations for coals from Arizona, Louisiana, and Washington
were needed for, respectively, five, one, and two utility units.
Because no data were available for coal from Louisiana, data from
Texas lignite coal were used to represent the concentration of
trace elements in Louisiana coal.12
Additional data on the concentrations of the trace elements
in utility coal shipments were received from ARCO Coal Company on
145 samples of Wyoming Coal and on 30 samples of bituminous
Colorado coal,13 and from the Illinois State Geological Survey
(ISGS) on 34 samples of Illinois cleaned coal.14 Arithmetic
averages of the trace element concentrations provided by ARCO
Coal Company and ISGS were converted to an as-received basis and
used directly, without application of cleaning factors, in the
EFP.15
For a unit that burned bituminous coal, the kilogram/year
feed rate of trace elements to the boiler was determined from the
average trace element concentration in the coal, a coal cleaning
factor, and the annual fuel consumption rate. No coal cleaning
factors were applied to lignite and subbituminous coals (see
Equations 1 and 2 in Table D-2).
If the fuel type was oil, the program accessed a database
containing the arithmetic average of trace element concentrations
in residual oil (see Figure D-2). Each concentration data point
was the arithmetic average of repeated measurements, and at least
one of the repeated measurements had to be a detected
concentration (see discussion of nondetected data in section
D.10). Because trace element data were available only on
residual oil-fired units, and since 95 percent of the oil-fired
units burn residual oil, all units were assumed to burn residual
oil. Although densities of residual oils vary, an average
density of 8.2 Ib/gal was chosen for the feed rate calculation
for oil. The concentration data and density were used, as shown
in Equation 3 in Table D-2, to calculate a kilogram/year rate of
each trace element entering the unit's oil-fired boiler.
Oil-fired organic HAP exit concentration calculations included a
150,000-Btu/gallon heating value for oil.
An emission rate for each organic HAP emitted from gas-fired
units was extracted from the test reports. Only two test reports
analyzed organic HAPs, and a geometric mean emission rate of each
D-ll
-------
CO
cu
>c
u
•H
c
(C
tn
M
O
G
H
CO
C
O
4J
(0
rH
d
u
iH
U
Q)
-j
1
U
(N
I
D
Q)
rH
•9
EH
M
O
-H
y
•^
0
n
4J
i
H
0
O
U
J
ti
«M
0
O
d
•0
o
o
fa
i
H
«J
0
U
tuminous
•H
A
0
fa
•5
&
c
o
|
*
§
^
•^
****/
I
«•*
S
S
jj
^
I
g
^f.
I
"*5
§ ^
E
^>
&
•§
o
>
g
ts*
).
-**
8
1
1— 4
u
s
•C ^J
"" -a,
: I
S ^
O "^
1
•? K
1^
c
i
0 "Q
*» ^
« -ts
H
O
U
n
|
5
•H
1
n
!
o
•H
g
•H
i-H
0
fa
K
i
•^
1
J5
•«
II
^
_60
C
1
^
**^i
^
1
^
s
1
1
1
V
s
2
k
^
•
*-*
2
^
25^
|
<
"~
<
§>
H
||i
K
'S
1
,
k.
. §
:
G o
0 u
•rl ^3
*> 0
a .^
••
CQ
8.
H
«
M
5
S
o
S
1
.a
s
"1
"g!
^j
s
-Si
u
2
X
>.
OS
o
H
"5
^S
J
—
"1
*
""•*
(
as
o
oo
SO
+
S s
^ *>
I*
1 ^
i "**,
C3 *O
i5i "^
cn ^
O
'0
«M
>
t?
1
^
7
0
|
•<
^
1
- t
§ s
•H §
A) oo
1 ^
D-12
-------
observed organic HAP was used. This rate in kilogram HAP/109
cubic feet was then multiplied by the unit's gas consumption to
obtain a kilogram HAP/year stack emission rate of each specific
HAP (see Equation 4 in Table D-2). This result was equivalent to
a stack emission because there were no PM control or SO2 control
devices on gas-fired units. The geometric mean of the
concentrations were averaged and used in the gas-fired boiler
calculations (see Figure D-2). The few trace elements found in
the gas database were estimated by this procedure. Fuel gas
density was assumed to follow the ideal gas law.
D.8 HYDROGEN CHLORIDE (HCL) AND HYDROGEN FLUORIDE (HF)
CONCENTRATION IN FUEL
To obtain HCl or HF emissions from the boiler, emission
factors were derived by performing mass balances for chloride and
fluoride, then converting these balances to the equivalent levels
of HCl or HF throughout, the boiler system.16 For example, for
each part per million of chloride in the feed coal at one of the
test sites, 0.61 Ib/hr of HCl was found in the gas stream leaving
the boiler and 0.00145 Ib/h in the stack gas. Similarly for HF,
the boiler emissions were 0.56 Ib/h for each part per million of
fluoride in the coal and 0.00448 Ib/h in the stack. For ease of
programming, the HCl and HF emissions were addressed starting
with the fuel. This programming was done by multiplying the
chloride and fluoride concentrations in the fuel constituents by
0.61 or 0.56, respectively. The resulting numbers allowed direct
conversion into boiler emissions that could be further modified
for systems with PM control or S02 control.
The chloride concentrations were not available for coals
from the following States: Alaska, Illinois, Indiana, Iowa,
Missouri, Utah, and Washington. Chloride concentrations were
assigned, as shown in Table D-3, for coals originating from these
States.17
D.9 EMISSION MODIFICATION FACTORS FOR INORGANIC HAPS
To address the partitioning of the HAP stream through the
combustion and pollution control process, partitioning factors
known as emission modification factors (EMFs) were developed from
inorganic HAP testing data. The EMFs are fractions of the amount
of a HAP compound exiting a device (boiler or air pollution
control device [APCD]) divided by the amount of the same HAP
compound entering that device.18 These EMFs were averaged by
taking the geometric mean of similar devices (e.g., all oil-fired
tangential boilers, all cold-side ESPs). Geometric means were
D-13
-------
Table D-3. Assigned Chloride ppmw and HC1 ppmw Concentrations in
Coal, by State of Coal Origin 17
State
Alaska
Illinois
Indiana
Iowa
Missouri
Utah
Washington
Conversion of assigned ppmw
chloride to assigned HC1 ppmw
54 x 0.61 =
1,136 x 0.61 =
1,033 x 0.61 =
1,498 x 0.61 =
1,701 x 0.61 =
220 x 0.61 =
104 x 0.61 =
Assigned ppmw HC1
in coal
32.9
693.0
630.0
914.0
1,038.0
134.0
63.0
used because of the presence of outlying data points, the small
amount of data, and the general fit of the data to a log-normal
curve. These geometric means were then applied to the kg/yr feed
rates entering the boiler, the effect of which either reduced or
left unchanged the emissions that passed through them. Those
EMFs calculated as being greater than 1.0 (i.e., more material
exiting a device than entering it) are set to equal 1.0. The
EMFs are based on emission test report data collected and
analyzed after 1990.
Nearly all EMFs were computed from three data samples before
and three data samples after the particular device. When all six
data samples for a particular EMF computation were nondetects,
the EPA decided to disregard the EMF. As such, EMFs were
computed when there was at least one detected sample among the
six measured samples.
The EMFs were computed with data from different test reports
but for similar devices (i.e., cold-side ESPs, front-fired
boilers in oil-fired units). The data from coal-fired units were
not segregated by State of coal origin. The EMFs from devices
are segregated into only coal-, oil-, or gas-fired bins.
The EFP itself uses EMFs to partition the emissions as they
proceed from the fuel through the unit to the stack exit as
follows. The average concentrations of metallic HAPs in an
individual fuel by State (based on USGS data) were multiplied by
the amount of fuel that the unit burned in 1990. After
accounting for variables such as coal cleaning (bituminous coal
D-14
-------
only) and coal type (higher heating value), the emission
concentration of an inorganic HAP was converted into an emission
rate in kg/yr entering the boiler. The emission rate entering
the boiler was then modified by EMFs for the boiler, the
particulate control device (when applicable), and the S02 control
device (when applicable).
As stated above, these geometric mean EMFs were then applied
to the fuel HAP concentration estimates and the kg/yr fuel feed
rates entering the boiler, which either reduced or left unchanged
the emissions that passed through it, depending on the value of
the EMF.
Table C-l (Appendix C) shows two sets of EMF data for the
DOE Niles test site. One unit with NOX control is in a section
designated without NOX control. This apparent anomaly occurred
because the NOX control method used, SCR, is a postcombustion NOX
control and does not effect the boiler EMFs. The data are
labeled this way to identify the data obtained from a separate
test report.
The HCl and HF emission factors were addressed in the fuel,
therefore, all HCl and HF boiler EMFs for all fuel types, were
made equal to 1 in the EFP.
Appendix C contains all of the EMFs used to develop the unit
emission estimates for inorganic HAPs.
D.10 ORGANIC AND MINERAL ACID HAPS
Organic and mineral acid HAP emissions were handled in one
of two ways. The first method was used only with HCl or HF
emissions and was described in section D.8. The numbers
resulting from this method allowed direct conversion into boiler
emissions that could be further modified for systems with PM
control or S02 control.
Hydrochloric acid and HF EMFs for PM and SO2 control devices
were developed with data from four test reports where contractors
conducted tests individually for HCl and chlorine as well as HF
and fluorine, before and after each control device. The rest of
the available reports showed tests only for chlorine and fluorine
and estimated the fractions that were HCl or HF. In developing
the HCl and HF EMFs for wet FGDs and dry scrubbers, the EPA
decided to address the effect of flue gas bypass. After
analyzing test data and having discussions with industry
representatives, EPA decided to assume an industry average flue
gas bypass of 17 percent for wet FGDs and 14 percent for dry
D-15
-------
scrubber systems. This assumption was used only in developing
HC1 and HF EMFs.19 Because each of the four test sites was
different than the others regarding S02 and PM control, the
emission factors for chlorine and fluorine were maintained
separately for the four system types rather than averaging them.
The second method of handling organic and mineral acid HAPs
was for organics. Because they were not always tested at the
entrance and exit of each control device in the emissions
testing, all organic HAP emissions were addressed by examining
the test data and determining the concentration of a particular
HAP exiting the stack. Organic HAP concentrations were obtained
from emission test reports. Table D-4 gives the equations used
to estimate organic HAP emissions from coal-, oil-, and gas-fired
boilers.
If stack emission or APCD exit emission data were
unavailable or reported as nondetected, and if at least one-third
of the data samples at the inlet of the APCD were detected
concentrations, EPA used organic emissions at the inlet of the
APCD and accounted for the effect of the APCD with EMFs. For
each individual organic HAP observed in testing, a median
concentration was obtained. This fuel-specific median
concentration was then individually multiplied by each utility
unit's fuel consumption. The result was a fuel-specific emission
rate for all organic HAPs that were observed at least once during
testing.
D.ll TREATMENT OF NONDETECTED DATA IN THE DEVELOPMENT OF EMFS
In the raw data taken from the test reports, the EPA used a
protocol to analyze detected and nondetected compounds in the
test samples. The protocol is as follows:
• When all values for a specific compound are above the
detection limit, the mean arithmetic concentration is
calculated using the reported quantities.
• For results that include values both above and below the
detection limit (with the detection limit shown in
parentheses), one half of the detection limit is used for
values below the detection limit to calculate the mean. For
example:
Analytical values Calculation Mean value
10,12,ND(8) (10+12+[8/2])/3 8.7 ND
D-16
-------
The calculated mean cannot be smaller than the largest
detection limit value. In the following example, the
calculated mean is 2.8. This quantity is less than the
largest detection limit, so the reported mean becomes ND(4).
Analytical values Calculation Mean value
5,ND(4),ND(3) (5+[4/2]+[3/2])/3 ND(4)
• When all sample results are less than the detection limit,
the data are not used.
D.12 MODEL CHANGES FOR ESTIMATES IN THE YEAR 2010
Emission estimates for 2010 were derived from the same basic
model described above. However, changes to input files were made
to accommodate expected changes in fuel usage, generating
capacity, and responses to Phases I and II of the 1990 amendments
under Title IV. The details of these expected changes, except
for coal usage, are described in section 2.7 of this report.
Details of coal usage are described below.
To approximate the projected increase in the use of coal,
and particularly lower sulfur coals, the 2010 coal consumption
was determined as follows. First the estimated overall increase
in electric utility coal consumption was determined
(37 percent).20 Then, instead of using an overall percentage
increase for each coal-fired unit, a factor was derived for each
coal State of origin to represent the expected increase or
decrease in consumption for that State's coal in 2010. The 1990
coal consumption was then multiplied by the 2010 factor, listed
in Table D-5, that corresponded to the State of coal origin
assigned to each unit.21
D-17
-------
f „
w
04
s
u
•H
c
(0
O)
M
O
to
o
-H
4J
03
i — 1
U
fd
u
Q)
rw
Qi
a
i
Q
a)
1-1
(C
u
u
at
a
CD
JK
|fj
i
h
•0
0)
4J
"1
w
n
cu
jjj
*>
f
H
0
1
0
U
<0
n
^J
d
§
1-1
^
••
M
O
H
•rf
2
•d
o
M
•H
«H
H
aJ
o
U
2
§
c
:£»
'S
tt
5
as
":P a.
o s w
•» •*' ^
H "^
^ »
i 6,
^j a: a.
2 ^S
BJ ^ -ft
^ Jo
^^ O
S
oa
1 K
u S
-a Q-
|
^
S
a
^
*-7
S
,
^
1
1
H I
c 8
o -a
•H §
*j 8
H
••
\^
9
H
•H
&
•d
S
•H
«W
1
H
•H
O
»
S
c
•2
i:
s
OJ
^~vMMn>*-
—
^•^^
n
2 a,
— ' S *•
H " ""
v -i-*
2
5
o
—
T2 H
C Q^
"w "^
*o *5l ft.
#5
•^^ ^~ *c
"5 £ -
"Q O
*^s ^
'S 5
S Q, 55
4 ^ I
| * 1
* ^
K 2
"S
—•*
Sj
cs
1
'5
^
1
^
-M*
^
r~ *2»
o E:
— -2
"S Q
1
•:
c
_e
^
^ ^
K
1
«s 1
6 i
0 <•>
i 1
H
••
^4
0
rt
•H
&
•«
I
•H
«H
•
0
^
H; ^
••«
u
a. t?
S s
•*; c
ao -3
•*» .a
XJV
s
•a,
•2
•S
S
u
1
1
%)
s
1
*
-2
§
X""~"o^
— O
^ ""'
^
s=
^
^ .
8 ^
-
i
I
0 ^
•H §
4) BO
I i
H
D-18
-------
Table D-5. Coal consumption scaling factors for 2010
State of coal origin
Kentucky
Pennsylvania
West Virginia
Maryland
Ohio
Alabama
Louisiana
Texas
Virginia
Illinois
Indiana
Iowa
Kansas
Missouri
Oklahoma
Alaska
Arizona
Colorado
Montana
New Mexico
North Dakota
Utah
Washington
Wyoming
2010 factor*
1.27
1.23
1.24
0.872
0.872
1.41
1.41
1.41
1.41
1
1
1
1
1
1
1.599
1.599
1.599
1.599
1.599
1.599
1.599
1.599
1.599
For each coal-fired unit, the 2010 coal consumption was determined as
follows: The 1990 coal consumption was multiplied by the 2010 factor
that corresponded to the State of coal origin assigned to the unit.
D-19
-------
D.I5 REFERENCES
1. Utility Data Institute. EEI Power Statistics Data Base.
Washington, DC. 1992. (excluding nonutility units)
2. Singer, J.G., ed. Combustion Fossil Power. 4th ed.
Combustion Engineering, Incorporated, Windsor, CT. 1991.
p. 2-3, modified table.
3. Memorandum from Cole, J.D., Research Triangle Institute to
Maxwell, W.H., Environmental Protection Agency. February 3,
1993 . Addressing fuel consumption gaps in the EEI power
statistics data base data.
4. Ref. 2.
5. Ref. 1.
6. Memorandum from Heath, E., RTI to Maxwell, W.H., EPA.
July 14, 1993. State of coal origin used in the computer
emission program.
7. Memorandum from Turner, J.H., RTI to Cole, J.D., RTI.
April 18, 1994. Collapsing utility boiler types for
emissions modeling.
8. Akers, D., C. Raleigh, G. Shirley, and R. Dospoy, The Effect
of Coal Cleaning on Trace Elements, Draft Report,
Application of Algorithms. Prepared for EPRI by CQ Inc.,
February 11, 1994.
9. Memorandum from Heath, E., RTI to Maxwell, W.H., EPA.
April 5, 1994. Proposed coal cleaning factors.
10. Memorandum from Heath, E., RTI to Maxwell, W.H., EPA. March
19, 1993. Review of the U.S. Geological Survey data.
11. Memorandum from Heath, E., RTI to Maxwell, W.H., EPA. July
8, 1993. Assignment of concentration of trace elements in
coals from Arizona, Louisiana, and Washington.
12. Ref. 11.
13. Letter from Bowling, C.M., ARCO Coal Company to Maxwell,
W.H., EPA. June 9, 1993. Concerning the use of USGS data
to represent the concentration of trace elements in coal
shipments.
D-20
-------
14. Demir, I., R.D. Harvey, R.R. Rush, H.H. Damberger,
C. Chaven, J.D. Steele, W.T. Frankie, and K.K. Ho,
Characterization of Available Coal from Illinois Mines,
draft report. Illinois State Geological Survey file number
to be assigned. December 28, 1993.
15. Memorandum from Heath, E., RTI to Maxwell, W.H., EPA. April
29, 1994. As-shipped coal data and how gaps in the USGS and
UDI database were filled for the computer emission program.
16. Memorandum from Turner, J.H., RTI, to Cole, J.D., RTI.
April 19, 1994. Methodology for determining HC1 and HF
concentrations fom utility boilers.
17. Memorandum from Heath, E., RTI to Maxwell, W.H., EPA. May
27, 1994. USGS data gaps in chloride concentrations for
seven states.
18. Memorandum from Cole, J.D., RTI, to Maxwell, W.H., EPA.
March 31, 1994. Emission factor memorandum.
19. Memorandum from Cole, J.D., RTI to Maxwell, W.H., EPA.
May 9, 1994. Emission modification factors for HCl and HF
including FGD system bypass.
20. Economic Analysis of the Title IV Requirements of the 1990
Clean Air Act Amendments. Prepared for the U.S. EPA, Office
of Air and Radiation, Acid Rain Division. Prepared by ICF
Resources Incorporated. February 1994.
21. Memorandum from Heath, E., RTI to Maxwell, W.H., EPA. May
9, 1994. Proposed method to account for utilities switching
to lower sulfur coals in 2010.
D-21
-------
Appendix E - Health Effects Summaries: Overview
-------
This page is intentionally left blank.
-------
This section contains summaries of health effects data for
seven hazardous air pollutant (HAPs) emitted from utilities
(i.e., arsenic, chromium, nickel, mercury, hydrogen chloride,
hydrogen fluoride, and dioxins). Radionuclides are discussed in
Chapter 9 of the interim report. All of the numbers presented in
these summaries are subject to change, if EPA obtains new data in
the future indicating that the risk is higher or lower than that
currently being considered. For more information on health
effects, readers can refer to the referenced sources at the end
of Appendix E. Also, health effects information for these HAPs
and other HAPs can be obtained from the EPA's Integrated Risk
Information System1 or from an EPA document titled Health Effects
Notebook for Hazardous Air Pollutants.2 Each summary, except the
one for mercury, contains the following sections:
E.1 INTRODUCTION
E.2 CANCER EFFECTS
E.3 NONCANCER EFFECTS
E.3.1 Acute (Short-Term)
E.3.2 Chronic (Lona-Term)
E.3.3 Reproductive and Developmental
The following is a discussion of the information contained in
each of these sections:
E.1 INTRODUCTION
This section presents a brief overview of the chemical, with
information on its chemistry, physical properties, and major
uses. If available, EPA's National Ambient Air Quality Standard
(NAAQS) and/or Maximum Contaminant Level Goal (MCLG) or Maximum
Contaminant Level (MCL) are also presented in this section.
EPA's NAAQS are legally enforceable air standards set under the
Clean Air Act Amendments of 1990; these are health-based
standards with considerations such as economics and technical
feasibility factored in. EPA's MCLGs are nonenforceable health
goals that are set at levels at which no known or anticipated
adverse health effects occur and that allow an adequate margin of
safety. Maximum contaminant levels are legally enforceable
drinking water standards which are set as close to the MCLGs as
feasible.
E.2 CANCER EFFECTS
The results of available cancer studies in animals and/or
humans are presented in this section. In addition, the EPA's
cancer weight-of-evidence classification system is included. EPA
uses a weight-of-evidence, three-step procedure to classify the
E-l
-------
likelihood that the chemical causes cancer in humans. In the
first step, the evidence is characterized separately for human
studies and for animal studies. The human studies are examined
considering the validity and representativeness of the
populations studied, any possible confounding factors, and the
statistical significance of the results of the studies. The
animal studies are evaluated to decide whether biologically
significant responses have occurred and whether the responses are
statistically significant increases in treated versus control
animals. Secondly, the human and animal evidence is combined
into an overall classification. This classification is based on
an analysis of both the human and animal evidence, considering
the number and quality of both types of studies. In the third
step, the classification is adjusted upward or downward, based on
an analysis of other supporting evidence. Supporting evidence
includes structure-activity relationships (i.e., the structural
similarity of a chemical to another chemical with known
carcinogenic potential), studies on the metabolism and
pharmacokinetics of a chemical, and short-term genetic toxicity
tests. The result is that each chemical is placed into one of
the following six categories:
Group
A
B1
B2
C
D
E
Description
Known human carcinogen
Probable human carcinogen, limited human data are available
Probable Human carcinogen, sufficient evidence in animals and
or no evidence in humans
inadequate
Possible human carcinogen
Not classifiable as to human carcinogenicity
Evidence of noncarcinogenicity for humans
This section also includes information on the inhalation
cancer risk and oral unit cancer risk. If EPA has calculated
both inhalation and oral unit cancer risk values, then this
section is divided into two subsections.
The inhalation unit risk estimate (IURE) for the chemical
is the estimated increased probability of a person's developing
cancer from breathing air containing a concentration of
1 microgram pollutant per cubic meter (//g/m3) of air for
70 years. The IURE is derived using mathematical models that
assume a nonthreshold approach: i.e., there is some risk of
cancer occurring at any level of exposure. The methods used to
derive these values typically result in an "upper bound"
E-2
-------
estimate; i.e., the true risk is unlikely to exceed this value
and may be lower. However, some unit risk estimates are not
"upper bound" estimates but rather are based on a "maximum
likelihood" estimate (e.g., arsenic).
The risk- specif ic dose, which is an estimate of the dose
corresponding to a specified level of cancer risk, is also
included. This section presents risk-specific doses
corresponding to a one-in-a-million and one- in-a-hundred- thousand
excess risk attributed to exposure to the chemical. This means
that EPA has estimated that if an individual were to breathe air
containing these concentrations of the chemical, over his or her
lifetime, that person would theoretically have no more than a
one-in-a-million or one- in-a-hundred- thousand increased chance of
developing cancer as a direct result of breathing air containing
the chemical .
If available, the oral unit cancer risk is also presented.
Both the oral cancer risk and the corresponding risk-specific
dose are developed for an exposure of 70 years to the chemical
through the drinking water. The oral unit risk estimate (OURE)
is the estimated increased risk of cancer for drinking for
70 years 2 liters/day of water that contains a concentration of
of pollutant per liter. It is expressed in units of //g/L.
E.3 NONCANCER EFFECTS
E.3.1 Acute (Short-Term)
Results from acute animal tests or acute human studies are
presented in this section. Acute animal studies usually report
an estimated median lethal dose (LD50) or median lethal
concentration (LC50) . This is the dose (or concentration)
estimated to kill 50 percent of the experimental animals.
Results from these tests are divided into the following toxicity
categories :
Lethality
Oral LD50
Dermal LDen
Inhalation LC50
Extreme
< 50 mg/kg
<200 mq/kg
<200 mg/m3
High
50 to 500 mg/kg
200 to 2,000 mg/kg
200 to 2,000 mg/m3
Moderate
500 to 5,000 mg/kg
2,000 to 20,000 mg/kg
2,000 to 20,000 mg/m3
Low
>5,000 mg/kg
>20,000 mg/kg
>20,000 mg/m3
Source: U.S. EPA. Office of Pesticide Programs, Registration and Classification Procedures, Part II. Federal Register. 40:28279.
E-3
-------
Acute human studies usually consist of case reports from
accidental poisonings. These case reports often help to define
the levels at which acute toxic effects are seen in humans.
E.3.2 Chronic f Loner-Term}
This section summarizes the major chronic noncarcinogenic
effects seen from exposure to the chemical. Chronic animal
studies usually range from 90 days to 2 years. Human studies
investigating effects ranging from exposure of a few years to a
lifetime are also included. In addition, subchronic studies may
be included in this section. Subchronic studies are usually
animal studies of several weeks to 90 days.
The Inhalation Reference Concentration (RfC) is presented in
this section. The RfC is an estimate (with uncertainty spanning
perhaps an order of magnitude) of the daily exposure of a
chemical to the human population by inhalation (including
sensitive subpopulations) that is likely to be without
deleterious effects during a lifetime of exposure. The RfC is
derived based on the assumption that thresholds exist for
noncancer effects; i.e., there is a level below which no toxic
effects would occur. The RfC is calculated as follows: EPA
reviews many human and/or animal studies to determine the highest
dose level tested at which the critical adverse effect does not
occur—i.e., the no observed adverse effect level (NOAEL)—or the
lowest dose level at which the critical adverse effect is
observed, the lowest observed adverse effect level (LOAEL). The
NOAEL from an animal study is adjusted for exposure duration and
respiratory tract differences between animals and humans. EPA
then applies uncertainty factors to adjust for the uncertainties
in extrapolating from animal data to humans (10), and for
protecting sensitive subpopulations (10). Also, a modifying
factor is applied to reflect professional judgment of the entire
data base.
The RfC is not a direct or absolute estimator of risk, but
rather a reference point to gauge the potential effects. Doses
at or below the RfC are not likely to be associated with any
adverse health effects. However, exceedance of the RfC does not
imply that an adverse health effect would necessarily occur. As
the amount and frequency of exposures exceeding the RfC
increases, the probability that adverse effects may be observed
in the human population also increases. The RfC is expressed in
milligrams of pollutant per cubic meter of air (mg/m3) . If
available, the Oral Reference Dose (RfD) is also presented in
this section. The RfD is the oral equivalent of the RfC.
E-4
-------
EPA's confidence in the RfC and/or RfD is also presented in
this section. EPA ranks each RfC and RfD as low, medium, or high
in three areas: (1) confidence in the study on which the RfC or
RfD was based; (2) confidence in the data base; (3) overall
confidence in the RfC or RfD. All three rankings are presented
in this section.
E.3.3 Reproductive and Developmental
This section presents the results of reproductive and
developmental studies on the effects of the chemical in animals
and humans. Reproductive effects are those effects that
adversely affect the female or the male reproductive system.
Examples in the female include reduced fertility, a decrease in
the survival of offspring, and alterations in the reproductive
cycle. Male reproductive effects include a decrease in sperm
count or an increase in abnormal sperm morphology. Developmental
effects are adverse effects on the developing organism that
result from exposure prior to conception (either parent), during
prenatal development, or postnatally to the time of sexual
maturation. Examples include altered growth, death of the
developing organism, and malformations or birth defects.
Reproductive and developmental effects may be observed after
short-term or long-term exposure to the chemical, as some effects
can be attributed to one time or short-term exposures during a
critical biological cycle.
E.4 ARSENIC HEALTH EFFECTS SUMMARY
Arsenic is a naturally occurring element in the earth's
crust that is usually found combined with other elements.
Arsenic combined with elements such as oxygen, chlorine, and
sulfur is referred to as inorganic arsenic; arsenic combined with
carbon and hydrogen is referred to as organic arsenic. In this
health effects summary, arsenic refers to inorganic arsenic and
its associated compounds. Organic arsenic compounds, such as
arsine gas, are not discussed. EPA has set a Maximum Contaminant
Level (MCL) of 0.05 mg/L for inorganic arsenic.3
E.4.1 Cancer Effects — Arsenic
There is clear evidence that chronic exposure to inorganic
arsenic in humans increases the risk of cancer. Studies have
reported that inhalation of arsenic results in an increased risk
of lung cancer. In addition, ingestion of arsenic has been
associated with an increased risk of nonmelanoma skin cancer and
bladder, liver, and lung cancer. No information is available on
the risk of cancer in humans from dermal exposure to arsenic.
Animal studies have not clearly associated arsenic exposure, via
ingestion exposure, with cancer. No studies have investigated
E-5
-------
the risk of cancer in animals as a result of inhalation or dermal
exposure.4
EPA has classified inorganic arsenic in Group A - Known
Human Carcinogen. For arsenic, the Group A classification was
based on the increased incidence in humans of lung cancer through
inhalation exposure and the increased risk of skin, bladder,
liver, and lung cancer through drinking water exposure.5
E.4.1.1 Inhalation Cancer Risk for Arsenic. EPA used the
absolute-risk linear extrapolation model to estimate the
inhalation unit risk for inorganic arsenic. Five studies on
arsenic-exposed copper smelter workers were modeled for excess
cancer risk. All five studies showed excess risks of lung cancer
that were related to the intensity and duration of exposure and
the duration of the latency period. The estimates of unit risk
obtained from the five studies were in reasonably good agreement,
ranging from 1.25 x 10~3 to 7.6 x 10~3 (yug/m3)'1. Using the
geometric mean of these data, EPA calculated an inhalation unit
risk estimate of 4.29 x 10"3 (^g/m3)"1 (EPA).6 Based on this unit
risk estimate, EPA estimates that if an individual were to
breathe air containing arsenic at 0.0002 /ug/m33 over his or her
entire lifetime (70 years), that person would theoretically have
an increased chance of one in a million of developing cancer as a
direct result of breathing air containing this chemical.
Similarly, EPA estimates that breathing air containing
0.002 //g/m3 would result in an increased chance of up to one in a
hundred thousand of developing cancer. EPA has high confidence
in the arsenic cancer unit risk estimate for inhalation exposure
because the studies examined a large number of people, the
exposure assessments included air measurements and urinary
arsenic measurements, and lung cancer incidence was significantly
increased over expected values.7
The Electric Power Research Institute (EPRI) has proposed a
revision to EPA's IURE for inorganic arsenic. EPRI used standard
EPA risk assessment methodology to recalculate the estimated
risk. They calculated a new unit risk of 1.43 x 10~3 (//g/m3)"1,
which is one-third the value on IRIS presented above. EPRI's
risk estimate is based on updated exposure data from an
epidemiology study of workers at a smelter in Tacoma, Washington,
which indicated that the workers were much more highly exposed
than previously thought. EPRI also used results from a recent
Swedish smelter study.8
a 0.0002 Mg/m3 (concentration corresponding to a 10~6 risk level) =10"6 (risk
level)/4.29 x 10~3 (/ug/m3)"1 (unit risk estimate).
E-6
-------
E.4.1.2 Oral Cancer Risk for Arsenic. To estimate the
risks posed by ingesting arsenic, EPA obtained in Taiwan
concerning skin cancer incidence, age, and level of exposure via
drinking water. In 37 villages that had obtained drinking water
for 45 years from artisan wells with various elevated levels of
arsenic, 40,421 individuals were examined for hyperpigmentation,
keratosis, skin cancer, and blackfoot disease (gangrene of the
extremities caused by injury to the peripheral vasculature). The
local well waters were analyzed for arsenic, and the age-specific
cancer prevalence rates were found to correlate with both local
arsenic concentrations and age (duration of exposure). Based on
these data, although EPA has not presented the calculations for
the oral unit risk estimate for arsenic9, they did propose that a
unit risk estimate of 5 x 10~5 (//g/L) ~l from oral exposure to
arsenic in drinking water be used.10
The Taiwan cancer data have the following limitations:
(1) the water was contaminated with substances such as bacteria
and ergot alkaloids in addition to arsenic; (2) total arsenic
exposure was uncertain because of intake from the diet and other
sources; (3) early deaths from blackfoot disease may have led to
an underestimate of prevalence; and (4) there was uncertainty
concerning exposure durations. Due to these limitations, and
also because the diet, economic status, and mobility of
individuals in Taiwan are different from those of most
U.S. citizens, EPA has stated "the uncertainties associated with
ingested inorganic arsenic are such that estimates could be
modified downwards as much as an order of magnitude, relative to
risk estimates associated with most other carcinogens."ll
E.4.2 Noncancer Effects — Arsenic
E.4.2.1 Acute (Short-Term) Effects for Arsenic. Arsenic
has been recognized as a human poison since ancient times, and
large doses, approximately 600 //g/kg/day or higher, taken orally
have resulted in death. Oral exposure to lower levels of arsenic
has resulted in effects on the gastrointestinal system (nausea,
vomiting); central nervous system (headaches, weakness,
delirium); cardiovascular system (hypotension, shock); and the
liver, kidney, and blood (anemia, leukopenia). Acute arsenic
poisoning of humans, through inhalation exposure, has resulted in
similar effects, including effects on the gastrointestinal system
(nausea, diarrhea, abdominal pain), blood, and central and
peripheral nervous system. The only effect noted from dermal
(skin) exposure to arsenic in humans is contact dermatitis, with
symptoms such as erythma and swelling. This effect has been
noted only at high arsenic levels.12
E-7
-------
Because significant information is available on the acute
effects of arsenic poisoning in humans, few animal studies have
been carried out. The limited available data have shown arsenic
to have moderate to high acute toxicity to animals by the oral
route. This is based on data showing the LD50 for arsenic to
range between 50 and 5,000 mg/kg.13
E. 4 . 2 . 2 Chronic (Loner-Term) Effects for Arsenic . The
primary effect noted in humans from chronic exposure to arsenic,
through both inhalation and oral exposure, is effects on the
skin. The inhalation route has resulted primarily in irritation
of the skin and mucous membranes (dermatitis, conjunctivitis,
pharyngitis, and rhinitis), while chronic oral exposure has
resulted in a pattern of skin changes that include the formation
of warts or corns on the palms and soles along with areas of
darkened skin on the face, neck, and back. Other effects noted
from chronic oral exposure include peripheral neuropathy,
cardiovascular disorders, liver and kidney disorders, and
blackfoot disease. No information is available on effects in
humans from chronic low-level dermal exposure to arsenic.14
No studies are available on the chronic noncancer effects of
arsenic in animals, from inhalation or dermal exposure. Oral
animal studies have noted effects on the kidney and liver.15
EPA has established an RfD (Reference Dose) for inorganic
arsenic of 0.0003 mg/kg/day, based on a NOAEL (adjusted to
include arsenic exposure from food) of 0.0008 mg/kg/day, an
uncertainty factor of 3, and a modifying factor of I.16 This RfD
was based on two studies17 that showed that the prevalence of
blackfoot disease increased with both age and dose for
individuals exposed to high levels of arsenic in drinking water.
This same population also displayed a greater incidence of
hyperpigmentation and skin lesions. Other human studies support
these findings, with several studies noting an increase in skin
lesions from chronic exposure to arsenic through the drinking
water. The EPA has not established a RfC for inorganic
• 1 R
arsenic. °
EPA has medium confidence in the studies on which the RfD
was based and in the RfD. The key studies were extensive
epidemiologic reports that examined effects in a large number of
people.- However, doses were not well characterized, other
contaminants were present, and potential exposure from food or
other sources was not examined. The supporting studies suffer
from other limitations, primarily the small populations studied.
However, the general database on arsenic does support the
E-8
-------
findings in the key studies; this was the basis for EPA's "medium
confidence" ranking of the RfD.19
E.4.2.3 Reproductive and Developmental. Limited
information is available on the reproductive or developmental
effects of arsenic in humans. The only available information
consists of several studies that suggest that women who work in,
or live near, metal smelters may have higher than normal
spontaneous abortion rates, and their children may exhibit lower
than normal birth weights. However, these studies are limited
and contain significant uncertainties because they were designed
to evaluate the effects of smelter pollutants in general and are
not specific for arsenic.20
Animal studies on arsenic exposure via oral and inhalation
routes have reported that arsenic at very high doses may cause
death to the fetus or birth defects. No information is available
on reproductive or developmental effects of arsenic in animals
from dermal exposure.21
E.5 CHROMIUM HEALTH EFFECTS SUMMARY
Chromium is a metallic element that occurs in the
environment in two major valence states: trivalent chromium
(chromium III) and hexavalent chromium (chromium VI).
Chromium VI compounds are much more toxic than chromium III
compounds; chromium III is an essential element in humans, with a
daily intake of 50 to 200 micrograms per day recommended for an
adult, while chromium VI is quite toxic. However, the human body
can detoxify some amount of chromium VI to chromium III. EPA has
set a Maximum Contaminant Level (MCL) of 0.1 mg/L for total
chromium.22
E.5.1 Cancer Effects for Chromium
Epidemiological studies of workers have clearly established
that inhaled chromium is a human carcinogen, resulting in an
increased risk of lung cancer. These studies were not able to
differentiate between exposure to chromium III and chromium VI
compounds. No information is available on cancer in humans from
oral or dermal exposure to chromium.23'24
Animal studies have shown chromium VI to cause lung tumors
via inhalation exposure. No studies are available that
investigated cancer in animals from oral or dermal exposure to
chromium VI. Chromium III has been tested in mice and rats by
the oral route, with several studies reporting no increase in
tumor incidence. No studies are available on cancer in animals
from inhalation or dermal exposure to chromium III.25'26
E-9
-------
EPA has classified chromium VI in Group A - Known Human
Carcinogen.27 Since the human studies could not differentiate
between chromium III and chromium VI exposure, and only
chromium VI was found to be carcinogenic in animal studies, EPA
concluded that only chromium VI should be classified as a human
carcinogen.28 EPA has classified chromium III in Group D - Not
Classifiable as to Human Carcinogenicity.29
EPA used the multistage extrapolation model, based on data
from an occupational study of chromate production workers, to
estimate the unit cancer risk for chromium VI. EPA calculated an
IURE of 1.2 x 10~2 (yug/m3)'1.30 Based upon this unit risk
estimate, EPA estimates that if an individual were to breathe air
containing chromium VI at 0.00008 //g/m3 b over his or her entire
lifetime, that person would theoretically have an increased
chance of up to a one in one million of developing cancer as a
direct result of breathing air containing this chemical.
Similarly, EPA estimates that breathing air containing
0.0008 jug/m3 would result in an increased chance of up to one in
one hundred thousand of developing cancer.31 EPA has not
calculated a risk estimate from oral exposure to chromium VI32 or
from inhalation or oral exposure to chromium III.33
EPA has confidence in the risk estimate for chromium VI,
based on the fact that the results of studies of chromium
exposure are consistent across investigators and countries and
because a dose response for lung tumors has been established.
However, an overestimation of risk may exist due to the implicit
assumption that the smoking habits of chromate workers were
similar to those of the general white male population, because it
is generally accepted that the proportion of smokers is higher
for industrial workers than for the general population.34
The International Agency for Research on Cancer (IARC) has
stated that there is sufficient evidence in humans for the
carcinogenicity of chromium VI compounds and inadequate evidence
in humans for the carcinogenicity of chromium III compounds.35
E.5.2 Noncancer Effects
This section presents information from human and/or animal
studies on the acute (short-term), chronic (long-term), and
reproductive/developmental effects of chromium VI and
chromium III.
0.00008 yug/m3 (concentration corresponding to a 10"6 risk level) = 10 6 (risk
level)/1.2 x 10"2 (Mg/ro3)"1 (unit risk estimate).
E-10
-------
E.5.2.1 Acute (Short-Term) for Chromium. The respiratory
tract is the major target organ for chromium VI following
inhalation exposure in humans. Dyspnea, coughing, and wheezing
were reported in cases in which individual inhaled very high
concentrations of chromium VI. Other effects noted from acute
inhalation and oral exposure to very high concentrations of
chromium VI include gastrointestinal and neurological effects,
while dermal exposure causes skin burns.36
Acute animal studies have reported chromium VI to have
extreme toxicity from inhalation and oral exposure. This is
based on data showing the LC50 for chromium VI to be less than
200 mg/m3 and the LD50 to be less than 50 mg/kg. Chromium III
has been shown to have moderate toxicity from oral exposure,
based on LD50 data in the range of 500 to 5,000 mg/kg. The
kidney is the major target organ for chromium VI acute toxicity
in animals, with high doses resulting in kidney failure. Other
target organs include the brain and the liver.37
E.5.2.2 Chronic (Lona-Term) for Chromium. Chronic
inhalation exposure to chromium VI in humans results in effects
on the respiratory tract, with perforations and ulcerations of
the septum, bronchitis, decreased pulmonary function, pneumonia,
asthma, and nasal itching and soreness reported. Chronic
exposure to high levels of chromium VI by inhalation or oral
exposure may also produce effects on the liver, kidney,
gastrointestinal and immune systems, and possibly the blood.
Dermal exposure to chromium VI may cause contact dermatitis,
sensitivity, and ulceration of the skin.38
Limited information is available on the chronic effects of
chromium in animals. The available data indicate that, following
inhalation exposure, the lung and kidney have the highest tissue
levels of chromium. No effects were noted in several oral animal
studies with chromium VI and chromium III.39
EPA has established RfD for chromium VI of 0.005 mg/kg/day,
based upon a NOAEL (adjusted) of 2.4 mg/kg/day, an uncertainty
factor of 500, and a modifying factor of I.40 This was based on
a study of rats, which reported no adverse effects after their
exposure to chromium VI in the drinking water for 1 year. Other
studies support these findings; one study reported no significant
effects in female dogs given chromium VI in the drinking water
for 4 years, and a case study on humans reported no adverse
health effects in a family of four who drank water for 3 years
from a private well containing chromium VI at 1 mg/L.41
E-ll
-------
EPA has low confidence in the study on which the RfD for
chromium (VI) was based and in the RfD. Confidence in the key
study was ranked low due to the small number of animals tested,
the small number of parameters measured, and the lack of toxic
effects at the highest dose tested. The low ranking of the RfD
was due to lack of high-quality supporting studies and the fact
that developmental and reproductive effects are not well
studied.42
The RfD for chromium III is 1 mg/kg/day, based upon a NOAEL
(adjusted) of 1,468 mg/kg/day, an uncertainty factor of 1,000,
and a modifying factor of I.43 This was based on no effects
observed in rats fed chromium III in the diet for 2 years. EPA
has low confidence in the study on which the RfD was based and in
the RfD. The low ranking of the key study was due to the lack of
explicit detail on study protocol and results, while the low
ranking of the RfD was due to the lack of supporting data and the
lack of an observed effect level in the key study.44 EPA has not
established an RfC for chromium III45 or chromium VI.46
E.5.2.3 Reproductive and Developmental for Chromium.
Limited information is available on the reproductive or
developmental effects of chromium in humans. The only available
data suggest that exposure to chromium (VI) by inhalation in
women may result in complications during pregnancy and
childbirth.47
Animal studies have not reported reproductive effects from
inhalation exposure to chromium (VI). Oral studies on chromium
(VI) have reported severe developmental effects in mice such as
gross abnormalities and reproductive effects including decreased
litter size, reduced sperm count, and degeneration of the outer
cellular layer of the seminiferous tubules. No information is
available on the reproductive or developmental effects of
chromium (III) in humans or animals.48
E.6 HYDROCHLORIC ACID HEALTH EFFECTS SUMMARY
Hydrochloric acid is an aqueous solution of hydrogen
chloride gas and is commercially available in several
concentrations and purities. Because of impurities, commercial
varieties of hydrochloric acid are generally yellow.
Hydrochloric acid is used in refining metal ore, as a lab
reagent, and in the removal of scale from boilers.49
E-12
-------
E.6.1 Cancer Effects
Limited information is available on the possible
carcinogenic effects of hydrochloric acid. No information is
available on the cancer risk to humans from exposure to
hydrochloric acid. The carcinogenic effects of combined and
separate exposures via inhalation to formaldehyde and
hydrochloric acid were investigated in a study on rats. No
carcinogenic response was observed when rats were exposed only to
hydrochloric acid at concentrations of 10 ppm.50 No studies have
investigated risk of cancer in animals as a result of oral or
dermal exposures.
EPA has not classified hydrochloric acid with respect to
potential carcinogenicity and has not estimated the unit cancer
risk associated with hydrochloric acid.51
E.6.2 Noncancer Effects — Hydrogen Cloride
E.6.2.1 Acute (Short-Term) Effects for Hydrogen Cloride.
The acute effects on humans exposed by inhalation to hydrochloric
acid include coughing, choking, inflammation and ulceration of
the respiratory tract, chest pain, and pulmonary edema. Oral
exposure may result in corrosion of the mucous membranes,
esophagus, and stomach, with nausea, vomiting/ intense thirst,
and diarrhea. Dermal contact with hydrochloric acid can cause
burns, ulcerations, and scarring.52
Animals exposed to 320 parts per million (ppm) for 6 minutes
suffered sensory irritation, while levels of 680 ppm or higher
for 1 minute resulted in less severe effects; inhalation of air
containing 6,400 mg/m3 hydrochloric acid for 30 minutes resulted
in death from laryngeal spasm, laryngeal edema, or rapidly
developing pulmonary edema.53 Acute inhalation exposure tests
resulted in an LC50 of 1,108 ppm for exposed mice and 3,124 ppm
for exposed rats (moderate to high acute toxicity). An LD50 of
900 mg/kg (moderate acute toxicity) was reported for rabbits
exposed orally to hydrochloric acid.54 No information is
available on effects in animals from acute dermal exposure to
hydrochloric acid.
E.6.2.2 Chronic (Loncr-Term) Effects for Hydrogen Cloride.
In humans, cases of gastritis, chronic bronchitis, dermatitis,
and photosensitization have been reported among individuals
exposed occupationally to hydrochloric acid.55 No other data are
available specifically on the effects of long-term human exposure
dermally or via inhalation or ingestion.
E-13
-------
In animals, the only study of the effects of long-term
inhalation of hydrochloric acid reported epithelial or squamous
hyperplasia of the nasal mucosa, larynx, and trachea. In a
90-day inhalation study, decreased body weight gains, minimum to
mild rhinitis, nasal cavity lesions, and eosinophilic globules in
the epithelial lining of the nasal tissues were reported in test
animals.56 No studies are available on the long-term effects on
animals from low-level oral or dermal exposures to hydrochloric
acid.
EPA has established an RfC for hydrochloric acid of
0.02 mg/m3. This concentration was based on a rate study in
which hyperplasia of the nasal mucosa, larynx, and trachea were
seen. An uncertainty factor of 300 was applied to an LOAEL of
6.1 mg/m3.57 The EPA has low confidence in the study, database,
and RfC because the study used only one dose and the database did
not provide any additional chronic or reproductive studies.58
E.6.2.3 Reproductive and Developmental for Hvdroaen
Cloride. No information is available on reproductive or
developmental effects of hydrochloric acid in humans. In animal
studies in which female rats were exposed via inhalation prior to
mating and during gestation, severe dyspnea, cyanosis, and
altered estrus cycles were noted in the dams; increased fetal
mortality and decreased fetal weight were also reported in
offspring.59 No animal studies are available on reproductive or
developmental effects of oral or dermal exposure.
E.7 HYDROGEN FLUORIDE HEALTH EFFECTS SUMMARY
Hydrogen fluoride (HF) is a colorless gas that is used in
making aluminum and in making chlorofluorocarbons. HF readily
dissolves in water, is present in the air or other media, and, in
the dissolved form, is known as hydrofluoric acid. Air around
hazardous waste sites or factories that use or produce HF may
contain this chemical.60 EPA has set a maximum contaminant level
(MCL) of 4 mg/L for HF.61
E.7.1 Cancer Effects — Hydrogen Fluoride
A cohort of workers in Denmark exposed to hydrofluoric fumes
or dust reported an increase in mortality and morbidity from
respiratory cancer. Increased lung cancer rates have been
reported in aluminum industry workers, although no correction was
made for smoking and exposure to other chemicals.
.Epidemiological studies of populations exposed to fluorides
through drinking water have not shown an increased risk of
cancer. No data are available on cancer in humans following
dermal exposure to HF.62 No animal studies have been identified
E-14
-------
regarding the carcinogenic effects of HF. EPA has not classified
HF with respect to carcinogenicity and has not estimated a unit
risk for HF.63
E.7.2 Noncancer Effects — Hydrogen Fluoride
E.7.2.1 Acute (Short-Term) Effects for Hvdroaen Fluoride.
Acute (short-term) inhalation exposure to HF can cause severe
respiratory damage in humans, including severe irritation and
pulmonary edema. Many of the human studies regarding inhalation
of HF also involved dermal exposure, making it difficult to
determine which effects are specific to the inhalation route.
The results of ingestion include necrosis of the esophagus and
stomach with nausea, vomiting, diarrhea, circulatory collapse,
and death. Severe ocular irritation and dermal burns may occur.
following eye or skin exposure.64'65
In animals, acute inhalation exposure has resulted in renal
and hepatic damage. HF produces irritation of the eyes, skin,
and conjunctivae in rats as a result of dermal exposure. No
information was found on the effects on animals from oral
exposures to HF.66
E.7.2.2 Chronic (Long-Term) Effects for Hydrogen Fluoride.
The major health effect of chronic inhalation exposure to HF and
fluoride dusts, either individually or in combination, is
skeletal fluorosis.67 Chronic inhalation exposure of humans to
HF has resulted in irritation and congestion of the nose, throat,
and bronchi at low levels.68 In addition, persons exposed
occupationally to HF and fluoride dusts in an aluminum smelter
reported reduced expiratory volume and increased cough and sputum
production. No information is available on the chronic effects
of oral or dermal exposure to HF in humans.69
Limited information exists on the chronic effects of HF in
animals. Damage to the liver, kidneys, and lungs has been
observed in animals chronically exposed to HF by inhalation.70
No information was found on the long-term effects of oral or
dermal exposure in animals. EPA is reviewing the RfC and Rf D for
HF.71
E. 7 . 2 . 3 Reproductive and Developmental Effects for Hydrogen
Fluoride. No studies were located regarding the developmental
and reproductive effects in humans from inhalation, oral, or
dermal exposure to HF.72
Dogs exposed via inhalation to HF developed degenerative
testicular changes and ulceration of the scrotum. No studies
E-15
-------
were found regarding the reproductive and developmental effects
in animals from oral or dermal exposure.73
E.8 MERCURY HEALTH EFFECTS SUMMARY
Mercury is a naturally occurring element that exists in
three forms: elemental mercury, inorganic mercury (primarily
mercuric chloride), and organic mercury (primarily methyl
mercury). Elemental mercury is a shiny, silver-white, odorless
liquid; inorganic mercury compounds are usually white powders or
crystals; and organic mercury compounds are white crystalline
solids. The majority of mercury in air is elemental mercury
vapor, which is released to the air by natural and industrial
sources. EPA has set a Maximum Contaminant Level (MCL) of
0.002 mg/L for mercury.74 The health effects of mercury are
discussed in the draft Mercury Study Report to Congress, which is
currently being reviewed by the Science Advisory Board (SAB).
See Chapter 7 of this interim utility study report for
information on how to obtain the draft mercury study.
E.9 NICKEL HEALTH EFFECTS SUMMARY
Nickel is a silvery-white metal that is usually found in
nature as a component of silicate, sulfide, or arsenide ores.
The following table presents the physical properties of some of
the major forms of nickel.
The most predominant forms of nickel in the atmosphere are
probably nickel sulfate, nickel oxides, and the complex oxides of
nickel. Each form of nickel exhibits different physical
properties. Nickel compounds may be divided into two groups:
soluble and insoluble nickel compounds. The soluble compounds
include nickel sulfate and nickel acetate. Insoluble compounds
include nickel monoxide, metallic nickel, nickel hydroxide,
nickel subsulfide, and nickel carbonyl. Most nickel is used to
make stainless steel; other uses include the manufacture of
batteries, electroplating baths, textile dyes, coins,
spark-plugs, and machinery parts.
E.9.1 Cancer Effects — Nickel
Human studies have reported an increased risk of lung and
nasal cancers among nickel refinery workers exposed to nickel
refinery dust and to nickel sulfate.75 Nickel refinery dust is
defined as the "dust from pyro-metallurgical sulfide nickel
matte" refineries and is a mixture of many nickel compounds,
including nickel subsulfide. It is not clear which compound is
carcinogenic in the nickel refinery dust.76 No information is
E-16
-------
Chemical Name
Metallic Nickel
Nickel Hydroxide
Nickel Subsulfide
Nickel Carbonyl
Nickel Sulfate
(anhydrous)
Nickel Monoxide
Nickel Acetate
Formula
Ni
Ni (OH)2
Ni3S2
Ni (0)4
NiS04
NiO
Ni(OCOCH3)2
Description
Lustrous white, hard
ferromagnetic metal or grey
powder
Green crystals or
amorphous solid
Lustrous pale yellow or
bronze metallic crystals
Colorless to yellow liquid
Pale-green to yellow crystals
Grey, black, or green
powder
Dull-green crystals
Solubility
Soluble in dilute nitric acids; slightly
soluble in hydrochloric or sulfuric acids;
insoluble in cold or hot water
Nearly insoluble in cold water; soluble in
acid, ammonium hydroxide
Insoluble in cold water; soluble in nitric
acid
Nearly insoluble in water; soluble in
ethanol, benzene, and nitric acid;
insoluble in dilute acids or dilute alkali
Soluble in water; insoluble in ethanol
Insoluble in water; soluble in acid
Soluble in water; insoluble in ethanol
Source: IARC199035
available on the carcinogenic effects of
oral or dermal exposure.77'78
nickel in humans from
Animal studies have reported lung tumors from inhalation
exposure to the following nickel compounds and mixtures: nickel
refinery dusts, nickel sulfate, nickel subsulfide, nickel
carbonyl, and metallic nickel. Studies in animals have reported
tumors from intramuscular and other routes of administration from
exposure to nickel monoxide and nickel hydroxide. Oral animal
studies have not reported tumors from exposure to nickel acetate
in the drinking water. No information is available on the
carcinogenic effects of nickel in animals from dermal
exposure
79-82
E.9.1.1 Cancer Effects for Nickel Refinery Dust. EPA has
classified nickel refinery dust in Group A - Known Human
Carcinogen. For nickel refinery dust, the Group A classification
was based on an increased risk of lung and nasal cancer in humans
through inhalation exposure and increased lung tumor incidences
in animals.83 The International Agency for Research on Cancer
(IARC) has classified nickel refinery dust as having sufficient
evidence in humans for carcinogenicity. This is based on the
same information EPA used.
EPA used the additive and multiplicative extrapolation
method, based on human data, to estimate the unit cancer risk for
E-17
-------
nickel refinery dust. EPA calculated an inhalation unit risk
estimate of 2.4 x 10~4 (/ug/m3*'1.84 Based upon this unit risk
estimate, EPA estimates that if an individual were to breathe air
containing nickel refinery dust at 0.004 /ug/m3 over his or her
entire lifetime (70 yrs, 24 hrs/day), that person would
theoretically have an increased chance of up to one in one
million of developing cancer as a direct result of breathing air
containing this chemical. Similarly, EPA estimates that
breathing air containing 0.04 /ug/m3 would result in an increased
chance of an increased chance of up to one in one hundred
thousand of developing cancer.85
EPA used four data sets, all from human exposure, to
calculate the unit risk estimates for nickel refinery dusts. A
range of incremental unit risk estimates was calculated from
these data sets that were consistent with each other.86
E.9.1.2 Cancer Effects for Nickel Sulfate. The National
Toxicology Program (NTP) has recently completed a draft report on
the carcinogenic effects of nickel sulfate hexahydrate. They
have concluded that there was no evidence of carcinogenic
activity of nickel sulfate hexahydrate in male or female rats or
male or female mice. These conclusions are based on the results
of 2-year inhalation studies.87
The International Committee on Nickel Carcinogenesis in Man
summarized the available epidemiologic data on nickel and
concluded that there was strong evidence that exposure to soluble
nickel (primarily nickel sulfate) was associated with an
increased respiratory cancer risk.88
The International Agency for Research on Cancer (IARC) has
classified nickel sulfate as having sufficient evidence in humans
for carcinogenicity.89 This is based on epidemiological studies
that showed an increased risk of lung and nasal cancer through
inhalation exposure. In addition, animal studies have reported
malignant tumors in the peritoneal cavity when nickel sulfate was
applied by intraperitoneal injections.90
E.9.1.3 Cancer Effects for Nickel Subsulfide. EPA has also
classified nickel subsulfide in Group A, based upon the same
studies as those that were used to classify nickel refinery
dust.91 For nickel subsulfide, EPA also used human data to
estimate the unit cancer risk. EPA calculated an inhalation unit
risk estimate of 4.8 x 10~4 (//g/m3) ~1.92 EPA estimates that if an
individual were to breathe air containing this nickel compound at
0.002 //g/m3 over his or her entire lifetime, that person would
theoretically have an increased chance of up to one in one
E-18
-------
million chance of developing cancer as a direct result of
breathing air containing this chemical. Similarly, EPA estimates
that breathing air containing 0.02 /ug/m3 would result in an
increased chance of up to one in one hundred thousand chance of
developing cancer. EPA has also calculated unit risk estimates
for nickel subsulfide from a rat inhalation study. These
estimates were approximately one order of magnitude greater than
those calculated from the human studies.93
The National Toxicology Program has recently completed a
draft report on the carcinogenic effects of nickel subsulfide.
They have concluded that there was clear evidence of carcinogenic
activity of nickel subsulfide in male and female rats and no
evidence of carcinogenic activity for male and female mice.
These conclusions are based on the results of 2-year inhalation
studies.94
IARC has classified nickel subsulfide as having sufficient
evidence in humans and experimental animals for
carcinogenicity.95 The International Committee on Nickel
Carcinogenesis in Man concluded that there was some evidence to
suggest that exposure to nickel subsulfide presents on increased
risk of lung and nasal cancer.96
The State of California has calculated an estimated unit
risk for continuous lifetime exposure to nickel subsulfide at
1 //g Ni/m3. This risk ranges from 2.8 x 10"3 for the maximum
likelihood estimate to 3.7 x 10~3 for the upper 95 percent
confidence limit. This risk estimate was based on animal data.97
E.9.1.4 Cancer Effects for Nickel Carbonvl. EPA has
classified nickel carbonyl in Group B2 - Probable Human
Carcinogen. For nickel carbonyl, this classification was based
on an increase in lung tumors in animals exposed via
inhalation.98 IARC has classified nickel carbonyl as having
limited evidence in experimental animals for carcinogenicity.99
This is based on the same information as that EPA used.
EPA has not calculated an inhalation or an oral unit cancer
risk estimate for nickel carbonyl, due to the lack of appropriate
data. In one study, the survival rate of the animals was very
low, and another study used the intravenous route of exposure.100
E.9.1.5 Cancer Effects for Nickel Monoxide. The NTP has
recently completed a draft report on the carcinogenic effects of
nickel monoxide. They have concluded that there was some
evidence of carcinogenic activity of nickel monoxide in male and
female rats, no evidence of carcinogenic activity in male mice,
E-19
-------
and equivocal evidence of carcinogenic activity in female mice.
These conclusions are based on the results of 2-year inhalation
studies.101
IARC has classified nickel monoxide as having sufficient
evidence in experimental animals for carcinogenicity.102 This is
based on animal studies that showed an increased incidence of
tumors in rats exposed via intrapleural, intramuscular, and
intraperitoneal administration. The International Committee on
Nickel Carcinogenesis summarized the available epidemiologic data
on nickel and concluded that there was some evidence to suggest
that exposure to oxidic nickel (including nickel monoxide) may
result in increased lung and nasal cancer risks.103
E.9.1.6 Cancer Effects for Nickel Hydroxide. IARC has
classified nickel hydroxide as having sufficient evidence in
experimental animals for carcinogenicity.104 This is based on
animal studies that showed an increase in tumors in rats exposed
via intramuscular injection.
E.9.1.7 Cancer Effects for Metallic Nickel. IARC has
classified metallic nickel as having sufficient evidence in
experimental animals for carcinogenicity.105 This is based on
animal studies that showed an increase in tumors from exposure
via inhalation and intratracheal, intraperitoneal, and
intravenous administration. The International Committee on
Nickel Carcinogenesis in Man summarized the available data on
nickel and concluded that the available information gave no
evidence of increased respiratory cancer risks from exposure to
metallic nickel.106
E.9.1.8 Nickel Acetate. IARC has not classified nickel
acetate as to carcinogenicity. 7
E.9.1. 9 Overall Assessment for Nickel Compounds. IARC
examined all of the data on nickel and stated that for an overall
evaluation, it considers nickel compounds to be carcinogenic to
humans and metallic nickel to be possibly carcinogenic to
humans.108
The State of California has calculated an estimated unit
risk for continuous lifetime exposure to nickel compounds at
1 //g/m3. This risk ranges from 2.1 x 10~4 for the maximum
likelihood estimate to 2.57 x 10~4 for the upper 95 percent
confidence limit. This risk estimate was based on human data.
They also concluded that all nickel compounds should be
considered potentially carcinogenic to humans by inhalation.109
E-20
-------
The American Conference of Governmental Industrial
Hygienists (ACGIH) have stated that all nickel compounds should
be considered as confirmed human carcinogens, based on the weight
of evidence from epidemiologic studies of, or convincing clinical
evidence in, exposed humans.110
The International Committee on Nickel Carcinogenesis in Man
concluded that more than one form of nickel gives rise to lung
and nasal cancer. They stated that although much of the
respiratory cancer risk seen among nickel refinery workers could
be attributed to exposure to a mixture of nickel oxides and
sulfides, exposure to large concentrations of nickel oxides in
the absence of nickel sulfides was also associated with increased
lung and nasal cancer risks. In addition, there was evidence
that soluble nickel exposure (such as nickel sulfate) increased
the risk of these cancers. They concluded that respiratory
cancer risks are primarily related to exposure to soluble nickel
at concentrations greater than 1 mg/m3 and to exposure to less
soluble forms at concentrations greater than 10 mg/m3.111
E.9.2 Noncancer Effects — Nickel
E.9.2.1 Acute (Short-Term) Effects for Nickel. Nickel
carbonyl appears to be the most acutely toxic nickel compound.
Symptoms from acute inhalation exposure in humans include
headache, vertigo, nausea, vomiting, insomnia, and irritability,
followed by chest pains, dry coughing, cyanosis, gastrointestinal
symptoms, sweating, visual disturbances, and severe weakness.
Acute oral exposure to high levels of nickel sulfate and nickel
chloride in humans has resulted in vomiting, cramps, impaired
vision, giddiness, headache, and cardiac arrest in humans. No
information is available on the acute effects of nickel via
dermal exposure in humans.112
The lungs and kidneys appear to be target organs for acute
nickel carbonyl toxicity, via inhalation and oral exposure in
animals, with pulmonary fibrosis and renal edema reported. No
information is available on acute effects of nickel via dermal
exposure in animals.113 Acute animal tests, such as the LD50 test
in rats, have shown nickel compounds to exhibit acute toxicity
values ranging from low to high, based upon LD50 data in the
range of 50 mg/kg to greater than 5,000 mg/kg. The soluble
compounds, such as nickel acetate, were most toxic, and the
insoluble compounds, such as metallic nickel powder, were the
least toxic.114
E.9.2.2 Chronic (Long-Term) Effects for Nickel. Contact
dermatitis is the most common effect in humans from exposure to
E-21
-------
nickel, via inhalation, oral, and dermal exposure. Cases of
nickel contact dermatitis have been reported following
occupational and nonoccupational exposure, with symptoms of
itching of fingers, wrists, and forearms. Chronic inhalation
exposure to nickel in humans also results in respiratory effects.
These effects include direct respiratory effects such as asthma
due to primary irritation or an allergic response and an
increased risk of chronic respiratory tract infections.115'116
Animal studies have reported effects on the lungs, kidneys,
and immune system from inhalation exposure to nickel, and effects
on the respiratory and gastrointestinal systems, heart, blood,
liver, kidney, and decreased body weight from oral exposure to
nickel. Dermal animal studies have reported effects on the
skin.117'118
E.9.3 Essentiality for Nickel
Nickel has been demonstrated to be an essential nutrient for
some mammalian species, and it has been suggested that it may
also be essential for human nutrition. A requirement for nickel
has not been conclusively demonstrated in humans, and a
recommended daily allowance has not been set. By extrapolation
from animal data, there have been various estimates of the human
daily requirement for nickel. The National Academy of Sciences
estimated that a 70 kilogram person would have a daily
requirement of 50 /^g of nickel.119 Other researchers have
estimated requirements ranging from 30 /u.g to 120 fj.g of nickel.120
E.9.4 Reproductive and Developmental Effects for Nickel
No information is available regarding the reproductive or
developmental effects of nickel in humans. Animal studies have
reported developmental effects, such as a reduction in fetal body
weight, and reproductive effects, including testicular
degeneration from inhalation exposure to nickel. Oral animal
studies have reported deaths in females due to pregnancy
complications and a significant decrease in number of offspring
per litter from exposure to nickel.121
E.9.5 Noncancer Health-Based Numbers for Nickel
EPA has established a Reference Dose (RfD) for nickel
(soluble salts) of 0.02 mg/kg/day, based upon a NOAEL (adjusted)
of 5 mg/kg/day, an uncertainty factor of 300, and a modifying
factor of I.122 This was based on a study in rats that showed
decreased body and organ weights from chronic (2-year) exposure
to nickel in the diet. Other studies showed similar results,
with decreased body and organ weights after exposure to nickel
chloride via gavage and through the drinking water. EPA has not
E-22
-------
established a Reference Concentration (RfC) for any nickel
compound.123
EPA has medium confidence in the RfD for nickel (soluble
salts) and low confidence in the study on which it was based.
The Ambrose et al.124 study was properly designed and provided
adequate toxicological endpoints; however, high mortality
occurred in the controls. The database provided adequate
supporting subchronic studies; this was the basis for EPA's
medium confidence level in the RfD.125
The EPRI has recommended a RfC of 2.38 x 10~3 mg(Ni)/m3 for
all nickel compounds. This was based on the ACGIH Threshold
Limit Value (TLV). It was translated for community exposure by
scaling for exposure time differences between community and
occupational exposure assumptions.126
Calabrese has calculated an ambient air level goal (AALG)
for soluble nickel compounds of 0.36 ng (Ni)/m3 and an AALG for
insoluble nickel compounds of 7.1 ng (Ni)/m3. An AALG is a
health-based guideline based on risk assessment methodology
similar to that used by EPA.127
The California Air Resources Board has stated that the most
sensitive noncancer endpoint reported in humans is allergic
sensitization, while immune suppression is the most sensitive
endpoint reported in animal studies. The board has concluded
that because these noncancer effects occur at concentrations
greater than 3 orders of magnitude above a 24-hour maximum
concentration of nickel (0.024 ng(Ni)/m3) measured in California
near an industrial source, it is unlikely that noncancer health
effects would be caused by the levels of nickel compounds
currently in the air.128
The Agency of Toxic Substances and Disease Registry78 has
recommended a minimum risk level (MRL) for intermediate duration,
inhalation exposure to nickel of 9.5 x 10~5 mg(Ni)/m3. They have
stated that this MRL may not be protective for some
hypersensitive individuals.129 An MRL is a health-based
guideline based on similar risk assessment methodology to that
used by EPA.
E.9.6 Federal Regulations and Guidelines for Nickel
The Occupational Safety and Health Administration (OSHA) has
established a maximum allowable level of nickel in workplace air
for an 8-hour workday, 40-hour workweek of 1 mg(Ni)/m3 for
metallic nickel and insoluble compounds, and 0.1 mg(Ni)/m3 for
soluble nickel compounds.130
E-23
-------
The National Institute of Occupational Safety and Health
has a recommended exposure level for workplace air of
0.15 mg (Ni)/m3 for all nickel compounds except nickel carbonyl
and 7 yug (Ni)/m3 for nickel carbonyl.131
The ACGIH has recommended a TLV of 0.05 mg(Ni)/m3 for an
8-hour exposure in the workplace to all nickel compounds
(elemental, insoluble, and soluble),132
The EPA has set a maximum contaminant level (MCL) of
0.1 mg/L for nickel. This is the maximum level allowed in
drinking water.133
E.10 2,3,7,8-TETRACHLORODIBENZO-P-DIOXIN HEALTH EFFECTS SUMMARY
2, 3,7,8-Tetrachlorodibenzo-p-dioxin (2,3,7,8-TCDD) belongs
to the class of compounds, chlorinated dibenzo-p-dioxins, which
are referred to as dioxins. 2,3,7,8-TCDD is a colorless solid
with no known odor. It does not occur naturally, nor is it
intentionally manufactured by any industry, although it can be
produced inadvertently in small amounts as an impurity during the
manufacture of certain herbicides and germicides and has been
detected in products of incineration of municipal and industrial
wastes. The only present use for 2,3,7,8-TCDD is in chemical
research.134
E.10.1 Cancer Effects — Dioxins
An increase in lung cancer risks was observed among Japanese
males exposed as a result of an oil poisoning accident. Human
studies have also found an association between 2,3,7, 8-TCDD and
soft-tissue sarcomas, lymphomas, and stomach carcinomas, although
for malignant lymphomas, the increase in risk is not consistent.
The increase in risk is of borderline significance for highly
exposed groups and is less-significant among groups exposed to
lower levels of 2,3,7,8-TCDD. Although there are problems with
the studies of human effects, such as confounding factors, short
follow-up period, and lack of exposure information, the overall
weight of evidence from epidemiological studies suggests that the
generally increased risk of cancer in humans is likely due to
2,3,7,8-TCDD.135
Information on the carcinogenicity of 2,3,7,8-TCDD following
inhalation exposure of animals is not available. In animal
studies of oral exposure to 2,3,7,8-TCDD, multisite tumors in
rats and mice including the tongue, lung, nasal turbinates,
liver, and thyroid have been reported. Estimates derived from
human data suggest a unit risk for lung cancer of 3 x 10~4 to
E-24
-------
5 x 10~4 pg/kg-day) ~1; for all cancers combined the unit risk
estimate is 2 x 10~3 to 3 x 10'3 (pg/kg-day)-1.136
E.10.2 Noncancer Effects — Dioxins
E.10.2.1 Acute (Short-Term) Effects for Dioxins. The acute
effects on humans exposed through the spraying in Vietnam of
herbicides that contained 2,3,7,8-TCDD include diarrhea,
vomiting, skin rashes, fever, and abdominal pain.137 Routes of
exposure in these instances are not well defined and may include
inhalation as well as oral and dermal exposures.
No information is available on effects in animals from acute
inhalation exposure to 2,3,7,8-TCDD. In oral exposure studies,
2,3,7,8-TCDD is highly toxic to all laboratory animals tested
even though there are large differences in species sensitivity.
LD50 values range from 0.6 //g/kg in male guinea pigs to
5,500 ^g/kg in hamsters. Other effects on animals from acute
oral exposure include loss of body weight, hepatotoxicity, and
decreased thymus weight.138 Information on the effects of acute
dermal exposure in animals is limited, although dermal effects
have been reported.139
E.I0.2.2 Chronic (Long Term) Effects for Dioxins. No
studies are available on the inhalation toxicity of 2,3,7,8-TCDD
in humans, although such exposure may have occurred in
populations exposed to chemicals contaminated with 2,3,7,8-TCDD.
Oral exposure of humans to chemicals contaminated with
2,3,7,8-TCDD has resulted in chloracne, immunotoxicity,
hyperpigmentation, hyperkeratosis, possible hepatotoxicity,
aching muscles, loss of appetite, weight loss, digestive
disorders, headaches, neuropathy, insomnia, sensory changes, and
loss of libido.140
Chloracne is the only substantiated effect in humans
produced by dermal exposure to compounds contaminated with
2,3,7,8-TCDD.141
No information on chronic inhalation and dermal exposure is
available for animals. Oral exposure to 2,3,7,8-TCDD has
resulted in dermatitis, extreme loss of body weight, and effects
on the liver and immune system.142 EPA has not established an
RfC or RfD for 2,3,7,8-TCDD.
E.10.2.3 Reproductive and Developmental Effects for
Dioxins. Several studies have investigated the incidence of
birth defects and reproductive effects in humans exposed to
2,3,7,8-TCDD through accidental releases or the spraying of
E-25
-------
2,3,7,8-TCDD-contaminated herbicides. EPA has concluded that the
data were not inconsistent with 2,3,7,8-TCDD's adversely
affecting development, but as a result of the limitations of the
data, these studies could not prove an association with
2,3,7,8-TCDD exposure and the observed effect. The major
limitations in these human studies were the concomitant exposure
to other potentially toxic chemicals, the lack of any specific
quantitative data on the extent of exposure of individuals within
the study group, and the lack of statistical power of the
studies.143
No studies are available on the reproductive and
developmental effects in animals caused by inhalation or dermal
exposure to 2,3,7,8-TCDD.144 In oral exposure studies,
2,3,7,8-TCDD has produced fetal anomalies, including cleft palate
and hydronephrotic kidneys in mice and internal organ hemorrhage
in rats, and resulted in spontaneous abortions in monkeys and
decreased fetal survival.
E-26
-------
E.23 REFERENCES
1. U.S. Environmental Protection Agency. Integrated Risk
Information System (IRIS) Database, National Center for
Environmental Assessment, Cincinnati, OH.
2. U.S. EPA. Health Effects Notebook for Hazardous Air
Pollutants. EPA-456-d-94-1003. Air Risk Information
Support Center, RTP, NC"27711.
3. EPA. Drinking Water Regulations and Health Advisories.
Office of Water. Washington, D.C., 1994.
4. ATSDR. Toxicological Profile for Arsenic (Update). Agency
for Toxic Substances and Disease Registry. U.S. Public
Health Service. Section 2.2, 1993.
5. EPA. Integrated Risk Information System on Arsenic.
Carcin-1, 1995.
6. Ref. 5, Carcin-3.
7. Ref. 5, Carcin-4.
8. Electric Power Research Institute. Electric Utility Trace
Substances Synthesis Report. EPRI TR-104614-V4. pg. G-l,
1994.
9. Ref. 3.
10. Ref. 5, Carcin-3.
11. Ref. 5, Carcin-3.
12. Ref. 4, Section 2.2.
13. Ref. 4, Section 2.2.2.1.
14. Ref. 4, Section 2.2.
15. Ref. 4, Section 2.2.2.
16. Ref. 5, RfD-1.
17. Tseng, W.P., H.M. Chu, S.W. How, et al. Prevalence of
skin cancer in an endemic area of chronic arsenium in
Taiwan. J. Natl. Cancer Inst. 40:453-463. 1968. (as
cited in Reference 5)
E-27
-------
18. Ref. 5, RFC-1.
19. Ref. 5, RfD-5 & 6.
20. Ref. 4, Section 2.2.1.
21. Ref. 4, Section 2.2.
22. Ref. 3.
23. ATSDR. Toxicological Profile for Chromium. Agency for
Toxic Substances and Disease Registry. U.S. Public Health
Service. Section 2.2. 1991.
24. EPA. Integrated Risk Information System on Chromium VI.
Carcin-1. 1995.
25. Ref. 23, Section 2.2.
26. Ref. 24, Carcin-1.
27. Ref. 24, Carcin-1.
28. Ref. 24, Carcin-1.
29. EPA. Integrated Risk Information System on Chromium III.
Carcin-1. 1995.
30. Ref. 24, Carcin-2.
31. Ref. 24, Carcin-3.
32. Ref. 24, Carcin-2.
33. Ref. 29, Carcin-2.
34. Ref. 24, Carcin-3.
35. International Agency for Research on Cancer. IARC
Monographs on the Evaluation of Carcinogenic Risks to
Humans: Chromium, Nickel, and Welding. Volume 49. Lyon,
France, pp. 208-214. 1990.
36. Ref. 23, Section 2.2.
37. Ref. 23, Section 2.2.
38. Ref. 23, Section 2.2.
E-28
-------
39. Ref. 23, Section 2.2.
40. Ref. 24, RfD-1.
41. Ref. 24, RfD-2.
42. Ref. 24, RfD-3.
43. Ref. 29, RfD-1.
44. Ref. 29, RfD-3.
45. Ref. 29, RfC-1.
46. Ref. 24, RfC-1.
47. Ref. 23, Section 2.2.1.
48. Ref. 23, Section 2.2.
49. The Merck Index: An Encyclopedia of Chemicals, Drugs, and
Biologicals (llth edition). Ed. S. Budavari. Merck & Co.,
Inc. Rahway, NJ. p.756. 1989.
50. HSDB. Hazardous Substances Databank (online database).
U.S. Department of Health and Human Services, National
Library of Medicine, National Toxicology Information
Program. Bethesda, MD. 1992.
51. EPA. Integrated Risk Information System on Hydrogen
Chloride. Carcin-1. 1995.
52. Ref. 50.
53. Ref. 50.
54. RTECS. Registry of Toxic Effects of Chemical Substances.
U.S. Department of Health and Human Services, National
Library of Medicine, Bethesda, MD. 1992.
55. Ref. 50.
56. Ref. 51, RfC-2.
57. Ref. 51, RfC-1.
58. Ref. 51, RfC-4.
59. Ref. 51, RfC-4.
E-29
-------
60. ATSDR. Toxicological Profile for Fluorides, Hydrogen
Fluoride, and Fluorine. Agency for Toxic Substances and
Disease Registry. US Public Health Service. Section 1.
1993.
61. Ref. 3.
62. Ref. 60, Section 2.2.
63. EPA. Integrated Risk Information System on Hydrogen
Fluoride. Carcin-1. 1995.
64. Ref. 50.
65. Ref. 60, Section 2.2.
66. Ref. 60, Section 2.2.
67. Ref. 60, Section 2.2.1.
68. EPA. Health Issue Assessment: Summary Review of Health
Effects Associated with Hydrogen Fluoride and Related
Compounds. Environmental Criteria and Assessment Office,
Office of Health and Environmental Assessment, ORD,
Cincinnati, OH. Section 7.2. 1989.
69. Ref. 60, Section 2.2.
70. Ref. 68, Section 7.2.
71. Ref. 63, RfC-1, RfD-1.
72. Ref. 60, Section 2.2.
73. Ref. 60, Section 2.2.
74. Ref. 3.
75. Ref. 35, pp. 407, 408.
76. EPA. Integrated Risk Information System on Nickel Refinery
Dust. Carcin-1. 1995.
77. Ref. 76, Carcin 1 & 2.
78. ATSDR. Toxicological Profile for Nickel. Agency for Toxic
Substances and Disease Registry. U.S. Public Health
Service. Section 2.2. 1993.
E-30
-------
79. Ref. 76, Carcin-2.
80. Ref. 78, Section 2.2.
81. EPA. Integrated Risk Information System on Nickel Carbonyl.
Carcin-2. 1995.
82. EPA. Integrated Risk Information System on Nickel
Subsulfide. Carcin-2. 1995.
83. Ref. 76, Carcin-1.
84. Ref. 76, Carcin-3.
85. Ref. 76, Carcin-3.
86. Ref. 76, Carcin-3.
87. National Toxicology Program. NTP Draft Technical Report on
the Toxicology and Carcinogenesis Studies of Nickel Sulfate
Hexahydrate in F344/N Rats and B6C3F1 Mice. U.S. Dept. of
Health and Human Services. Public Health Service. National
Institute of Health. 1994.
88. International Committee on Nickel Carcinogenesis in Man.
Scand. J. Work. Environ. Health. 16:1-82. 1990.
89. Ref. 35.
90. Ref. 35.
91. Ref. 82, Carcin-1.
92. Ref. 82, Carcin-2.
93. Ref. 82, Carcin-3.
94. National Toxicology Program. NTP Draft Technical Report on
the Toxicology and Carcinogenesis Studies of Nickel
Subsulfide in F344/N Rats and B6C3F1 Mice. U.S. Dept. of
Health and Human Services. Public Health Service. National
Institute of Health. 1994.
95. Ref. 35.
96. Ref 88, p. 71-72.
E-31
-------
97. California Air Resources Board. Initial Statement of
Reasons for Rulemaking. Proposed Identification of Nickel
as a Toxic Air Contaminant, p. 4-Part B. 1991.
98. Ref. 81, Carcin-1.
99. Ref. 35, p. 410.
100. Ref. 81, Carcin-2.
101. National Toxicology Program. NTP Draft Technical Report on
the Toxicology and Carcinogenesis Studies of Nickel Oxide in
F344/N Rats and B6C3F1 Mice. U.S. Dept. of Health and Human
Services. Public Health Service. National Institute of
Health. 1994.
102. Ref. 35, p. 410.
103. Ref. 88, p. 72.
104. Ref. 35, p. 410.
105. Ref. 35, p. 410.
106. Ref. 88, p. 72-73.
107. Ref. 35, p. 410-411.
108. Ref. 35, p. 411.
109. Ref. 97, p. 4-5 Part B.
110. American Conference of Governmental Industrial Hygienists.
Threshold Limit Values for Chemical Substances and Physical
Agents and Biological Exposure Indices. Cincinnati, OH.
1995.
111. Ref. 88, p. 74-75.
112. EPA. Health Assessment Document for Nickel. Office of
Health and Environmental Assessment. EPA/600/8-83/012F.
Section 5.5.1. 1985.
113. Ref. 112, Section 5.1.2.
114. Ref. 78, Section 2.2.2.
115. Ref. 78, Section 2.2.
E-32
-------
116. Ref. 112, Section 5.2.2.
117. Ref. 78, Section 2.2.
118. Ref. 112, Section 5.2.1.3.
119. National Academy of Sciences. Drinking Water and Health.
Volume 3. Safe Drinking Water Committee. National Academy
Press. Washington, D.C. 1980.
120. Ref. 97, p. 26-31.
121. Ref. 78, Section 2.2.
122. EPA. Integrated Risk Information System on Nickel, Soluble
salts. RfD-1. 1995.
123. Ref. 122, RfC-1.
124. Ambrose et. al. Long-term toxicologic assessment of nickel
in rats and dogs. J. Food Sci. Technol. 13:181-187. 1976.
(as cited in EPA 1994d)
125. Ref. 122, RfD-6.
126. Ref. 8, p. 1-3.
127. Calabrese, E. Air Toxics and Risk Assessment. Lewis
Publishers, Inc. Chesea, MI. p. 463-471. 1991.
128. Ref. 97, Executive Summary.
129. Ref. 78, Section 2.2.
130. Ref. 78, Section 7.
131. Ref. 78, Section 7.
132. Ref. 78, Section 7.
133. Ref. 3.
134. ATSDR. Toxicological Profile for
2,3,7,8-Tetrachlorodibenzo-p-dioxin. Agency for Toxic
Substances and Disease Registry. U.S. Public Health Service.
U.S. Department of Health and Human Services. Section 1.
1989.
E-33
-------
135. EPA. Health Assessment Document for
2,3,7,8-Tetrachlorodibenzo-p-Dioxin (TCDD) and Related
Compounds. Vol II. (Draft). Office of Research and
Development. Washington, DC. Volume II, Chapter 7. 1994
136. Ref. 135, Volume I, Chapter 6; Volume III.
137. HSDB. Hazardous Substances Databank (online database).
U.S. Department of Health and Human Services, National
Library of Medicine, National Toxicology Information
Program. Bethesda, MD. 1993.
138. Ref. 134, Section 2.2.
139. Ref. 135, Volume I, Chapter 6.
140. Ref. 134, Section 2.2.
141. Ref. 134, Section 2.2.3.
142. Ref. 134, Section 2.2.2.
143. Ref. 134, Section 2.2.1.
144. Ref. 134, Section 2.2.2.
E-34
-------
Appendix F: Documentation of The Inhalation Human Exposure
Modeling for the Utility Study
-------
This page is intentionally left blank.
-------
F.1 INTRODUCTION
The model used to calculate direct inhalation risks from
(HAPs) emitted from utility boilers is the Human Exposure Model
Version 1.5 (HEM). It was developed by the Pollutant Assessment
Branch (PAB) of the EPA's OAQPS and was designed for screening
assessments. The model is used in source ranking to assess the
relative risks associated with exposure to different pollutants
and to characterize human exposure, cancer risks, and
noncarcinogenic hazards for stationary sources that emit HAPs.
The HEM uses the Industrial Source Complex - Long Term Version 2
(ISCLT2) air dispersion model, updated 1990 census population
data, meteorological, temperature, and mixing height databases,
and chemical-specific health effects numbers (see Table F-l.)
The remainder of this technical report contains a
description of ISCLT2, the population and meteorological
databases, human exposure algorithms, and risk estimating
methodology applied in HEM to arrive at direct inhalation risk
estimates for this utility study.
F.2 ISCLT2 DISPERSION MODELING
Air dispersion modeling is used to estimate atmospheric fate
and transport of pollutants from the point of emission to the
location of exposure to arrive at long-term average ambient air
concentrations of the pollutant. ISCLT2, the air dispersion
model used in HEM, is the Agency's regulatory air dispersion
model for the types of sources represented in this study. ISCLT2
is one of the primary models used to support EPA studies and
regulatory programs for air pollutants. ISCLT2 uses emission
parameters and meteorological data to estimate the transport and
dispersion of pollutants in the atmosphere.
The ISCLT is a steady-state, Gaussian plume, atmospheric
dispersion model that applies to multiple-point, area, and volume
emission sources. It is designed specifically to estimate
long-term ambient concentrations resulting from air emissions
from these source types in a computationally efficient manner.
ISCLT2 is recognized by the Guideline on Air Quality Models1 as a
preferred model for dealing with complicated sources (i.e.,
facilities with point, area, and volume sources) when estimating
long-term concentrations (i.e., monthly or longer).
As described in the Guideline on Air Quality Models, the
ISCLT is appropriate for modeling industrial source complexes in
either rural or urban areas.2 With this model, long-term ambient
concentrations can be estimated for transport distances up to
F-l
-------
Table F-l. Summary of HEM Features
Characterization
Dispersion model
Meteorological database
Population database
Exposure calculations
Single pollutant, multiple source, nationwide
ISCLT2
Data set from locations/years available on OAQPS TTN and
from the National Weather Service
1990 Census Databases
Block Level
6.9 million records
>0.5 km Interpolate air concentration to population
<0.5 km Assign population to air concentration
HEM = Human Exposure Model
ISCLT2 = Industrial Source Complex Model - Long Term Version 2
OAQPS = Office of Air Quality Planning and Standards
TTN = Technology Transfer Network
50 km. The ISCLT2 incorporates separate point, area, and volume
source computational algorithms for calculating ambient
concentrations at user-specified locations (i.e., receptors).
The locations of the receptors relative to the source locations
are determined through a user-specified Cartesian coordinate
reference system. For the utility study, receptors were placed
around the source along 16 radials, spaced every 22.5 degrees, at
distances of 0.2, 0.5, 1.0, 2.0, 5.0, 10.0, 20.0, 30.0, 40.0, and
50.0 kilometers from the source.
ISCLT2 source inputs vary according to source type. For the
point sources in this study, the inputs include emission rate,
physical stack height, stack inner diameter, stack gas exit
velocity, and stack gas exit temperature.
The ISCLT2 is a sector-averaged model that uses statistical
summaries of meteorological data to calculate long-term,
ground-level ambient concentrations. The principal
meteorological inputs to the ISCLT2 are stability array (STAR)
summaries that consist of a tabulation of the joint frequency of
occurrence of wind speed categories, wind-direction sectors, and
Pasquill atmospheric stability categories. Other meteorological
data requirements include average mixing heights for each
stability class and average ambient air temperatures.
F-2
-------
As described above, the ISCLT2 model computes long-term
ambient concentrations at user-specified receptor points that
occur as a result of air emissions from multiple sources . These
computations are done on an emission point (stack) -by-stack basis
such that the ambient concentration from each stack at each
receptor is computed. Total ambient concentrations at a
particular receptor are obtained by summing the contributions
from each of the stacks. With Gaussian plume algorithms such as
those included in the ISCLT2, the source contributions at each
receptor are directly proportional to the source emission rate.
Normalized ambient concentrations for each source-receptor
combination were computed such that they would correspond to a
unit emission rate of 1 gram per second (g/s) for each stack in
the facility. The total ambient concentration at a receptor is
then computed as the sum of the contributions from each stack,
where the latter are computed as the product of the normalized
concentration and the desired emission rate. Mathematically,
this can be expressed as follows:
7=1
Where:
Xj = total ambient concentration at receptor i, ug/m3
9i = emission rate for stack, g/s
Xj: = normalized contribution from stack j to receptor i, ug/m3
J = total number of stacks.
Thus, the principal output of the dispersion modeling is a set of
normalized stack contributions, i.e., x^ in the above equation
for each scenario modeled.
F.2.1 Assumptions Used
For the utility study, HEM analysis flat terrain was assumed
because of the lack of information. Building downwash was not
considered because of the tall stacks used by the utility
boilers. The assumption was made that all particles were small
enough to behave as gases. All emissions from one site are
assumed to originate from stacks that are collocated.
F.2.2 Model Options
Air dispersion is affected by surface roughness. The ISCLT2
model provides two 'regimes of surface roughness based on land
F-3
-------
classification: urban and rural. When there is no information
available regarding the land classification around a particular
source of interest, the air quality modeling guidelines suggest a
surrogate, population density, to make a land classification
determination. Because the population database which is part of
the HEM model can easily provide population density estimates,
this option was selected for the utility study for conducting the
more detailed analyses. Initial screening analysis assumed the
plant setting of "urban," which earlier sensitivity analysis
indicated would maximize surrounding ambient concentrations
estimates.
EPA's Guideline on Air Quality Models3 distinguishes between
urban and rural settings based on population density. "Urban" is
defined as a population density greater than 750 people per km2
in the area between the point source and a 3 km distance from the
source; "rural" is assumed for a population density of less than
750 people per km2.
ISCLT2 can be run in a number of different ways by changing
various modeling options. For consistency in regulatory modeling
applications, a set of choices has been defined as the default
option. The default option set determines how the model
calculates ambient air concentrations and includes:
• default stack-tip downwash calculations
• buoyancy-induced dispersion calculations
• final plume rise in all calculations
• calms processing routines
• upper-bound concentration estimates for sources
influenced by building downwash from super-squat
buildings
• default wind profile exponents
• default vertical potential temperature gradients.
The default option set was used in the utility study with
one change. The plume rise was changed from the final plume rise
option of the default selections to the use of a transitional
plume rise. Plume rise accounts for how the plume behaves near
the stack as a function of the momentum of release of the plume
and the buoyant rising of the plume resulting from the high plume
temperature in comparison to the surrounding air. The use of the
F-4
-------
transitional plume rise would be expected to produce more
realistic estimates of ambient air concentrations near the stack
where the maximum concentrations occur. Each of these defaults
is defined further in the ISCLT2 User's Guide.4
F.3 HEM DATABASES
Four databases are contained in the HEM model. The
meteorological database contains long-term summaries for selected
locations across the country. HEM pairs plant locations with the
nearest location for meteorological data contained in the
database. The second database is the population database, which
contains population data from the 1990 census. Ambient air
concentrations of the modeled pollutant are coupled with the
population numbers and location to develop nationwide exposure
estimates. The two remaining databases contain estimates of
ambient temperatures and mixing height.
F.3.1 Meteorological Database
The ISCLT2 meteorological database contains long-term
meteorologic data, primarily from National Weather Service (NWS)
airport locations, in the form of STablility ARrays (STAR
summaries). STAR summaries display joint frequencies of
occurrence of wind direction, wind speed, and air stability by
combining these factors into a frequency distribution.
HEM chooses the STAR Data set for each plant based on proximity
of the plant to the location where the meteorological data were
collected.5
The meteorological database used for the utility study
contains data from hourly surface observations obtained from the
OAQPS Technology Transfer System (TTN). The Support Center for
Regulatory Air Models Bulletin Board System (SCRAM-BBS) contains
annual data files of surface observations from 349 NWS locations
(primarily airports) across the United States and its Territories
for the years 1984-1989. From each location's surface
observations, STAR summaries were created that encompass all
available years into one long-term estimate of the location's
dispersion characteristics. Figure F-l depicts the coverage of
the HEM meteorological database. The range of averaging years
over which the data are averaged is from 1 to 6 years, with a
typical average of 6 years (225 sites).
F.3.2 Population Database
The population database contains "block level" 1990 census
data collected by the U.S. Census Bureau for reapportionment as
specified in Public Law 94-17. It is used by the model to
F-5
-------
CO
£
35
«
t5
D
DC
c
o
o
o
0)
M
F-6
-------
estimate the location and number of people exposed to the modeled
pollutants. The 1990 population has been aggregated into
6.9 million blocks.
F . 3 . 3 Mixing Height Database
The mixing height database is more limited in scope than the
other databases mentioned above. Only 73 sites were available
from the NWS for the years 1984-1989. Also, the mixing heights
are calculated from observations taken once daily. Forty of the
73 sites are based on 6 years of observations.
F . 3 . 4 Temperature Database
The temperature database provides an arithmetic average of
ambient temperatures for each atmospheric stability class for
each STAR site. Because the temperature was recorded for every
set of wind speed and direction observations in the NWS raw data,
the temperature database is similar to the meteorological
database; that is, each database has the same number of sites
(349), the same number of years of data to calculate the averages
at each site, and the same typical number of years (6) on which
averages are based. By default, the site closest to the plant is
selected for air dispersion calculations and is, for this
database, the nearest STAR site.
F.4 EXPOSURE ALGORITHMS
Exposure is calculated in HEM through pairing population
information from the census database with modeled ambient air
concentrations of each specific pollutant. The output of the
dispersion model is an air concentration array around the plant.
HEM calculates exposure by integrating the HAP air concentration
at the population center (centroid) of the census block through
interpolation of the air concentration values at the surrounding
modeled points. All persons residing in the census block are
treated as being exposed to the air concentration at the
centroid.
F . 4 .1 Air Concentration - Population Pairing
ISCLT2 calculates air concentrations at user-specified
receptors. For the utility study, receptors were placed around
the source along 16 radials, spaced every 22.5 degrees, at
distances of 0.2, 0.5, 1.0, 2.0, 5.0, 10.0, 20.0, 30.0, 40.0, and
50.0 km from the source, for a total of 160 receptors. Except
for receptors located very close to the stack, HEM calculates
exposure by interpolating the air concentration at the population
centroid (the population center of the census block) between the
values at the receptors surrounding the centroid. There is a
linear relationship between the logarithm of the concentrations
F-7
-------
and the logarithm of the radial distances. This linear
relationship is used to estimate the concentration along the
radial nearest the centroid at the same distance from the stack
as the centroid. The estimates are then interpolated linearly
between the radials of the receptors surrounding the population
centroid. Figure F-2 depicts the relationship between the
receptor locations and a hypothetical block population centroid.
F . 4 . 2 Exceptions for Population Close to Source
Within 0.5 km of the stack, the exposure is calculated
differently than described above because close to the stack, the
receptors are much closer together. Here, the population is
estimated at the points where the air concentration is
calculated, rather than the air concentrations' being estimated
at the known population point. This more complicated scheme is
described in detail in the HEM User's Manual.6
F.5 RISK CALCULATIONS
In general, long-term exposure estimates are paired with
chemical-specific health benchmarks, such as inhalation unit risk
estimates (lUREs), to calculate the risk to the population of
developing cancer or the potential for developing other adverse
health effects. Health benchmarks are input for each chemical
modeled. Health benchmarks and other toxicity information are
discussed in Health Effects Summaries for the Utility Study.1
Risk is calculated for the exposed population on a single-
pollutant basis. For carcinogens, HEM produces distributions of
exposure and risk, as well as estimates of annual incidence,
number of people exposed at various risk levels, and maximum
individual risk (MIR). A comparison of the modeled ambient air
concentration to the reference concentration is used to estimate
the extent of adverse health effects for noncarcinogens.
Aggregate risk associated with exposure to multiple pollutants is
evaluated by adding the risks from individual pollutants.
The utility boiler HEM modeling application requires the
input of chemical-specific toxicity information. HEM uses the
lUREs for carcinogens to estimate cancer risks or other adverse
health effects for each individual chemical according to that
chemical's particular level of toxicity. The more toxic a
chemical, the lower the ambient air concentration necessary to
produce high risk levels.
F. 5 .1 Reoruired Health Number Input
An IURE is entered in the risk calculation for each
carcinogenic pollutant. The IURE represents an estimate of the
increased cancer risk from a lifetime (70-year) exposure to a
F-8
-------
North
^\
\ 10
I
^
\
20
, /
i
30
/
i
40
50km
South
A Block Centroid
© Census Block
Figure F-2. The exposure algorithms interpolate between the estimated air concentrations and
the population data. Air concentrations are calculated at the points where the
circles and lines intersect. Population is known at the block centroid locations.
The concentration at the centroid is calculated based on the concentration
estimated at the 4 points surrounding the centroid.
F-9
-------
concentration of one unit of exposure. The IURE for inhalation
is normally expressed as risk per /ug/m3 of air contaminant.
Hazard quotients for noncarcinogens are calculated by
comparing the ambient air concentration of the pollutant with its
reference concentration (RfC). The RfC is an estimate (with
uncertainty spanning perhaps an order of magnitude) of the daily
exposure of the human population to the chemical by inhalation
(including sensitive subpopulations) that is likely to be without
deleterious effects during a lifetime.
F.5.2 Risk Calculations
HEM calculates carcinogenic risk using standard EPA risk
equations and assumptions. Maximum individual risk (MIR) is
defined as the increased probability of an individual to develop
cancer following exposure to a pollutant at the maximum modeled
long-term ambient concentration assuming a lifetime of exposure.
It is calculated by multiplying the estimated ambient air
concentration of a HAP by the IURE.
Unlike cancer risk characterization, noncancer risks
typically are not expressed as a probability of an individual
suffering an adverse effect. Instead, the estimated exposure
concentration is compared with a noncancer health benchmark such
as an RfC. This is usually expressed as a hazard quotient. the
hazard quotient is the ratio of the exposure (ambient air
concentration of the pollutant) to the RfC. The RfC represents
the highest protective concentration, and a ratio value greater
than or equal to one would represent an exposure that may be a
public health concern and should be evaluated further.
For additional information on the carcinogenic and
noncarcinogenic effects of HAPs, refer to Health Effects
Summaries for the Utility Study.8
F.6 ASSUMPTIONS
Simplifying assumptions are used in the HEM utility boiler
analysis to enable estimation of the potential health effects due
to HAP emissions from utility boilers. The following assumptions
are made from HEM:
1. Direct inhalation of pollutants is the only source of
exposure.
2. Average exposures are equivalent to those experienced
if one constantly stayed at home; no adjustment is made
F-10
-------
for exposure changes resulting from population movement
between home, school, work, etc.
3. Homes are located at population-weighted centers
(centroids) of census blocks (or at nodes of the polar
grid within 0.5 km) because the locations of actual
residences are not included in the database.
4. For the most exposed individuals, it is assumed that
people reside at the home for their entire lifetimes
(in modeling carcinogens, a lifetime is assumed to be
70 years).
5. Indoor concentrations are the same as outdoor
concentrations.
6. The plant emits pollutants at the same level for the
70-year lifetime of exposure.
7. No resuspension of pollutants via dust occurs.
8. There is no population migration or growth.
9. Varying exposures that might arise as a result of
differences in age, sex, health status, degree of
activity, etc. do not exist.
10. Because the model does not handle complex terrain, each
plant is located in flat terrain. An additional
complex terrain analysis was conducted using specially-
designed models.
11. The nearest meteorological location provides the most
appropriate STAR, temperature, and mixing height data
for the plant.
12. No pollutants are emitted from point sources other than
stacks.
F.7 HEM OUTPUT
For carcinogens, HEM produces estimates of annual incidence
(population risk), number of people exposed to various risk
levels, and maximum individual lifetime risk. For
noncarcinogens, HEM estimates the number of people exposed at •
various concentrations and the maximum individual concentration.
These values are for individual pollutants; no summing of risks
across chemicals is performed.
F-ll
-------
F.8 REFERENCES
1. U.S. EPA. Guideline on Air Quality Models (Revised).
United States Environmental Protection Agency, Office of Air
Quality Planning and Standards, Research Triangle Park, NC.
EPA-450/2-78-027R. 1993. p. 4-4.
2. Ref. 1, p. 4-4.
3. Ref. 1, p. 8-10.
4. U.S. EPA. User's Guide for the Industrial Source Complex
(ISC2) Dispersion Models, Volume 1 - User's Instructions.
United States Environmental Protection Agency, Office of Air
Quality Planning and Standards, Research Triangle Park, NC.
EPA-450/4-92-008a. 1992.
5. Ref. 4, pp. 3-58 to 3-60
6. U.S. EPA. User's Manual for the Human Exposure Model (HEM).
United States Environmental Protection Agency, Office of air
Quality Planning and Standards, Research Triangle Park, NC.
EPA-450/5-86-001. 1986. pp. 2-12 to 2-19.
7. Health Effects Summaries - Overview (Apendix E of this
report).
8. Ref. 7.
F-12
-------
Appendix 6 - Preliminary Uncertainty Analysis for the
Characterization of the Human Health Risks from Direct Inhalation
Exposures to Electric Utility HAP Emissions
-------
This page is intentionally blank.
-------
EXECUTIVE SUMMARY
GES.1 INTRODUCTION
Section 112(n)(1)(A) of the 1990 Clean Air Act Amendments
requires the U.S. Environmental Protection Agency (EPA) to report
on the hazards to the public health that could reasonably occur
as a result of emissions of hazardous air pollutants (HAPs) from
electric utility steam generating units (utilities).
As part of the study, EPA completed an assessment of the
risks to human health from direct inhalation of utility plants'
HAP emissions within 50 km of each plant for the year 1990. The
EPA conducted an uncertainty analysis on the estimates of risks
from direct inhalation of utility HAP emissions. The need for
uncertainty analysis as a part of any risk assessment and its aid
in conveying results of the risk assessment is widely accepted,
having been proposed in both EPA Risk Characterization Guidance1
and the NRC Committee Report: Science and Judgement in Risk
Assessment.2 Furthermore, the initial risk assessment relies on
a combination of point values - some conservative and some not
conservative. This method yields point estimates of exposure and
risk that fall at unknown percentiles of the full distributions
of exposure and risk. For this reason, the.EPA conducted an
uncertainty analysis to determine the range of possible values
and to estimate the degree of conservatism in the point estimates
calculated in the original risk assessment.
The original inhalation risk assessment is presented in
Chapters 4 to 6 of the interim report. However, a summary of the
results are provided below to provide the reader with the risk
estimates before the uncertainty analysis is presented.
GES.2 ESTIMATES OF HUMAN HEALTH RISKS FROM DIRECT INHALATION OF
HAPS EMITTED BY ELECTRIC UTILITIES
The inventory of sources modeled included all coal-, oil-
and gas-fired plants that have at least one boiler with a
capacity of 25 MW or more: 426 coal-fired plants, 137 oil-fired
plants, and 267 gas-fired plants. Risks were estimated for 12
carcinogens and four noncarcinogens.
GES.3 CARCINOGENS
Direct inhalation exposures to emissions from 24 (22
oil-fired and 2 coal-fired) plants are estimated to result in
individual risks of 1 in 1,000,000 (10~6 or greater). Emissions
from oil- and coal-fired facilities are the major contributors to
G-l
-------
these risks from direct inhalation, while all gas-fired utilities
have been shown to present lower risks. The total annual cancer
incidence estimated to result from direct inhalation exposure to
HAP emissions within 50 km of plants was estimated as 0.56
cases/year. HAP emissions from oil-fired utilities accounted for
0.47 cases/yr or about 84 percent of the total incidence. Table
G-l summarizes the population exposed within 50 km of individual
facilities at different levels of individual risk. The highest
MEI across all utilities was 1 xlO"4 for the mixture of HAPs.
GES.4 NONCARCINOGENS
The highest HQ across all noncarcinogen HAPs across all
plants was 0.12 from hydrogen chloride and 0.046 from manganese
exposures from a coal-fired plant. All HQ values were less than
one. These risk estimates are for inhalation exposure within
50 km alone and do not include risks associated with long-range
transport or indirect exposures (e.g., ingestion of contaminated
foods). The risks from indirect exposure may potentially be
significant for those compounds which are environmentally
persistent and have the tendency to bioaccumulate.
GES.5 DISCUSSION
The methods used up to this stage of the risk assessment
incorporate a few generally conservative assumptions (i.e., more
likely to overestimate rather than underestimate risk) to address
some of the inherent uncertainties. The risk estimates presented
in this section are intended to represent conservative estimates
of risks due to inhalation exposure within 50 km of each
facility. Table G-l summarizes the risk estimates across all
utilities modeled.
More information on the assumptions, models, methods, data,
and uncertainties associated with the exposure and risk
assessment is provided below. An uncertainty analysis was
conducted and the results are used to determine the extent of
conservatism in the risk analysis. The uncertainty analysis also
helps identify errors of either overestimation or
underestimation. Methods, models, data, and assumptions used in
the analysis are identified, including their rationale and the
effect of reasonable alternative assumptions on the conclusions
and estimates. The uncertainty analysis focused on the three
HAPs (nickel, arsenic, and chromium) which accounted for over
95 percent of total incidence due to inhalation exposure to HAPs.
Since these HAPs accounted for most of the inhalation risk, an
analysis of uncertainty on these three is believed to account for
G-2
-------
O
0
H
I
c
CD
t»
§
"
g
z
0}
D)
rcino
CO
X
cf
J.
0
A
i
CA
C
Q.
**
o
A
oc
^
0
CD
a
o
Q.
LU
CA
CO
a
o
"co
o
u
CA
to
6
"5
0
u
CA
CD
(D
0
§
o
en
CD
(3
O
"5
o
O
CA
§ £
1 1
N =
CD .!= ^5
I < Q.
Z
Z
z
o
CM
CM
O
\S
CM
.j
•4-
CM
LU
CM
in
LU
CO
UJ
CO
H Arsenic
Z
z
z
o
o
o
o
0
o
1
LU
00
^
LU
CO
I Beryllium
z
z
z
o
.-
o
o
in
o
CD
UJ
CM
^
UJ
CM
| Cadmium
Z
z
z
o
»—
o
\y
CO
CM
o
r—
i
CO
UJ
in
CO
UJ
CN
| Chromium
Z
z
z
o
o
o
o
o
o
1
^
UJ
00
LU
in
c
'x
o
o
'
in
o
o
0
CM
d
z
z
z
0
o
z
z
z
1 Hydrogen
Chloride
LU
s
o
o
I—
o
o
C)
z
z
z
o
o
z
z
z
•D
CO
•
CO
o
0
co
s
d
z
z
z
o
o
z
z
z
CD
nManganes
1
I
•
z
z
z
o
o
o
z
z
z
H Mercury
II
Z
^
z
z
o
o
CM
O
o
^5
f-.
o
^
LU
CM
in
UJ
^
UJ
1
Z
Z
<
z
z
o
1
o
1
1
J
o
1
^
LU
00
CD
In-Nitrosod
methylami
Z
^
z
z
o
CN
CN
CM
Z
z
^
LU
CO
t
UJ
CO
UJ
"5
B
.— w
I 1
i= c
c _,; CD
5 o fi
0 « C T5
X 9> £ ffl
1 c -S '£
1 =8
> . " o
i SS-S-a
;j? I = S
«| aj?
! -o a- E .2
; fl? o O *~
. £ 'I ° o
1 E S, s-g
! i 5 1 S
I O a > 3
: 2 "S "S .«
• O 2 CD -O
*^ ^_j V *_
•fc co
III
•5 to ~
o5 8
G-3
-------
a majority of the uncertainty in the overall risk estimates due
to inhalation exposure.
GES.6 APPROACH TO UNCERTAINTY ANALYSIS
Uncertainty has been classified into four types (parameter
uncertainty, model uncertainty, decision-rule uncertainty, and
variability).3 The first two, parameter and model uncertainty,
are generally recognized as major sources of uncertainty.
Parameter uncertainty occurs through measurement errors, random
errors, or systematic errors when variables cannot be measured
precisely either because of equipment limitations or because the
quantity being measured varies spatially or temporally. Model
uncertainty can result from surrogate variables, excluded
variables, abnormal conditions, or incorrect model form.
Decision-rule uncertainty arises out of the need to balance
different social concerns when determining an acceptable level of
risk which can affect the choice of model, data, or assumptions.
Variability is often used interchangeably with the term
"uncertainty," but this is not strictly correct. Variability is
the unchanging and underlying distribution of a parameter based
on physical, chemical, and/or biological processes (e.g., body
weight within a population). Even if variability is known
(hence, not in itself uncertain) it still contributes to overall
uncertainty of the risk assessment.
This uncertainty analysis focused on parameter uncertainty
within the models and data available for the local inhalation
risk assessment only. Long-range transport and non-inhalation
exposures and risks were not assessed in this uncertainty
analysis. Table G-2 briefly summarizes information regarding the
parameters used in the inhalation risk estimation process. Model
uncertainty was not addressed. However, these uncertainties are
described qualitatively. Variability has been evaluated
separately for exposure-response, but is included in the overall
estimate of uncertainty related to emissions and exposure. The
goal of this uncertainty analysis is to estimate the range of
possible risk estimates due to inhalation exposure within 50 km
of plants considering the parameter uncertainty and variability.
It should also be noted that there are other sources of
uncertainty, some of which may be significant, which could not be
evaluated quantitatively. These uncertainties are qualitatively
discussed.
The approach used in this analysis was to identify the
uncertainty with each of the parameters used in the risk
estimation process. First, the uncertainty associated with each
G-4
-------
o
(V
i—i
X!
§
3
I
§•
ll
m =
§
.
± =
§•§
8 _
°
^ 2 S ^? -cf
• O • J^CQ
liflil
ii
§00-
-llfl
«_-«« =
Q. • §
•a o
5S
& I
5 «-2
£ %-S:
£ S£ .
f!
la
8 E
en i
I!
11
I
»p
III
h
s«
-
tl
l\
• 2 i
•5 f'
Si
ex
«:
11; g E
io w S
ii
•si
G-5
-------
II
II
II
•o
ID
3
•H
4J
0
o
CN
I
O
It
-I
a.
•o
m 5.-5
11
•S-g5! |
"5 2 *o ^ 2
CO * —
I
S*
5 « £
iiil
^ i1 *n
ill
I! I
Hi!
lil§
"5 O) O) O
S!?l
. S
N g
•a
S.
p_ -
« CD
* i
c •
Ilil*
li>?ll
? » O o "C
O CO — O 0
_l CO CD « >
-------
Q)
C
•H
4J
c
O
U
CN
I
O
0)
s
E i
os SSs E
ooo „ g-o
li! fit
—^ 5 S * m tt C
. ?• S ^ ? ^ a
H1
III
II
ill!
£ i -s 5
lilt
lim'i'-i
_
o o •= «.
1=11*
ll
ril
Scl-
fl:
^!
liii
£ 8 | e. |
f * 1 § °r
| 0= ~
11-
S
ft
« a
o g S|5
iSS^I §
lilt
I Hi
8 § £
S £
II
11
•i
i
a.
I c
I
f
If
il
Ii
G-7
go, £ 2
!•= tS
Hill
ril
i
i.
§
1 w
ll
is
s «
Si
I «f 1
lilli
° t
1s-
-
H
O O
if
is
ts
l-s
;-°
if
-------
of these variables was described using an appropriate statistic
(e.g., mean and standard error of means) or as a probability
density function (the relative probability for discrete parameter
values). The standard error of the mean (SE) for each parameter
was used as the estimate of uncertainty and variability rather
than the standard deviation for each parameter. The SE is
considered a more appropriate statistic because the available
dose-response data are based on long-term average exposures.
However, it should be noted that using the SE from a sample may
be an overconfident estimate (i.e., too narrow a range) of
uncertainty. The SE of a normal distribution underestimates the
range of uncertainty for small sample sizes. In this case, the
t-distribution is the more appropriate distribution. Also the
sample population must be representative of the total population
and the underlying variability in the population. To address
these concerns, adjustments were made to the observed SE to
expand the estimate of uncertainty.
In general, numerical methods (i.e., Monte Carlo simulation)
were then used to develop a composite uncertainty distribution by
combining the individual distributions. In Monte Carlo
simulations, the risk and/or model equations are repeatedly
solved using randomly sampled values from the specified
distributions to calculate a distribution of estimated risk
values. These risk distributions were derived for both estimates
of MIR and population risks and can be expressed as a probability
density function or cumulative probability density function. The
distribution allows the risk assessor to choose the value
corresponding to the appropriate percentile in the overall
distribution. For example, an exposure level or risk level can
be selected which corresponds to the 95th percentile of the
overall risk distribution. Because variability was not
specifically differentiated in the analysis of emissions and
dispersion modeling, uncertainty and variability were simulated
together in a one-dimensional Monte Carlo simulation. This
yields a hybrid distribution which applies to single "typical"
individual from the exposed population.
The uncertainty analysis was conducted on the three major
components of the risk assessment process, emissions
characterization, dispersion and exposure modeling, and
exposure-response assessment. Each of these is summarized
briefly below. Figure G-l provides an example of how the
uncertainty from each of these components is combined into an
overall distribution. A detailed uncertainty analysis could not
be conducted on all of the utility plants. Therefore, a total of
four plants (2 oil-fired and 2 coal-fired plants) were selected
which appear to be 4 of the highest risk plants, based on both
G-8
-------
LU
CO
G-9
-------
the estimated incidence and the maximum individual risk. Each of
these plants was analyzed for arsenic, nickel, and chromium.
Plant #29 is used for illustration purposes and was selected
because it is the oil-fired plant contributing most to cancer
incidence.
GES.7 EMISSIONS AND EXPOSURE CHARACTERIZATION
An emissions factor program was developed by EPA to estimate
plant-specific emissions rates based on fuel type, fuel origin,
plant configurations, and emissions testing results. The
emission factor program and the data used are described in
Chapter 3. This program is based on a mass-balance
concept—reducing concentrations in the fuel due to the impact of
the boiler and control devices. This model was designed to use
estimates of emissions from individual plants to derive a
nationwide estimate of emissions. The accuracy of the model in
predicting emissions for individual plants may vary
significantly.
Parameters used in the emissions characterization were: fuel
consumption (Coal: tons/yr, Oil: barrels/yr), HAP (trace element)
concentration in fuel, coal cleaning factor (if needed),
emissions modification factors for the boiler (EMFb,
boiler-specific factor to account for the amount of HAP entering
the boiler to that exiting the boiler) and the air pollution
control device (APCD), if present, and EMFa. (APCD-specific
factor to account for the amount of HAP entering the APCD to that
exiting the APCD)
GES.7.1 Plant-Specific Emission Rates
Monte Carlo simulation was used to develop a distribution of
possible plant-specific emissions rates. Simulations were
carried out randomly sampling values for fuel consumption, HAP
concentration, and EMFs. For illustration purposes, Table G-3
and Figure G-2 present the summary statistics and graphical
representation, respectively, of the emissions forecast
predicted for Plant #29. This distribution gives some indication
of the degree of uncertainty and the possible range of emissions
estimates which may be experienced. The original emissions
estimation program had been designed to be an unbiased estimator
of emissions. The results of the uncertainty analysis tend to
support this assertion. The original baseline estimates ranged
from the 22nd percentile of the overall distribution (for arsenic
using the SGS oil concentration data) to the 95th percentile (for
chromium using the SGS oil concentration data). This broad range
is attributed to the re-analysis of the oil concentration data in
how non-detects were evaluated (using the probability plotting
G-10
-------
-S|SfeSSgggS
^^^CMcoincn^^ctjg
o ,—^
(<•» en
5 *£•
SO GO T* 00 ^ 2\ rS i—
ocoo,no||5
co •<*• o
T>
_« CSJ
11
-2 Jl
CO 0-
O -0
0) £
° E
« 2
«-•* H—
u ^
t^^"'
cB g
E o
E «
CO
. HI
^CL
O <
_0) I
JD
CO
s
03
CO
£
tr\ •+• t*"" m T— o>
^ ^ CNJ ^ oo w
Ki rn '•^ ^J *•J
81 S^$E^
•^ oo $::
S8S
^J- Cn T-
r*- CM in
•»- CM in
o
co
en in
co en
CM cp
6 d
cMincncqcqi^oco-i-1^00
dddT^coocicM^f-^Ico
0 ^
o
LL
•o
0>
C
in
go
0
CO
§
1
I
^£88S!g§
o , , o q oo g g
o ' ' CM co en o g
g
15
CO
2! oo
CM 0
in b
in ^:
"* CM
§^
S
CO CO 00 00 S
T- CO h- O5 ^
en oo h-
co in i- h- en
T- CM CO
00
o ^f oo co en T- ^
CM
ooc3T-cocncoe»£:$J£j
co
CO
LLJ
•e oT w
•i •= ^B
£
_ *
«= CD
"c Q.
1 8
P Q
CO CL
Q. 5-
O
3
-Q
•
W
'
I I
^3 CZ (0
JO CO T>
co
c|
|¥
w °2
o
W
255SS|!1§ I
03
O
Q.
CO
= CD ^8 -5
o fn co ••"-
i CO co oj 'I
r» ^^ ^ *±2 —
03 Q)
•D
1
E " 8.
CO
8
O HI ® fi r-n
^ ^ g LL .E
U 1 I O
= -^ CD c Q-
R "6 -° CO
„ -^ cp £ t-
O Q. 5 c g
EC
Q.
•I 1
.t; . ^
C O CO
i §-§
S s S 8
C CX 3 C
O « « 8
C T C
•»- =. Q CO
rn CT .:= m
O
O
5
(0
O "CO
E - w
w 3 "5 CO {0
jo « co * °-
re c co
•2 S E « E 5 «
-2 TO g g ~ "g -5
C— *~ W r ^ *^ m —
.U >, •= T3
|f -S g
Ji-g |5 re
^ o §
2 2 o-
«
« i
£ CO
y
CO
CL
u.
2
3 -0 C
co ® '6
M -IB;
CO E .2
85-i
G-ll
-------
Figure G-2. Summary of Results to Monte Carlo Simulation of
HAP Emissions from Oil-Fired Plant#29
FCEM Concentration Data
SGS Concentration Data
CellElt
.336
Forecast: Plant 29 As Emissions
Frequency Chart
2,942 Trials Shown
100.00 200.00 300.00
kg/yr
Cell E12
.027 1
Forecast: PI 29 As Emission (SGS)
Frequency Chart 2,941 Trials Shown
Cell H11
.036
Forecast: Plant 29 Cr Emissions
Frequency Chart
2,937 Trials Shown
10:
Forecast PI 29 Cr EMissions (SGS)
Frequency Chart 2,444 Trials Shown
CellK11
.036 ->
Forecast: Platn 29 Nl Emissions
Frequency Chart
2,930 Trials Shown
106
CellK12
.030 -
Forecast: PI 29 Nl Emissions (SGS)
Frequency Chart 2,940 Trials Shown
G-12
-------
routine versus treating them as half the detection limit). When
the combined data set (SGS and FCEM oil concentration data) is
considered the original baseline emissions estimates were
correspond to the 58th percentile for arsenic, 68th percentile
for nickel, and the 80th percentile for chromium. Furthermore,
the 95th percentile of the simulated range of long-term average
emissions estimates are within a factor of two or three of the
mean and original baseline estimate and within a factor of less
than six within the median.
To compare how estimates differ for the coal-fired plants,
Plant #343 was selected for illustration purposes. Plant #343
was the coal-fired plant that contributed most to cancer
incidence. The original point estimate of emissions ranged from
the 81st to 96th percentile of the range of emissions predicted
under the uncertainty analysis. This is significantly higher and
may be attributed to the fact that in the original analysis coal
from a single state was assumed (defined by the state whose coal
was consumed most). However, for this plant the other coals used
were generally of lower HAP concentrations which would tend to
drive the emissions estimate lower than estimated during the
baseline assessment. The 95th percentile on the simulated range
of emissions ranged from 0.9 times the baseline estimate for
nickel to 2.5 times the baseline estimate for arsenic. As with
Plant 29 the 95th percentile of the overall range of predicted
emissions typically was within a factor of two or three of the
mean estimate and the original baseline estimate.
A preliminary analysis of the emissions estimation model was
also conducted. Comparisons were made of test data from 19
utility boiler stacks (17 coal-fired, 2 oil-fired) against
predicted emissions for the same plants using the emissions
model. For each facility, the emission estimate from the model
was divided by the reported value from the corresponding test
report. A value of 1 meant that the model exactly predicted the
test results, values lower than 1 indicated the model
under-predicted emissions, while values higher than 1 indicated
the model over-estimated emissions. In general, the results
suggested that the model performs as expected (i.e., across a
range of boilers and constituents to estimate overall emissions).
The average of the ratios across all stacks and constituents was
1.0, while averages for the three HAPs are 1.4, 0.7, and 0.9 for
arsenic, chromium, and nickel, respectively.
However, while the model did well in predicting overall or
average emissions across a range of utility boilers, large
differences between predicted and reported values are found for
individual boilers and constituents. The largest difference for
G-13
-------
an individual boiler estimated emissions was about 5,000 times
lower than reported test results. Furthermore, the model tended
to underestimate rather than overestimate emissions about 60
percent of the time within this sample of boilers. A preliminary
evaluation of facilities with large differences between projected
and actual emissions found that these facilities were likely to
burn multiple fuel types. Petroleum coke was found to be burned
at the facility with the greatest difference. Therefore, while
the model appears to predict fairly well, emissions may be over-
or under-estimated at individual facilities. This difference may
be due to the combinations of fuel being burned since the model
assumes a single fuel type. Furthermore, the plants in which the
model underestimated significantly were some of the lowest
emitting plants predicted by the model. Using measured rather
than modeled emissions for those plants would bring their
emissions more in line with the rest of the plants, but they
would not become high-risk plants when compared to others.
However, it should be noted that the emissions test data used in
comparing modeled to measured emissions were based on short-term
(and sometimes non-simultaneous) sampling. Whether these data
are indicative of long-term emissions from the same stack are not
known.
GES.7.2 Dispersion and Exposure
Air dispersion modeling is complex and nonlinear, cannot be
carried out with the use of spreadsheets, and requires
significant time to conduct the modeling and process the data for
each run. In order to better estimate percentile values above
90 percent, a stochastic (Monte Carlo) approach requires large
number (thousands) of repetitive runs (3,000 was used for the
emissions estimates) needed to generate a distribution. Given
the time and resources required for single runs, the Monte Carlo
approach was not feasible and an alternative approach was needed
to evaluate the uncertainty in dispersion and exposure modeling.
The degree of dispersion and resulting exposure is affected
by three major parameters: plant stack parameters (e.g., stack
height, stack gas temperature, and exit velocity), meteorologic
conditions, and surface roughness (urban versus rural). The
uncertainty analysis therefore, focused on the three parameters.
The three factors being evaluated are non-linear with respect to
each other and, require a separate HEM run for each parameter
value. Therefore, a test matrix approach was used to evaluate
uncertainty in the exposure modeling component of the exposure
assessment. A limited number of options were developed to
represent the expected range of uncertainty for each of these
three categories of parameters as follows:
G-14
-------
Surface roughness: urban or rural mode
Stack parameters: represented as high (1.1 x UDI values),
medium (UDI values) and low (0.9 x UDI
values) estimates for stack gas temperature
and flue gas exit velocity.
Meteorology: three closest meteorology locations in the
STAR database.
As a result, for each plant a total of 18 different HEM runs were
made covering each combination of dispersion parameters. For the
purposes of this uncertainty analysis it was assumed that there
is insufficient information to determine the relative correctness
of each combination and, therefore, each was considered equally
likely to represent the possible range of values. The
coefficients for estimating maximum concentration and total
exposure (per 1 kg/yr emission) resulting from each of these
18 HEM runs was summarized for each plant.
GES.8 EXPOSURE-RESPONSE ASSESSMENT
The variability of the quantitative relationship between
exposure and the excess probability of cancer for different
humans and the uncertainty in the mean (taken here also to be the
"best estimate" or "maximum likelihood estimate") quantitative
relationship between exposure and the excess probability of
cancer, were all addressed. As with the uncertainty analysis for
emissions and exposure estimates, efforts were limited to the
three HAPs-- arsenic, chromium, and nickel. Specific parameters,
for which uncertainty about the mean value (or best estimate for
a given parameter within the exposed population) was addressed,
include exposure frequency, exposure duration, breathing rate,
deposition fractions, and retention half-times. Uncertainty
related to the cancer slope factor (CSF) focused on data, and the
use of epidemiologic data (typically from workers) extrapolated
to the general population.
The study of variability focused on how parameter values
would be expected to vary among individuals within the general
population and how that would affect the risk estimations. The
parameters for which some measure of variability among
individuals within the general population was addressed include
exposure duration, exposure frequency, breathing rate, deposition
rate, and retention times in the lung. No specific measures of
variability were available for how the CSF for these three HAPs
may differ among individuals. However, limited data indicate
that the CSF differs between smokers and non-smokers and was
G-15
-------
incorporated in the analysis. Figure G-3 displays the
distribution of uncertainty in the CSF for estimating both MIR
and annual cancer incidence.
GES.9 RISK CHARACTERIZATION
The distribution of estimates of MIR and incidence for Plant
29 are presented in Table G-4. The results indicate that the
baseline risk assessment resulted in conservative estimates of
risk. It should be noted that the estimate of MIR is influenced
by which of the oil concentration data sets (FCEM or SGS) are
used. The original estimates of MIR for Plant 29 generated
during the baseline risk assessment ranges from the 71st
percentile for arsenic (using the SGS data) to the 98th
percentile for chromium (also using the SGS data). Similar
results were obtained for the estimates of annual cancer
incidence.
For comparison, analysis of MIR and annual cancer incidence
was also conducted for Plants 343, 240, and 133. In general, the
results support the conclusions that the baseline assessment
resulted in reasonably conservative estimates of risk while still
being within the range of plausible values. For the coal-fired
plants (343 and 240), the point estimates of MIR generated during
the baseline assessment were associated with the 91st to 94th
percentile on the overall distribution of possible MIR risk
estimates. The 95th percentile (a typical high-end risk estimate)
of the overall distribution is roughly twice the original risk
estimate, four times the mean and within an order of magnitude to
the median MIR risk estimate. For oil-fired Plant 133 the
results were similar, though the uncertainty was a little more
broad. The original estimates of MIR from the baseline
assessment were between the 86th and 99th percentile on the
estimated overall distribution of possible MIR values.
GES.10 DISCUSSION OF RESULTS
The risk estimation process used in the baseline assessment
utilized a combination of parameters each with varying degrees of
conservatism (the degree of over-estimation, or under-
estimation) . In general, the estimates of maximum individual
risk (MIR) and annual cancer incidence derived in the initial
risk assessment were conservative, generally around the 95th
percentile on the distribution. The 95th percentile is roughly
10 times the median, and about 5 times the mean. The sensitivity
analysis indicated that the dispersion coefficient (surface
roughness) was the most significant parameter for estimating
uncertainty MIR and incidence, followed by the EMFs. The
G-16
-------
0>
o
O)
0>
P
(0
o
o
8
I
o
T3
CO
CO
4?
CO
LU
CD
W
CO
o
CO
o
"•3
.0
*
CD
£
D
O)
G-17
-------
•—
to
CO
Q «
c -
-2 i
co Z >,
j= .,5^
c 8 = |
CD c a .2
ri «S|
0 1 =^
O i 5 °
co -5
03
^
rn c
UJ c
^ <
co
2
1 1 1
LLJ ol
O 'E
n j>
-, CO
O co
O
.CO
CO
Q.
E
o
0
CM ]|
o
£2 Z
JO
^
CD
O
C
CD
•n
\j
j| 3
_ 'E
•3 2
C .C
C en O
< 8 •£
"c? C S
I y
M
c •?
.2 g
** 0
= 1
.g <
!s2
Q"^~
•
i
O
CD
en
!_.
« S ~
1
UJ
CO
21 g CJ-
Lyj • 2>.
u_| CM
*t CO
o
LU
CO
Lul 9 &
O UJ *~*
ill ^—
ol 9 £
w ^ 8~
LLJ
CO
I 1 f
8 1 »g
o
LLJ
^
51 § S"
ol LU N"-'
Q | CM
3 9 f
COl LU ^
"* t»-
o
1
LU
CM
Lul 9 fe
O UJ *~~
II [ T—
COl t ^
d 9 £
COl LU *-
r***
o
S. i.
uu
CD
§o »"
9 c».
^
£
CO
c
'co
LU
I 2"
£1
S 3 1
1 «!
§
s
LLJ
•^
O
LLJ
CO
CD
O
LU
^
s
UJ
m
9
UJ
9
UJ
co
a>
o
i
UJ
CM
^
LLJ
"*"
0
CM
9
LLJ
CM
1
LU
CM
S?
O
d
S
1
§
o
a.
9
UJ
9
LU
*~
S
g
in
o
LU
S
LLJ
CO
in
o
LU
CM
9
LJJ
CO
O
1
LU
CM
O
T-
LLJ
N
9
LLJ
CM
O>
O
LLJ
CO
?
UJ
CO
gS
in
c\i
CO CO
9 9
LLJ LU
CM "*
88
LU UJ
CM CM
in in
9 9
LLJ LU
^- CM
in •*
9 9
LLJ UJ
?s
UJ UJ
co o>
in in
9 9
LLJ LU
•* CO
Sfe
LJJ LU
CO CO
o o
1 1
LU LU
m a>
en en
o o
LU UJ
i- CO
g§
LU LLJ
co in
CO CO
9 9
LLJ LLJ
CM CO
§§
UJ UJ
CD T-
sS sS
0 0
in *"
CO
o
g
?
LU
in
in
o
LU
in
o
LJJ
CO
8
LLJ
CM
LU
CM
LJJ
CO
0
LU
CM
O>
0
UJ
^
9
LU
CO
9
UJ
8
LLJ
CO
>s
in
CM
ss
LJJ LLJ
CM f
ss
LU LLJ
••- CM
ss
LLJ LLJ
i- CO
99
LU LLJ
CD t-
CO CO
9 9
LU UJ
<*• co
9 9
LU LLJ
•* CO
99
LLJ LLJ
O> CO
feg
1 1
LU UJ
CO CM
CO CO
O O
UJ LU
*- CM
99
LLJ LU
9 9
UJ LU
CM h-
88
UJ UJ
CD 00
>O xP
o So
in f-
So
LJJ LU
ss
UJ LU
TC CO
99
LU LLJ
C» i-
88
LLJ LLJ
CO •«}•
CM CM
0 0
LJJ UJ
CM CM
CO CO
O O
LLJ LU
CM CO
co in
9 9
LJJ LLJ
CO T-
gg
1 1
LU LU
8£
LU LU
CO i-
O O
I |
UJ LU
co in
CO CO
0 0
LU LU
i- CM
99
UJ LU
CM in
vP xP
0s* O**
o q
Q]
I
9
LU
O5
UJ
CO
9
LU
CD
9
LLJ
CO
CM
O
LLJ
?
UJ
CM
ID
O
i
LU
l»-
O
LU
CM
O
LU
LLJ
g
LU
g5
in
i^;
en
o
LU
c»
o
LU
^
di
co
CM
O
LJJ
co
i
UJ
CM
9
UJ
CM
in
o
LLJ
in
o
i
UJ
05
CO
9
UJ
CO
CD
9
LU
CD
LU
CM
s
UJ
CO
3?
o
d
o
cq
o
in
d
^.
•
o
CO
00
d
in
CM
.
T~
d
co
CNJ
'"I
"*
CO
d
ca
CD
f£
ffi
a>
CO CO
in co'
CO CO
in co
o co
$2 co
T- CO
N- CO
co t
in co
. o
* °°
o co
^ f~\
• l~J
1— ^
CD Cft
oS co
<°. CM
CO rf*
O
" CO
r>- in
CO CO
a>
c c
|1
E J3
£ £
in &
o> o>
•Q. S
* 8 1
JS «
o
=5 c Z
II _ ~
G-18
-------
deposition fraction, retention time, and exposure frequency also
contributed significantly in the variability of these estimates.
These results indicate that the EPA baseline risk estimates
are generally conservative estimates of risks due to inhalation
esposure to HAP emissions within 50 km of plants. However, it is
important to note again that this uncertainty analysis did not
include an evaluation of uncertainty associated with long-range
transport or multipathway exposures. This conservatism is
usually appropriate given EPA's mandate of public health
protection. Often there is a concern that the use of several
conservative assumptions results in risk estimates which are
unrealistic and beyond the range of possible risks (i.e., overly
conservative). The results of the uncertainty analysis indicate
that the baseline risk estimates are reasonably conservative
(generally around the 90th or 95th percentile). The uncertainty
analysis supports the baseline risk estimates as they are within
the range of predicted risks.
However, it should be noted that this analysis has focused
only on parameter uncertainty. As a result, the uncertainty
presented here may underestimate the overall uncertainty. The
uncertainty within those parameters evaluated may also have been
constrained by the limits of the models. For example, the effect
of surface roughness on dispersion is addressed by the choice of
the rural or urban setting within the HEM model. However, the
binary choice of urban versus rural does not fully encompass the
range of surface roughness. The range of surface roughness found
in the environment is broader than the range represented by the
urban and rural default settings within the model. Likewise, the
analysis of uncertainty in the exposure-response relationship did
not consider the uncertainty related to alternative
dose-extrapolation models (e.g., Weibull versus linearized
multi-stage) which have been shown to differ by several orders of
magnitude.
G-19
-------
G.1 INTRODUCTION
This appendix presents the uncertainty analysis EPA has
conducted on the estimates for human health risks from direct
inhalation of electric utility HAP emissions within 50 km of
plants. The general principles of uncertainty analysis are
presented in Section G.I.I. Section G.2 of the report summarizes
the risk estimates from the EPA study of direct inhalation of HAP
emissions. Sections G.3 through G.5 provide the detailed
uncertainty analysis for the exposure assessment,
exposure-response (dose-response) assessment, and risk
characterization portions of the assessment, respectively.
Sections G.3 through G.5 also describe and evaluate the models,
data, and methods used to conduct the analysis and their
attendant uncertainties. Where possible, quantitative estimates
of uncertainty have been generated. In the absence of
quantitative information, a qualitative description is given.
G.I.I The Need For Uncertainty Analysis
Unlike the other components of risk assessment, risk
characterization did not have its own specific guidelines in the
original 1986 EPA risk assessment guidelines. The general
content of risk characterization was defined by the National
Academy of Science4 and, to a limited degree, in each EPA Risk
Assessment Guideline.5 However, much was left to the
professional judgment of those involved in risk assessment
preparation. As a result, a high degree of variability in how
risk characterization has been practiced has developed.1
Although a great deal of careful analysis and scientific judgment
goes into the development of EPA risk assessments, significant
information is often omitted as the results of the risk
assessment are passed along to the decision-makers. Often, when
risk information is presented to the ultimate decision maker and
to the public, the results have been reduced to a single
numerical risk value. Such an approach may not often provide
sufficient information to the decision maker.
G.I.1.1 RAC Guidance on Risk Assessment. To address these
concerns, the EPA's Risk Assessment Council (RAC), composed of
risk assessors and risk managers from within the Agency,
evaluated EPA risk assessment practices. The RAC, after careful
evaluation, recommended guidance on risk assessment focusing on
the risk assessment-risk management interface, risk
characterization, and exposure and risk descriptors. The major
elements of the guidance are summarized below.
The RAC reached several conclusions in their evaluation of
EPA risk assessment practices including:
G-20
-------
• A full and complete presentation of risk is needed,
including a statement of confidence about data and methods
used to develop the assessment.
• A basis for greater consistency and comparability should be
developed.
• Professional judgment plays a necessary and important role
in the overall statement of risk.
As practiced at EPA, the risk assessment process depends on
many different kinds of scientific data (e.g., exposure,
toxicity, epidemiologic), all of which are used to "characterize"
the expected risk to human health or the environment. Informed
use of reliable scientific data from many different sources is a
central feature of the risk assessment process.
Two elements are required for full characterization of risk.
First, the characterization must address qualitative and
quantitative features of the assessment. That is, along with
quantitative estimates of risk, full risk characterization must
clearly identify all assumptions, their rationale, and the effect
of reasonable alternative assumptions on the conclusions and
estimates. Second, it must identify any important uncertainties
in the assessment as part of a discussion on confidence in the
assessment. This statement on the confidence of the assessment
must identify all major uncertainties and comment on their
influence in the assessment. The uncertainty statement is
important for several reasons:
• Information from various sources carries different kinds of
uncertainty, and knowledge of these differences is important
when uncertainties are combined to characterize risk.
• Decisions must be made about expending resources to acquire
additional information to reduce uncertainties.
• A clear and explicit statement of the implications and
limitations of a risk assessment requires a clear and
explicit statement of related uncertainties.
• Uncertainty analysis gives the decision-maker a better
understanding of the implications and limitations of the
assessments.
G.I.1.2 NRC Committee Report: Science and Judgement in Risk
Assessment. Section 112(o) of the Act, required the EPA to
G-21
-------
commission a study by the National Research Council2 on risk
assessment methods used by the EPA as related to HAPs.
Recommendations developed by the NRC relevant for this
uncertainty analysis include the following:
• Conduct formal uncertainty analyses, to show where
additional research might resolve major uncertainties.
• Consider the limits of scientific knowledge, the remaining
uncertainty, and the desire to identify errors of either
overestimation or underestimation of risk.
• Consider uncertainties when ranking risks, rather than
ranking risk based solely on the point estimate.
G.I.2 Background Information on Uncertainty
Uncertainty can be introduced into a risk assessment at
every step in the process. It occurs because risk assessment is
a complex process, requiring the integration of:
• The fate and transport of pollutants in a variable
environment by processes that are often poorly understood or
too complex to quantify accurately
• The potential for adverse health effects in humans as
extrapolated from animal bioassays
• The probability of adverse effects in a human population
that is highly variable genetically, in age, in activity
level, and in life style.
Even using the most accurate data with the most sophisticated
models, uncertainty is inherent in the process.
Finkel6 classified all uncertainty into four types
(parameter uncertainty, model uncertainty, decision-rule
uncertainty, and variability), summarized in Table G-5. The
first two, parameter uncertainty and model uncertainty, are
generally recognized as major sources of uncertainty.
Parameter uncertainty occurs when parameters appearing in
equations cannot be measured precisely and/or accurately either
because of equipment limitations or because the quantity being
measured varies spatially or temporally. Random, or sample
errors, are a common source of parameter uncertainty that is
especially critical for small sample sizes. More difficult to
G-22
-------
TABLE G-5. SOURCES OF UNCERTAINTY IN RISK ASSESSMENT"
General Type
Parameter
uncertainty
Model
uncertainty
Decision-rule
uncertainty
Variability
Specific Source of
Uncertainty
Measurement errors
Random errors
Systematic errors
Surrogate variables
Excluded variables
Abnormal conditions
Incorrect model form
Comments/Examples
Include limitations of equipment, methodology, and human error
Some processes impossible to measure exactly
Sampling errors
Can be minimized by increasing sample size
Nonrandom errors
Result of inherent flaw in data gathering processes
Minimize by external peer review
Use of animal bioassays to determine effect on humans, for example
May result from model simplification or failure to recognize an important
variable
Failure to recognize importance of episodic meteorological events, for
example
Choice of dose-response model for carcinogens, for example
More important for risk management, but need to recognize that value
judgments affect choice of model and interpretation of results
Those important for health risk assessment include environmental factors,
genetic variability, and lifestyle differences
Even if variability is known (therefore, not in itself uncertain) it still
contributes to overall uncertainty of the risk assessment
* Adapted from Finkel, 1990.
recognize are nonrandom or systematic errors that result from
bias in sampling, experimental design, or choice of assumptions.
Model uncertainty is associated with all models used in risk
assessment. These include the animal models used as surrogates
for human carcinogenicity, dose-response models used in
extrapolations, and the computer models used to predict the fate
and transport of chemicals in the environment. The use of
rodents as surrogates for humans introduces uncertainty into the
risk factor since there is considerable interspecies variability
in sensitivity. Computer models are simplifications of reality,
requiring exclusion of some variables that influence predictions
but cannot be included in models either because of increased
complexity or a lack of data on that parameter. The risk
assessor needs to consider the importance of excluded variables
on a case-by-case basis, because a given variable may be
important in some instances and not in others. A similar problem
can occur when a model that is applicable under average
G-23
-------
conditions is applied under conditions that differ from the
average. Large bodies of water, for example, can cause
meteorological conditions that are not adequately modeled by air
dispersion models such as the Industrial Source Complex - Long
Term (ISCLT) used in this analysis. Choosing the correct model
form is often difficult because conflicting theories seem to
explain a phenomenon equally well.
The third type, decision-rule uncertainty, is probably of
more concern to risk managers. This type of uncertainty arises,
for example out of the need to balance different social concerns
when determining an acceptable level of risk. Finkel provides a
complete discussion of decision-rule uncertainty.3
Variability, the fourth source of uncertainty, is often used
interchangeably with the term "uncertainty," but this is not
strictly correct. Variability may be tied to variations in
physical and biological processes and cannot be reduced with
additional research or information, although it may be known with
greater certainty (e.g., age distribution of a population may be
known and represented by the mean age and its standard
deviation). "Uncertainty" is a description of the imperfection
in knowledge of the true value of a particular parameter or its
real variability in an individual or a group. In general,
uncertainty is reducible by additional information-gathering or
analysis activities (better data, better models), whereas real
variability will not change (although it may be more accurately
known) as a result of better or more extensive measurements.7
Variability will introduce uncertainty if risks are calculated
for an individual drawn at random within a population.
The degree to which all types of uncertainty need to be
quantified and the amount of uncertainty that is acceptable
varies with the intent of the analysis. For a screening level
analysis, a high degree of uncertainty is often acceptable,
provided that conservative assumptions are used to bias potential
error toward protecting human health. A region-wide or
nationwide study will be less uncertain than a site-specific one
in determining average risks since, in the former case, it may be
possible to use the average of a parameter value over many sites
(which often can be estimated better than a site-specific value).
However, the general analysis may be highly uncertain in defining
the range of possible risks, which are influenced by
site-specific conditions. In general, the more detailed or
accurate the risk characterization, the more carefully
uncertainty needs to be considered.
G-24
-------
G.I.2.1 Qualitative Description of Uncertainty. Often, the
sources of uncertainty can be determined, but they cannot be
quantified. This can occur when a factor is known or expected to
be variable, but no data are available (e.g., the amount of time
people at a specific site spend out of doors). In this case,
sometimes default data are available that can be useful for
estimating a possible range of values. Uncertainty often arises
out of a complete lack of data. A process may be so poorly
understood that the uncertainty cannot be quantified with any
confidence. In addition, some sources of uncertainty (such as
uncertainty in theories used to deduce models) are inherently
qualifications reflecting subjective modes of confidence rather
than probabilistic arguments. When uncertainty can only be
presented qualitatively, the possible direction and orders of
magnitude of the potential error should be considered.
G.I.2.2 Quantitative Description of Uncertainty. Knowledge
of experimental or measurement errors can also be used to
introduce a degree of quantitative information into a qualitative
presentation of uncertainty. For example, standard laboratory
procedures or field sampling methods may have a known error level
that can be used to quantify uncertainty. In many cases, the
uncertainty associated with particular parameter values or for
the estimated risks can be expressed quantitatively. A six-step
process for producing a quantitative uncertainty estimate has
been identified as follows6:
• Define the measure of risk (e.g., deaths, life-years lost,
maximum individual risk (MIR), population above an
"unacceptable" level of risk). More than one measure of
risk may result from a particular risk assessment; however,
the uncertainty should be quantified for each individually.
• Specify "risk equations" that present the mathematical
relationships that express the risk measure in terms of its
components. This step is used to identify the important
parameters in the risk estimation process.
• Generate an uncertainty distribution for each parameter or
equation component. These uncertainty distributions may be
generated by the use of analogy, statistical inference
techniques, or elicitation of expert opinion, or some
combination of these.
• Combine the individual distributions into a composite
uncertainty distribution. Monte Carlo simulation, frequently
G-25
-------
used for this step, is discussed in greater detail later in
this section, and was used in this analysis.
• Recalibrate the uncertainty distributions. Inferential
analysis could be used to "tighten" or "broaden" particular
distributions to account for dependencies among the
variables and/or to truncate the distributions to exclude
extreme values.
• The output should be summarized in a manner that is clear
and highlights the important risk management implications.
Specific factors should be addressed including: the
implication of supplanting a point estimate produced without
considering uncertainty, the balance of the costs of under-
or overestimating risks, unresolved scientific
controversies, and implications for research.
When a detailed quantitative treatment of uncertainty is
required, statistical methods are employed. Two approaches to a
statistical treatment of uncertainty with regard to parameter
values are described here and were used in this analysis where
appropriate. The first is simply to express all variables for
which uncertainty is a major concern using an appropriate
statistic. For example, if a. value used is from a sample (e.g.,
yearly emissions from a stack) , both the mean and standard
deviation should be presented. If the sample size is very small,
it may be appropriate to give the range of sample values and use
a midpoint as a best estimate in the model; or, both the smallest
and largest measured value could be used to get two estimates
that bound the expected true value. The appropriate statistic to
use depends on the amount of data available and the degree of
detail required. Uncertainties can be propagated using
analytical or numerical methods.
A second approach is to use the probability distributions of
major variables to propagate parameter value uncertainties
through the equations used in a risk analysis. A probability
distribution of expected values is developed for each parameter
value. These probability distributions are typically expressed
as either probability density functions (PDF) or as cumulative
probability density functions (CPF). The PDF presents the
relative probability for discrete parameter values while the CPF
presents the cumulative probability that a value is less than or
equal to a specific value. The PDF approach was used here to
represent a propagate parameter uncertainty.
Uncertainties are propagated by developing a composite
uncertainty distribution by combining the individual
G-26
-------
distributions with the equations used to calculate probability of
cancer. Numerical methods are often employed for this phase,
with Monte Carlo simulations gaining wide acceptance for this
purpose. In Monte Carlo simulations, a computer program (e.g.,
Crystal Ball) is used to repeatedly solve the model equations
under different selections of parameter values to calculate a
distribution of exposure (or risk) values. Each time the
equations are calculated, values are randomly sampled from the
specified distributions for each parameter. The end result is a
distribution of exposure (or risk). These can again be expressed
as PDFs or, more appropriately, as CPFs. The distribution allows
the risk assessor to choose the value corresponding to the
appropriate percentile in the overall distribution. For example,
an exposure level or risk level can be selected that corresponds
to the 95th percentile of the overall risk distribution rather
than relying on a point estimate of risk based on the 95th
percentile values for each parameter. This allows the risk
analyst to reflect quantitatively the confidence of that risk
estimate with respect to the range of possible risks.
G.I.3 Approach to Uncertainty Analysis
This uncertainty analysis addresses only the cancer risks
associated with direct inhalation of HAPs emitted from electric
utilities and focused on parameter uncertainty within the models
and available data. Model uncertainty was not addressed
quantitatively due to limits in time and resources but are
described qualitatively. Variability has been evaluated
separately for exposure-response, but is included in the overall
estimate of uncertainty related to emissions and exposure.
Variability was not specifically addressed for emissions and
dispersion modeling because the available data did not permit
this distinction. The data presented to EPA was in different
forms, some raw data while others limited to summary reports. As
a result, it was not possible to identify inherent variability
and the impact of sampling and analysis errors.
The overall goal of this uncertainty analysis is to estimate
the range of possible estimates due to direct inhalation of
nonradionuclide HAPs emitted from utilities only. Risks from
indirect exposure, radionuclides, and long-range transport are
addressed not addressed in this appendix. This analysis focused
on parameter uncertainty and, where possible, variability. It
should also be noted that there are other sources of uncertainty,
some of which may be significant, which could not be evaluated
quantitatively but are qualitatively discussed.
The approach used in this analysis was to identify the
uncertainty with each of the parameters used in the risk
G-27
-------
estimation process. First, the uncertainty associated with each
these all variables was described using an appropriate statistic
(e.g., mean and standard error of means) or as a probability
density function (their relative probability for discrete
parameter values). The characterization of cancer risks from
direct inhalation is estimated based on the long-term average
concentration and exposures. Therefore, in most cases the
standard error of the mean (SE) for each parameter was used as
the estimate of uncertainty and variability rather than the
standard deviation (SD). The SE is a more appropriate statistic
given that the risk assessment is concerned with long-term
average exposures and intermittent extreme values of exposure
levels are not as important. However, directly using the SE from
the sample population would tendency be an overconfident estimate
of (i.e. too narrow a range) of uncertainty due to:
• The SE is used as an estimate of the SD of a normal
distribution while the t-distribution is used to represent
the sampling distribution. For small sample sizes, the
t-distribution is significantly different, and approaches a
normal distribution as samples sizes increase (n~30).
Therefore, error is introduced by assuming a normal
distribution.
• The mean of a distribution represents a long-term average
only when the population samples are representative of
values that would be obtained over the long-term. When data
are obtained from a convenience sample, additional error is
introduced due to potential nonrepresentativeness of the
sample.
• The mean of a distribution represent a long-term average
only if the population sample represents true variability.
Measurement error must also be considered. Generally the
most appropriate estimate of uncertainty in the long-term
average is to use the larger of the SE or the SD due to
measurement error.
Therefore, adjustments to the SE have been made to more
accurately reflect uncertainty. The procedure used is described
in Section G.3.1.
In general, once uncertainty was defined for each parameter,
numerical methods (i.e., Monte Carlo simulation) were then used
to develop a composite uncertainty distribution by combining the
individual distributions. The resulting distributions provides
an estimate of the range of values which can be expected and
allows the risk assessor to choose the value corresponding to the
G-28
-------
appropriate percentile in the overall distribution, or to
determine where the point estimate lies in the overall
distribution. For example, an exposure level or risk level can
be selected which corresponds to the 95th percentile of the
overall risk distribution rather than relying on a point estimate
of risk based on the 95th percentile values for each parameter.
Since variability was not differentiated in many cases from
uncertainty, they were simulated together in a one-dimensional
Monte Carlo simulation yielding a hybrid distribution. This
hybrid distribution applies to a single individual taken at
random from the exposed population. If either variability or
uncertainty dominates the analysis, then this one-dimensional
approach may also be used to characterize wether a specified
percentile of the population or a specified upper confidence
level, respectively.
Ideally, if data and resources permitted, variability and
uncertainty would be characterized separately and a two-
dimensional simulation could be conducted.8 The distributions
from a two-dimensional simulation would then allow a multi-
dimensional characterization of individual risks -- X level of
risk (e.g., 10"5) for the Yth percentile of the population
(addressing variability) with Z degree of confidence (addressing
uncertainty). However, the available data and resources did not
allow for this approach.
The uncertainty analysis was conducted on the three major
components of the risk assessment process, emissions
characterization, dispersion and exposure modeling, and
exposure-response assessment. A detailed uncertainty analysis
could not be conducted on all of the electric utility plants or
HAPs included in the original analysis. Therefore, a total of
four plants (2 oil-fired and 2 coal-fired plants) were selected
which contribute most to risk, based on both the estimated
incidence and the maximum individual risk. The highest risk
oil-fired and coal-fired plant were used for illustration
purposes in the text. Each of these plants was analyzed for
arsenic, nickel, and chromium which were shown to account for a
majority of the risk (which is explained in greater detail in
Section G.2).
G.2 QUANTITATIVE BASELINE ESTIMATES OF RISKS FROM DIRECT
INHALATION OF ELECTRIC UTILITY HAP EMISSIONS
This section summarizes the risk estimates that were
generated as part of the EPA risk assessment of direct inhalation
exposures to electric utility hazardous air pollutant (HAP)
emissions. The risk assessment is described in detail in
G-29
-------
Chapters 4 to 6. The risk estimates are summarized here as a
foundation and reference point for the uncertainty analysis.
They are referred to throughout this document as the "baseline".
Three measures of risk were estimated under the baseline case:
individual risk, population risk, and noncancer risk.
G.2.1 Risks Associated With Coal-fired Electric Utility Boilers
A total of 426 coal-fired were modeled using the HEM, and
1990 emissions and population data. The HEM estimated the
ambient HAP concentrations within 50 km of the plant, the
population exposed, and the distribution of exposure within the
population. Table G-6 summarizes the MEI risks, the number of
persons exposed above individual cancer risk levels of 10"7 and
10"6, the number of plants whose emissions result in those risk
levels, and the maximum HQ for HAPs evaluated.
G.2.1.1 Individual Cancer Risk. The MEI risk was highest
for arsenic (Class A, human carcinogen) at 3 x 10"6. Table G-6
also shows that arsenic emissions from 42 plants resulted in MIRs
of greater than or equal to 10~7.
G.2.1.2 Population Cancer Risk. As with the MIR, arsenic
and chromium are the major contributors to the total population
exposed to risk levels of one in 1 million (1 x 10"6) or more.
Approximately 850 people are estimated to have a risk of 1 x 10"6
or greater from exposure to arsenic, and about 100 people have a
comparable risk from exposure to chromium. The HEM also
calculated the annual incidence of cancer expected for each of
the chemicals. The total cancer incidence from all HAPs was
estimated as 0.09 cases per year. As shown in Figure G-4,
arsenic and chromium are again the major contributors and account
for almost 90 percent of the estimated cancer incidence.
G.2.1.3 Noncancer Risk. The maximum HQ estimated for
noncarcinogenic HAPs emitted from coal-fired plants was 0.12 for
hydrogen chloride. HQ values for all other HAPs were lower.
G.2.2 Risks Associated With Oil-fired Electric Utilities
A total of 137 oil-fired plants with boilers of 25 MW or
were evaluated using 1990 HAP emissions and population data.
Table G-7 summarizes some of the risk values for oil-fired
plants.
G.2.2.1 Individual Cancer Risk. However, there are maj or
uncertainties in these estimates because of the speciation of
nickel emissions from oil-fired utilities. In this analysis, as
a conservative assumption (i.e., more likely to overestimate
rather than underestimate risk), all nickel was assumed to be
G-30
-------
Table G-6. SUMMARY OF RISK BY HAP: COAL-FIRED POWER PLANTS
Pollutant
Arsenic
Beryllium
Cadmium
Chromium'
Dioxin/furans
Hydrogen
chloride
Lead
Manganese
Mercury
Nickel"
n-Nitrosodi-
methylamine
Total
Carcinogens
MEI
3x Id'6
3x 10'7
2x 10'7
2x 10-6
5x 10'8
NA
NA
NA
NA
7x 10'7
8x TO'7
4x 10-*
Population
MIR > 10 7
1.7M
1,280
107
80,500
0
NA
NA
NA
NA
5,100
9,150
NA
# Plants
MIR > 10 7
42
2
1
10
0
NA
NA
NA
NA
3
4
42
Population
MIR > ID"6
2,370
0
0
107
0
NA
NA
NA
0
0
0
NA
# Plants MIR
> 10*
2
0
0
1
0
NA
NA
NA
0
0
0
2
Noncarcinogen
HCU,
NA
NA
NA
NA
NA
0.12
0.001
0.046
0.002
NA
NA
NA
MEI = Maximum exposed individual, calculated using the highest concentration. An individual may or may
not be exposed at that point. This value may be greater than the MIR, which is calculated at the
centroid of a census block.
MIR = Maximum Individual Risk, highest risk identified at the centroid of a census tract to which a
population is assigned.
NA = Not available.
HQ = Hazard quotient, the ratio of exposure concentration to the Reference Concentration (RfC). HQ
values below 1 are not expected to result in adverse effects.
Total = Total MEI and Total MIR are the sum of the MIR and MEI for individual HAPs within a plant.
The total HQ ( = HI) is the sum of the HQs within a plant.
• Assumes that 11 percent of total chromium emitted is hexavalent chromium, the species of chromium
responsible for carcinogenic potential. Trivalent chromium, which would also be present, is not thought
to have carcinogenic potential.
b The nickel emitted is a mixture of various nickel compounds such as soluble nickel. This analysis
assumes that all nickel emitted has the same carcinogenic potency as nickel subsulfide.
G-31
-------
CD
•5
0
O
c
0
;g
'o
c
*_
CD
o
C CO
CO 0
O j|
§5
.
M- LL
O _L
C *
is
•c o
•4— • i_
CO «•-
B 8
O)
Li.
G-32
£
co
£
CD
TJ
I
a
I
I
T3
1
CD
Q.
§
co
o
£
o>
c
1
^
1
1
CD
o
m
c
E
£
§
Q.
J8
^^
_£ 3
•fr CD
g£
=5^
C T3
||
-------
Table 6-7. SUMMARY OF RISK BY HAP: OIL-FIRED POWER PLANTS
Pollutant
Arsenic
Beryllium
Cadmium
Chromium
Dioxin/furans
Hydrogen
chloride
Lead
Manganese
Mercury
Nickel"
Total
Carcinogens
MEI
1 x 10'5
8x 10'7
2x 10-*
5x 10-6
1 x 10'7
NA
NA
NA
NA
1 x Id"4
1 x ID"4
Population
MIR > 10'7
2.1M
2,280
3,040
257,000
45
NA
NA
NA
NA
73M
NA
# Plants MIR
> 10 7
22
2
2
10
1
NA
NA
NA
NA
79
85
Population
MIR > 10-*
2370
0
45
2,280
0
0
0
0
0
1.65M
NA
# Plants MIR
> 10*
2
0
1
1
0
0
0
0
0
20
22
Noncarcinogen
HO™
NA
NA
NA
NA
NA
0.0055
0.0004
0.037
NA
NA
NA
MEI = Maximum exposed individual, calculated using the highest concentration. An individual may or may
not be exposed at that point. This value may be greater than the MIR, which is calculated at the
centroid of a census block.
MIR = Maximum Individual Risk, highest risk identified at the centroid of a census tract to which a
population is assigned.
NA = Not available.
HQ = Hazard quotient, the ratio of exposure concentration to the Reference Concentration (RfC). HQ
values below 1 are not expected to result in adverse effects.
Total = Total MEI and Total MIR are the sum of the MIR and MEI for individual HAPs within a plant.
The total HQ (= HI) is the sum of the HQs within a plant.
' Assumes that 18 percent of total chromium emitted is hexavalent chromium, the species of chromium
responsible for carcinogenic potential. Trivalent chromium, which would also be present, is not thought
to have carcinogenic potential.
b This analysis conservatively assumes that all nickel emitted from electric utilities has the same
carcinogenic potency as nickel subsulfide (the highest potency of nickel compounds tested). However,
the nickel emitted is a mixture of various nickel compounds such as soluble nickel. Emissions tests have
found nickel subsulfide to be present as less than 10 percent of total nickel emitted. Most nickel
compounds are assumed to have carcinogenic potential though the potency is not known. If the relative
potency (URE) of the mixture of nickel compounds emitted from oil-fired utilities was 50 percent of
nickel subsulfide, about 23 million persons would be exposed at an MIR > 10~7, and 100,000 at an MIR
> 10"6, if the URE were 20 percent of nickel subsulfide, about 7.5 million persons would be exposed at
an MIR > 10'7, and 9,930 at an MIR > 10"6.
G-33
-------
equipotent to nickel subsulfide, which has the highest cancer
potency of nickel compounds evaluated by EPA. The limited
speciation data indicate that less than 10 percent of all nickel
emissions may be nickel subsulfide. The remainder of the nickel
is a combination of various nickel compounds for which the EPA
has not yet determined the carcinogenic potency.
There were three oil-fired plants with risks exceeding one
in 100,000 (1 x 10"5) due to nickel exposures and one plant that
exceeded this level due to arsenic exposure. Nickel was
responsible for MIRs that exceeded one in 10,000,000 (1 x 10~7)
at 79 plants and exceeded 10"6 at 20 plants.
G.2.2.2 Population Risk. As with the MIR, nickel and
arsenic are the major contributors to the total population
exposed to risk levels of one in 1,000,000 (1 x 10"6) or more.
The total cancer incidence associated with HAP emissions
from the 137 oil-fired plants was estimated as 0.47 cases per
year. As shown in Figure G-5, nickel accounts for about 86
percent of the total annual incidence.
G.2.2.3 Noncancer Risks. The highest reported HQ
resulting from oil-fired power plant emissions was 0.055 for
manganese. The HQs for other HAPs and other plants were lower.
G.2.3 Risks From Gas-fired Plants
Risks were estimated for 1990 HAP emissions from 267
gas-fired facilities. Table G-8 summarizes the results. HAP
emissions from only one plant resulted in risks greater than one
in 10 million (10~7) with only 23 persons exposed above that
level. For noncarcinogens, the maximum HQ estimated was
1.32 x lO'7.
G.2.4 Risks from All Electric Utilities
G.2.4.1 Individual Cancer Risk. HAP emissions from oil-
and coal-fired facilities are the major contributors to risk;
gas-fired utilities have been shown to present lower risks. As
expected from the previous discussion, nickel, arsenic, and
chromium are the major contributors to the MIR.
G.2.4.2 Population Cancer Risk. Table G-9 summarizes the
population exposed to the major carcinogens at various risk
levels for coal-fired and oil-fired facilities. The table shows
that arsenic and chromium contribute the most to the risk from
coal-fired plants compared to nickel and arsenic for oil-fired
plants.
G-34
-------
O)
c
CO
0
- '£:
CD o
O )
O
it
£6
I!
11
II
Q LU
CD <
£1
CD c
°:i
LO *-
O)
G-35
-------
TABLE G-8. SUMMARY OF RISK BY HAP: GAS-FIRED POWER PLANTS
Pollutant
Arsenic
Lead
Mercury
Nickel*
Carcinogens
MEI
2x 10'7
NA
NA
2x 10'7
Population
MIR > 10 7
0
NA
NA
23
# Rants MIR
> 10'7
0
NA
NA
1
Population
MIR > 10*
0
NA
NA
0
# Plants MIR
> 10*
0
NA
NA
0
Noncarcinogen
HO™
NA
1.32x 10'7
NA
NA
MEI = Maximum exposed individual, calculated using the highest concentration. An individual may or may not be
exposed at that point. This value may be greater than the MIR, which is calculated at the centroid of a census
block.
MIR = Maximum Individual Risk, highest risk identified at the centroid of a census tract to which a population is
assigned.
NA = Not available.
HQ = Hazard quotient, the ratio of exposure concentration to the Reference Concentration (RfC). HQ values below 1
are not expected to result in adverse effects.
• The nickel emitted is a mixture of various nickel compounds such as soluble nickel. This analysis assumes that all
nickel emitted has the same carcinogenic potency as nickel subsulfide.
Table G-9. SUMMARY OF ESTIMATED POPULATION EXPOSED AT VARIOUS
LEVELS OF RISK OR GREATER
Number of Persons Exposed at Various Risk Levels or Greater by HAP
COAL-FIRED PLANTS
Risk Level
5E-6
2.5E-6
1E-6
5E-7
2.5E-7
1E-7
Arsenic
0
0
852
5,990
88,800
1,710,000
OIL-FIRED PLANTS
Risk Level
5.E-5
2.5E-5
1E-5
Nickel
89
2,240
2,310
Chromium
0
0
107
2,160
8,630
80,500
Arsenic
0
0
45
n-Nitroso
0
0
0
399
585
9,150
Chromium
0
0
0
Nickel
0
0
0
0
947
5,100
Cadmium
0
0
0
Beryllium
0
0
0
0
0
1,280
Beryllium
0
0
0
Cadmium
0
0
0
0
0
107
Dioxins/
furans
0
0
0
Dioxins/
furans
0
0
0
0
0
0
G-36
-------
Table 6-9. Continued
OIL-FIRED PLANTS
Risk Level
5E-6
2.5E-6
1E-6
5E-7
2.5E-7
1E-7
Nickel
9,930
100,000
1,650,000
7,460,000
23,100,000
73,300,000
Arsenic
89
2,280
2,370
32,600
287,000
2,140,000
Chromium
45
89
2,280
2,280
9,490
257,000
Cadmium
0
0
45
89
2,280
3,040
Beryllium
0
0
0
45
89
2,280
Dioxins/
furans
0
0
0
0
0
45
Note: There may be multiple counting of population around facilities within 50 km of each other. Exposed
individuals would be included in the statistics for each plant within 50
Approximately 0.56 cancer cases per year are estimated to
result from HAP emissions dispersed within 50 km utilities.
Figure G-6 shows the contribution of each compound to the
incidence. When all of the plants are combined, nickel, arsenic,
and chromium account for over 95 percent of the incidence.
G.2.4.3 Noncancer Risks. Emissions of noncarcinogenic HAPs
from utilities did not result in an estimated HAP air
concentration that exceeded the RfC. The highest HQ was 0.12 for
hydrogen chloride from a coal-fired plant.
G.2.5 Discussion
Direct inhalation exposure to HAPs emitted from a total of
178 utilities result in individual risk of 1 in 1,000,000 (10~7
or greater); exposures to emissions from 24 plants are estimated
to result in individual risks of 1 in 10,000,000 (10~7 or
greater). Emissions from oil- and coal-fired facilities are the
major contributors to these risks from inhalation, while
gas-fired utilities have been shown to present lower risks. The
highest MET risk accross all plants was estimated to be 1 x 10"4.
These risk estimates are for direct inhalation exposure
alone and do not include risks associated with long-range
transport or indirect exposures. The risks from indirect
exposure may be significant for those compounds which are
environmentally persistent and have the tendency to
bioaccumulate.
G-37
-------
£
CD
I
G-38
-------
The risk assessment process typically uses a few
conservative assumptions and default options to address some of
its inherent uncertainties. However, in some cases assumptions
and defualt options are used that are likely to underestimate
exposures and risks. For example, the HEM inherently
underestimates individual exposure in an exposed population
impacted by more than one plant. The HEM calculates risks for an
individual plant for the exposed population within 50 km of the
plant. However, some individuals may be within 50 km of more
than one power plant and would be double-counted and included in
the risk distribution for each plant presumably at different
individual risk levels. The risk for these individuals is the
sum of the risks for each plant to which they are exposed. As a
result, the HEM underestimates exposures for individuals exposed
to emissions from more than one power plant. However, a modeling
analysis (section 6.5 of Chapter 6) was conducted to evaluate the
impact of the uncertainty associated with overlapping plumes.
This analysis indicates that overlapping plumes leads to only a
slight, if any, increase in exposures for the MEIs. Also, the
method of double-counting used in the HEM does not affect the
estimation of cancer incidence.
A risk characterization must address the uncertainties and
limitations of the analysis. Therefore an uncertainty analysis
evaluating each of the major models, assumptions, data, and
variability was conducted on the results of this risk assessment
The uncertainty analysis helps identify errors of either
overestimation or underestimation. Methods, models, data, and
assumptions used in the analysis are identified, including their
rationale and the effect of reasonable alternative assumptions on
the conclusions and estimates. These are presented in Section
G.3 for exposure assessment, and Section G.4 for the
exposure-response relationship. The uncertainties identified in
Sections G.3 and G.4 are propagated (combined) to generate
overall estimates of uncertainty in the risk estimates
themselves. This process and the results are described in
Section G.5.
G.3 ANALYSIS OF UNCERTAINTY OF DIRECT INHALATION EXPOSURE TO
HAPS EMITTED FROM ELECTRIC UTILITIES
Exposure assessment is the determination or estimation
(qualitative or quantitative) of the magnitude, frequency,
duration, and route of exposure. An exposure assessment has four
major components:
• Emissions characterization,
• Environmental fate and transport,
G-39
-------
• Characterization of the study population, and
• Exposure calculation.
The overall exposure assessment is summarized in chapters 4 to 6
of the interim report along with a discussion of the
uncertainties and limitations of this exposure assessment,
including default options; models, methods, and data;
uncertainty; variability; and aggregation. This section
summarizes a more detailed uncertainty analysis that was
conducted on the exposure assessment. The exposure assessment
comprised two distinct phases: emissions characterization and
exposure modeling (which includes environmental fate and
transport, characterization of the study population, and exposure
calculation). Emissions characterization for this utility study
is described in detail in Chapter 3. The environmental fate and
transport, population characterization, and exposure calculations
are described in detail in Appendix F.
The baseline exposure assessment used a combination of
assumptions, parameters, and input data. Some of these are
considered conservative (more likely to overestimate than
underestimate exposure) and others are not conservative. Any
procedure that relies on a combination of point values-some
conservative and some typical-yields a point estimate of exposure
and risk that falls at an unknown percentile of the full
distributions of exposure and risk.7 For this reason, the degree
of conservatism in the exposure calculation cannot be determined.
A quantitative uncertainty analysis was conducted to identify the
degree of uncertainty, degree of uncertainty, and overall
confidence in the inhalation risk assessment results for the
local analysis (i.e., within 50 km of plants).
The uncertainty analysis was conducted separately for the
two phases of the exposure assessment identified above:
emissions characterization and exposure modeling. These analyses
are described in Sections G.3.1 and G.3.2, respectively.
The uncertainty analysis focuses on three HAPs (arsenic,
chromium, and nickel), which contributed most to inhalation risks
and is considered to be generally representative of uncertainties
for the full complement of HAPs. Furthermore, a detailed
uncertainty analysis could not be conducted on all of the utility
plants included in the original analysis. Therefore, four plants
(2 oil-fired and 2 coal-fired plants) were selected that
contribute most to risk, based on both the estimated incidence
and the maximum individual risk. Again, these are considered to
be generally representative of other plants.
G-40
-------
> .^^^^°»»°-
.el
al
. 100
(YQ
., . 100
o^-*- •
Too:
G-<
-------
'at
to
* 0.
to
-3
c3ata r>J-1.2 e
^ti^f^tecj I^SlCo^
Po^JV^te/RM? ^yeTr^S.
POg
Vari^ s^
-5:?^t- «*
^^^ to00 ^
-------
U)
O)
•a
£
il
6
(0
2
3
O)
O
O
T3
CD
.O
"o.
w
o
O
"33
CO
E
3
CO
I
I
o
_a
CO
1
u
S
J5
^^
s
(L
O
a
«
i
§
'i -
| >
> *>
*5
c o
CD O
'.§ ?
"S
o
U
c
o t:
*5 •£•
i 1
™ ft
» •*•
c o
O o
O o
ss *""
O
a
a
Z
^
c
3
1
O
cc
z
o
o
o
z
XONAUQJ
o
cc
1
o
CO
in
o
CM
en
CM
CO
1
t~
0)
CM
O
cc
o
u
O
Z
FPRYNOX
O
cc
1
o
CO
CO
o
CO
in
CO
CO
CM
O)
CM
o
cc
o
o
O
Z
FPRYNOX
O
cc
O
CO
in
CO
0
o
CO
O
CO
en
CM
O
cc
Z
o
u
o
z
X
PRYNONO
0
O
en
CO
o
,_
CD
CO
CD
^f
en
CM
O
cc
Z
o
o
0
z
X
PRYNONO
(D
^C
i
0
en
CO
CM'
CM
0
.
CO
•
*~
in
en
CM
g
Z
O
0
0
z
X
o
z
cc
0
0
cc
0
<*•
o
o
en
O
CM
CO
CO
CO
o
cc
z
o
o
o
z
XONAUQJ
O
cc
O
r-
^
0
T—
CD
CM
^
CO
CO
t—
O
cc
z
o
u
o
z
XONAUQJ
0
cc
o
o
CO
Q
^-
00
r—
^
in
CO
CO
0
cc
h-
O
U
o
z
XONAUQJ
0
cc
0
CO
!v.'
O
in
en
CO
CO
CD
CO
CO
O
cc
O
U
O
Z
X
PRYNONO
a
L_
"7
O
CD
CO
in
O
ix
,_
«—
00
p»»
CO
CO
O
cc
z
o
o
o
z
X
PRYNONO
0
U-
~
6
CO
^
CO
0
en
00
CO
CO
G-43
-------
CO
CD
T3
O
§
O
co
c
g
to
o
O
•o
CO
o
Q.
3
CO
O
O
o
LL
CO
I
V)
.
o
O
£
1
|
U
I
SO2 Control
i 1
1°
Boiler type
Coeff. of
variation (1000
tons/yr)
Coal
Consumption
(1000 tons/yr)
COM ongln
used hi EFP
I
I
8
*-
z
1
CO
8
en
CO
CO
CM
*
IO
CO
CO
CO
d
§
NO CONTROl
CO
X
Q.
CO
UJ
X
o
1
X
CO
CN
IO
o
CO
cc
IO
d
-
e*
I
r-
z
3
CD
3
n
CO
CO
CM
1
o
CO
CO
CD
-J
§
NO CONTROl
CO
X
0-
n
UJ
rANGNONOX
X
i
1,135.60
d
CM
C*
35
8
«-
w
3
CO
3
n
CO
CO
CM
5
o
CO
CO
CD
-J
i
NO CONTROl
en
a.
n
UJ
X
O
3
<
X
en
I
1,424.60
-
CO
CO
p
8
*-
t
3
CO
3
»
CO
CO
CM
5
D
CO
CO
=!
3
NO CONTROl
CO
a.
co
UJ
rANGNONOX
X
IO
CO
§
I
d
-
CO
CO
*
CO
0
5
DO
CD
•-
^
CM
CO
CM
CO
1
O
IO
CO
2
§
O
NO CONTROl
CO
X
Q.
CO
UJ
XONONONVJ
X
CM
CT
CO
UJ
rANGNONOX
X
CO
IO
c
*
o
CM
G-44
-------
• Long-term average exposure is used to calculate cancer risks
minimizing the importance of short-term variations.
• Data on fuel consumption rates are not reported for yearly
consumption only.
• Use of individual units as base or peak load is likely to
change over time, providing for a cycle of repair and
maintenance for individual units.
Uncertainty (including variability) in fuel consumption
estimates can be expected to occur as a result of changes in
demand for electricity (seasonally and annually), varying value
of the fuel used (e.g., bituminous vs. anthracite coal), and
changes in the source of fuel (e.g., State of coal origin).
However, there are no data on which to estimate the uncertainty
introduced by these factors and for uncertainty in fuel
consumption rates. Data are not readily available on the rate of
fuel consumption for individual units by month or season. The
data are self-reported and to evaluate the accuracy of these data
and to evaluate trends in fuel consumption over time would
require significant primary data collection from individual
utilities, which is beyond the scope of this study and may be
restricted by guidelines established by the Office of Management
and Budget. There were no data readily available which provided
insight into the uncertainty or variability or uncertainty in
fuel consumption. The only data available were from emissions
tests conducted under optimal or steady state conditions which
are not representative of uncertainty in long-term consumption.
Therefore, a normal distribution with a coefficient of variation
of +5 percent was arbitrarily selected as a lower bound on the
uncertainty in long-term average fuel consumption.
G.3.1.3 HAP Concentration in Fuel. Separate approaches to
estimate HAP concentration in coal and oil were used in the
emissions characterization. These were necessitated by the broad
range of types and sources of coal and a lack of concentration
data for fuel oil. Therefore, the uncertainty analysis was
conducted, and is discussed, separately for each.
G.3.1.3.1 HAP Concentration in Oil. The HAP concentration
data in oil are summarized in Table G-ll. Limited data on trace
element concentrations in oil were available. Two distinct
groups of test data of trace element concentrations in residual
oil have been identified. The risk assessment was conducted
using one data set (FCEM) of a total of 12 different samples of
residual oil and initial data analysis assumed that non-detects
were at one-half the detection limit. Residual oil is assumed to
G-45
-------
CO
LU
O
CC
5
LU
LU
a
LU
CC
CO
o
1-
m
co
o
DC
P
LU
LJL
o
CC
CO
CD
_eo
JD
o
».
g
0
CM
CO
O
0
00
CM
O
1
.a
c
•c
"c
*-
0
in
d
CO
3
T"
^
O
3)
d
Tj-
co
0
a;
|
1
1
T—
h
o
8
o
CO
in
d
CO
in
0
CO
ex
.CO
X
o
o
1
c
CC
"-
o
o
d
CM
g
^_
en
00
0
%
d
en
C-.
o
re
X
o
o
73
C
u.
Oil/APCD EM
8
o
CO
o
d
CM
in
in
o
in
o
CM
in
d
O)
Tf
O
"5
.a
ESPcold
IIArsenic
8
0
Q
d
CM
5
o
5
o
d
^f
^
o
CO
1
ESPcold
llChromium
o
o
o
d
CM
N
0
2
0
d
fx.
o
o
«
ESPcold
a
c
Z
Q
CO
O
a
±>
CO
C3)
d
rj-
•^
CO
T-
D
Z
CO
E
o
en
o
^5
III
LL
1
,_
d
CM
CM
b
o
1
co
E
o
ex
o
CO
o
CO
c/j
(/
^_
8
d
CM
CM
T—
1
co
E
o
ex
o
C/]
CS
01
LJ
o
o
•
-------
be representative of all oil fuel burning since 95 percent of
oil-fired boilers burn residual oil, and that the 12 samples were
also representative. The average concentrations taken to
represent the "best estimate" were 0.699 ppm for arsenic, 0.3 ppm
for chromium, and 24.37 ppm for nickel.
Subsequently a second data set (SGS) of about 22 samples of
residual fuel oil were obtained which initially appeared to be
similar with respect to average concentration (Baker and
Robertson, 1994). Both data sets were re-analyzed using an
alternative treatment of non-detects (using a probability
plotting routine) and assumed to be log-normally distributed
(prohibiting negative concentrations with no upper bound limit).9
The results of the analysis are summarized in Table G-ll. The
results indicate that the trace element concentration differs
from the averages previously calculated (using non-detects equal
to half the detection limit), and that the two data sets may
reflect sampling from appreciably different "population of oil
samples." The distributions for nickel are
roughly comparable between the two data sets, but differ
dramatically for both chromium and arsenic. The difference in
the two "populations" of oil samples may reflect some combination
of real changes over time in residual oil HAP concentrations,
seasonal differences, and/or differences in analytical chemistry.
Residual fuel oil is the leftovers of the oil refining process
and does vary by season (reflecting the shift of gasoline
production in summer for heating oil in the winter), and with
shifts in the composition of the crude and demands for different
volatility and other characteristics imposed by VOC control
requirements.
The major uncertainties associated with oil HAP
concentration are whether there are significant differences in
the oil burned throughout the country and whether differences in
concentrations may occur spatially (geographic origin),
temporally, or by type of oil. Since 95 percent of oil-burning
units burn residual oil, use of the two residual oil data sets is
reported. However, how well that residual oil data represent the
universe of residual oil burned or which sample "population" is
more representative is not known.
Therefore, the two data sets will be treated as two distinct
cases and are assumed to be representative of the range of oil
being burned by utilities. It is also assumed that the
uncertainty (including variability) in the mean concentration
within that data sets also represents the uncertainty in the
long-term mean HAP concentration for each utility. Given that
little information is available on how representative these data
G-47
-------
are of the oil being burned, and that regional differences are
expected, the standard deviation (SD), rather than the SE, was
used as a measure of uncertainty. The use of the SD is supported
because any one individual facility may use oil which is
associated with any one of the samples analyzed, and therefore,
the standard deviation would more accurately reflect the
uncertainty in the HAP concentration.
G.3.1.3.2 HAP Concentration in Coal. The estimation of HAP
concentration in coal can be rather complex because
concentrations differ among types of coal burned (bituminous,
sub-bituminous, lignite, and anthracite) and the geographic
origin (State) of that coal. In addition, individual plants may
burn coal of multiple types and origin. For units burning coal,
the risk assessment assigned each unit a single State of coal
origin based on the State of origin for the majority of coal
burned at that plant. The average concentrations of metallic
HAPs reported in the U.S. Geological Survey [USGS] database for
that State were multiplied by the amount of fuel that the unit
burned in 1990. The estimation of HAP concentration is further
complicated by the fact that bituminous coal is cleaned prior to
shipment and burning, which reduces the trace element
concentration. Therefore, an additional coal cleaning factor
(CCF) (the proportion of HAP remaining after cleaning) was
applied to those States with regulations requiring cleaning. The
average of coal cleaning based on data from cleaning samples from
Alabama, Illinois, Kentucky, and Pennsylvania was applied to
bituminous coal to obtain the HAP concentration to be used in the
emissions estimating equation (Equation G-l).
The major uncertainties associated with the HAP
concentrations in coal include:
• Trace element concentration in coal has been shown to vary
widely among States, within a State geographically, and even
within a single coal seam. The use of a single-point
estimate may not represent the range of concentrations (or
average concentrations) that may be experienced at a single
plant. Therefore, it is possible that the risk assessment
may over- or underestimate actual emissions.
• The USGS database contains concentrations of trace elements
in coal extracted from the ground but does not include
analyses of coal shipments. The concentrations of trace
elements in coal in the ground and in coal shipments to
utilities may differ because, in the process of preparing a
coal shipment, some of the mineral matter in coal may be
removed. As a result, emissions from coal-fired units may
G-48
-------
be lower than predicted, resulting in an overestimation of
risk, though the degree of overestimation cannot be
determined.
• Individual units may burn coal that originated from several
States. These coals differ in values and trace element
concentrations. For this analysis, the EF model assigns
each coal-fired unit a single State of coal origin.
Therefore, actual emissions from a utility may differ
significantly from that estimated if alternative coals are
used with higher or lower trace element concentrations.
This analysis quantitatively evaluated the uncertainty
introduced with assignment to a single State of coal, and the HAP
concentration within individual States, and the effectiveness of
coal-cleaning. The uncertainty in relating the USGS data
measured within the seam to the actual coal shipments could not
be quantitatively assessed because of the lack of data.
The uncertainty analysis focused on developing an estimate
of composite coal concentration for each facility that would be
used in the emissions estimation equation (Equation G-l). This
estimate included both an average or "best" estimate and some
measure of the uncertainty around that estimate. The SE from the
data were used because of the large number of samples and
exhaustive sampling by the USGS of economically viable veins.
This was accomplished by first defining distributions for HAP
concentration in coal for each state of coal origin. Coal from
nine States (Illinois, Indiana, Kentucky, New Mexico,
Pennsylvania, Utah, Virginia, West Virginia, and Wyoming) was
burned in the nine coal-fired plants evaluated. The HAP
concentration is assumed to be log-normally distributed (which
supports the physical constraint of negative concentration with
no upper limit). The raw data for those States were reviewed and
the average HAP concentration and SEM for each of the three HAPs
were calculated for each State and are presented in Table G-12.
The raw data on coal cleaning were also reviewed and an
average and SEM were calculated for each of the three HAPs. The
HAP concentration and the CCF were multiplied together to
estimate the HAP concentrations for those States subject to coal
cleaning to estimate the HAP concentration in "cleaned" coal.
The SEMs for both the CCF and HAP concentrations were used to
estimate the uncertainty in the HAP concentration in "clean" coal
using the following formula:
G-49
-------
Table G-12. SUMMARY OF PARAMETER DISTRIBUTIONS FOR COAL-FIRED ELECTRIC UTILITIES
'arameter
TRACE ELEM CONC
Arsenic
Cleaned
Cleaned
Cleaned
Cleaned
Cleaned
Chromium
State of origin
CO
IN
IN
KY
KY
NM
PA
PA
VA
VA
WV
WV
WY
CO
IN
Cleaned IN
KY
Cleaned KY
INM
Cleaned
Cleaned
Cleaned
Nickel
PA
PA
VA
VA
WV
WV
WY
CO
UN
Cleaned
Cleaned
Cleaned
Cleaned
Cleaned
IN
KY
KY
NM
PA
PA
VA
VA
WV
WV
WY
I
Distribution
Log normal
Log normal
Log normal
Log normal
Log normal
Log normal
Log normal
Log normal
Log normal
Log normal
Log normal
Log normal
Log normal
Log normal
Log normal
Log normal
Log normal
Log normal
Log normal
Log normal
Log normal
Log normal
Log normal
Log normal
Log normal
Log normal
Log normal
Log normal
Log normal
Log normal
Log normal
Log normal
Log normal
Log normal
Log normal
Log normal
Log normal
Log normal
Log normal
Unit
ppm
ppm
ppm
ppm
ppm
ppm
ppm
ppm
ppm
ppm
ppm
ppm
ppm
ppm
ppm
ppm
ppm
ppm
ppm
ppm
ppm
ppm
ppm
ppm
ppm
ppm
ppm
ppm
ppm
ppm
ppm
ppm
ppm
ppm
ppm
ppm
ppm
ppm
ppm
Min.
0.5
0.5
0.1
0.2
0.31
1
0.08
0.28
1.09
1.4
3.1
0.07
4.6
3.8
2.6
0.6
0.9
5.3
2.4
0.45
3.1
3.2
1.7
0.7
Central
Arith
1.3
10.1
5.5
19.1
10.4
1.8
32.11
17.5
11.2
6.1
10.7
5.8
0.7
1.9
15.5
7.9
16.3
8.3
6.0
20.1
10.3
12.5
6.4
15.3
7.8
2.8
1.3
17.9
10.1
17.5
9.9
4.6
20.4
11.6
11.2
6.4
14.3
8.1
2.2
Tendency
Geo
Max
2.9
42.7
169.0
13.9
16920
79.1
237.4
2.09
3.83
75.3
86
19.1
67
28.2
131.8
10.95
3.64
48.14
108.9
14.5
223.6
26.9
280
5.8
N
30
73
267
96
527
39
431
145
30
69
266
96
532
39
436
145
30
73
266
93
547
39
437
145
Std Err
1.07
0.72
1.67
1.20
32.10
17.56
2.76
1.57
0.44
0.06
1.14
0.81
0.57
0.65
0.27
0.41
Std Dev
0.75
0.84
0.62
0.52
0.61
0.07
0.13
1.26
0.92
0.83
0.74
0.26
0.65
0.76
0.95
0.65
0.76
0.63
0.07
G-50
-------
Table G-12 (continued)
Parameter
COAL CLEAN FACTOR
Arsenic
Chromium
I Nickel
i
1
EMF: Boiler
Arsenic
Chromium
cyclwetnonox
frontdrynonox
htangnonox
oppdrynox
tangdrynonox
cycwetnonox
frontdrynonox
htangnonox
oppdrynox
tangdrynonox
Nickel !
icycwetnonox
frontdrynonox
l htangnonox
i oppdrynox
tangdrynonox
EMF: Control device
Arsenic ;
i Baghouse
Chromium
Nickel
ESPcold
ESPhot
wetscrub
Baghouse
ESPcold
ESPhot
wetscrub
Baghouse
ESPcold
ESPhot
wetscrub
Distribution
normal
normal
normal
beta
beta
beta
triangular
beta
beta
beta
beta
triangular
beta
beta
beta
beta
triangular
beta
beta
beta
beta
beta
beta
beta
beta
beta
beta
beta
beta
beta
beta
beta
Unit
unitless
unrtless
unitless
unitless
unitless
unitless
unitless
unitless
unitless
unitless
unitless
unitless
unitless
unitless
unitless
unitless
unitless
unitless
unitless
unitless
unitless
unitless
unitless
unitless
unitless
unitless
unitless
unitless
unitless
unitless
unitless
unitless
Min.
0.23
0.25
0.27
0.25
0.22
0.89
0.11
0.60
0.22
0.61
1.00
0.20
0.58
0.12
0.76
0.91
0.29
0.44
0.00
0.00
0.02
0.04
0.00
0.00
0.02
0.03
0.00
0.00
0.00
0.05
Central
Arith
0.54
0.51
0.57
0.53
0.71
0.89
0.73
0.80
0.27
0.87
1.00
0.61
0.79
0.28
0.85
0.91
0.44
0.64
0.01
0.03
0.06
0.32
0.06
0.03
0.03
0.42
0.04
0.05
0.01
0.79
Tendency
Geo
0.51
0.48
0.54
0.50
0.59
0.89
0.59
0.77
0.26
0.85
1.00
0.55
0.76
0.26
0.84
0.91
0.42
0.60
0.01
0.02
0.04
0.20
0.02
0.02
0.03
0.26
0.01
0.02
0.00
0.53
Max
0.97
0.89
0.85
0.85
0.99
0.89
1.00
1.00
0.35
1.00
1.00
1.00
1.00
0.72
1.00
0.91
0.67
0.84
0.03
0.06
0.11
0.76
0.25
0.07
0.04
1.00
0.20
0.37
0.01
1.00
N
27
27
27
6
3
1
6
2
6
3
1
6
2
6
3
1
5
2
6
17
2
5
5
17
2
5
6
17
2
5
StdErr
0.04
0.04
0.03
SEM+SD
0.12
0.30
0.19
0.41
0.02
0.18
0.13
0.30
0.10
0.11
0.09
0.29
0.01
0.01
0.10
0.17
0.06
0.01
0.02
0.20
0.05
0.03
0.01
0.19
Std Dev
0.214
0.186
0.168
SEM
0.10
0.24
0.15
0.20
0.02
0.13
0.11
0.21
0.08
0.08
0.07
0.20
0.01
0.01
0.07
0.13
0.05
0.01
0.01
0.17
0.03
0.02
0.00
0.19
G-51
-------
SEMcleancoal =,/(SEMhap2*SEMccf2) + (MEANhap2xSEMccf2) + (SEMhap2xMEANccf2) [G-2
The resulting average concentrations for the three HAPs in
"clean" coal (Indiana, Kentucky, Pennsylvania, Virginia, and West
Virginia) and the SEM for those concentrations are included in
Table G-12.
SBMcomposite=v/(%statelx5£Mstatel2)+(%state2xS£jtetate22)+(%state3x5£Mstate32) [G-3 ]
Once HAP concentrations had been determined for each State,
the next step was to estimate the average HAP concentrations in
the composite coal (the overall mixture of coal) burned at each
utility as well as some measure of the uncertainty related to
that average. In all cases, it was assumed that each unit within
a utility burned the same mixture of coal as was indicated by the
data reported in the UDI/EEI database. It is likely that the
mixture of coal and the relative contribution of any one state
used at a facility would change over time, depending on the
sulfur content, BTU heating value, cost, and availability.
However, no data were available to allow for a prediction of how
the mixture of coal would change with respect to the long-term
average. A weighted average for HAP concentration in the
composite coal was estimated for each utility using the data in
Table G-lOa and G-lOb. The SEM of the HAP concentration per
State was used to estimate the measure of uncertainty (SEM) in
the average HAP concentration in the composite coal using the
following formula (illustrated for plant using coal from three
States): This was carried out for each of the three HAPs for
each of the nine utilities being evaluated. These values and
associated SEMs were used as inputs for parameter TE in the
emissions estimation equation (Equation G-l).
G.3.1.4 Emissions Modification Factors. The EMFs were
developed and used to represent the amount of a HAP compound
entering and exiting a device (boiler or air pollution control
device) . The EMFs were based on the results of emissions testing
conducted by the Electric Power Research Institute (EPRI),
Department of Energy (DOE), EPA, and selected electric utilities.
Sampling was conducted at the boiler inlet, the boiler outlet,
and before and after the air pollution control devices. For each
G-52
-------
trace element in the feed, partitioning factors associated with
each boiler type were developed and applied. Control device
removal efficiencies were then applied to arrive at the final
emission factor (the amount of a HAP emitted divided by the
average amount of the same HAP entering the device). For each
test site, multiple inlet and outlet samples were taken, these
samples may or may not have been taken simultaneously. The EMF
for each test site was calculated as the average of the inlet HAP
amount divided by the average of the outlet measurement. It
should be noted that the ratio of the mean inlet and outlet
concentrations may differ from an average of EMFs calculated for
each data point. However, raw data was not available for all
test results to make a comparison of the possible errors
introduced by this approach. In some cases, individual test
values were reported while for others only average values were
reported. Therefore, it was not possible to evaluate the true
variability in the EMFs and the errors introduced through
sampling and analysis.
The values for any EMF are constrained to between zero and
one. A value of zero implies that the boiler and/or APCD is 100
percent efficient in removing the HAP from the exit flue gas and
containing it entirely in the bottom ash of the boiler or in the
APCD respectively. Values of one imply that the boiler and/or
APCD have no effect on HAP removal and all that enters is
emitted.
Basically, EMFs are fractions of the amount of a HAP exiting
a device (boiler or air pollution control device) divided by the
amount of the same HAP compound entering that device. These
factors are calculated by taking the geometric mean of similar
devices (e.g., oil-fired tangential boilers). Geometric means
for EMF values were used in the initial risk assessment based on
the limited number of samples for many of the EMFs. However, it
was later decided that the arithmetic mean may be a more
appropriate statistic. Therefore, for this uncertainty analysis,
the mean for each EMF was used.
The major uncertainties associated with the EMFs include:
• Limited testing may have occurred at facilities where the
device (boiler or APCD) was not operating in a manner
representative of those of the same type.
• Data on which the EMFs have been estimated were the result
of non-simultaneous sampling.
• Performance may vary significantly among similar APCDs.
G-53
-------
• Device performance may be affected by load changes, process
upsets, process modifications, environmental factors, age,
maintenance, or production details.
The uncertainty analysis was limited to existing data, which vary-
significantly among the various devices for which EMFs were
estimated. The uncertainty represented in the raw data was
assumed to be representative of the overall uncertainty in the
EMF. The data did not allow for a differentiation between
uncertainty and variability. The raw data for each device were
reviewed and the mean, geometric mean, and standard errors were
calculated for each HAP. For some devices (e.g., oil
tangentially fired dry without NOX controls) , only a single data
point was available. To estimate the uncertainty for those EMFs,
a triangular distribution was used. The single value was used as
the apex (most likely) and the distribution was bound between
0 and 1.
As described above, the EMFs are constrained to values
between zero and one. Using a normal distribution would allow
for negative values (implying that a HAP is removed from
surrounding air) and for values greater than one (implying that
the HAP is created in the device), neither of which is physically
possible. (Note: EMFs > 1 are possible for HAPs that are
produced during combustion, but this does not apply to the trace
elements.) The use of a log normal distribution prevents the
occurrence of negative EMFs, but still allows for values greater
than one. Therefore, the beta distribution was selected to
represent the uncertainty associated with the EMFs. The beta
distribution is commonly used to represent variability (or
uncertainty) over a fixed range, in this case between zero and
one. The value of the beta distribution lies in the variety of
shapes it can assume by varying two parameters, alpha (A) and
beta (B), which are explicitly defined by the available data,
specifically using the mean and SEM (or SD). Alpha and beta are
defined in terms of the mean and SEM as follows:
A-Jfeanx[Jtfeanx-i!l25S5L-l] B= (1-Mean) x [Mean* ^~Me^ -i]
SE2 SE2
If the two parameters are equal, the distribution is symmetrical.
If a < (3, the distribution is said to be positively skewed (most
of the values near the minimum) . If a > (3, the distribution is
said to be negatively skewed (most of the values near the
maximum). The EMFs (mean) and their associated uncertainty (SEM)
are presented in Table G-ll for oil-fired boilers and in Table
G-54
-------
G-12 for coal-fired boilers. These values and their associated
SEMs were used as inputs for parameter EMFa and EMFb in the
emissions estimation equation (Equation G-l) .
It should be noted that the beta distribution is defined by
the existing data and does not necessarily have foundations to
describe known processes. Therefore, an alternative approach was
also investigated assuming that EMF = e"m, and values of m were
estimated from the available data and then described as a
log-normal distribution. In general, the results were similar
but, in all cases, the beta distribution yielded a greater spread
in the range of possible values and uncertainty. Therefore, as a
conservative approach, the beta was used in this analysis.
As described in Section G.I, directly using the SE from the
sample population would tendency be an overconfident estimate of
(i.e. too narrow a range) of uncertainty due to:
• The SE is used as an estimate of the SD of a normal
distribution while the t-distribution is used to represent
the sampling distribution. For small sample sizes, the t-
distribution is significantly different, and approaches a
normal distribution as samples sizes increase (n~30).
Therefore, error is introduced by assuming a normal
distribution.
• The mean of a distribution represents a long-term average
only when the population samples are representative of
values that would be obtained over the long-term. Data
obtained from a convenience sample, additional error is
introduced due to potential nonrepresentativeness of the
sample.
• The mean of a distribution represent a long-term average
only if the population sample represents true variability.
Measurement error must also be considered, generally the
most appropriate estimate of uncertainty in the long-term
average is to use the larger of the SE or the standard
deviation due to measurement error.
An estimate in the uncertainty of the SE was made to account for
some of these errors and uncertainties, estimating the SD of the
expected SE repeatedly sampling from the overall population.
Using the method described in Hattis and Silver27 the following
steps were carried out:
• estimate mean and SD for the existing data set and assume to
describe the "true" distribution of the population.
G-55
-------
• randomly sample from a normal distribution with a "true"
mean and SD the same number of samples as in the original
data (repeated 2500 times).
• determine the standard error for each sample,
• calculate the SD of the sample SEs.
The measure of uncertainty used for these parameters was the
initial SE + the SD of the simulated SEs and are reported in
Tables G-ll and G-12.
G.3.1.5 Plant-Specific Emission Rates. Stochastic (Monte
Carlo) simulation was used to determine the uncertainty in the
risk estimates based on the uncertainty and variability within
each of the parameters used in the emissions model. Crystal Ball
(Decisioneering, Inc., Denver, CO) was used to conduct the
simulation. A probability distribution that best represents the
variable, its average value, and a measure of variability or
uncertainty about the average value was developed for each of the
parameters used in the model. The standard error of the mean
(SE) for each parameter was used as the estimate of uncertainty
rather than the standard deviation for each parameter. The SE is
a more appropriate statistic given that the risk assessment is
concerned with long-term average exposures and intermittent
extreme values of exposure levels are not as important.
After the assumptions about each variable in Equation G-l
are entered into the program, a three-step process is used to
generate a forecast chart containing a frequency distribution of
probable values for the calculation of interest (e.g., arsenic
emissions). The 3 steps are as follows:
1. Generate a random number for each variable based on its
type of probability distribution, average, and measure
of uncertainty.
2. Recalculate the spreadsheet using the randomly
generated numbers.
3. Enter the results into the forecast frequency chart.
These three steps were repeated a total of 3,000 times to produce
a frequency distribution of calculated HAP emissions values.
Running the simulation 3,000 times was arbitrarily selected based
on observations that the forecast curve was smooth. A quick
review of data from 2,000, 2,500, 3,000 and 4,000 simulations
G-56
-------
showed little change in resulting statistics and percentile
between 3,000 and 4,000 trials. It should be noted that a single
utility may consist of several individual units. To estimate
emissions from a given utility, each unit is modeled separately
and then the emissions are totaled across all units for each
simulation. The same value for trace element concentration in
fuel was used for each unit within a facility based on the
assumption that each unit would draw on an overall fuel supply
shared by all units within a facility. The remaining parameters
(FC, EMFb, and EMFa) were treated independently, which accounts
for the fact that fuel consumption and device performance (EMF)
are independent for individual units within a utility.
For illustration purposes, Figure G-7 summarizes the
distributions used for each of the parameter inputs for Plant
#29, the oil-fired plant that contributed most to cancer
incidence. Table G-13 summarizes the distribution of predicted
emissions from Plant 29 while Figure G-8 graphically presents the
distributions predicted for Plant #29. This distribution gives
some indication of the degree of uncertainty and the possible
range of emissions estimates that may be experienced. Figure G-9
graphically presents the cumulative probability distribution for
emissions from Plant 29 using the FCEM and SGS concentration
data. Furthermore, Figure G-9 and Table G-13 presents a
cumulative probability plot of the combined concentration data.
These distributions give some indication of the degree of
uncertainty and the possible range of emissions estimates which
may be experienced. The original emissions estimation program
had been designed to be an unbiased estimator of emissions. The
results of the uncertainty analysis tend to support this
assertion. The original baseline estimates ranged from the 22nd
percentile of the overall distribution (for arsenic using the SGS
oil concentration data) to the 95th percentile (for chromium
using the SGS oil concentration data). This is attributed to the
reanalysis of the two oil concentration (FCEM and SGS) data sets
using the probability plotting routine rather than treating non-
detects as half the detection limit used in the initial risk
assessment. When the combined data set (SGS and FCEM oil
concentration data) is considered the original baseline emissions
estimates were correspond to the 58th percentile for arsenic,
68th percentile for nickel, and the 80th percentile for chromium.
Furthermore, the 95th percentile of the simulated range of long-
term average emissions estimates are within a factor of two or
three of the mean and original baseline estimate and within a
factor of less than six within the median.
To compare how estimates differ for the coal-fired plants,
Plant #343 was selected for illustration purposes. Plant #343
G-57
-------
Figure G-7. Assumptions Used in the Modeling Uncertainty in Emissions from Plant #29
Assumption: oil consumption units 1,2
Normal distribution with parameters:
Mean 1,341.00 (=C21)
Standard Dev. 67.05 (=C27)
Selected range is from -Infinity to +lnfinity
Mean value in simulation was 1,340.90
Assumption: oil consumption units 3,4
Normal distribution with parameters:
Mean 1,547.80 (=D21)
Standard Dev. 77.39 (=D27)
Selected range is from -Infinity to -i-lnfinity
Mean value in simulation was 1,546.88
Assumption: As concentration in oil
Lognormal distribution with parameters:
Mean -3.32 (log)
Standard Dev. 1.80 (log)
Selected range is from -Infinity to +lnfinity
Mean value in simulation was 0.17
Assumption: Cr concentration in oil
Lognormal distribution with parameters:
Mean 0.28
Standard Dev. 0.20
Selected range is from 0.00 to +lnfinity
Mean value in simulation was 0.29
Assumption: Ni Concentration in oil
Lognormal distribution with parameters:
Mean 26.31
Standard Dev. 21.37
Selected range is from 0.00 to +lnfinity
Mean value in simulation was 25.93
Assumption: SGSCONCAs
Lognormal distribution with parameters:
Mean -0.24 (ln=2.3*log)
Standard Dev. 0.39 (ln=2.3*log)
Selected range is from -Infinity to +lnfinity
Mean value in simulation was 0.85
Assumption: SGSConcCr
Lognormal distribution with parameters:
Mean 5.74
Standard Dev. 10.68
Selected range is from 0.00 to +lnfinity
Mean value in simulation was 6.02
Assumption: SGSConcNi
Lognormal distribution with parameters:
Mean 36.03
Standard Dev. 21.53
Selected range is from 0.00 to ^Infinity
Mean value in simulation was 35.54
l_
G-58
-------
Figure G-7. Continued.
Assumption: frontdrynox As EMF
Beta distribution with parameters:
Alpha 1.69 (=C13)
Beta 1.17 (=C14)
Scale 1.00
Selected range is from 0.00 to 1.00
Mean value in simulation was 0.59
Assumption: Frontdrynox Cr EMF
Beta distribution with parameters:
Alpha 0.74 (=F13)
Beta 0.39 (=F14)
Scale 1.00
Selected range is from 0.00 to 1.00
Mean value in simulation was 0.66
Assumption: Frontdrynox Ni EMF
Beta distribution with parameters:
Alpha
Beta
Scale
Selected range is from 0.00 to 1.00
Mean value in simulation was 0.59
1.00 (=C13;
0.69 (=C14;
1.00
Assumption: Tangdrynonox Ni EMF
Triangular distribution with parameters:
Minimum 0.00
Likeliest 0.53 (=J83)
Maximum 1.00
Selected range is from 0.00 to 1.00
Mean value in simulation was 0.52
Assumption: tangdrynonx Cr EMF
Triangular distribution with parameters:
Minimum 0.00
Likeliest 1.00 (=D23)
Maximum 1.00
Selected range is from 0.00 to 1.00
Mean value in simulation was 0.67
Assumption: tangdrynonox Ni EMF
Triangular distribution with parameters:
Minimum 0.00
Likeliest 0.66 (=G53)
Maximum 1.00
Selected range is from 0.00 to 1.00
Mean value in simulation was 0.56
G-59
-------
.g
"+3
C\J
o .
— Q.
ca -j-,
O o
lit
16
CO
CO
O C^
f^ CO
C
c 2
£ '«
E .52
WL§
crj ^
CO
TJI
iombine
8
jo
•2 0
E col
IU
z
LU
O
LL
•o
CD
t"~
o
O
g
_o
8 col
col
'i wl
IU
o
1
II 1
1
1
Combined '
VI
1
«o
E co
IU
(0
^~
LU
o
1 1
CM ccT
S s.
CM O ^K
1- r» a>
CD T- m
00 00 ^*^"
CO .-^
CM CO
S fc-
in o"
•* S,
o o in"
CM CD O),
O) ^
CD CO,
? 1
n SI
•* eo1
•sr o>
^"^*
CB
*g
c
•s
LU
•j- 5"
"5 "**^
CL g
•— d)
S 2 g
1 1£
III
°§1
10 CM CM
§i|
CM CD in
O O i-
CM in c»
000
<«• - «
9 . .
0 ' '
CO CO 00
i- CO 1^
o •* oo
000
0 W 0
o CM in
CD
rs *"7*
c
o 2
P Q.
It
CM 00
535
m co
T- CO
83
SCO
Is.
i- CM
T- 1—
CM 00
CD 00
i- CO
O CO
CM CO
0 0
CM CO
00 S!
o> 5
co o)
T- CO
•^ >p
o in
T- CM
CM CM
CD CO
o> in
in o)
is
° °-
O> O)
in o
^- oo
CO O
CM CO
Is- O
00 CM
5LO
CO
«co
2T
05 CM
r>- o>
O> CO
T- CM
T- ,_
O) CO
o in
in i^
in o o
CM co in
CD o r-
•* O> it
•r- T- CM
CO O Tf
T- CM CM
CO T- CO
CO ^f CO
S- ^t" O)
CM h- CM
T- i- CM
CO O) 00
° 2 CM
CO -r- ^
•* N- ^
in ^t ^-
1- T- CM
CO CO ^
00 h^ ^
m T- r^
CO ^* ^*
•* fc CM
03 ^ CO
S5 S5 ^
o o in
o> o
g
§
o
in
CM
00
o
T—
in
in
CO
T—
CO
CO
§
^~
£
00
{i
^~
CM
b
d
o
''"
in
c\i
CO
c\i
CM
CO
CO
CO
CO
m
CM
O)
CM
S
1-
Tf
c
CO
CD
E
•2| x:
(T ^
CM CO
CO CM
en in
PJ CM
N> T™
co c\i
r^. CM
If) CM
T- O
CO T^
co in
CO CM
cr> oo
CO CM
r— T™
CM CO
"*. CO
*" T—
'
c g
co =
^ co
fi CO
E n
•£ £
LO K
en o>
.cr
e-
03
1
Q.
£
S
ii
Q.
>.
^
CD
•i
c
.0
to
^
1
•o
c
-2
2
^
|
"o
^5
*5i.
6
u
LU
o
LL
CO
•co £
>, CD
CO OT
C CD
„, CO
\D ^3
~ CO
o O
i co
.2 -o
T3 «=
CO TO
£ LU
•o P
Q U.
tS CD
^£1 4-*
8 "o
m —
"5. 15
S 2
E °"
2 S
tal analysis f
issuming eqi
Q) CD
E B
11
co" *-
11
"c i
§ E
D" / -\
Q) U
« II
3 ^
CO f2
II 3
co £
O o
CO O
52
To
- -i
c 1o
CD 0)
£ £
w .E
CO o
Q.
rn ^~
O3 ^
£ co c
|1|
0.2 o
!o^
CD £ C
'TI
"55 .c ?
co *= o
to c g-
O .-* W
r- CD
03 t- d
1 -2 o
c co o
T3 C 0
m CD ^
CO 0 3
3 C J3
CO R '^
c w co
te of emissio
the average
predicted di
1 §|
"S3 CD "o
HI
1— CO CD
ii > E
2 Q §.
CO ^ QJ
C ^0 f"
""ffi
III ^ II
c 1- "»"
"2 1
co y
S CD
£ i.
G-60
-------
Figure G-8. Summary of Results fo Monte Carlo Simulation of
HAP Emissions from Oil-Fired Plant#29
FCEM Concentration Data
SGS Concentration Data
Forecast: Plant 29 As Emissions
Frequency Chart
2,942 Trials Shown
us
200.00
kg/yr
4
40000
CellE12
.027 -r
Forecast: PI 29 As Emission (SGS)
Frequency Chart 2,941 Trials Shown
CellH11
.036 -
Forecast Plant 29 Cr Emissions
Frequency Chart
2,937 Trials Shown
105
CellH12
087 <
Forecast PI 29 Cr Emissions (SGS)
Frequency Chart 2,444 Trials Shown
CellK11
Forecast: Platn 29 Ni Emissions
Frequency Chart
2,930 Trials Shown
106
Cell K12
030 -r
Forecast: PI 29 NI Emissions (SGS)
Frequency Chart 2,940 Trials Shown
5,62500
11,250.00 16.B7500
kj/yr
G-61
-------
Figure G-9. Cumulative Probability of Uncertainty in Emissions from Plant #29:
Comparison of FCEM and SGS Oil Concentration Data
50.00 100.00 150.00 200.00 250.00 300.00 350.00 400.00 450.00 500.00
Plant 29: Arsenic Bniaeions(kg/yr)
93SConc. Data
FCBA Cone. Data
Combined
50.00
100.00 150.00
Chromium Bnisetons(kg/year)
200.00
250.00
5,000.00
10,000.00 15,000.00
Nickel Bntatons(kg/yeai)
20,000.00
25,000.00
G-62
-------
was the coal-fired plant that contributed most to cancer
incidence. Figure G-10 graphically presents, and Table G-14
summarizes the distribution for emissions estimates resulting
from the uncertainty analysis. The original point estimate of
emissions ranged from the 81st to 96th percentile of the range of
emissions predicted under the uncertainty analysis. This is
significantly higher and may be attributed to the fact that in
the original analysis coal from a single state was assumed
(defined by the state whose coal was consumed most). However,
for this plant the other coals used were generally of lower HAP
concentrations which would tend to drive the emissions estimate
lower than estimated during the baseline assessment. The 95th
percentile on the simulated range of emissions ranged from 0.9
times the baseline emissions estimate for nickel to 2.5 times the
baseline estimate for arsenic. As with Plant 29 the 95th
percentile of the overall range of predicted emissions typically
was within a factor of two or three of the mean emissions
estimate and the original baseline emissions estimate.
Additional analysis was also conducted on two additional
plants, Plant 133 and Plant 240, the highest oil and coal-fired
plants with the highest estimated MIR, respectively. The results
from this analysis are summarized in Table G-14. For Plant 133,
the estimate of emissions is influenced by which of the oil
concentration data sets are used. The original emissions
estimate used in the baseline assessment ranged from the 32nd
percentile for arsenic (using the SGS data) to the 94th
percentile for arsenic (using the FCEM data) and chromium (using
the SGS data). For Plant 240, the original emissions estimate
used in the baseline assessment was between the 82nd and 92nd
percentile on the range of predicted emissions.
G.3.2 Dispersion and Exposure
The HEM was used to conduct exposure modeling. The
Industrial Source Category- Long-Term (ISCLT2) air dispersion
model within HEM, was used to estimate long-term average ambient
air HAP concentrations within 50 km of each source. The exposure
modeling conducted for the baseline risk assessment is based on
the following default assumptions:
• Dispersion occurs as predicted by a Gaussian-plume model in
a rural setting in flat terrain.
• All exposures occur at one location; no adjustment is made
for exposure changes resulting from population movement
between home, school, work, etc.
G-63
-------
Figure G-10. Results of Monte Carlo Simulation of HAP Emissions (kg/year) for Coal-Fired Plant #343
CellE11
£97 •
223 .
£
15 .149 .
IB
.a
£ -074 -
Forecast: As Emissions: Plant #343
Frequency Chart 2,909 Tn
,
linn,.,, Lt™..:***0.!.
als Shown
- 865
648
m
n
*13? >0
S
ft
r 216 Q
0.00 750.00 1.500.00 2,250.00 3,000.00
kg/yr
CellF11
.025 -r
.000
0.00
Forecast: Cr Emission, PI 343
Frequency Chart
2,937 Trials Shown
74
55.5
18.5
CeMG11
.041 -r
.031 -. .
.021 - .
.010 ..
.000
o.oo
Forecast: Ni emissions, PI 343
Frequency Chart
Iliiiii .... i ..
Moan =113.87
112.50
225.00
ktfrear
2,941 Trials Shown
121
90.7
60.5
30-2
n
337.50
<
450.00
G-64
-------
Table G-14. Summary of Results of Monte Carlo Simulation of HAP Emissions from Coal-Fired plant #343
Forecast: Emissions: Plant #343
arsenic chromium nickel
Mean 552 778 114
Standard Deviation 901 540 116
Mean Std. Error 16 10 2
Original Point Estimate 969 1631 360
Percentile 81 92 96
Percentile
0.0% 0 10 0
2.5% 2 93 4
5.0% 5 140 7
10% 10 208 13
25% 25 387 33
50% 169 653 78
75% 688 1061 151
90% 1619 1500 262
95.0% 2462 1836 334
97.5% 3198 2163 430
100.0% 8490 4623 985
Ratio
95th: mean 4.5 2.4 2.9
95th: median 14.5 2.8 4.3
95th: baseline 2.5 1.1 0.9
Original Point Estimate = Estimate derived in the baseline assessment using EPA default values
Percentile = the percentile of the original point estimate within the Monte Carlo derived distribution
G-65
-------
• Exposure occurs at centroids of census blocks or at
user-specified location for which ambient concentrations are
estimated since the locations of actual residences are not
included in the database.
• People reside at the population centroids for their entire
lifetimes (assumed to be 70 years for cancer risk
estimation) and indoor concentrations are assumed to be the
same as outdoor concentrations. (Note: the concentration
at the centroid is assumed to represent a person's average
lifetime exposure allowing for activity patterns between
areas of higher and lower ambient concentration.)
• The power plant emits HAPs at the same level for the 70-year
lifetime of exposure.
• The only source of exposure is the stack emissions dispersed
in the ambient air; no resuspension of pollutants via dust
is considered.
• There is no population migration or growth.
• Varying exposures that might arise as a result of
differences such as age, sex, health status, and degree of
activity do not exist (although variations in these with
respect to issues of exposure-response are considered in
Section G.4).
Dispersion modeling and exposure calculation are complex
processes and nonlinear in nature. A detailed discussion of the
mathematical equations, models, and data can be found in Appendix
F and in several EPA reference documents. 10#owever, the
parameters and data used in the modeling can be described briefly
as follows:
• Plant-specific parameters: stack heights, stack diameter,
exit gas temperature, exit gas velocity, latitude and
longitude (obtained from the UDI/EEI database).
• Meteorological data: STability ARrays (or STAR data
summaries), joint frequencies of occurrence of windspeed,
wind direction, and air stability combining into a frequency
distribution (from the nearest National Weather Service
location).
• Population data: 1990 U.S. Census data, on the "block
level," containing 6.9 million records; for each plant, the
census blocks within 50 km were identified and assigned to
G-66
-------
be located at the centroid of the census block if within 0.5
km, or to user-specified receptors where ambient air
concentrations are modeled.
Some of the major uncertainties associated with the exposure
modeling include:
• Meteorological conditions at the nearest STAR site may
differ from the conditions near the utility site.
• The HEM assumes that exposure occurs at estimated ambient
air concentrations of the pollutant. However, it has been
shown that people spend a majority of their time indoors.
Indoor concentrations of ambient pollutants would likely be
less than those found outdoors. As a result, exposures
would be less.
• Gaussian dispersion models do not accurately predict
pollutant dispersion in areas of complex terrain (e.g.,
hills, valleys, mountains, over water), where deposition is
significant, or where transformation occurs, and might not
correctly predict where maximum concentrations will be
realized. The uncertainty in using a Gausian plume model
was not evaluated due to the lack of alternative verified
models and resources.
• Surface roughness is addressed in HEM as one of two model
defaults, urban or rural, using population density as a
proxy. The selection of urban or rural can dramatically
affect dispersion and concomitant exposure. Furthermore,
these two default modes may also not cover the possible
range of surface roughness which may be encountered.
• Exposure is also tied to emission rates and atmospheric
processes that affect pollutant concentrations in the
microenvironment where the person is exposed (e.g.,
infiltration of outside air indoors, transport by the
prevailing wind, diffusion/transport by atmospheric
turbulence, chemical and physical transformation,
deposition, and reentrainment), each of which has its own
underlying variability. For example, personal activities
and pollutant concentrations at a specific location might
change in response to outdoor temperature (e.g., open
windows).
G.3.2.1 Approach to Uncertainty Analysis. Air dispersion
modeling is complex and nonlinear and cannot be carried out with
the use of spreadsheets (hence, cannot be used within Crystal
G-67
-------
Ball). In addition, significant time is required to conduct the
modeling and process the data for each run. A stochastic (Monte
Carlo) approach requires a large number (thousands) of repetitive
runs (3,000 was used for the emissions estimates) to generate a
distribution. Given the time and resources required for single
runs, the Monte Carlo approach was not feasible and an
alternative approach was needed to evaluate the uncertainty in
dispersion and exposure modeling.
The degree of dispersion and resulting exposure is affected
by three major parameters: plant stack parameters (e.g., stack
height, stack gas temperature, and exit velocity), surface
roughness (urban versus rural), and meteorologic conditions. The
uncertainty analysis, therefore, focused on the three parameters.
Separate model runs were conducted for different combinations of
these parameters and uncertainty was estimated by examining the
range of resulting exposure estimates.
G.3.2.1.1 Plant Stack Parameters. The effective stack
height or plume rise above the physical stack height will impact
the degree of dispersion. In general, the greater the plume
rise, the greater the degree of dispersion and area impacted; the
lower the plume rise, the higher the maximum ground-level
concentration with a smaller impacted area. Plume rise is a
nonlinear relationship between exit gas velocity, stack diameter,
exit gas temperature, windspeed, and ambient temperature.
Simply, to represent the uncertainty related to effective stack
height, one could calculate representative high, medium, and low
values for effective stack height across all plants (e.g., +_ 10
percent). It should be noted that over time a much greater
degree of variation in exit velocity and stack gas temperature
would be expected related to changes in plant operation.
However, there were no data available to indicate how these
variations would impact the long-term averages experienced at a
specific plant. Ambient temperature and windspeed are the two
factors that have the greatest impact on plume rise and would be
a site-specific calculation. Exit gas temperature and velocity
may vary according to plant utilization. No data on variations
in plant operation is available. Therefore, the approach used
may underestimate actual uncertainty.
G.3.2.1.2 Surface Roughness. Surface roughness influences
the ground-level turbulence and dispersions and includes such
things as downwash. Because the predominance of physical
structures (e.g., buildings) is a measure of surface roughness,
the distinction is represented by using either the urban or rural
setting in HEM (Population density is used as surrogate for this
parameter, hence the reference to urban versus rural.) It should
G-68
-------
be noted that similar surface roughness can occur in entirely
natural settings based on terrain and vegetation. The urban and
rural mode of HEM approximates the bounds of surface roughness.
In actuality, surface roughness at a particular site may be
somewhere in between. To address this uncertainty, HEM was run
in both the urban and rural mode. It is not possible to
determine where a specific site may fit on the continuum of
surface roughness; therefore, the two modes are considered to be
•equally plausible.
G.3.2.1.3 Meteorology. Local meteorology has a great
impact on dispersion. Ambient temperature, stability class,
windspeed, and wind direction will all influence the degree and
direction of dispersion and subsequent exposure. Meteorology can
vary dramatically over a short distance and the nearest STAR site
may not be representative of, or even appropriate for, the local
meteorology in the vicinity of a utility. Prevailing or
intervening physical features (e.g., hills or mountains, large
bodies of water) may influence the prevailing wind direction and
precipitation and indicate that an alternative STAR site may be
more appropriate. Given the large number of utilities and STAR
sites, it is not possible to determine which of the STAR sites is
most appropriate, or the degree to which local meteorology may
differ. Therefore, a limited number of alternative STAR sites
were used to represent local meteorology.
The three factors being evaluated were nonlinear with
respect to each other and required a separate HEM run for each
parameter value. Therefore, a test matrix approach was used to
evaluate uncertainty in the exposure modeling component of the
exposure assessment. A limited number of options were developed
to represent the expected range of uncertainty for each of these
three categories of parameters as follows:
• Surface Roughness: Urban or rural mode
• Stack parameters: Represented as high (1.1 x UDI values),
medium (UDI values) and low (0.9 x UDI
values) estimates for stack gas
temperature and flue gas exit velocity
• Meteorology: Three closest meteorology locations in
the STAR database.
As a result, HEM runs were made for 18 different combinations of
dispersion parameters that may apply to a given plant. For the
purposes of this uncertainty analysis, it was assumed that there
was insufficient information to determine the relative
G-69
-------
correctness of each combination and, therefore, each was
considered equally likely to represent the possible range of
values.
G.3.2.2 Uncertainty in Exposure Modeling. For each of the
four plants, HEM was run using each combination of input
parameters (for a total of 18 runs) for a unit emission rate
(1 kg/yr). The maximum concentration, total exposure, and number
of people exposed at various concentrations were estimated for
each run. The coefficients for estimating maximum concentration
and total exposure (per 1-kg/yr emission) resulting from each of
these 18 HEM runs were summarized for each plant. For
illustration purposes, Figure G-ll summarizes the dispersion
coefficients for estimating the maximum concentration and total
exposure over the 18 different parameter combinations for Plant
#29. To estimate the maximum concentration or incidence, the
coefficient is multiplied by the actual emissions (kg/yr).
Maximum concentration is useful in estimating the MEI risk (Max
Cone x Unit Risk Estimate [URE] ) and the total exposure is used
to estimate cancer incidence (Exposure x URE/70 years).
The range of values of maximum concentration and exposure
over the 18 combinations can be a useful measure to evaluate
uncertainty. Furthermore, the uncertainty in each measure may be
affected independently with regard to each parameter combination.
For example, the maximum concentration may not change
significantly using data from one meteorological location,
however, the location where that concentration occurs and the
number of exposed persons (and the resulting incidence) may
change significantly due to changes in prevailing wind direction.
For specific locations, the HEM is believed to predict
concentrations to within a factor of two. However, for long-term
estimates of maximum location (without regard to location), HEM
predictions are believed to have a coefficient of variation of 10
to 40 percent, which is significantly less than the variation
between individual scenarios.
G.3.2.2.I Complex Terrain. The current version of the air
dispersion model contained in the HEM is the ISCLT2 version
92062, which was designed for sources located in relatively
simple or flat terrain where land elevations were below the top
of the stack. The EPA conducted a screening analysis on the
effect of complex terrain. Due to the recent availability of
elevation data and retrieval software, a screening analysis for
complex terrain around six coal- and six oil-fired higher-risk
utility plants was conducted.
G-70
-------
D
(guu/Bn) uoiiBJiueouoQ wnw!XB|/\|
(9-301 sewn)
in ^ co CM
ejnsodxg
c
Q)
'o
O
O
c
g
'en
Q_
£
!5
CO
n
o
£
0)
D
|
O
O
2
D
O)
LL
eo
C
0
O
*
O
0)
CL
CO
<
i_
0
C
D
G-71
-------
The model recommended by the Air Quality Modeling Group for
use in a screening analysis for analyzing the impact of complex
terrain was the Industrial Source Complex - Short Term - Draft
model (ISCSTDFT), version 94340. The advantage of this model was
that it contained the algorithms from the COMPLEXl model, which
is expressly designed for air dispersion when terrain features
are above the tops of the stacks. The model has three air
dispersion regimes. For those areas in the first regime (terrain
below the stack top), the model applies the ISCST algorithms with
elevated receptors. For those areas in the second regime
(terrain between the top of the stack and the effective plume
height), the model uses the larger of the concentrations
estimated by the ISCST and the COMPLEXl models. For the third
regime (terrain above the effective plume height), the model uses
the COMPLEXl algorithms only.
Unlike the ISCLT2, the ISCSTDFT air dispersion model
requires hourly meteorological observations for periods of up to
I year. For this study, the 1984 data were used for all plants.
As a preliminary study, one site was chosen and several years of
meteorological data were used. For this one site, the long-term
maximum concentrations (shown in Table G-15) for arsenic did not
vary much from year to year; therefore, the multiple-year
analysis was not conducted for other plants.
The ISCSTDFT model requires elevation estimates for each
point on the polar grid system. The algorithm used to estimate
the elevation of a polar grid point was to choose the maximum
elevation for any of the four points, from the USGS 3-arc-second
database, surrounding the polar grid point. At latitudes
contained in the continental U.S., 3 arc-seconds on a map is
roughly equivalent to 300 feet or 100 meters.
The ISCSTDFT model was set up to run elevation data for the
polar grid system. The regulatory default mode was used, with
the exception of the gradual plume rise option (see the sample
outputs in the attachments). The elevation data file was created
using a format that could be used directly in the air dispersion
model and this greatly simplified the procedures for setting up
the ISCSTDFT runs. However, source parameters, such as location
and stack data were entered "by hand." The model output, the
arsenic concentration profile, was then processed by a simple
FORTRAN program to rearrange the data into a 'format suitable for
exposure analysis. The rearranged arsenic concentration data
were then used to run the exposure model. Exposure calculations
were conducted using the exposure algorithms from the HEM.
G-72
-------
Table 6-15. Multiple Year ISCSTDFT Analysis
Year
1984
1985
1986
1987
Max. Cone. U/g/m3>
.00245
.00238
.00260
.00244
The results of the analysis are provided in Table G-16.
Although estimated maximum concentrations to which people were
being exposed increased for several of the higher risk plants,
the overall maximum individual risk for each source category
remained at about the same level. Also, the overall population
risk estimates, i.e., cancer cases per year per plant and cancer
cases per source category, remained at about the same level for
both source categories. Some of the estimated risks decreased
because of lower terrain features. Thus, the addition of complex
terrain to the analysis did not significantly change the risk
assessment results. The analysis was conducted for arsenic but
is representative of other toxic emissions.
G.3.3 Qualitative Discussion of Uncertainty
The results (distributions), derived in Sections G.3.1 and
G.3.2 represent only a portion of the uncertainty embodied in the
risk estimates. This uncertainty analysis focused only on the
parameter uncertainty within the models used in the baseline
analysis. Analysis was also limited to those parameters for
which data or information were available. Model uncertainty has
not been formally evaluated. Alternative models may yield
dramatically different results.
A preliminary evaluation of the emissions factor program was
conducted. Comparisons were made of test data from 19 utility
boiler stacks (17 coal-fired, 2 oil-fired) against predicted
emissions for the same plants using the emissions model. For
each facility, the emission estimate from the model was divided
by the reported value from the corresponding test report. A
value of 1 meant that the model exactly predicted the test
results, values lower than 1 indicated the model under-predicted
emissions, while values higher than 1 indicated the model
over-estimated emissions. In general, the results suggested
G-73
-------
TABLE G-16. RESULTS OF THE COMPLEX TERRAIN MODELING
AND EXPOSURE/RISK ANALYSIS
Source
People
HEM
ISCSTDFT
Max Risk
HEM
ISCSTDFT
Total People
HEM
ISCSTDFT
Annual Incidence
HEM
ISCSTDFT
Coal Plants
Plant A
Plant B
Plant C
Plant D
Plant E
Plant F
Oil Plants
Plant G
Plant H
Plant 1
Plant J
Plant K
Plant L
399
6
3
14
5
71
45
1
399
68
69
6
19
4
45
1
1 .94E-06
5.00E-07
3.61 E-07
1 .98E-07
1 .78E-07
1 .27E-07
1 .38E-05
1 .41 E-06
8.40E-07
5.62E-07
3.12E-07
2.42E-07
1.10E-06
4.00E-07
2.85E-06
9.26E-07
1.67E-07
1.17E-06
5.65E-06
6. 41 E-07
M6E-06*
9.01 E-07*
2.15E-7*
1.72E-07*
3,980,000
361,000
683,000
407,000
58,100
347,000
3,980,000
361,000
683,000
407,000
58,100
347,000
836,000
836,000
836,000
836,000
0.0024
0.0003
0.0007
0.0002
<.ooo
1
0.0002
0.0026
0.0003
0.0010
0.0003
<.0001
0.0004
0.0014
0.0002
0.0013
0.0002
* Maximum risk level based on maximum concentration from the air dispersion model with no regard to
actual exposure. Exposure modeling will produce a similar or lower MIR value.
that the model performs as expected (i.e., across a range of
boilers and constituents to estimate overall emissions). The
average of the ratios across all stacks and constituents was
1.01, while averages for the three HAPs are 1.4, 0.7, and 0.9 for
arsenic, chromium, and nickel, respectively.
However, while the model did well in predicting overall or
average emissions across a range of utility boilers, large
differences between predicted and reported values are found for
individual boilers and constituents. The largest difference for
an individual boiler was were estimated emissions were about
5,000 times lower than reported test results. Furthermore, the
model tended to underestimate rather than overestimate emissions
about 60 percent of the time within this sample of boilers. A
preliminary evaluation of facilities with large differences
between projected and actual emissions found that these
facilities were likely to burn multiple fuel types. Petroleum
G-74
-------
coke was found to be burned at the facility with the greatest
difference. Therefore, while the model appears to predict fairly
well, emissions may be over- or under-estimated at individual
facilities. This difference may be due to the combinations of
fuel being burned since the model assumes a single fuel type.
The plants in which the model underestimated significantly were
some of the lowest emitting plants predicted by the model, and
using measured rather than modeled emissions would bring their
emissions more in line with the rest of the plants, but would not
become high-risk plants when compared to others. However, it
should also be noted that the emissions test data on which these
comparisons were made were from short-term (often
nonsimultaneous) sampling. Whether these test data reflect
actual long-term average emissions from these stacks are not
known.
Likewise, significant uncertainty may also be associated
with the use of HEM to conduct dispersion and exposure modeling.
The sensitivity analysis (described in Section G.5 below)
indicated that the dispersion coefficient is a significant
parameter. Therefore, it is likely that alternative models may
yield dramatically different results, and that plant-specific
estimates have the potential to vary from that estimated. An
evaluation of model uncertainty of both the emissions factor
program and HEM is indicated given their relative contribution to
overall parameter uncertainty. Some studies indicated that the
estimates from dispersion models may have a coefficient of
variation of +_ 10 to 40 percent in predicting short-term
concentrations or identifying the location of maximum
concentration. It is believed that with the models have greater
precision in estimating long-term concentrations and exposures.
A preliminary analysis was attempted to adjust the dispersion
coefficients accordingly. However, the spread in the dispersion
coefficients for each of the 18 scenarios was overwhelmed by the
variation in the dispersion coefficients between the scenarios.
G.4 EVALUATION OF UNCERTAINTY AND VARIABILITY ANALYSIS IN THE
EXPOSURE-RESPONSE ASSESSMENT FOR SELECTED HAPs
This section presents uncertainty and variability analyses
for the exposure-response component of the calculation of risks
from electric power generation. Uncertainty and variability of
exposures were reviewed in previous sections. The goal of the
present section is to estimate the variability of the
quantitative relationship between exposure and the excess
probability of cancer for different humans exposed to releases
from electric power generation in the U.S.; and the uncertainty
G-75
-------
in the average or mean quantitative relationship between exposure
and the excess probability of cancer.
There are two basic questions addressed in this section:
• In a group of people, all exposed to the same concentration
of a particular material in air as a result of electric
power generation (and as provided in earlier sections of
this report), what would we expect to be the variation of
the excess probability of adverse health effects resulting
from this exposure? Would everyone have the same probability
(this is unlikely) ? Would this probability of effect vary
between exposed people due to differences such as the rates
at which different people breathe, the length of time
different people are exposed, and so on? If it does vary
(which is likely), what is the cause of this variability and
how can we describe the variation between people? The
answers to these questions constitute the variability
analysis in this section. The reason for performing this
analysis is to examine the degree of risk equity within
groups that are exposed to the same concentration of
materials in air; to identify sensitive subpopulations of
people with risks larger than the average for the
population; and to provide this information to risk managers
for consideration in decisions. Of particular interest is
the likely magnitude of the risk to the individual in the
population with the maximal risk. The exposure assessment
identified the maximally exposed individual (or MEI);
assessment of the variability in the exposure-response
relationship provides information on the variability of
response within a subpopulation of individuals located at
the point of the MEI.
• In a group of people, all exposed to the same concentration
of a particular material in air as a result of electric
power generation, are we certain about the average (mean)
excess probability of adverse health effects resulting in
this population from this exposure (this is unlikely) ? If we
are not certain (which is likely), how can we express this
uncertainty? Can we give an upper and a lower bound on where
the true value of the average probability of effect is
likely to be found? What are these values? Can we give a
distribution of estimates of the average probability, with
this distribution describing how confident we are that the
true value of the average probability is any particular
number (such as 10'€ or 1Q-5)? What is this distribution? The
answers to these questions constitute the uncertainty
G-76
-------
analysis in this section. The reason for performing this
analysis is to examine the confidence with which we can
assert that the average probability of effect in a
population for which the average exposure has already been
estimated is less than any particular number.
There also are compound questions that can be addressed once
answers have been supplied to the above two questions. These two
answers can be combined in risk characterization (along with the
assessment of variability and uncertainty in exposure from
previous sections of this report) to provide answers to the
questions:
• How certain are we about the amount of variability in a
population?
• How certain are we about the number of people whose excess
probability of effect exceeds some particular value (such as
ID'6 or 1Q-5) ?
Attention is restricted here to three HAPs which constitute
in excess of 90 percent of the inhalation cancer risks from
utilities in the U.S., arsenic, nickel and chromium compounds. It
is assumed that the variability and uncertainty characterizing
these three HAPs approximates the variability and uncertainty
characterizing the entire mixture of HAPs released from
utilities. This is a reasonable assumption given the degree to
which these three HAPs contribute to the cancer risks. Still, it
must be recognized that this uncertainty and variability analysis
considers only the inhalation exposure pathway. It is likely
that the indirect pathways of exposure would increase the mean
value of the risk to the exposed population (perhaps also
affecting the distributional characteristics, such as the SDs).
The analyses reported in this section represent a lower limit on
both the mean probability of effect and on the upper end of the
confidence interval for this probability. In other words, both
the mean and the upper limit on the confidence interval are
likely to increase when the indirect exposure pathways (e.g.,
ingestion) are considered.
In addressing uncertainty and variability for any quantity
such as the excess probability of cancer, there are two broad
approaches typically used. In the first, the analysis generates
an estimate of the average value in the population by using the
average values for all parameters needed to calculate the risk;
an estimate of the highest value by using the highest values for
all parameters needed to calculate the risk; and an estimate of
the lowest value by using the lowest values for all parameters
G-77
-------
needed to calculate the risk. For variability analysis, the
"highest value" means the highest value actually found in the
population (such as the highest probability of cancer); the
"lowest value" means the lowest value actually found in that same
population (such as the lowest probability of cancer). The
variability analysis then is summarized by the average
probability of cancer in the population; the highest probability
of cancer in that population (probably in sensitive individuals);
and the lowest probability of cancer in that same population
(probably in insensitive individuals).
For uncertainty analysis, the "highest value" means the
largest estimate of the mean probability of cancer which is
consistent with existing information (in all cases in this
report, the mean is taken also to be the "best estimate"); the
"lowest value" means the smallest estimate of the mean
probability of cancer which is consistent with existing
information. The uncertainty analysis then is summarized by a
"best estimate" of the average probability of cancer in the
population (using the best estimate for each of the parameters
needed to calculate this probability); an "upper bound" estimate
of the average probability of cancer for which it can be said
that it is very likely that the true value of the average
probability of cancer is greater than this "upper bound" (using
the upper bound estimate for each parameter); and a "lower bound"
estimate for which it can be said that it is very unlikely that
the true value of the mean probability of cancer is less than
this "lower bound" (using the lower bound estimate for each
parameter).
The problem with this first approach is that it supplies
information only at three numerical values (the average, the
"upper bound" and the "lower bound"), and it usually is not
possible to state quantitatively the level of frequency (for
variability analysis) or confidence (for uncertainty analysis)
associated with the upper and lower bound estimates. A second
approach that provides more information is to produce a
distribution of values for each parameter used in calculations,
each with an assigned level of probability or confidence. The
result is a probability density function for the probability of
cancer in the exposed population to which the risk manager may
turn in asking questions such as:
• What fraction of people have an excess probability of cancer
above ID'6? Above ID'5? (a question of variability) .
G-78
-------
• With what confidence can we assert that the average
probability of cancer in a population is 10~6?
10'5? (a question of uncertainty) .
While this second approach requires more judgment (distributions
must be selected, rather than simply upper and lower bounds), it
provides more complete and useful information to the risk
manager. It is for this reason that the analysis in this section
focuses primarily on the distributional approach, although the
first approach also is examined at selected points.
It is important to note that estimates of the probability of
effect require information on a number of parameters describing
the biological properties of exposed people as well as the
chemical and physical characteristics of the arsenic, nickel and
chromium in the environment. Where appropriate, each of these
components of variability and uncertainty (e.g., breathing rates,
retention halftimes, etc.) are described here as quantitative
probability density functions reflecting either the frequency
(for variability analysis) or the degree of confidence (for
uncertainty analysis) associated with any particular numerical
value. This allows the uncertainty and/or variability of the
separate components to be combined mathematically (here, through
Monte Carlo techniques) in estimating the uncertainty and/or
variability of the probability of cancer.
In some instances, however, the components are described
here qualitatively. This is true primarily for the identification
of sensitive subpopulations on the basis of biological
characteristics, for the treatment of non-cancer endpoints, and
for uncertainties introduced by the existence of different model
forms for dose-response extrapolation (which is treated here
semi-quantitatively). In each case, an indication is given of
the potential impact of these qualitative or semi-quantitative
analyses on the final estimates of the probability of cancer.
The important characteristics of risk for which uncertainty
and variability of exposure-response assessment are performed in
this report include:
1. Uncertainty in the Mean of the MIR. In this section,
it is assumed that the highest concentration of the
material in air has been identified for each source.
This concentration is shown as C^^. It is assumed that
a population is exposed to this concentration over a
normal lifetime. There is uncertainty about the mean
numerical value for each of the biological and
environmental factors (described below) which must be
G-79
-------
specified to convert C^ to an estimate of the
probability of cancer. The question being addressed is:
How confident are we that the mean excess probability
of cancer in a population uniformly exposed at a
concentration of C^ is any particular value (such as
10-6, 10-5, etc)?
2. Variability in the Mir for Different Individuals in a
Population Located at the Point Where the Concentration
Equals C,^. It is assumed that a population is exposed
to this concentration over a lifetime. There is
variability between people in the numerical value for
each of the biological and environmental factors
(described below) which must be specified to convert
Cnax to an estimate of the probability of cancer. As a
result, there is variability in the probability of
cancer for individuals within a population all exposed
at the concentration CB^. The question being addressed
is: Within a population uniformly exposed at a
concentration of C^ , what fraction of the people have
an excess probability of cancer equal to any particular
value (such as 10~6) ?
3. Uncertainty in the Estimate of Cancer Incidence for the
Entire U.S. Population When the Average Concentration
jLs_Cave. It is assumed that the U.S. population is
exposed to this concentration over a lifetime. There is
uncertainty about the mean value for each of the
biological and environmental factors (described below)
which must be described to convert Cave to an estimate
of the probability of cancer. The question being
addressed is: How confident are we that the excess
number of cancers in a population of the size of the
U.S., and exposed at a concentration of Cave , is any
particular value (such as 1, 0.1, etc}?
Calculating either probability of cancer or number of
cancers requires a range of assumptions (usually given as
parameter values for an equation) concerning the relationship
between exposure and response, as discussed in later sections.
The analyses of variability and uncertainty for these assumptions
are divided here into two groups of parameter values. The first
contains analysis of uncertainty and variability for parameter
values that are common across the three HAPs considered (arsenic,
nickel and chromium). The statistical properties of the
distributions describing these parameter values are the same for
all three compounds. If a specific individual is above the
population average for the given parameter value, this will hold
G-80
-------
true equally for all three compounds. An example of this in
variability analysis is breathing rate (liters of air per
minute), where if an individual has a breathing rate higher than
the average for chromium, he or she also will have a higher
breathing rate for nickel. An example in uncertainty analysis is
that an underestimate of the average breathing rate for the
population (perhaps due to sampling error) will produce an
underestimate of the cancer risk for all three HAPs.
The second group contains analysis of uncertainty or
variability for parameter values that differ between the three
HAPs. The statistical properties of the distributions describing
these parameter values are not necessarily the same for the three
compounds. An example of this in variability analysis is the CSF.
The degree of variability in the CSF for nickel differs from that
for chromium due to differences in the data base. An example in
uncertainty analysis is that the measure of uncertainty (such as
the SD) for the CSF may differ between the three HAPs. If the
estimate of the CSF for a HAPs such as nickel (for example) is
larger than the true value for nickel due to sampling error, this
is not inconsistent with a finding that the estimate of the CSF
for chromium is smaller than the true value for chromium.
This distinction between uniform and compound-specific
components of variability and/or uncertainty is treated here by
subdividing exposure-response assessment into three separate
categories: exposure times, pharmacodynamics and dose-response.
Exposure times express the fraction of time an individual is
indoors or outdoors (reflected by the mean concentration to which
they are exposed, taking into account potential differences in
the indoor and outdoor concentrations) and the length of
residence at the point of exposure. The uncertainty and/or
variability of exposure is the same for the three compounds.
Pharmacodynamics is taken here to include all
considerations from exposure to target dose. It is divided
conceptually into considerations related to lung intake (here,
the volume of air breathed per unit time); lung uptake (the
amount of the inhaled material deposited in the lung after
intake, affected primarily by deposition fractions); burden (the
equilibrium amount of the inhaled material in organs, affected
primarily by retention half-times); biologically significant
burden (the equilibrium amount of the biologically active form of
the compound, affected primarily by bioactivation and/or de-
activation fractions); and dose (either the amount of interaction
between the biologically active form and target structures such
as DNA, or the integral of the biologically significant burden
over the period of exposure, often called the "area under the
G-81
-------
curve" or AUTC). For the three compounds considered here,
variability for any particular pharmacodynamic property is the
same for the three compounds either because the properties are
identical for the three compounds (such as in the case of the
volume of air breathed per unit time) or because there is
insufficient information to distinguish between the compounds
(such as variability and uncertainty in retention half-times).
The same is true for analysis of uncertainty for these parameter
values. This means the distribution describing uncertainty
introduced by any particular assumption such as breathing rate,
deposition fractions, retention half-times, biotransformation and
dosimetry will be characterized identically for the three
compounds. The distributions differ, of course, between different
parameters.
The third component of exposure-response assessment,
dose-response, is the relationship between dose and excess
probability of cancer. Here, uncertainty is different for the
three HAPs due to differences in the quality of epidemiologic
data on which the CSF are estimated and uncertainty as to the
assumption of speciation. The statistical properties of these
three distributions, therefore, are different. There is little
quantitative information on which to base an analysis of
variability of the CSF, other than to note the effect of smoking
(which is considered in the analyses which follow). This issue
is treated more qualitatively through a discussion of the
existence of sensitive subpopulations.
Dose-response also differs in degree, if not in kind, of
uncertainty due to the need to extrapolate from high to low
exposures using models that are only partially established
scientifically. The issue of how to quantify uncertainty in
dose-response relationships when competing biologically plausible
models is one of the most highly debated areas in uncertainty
analysis. This issue will be discussed semi-quantitatively in
this report. The approach will be to first perform an analysis of
uncertainty conditional upon adoption of the dose-response models
employed in the base-line EPA risk calculations, and then to
consider the potential impact of alternative scientifically-based
extrapolation models.
For non-cancer endpoints, expressed in the analysis as
Hazard Quotients (HQ), the treatment of uncertainty and
variability is primarily qualitative. In the baseline analysis of
non-cancer risks from the inhalation exposure pathway, the HQs
for arsenic, nickel and chromium have not been estimated due to a
lack of RfCs for these HAPs.
G-82
-------
The highest inhalation HQ for any HAP analyzed was 0.12 for
HCl. Therefore an uncertainty analysis was not performed for
inhalation noncancer effects.
It must be noted, however, that the exposure assessment used
in this analysis is based on an assumption of continuous exposure
at a uniform ambient concentration. In real situations of
exposure, the concentrations of all HAPs will fluctuate in time,
rising through high values above the average and falling to low
values below the average. During the excursions to high
concentrations, such as might occur soon after accidental
releases from a facility, the concentrations in air might exceed
the time-weighted averages significantly. At the present time,
however, the analysts do not have information on the distribution
of concentrations in time at a single point located near a
facility. Future research should focus on characterizing this
source of temporal variability and its influence on the frequency
of non-cancer effects in the exposed population. The same can be
said for the contribution from indirect exposure pathways (which
also could increase the average daily rate of intake.
A related issue is the presence of background levels of
several of the HAPs (such as cadmium) in the tissues of exposed
individuals. Where these background levels are close to the NOELs
and LOELs, a finding that the utilities do not produce
concentrations at the RfC does not ensure that non-cancer effects
will fail to appear. Concentrations below the RfC could produce
the effect if the background levels already are high. Future
research should focus on characterizing these background levels
and assessing their influence on the frequency of non-cancer
effects in the exposed population.
G.4.1 The Quantitative Depiction of Uncertainty and Variability
Both uncertainty and variability analyses begin by
specifying distributions for each of the parameter values
appearing in equations used to estimate the probability of an
effect such as cancer (the relevant equation is discussed in the
following section). For uncertainty analysis, the distribution
expresses the probability that the true value for the parameter
(usually taken to imply the true mean for some population) is
equal to any particular numerical value between a lower bound and
an upper bound. For variability analysis, the distribution
expresses the frequency with which the parameter has any
particular numerical value within individuals in a population for
which the risk is being estimated. This frequency is the same as
the fraction of individuals in the population whose parameter
value equals a particular numerical value.
G-83
-------
A key problem, both conceptually and methodologically, is
how these distributions should be developed. Ideally, the
distributions for variability should be based on measurements of
the variation of the parameter between individuals in a well
specified population. Both biological and environmental
parameters usually are found to be distributed lognormally,
meaning the logarithms of the parameter values for individuals in
the population are distributed normally. The lognormal
distribution is characterized by a median, which is the numerical
value of the parameter for which it may be said that 50 percent
of the individuals have a parameter value which is larger and 50
percent have a value which is smaller; and by a geometric
standard deviation (or GSD), which is a quantitative measure of
the degree of variability. In the population, 68 percent of the
numerical values are between the median divided by the GSD (the
"lower bound") and the median times the GSD (the "upper bound").
In the same population, 95 percent of the numerical values are
between the median divided by the square of the GSD; and the
median times the square of the GSD.
Often, the GSD can be estimated directly from the available
data by fitting the data to a lognormal distribution. In some
cases, however, part of the measured variability is due to
measurement error (properly included in uncertainty analysis)
rather than variability. In such cases, the GSD of the measured
distribution for variability will be larger than the true GSD. An
attempt has been made in the present section to separate these
two contributions to measurements of variability, primarily by
selecting data in which the uncertainty in individual
measurements is small compared to the standard deviation of the
intersubject variability. In cases where this does not hold, the
most complete approach would be to (I) estimate the uncertainty
in individual measurements and (ii) de-convolve the true
variability from the measured variability. That approach could
not be used here since in essentially all cases of parameter
values, the analysis employs ratios of parameter values in the
general U.S. population over those in the epidemiological
populations from which slope factors were obtained. As a result,
subjective judgment was used to narrow the distribution of
variability after accounting for the uncertainty in individual
measurements. Throughout the present section on-
exposure-response, variability will be characterized by a
lognormal distribution. The GSD is calculated from fitting of the
data to a lognormal distribution, or from the procedure described
above. The reasoning behind selection of the median and GSD are
explained for each parameter value.
G-84
-------
The situation is more complex (and contentious) for
instances of uncertainty analysis. The reason is that uncertainty
has several sources2 with varying degrees of susceptibility to
mathematical and statistical analysis. Measurement errors
typically can be summarized by techniques of statistics, often
displaying statistical properties characteristic of the normal
distribution when only precision is considered. If
representativeness of a sample is considered, however, it is
equally valid to use a lognormal distribution since the mean
values of a parameter in subpopulations usually is distributed
lognormally across subpopulations. In the present analysis, a
normal distribution is used when issues of precision dominate,
and a lognormal distribution is used when issues of
representativeness dominate.
The characteristics of these distributions can at times be
estimated directly from the available data when interpreted
through statistical procedures (such as calculation of the
standard error of the mean based on knowledge of the standard
deviation of the population of samples, which is the method used
in much of the analysis of uncertainty in this report) . In other
instances, however, the data either are insufficient to determine
the standard deviation (or GSD) by such techniques or contain
sources of uncertainty not captured by standard statistical
techniques. An example of the latter consideration is study
design, where there may be a question as to whether the study was
conducted properly, accounted for confounding influences, etc. In
these instances, analysts often employ more subjective judgments
of the distribution's properties. While the selection of these
properties depends to some degree on examination of the available
data and on past experience (and, hence, is partially objective),
it also involves a judgment about factors such as study design,
reliability of reporting and conceptual ambiguity which are not
purely statistical issues and must be rooted in personal
judgments. It is important to note that this subjective element
is a part of all uncertainty analysis and does not appear only in
instances where data are scarce. Such subjective judgments are
noted in the present analysis when they occur. A reason is given
for the judgment, although it is recognized that rational
individuals may differ in their judgments.
Finally, it should be noted that throughout this analysis
some parameter values have been given quantitative
characterizations of uncertainty and/or variability, while others
are discussed qualitatively. The Agency recognizes that it is
common practice in risk analysis to attempt to quantify all
sources of variability and uncertainty, even where data are
insufficient to provide a statistical basis for the development
G-85
-------
of distributions. In these cases, subjective measures of
variability and uncertainty typically have been used.2 While
such approaches could be used in the current analysis, the Agency
has concern over the justification for subjective assignments. In
many cases in this analysis, therefore, it has chosen to avoid
these assignments by relying on qualitative discussions of
uncertainty and variability for parameters where data are
insufficient to establish distributions statistically. Reviewers
of this preliminary document should not, therefore, rely solely
on the quantitative aspects of the analysis provided in the
various figures and tables, but should include consideration of
the discussions providing qualitative characterizations of
important sources of uncertainty and variability.
G.4.2 The Defining Exposure-response Relationship
Regardless of whether one is calculating the Maximum
Individual Risk, the average lifetime risk for the U.S.
population, the number of cancers, or the fraction of individuals
with a probability of cancer above a specified numerical value,
the equation relating exposure and the probability of that
exposure producing cancer is:
LPC = C x SF x EFCF x EDCF x BRCF x DFCF x TCF x
TFCF X DCF X SFCF x SCF [G-4]
where
LPC = the excess lifetime probability that an individual
develops cancer as a result of exposure to emissions
of a compound; the values on the right-hand side of
this equation must be the values for the individual
or group being considered; if the values on the
right-hand side are the average for a population,
then LPC is the fraction of people in the population
developing cancer as a result of the exposure; if
the values on the right-hand side are for a MEI,
then LPC is the probability of cancer for this MEI;
C = the outdoor concentration (mg/m3) to which this
individual or group is exposed; this is specified in
the exposure assessment component of this analysis;
SF = the slope factor for this individual or group in
units of (mg/m3)"1; this slope factor does not
account for differences in the relationship between
exposure and response that might exist in
extrapolating from occupational study populations
from which the slope factors originally were
G-86
-------
developed (adults under conditions of activity) to
the general U.S. population of interest in this
report (including all age groups and levels of
activity); factors affecting the relationship
between exposure and response are accounted for in
the remaining terms below;
EFCF = correction factor for the fact that a part of each
day is spent indoors rather than outdoors; it is
equal to the sum of (I) the fraction of time spent
outdoors multiplied by the concentration outdoors
and (ii) the fraction of time spent indoors
multiplied by the concentration indoors, with this
sum then divided by the concentration outdoors;
EDCF = correction factor for the exposure duration for this
individual or group, taken to be the ratio of the
actual number of years of residence to the number of
years assumed in the base-line EPA calculations (70
years);
BRCF = correction factor for the breathing rate, taken as
the ratio of the breathing rate in the individual or
group (averaged over a 70 year lifetime with age-
dependent levels of activity) to that in the
population from which the slope factor was estimated
(adults engaged primarily in light to heavy
activity); there is no need for such a correction
factor in instances where the development of the
slope factor by the Agency already includes the
necessary adjustment;
DFCF = correction factor for the lung deposition fractions,
taken as the ratio of the deposition fraction in the
individual or group to that in the population from
which the slope factor was estimated; there is no
need for such a correction factor in instances where
the development of the slope factor by the Agency
already includes the necessary adjustment;
TCF = correction factor for retention half-times, taken as
the ratio of the retention half-time in the
individual or group to that in the population from
which the slope factor was estimated; only single-
exponential retention functions are considered here;
there is no need for such a correction factor in
instances where the development of the slope factor
G-87
-------
by the Agency already includes the necessary
adjustment;
TFCF = "transformation fraction" correction factor for the
conversion (activation or detoxification) from
burden to biologically significant burden; since the
relevant biotransformation information is not
available for the three compounds, this is not
considered in the following analysis;
DCF = "dosimetry" correction factor for the conversion
from biologically significant burden to dose-rate;
since the relevant dosimetric information is not
available for the three compounds, this is not
considered in the following analysis;
SFCF = correction factor for the CSF, taken as the ratio of
the CSF for the individual or group to that used in
the baseline risk assessment; this presumes the same
speciation as in the epidemiological studies used to
generate the CSF initially;
SCF = correction factor for speciation, taken as the ratio
of the weighted average of the CSFs for the
different species (weighted by fraction of the
concentration contributed by each species) in the
individual or group 'to the weighted average in the
epidemiological studies used to generate the CSF
initially.
It is important to note that in all cases, the correction factors
above are in the form of ratios between the general U.S.
population (for which inferences are being drawn in the risk
assessment) and the populations from which the CSFs were drawn
(usually, epidemiological populations characterized by adult
ages, relatively healthy individuals, and physiological
parameters characteristic of at least light activity rather than
rest). It is the uncertainty and variability of these ratios,
and not of individual-specific values, that is addressed in the
following sections.
G.4.3 Variability in the Exposure-response Relationship
G.4.3.1 Variability of Compound-independent Values in
Equation G-4. The values necessary for Equation G-4 that are
assumed not to be a function of the compound being considered are
EFCF, EDCF, BRcF, and DFCF.
G-88
-------
Exposure Frequency (EFCF) : The exposure frequency assumed in the
base-line calculation of the probability of cancer was 1.0,
interpreted as an assumption that (I) indoor and outdoor
concentrations are equal and (ii) that an individual spends all
of his or her time within the assigned exposure grid-block during
each year. The value of EFCF is the "weighted exposure
frequency". This is found by asking the following question: if
the individual spends a fraction of time indoors, where the
concentration is equal to the indoor concentration, and a
fraction of time outdoors, where the concentration is that used
in the base-line calculation, what is the equivalent fraction of
time the individual is exposed to a concentration only equal to
the outdoor concentration?
The equation for calculating EFCF is:
EFCF = fi x Ci + fc x C0 [G-5]
For example, if the indoor concentration is 0.2 and the outdoor
concentration is 1, and if the fraction of time indoors is 0.5
(with a fraction outdoors equal to 0.5), the "weighted exposure
frequency" is 0.5 x 0.2 + 0.5 x 1 = 0.6. This means the actual
exposure pattern (50 percent time indoors and 50 percent
outdoors) produces the same level of exposure as a pattern in
which 60 percent of each day was spent at the outdoor
concentration (and the remainder at 0 concentration).
The median value for the ratio of the concentration indoors
to that outdoors is 0.4 (EPA, 1989). This ratio will be nearly
1.0 for houses with very large rates of ventilation (so long as
ventilation is not dominated by infiltration through cracks,
which might act as filters for the arsenic, nickel and chromium
particles); and significantly less than 0.4 for homes with very
low rates of ventilation. It is assumed here that the upper 95
percent confidence limit is approximately 0.8. Since the median
is 0.4, the ratio of the upper limit to the median is 2.0. The
GSD then is equal to the square root of 2.0, which is 1.4.
Variability of the ratio of indoor to outdoor concentrations is
characterized here by a lognormal distribution with a 95 percent
confidence interval of [0.2, 0.8], with a median concentration
ratio of 0.4 and a geometric standard deviation of 1.4. This
distribution is truncated at 0.0 and 1.0 since the ratio of
concentrations cannot lie outside these bounds.
The ratio of concentrations indoors to outdoors must be
supplemented by an estimate* of the fraction of time spent indoors
G-89
-------
and outdoors, taken here to be given by the fractions fz
(fraction of time indoors) and f0 (fraction of time outdoors) .
These fractions depend on age and occupation, but the fraction of
time outdoors (f0) typically averages approximately 0.15 and the
fraction indoors (fi) 0.85.15 The median value of the "weighted
exposure frequency" is approximately (1.Ox 0.15) + (0.4x0.85)
or 0.49, which is interpreted as 11.8 hours per day ( 0.49 x 24)
exposed to the outdoor concentration.
Information on the variability of fx and f0 is not
available, but the fraction of time indoors cannot exceed 1.0 and
probably rarely is below 0.5 (allowing for sleeping and normal
household activities). This sets an upper bound on fx of 1.0 and
a lower bound of 0.5. The values of fz and f0 for individuals
clearly are completely correlated, since they must sum to 1.0 in
all instances. It is assumed here, therefore, that fx is
distributed lognormally with a geometric standard deviation of
1.2 truncated at the boundaries of the interval [0.5, 1] . The
value of f0 always is I -fI. The final "weighted" exposure
frequency then is distributed lognormally with a median of 0.49,
mean of 0.52 and a geometric standard deviation of approximately
1.4. The distribution is truncated at [0, 1].
The above distribution is representative of a first scenario
in which the actual conditions in US homes is considered. As a
matter of policy, however, the analysts also have chosen to
examine the scenario in which the exposure frequency is 1.0 (i.e.
to examine variability conditional upon the assumption that
exposures always are at the outdoor concentration). In this
second scenario, the mean value for EFCF is 1.0 and there is no
variability. Both scenarios are considered in this analysis.
Exposure Duration (EDCF) : The exposure duration assumed in the
base-line calculations is 70 years. This is consistent with the
assumption either that an individual resides the entire 70 years
at the same location, or moves to grid blocks with the same
average concentration. In empirical studies16, the median length
of time an individual resides in a single location is 9 years.
This time is distributed approximately lognormally with a
geometric standard deviation of 2.5.16
The use of values of EDCF other than 1 is justified only for
the calculation of the MIR, and only if it is assumed that ALL
individuals living within the grid block characterized by the MEI
move out of that grid block upon changing residence. If even one
individual remains in the grid block throughout the 70 years, his
or her risk is by definition the MIR. Since this does occur,
there is no need to incorporate variability of EDCF into the
G-90
-------
calculation of the MIR. Variability in this factor is not,
therefore, considered further here, and a value of 70 years is
employed in all calculations.
As to the issue of the variability of the Lifetime
Probability of Cancer for the entire U.S. population, it must be
noted that the exposed population exceeds 190 million, or more
than 70 percent of the U.S. population. It is a reasonable
assumption that exposed individuals, upon changing residences,
simply change positions with another individual in the exposure
field. If the individual moves out of the exposure field, a
previously unexposed person moves to that location and
contributes equally to the calculation of the mean Lifetime
Probability of Cancer for the entire U.S. population and to the
number of deaths. The authors know of no empirically-verified
assumption as to the distribution of the ratio of initial air
concentration (i.e. at the location from which a person moves)
over the air concentration at the new place of residence for
individuals who move. It is assumed here that all individuals
reside in one randomly chosen location throughout their 70 year
lifespan. This will tend to slightly overestimate the variation
in the Lifetime Probability of Cancer between individuals in the
U.S. population, but the magnitude of this overestimate cannot be
determined at present due to a lack of data on the distribution
of ratios of concentrations before and after migration.
Breathing Rate (BRcF) : Measurements of variability (Hattis
and Silver, 1994), after factoring out measurement error,
indicate the breathing rate is distributed lognormally in adults
with a geometric SD of between 1.2 and 1.3. A geometric SD of
1.25 is assumed here as a best estimate.
Mean breathing rates in occupational populations (such as
the ones on which the CSFs are based) generally are higher than
the mean of 20 m3/day for the general population (which includes
consideration of the difference in breathing rates indoors and
outdoors). Assuming primarily light activity in the workers,
the breathing rate would be approximately 30 mVday. This
indicates a mean value of the ratio (BR^) of 0.67. If the CSFs
used in this analysis do not reflect these differences, the value
of BRCF would be distributed lognormally with a geometric SD of
1.25, a mean of 20/30 or 0.67, and a median of 0.67 /
exp(ln21.25/2) or 0.65.
However, it is not currently clear whether the original CSFs
included consideration of the different breathing rates in the
two populations. Until this issue is resolved, the analysis of
variability developed here includes no additional correction for
G-91
-------
breathing rates, employing a lognormal distribution with a mean
of 1.0 and a geometric SD of 1.25. It is recommended that future
research focus on resolving this issue.
Deposition Fraction (DFCF): Deposition fractions vary between
individuals due to variation in breathing rates, particle sizes
and the sizes of lung passages.17 Crawford-Brown and Hofmann18
have estimated variability of deposition fractions for a constant
particle size and found it to be distributed lognormally with a
geometric SD of approximately 2 after accounting for uncertainty
due to measurement errors. The differences in deposition
fractions between the epidemiological study populations and the
general population cannot be determined due to a lack of
information on particle size distributions in the two
populations. It is assumed here, therefore, that the values of
DFCF are distributed lognormally with a geometric SD of 2, mean
of 1, and a median of 1.0 / exp(ln22/2) or 0.79. Future
research should focus on determining the particle size
distributions of HAPs, and on the implications of this for DFCF.
G.4.3.2 Variability of Compound-Specific Values in Equation
G-4. The values necessary for Equation G-4 that are a function
of the compound being considered, at least in the present
analysis, are TCF, TFCF, DCF, SFCF and SCF.
Retention Half-times (TCF) : Retention half-times typically are
distributed lognormally19'20, with a geometric SD of approximately
1.3 after accounting for uncertainty due to measurement error.
One problem with such measurements, however, is that they often
are over a short-term and conducted only once on each individual
in the study population. For the current analysis, it is
necessary to determine the variability in the long-term average
retention characteristics in individuals (i.e. the half-time
averaged over a lifetime of intake, rather than following a
single acute exposure as in the experimental measurements). The
information on which the analysis of Crawford-Brown20 was based,
however, includes instances of long-term retention of materials
such as radium in the bone and plutonium in the lungs.
There is no information to suggest that these long-term
retention half-times in the epidemiological study populations
differs systematically from those in the general population,
although life-time averaged retention half-times should be
slightly lower in the general population due to the inclusion of
young ages where the retention half-times usually are lower than
adult values.21 The effect of this age-dependence on the three
compounds considered here, however, cannot be determined.
Variability in TCF for this analysis is assumed to be described
G-92
-------
by a lognormal distribution with a geometric standard deviation
of 1.3, a mean of 1.0, and a median of 1.0 / exp(ln2!.3/2) or
0.97.
Other (TFCF and DCF) : As mentioned previously, there is no
information on which to base estimates of the variability of
either TFCF or DCF , so these parameters are set here to 1. 0 with
no variability. This should not introduce significant errors into
the analysis of variability since the three compounds being
considered are not significantly bioactivated, which is the
common source of variability in these two factors when dealing
with more complex chemical forms.
Slope Factor (SFCF) : The variability in the CSF has not been
determined quantitatively for the general population. It has been
noted, however, that the arsenic CSF for heavy smokers may be
larger than that for never-smokers by as much as a factor of 822,
suggesting variability in the population with respect to smoking
status. The variability of CSFs for arsenic is characterized here
by a lognormal distribution with a mean of 1.0 and a geometric SD
of 1.5 (a factor of 5 between the upper and lower 95 percent
limits). It is not possible at present, however, to provide a
quantitative characterization of this variability for nickel and
chromium. This variability is discussed in more detail
qualitatively in a later section on sensitive subpopulations.
Speciation (SCF) : The variability in the speciation factor
reflects the difference in CSFs for the different species of each
HAPs, and the likely variation in this speciation for exposures
to individuals in well defined groups. The variation will be
largest for the entire U.S. population (which includes variation
in the source terms that contribute to speciation), and smallest
for consideration of the MIR (where the speciation within a grid
block or at an exposure point will not vary as greatly between
exposed individuals). These sources of variability is considered
separately here for each of the three HAPs.
Arsenic: Arsenic is found primarily in oxidation states III
and V. The best estimate is that 18 percent is in State III and
43 percent in State V, with the remainder in an unspecified
oxidation state. At the upper bound, 33 percent is in State III
and 67 percent in State V. At the lower bound, 3 percent is in
State III and 19 percent in State V, with the remainder in an
unspecified oxidation state.
Neither state has been shown to produce mutations23, but
there is limited evidence that both produce chromosomal
changes.24 It is not possible at this time to differentiate the
G-93
-------
tumorigenicity of these two compounds, so they are considered
here to have the same CSF of 0.00429 (ug/m3)-1. Since this is the
same CSF as was used in the base-line calculations, the mean for
this correction factor then is 1.0. No variability is assumed
since the two species are assumed to have equal CSFs, although
uncertainty is treated in a later section.
Nickel: There are several forms of nickel, with soluble,
sulfides and oxides being present. For nickel refinery dust
(characteristic of exposures in the epidemiological population) ,
the best estimate of the mean value of the CSF for an exposed
population is 2.4 x 10"4 (ug/m3)-1. For subsulfidic dust (present
in exposures of the general population), the best estimate of the
mean CSF is 4.8 x 10"4 (ug/m3)'1. In exposures of the public,
soluble nickel and oxides dominate, constituting between 90 and
95 percent of the total nickel. In a review by the International
Committee on Nickel Carcinogenesis25, soluble nickel was
determined to be the primary carcinogenic form, being
approximately a factor of 10 more potent than the less soluble
forms. Recent studies in rats and mice conducted by the NTP,
however, have found soluble nickel to be not carcinogenic in
these species (personal communication, Hudson Bates, NIPERA,
September, 1995). The soluble fraction for the general public
averages approximately 60 percent of total nickel, but it is not
possible to relate this accurately to the fraction in the
epidemiological studies since the composition of refinery dust is
generally poorly characterized. Variability of speciation, and
its effect on variability of CSFs for the general population,
cannot, therefore, be considered here. Since the base-line
calculation employed the CSF for nickel subsulfide, the mean of
this correction factor is assumed to be 1.0.
Chromium: The two compounds of interest are Cr-III+ and
Cr-VI+. Initial data indicate that the primary carcinogenic form
is Cr-VI+23, although the issue of the carcinogenicity of Cr-III+
has not been resolved. Until that is resolved, it is not possible
to quantitatively characterize the variability of the speciation
correction factor, even if data on the degree of variability in
speciation itself is available.
G.4.3.3 Summary of Variability. Each of the factors above
may be incorporated into Equation G-4 to yield a determination of
variability in the cancer risk for an individual at a selected
level of exposure. For the present section, which deals only with
exposure-response assessment, it is assumed that the
concentration of each of the three HAPs is 1.0 (ug/m3) . This
will be combined with information on variability of exposures in
G-94
-------
a later section on risk characterization. A summary of the
various distributions is provided in Table G-17.
Figure G-12 displays the probability density functions for
the separate parameter values used in Equation G-4 for evaluating
variability for both estimating incidence in which the exposure
frequency correction factor has a mean of 0.52 and for estimating
MIR in which the exposure frequency is set a t 1.0. Table G-17
summarizes and Figure G-13 graphically displays the resulting
probability density functions for the CSF used to estimate annual
incidence for exposures to arsenic, chromium, and nickel at a
concentration of 1 ug/m3 (uniform throughout the exposed
population). From these, it may be seen that the 95 percent
interval for the relationship between exposure and lifetime
probability of cancer for arsenic is approximately 3.8 x 10"4 to
7.9 x 10'3 (the baseline calculation yields 4 x 10'3) ; the 95
percent interval for the relationship between exposure and
lifetime probability of cancer for nickel is approximately 4.3 x
10"5 to 8.9 x 10'4 (the baseline calculation yields 4.8 x 10'4) ;
and the 95 percent interval for the relationship between exposure
and lifetime probability of cancer for chromium is approximately
2.2 x 10~4 to 4.4 x 10"3 (the baseline calculation yields 0.2 x
1.2 x 10'2 = 2.4 x 10 -3; the factor of 0.2 accounts for the
assumption in the baseline calculation that Cr(VI) represents 20
percent of the total exposure to chromium). These confidence
intervals are particularly valid for the average cancer incidence
estimates.
Figure G-14 displays the resulting probability density
functions for the Lifetime Probability of Cancer for exposures to
arsenic, chromium, and nickel at a concentration of 1 ug/m3
(uniform throughout the exposed population). This distribution
is also summarized in Table G-17. From these, it may be seen
that the 95 percent interval for the relationship between
exposure and lifetime probability of cancer for arsenic is
approximately 8.3 x 10"4 to 1.4 x 10"2 (the baseline calculation
yields 4 x 10"3) ; the 95 percent interval for the relationship
between exposure and lifetime probability of cancer for nickel is
approximately 9.3 x 10"5 to 1.5 x 10~3 (the baseline calculation
yields 4.8 x 10"4) ; and the 95 percent interval for the
relationship between exposure and lifetime probability of cancer
for chromium is approximately 4.6 x 10"4 to 7.5 x 10"3 (the
baseline calculation yields 2.4 x 10"3) . These confidence
intervals will be particularly valid for calculations of the MIR
estimates.
G-95
-------
Table G-17. Summary of Monte Carlo Simulation of Variability
in CSF for Estimating MIR and Incidence
Variability
Mean
EPA Verified URE
(Percentile)
Percentiles
JRE
0.0%
2.5%
5.0%
10%
25%
50%
75%
90%
95.0%
97.5%
100.0%
Arsenic
2.3E-03
4.3E-03
(89)
1.4E-04
3.9E-04
4.8E-04
6.2E-04
9.6E-04
1.7E-03
2.8E-03
4.6E-03
6.1E-03
7.9E-03
2.5E-02
Incidence
Chromium
1.3E-03
2.4E-03
(89)
7.7E-05
2.2E-04
2.7E-04
3.5E-04
5.4E-04
9.3E-04
1.5E-03
2.6E-03
3.4E-03
4.4E-03
1.4E-02
Nickel
2.5E-04
4.8E-03
(89)
1.5E-05
4.3E-05
5.3E-05
6.9E-05
1.1E-04
1.9E-04
3.1E-04
5.2E-04
6.8E-04
8.8E-04
2.8E-03
Arsenic
4.3E-03
4.3E-03
(64)
2.9E-04
8.3E-04
1.1E-03
1.4E-03
2.1E-03
3.4E-03
5.4E-03
8.4E-03
1.1E-02
1.3E-02
4.1E-02
MIR
Chromium
2.4E-03
2.4E-03
(64)
1.6E-04
4.6E-04
5.9E-04
7.7E-04
1.2E-03
1.9E-03
3.0E-03
4.7E-03
6.3E-03
7.5E-03
2.3E-02
Nickel
4.8E-04
4.8E-03
(64)
3.2E-05
9.3E-05
1.2E-04
1.5E-04
2.3E-04
3.8E-04
6.0E-04
9.4E-04
1.3E-03
1.5E-03
4.6E-03
G-96
-------
Figure G-12. Assumptions Used in the Analysis of Variability in CSFs
Assumption: BR, Variability
Lognormal distribution with parameters:
Mean 1.00E+00
Standard Dev. 2.23E-01
Selected range is from O.OOE+0 to +lnfinrty
Mean value in simulation was 1.01E+0
Assumption: DF, Variability
Lognormal distribution with parameters:
Mean 1.00E+00
Standard Dev. 6.93E-01
Selected range is from O.OOE+0 to +lnfinity
Mean value in simulation was 1.01 E+0
Assumption: T, Variability
Lognormal distribution with parameters:
Mean 1.00E+00
Standard Dev. 2.62E-01
Selected range is from O.OOE+0 to ^Infinity
Mean value in simulation was 1 .OOE+0
Assumption: SFas Variability
Lognormal distribution with parameters:
Mean 1 .OOE+00
Standard Dev. 4.05E-01
Selected range is from O.OOE+0 to +lnfinity
Mean value in simulation was 9.97E-1
Assumption: EF, Var., Incidence (Used in CSF for estimating incidence only)
Lognormal distribution with parameters: "•*"•*
Mean 5.20E-01
Standard Dev. 1.75E-01
Selected range is from O.OOE+0 to +lnfinity
Mean value in simulation was 5.15E-1
G-97
-------
Figure G-13. Distribution of CSFs Used to
Estimate Annual Incidence: Variability
Cell D105
.044 1
Forecast: CSF, Arsenic, Variability, Incidence
Frequency Chart
2,941 Trials Shown
128
Forecast: CSF, Chromium, Variability, Incidence
Frequency Chart
2,938 Trials Shown
127
Forecast: CSF, Nickel, Variability, Incidence
Frequency Chart
2,938 Trials Shown
127
G-98
-------
Figure G-14. Distribution of CSFs Used to
Estimate MIR: Variability
C«IID107
.040 H
Forecast: CSF, Arsenic, Variability, MIR
Frequency Chart 2,947 Trials Shown
> 119
Forecast: CSF, Chromium, Variability, MIR
Cell E107 Frequency Chart 2,929 Trials Shown
.039 H
V 115
86.2
57.5
28.7
n
Cell F107
.038
Forecast: CSF, Nickel, Variability, MIR
Frequency Chart
.028 .. . .
.019 .. . .
2,923 Trials Shown
110
82.5
275
G-99
-------
G.4.4 Uncertainty in the Exposure-response Relationship
G.4.4.1 Uncertainty of Compound- Independent Values in
Equation G-4 . The values necessary for Equation G-4 that are
assumed not to be a function of the compound being considered are
EFCF, EDCF/ BRcp, and DFCF.
Exposure Frequency (EFCF) : The exposure frequency assumed in the
base-line calculation is 1.0, interpreted as an assumption that
(I) indoor and outdoor concentrations are equal and (ii) that an
individual spends all of his or her time within the assigned
exposure grid-block. The calculation of EFCF is reviewed in
section G.4.3.1 and is not repeated here.
As to the first assumption, the median value for the ratio
of the concentration indoors to that outdoors is 0.4 and the mean
approximately 0 . 42 . 16 The uncertainty in this value is due
primarily to measurement variability and consideration of the
representativeness of the sample. The standard error of the mean
is approximately 0.05. Given the sample size and the random
nature of selection for the population measured, the mean value
for the concentration ratio is unlikely to be less than 0.3 or
greater than 0.5. A normal distribution with a SD of 0.05 is
assumed here.
The ratio of concentrations indoors to outdoors must be
supplemented by an estimate of the fraction of time spent indoors
and outdoors, taken here to be given by terms fI (indoors) and f0
(outdoors) . These fractions depend on age and occupation, but
the fraction of time outdoors typically averages approximately
0.15 and the fraction indoors 0.85.26 This yields a mean
"weighted" exposure frequency of (1.0 x 0.15) + (0.42 x 0.85) or
0.5. Information on the uncertainty of f x and f 0 is not
available, but the fraction of time indoors cannot exceed 1.0 and
probably rarely is below 0.5 (allowing for sleeping and normal
household activities) . The values of fz and f0 for individuals
clearly are completely correlated, since they must sum to 1.0 in
all instances. In the absence of detailed information on which to
estimate uncertainty, the subjective judgment is made here that
uncertainty in fj is characterized by a normal distribution with
a mean of 0.85 and SD of 0.05, truncated at the boundary value
of 1.0. The value of f0 always is 1 - f t. The "weighted" exposure
frequency correction factors then is distributed normally with a
mean of approximately 0.5 and a SD of approximately 0.05.
The above distribution is representative of a first scenario
in which the actual conditions in U.S. homes is considered.
However, the scenario in which the exposure frequency is 1.0 was
G-100
-------
also examined. In this second scenario, the mean value for EFCF
is 1.0 and there is no variability. Both scenarios are considered
in this analysis. For the MEI, the second scenario is considered
more appropriate.
Exposure Duration (EDCF) : The exposure duration assumed in the
base-line calculation is 70 years. This is consistent with the
assumption either that an individual resides the entire 70 years
at the same location, or moves to grid blocks with the same
average concentration.
The use of values of EDCF other than 1 is justified only for
the calculation of the MIR, and only if it is assumed that ALL
individuals living within the grid block characterized by the MEI
move out of that grid block upon changing residence. If even one
individual remains in the grid block throughout the 70 years, his
or her risk is by definition the MIR. Since this does occur,
there is no need to incorporate uncertainty of EDCF into the
calculation of the MIR. Uncertainty in this factor is not,
therefore, considered further here, and a value of 70 years is
employed in all calculations.
As to the uncertainty in cancer incidence, the exposed
population exceeds 190 million, or more than 70 percent of the
U.S. population; therefore, it is a reasonable assumption that
exposed individuals, upon changing residences, simply change
positions with another individual in the exposure field. If the
individual moves out of the exposure field, a previously
unexposed person moves to that location and contributes equally
to the calculation of the number of deaths. This change will have
no effect on the estimated incidence. It is assumed here that
individuals reside in one randomly chosen location throughout
their 70 year lifespan, and that uncertainty in EDCF is not
significant.
Breathing Rate (BRCF) : Measurements of variability27, corrected
for measurement error, indicate the breathing rate is distributed
lognormally in adults with a geometric SD of between 1.2 and
1.3. These same measurements indicate that the SE is
approximately 3 percent of the mean value. This must be broadened
since it does not reflect uncertainty in the effect of
introducing differences in age distribution and level of
activity, issues of accuracy of the measurement techniques, or
issues of the representativeness of the sampled population. The
subjective judgment here is that these factors should produce a
distribution which is characterized by lognormality with a
geometric SD of at least 1.1 for the ratio of lifetime-averaged
breathing rates in the two populations, given that inclusion of
G-101
-------
age-dependence in the calculation of life-time averaged breathing
rates can cause changes as large as 10 percent.
The mean breathing rates in occupational populations (such
as the ones on which the CSFs used in the present analysis are
based) generally are higher than the mean of 20 mVday for the
general population. Assuming primarily light activity in the
workers, the breathing rate would be approximately 30 mVday. As
of the publication of this preliminary report, however, it is not
clear whether the original CSFs included consideration of the
different breathing rates in the two populations.
Deposition Fraction (DFCF) : Deposition fractions vary between
individuals due to variation in breathing rates, particle sizes
and the sizes of lung passages.17 Crawford-Brown and Hofmann18
have estimated variability of deposition fraction for a constant
particle size and found it to be distributed lognormally with a
geometric SD of approximately 2 after accounting for measurement
uncertainty. The SD of the estimate of the mean deposition
fraction is on the order of 10 percent of the mean when particle
size is known. This must be broadened since it does not reflect
uncertainty in the effect of introducing differences in age
distribution and level of activity, issues of accuracy of the
measurement techniques, or issues of the representativeness of
the sampled population. The differences in deposition fraction
between the epidemiological study populations and the general
population were not included in the original development of the
CSF by the Agency and cannot be determined due to a lack of
information on particle size distributions in the two
populations. This should broaden the distribution for
uncertainty, but the degree of broadening cannot be well
specified at present quantitatively. The subjective judgment here
is that these factors should produce a distribution which is
characterized by lognormality with a geometric SD of at least
1.25 for the ratio of lifetime-averaged deposition fractions in
the two populations, given that changes in particle size within
the range of values typically found in the ambient air (between
0.05 and 0.5 microns) can cause changes as large as a factor of
two or more.
G.4.4.2 Uncertainty of Compound-Specific Values in Equation
G-4. The values necessary for Equation G-4 that are a function
of the compound being considered, at least in the present
analysis, are EFCF, EDCF, BRcF, and DFCF.
Retention Half-times (TCF) : Retention half-times typically are
distributed lognormally19-20, with a median value of the geometric
SD of approximately 1.3. There is no information to suggest that
G-102
-------
retention half-times in the epidemiological study populations
differs systematically from those in the general population,
although life-time averaged retention half-times should be
slightly lower due to the inclusion of young ages where the
retention half-times usually are lower than adult values
(Crawford-Brown, 1984).
The Agency did not include a correction for retention
half-times in the original analyses for the CSFs. The effect of
this age-dependence on the three compounds considered here,
however, cannot be determined directly due a lack of data on the
effects of age on these compounds. It is assumed here that the
study populations consisted primarily of adults, and that the
general U.S. population contains all ages. Inclusion of younger
ages into a calculation of the lifetime-averaged value of the
half-time generally produces values that are within 10 percent of
the adult value (based on information provided in Crawford-Brown,
1984), although it should be noted that the life-time averaged
value usually is lower than the adult value. For this analysis,
the subjective judgment of the analysts is that uncertainty in
the estimate of the mean value of TCF for the exposed population
is described by a normal distribution with a mean of 1.0, and a
SD of 0.1.
Other (TFCF and DCF) : There is no information on which to base
estimates of the uncertainty of either TFCF or DCF , so these
parameters are set here to 1.0 with no uncertainty. This should
not introduce significant errors into the analysis of uncertainty
since the three compounds being considered are not significantly
bioactivated, which is the common source of variability in these
two factors when dealing with more complex chemical forms.
Slope Factor (SFCF) : The uncertainty in the CSF has several
components. There is uncertainty due to parameter estimation for
a selected extrapolation function (such as the linearized
multistage model) applied to a single data set, which may be
summarized by the standard error of the CSF for that data set,
broadened by consideration of study design. There is uncertainty
due to the existence of different data sets, which may be
characterized by the standard error of the mean value of the CSF
across those data sets, broadened by issues of representativeness
of the sampled populations. There is uncertainty due to the
inability to choose between competing exposure-response models
used for extrapolation. And there is uncertainty introduced by
potential differences in concurrent exposures to synergistic
and/or antagonistic risk agents (such as cigarette smoke) in the
general U.S. and study populations.
G-103
-------
To characterize uncertainty from these sources, the authors
analyzed the results of selected epidemiological studies on
exposures to airborne arsenic28, nickel29 and chromium30, as well
as the differences in CSFs determined from different populations
exposed to these compounds and from using different exposure-
response functions. These uncertainties are discussed here for
the three separate materials. Consideration of the uncertainty
introduced by use of extrapolation functions is discussed
qualitatively in a later section.
For each of the three compounds, normal distributions were
converted to lognormal distributions with equivalent 68 percent
confidence intervals to avoid negative values for the CSFs, and
to better reflect issues of representativeness and study design
(best described by lognormal distributions). The "equivalent
lognormal" characteristics are described below.
Arsenic: The studies examined for the maximum likelihood
estimates of the CSF varied between 1.25 x 1CT3 and 7.6 x 1CT3
(ug/m3)'1.28'31'32'33 The base-line calculations of the risk from
arsenic employed a mean value of 0.00429 (ug/m3) "1; the standard
error of the mean for these studies is approximately 0.002
(ug/m3)'1. A recent analysis by Virens and Silvers34 yielded a
value of 0.00143 (ug/m3)'1, which is within the range of values
noted above.
Uncertainty due to imprecision in the estimate of the CSF
from a single data set may be approximated by examining a
representative study on arsenic28, as shown in Figure G-15. The
standard error in the CSF for these data obtained from a linear
model is approximately 70 percent of the mean (maximum
likelihood) estimate, with a lower bound estimate that includes a
CSF of 0.0. The best estimate is shown as the solid line in that
figure, with the upper 68 percent function shown as the dashed
line. This suggests an intra-study standard error of
approximately 0.003 (ug/m3)"1. This is similar in magnitude to the
standard error noted in the previous paragraph and also is
similar in magnitude to the difference in maximum likelihood
estimates when the linearized multistage and Weibull models are
applied to the same data.
The effect of differences in smoking prevalence for the
study and general U.S. populations cannot be determined
quantitatively at present. The arsenic CSF for heavy smokers may
be larger than that for never-smokers by as much as a factor of
822, but the effect on the present analysis cannot be determined
rigorously at present.
G-104
-------
•ooo «ooo
mg-yrs/U
10000 12000 14000
Chromium
400-
3SO-
100.
250-
cc
Z zoo-
Years ol Experience
Nickel
200 300 400
mg-months/m3
This figure displays the data and model fits used in the uncertainty analyses for slope factors In the case of arsenic, nickel and chromium. In each case, the measure of effect is
the SMR The value of (SMR-100 percent) is assumed to be proportional to the excess lifetime probability of cancer The open squares are the original data points from the
studies cited in the text. The vertical lines around these squares are the 95 percent confidence limits, determined through Poisson statistics. The solid sloped line is the best
fitting linear model as determined from least squares in which the weighting factor for each data point is the inverse of the standard error of the mean. The dashed line is the
upper 68 percent confidence interval on the slope.
For each substance, the Weibull model is given by the equation:
(SMR-100%yiOO% = l-expHa+bD"1))
where a, b and m are constants. The value of m is equal to 1.0 for all three substances examined here. Best-fitting values of a and b were determined from least squares
analysis in which the weighting factor for each data point is the inverse of the standard error of the mean. The best-fitting linear equation and the Weibull model then were used
to predict the value of (SMR-100 percent) at a dose equal to 5 percent of the maximal dose in the epidemiological studies. The ratio of these two predictions is reported in the
text.
Figure G-15.
Data and Model Fits for Slope Factors for Arsenic,
Chromium, and Nickel.
G-105
-------
Summarizing, the above sources of uncerta
of 0.002 (ug/m3)'1 from intra-study uncertainty;
of 0.003 (pg/m3)'1 from inter-study uncertainty,
of exposure-response models; and an uncertainty ch^ 3
exposures that cannot be characterized quantitative •*"
other than to state that it should be of a magnitud?
the sources noted above. Taking these considerations1^. "
the equivalent lognormal distribution has a mean of 1\ "*~
geometric SD of 3.
o
Nickel: The different studies of the CSF for nickel
exposures7 varied by a factor of approximately 5 from low
around the mean estimate of 0.00048 (ug/m3)'1. This is simij.
the case for arsenic. If the study by Enterline and Marsh29 .
shown in Figure G-15 is used to estimate uncertainty conditit
on a single data set and the linearized multi-stage model, tht
of the CSF is approximately 80 percent of the mean, which is
similar to that found for arsenic and also is similar in
magnitude to the difference in maximum likelihood estimates when
the linearized multistage and Weibull models are applied to the
same data. Taking these considerations together, the equivalent
lognormal distribution has a mean of 1.0 and a geometric SD of 3.
Chromium: Only one study35 provides a quantitative estimate
of the CSF for chromium exposures, with the other studies being
restricted to length of exposure. As seen in Figure G-15, the SD
of the CSF is approximately 85 percent of the mean, which is
similar to that found for nickel and arsenic and also is similar
in magnitude to the difference in maximum likelihood estimates
when the linearized multistage and Weibull models are applied to
the same data. The mean estimated by the Agency is 0.012
(ug/m3)"1 Taking these considerations together, the equivalent
lognormal distribution has a mean of 1.0 and a geometric SD of 3.
Speciation (SCF) : The uncertainty in the influence of speciation
on the CSF cannot be determined accurately since the composition
in the epidemiological studies is characterized poorly. For
chromium, uncertainty in SCF is related directly to the
uncertainty in the difference in Cr-VI+/Cr-III+ speciation
fraction in the general U.S. and study populations, and the
influence of this difference on the "species-averaged" CSF (i.e.
the CSF weighted by the fraction of the exposure from each
species). It is not possible at present to quantitatively
characterize the uncertainty in the CSF for chromium. The same is
true for arsenic.
The distribution for nickel is based on the recognition that
the CSF for the subsulfide nickel (characteristic of general U.S.
G-106
-------
exposures) is a factor of 2 above that for refinery nickel
(characteristic of the exposures for the epidemiological
population). The more recent NTP studies, however, suggest that
soluble nickel may not be carcinogenic in .rats and mice, which
would tend to increase the confidence in CSFs below the best
estimates used in the baseline analysis of risk. The CSF for the
general population should be intermediate between these values,
given the likely inclusion of both species of nickel to some
degree. The CSF chosen in the base-line analysis, however,
included only subsulfide nickel. As a conservative assumption,
and based on subjective judgment (supported by the fact that the
range of a factor 2 between the two species implies that the mean
will not be incorrect by more than 40 percent even if one species
dominates) , the uncertainty for SCF for nickel is given here by
an equivalent lognormal distribution has a mean of 1.0 and a
geometric SD of 1.2.
G.4.4.3 Summary of Uncertainty. Uncertainty in each of the
factors above may be incorporated into Equation G-4 to yield a
determination of uncertainty in the Lifetime Probability of
Cancer for an exposed individual at a selected level of exposure.
For the present section, which deals only with exposure-response
assessment, it is assumed that the concentration is 1.0 (ug/m3) .
This will be combined with information on uncertainty of
exposures in a later section on risk characterization.
Figure G-16 displays the probability density functions for
the separate parameter values used in Equation G-4 for the first
scenario in which the exposure frequency correction factor has a
mean of 0.52. Figure G-17 displays the resulting probability
density functions for the CSF used to estimate annual incidence
for exposures to arsenic, chromium, and nickel at concentrations
of 1 ug/m3 (uniform throughout the exposed population). Table
G-18 summarizes the distribution. From these, it may be seen
that the 95 percent interval for the relationship between
exposure and lifetime probability of cancer for arsenic is
approximately 2.2 x 10"4 to 8.0 x 10"3 (the baseline calculation
yields 4 x 10~3) ; the 95 percent interval for the relationship
between exposure and lifetime probability of cancer for nickel is
approximately 2.3 x 10"5 to 9.2 x 10~4 (the baseline calculation
yields 4.8 x 10"4) ; and the 95 percent interval for the
relationship between exposure and lifetime probability of cancer
for chromium is approximately 1.3 x 10~4 to 4.5 x 10"3 (the
baseline calculation yields 2.4 x 10"3). These confidence
intervals will be particularly valid for calculations of the
average incidence in the U.S. population.
G-107
-------
Figure G-16. Assumptions Used in the Analysis of Uncertainty in CSFs
Assumption: BR, Uncertainty
Lognormal distribution with parameters:
Mean 1.00E+00
Standard Dev. 9.50E-02
Selected range is from O.OOE+0 to +lnfinity
Mean value in simulation was 9.98E-1
Assumption: DF, Uncertainty
Lognormal distribution with parameters:
Mean 1.00E+00
Standard Dev. 2.23E-01
Selected range is from O.OOE+0 to +lnfinity
Mean value in simulation was 1 .OOE+0
Assumption: T, Uncertainty
Normal distribution with parameters:
Mean 1.00E+00
Standard Dev. 1.00E-01
Selected range is from -Infinity to -t-lnfinity
Mean value in simulation was 1 .OOE+0
Assumption: SFAs, Uncertainty
Lognormal distribution with parameters:
Mean 1.00E+00
Standard Dev. 1.10E+00
Selected range is from O.OOE+0 to +lnfinity
Mean value in simulation was 9.85E-1
Assumption: SFNi, Uncertainty
Lognormal distribution with parameters:
Mean 1.00E+00
Standard Dev. 1.10E+00
Selected range is from O.OOE+0 to +lnfinity
Mean value in simulation was 9.85E-1
Assumption: SFCr, Uncertainty
Lognormal distribution with parameters:
Mean 1.00E+00
Standard Dev. 1.10E+00
Selected range is from O.OOE+0 to +lnfinity
Mean value in simulation was 9.90E-1
Assumption: SNi, Uncertainty
Lognormal distribution with parameters:
Mean 1.00E+00
Standard Dev. 1.82E-01
Selected range is from O.OOE+0 to -(-Infinity
Mean value in simulation was 1.OOE+0
Assumption: EF, Uncertainty (Used for CSF to Estimate Incidence Only)
Normal distribution with parameters:
Mean 5.00E-01
Standard Dev. 5.00E-02
Selected range is from -Infinity to +lnfinity
Mean value in simulation was 5.01 E-1
G-108
-------
Figure G-17. Distribution of CSFs Used
to Estimate Annual Incidence: Uncertainty
Forecast: CSF, Arsenic, Uncertainty, Incidence
Frequency Chart
.000
O.OOE+0
2,929 Trials Shown
141
70.5
35.2
Forecast: CSF, Chromium, Uncertainty, Incidence
Frequency Chart
2,927 Trials Shown
1 137
102
68.5
34.2
Cell F104
.048 -r
Forecast: CSF, Nickel, Uncertainty, Incidence
Frequency Chart
.036 ...
a
e
.024 -. .
.012 -. .
O.OOE+0
2.50E-4
5.00E-4
probability
7.50E-4
2,919 Trials Shown
h 140
illlliil...... n i- . . I,
105
70
35
<
1.00E-3
G-109
-------
Table G-18. Summary of Monte Carlo Simulation of Uncertainty
in CSF for Estimating MIR and Incidence
Uncertainty
Mean
EPA Verified URE
(Percentile)
Percentiles:
JRE
0.0%
2.5%
5.0%
10%
25%
50%
75%
90%
95.0%
97.5%
100.0%
Arsenic
2.1E-03
4.3E-03
(88)
6.7E-05
2.2E-04
3.0E-04
4.1E-04
7.3E-04
1.4E-03
2.6E-03
4.4E-03
6.2E-03
8.1E-03
2.7E-02
incidence
Chromium
1.2E-03
2.4E-03
(88)
3.7E-05
1.3E-04
1.7E-04
2.3E-04
4.1E-04
7.8E-04
1.4E-03
2.5E-03
3.5E-03
4.5E-03
1.5E-02
Nickel
2.4E-04
4.8E-03
(88)
6.8E-06
2.3E-05
3.2E-05
4.5E-05
8.0E-05
1.5E-04
2.9E-04
5.1E-04
7.1E-04
9.2E-04
3.5E-03
Arsenic
4.3E-03
4.3E-03
(67)
9.1 E-05
4.8E-04
6.2E-04
8.6E-04
1.5E-03
2.7E-03
5.1E-03
9.4E-03
1.3E-02
1.7E-02
5.2E-02
MIR
Chromium
2.4E-03
2.4E-03
(67)
5.1 E-05
2.7E-04
3.5E-04
4.8E-04
8.6E-04
1.5E-03
2.9E-03
5.3E-03
7.5E-03
9.7E-03
2.9E-02
Nickel
4.8E-04
4.8E-03
(67)
8.5E-06
4.9E-05
6.7E-05
9.3E-05
1.6E-04
3.0E-04
5.7E-04
1.1E-03
1.4E-03
2.1E-03
5.7E-03
G-HO
-------
Figure G-18 displays the resulting probability density
functions for the CSF used to estimate MIR for exposures to
arsenic, chromium, and nickel at a concentration of 1 ug/m3
(uniform throughout the exposed population). From these, it may
be seen that the 95 percent interval for the relationship between
exposure and lifetime probability of cancer for arsenic is
approximately 4.8 x 1CT4 to 1.7 x 10"2 (the baseline calculation
yields 4 x 10"3) ; the 95 percent interval for the relationship
between exposure and lifetime probability of cancer for nickel is
approximately 4.9 x 10"5 to 2.1 x 10 ~3 (the baseline calculation
yields 4.8 x 10 ~4) ; and the 95 percent interval for the
relationship between exposure and lifetime probability of cancer
for chromium is approximately 2.7 x 10"4 to 9.7 x 10"3 (the
baseline calculation yields 2.4 x 10 ~3). These confidence
intervals will be particularly valid for calculations of the MIR.
These distributions do not include detailed quantitative
consideration of uncertainties treated purely qualitatively, such
as the uncertainty introduced by dose-response extrapolation (for
all compounds) and speciation (for chromium and arsenic). As a
result, the quantitative confidence intervals and distributions
presented here should be viewed as being more narrow than those
that would be produced if more formal consideration were given to
these other sources of uncertainty.
G.4.5 Susceptibility and Additional Factors Affecting
Uncertainty in Exposure-Response
Factors leading to susceptibility for cancer induction
following exposure to environmental pollutants (such as arsenic,
chromium and nickel) have been reviewed recently by the National
Research Council.2 These factors relate to each of the stages of
calculation employed in the present analysis: exposure frequency
and duration; breathing rate; deposition fraction; retention
half-time; carcinogen metabolism; and dose-response
relationships. Susceptible individuals will be those:
1. At the upper end of the distribution of indoor/outdoor
concentration ratios;
2. At the upper end of the distribution of fraction of
time spent outdoors;
3. At the upper end of the distribution of breathing
rates;
4. At the upper end of the distribution of deposition
fractions;
G-lll
-------
Figure G-18.
Distribution of CSFs Used to Estimate MIR: Uncertainty
Cell D106
.043 -
.033 ...
£
3 .022 ...
a. -011 ••
.000
Forecast: CSF, Arsenic, Uncertainty, MIR
Frequency Chart 2,929 Trials Shown
T- 127
95.2
63.5
31.7
Forecast: CSF, Chromium, Uncertainty, MIR
Cell E106 Frequency Chart 2,933 Trials Shown
h 133
.034 -. . .
.023 -. .
.011 -.
.000
99.7
66.S
332
n
O.OOE-tC
Forecast: CSF, Nickel, Uncertainty, MIR
Frequency Chart
2,928 Trials Shown
151
I
37.7
O.OOE+O
2.00E-3
G-H2
-------
5. At the lower end of the distribution of retention half-
times;
6. At the upper end of the distribution of metabolic
activation rate constants (although this factor
probably is of little significance for elemental
compounds such as arsenic, nickel and chromium; it only
is significant for chemicals requiring bioactivation);
7. Experiencing depression and/or stress;
8. With concurrent respiratory tract infections and
bronchitis (disturbing pulmonary clearance or producing
scarring which has been associated with promotional
effects);
9. With depressed DNA repair rates (e.g. xeroderma
pigmentosum;
10. Significant histories of smoking, which has been
demonstrated to greatly increase susceptibility to
several of the carcinogens considered here). Active
smokers are likely to have a Lifetime Probability of
Cancer from exposure to these pollutants which is
higher than for non-smokers.
In addition to these factors which predispose individuals,
and which account for intersubject variability of
exposure-response relationships, it should be noted that there
are both synergistic and antagonistic effects of exposures. A
clear synergistic effect already has been noted in the case of
smoking; others include concurrent exposures to viruses and
alcoholic beverages. The CSFs employed in this study have been
developed primarily from occupational studies. These studies
often include concurrent exposures, which have not always been
accounted for in the epidemiological analysis. As a result, it is
not possible at present to quantify the effect of these
concurrent exposures, or even to determine whether their presence
will cause an under or over-estimation of the risks following
exposure of the general U.S. population.
The effects of some of these factors on inter-individual
susceptibility have been discussed in the analysis of variability
in Section G.4.3. Quantitative estimates of variability in CSFs
are more difficult to obtain, although it should be noted that
much of the intersubject variability in this factor usually
derives from differences in metabolic activation that will be
relatively unimportant here. Still, the NRC2 review indicates
G-113
-------
that variability of the CSF should be approximately lognormal
with a geometric SD on the order of 2.0 to 3.0. A similar value
was used in the present analysis (Section G.4.3.2.).
It must also be recognized that use of linearized
extrapolation models has been established as policy within the
Agency, at least until such time as other extrapolation functions
have' been shown to be appropriate. This determination has been
based partially on the assumption that dose-response equations
should be approximately linear at low doses and dose-rates.
Recent research into components of carcinogenesis other than
initiation, such as research on promotion, indicate that the
assumption of linearity (and even the assumption of a lack of
threshold) may not be correct as a general rule36, although it
might be correct in specific cases particularly when a substance
acts primarily by initiation and when there is large inter-
subject variability.18-20
The influence of alternative dose-response models, such as
distributed threshold models with no linearized term applied to
promoting agents and/or agents that produce significant
hyperplasia, generally is to lower estimates of the probability
of cancer from exposures at environmental levels below those
estimated by extrapolations using the linearized multistage
model. The possibility of thresholds, particularly for
promotional agents, introduces the possibility that confidence
intervals on the probabilities of cancer will include a
probability of 0.0 for some substances. For arsenic, nickel and
chromium, this possibility is lessened by findings of mutational
activity typically associated with initiation. Further
quantitative analysis of uncertainties introduced by alternative
dose-response models is not possible at present.
In their review, the NRC2 was divided as to the magnitude of
reasonable values of susceptibility factors (defined as the ratio
of the response in susceptible subpopulations to the average in
the general population). Some members suggested that a factor of
10 is appropriate, with others suggesting that this was too high
at present. As will be shown in Section G.5 of the present
analysis, the baseline calculations of probability of cancer fall
above the 90th percentile of most of the distributions. Since the
geometric SD for variability is on the order of 3 in this study,
the present analysis is consistent with the claim that the
baseline analysis already is representative of risks to
susceptible subpopulations. This claim is strengthened by the
lack of application of a dose-rate effectiveness factor which
accounts for the generally lowered susceptibility of individuals
to a given dose when the dose-rate is lowered to levels typical
G-114
-------
in environmental exposures (although such a dose-rate effect has
not been demonstrated to date for the three compounds considered
here, and the "general rule" of effectiveness increasing with
increasing dose-rate is not uniformly applicable across all risk
agents) .
G.5 EVALUATION OF UNCERTAINTY IN ESTIMATES OF RISK FROM DIRECT
INHALATION OF ELECTRIC UTILITY HAP EMISSIONS
As presented in Section G.2, risks from direct inhalation of
HAP emissions from utilities have been calculated. This section
presents the risk equations and presents overall uncertainty and
variability in the risk estimates.
G.5.1 Defining the Risk Characterization Equation
The generalized equation to estimate both individual and
population risk on a plant-specific basis can be summarized as:
= ExDxCSF [G-6]
where
Risk = excess probability of cancer (individual risk) or
cancer cases (population risk)3
E = emissions (kg/yr)
D = unit dispersion coefficient per emissions of 1
kg/yr
ug/m3/kg/yr (= maximum concentration: individual)
(ug/m3 x persons) /kg/yr (= total exposure:
population)
CSF = cancer CSF
per jag/m3 (individual risk)
per ug/m3 x persons (population risk) .
The uncertainty associated with each parameter in this
equation has been previously evaluated. Section G.3.1 addressed
the uncertainty associated with E (emissions); Section G.3.2,
with D (dispersion and exposure modeling), and Section G.4, with
CSF (CSF, exposure-response relationship) .
To obtain annual incidence, the number of cancer cases is divided by
70 years (the assumed human lifespan) to calculate annual incidence
in units of cancer cases per year.
G-115
-------
G.5.2 Uncertainty in Risk Estimates
The overall uncertainty associated with the risk estimates
can be estimated by combining the results (distributions) derived
in Sections G.3.1, G.3.2, and G.4. The forecast distributions
for each of these parameters were used as an input distribution
to calculate individual risk to the MEI and cancer incidence.
Crystal Ball was again used to conduct Monte Carlo modeling.
This was done for each of the four utility plants. The process
of deriving an overall distribution on risk estimates is depicted
in Figure G-19, and addresses the risks associated with nickel
emissions from Plant #29. Furthermore, for one oil-fired and one
coal-fired plant, the simulation was conducted using
distributions for all of the parameters directly so that an
overall sensitivity analysis could be conducted.
G.5.2.1 Uncertainty in Individual Risk Estimates. The
uncertainty in the estimates due to exposure to arsenic,
chromium, and nickel emissions from Plant #29 is statistically
summarized in Table G-19 and graphically depicted in Figure G-20.
The original risk estimates were developed using a combination of
point values — some conservative and some not conservative —
which yield a point estimate of exposure and risk that falls at
an unknown percentile of the full distributions of exposure and
risk. The distributions presented here now.give some indication
of the degree of conservatism in those calculations. Results of
the uncertainty analysis indicate that the baseline risk
estimates are conservative, but not overly conservative. The
previous point estimates of MIR are associated with about the
71st or 96th percentile on the overall distribution of possible
MIR risk estimates for arsenic (depending on which oil
concentration data are used), 87th or 98th percentile for
chromium, and 85 or 90th percentile for nickel when the
uncertainties related to emissions, dispersion and exposure
modeling, and exposure-response are incorporated. Figure G-21
presents a cumulative probability plot of MIR from Plant #29
considering the uncertainty in the MIR estimate. The MIR using
the FCEM, and SGS data are presented along with a combined plot
assuming that each of the oil concentration data sets are equally
likely. Comparison of a high-end risk descriptor (typically the
95th percentile) with a central tendency risk descriptor (median
or mean) is also a useful measure of the degree of uncertainty.
The 95th percentile (a typical high-end risk estimate) of the
overall distribution is roughly 10 to 20 times the median MIR
risk estimate for the three HAPs, indicating a geometric SD of
approximately 3.8. The 95th percentile was roughly 5 times the
mean MIR estimate for each of the three HAPs.
G-116
-------
O
6
0_
Ill
O
OJ
C3
1
Frequency
e
a
0
S
3
O
Frequency
w
3 £
S c O C
- ~ 5 D
z ^
O ^
0. '5
w c
LU Q>
c g
til z>
§?
ii5£
G-117
-------
TableG-19. Distribution of MIR and Annual Incidence: Plant #29:
Comparison of FCEM and SGS Concentration Data
MIR, Plant #29
Uncertainty
Arsenic Chromium Nickel
Mean
Initial Point Estimate
(percentile)
Percentiles:
0.0%
2.5%
5.0%
10%
25%
50%
75%
90%
95.0%
97.5%
100.0%
Ratio
95th: mean
95th: median
95th: baseline
FCEM
1E-07
(96)
2E-12
3E-09
6E-09
1E-08
3E-08
6E-08
8E-08
2E-07
5E-07
1E-06
3E-05
SOS
6E-07
6E-07
(71)
1E-09
8E-09
2E-08
3E-08
7E-08
2E-07
7E-07
1E-06
2E-06
4E-06
2E-05
FCEM
1E-07
(87)
2E-10
2E-09
3E-09
5E-09
1E-08
4E-08
1E-07
3E-07
5E-07
7E-07
6E-06
SGS
4E-08
2E-07
(98)
1E-11
7E-10
1E-09
3E-09
7E-09
1E-08
2E-08
8E-08
1E-07
2E-07
6E-06
FCEM
2E-06
(90)
2E-09
2E-08
5E-08
9E-08
2E-07
6E-07
2E-06
4E-06
7E-06
1E-05
9E-05
SGS
3E-06
4E-06
(85)
6E-09
4E-08
6E-08
1E-07
3E-07
9E-07
3E-06
6E-06
1E-05
2E-05
7E-05
0.8
8.7
3.5
4.1
10.6
3.8
2.3
12.8
4.2
Variability
0.7
9.6
3.9
1.7
11.1
4.0
2.5
10.4
3.8
Mean
Initial Point Estimate
(percentile)
Percentiles:
0.0%
2.5%
5.0%
10%
25%
50%
75%
90%
95.0%
97.5%
100.0%
Arsenic
FCEM SGS
1E-07 6E-07
6E-07
(95) (68)
5E-12 4E-09
1E-09 1E-08
3E-09 2E-08
6E-09 4E-08
1E-08 1E-07
3E-08 3E-07
6E-08 7E-07
2E-07 2E-06
5E-07 2E-06
1E-06 3E-06
1E-05 2E-05
FCEM
1E-07
(90)
4E-10
2E-09
3E-09
6E-09
1E-08
4E-08
1E-07
3E-07
4E-07
6E-07
3E-06
Chromium Nickel
SGS
3E-08
2E-07
(97)
1E-11
3E-10
7E-10
1E-09
3E-09
7E-09
3E-08
7E-08
1E-07
2E-07
2E-06
FCEM SGS
2E-06 2E-06
4E-06
(90) (90)
5E-09 2E-08
3E-08 6E-08
6E-08 1E-07
1E-07 2E-07
3E-07 4E-07
7E-07 1E-06
2E-06 3E-06
4E-06 6E-06
6E-06 9E-06
9E-06 1E-05
4E-05 5E-05
FCEM = Original oil concentration data, cSstribution defined by probability plotting technique.
SGS = Subsequent data, trace metal analysis from samples collected for nuSonucttde analysis.
Combined = Combined forecasts assuming equal probability of the FCEM and SGS data sets.
Initial Point Estimate = The estimate of emissions used in the baseline exposure assessment.
This value was based on the average concentration in the FCEM data.
(Percentile) = The percentile of the predicted distribution corresponding to the initial point estimate.
G-118
-------
03
ra
D
o
1
"c
a)
u
o
O
•q
03
LLI
O
u..
c>
as
Q.
o
o
of
C\J
JS
Q.
Frequency
I
E? 5 5 S 8
3 Affliqtqud
Frequency
1
1
R
!
f]
i
1
s
E
3
i
1 !
Freq
i
i
AMI
ticnc
3
s
•q!
OS 8 ? 1
o
w
.0
rt
b
C)
C\J
6
O)
I
2
D
c
s
UJ
I
f
E
1
S
l!
1
i
3
5 E
Freq
!
3
Atinc
uenD
e
"
pqu
c
J
i
d
*
-
I
•I
!
D
"
i
I
§
I
(A
Frequency
; 8 ?
it
8 S
JS
a
o>
CM
s I
o
u
UJ
|
• I
i.
IS
il
i
5
3
5 i
Freq
i
i
AMI
ucnc;
S
'
i*qu
0
^j
1
d
i
il
z
"
n
To
Q
I
i
(0
G-119
-------
Figure G-21. Cumulative Probability Distributions of Uncertainty in MIR: Plant #29
o.oo
OEfOO 5E07 1&06 2&06 2E06 3E06 3E-06 4&06 4E06
Maximum Individual FUk: Araente, Plant#29
0.00
0&-00 1&07 2E07 3&07 4&07 5E07 6&07 7&07 8&07 9E07
Maximum Individual FUk: Chromium, Plant #29
Combned
SQSConc Data
FCBJt Cone Data
OEfOO 2B06 4&06 6E06 8EO6 1S05 1E05
Maximum Individual Hak: Nickel, Rant #29
1E05
2EO5
G-120
-------
For comparison, a summary of the uncertainty in the
estimates due to exposure to arsenic, chromium, and nickel
emissions from Plants #343, 133, and 240 are statistically
summarized in Table G-20. In general, the results support the
conclusions that the baseline assessment resulted in reasonably
conservative estimates of risk while still being within the range
of plausible values. For the coal-fired plants (343 and 240) the
point estimates of MIR generated during the baseline assessment
were associated with the 91st to 94th percentile on the overall
distribution of possible MIR risk estimates. The 95th percentile
(a typical high-end risk estimate) of the overall distribution is
roughly twice the original risk estimate, four times the mean and
within an order of magnitude to the median MIR risk estimate.
For oil-fired Plant 133 the results were similar, though the
uncertainty was a little more broad, most likely attributable to
which of the oil concentration data sets are used. The original
estimates of MIR from the baseline assessment were between the
86th and 99th percentile on the estimated overall distribution of
possible MIR values.
G. 5 .2 .2 Uncertainty in Population Risk Estimate. The
process used to derive overall distributions on the risk
estimates for MIR was repeated for annual cancer incidence. The
distribution of possible estimates of annual cancer incidence due
to emissions from Plant #29 is statistically summarized in Table
G-21 and graphically depicted in Figure G-22. As with the
estimates of MIR, the point estimates for annual cancer incidence
associated with Plant #29 emissions generated in the risk
assessment are at the upper end of the distribution (near the
95th percentile). Furthermore, the 95th percentile was about 10
times the median, and about 5 times the mean.
For comparison, the distribution of possible estimates of
annual cancer incidence due to emissions from coal-fired Plant
#343 is statistically summarized in Table G-22. As with the
estimates of MIR, the point estimates for annual cancer incidence
associated with Plant #343 are at the upper end of the
distribution (near the 97th percentile). Furthermore, the 95th
percentile was about 8 to 30 times the median, and about 4 times
the mean.
G.5.2.3 Uncertainty in Total Annual Cancer Incidence Across
Plants. The total annual cancer incidence associated with all
utilities is the sum of incidence associated with individual
plants. The above analysis focuses on the uncertainty in cancer
risks associated with individual plants. An additional
simulation was conducted to evaluate the uncertainty associated
with the estimate of total annual cancer incidence across all
G-121
-------
CO
CM
CO
CO
«p
CO
UJ
o>
CO
CO
CO
co
UJ i-
o o>
UJ rt
cn en
CD
CO
O)
IV
IV
LU CM
O> CD
IV
CO
S^
CM ri
rv op
UJ UJ
CO »•;
coco
LULU
v P r » K » q
CM v CM
.LULLIUUJLJJ
LULUUJ
UJLU nSUJUJUJLUUJUJUJUJUJUJ
on® ^'T~®®v>ojcjsaQeo^rv
LUnj ^[JjiTjujiMLUUJUJUJUJUl
^eo CMri.h.inco1'?"?^'1:"0?
•" * *:a)rrimNa'N'fr~m
UJ UJ [TJUJUJUJUJUJUJUJIUIJUJ
m co -*r irt en r^. «> «— in «^ in m en
UJ LJJ jLULULULJJUJlijLLILJJUJUJ
%.%
"~ CM
LU UJ
00 O
CO IV
rv cp
UJ UJ
Oi - ^
enS^rir^^^^^"^
lyLULULUUJUJUJUJUJUJUJ — S.
js m Tf co cn o in co Ti* ^ co ^ *^
,XTl:cd'-:CO'«-:CMlrieOr-'cD •S'S
Jicnojcncpoprvivcpcpin
UJUJUJLULUUJUJUJUJUJ LJJO
»
CO
CM
rv cp
UJ UJ
COT-" T_T|:r^'-co'od
LU UJ L1J LU LU
^r^coco^
cMIOCCi»-:CM'
^o
.iLU
cncp
LULJJ
coco
LULU
ivh-cocoin
UJUJUJUJUJ
"S
is
£. °
ll
*" P
• 5S
CO C
(0 0
i
CO
P
^C)
£•
i i
21
gl
o
"• .Si
I
*
_
s
.s
I
r -
11
sl
(9
(0 .2
C
(T
s
0
"5
•X
^k -^
£• c
t E
3C c
c
o .a
U* C
<
in
LJJ
CO
en
CM
in
LU CM
in co
en
CD
LU CO
CD en
in
in
LU CO
en co
CM
in
LU tv
in co
cri
CD
LU en
CD CO
in
in
LU cn
a> en
CM
<•
UJ
'"
CO
LU
rv
CM
in
LU
CD
CM
T
UJ
o
»•
CO
LU
00
*
•p
LJJ
*~
T,
LU
CO
^
CD
UJ
CO
CM
in
LU
CM
CO
„
LU
CM
'-
•*"
CO
LU
in
fv
co
LU
in
00
LU
in
T-
^
LU
CO
CM
IV
11 1
|v
en
CO
LU
cn
*
CO
LU
rv
CO
CO
LU
CM
"~
cp
LU
O
CM
rv
LU
00
CM
^
Ul
CD
O)
CO
LU
in
cn
.s.
LU
in
»-
CD rv
••- •»
CO tv
LJJ LU
T- CM
cn CM
I- rv
UJ LU
|v CO
" "*
CP CD
LU LU
TT en
CM in
CO 00
LU UJ
o cn
« en
r-. CD
LU LU
»r co
m -
(D CO
LU LU
cn CD
*• ^t
iv rv
UJ UJ
cn co
*• "*
rv rv
LU LU
o in
CO tv
in in
LU LU
CO CM
CM |v
rv r-.
LU LJJ
in co
T- Tt
CD in
LU LU
"* "
in in
LU LU
in CM
«- in
tv CO
LU LU
o> t
CD CM
rv co
LU LU
iv co
CO T-
in in
LJJ LU
CO »-
i- CO
IV |v
LU LJJ
0 T
CM f
CD in
LU LU
O) CO
CO i-
in in
LU LU
CO CO
•^ •»
rv co
LU LU
CM CM
CO CM
CO CO
LU LU
in co
i- CM
Tf «f
LU LU
iv cn
T- CM
CO CD
LU LU
Tf |v
i- CM
in in
So *r
co in
Tt Tt
LU LU
CO r-
T- CM
co in
LU UJ
T- O
CD «-
CD CO
LU LJJ
•* 0>
* t
LU LU
K CO
r- CM
CD CD
LU LU
f Iv
r- CM
in in
UJ til
TT CO
co in
t •«
LU LU
CM *-
i- CM
co in
LU LU
en o
in i-
CD CD
LU LU
CO CM
Tf cn
« CO
LJJ LU
CO CO
•0- CM
cp in
LJJ LU
in i-
in in
in Tt
LU LU
CO O
CO Tf
TJ- CO
LU LU
i- CO
CO T-
in in
LU LU
in co
r- CO
in TT
LU LU
CO i-
" *
>» CO
LU LU
f CJ
Tt CM
co in
LU LU
1- CO
m •*
LU LU
m CM
cci uri
f CO
UJ LU
Tf CM
CO CM
in •«•
LU LU
co in
T- CO
m •«•
UJ til
Iv CO
•i- rv
G-122
-------
Table G-21. Distributions of Annual Cancer Incidence Plant 29
Comparison of FCEM and SGS Data
Annual Incidence Plant 29
Mean
(percentile)
Percentiles:
stimate
0.0%
2.5%
5.0%
10%
25%
50%
75%
90%
95.0%
97.5%
100.0%
FCEM
1E-03
(96)
1E-07
2E-05
4E-05
8E-05
2E-04
4E-04
8E-04
2E-03
6E-03
1E-02
2E-01
arsenic
SOS
7E-03
8E-03
(75)
5E-05
3E-04
6E-04
9E-04
2E-03
4E-03
8E-03
2E-02
2E-02
3E-02
2E-01
FCEM
1E-03
(92)
4E-06
4E-05
7E-05
1E-04
3E-04
6E-04
1E-03
3E-03
4E-03
6E-03
3E-02
Uncertainty
chromium
SGS
4E-04
3E-03
(98)
3E-07
5E-06
1E-05
2E-05
5E-05
1E-04
3E-04
9E-04
1E-03
3E-03
3E-02
FCEM
2E-02
(95)
1E-04
1E-03
2E-03
2E-03
5E-03
1E-02
2E-02
4E-02
6E-02
9E-02
4E-01
nickel
SGS
3E-02
6E-02
(89)
1E-04
1E-03
2E-03
4E-03
8E-03
2E-02
4E-02
7E-02
1E-01
1E-01
9E-01
Mean
(percentile)
Percentiles:
stimate
0.0%
2.5%
5.0%
10%
25%
50%
75%
90%
95.0%
97.5%
100.0%
FCEM
1E-03
(96)
2E-07
2E-05
3E-05
6E-05
2E-04
3E-04
9E-04
3E-03
6E-03
1E-02
1E-01
Variability
arsenic
SGS
8E-03
8E-03
(72)
8E-05
6E-04
8E-04
1E-03
2E-03
5E-03
9E-03
2E-02
2E-02
3E-02
2E-01
chromium
FCEM SGS
2E-03 7E-04
3E-03
(80) (96)
4E-06 5E-07
1E-04 9E-06
2E-04 2E-05
3E-04 3E-05
6E-04 9E-05
1E-03 2E-04
3E-03 6E-04
5E-03 2E-03
8E-03 3E-03
1E-02 4E-03
5E-02 3E-02
nickel
FCEM SGS
2E-01 3E-01
6E-02
(15) (26)
4E-03 5E-03
2E-02 2E-02
2E-02 3E-02
3E-02 5E-02
6E-02 9E-02
1E-01 2E-01
2E-01 3E-01
4E-01 6E-01
6E-01 8E-01
8E-01 1E+00
3E+00 8E+00
Initial Point Estimate = The estimate of emissions used in the baseline exposure assessment.
(Percentile) = The percentile of the predicted distribution corresponding to the initial point estimate.
G-123
-------
I
o
01
2
T3
i
CO
o
CO
**-
o
I
co
Q.
O
o
CM
CO
O
I
u
8
co
c
-
CN
- c
3 JO
C
cvi
6
9
Frequency
I 8 I
< o
a o
m O
co ~;
« S
W III
O O
LL u.
i
:i
•c
i
i
jj
f *
Si
i;
tk
5
g
Frequency
5 81 .
! J
5 8 g s i
J
-s
*
S
!
e
§ Frequency
J, , . .
i S 8
I
i
s
CO
o
CO
Frequency
Q.
IB
C «
0> %
2 Q
U r-
= I
« CB
3 ^
I*
< O
i: o
S°
P S
S iu
i
I
I Frequency
li I £
o 3
1
Frequency
Sill
_o
1
I
o
u
co
O
CO
G-124
-------
Table G-22. Summary of Results of Monte Carlo Simulation of
Annual Cancer Incidence (and MIR) for Coal-Fired Plant #343
Forecast: Annual Incidence, Plant 343 (cases/year)
Incidence, Uncert. Incidence, Var
arsenic chromium nickel arsenic chromium nickel
Mean 4.5E-4 3.3E-4 9.7E-6 4.8E-4 6.3E-4 1.0E-5
Initial Point Estimate 3.1 E-3 2.6E-3 1.0E-4 3.1 E-3 2.6E-3 1.0E-4
(Percentile) 97 99 98 97 96 97
Standard Deviation
Mean Std. Error
Percentile
0.0%
2.5%
5.0%
10%
25%
50%
75%
90%
95.0%
97.5%
100.0%
1.2E-3
2.2E-5
6.8E-17
2.9E-6
5.9E-6
1.2E-5
2.9E-5
6.9E-5
3.3E-4
1.1 E-3
2.2E-3
3.6E-3
1.7E-2
5.6E-4
1.0E-5
2.0E-7
7.0E-6
1.4E-5
2.7E-5
6.4E-5
1.5E-4
3.6E-4
8.0E-4
1.2E-3
1.8E-3
1.0E-2
2.1E-5
3.9E-7
1.2E-9
1.3E-7
2.6E-7
5.2E-7
1 .3E-6
3.6E-6
9.7E-6
2.3E-5
4.0E-5
6.0E-5
4.6E-4
1.4E-3
2.5E-5
1.4E-16
3.5E-6
7.0E-6
1.4E-5
3.5E-5
7.7E-5
3.7E-4
1.1 E-3
2.2E-3
3.6E-3
2.0E-2
1.0E-3
1.9E-5
1.0E-6
1.5E-5
3.0E-5
5.7E-5
1.3E-4
3.0E-4
7.1 E-4
1.5E-3
2.2E-3
3.2E-3
1.5E-2
2.0E-5
3.6E-7
2.3E-9
1 .3E-7
2.6E-7
5.2E-7
1.4E-6
4.0E-6
1.0E-5
2.3E-5
3.8E-5
5.8E-5
3.2E-4
Original Point Estimate = Estimate derived in the baseline assessment using EPA default values
Percentile = the percentile of the original point estimate within the Monte Carlo derived distribution
G-125
-------
utilities. For this analysis, the distribution of annual cancer
incidence for each individual plant was sampled and summed to
estimate the total cancer incidence for those four plants.
Figure G-23 and Table G-23 presents these results. Again, as
indicated in Table G-23, the results indicate that the original
estimate from the baseline risk assessment is reasonably
conservative, ranging from the 92nd to 97th percentile of the
overall distribution on total cancer incidence for those four
plants. Furthermore, the uncertainty in the total annual
incidence is somewhat less than the uncertainty associated with
individual plants. This is due to the fact that for any one
plant which experiences a high distribution value it is likely
that another plant would have a lower distribution value thereby
more approach the central tendency. The 95th percentile of the
distribution for total cancer incidence was typically within
about a factor of 3 to the mean and three to six to the median,
both less than would be expected for individual plants.
G.5.2.4 Sensitivity Analysis. Crystal Ball has the
internal capability to conduct a sensitivity analysis to
determine which parameter most influences the resulting
distribution. Crystal Ball calculates sensitivity by computing
contribution to variance between every assumption (parameter) and
forecast (output) while the simulation is running. A sensitivity
analysis was conducted for the estimation of uncertainty in MIR
and incidence. For all three HAPs, the dispersion coefficient
for maximum concentration contributed most to overall uncertainty
followed by the CSF correction factor (SF), the EMF, and
speciation correction factor (S). The trace element
concentration in oil, breathing rate adjustment (BR), and oil
consumption contributed least to uncertainty. Exposure frequency
adjustment (EF) is the additional factor used in estimating
incidence and only moderately contributed to overall uncertainty
compared to the other parameters identified above.
G.5.3 Summary
The baseline risk assessment conducted by EPA on HAP
emissions from utilities generated point estimates of risk. The
procedure used in that study was intended to be generally
conservative, more likely to overestimate than underestimate
actual risk. The methods used in that analysis relied on a
combination of point values, some conservative and some not
conservative, yielding a point estimate of exposure and risk that
falls at an unknown percentile of the full distributions of
exposure and risk.
A quantitative uncertainty analysis was conducted and is
presented in this Appendix. The large number of HAPs and utility
G-126
-------
m
CO
CO
s
Oi
CM
w
c
I
co
u
<
-------
Table G-23. Summary of Monte Carlo Simulation of Estimating Total Annual Cancer Incidence
Across Four Plants (Plants 29, 343, 240,133)
Forecast: Total Annual Cancer Incidence (cases/year)
(Combined estimates from Plants, 29, 343, 240,133)
Statistics:
Mean
Standard Deviation
Mean Std. Error
Percentile
0.0%
2.5%
5.0%
10%
25%
50%
75%
90%
95.0%
97.5%
100.0%
Baseline Estimate
Percentile
Arsenic
0.0057
0.0085
0.0002
0.0001
0.0001
0.0002
0.0004
0.001 1
0.0031
0.0069
0.0128
0.0196
0.0277
0.1011
0.015
92
Chromium
0.0013
0.0016
0.0000
0.0000
0.0001
0.0002
0.0003
0.0004
0.0009
0.0016
0.0030
0.0042
0.0056
0.0205
0.0067
97
Nickel
0.0302
0.0315
0.0006
0.0018
0.0042
0.0051
0.0071
0.0118
0.0205
0.0369
0.0617
0.0848
0.1183
0.3426
0.072
93
Total
0.0373
0.0328
0.0006
0.0026
0.0073
0.0090
0.0114
0.0174
0.0281
0.0457
0.0719
0.0965
0.1241
0.3537
0.094
95
Ratio
95th percentile: mean 3.4
95th percentile: median 6.3
95th percentile : baseline 1.3
3.1
4.8
0.6
2.8
4.1
1.2
2.6
3.4
1.0
G-128
-------
plants did not allow for an analysis of all HAPs or plants.
Therefore, the analysis was limited to three HAPs (arsenic,
chromium, and nickel), which accounted for over 95 percent of the
total cancer incidence, and a subset of plants selected as the
major contributors to incidence and which were responsible for
the highest MIR. The risk assessment process was divided into
three distinct components (emissions estimation, dispersion and
exposure modeling, and exposure-response) and the uncertainty was
analyzed independently in each. Overall distributions on the
uncertainty in the risk estimates were generated by combining the
uncertainties of the three components. Figure G-24 identifies
the major parameters and identifies the original assumptions used
in the baseline case and summarizes information on the parameter
values and associated uncertainty.
The results indicate that the point estimates of risks
generated in the original risk assessment were at the upper end
of overall distribution, usually around the 95th percentile (and
ranged from the 71st to 99th percentile). In evaluating risks,
the 95th percentile is typically used as a high-end risk
descriptor and the median or mean is used to describe central
tendency measures of risk. The 95th percentile was about 10
times the median value on the distribution, and roughly 5 times
the mean estimate of risk. A sensitivity analysis indicated
that the parameters that contribute most to the overall
uncertainty were the dispersion coefficient for maximum
concentration followed by the CSF correction factor (SF), the
EMF, and speciation correction factor (S). The trace element
concentration in oil, breathing rate adjustment (BR), and oil
consumption contributed least to uncertainty. Exposure frequency
adjustment (EF) -- the additional factor used in estimating
incidence -- only moderately contributed to overall uncertainty
compared to the other parameters.
It should be noted that the uncertainty estimates presented
here in this report are likely to underestimate the true overall
uncertainty for several reasons. The major reasons for
underestimation of uncertainty include the approach used in the
analysis (focusing on parameter uncertainty), limited data, and
the potential for unsuspected errors. Each is described below.
This uncertainty analysis focused on parameter uncertainty
alone and did not quantify uncertainty related to model choice
which in many cases (e.g., dose-response modeling) can lead to
dramatic and significant differences in risk estimates.
Experience indicates that the HEM and other models used by EPA to
estimate dispersion and exposure are limited in their ability to
estimate short-term concentrations and locations of maximum
G-129
-------
O)
CM
4^
JO
a.
I
CD
P
O
CD
O
03
"
UJ
c
.2
(0
<0
"5
t:
I
(A
1
cc
0)
CO
£
3
D)
LL
o>
CM
CD
cr
c z
1 I
11
l\
o>
I
5 E
I I
fe"
I
I
z
s g
M
l!
£
I
Z!
I>
UJ
O
Z
1
I i
5 g
I I
£
CM O O O O O
ui cr c
! a, i *
8
i
i i
§ I
Illl.
I
e
G-130
-------
concentrations. However, these models are more accurate when
predicting long-term average concentrations and exposure,
typically with a coefficient of variation of about 10 to 40
percent. The existing range of dose-response models can result
in estimates of risk which may differ by several orders of
magnitude.
The uncertainty analysis did not address all parameters
related to plant emissions due to constraints in available data.
Existing data were reported to EPA in various forms, ranging from
summary reports to raw data. Therefore, the available data did
not allow for the separate evaluation of uncertainty and
variability, given that sampling and analysis error could not be
quantified. Furthermore, data did not support the simulation of
many factors related to plant emissions, particularly in fuel
consumption. The rate of fuel consumption, and the fuel mix
consumed is expected to vary dramatically over time related to
plant operations and electricity demand. This will have a
significant impact on the pattern of emissions and flue gas
parameters (i.e., exit velocity and flue gas temperature). The
type of fuel consumed would be expected to be influenced by
sulfur content, seasonal demand, cost, and availability. These
factors may result in a higher degree of variation in emissions
and subsequent exposures than predicted. However, while these
factors would result in high degrees of short-term variation it
is unknown how this may impact the long-term average emissions.
The risk assessment and uncertainty analysis focused on the
cancer risks associated with direct inhalation of HAP emissions
within 50 km of plants. Additional risks and concomitant
uncertainties also exist for indirect exposures and long-range
transport, and re-entrainment of and subsequent exposure to
deposited particulates. This analysis was also limited to long-
term average exposures and did not account for the expected
short-term variations in exposure which may influence noncancer
risks, and total dose to carcinogens which are dose rate
dependent.
Evidence also indicates that commonly used 95 percent bounds
for normal and lognormal distributions of variables are very
sensitive to the underestimation of the true uncertainty and
typically do not address unsuspected errors. Empirical
probability distributions of the normalized deviations of the
measured quantities from the true values (in several data sets
from fields of nuclear physics, energy, and population
projections) do not follow the usually implied normal
distribution. These distributions are better described by an
exponential distribution.37 If much of the uncertainty comes
G-131
-------
from systematic errors, the usual justification for the normal
distribution between observed deviations of measured quantities
from the true values does not apply.
Unsuspected errors are quite common, and new measurements
are often far from the previous values. Environmental
measurements are rarely repeated with the same samples, and it is
hard to estimate how widespread the unaccounted errors are in
routinely collected data. One can hedge against unsuspected
uncertainties by multiplying the reported uncertainty range by a
default safety factor or to solve for the exponential factor to
describe the observed deviations. This safety factor or
exponential factor can be used to expand the confidence
intervals. This approach has been used in evaluating large
number of data sets related to nuclear physics, and energy
projections37 and suggested for use in the environmental field.27
One reviewer suggested using this approach to expand the
estimates of uncertainty for this study.7 However, the
exponential approach has not yet been documented to apply to
environmental measurements or models, and to test the
applicability would require greater amounts of data than are
available for this project. Therefore, EPA recognizes the
potential for unsuspected errors contributing to the overall
uncertainty, but feels as though at present there is insufficient
information to support the estimation of the proper exponential
function to be used to expand our confidence intervals.
EPA recognizes the potential for unsuspected errors in the
data and its associated representation of uncertainty. The
manner in which the data were collected and reported can
contribute to these errors. The emissions test data were
developed by EPRI in cooperation with the EPA and the Department
of Energy. Often these test data were limited in number and did
not encompass the entire universe of facilities with regards to
plant configurations, age, and operation and maintenance.
Furthermore, with a single short-term test it is not possible to
use conventional methods to estimate or subtract out short-term
variability and measurement errors from the long-term variability
of a parameter (e.g., emissions) Also it is not possible to
estimate the possible effects of realistic variations in fuels
and plant operating characteristics that could affect long-term
emissions. Data are often collected under relatively favorable
conditions in these respects and, therefore, may not entirely
reflect the full variability of important operating parameters
that would be seen over the lifetime of these facilities.7
In conclusion, the uncertainty analysis presented here
supports the original risk estimates. Given, EPA's mandate of
G-132
-------
protecting public health, EPA risk assessments are intended to
yield conservative estimates of risk (more likely to overestimate
than underestimate). This is typically accomplished with the use
of conservative assumptions. The original estimates were shown
to be realistically conservative ranging from the 71st to 98th
percentile of the overall distributions of risk predicted from
the uncertainty analysis. Past criticisms state that the use of
compounded conservative assumptions can lead to risk estimates
beyond the range of expected values. However, the uncertainty
analysis indicates that the original estimates for this study are
well within the range of possible values.
G-133
-------
G.5.5 References
1. U.S. EPA (Environmental Protection Agency) . Guidance on
Risk Characterization for Risk Managers and Risk Assessors.
Memorandum from F. Henry Habicht, II, Deputy Administrator,
Washington, B.C. 1992.
2. National Research Council. Science and Judgement in Risk
Assessment. Washington, DC., National Academy Press. 1994.
3. Finkel, A.M. Confronting Uncertainty in Risk Management. A
Guide for Decision Makers. Center for Risk Management,
Resources for the Future, Washington, D. C. 1990.
4. National Research Council. Risk Assessment in the Federal
Government: Managing the Process. Washington, D.C.
National Academy Press. 1983.
5. U.S. EPA (Environmental Protection Agency). Risk Assessment
Guidelines of 1986. EPA-600/8-87/045. (Guidelines for
Carcinogen Risk Assessment, Guidelines for Mutagenicity Risk
Assessment, Guidelines for Health Risk Assessment of
Chemical Mixtures, Guidelines for Health Assessment of
Suspect Developmental Toxicants, Guidelines for Estimating
Exposures), Office of Health and Environmental Assessment,
Washington, DC. EPA-600/8-87/045. 1987.
6. Finkel, A.M., Confronting Uncertainty in Risk Management.
A Guide for Decision Makers. Center for Risk Management,
Resources for the Future, Washington, D. C. 1990.
7. Hattis, D. and D.E. Burmaster, "Assessment of variability
and Uncertainty Distributions for Practical Risk Analysis",
Risk Analysis, Vol. 14, No. 5, p. 713-730. 1994.
8. Frey, H.C, Separating Variability and Uncertainty in
Exposure Assessment: Motivations and Method, Proceedings of
the 86th Annual Meeting of the Air & Waste Management
Association, Denver CO, June 1993.
9. Hattis, D. Review of EPA Electric Utility Report to Congress
and Uncertainty Analysis of Direct Inhalation Risks, memo to
Chuck French, US EPA, Risk and Exposure Assessment Group,
Office of Air Quality Planning and Standards,' July 27, 1995.
G-134
-------
10. U.S. Environmental Protection Agency, Guideline on Air
Quality Models (Revised). Office of Air Quality Planning
and Standards, Research Triangle Park, NC. EPA 450/2-78-
027R. 1986.
11. U.S. Environmental Protection Agency, User's Manual for the
Human Exposure Model (HEM). Office of Air Quality Planning
and Standards, Research Triangle Park, NC. EPA 450/5-86-
001. 1986.
12. U.S. Environmental Protection Agency, Supplement A to the
Guideline on Air Quality Models (Revised). Office of Air
Quality Planning and Standards, Research Triangle Park, NC.
EPA 450/2-78-027R. 1987.
13. U.S. Environmental Protection Agency, Human Exposure Model-
II User's Guide. Office of Air Quality Planning and
Standards, Research Triangle Park, NC. 1990.
14. U.S. EPA (Environmental Protection Agency). A Descriptive
Guide to Risk Assessment Methodologies for Toxic Air
Pollutants. Office of Air Quality Planning and Standards,
Research Triangle Park, NC EPA-453/R-93-038. 1993.
15. Hofmann, W. and Daschil, F., "Biological Variability
Influencing Lung Dosimetry for Inhaled 222-Rn and 220-Rn
Decay Products", Health Physics, 50, 345-367, 1986.
16. U.S. EPA (Environmental Protection Agency). Exposure
Factors Handbook. Office of Health and Environmental
Assessment, Exposure Assessment Group, Washington, DC.
EPA/600/8-89/043, NTIS PB90-106774. 1989.
17. National Research Council, Comparative Dosimetry of Radon in
Mines and Homes, National Academy Press, Washington, DC,
1991.
18. Crawford-Brown, D. and Hofmann, W, "The Role of Variability
of Dose in Dose-Response Relationships for Alpha Emitting
Radionuclides", Radiation Protection Dosimetry, 29, 293,
1989.
19. Hattis, D., Erdreich, L. and Ballew, M., "Human Variability
in Susceptibility to Toxic Chemicals- A Preliminary Analysis
of Pharmacokinetic Data from Normal Volunteers", Risk
Analysis, 7, 415-426, 1987.
G-135
-------
20. Crawford-Brown, D., "The Role of Dose Inhomogeneity in
Biological Models of Dose-Response", in Low Dose Radiation:
Biological Bases of Risk Assessment, ed. by. K.
Baverstock and J. Stather, Taylor and Francis, pp.
155-161, 1989.
21. Crawford-Brown, D., "A Unified Approach to Age Dependent
Metabolic Modeling", Health Physics, 46, 809, 1984.
22. EPRI (Electric Power Research Institute), Electric Utility
Trace Substances Synthesis Report, Volume 4: Appendix P,
Toxicology Profiles, EPRITR-104614-V4, November, 1994.
23. ATSDR (Agency for Toxic Substances and Disease Registry),
Toxicological Profile for Arsenic, Atlanta, Georgia, 1987.
24. Beckman, G., Beckman, L. and Nordenson, L., "Chromosome
Aberrations in Workers Exposed to Arsenic", Environmental
Health Perspective, 19, 145-146, 1977.
25. Doll, R., "Report of the International Committee on Nickel
Carcinogenesis in Man", Scandinavian Journal of the Work
Environment and Health, 16, 1-82, 1990.
26. Hofmann, W. and Daschil, F., "Biological Variability
Influencing Lung Dosimetry for Inhaled 222-Rn and 220-Rn
Decay Products", Health Physics, 50, 345-367, 1986.
27. Hattis, D. and Silver, K. , "Human Intersubject Variability-
A Major Source of Uncertainty in Assessing Risks for
Noncancer Health Effects", Risk Analysis, 14, 421-431, 1994.
28. Enterline, P. and Marsh, G., "Cancer Among Workers Exposed
to Arsenic and Other Substances in a Copper Smelter",
American Journal of Epidemiology, 116, 895-911, 1982.
29. Enterline, P. and Marsh, G., "Mortality Among Workers in a
Nickel Refinery and Alloy Manufacturing Plant in West
Virginia", Journal of the National Cancer Institute, 68,
925-933, 1982.
30. Hayes, R., Sheffet, A., and Spirtas, R., "Cancer Mortality
Among a Cohort of Chromium Pigment Workers", American
Journal of Industrial Medicine, 16, 127-133, 1989.
G-136
-------
31. Lee-Feldstein, "Arsenic and Respiratory Cancer in Man:
Follow up of an Occupational Study", in Arsenic: Industrial,
Biomedical and Environmental Perspectives, ed. by W.
Lederer and R. Fensterheim, Van Nostrand Reinhold, New
York, NY, 1983.
32. Welch, K., Higgins, M. and Burchfiel, C., "Arsenic
Exposure, Smoking and Respiratory Cancer in Copper Smelter
Workers", Archives of Environmental Health, 37, 325-335,
1982.
33. Brown, C. and Chu, K., "A New Method for the Analysis of
Cohort Studies: Implementations of the Multistage Theory of
Carcinogenesis Applied to Occupational Arsenic Exposure",
Environmental Health Perspective, 50, 293-308, 1983.
34. Virens, J. and Silvers, A., "Unit Risk Estimates for
Airborne Arsenic Exposure: An Updated View on Recent Data
from Two Copper Smelter Cohorts", Regulatory Toxicology and
Pharmacology (in press), 1995.
35. Mancuso, T, "Consideration of Chromium as an Industrial
Carcinogen", Proceedings of the International Conference on
Heavy Metals in the Environment, ed. by T. Hutchinson,
Institute for Environmental Studies, Toronto, Canada, pp.
358-363, 1975.
36. Crawford-Brown, D. And K. Brown, "An Integrative Approach to
Rational Discourse in Carcinogen Hazard Identification", in
Trace Substances in Environmental Health, ed. by R. Cothern,
Science Reviews Limited, 1994.
37. Shlyakhter, A.I., "An Improved Framework for Uncertainty
Analysis: Accounting for Unsuspected Errors", Risk Analysis,
Vol. 14, No. 4, p. 441-447. 1994.
G-137
-------
TECHNICAL REPORT DATA
(Please read Instructions on reverse before completing)
1 REPORT NO
EPA-453/R-96-013b
3. RECIPIENT'S ACCESSION NO
4 TITLE AND SUBTITLE
Study of Hazardous Air Pollutant Emissions from Electric
Utility Steam Generating Units — Interim Final Report
Volume 2. Appendices A - G
5. REPORT DATE
October 1996
6. PERFORMING ORGANIZATION CODE
7 AUTHOR(S)
8 PERFORMING ORGANIZATION REPORT NO.
9 PERFORMING ORGANIZATION NAME AND ADDRESS
10. PROGRAM ELEMENT NO.
U.S. Environmental Protection Agency
Emission Standards Division/Air Quality Strategies and
Standards Division
Office of Air Quality Planning and Standards
Research Triangle Park, NC 27711
11. CONTRACT/GRANT NO.
i: SPONSORING AGENCY NAME AND ADDRESS
13 TYPE OF REPORT AND PERIOD COVERED
14. SPONSORING AGENCY CODE
SUPPLEMENTARY NOTES
In ABSTRACT
This report has been prepared pursuant to section 112(n)(l)(A) of the Clean Air Act, and provides the
Congress and the public with information regarding the emissions, fate, and transport of utility HAPs.
The primary components of this report are: (1) a description of the industry; (2) an analysis of emissions
data: (3) an assessment of hazards and risks due to inhalation exposures to numerous HAPs (excluding
mercury); (4) an assessment of risks due to multipathway (inhalation plus non-inhalation) exposure to
radionuclides; and (5) a discussion of alternative control strategies. The assessment for mercury includes a
description of emissions, deposition estimates, control technologies, and a dispersion and fate modeling
assessment which includes predicted levels in various media based on modeling from four representative
utility plants using hypothetical scenarios. The EPA plans to publish a final report at a later date which
will include (1) a more complete assessment of the exposures, hazards, and risks: (2) conclusions, as
appropriate and feasible, regarding the significance of the risks and impacts to public health; and (3) a
determination as to whether regulation of utility HAPs is appropriate and necessary.
KEY WORDS AND DOCUMENT ANALYSIS
DESCRIPTORS
b. IDENTIFIERS/OPEN ENDED TERMS
c. COSATI Field/Group
Air Pollution
Atmospheric Dispersion Modeling
Electric Utility Steam Generating Units
Hazardous Air Pollutants/Air Toxics
Air Pollution Control
18 DISTRIBUTION STATEMENT
Release Unlimited
19. SECURITY CLASS (Report)
Unclassified
21 NO. OF PAGES
258
20. SECURITY CLASS (Page)
Unclassified
22 PRICE
EP\ Form 2220-1 (Rev. 4-77) PREVIOUS EDITION IS OBSOLETE
-------
"— Bl
.lbrary
U.S. Environmental Protection Agency
Region 5, Library (PL-12J)
77 West Jackson Boulevard, 12th Floor
Chicago, II 60604-3590
------- |