8458
Columbia
Road
(
440)
263­
8214
/
FAX:
{
W)
2354982
Email:
92riPo@
rrai.
com
/
Website:
www.
rrai.
com
0
Olmsted
Falls,
Ohio
44138­
2206
December
28,2001
whitman,
Administrator
ai
Protection
Agency
Headquarters
Washington,
D.
C.
2046
Subjecf:
Enforc
of
MACT
emissions
limitations
given
emissions
testing
methods.

Dear
Administrator
Whitman,

In
my
August
30,
1999
tetter
to
Administrator
Browner
on
this
subject,
I
explained
that
Document
(
TSD)
for
the
HWC
MACT
Rule
erroneously
stated
d
agreed
with
EPAs
assessment
of
the
expanded
uncertainty
of
methods,
the
range
of
a
d
emissions
represented
by
a
stack
i
at
there
were
significant
errors
in
the
TSD
analysis.

On
November
9
1999,
OSW'Director
Cotsworth
wrote
me
suggesting
that
review
of
the
rred
pending
completion
of
the
hexican
Society
of
Mechanical
h
Committee
on
Industrial
and
M
~
c
i
p
a
l
Waste's
ReMAP
(
Reference
and
Precision)
program
and
asked
that
I
involve
EMC's
Mr­
William
H.
Lamason
in
tie
foilow­
up.
,
­

The
inm
1
ReMAP
report
confirmed
the
methodologid
deficiencies
1
identified
1999
letter
and
identified
an
additional
problem
with
the
TSD
analysis.

In
­
A's
response
to
some
of
the
comments
I
made
on
the
Small
MWC
MACT
M
e
g
the
expanded
uncertainty
of
manual
sampling
methods
(
the
range
of
contain
the
measurand),
the
Agency
stated
that
dioxin
test
y
promulgated
and
that
measurement
uncertaimy
was
considered.
response
does
not
address
the
concern
because
EPA
never
established
fication
limit
for
Method
23
(
or
RCRA
Method
0023
for
that
#

d
could
not
have
because
the
Method
23
validation
oIumes
needed
to
calculate
concentration.
Hence,
the
capability
of
Method
23
at
the
regulatory
concentrations.

Environrnenra1
and
Statistical
Consuiting
c
Subject:
Enforceability
of
MACT
emissions
limitations
aven
the
uncertainty
of
manual
Page
2
December
28,2001
emissions
testing
methods.

Because
of
my
concerns
about
enforceability,
I
expanded
a
demonstration
project
I
was
conducting
in
November
1997
to
include
simultaneous
Method
23
sampling.
I
gave
those
data
to
EPA
and
ASME
used
them
in
ReMAP.
I
recently
incorporated
the
results
of
a
few
more
simultaneous
Method
23
tests
into
an
updated
assessment
of
dioxin
measurement
uncertainty,
which
I
have
expanded
to
include
"
 3098
dioxin
toxic
equivalency
as
well.
The
range
of
the
data
now
covers
the
entire
range
of
regulated
concentrations
instead
of
ing
at
27.6
ng/
dsm3
(
no
diluent
correction).
Those
updated
results
are
attached
to
­

I
determined
the
relationship
between
the
standard
deviation
of
simultaneous
test
results
d
concentration
using
the
procedures
developed
in
ReMAP.
That
relationship
d
along
with
Monte
Carlo
Simulation
(
MCS)
to
determine
the
credible
range
of
standard
deviations
and
measured
values
likely
to
be
observed
(
Cox,
M.
G.,
M.
P.
Dainton
and
P.
M.
Harris,
So@
are
Support
for
Metwhgy
Best
Practice
Guide
No.
6,
Uncertainty
and
Stdistical
Modeling,
National
Physics
Laboratory,
Toddington,
March
2001).
Specific
characteristics
of
the
measurement
method
such
as
the
limit
were
extracted
from
the
Monte
Carlo
Simulation
results.
the
attachment
to
this
letter
conform
to
the
A
m
k
a
n
National
Uncertainty
­
U­
S.
Guide
to
the
Expression
of
Uncertainty
in
SL
2540­
2­
199'
7).
The
data
and
results
are
attached
to
this
\

letter.

Method
301
is
the
only
promulgated
rule
establishing
the
lower
limit
of
applicability
of
an
emissions
test
method.
Both
Method
301
itself
and
the
preamble
make
it
clear
that
under
the
current
regulatory
scheme,
test
results
can
only
be
used
for
enforcement
when
above
the
practical
quantification
limit.
Of
course,
this
caveat
should
be
able
to
citly
waived
when
promulgated
emissions
limitations
explicitly
incorporate
ent
uncertainty.
I
The
statistical
concept
underlying
the
Method
30
1
definition
the
practical
quantification
limit
is
that
the
PQL
is
the
concentration
where
the
measured
concentration
(
measurement)
is
likely
to
be
within
10
percent
of
the
true,
but
unknowable
emitted
source
concentration
(
measurand).
I
used
the
statistical
concept
rather
than
the
formulas
found
in
Method
301.
because
the
HWC
MACT
TSD
correctly
states
that
the
Method
301
formulas
do
not
apply
for
manual
stack
emissions
testing
methods.
For
these
methods,
the
standard
deviation
between
simultaneous
replicate
measurements
is
neither
constant
nor
decreases
with
concentration;
rather,
it
increases
with
concent
,
Subject:
Enforceability
of
MACT
emissions
limitations
given
the
uncertzlinty
of
manual
Page
3
December
28,200
1
emissions
testing
methods.

The
practical
quantiation
limit
for
total
dioxins
is
34
&
dsm3.
Since
total
dioxin
MACT
limitations
include
7,
13
and
30
ng/
dsm3
corrected
to
7
percent
oxygen,
which
are
equivalent
to
5­
75
9­
14
and
21­
32
ng/
dsm3
on
an
uncorrected
basis
for
typical
MWCs,
it
is
clear
that
that
PQL
is
above
the
enforcement
levels.

The
new
data
did
not
materially
change
the
results
for
ITEQ
dioxins.
The
PQL
for
ITEQ
dioxins
is
1.6
ng/
dsm3.
So,
the
PQL
is
above
0.2
and
0.4
ng/
dsm3
corrected
to
7
percent
oxygen
emissions
limitations
found
in
the
now
remanded
HWC
MACT
rule.

I
remain
convinced
that
measurement
uncertainty
must
be
known
and
properly
considered
when
MACT
standards
are
established.
Absent
such,
I
am
deeply
concerned
that
truly
excessive
emissions
will
not
be
reduced
while
sources
that
are
actually
achieving
MACT
risk
being
punished
in
the
press
and
by
unwarranted
enforcement
actions
every
time
they
conduct
a
source
emissions
test.

Very
truly
yours,
H
G
Rig0
&
Associates,
Inc.

H.
Gregor
Rigo,
PhD,
PE,
QEP,
DEE
President
Attachments:
Updated
measurement
uncertainty
results
for
Total,
ITEQ
and
WHO98
TEQ
Dioxins
RCRA
Docket:
F96­
RCSP­
FFFF
Air
&
Radiation
Docket­
HWC:
A­
9045
­
MWC:
A­
98­
18
­
MWI:
A­
91­
6
­
WC:
A­
94­
63
Mi.
William
H.
Lamason,
EMC,
EPA
7
Updated
measurement
uncertainty
results
for
Total,
ITEQ
and
TEQ
Dioxins
The
following
three
sheets
are
all
similarly
arranged.
They
provide
summary
statistical
characteristics
and
measurement
uncertainty
results
for
Total
Dioxins,
ITEQ
Dioxins
and
WHO98
TEQ
dioxins.

The
upper
lee
hand
corner
of
each
page
is
a
tabulation
of
the
statistical
characteristics
of
the
data.
These
statistical
characteristics
describe
the
applicable
range
of
the
results
and
provide
the
input
parameters
needed
to
determine
measurement
uncertainty
using
either
main­
line
GUM
(
Guide
to
expressing
uncertainty
in
measurement)
or
Monte
Carlo
Simulation
WCS)
techniques.
MCS
is
the
correct
technique
to
apply
since
fbndamental
assumptions
implicit
in
main­
line
GUM
are
violated
by
the
nature
of
the
relationship
between
standard
deviation
and
concentration
for
manual
emissions
measurements.

Using
the
total
dioxins
sheet
as
an
example:

0
Simultaneous
replicate
data
has
been
obtained
between
0.5
and
399
ng/
dsm3.
The
average
and
standard
deviation
of
the
natural
logarithms
of
the
concentration
averages
are
1.
.9
and
2.06
ln(
ng/
dsm")
respectively.
of
the
sample
size
bias
corrected
standard
deviation
estimates
is
4
while
the
average
of
the
predicted
standard
deviations
is
2.6
leading
to
a
retransformation
bias
correction
factor
of
1.58.
The
natural
logarithm
of
this
factor
is
added
to
the
regression
intercept
to
correct
for
this
bias
The
retransformation
bias
corrected
equation
relating
average
concentration
and
the
standard
deviation
of
the
measurements
used
to
generate
the
average
is:
S
=
­
1.69
+
0.83(
concentration).
The
statistical
characteristics
are
based
on
27
dual­
train
samplings.

A
10,000
iteration
Monte
Carlo
Simulation
was
used
to
determine
that
for
total
dioxins:
Blank
trains
are
likely
to
yield
measured
total
dioxin
concentrations
up
to
0.07
ng/
dsm3.
The
detection
limit,
the
lowest
concentration
we
are
statistically
certain
is
greater
If
we
are
interested
in
making
sure
that
a
reported
result
has
at
least
one
significant
digit,
then
the
true
concentration
has
to
be
above
0.5
ng/
dsm3
and
the
lowest
level
at
which
two
significant
digits
may
be
properly
reported
is
24
ng/
dsm3.
than
a
blank
train,
is
0.9
ng/
dsm3.
I
The
long,
thin
dashed
line
on
the
graphs
is
the
central
value
estimated
&
om
the
data.
The
heavy
lines
demark
the
upper
and
lower
95%
(
2­
tailed)
confidence
intervals
and
the
surrounding
light
dotted
lines
indicate
the
uncertainty
range
associated
with
these
bounds.
­
2­

At
the
bottom
left
hand
comer:

0
The
relative
standard
deviation
is
plotted
against
measurand,
the
"
true"
but
unknown
and
unknowable
value
of
the
concentration.
PQL
is
the
concentration
corresponding
to
a
10
percent
relative
standard
deviation.
The
analytic
result
has
at
least
one
significant
digit
when
upper
confidence
limit
for
the
standard
deviation
equals
50
percent
and
may
have
more
than
one
significant
digit
when
the
lower
confidence
limit
for
the
standard
deviation
is
below
5
percent.
0
At
the
right
hand
side
of
each
sheet,
the
range
of
measured
values
is
plotted
against
the
measurand.
Like
a
typical
calibration
curve,
if
the
measurand
is
known
(
for
example,
when
a
standard
solution
is
analyzed)
then
reading
up
the
graph
indicates
the
range
of
Its
that
can
be
expected
using
the
Method.
In
stack
testing,
however,
the
s
are
what
we
have
and
the
measurand
is
both
unknown
and
unknowable.
If
I
conducted
5t
test
and
measured
100
ng/
dsm3,
the
total
dioxin
graphic
lets
me
figure
out
that
the
"
true"
measured
concentration
(
measurand)
is
between
80
and
130
ng/
dsm3.
Of
course,
these
concentrations
must
still
be
multiplied
by
the
dilution
correction
factor
to
express
the
result
in
the
units
of
a
diluent
corrected
regulatory
standard.
Summary
statistics
for
simultaneous
repIicate
Method
23
dioxin
measurements.

Run
ID
TO1
TO2
TO3
TO4
TO5
TO6
TO7
f
l
0
T11
T12
T13
T14
T15
T16
T17
T18
T19
EERTRCl
EERTRC2
EERTRC3
wes­
runl­
1
wes­
run
1­
2
wes­
run13
wes­
run2­
2
wes­
rur12­
3
TOTAL
DIOXINS
totdxna
totdxns
2.1840
0.1291
2.8496
0.2731
1.3521
0.3703
3.2676
0.6022
0.8941
0.0900
15.3254
3.9617
0.0558
2.0002
8.2172
0.1709
5.3342
0.1218
1.9771
0.9554
5.5100
,0.8090
0.9638
0.1478
1.0459
0.7357
4.7550
1.8872
15.7800
0.7797
27.5550
0.2924
ITEQ
DIOXINS
iteqa
itqs
0.2524
0.0572
0.1086
0.0016
0.1256
0.0227
0.0731
0.0261
0.1486
0.0361
0.0426
0.01
30
0.5208
0.0833
0.9128
0.0537
0.41
90
0.0727
0.3813
0.0002
0.2489
0.0202
0.0508
0.01
71
0.0582
0.0354
0.0357
0.0009
0.0459
0.0286
0.0408
0.0117
0.0423
0.0222
0.0209
0.001
5
0.0489
0.01
17
0.1776
0.001
7
0.3182
0.0019
7.7605
0.2028
7.5280
0.7809
3.5069
0.4250
4.3743
0.4541
3.5287
0.2606
WHO98
DIOXINS
who98a
who98s
0.161
1
0.0349
0.0709
0.0039
0.0876
0.0042
0.0482
0.0130
0.0896
0.0229
0.0332
0.0035
0.3268
0.0087
0.6133
0.0420
0.2793
0.0437
0.2430
0.0071
0.1620
0.0126
0.0508
004
0.0496
055
0.0286
0.0006
0.031
1
0.01
05
0.0314
0.0043
0.0310
0.0217
0.0223
0.0027
0.0203
0.0022
8.5165
0.1963
8.1251
0.8271
3.8095
0.4470
4.4339
1.1388
3.9597
0.201
3
Notes:
all
standard
deviations
are
sample
size
bias
corrected
­
multiplied
by
1.253
all
statistical
results
derived
from
pairs,
EERTRC
results
are
cross­
traverse
"
wes"
runs
are
new
since
ASMEIRCIMW
ReMAP
report
(
October
2000
testing)
0
0
0
'

8
3
Z
1
n
U
i
e
*
b
­
r
I
I
LI
U
¶
