MEMORANDUM
TO:
Docket
No.
OAR­
2002­
0058
FROM:
Jim
Eddinger,
U.
S.
Environmental
Protection
Agency,
OAQPS
(
C439­
01)

DATE:
February
2004
SUBJECT:
Statistical
Analysis
for
Determining
the
Appropriate
Percentile
Confidence
Level
for
Using
Fuel
Analysis
for
Compliance
BACKGROUND
The
fuel
analysis
requirements
in
the
proposed
rule
raised
significant
comments
during
the
public
comment
period.
The
main
concern
was
the
requirements
to
conduct
a
new
fuel
analysis
each
time
a
facility
would
receive
fuel
from
a
new
supplier.
Commenters
indicated
that
in
the
case
of
biomass
fuel
a
facility
may
receive
fuel
from
over
a
hundred
ever
changing
suppliers
and
that
this
would
result
in
a
significant
burden
on
many
small
wood/
biomass
facilities.

In
an
effort
to
address
these
concerns,
the
final
rule
has
been
revised
to
require
only
an
initial
sampling
and
testing
of
the
fuel
and
a
statistical
analysis
of
the
test
results
as
means
of
ensuring
ongoing
compliance.
There
is
no
longer
a
requirement
to
re­
test
fuel
from
new
suppliers
or
fuel
from
new
sources
or
locations
from
the
same
supplier.
There
is
also
no
general
periodic
testing
requirement
(
other
than
every
5
years).
We
have
attempted
to
account
for
fuel
variability
by
introducing
statistical
analysis
to
the
initial
compliance
demonstration.
This
memorandum
described
the
evaluation
of
coal
data
to
demonstrate
why
the
statistical
analysis
will
ensure
ongoing
compliance.
The
analysis
underlying
the
new
approach
are
presented
here.

Given
the
significant
variability
in
the
HAP
content
of
coal,
there
was
concerns
about
requiring
only
an
initial
sampling
and
testing
of
the
fuel
as
a
means
of
ensuring
ongoing
compliance.
Conceptually,
it
seems
like
we
could
make
an
argument
that
accounting
for
variability
would
eliminate
the
need
for
frequent
sampling;
but
only
if
we
could
show
that
we
have
adopted
a
statistical
analysis
that
accounts
for
the
variability
that
can
possibly
occur
in
the
applicable
fuel
type.
The
variability
factor
in
the
approach
outlined
derives
from
only
the
variability
observed
in
the
fuel
samples
taken
at
the
individual
facility.

RESPONSE
TO
COMMENTS
The
final
rule
specifies
the
testing
methodology
and
procedures
and
the
initial
and
continuous
compliance
requirements
to
be
used
when
complying
with
the
fuel
analysis
options.
Fuel
analysis
tests
for
total
chloride,
gross
calorific
value,
mercury,
metal
analysis,
sample
collection,
and
sample
preparation
are
included
in
the
final
rule.

If
you
elect
to
comply
with
the
emission
limits
based
on
fuel
analysis,
the
facilty
is
required
2
to
statistically
analyze,
using
the
z­
test,
the
fuel
analysis
data
to
determine
the
90th
percentile
confidence
level.
It
is
the
90th
percentile
confidence
level
that
is
required
to
be
used
to
determine
compliance
with
the
applicable
emission
limit.
The
statistical
approach
is
required
to
assist
in
ensuring
continuous
compliance
by
statistically
accounting
for
the
inherent
variability
in
the
fuel
type.

STATISTICAL
ANALYSIS
To
address
the
concern
on
whether
a
statistical
fuel
analysis
approach
would
reasonably
ensures
that
the
emission
limits
would
be
continuously
achieved
based
on
only
an
initial
fuel
sampling
and
analysis
program,
we
conducted
a
statistical
analysis
on
coal
analysis
data
obtained
for
the
proposed
Utility
MACT
rulemaking.
That
is,
we
determined
for
various
coal
data
the
mercury
emissions
level
achieved
for
a
source
based
on
the
average
and
the
90th
and
95th
percentile
using
the
one­
sided
z­
statistics
test
(
i.
e.,
the
emission
level
which
the
emission
point
is
estimated
to
be
able
to
achieve
90
or
95
percent
of
the
time).

PROCEDURE
In
the
determination
of
the
appropriate
percentile
confidence
level,
we
used
the
results
of
the
mercury
fuel
content
results
for
6
utility
facilities.
Monthly
fuel
analysis
were
conducted
over
a
12
month
period
on
each
of
these
units.
The
mercury
content
(
ppmw)
for
each
(
3­
sample)
fuel
analysis
are
presented
in
Table
1,
together
with
the
unit
average
(
mean)
emissions
of
mercury
and
the
standard
deviation
of
the
individual
test
runs.

The
statistical
approach
used
was
the
one­
sided
z­
statistics
test
using
the
equation:

confidence
limit
=
average
+
z
*
standard
deviation,

where
the
value
of
z
is
a
function
of
the
degrees
of
freedom
and
obtained
from
the
statistical
table
listing
t
distribution
critical
values.
The
number
of
degrees
of
freedom
for
each
set
is
11.
The
z
values
used
in
determining
the
90
and
95
percentile
confidence
limit
are
1.363
and
1.796,
respectively.

DATA
The
data
consisted
of
fuel
analysis
results
for
mercury
content
(
ppmw)
from
6
selected
utility
facilities.
The
data
were
obtained
from
the
ICR
database
developed
for
the
Utility
MACT
rulemaking.
In
general,
three
fuel
samples
were
analyzed
for
each
monthly
result.
The
averages
and
the
observed
standard
deviations
for
these
coal
analyses
are
listed
in
the
table.
3
Fuel
Analysis
Results
for
Mercury
Content
Mercury
Content
(
ppmw)
Average
Standard
Deviation
90%
Confidence
Limit
95%
Confidence
Limit
Mercury
Content
(
ppmw)
Average
Standard
Deviation
90%
Confidence
Limit
95%
Confidence
Limit
0.0867
0.0728
0.0450
0.0763
0.0900
0.0763
0.0767
0.0743
0.0867
0.0788
0.011658
0.0947
0.0997
0.0568
0.0733
0.0510
0.0890
0.024384
0.1222
0.0867
0.1040
0.0750
0.1160
0.0825
0.1150
0.0867
0.1150
0.0733
0.0910
0.0833
0.1198
0.0450
0.0651
0.0770
0.0723
0.0733
0.0499
0.0375
0.0713
0.0410
0.0608
0.01777
0.085
0.0926
0.0746
0.0477
0.0899
0.0784
0.012803
0.0958
0.1013
0.0817
0.0827
0.0653
0.0754
0.0323
0.0860
0.0765
0.0932
0.0795
0.0910
0.0730
0.0892
0.0783
0.0450
0.0614
0.0800
0.0787
0.0567
0.0693
0.0567
0.0707
0.0693
0.013587
0.0878
0.0937
0.0448
0.0457
0.0390
0.0631
0.02194
0.093
1.025
0.0803
0.0636
0.0585
0.0702
0.0720
0.0467
0.0898
0.0567
0.0823
0.0787
0.0450
0.1186
4
RESULTS
The
analysis
was
based
on
demonstrating
continous
compliance
with
a
benchmark
mercury
level
of
0.09
ppmw.
For
all
6
coals,
the
average
mercury
content
(
of
the
12
monthly
samples)
is
below
the
benchmark
level.
Therefore,
on
average,
all
would
demonstrate
compliance.
However,
3
coal
facilities
had
at
least
1
monthly
result
greater
than
the
benchmark
level
which
would
indicate
failure
to
comply
on
a
continuous
basis.
One
coal
set,
even
though
it
averaged
below
the
benchmark,
had
higher
monthly
results
than
the
benchmark
in
6
of
the
12
months.
Therefore,
it
would
apparently
not
be
reasonable
or
appropriate
to
based
continous
compliance
on
the
average
of
an
initial
sampling
program.

Investigating
the
appropriate
confidence
level
that
would
assist
in
ensuring
continuous
compliance
based
on
fuel
sampling,
we
first
considered
the
95th
percentile
confidence
limit.
However,
in
all
6
cases,
the
95th
percentile
confidence
limit
is
greater
that
the
benchmark
level,
even
though
in
2
cases
all
12
samples
were
below
the
benchmark.
At
the
90th
percentile
confidence
limit,
these
2
coal
sets
were
below
the
benchmark.

CONCLUSION
Based
on
this
statistical
analysis,
the
appropriate
percentile
confidence
level
for
ensuring
continuous
compliance
is
90.
Coal
sets
with
all
monthly
values
below
the
benchmark
had
their
90th
percentile
confidence
limit
also
below
the
benchmark.
For
coals
sets
with
any
monthly
values
at
or
above
the
benchmark,
the
90th
percentile
confidence
level
was
above
the
benchmark.
