MEMORANDUM 

SUBJECT:	Assessment of  Using Single Point Stack Test Data to Derive
30-Day Rolling Average Emissions Limits 

FROM:	Stephen Boone, Roy Neulicht, and Jeff Cole,  RTI

TO:	Bill Maxwell, EPA, SPPD, Energy Strategies Group

DATE:	August 18, 2011



Summary of Technical Directive

The WAM requested that RTI further assess using single-point data as the
basis for establishing MACT limits with 30-day rolling average
compliance periods. Specifically ESG asked, “… have we adequately
and accurately developed an emission limit from manual,
single-point-in-time testing that will truly reflect what would occur
through the use of CEMS and 30-day-day rolling averages (e.g., should
“m” in the statistical variability analysis be made equal to the 
“number of future runs” i.e., 1 … or 30 …)?”

Problem Statement for Evaluation of Hg Standard for >8,300 Btu/lb coal
units

Based on available CEMS data,  what is the relationship between a
unit’s 30 day average emission rate (including startups, shutdowns,
and malfunctions) and the emission rates occurring during short-term
periods at operating conditions equivalent to typical stack testing
conditions (i.e., steady stack flow rate and steady load at greater than
or equal to 90 percent of full load conditions). 

Background

Eight of the sources included in the initial MACT floor pool of 328
sources (in the >8,300 Btu/lb coal-fired subcategory), submitted Hg CEMS
data in lieu of manual stack test data.  However, only seven of these
CEMS data sets reflected the minimum mercury average of the source when
compared to available Part II manual stack test data and thus were used
in the initial floor ranking. The seven CEMS sources submitted 30 days
of daily Hg emissions averages, however none of the seven CEMS sources
fell into the top-12 percent of performers in the initial ranking of
sources. With the exception of these seven CEMS data sets, all other
data used for the initial floor ranking were comprised of Part II or
Part III manual stack test data submitted for the ICR.

Prior to proposal, the 30-day CEMs data from the seven units were
analyzed to determine the variability.  Using the UPL methodology, the
daily values for all the CEMS data from the seven units were averaged to
find the mean value and the variability of the daily values from the
mean was determined.  The variability of the CEMS data was about 2 times
the mean value.  The CEMS variability was then compared to the manual
stack test data variability, and it was found to be about one quarter
the variability of the manual stack tests. Since the CEMS data
variability was less than the variability of the data set used to
establish the MACT floor, EPA determined that the MACT floor emission
limit (corrected for variability) was achievable on a 30-day rolling
average basis, as EPA proposed in the rule. 

Some facilities also provided historical CEMS data with their Part II
ICR submittals.  These data were received from approximately 100
sources.  Review of these data, indicated that information sufficient to
conduct an analysis (i.e., unit load, lb/MMBtu emission rate, and data
quality flags present to exclude calibrations and other suspect data)
was provided from about 60 of the 100 sources.  This analysis is
described in the following paragraphs.  Note that, to date, only
historical CEMS data from 39 sources have been analyzed.

Technical Approach

In this analysis the historical CEMS data were used to derive a
mathematical relationship between the 30-day average emission limit that
is achievable (i.e., the maximum 30-day average emission rate in the
CEMS data set) and the minimum six-hour average emission rate that is
achievable at stack test conditions (i.e., the minimum emission rate
recorded by CEMS during periods of steady operations at greater than or
equal to 90 percent of capacity).

The Hg CEMS historical data emissions data covers Hg CEMS operated
between 1/1/2005 through 12/31/09 (with some data sets continuing a
month or two into 2010). Only one of the sources included in the Hg MACT
floor pool of 40 units (determined exclusively based on each unit’s
minimum manual stack test result), also provided Hg CEMS data under Part
II of the ICR.  This source (Public Service Company of New Mexico’s
San Juan Station, Unit 4, ORIS 2451) had the 33rd lowest Hg emissions
rate in the MACT floor pool.  Additionally, a second unit (not included
within the top 40 because the source did not provide manual stack test
data) provided CEMS data demonstrating 30-day averages at or below the
median MACT floor rate without variability at a higher frequency than
San Juan Unit 4.  Performance versus the MACT floor (without
variability) was based upon analysis of “surrogate emissions tests”
as explained in the following paragraphs. 

First, the Part II CEMS data were reviewed for completeness.  The hourly
CEMS data selected for analysis were then reduced to 30-day rolling
averages. Further analysis was done to identify 6-hour periods of CEMS
data at operating conditions typical of manual stack tests and to
determine the average emission rate during such periods (i.e., periods
of steady load and stack flow for at least six hours while the source
operated at 90 percent capacity or greater).  These periods are
referred to in this memorandum as of “surrogate emissions tests.” As
previously noted, San Juan Unit 4 was the only source that provided CEMS
data in the top 40 best performing units based on the manual stack
tests. Figure 1 compares the manual stack test data used in the floor
analysis for San Juan Unit 4 versus the 30-day rolling averages measured
by CEMS over approximately 8,000 hours (including the hours of the Part
III San Juan Unit 4 manual stack test used in the floor calculation). 
Figure 1a adds to the figure the calculated 6-hr CEMS average
“surrogate stack test” data.  The CEMS data averages included
periods of startup, shutdown, and malfunction and were calculated based
on the lbs Hg/MMBtu compliance option using a diluent cap in accordance
with §63.10005(l) of the proposed rule (if necessary) during startup
and shutdown periods.  In other words, the data were screened to
identify start-up and shut-down periods (based on an estimated startup
and shutdown threshold equal to 28 percent of full load), and for these
periods, if necessary based on the actual diluent concentration, the
emissions rates were recalculated using the appropriate cap value in
accordance with the proposed provisions of §63.10005(l).  Note that
based on these 30-day average CEMS data, San Juan Unit 4 complied with
the proposed MACT limit (1.2 x10-6 lb/MMBtu) during all 30 day averages
occurring in the submitted CEMS data.  

The overall objective of this analysis was to determine the averaging
period ratio (APR) or the ratio of the maximum average emission
limitation achieved on a 30-day rolling average basis divided by the
minimum surrogate stack test as measured by Hg CEMS. For the purposes of
this analysis, the APR was defined as follows:

Averaging Period Ratio (APR) =	(maximum 30-day rolling average recorded
by Hg CEMS)

	(minimum short-term Hg emission rate at stack test conditions*)

*Surrogate stack test conditions are steady flow operations at a
generating rate that is > 90% of capacity.

This technical approach provided for a very large data set for each unit
with hundreds of surrogate stack tests and hundreds of 30-day averages
during all operating conditions. Each APR is unit specific and provides
an estimate of differences between the short-term and the long-term
emission averages that are related to factors including 

A particular source’s fuel contracts (i.e., variability of fuel-Hg
content),

Hours since the last major maintenance event (for the larger data sets),
and

Seasonal factors (for the larger data sets) 

Routine cycling to match system demand

Other analyses such as the UPL method, applied across multiple best
performing units, may be necessary to account for variability associated
with unit age, design, measurement technique, and cycling profiles not
represented in the CEMS database.

Subsequently, Part II CEMS data files for 39 units (a total of
approximately 180,000 hours of Hg CEMS data) submitted in response to
the ICR were reviewed to evaluate periods of operation consistent with
the normal procedures for manual stack tests (i.e., surrogate stack
tests). Table 1 summarizes these CEMS data. The companies operating
units identified in bold text within Table 1 all provided 900 or more
hours of Hg CEMS data. These data were utilized to derive an equation
for predicting the APR for a unit based strictly on short-term stack
test data. A threshold of 900 hours of Hg CEMS data was chosen because a
data set of this size or greater provides at least 180 thirty-day
averages for comparison against the minimum short-term emission rate.
Thirty-six units had 900 or more hours of Hg CEMS data.

The APR, as defined above, was calculated for 35 of the units (all the
units other than San Juan Unit 4) identified in the bold text, and the
ratios were plotted (see Figure 2) versus the minimum “surrogate stack
test” in the same data set to obtain a relationship between the
lowest, achievable, short-term (i.e., 6-hour), emission rate at stack
test conditions and the maximum 30-day rolling average that was
achieved.  The relationship was evaluated by calculating the predicted
value for San Juan Unit 4 and comparing the predicted value with the
actual value recorded by CEMS.

Summary of Findings

The relationship between the maximum 30-day average emission rate was
found to have a good correlation to the value of the minimum
“surrogate stack test” as defined by Equation 1:

(Equation 1)	y = 0.0018x-0.569

r² = 0.75 

The amount of error in this prediction was determined by applying
Equation 1 to the San Juan Unit 4 actual minimum surrogate stack test (1
x 10-8 lbs Hg/MMBtu on a 6-hr average) to predict the APR and
subsequently the to predict the maximum 30-day average for comparison
against the actual maximum 30-day average in the CEMS data:

APR predicted for San Juan U4 = 0.0018(1 x 10-8)-0.569 = 64.2

Therefore, the 30- day average emission rate predicted for San Juan Unit
4 is estimated as follows: 

Predicted maximum 30-day average for San Juan Unit 4 = APR x 1 x 10-8 =
6.42 x 10-7 lbs Hg/MMBtu

The actual maximum 30-day average for San Juan Unit 4 was 4.94 x 10-7
lbs Hg/MMBtu.

The percent error of the predicted value versus the actual value for San
Juan U4 was 30 percent.

Therefore, for minimum emission rates about 1 x 10-8 lbs Hg/MMBtu,
Equation 1 is expected to predict the APR for a large CEMS data set with
an error margin of approximately 30 percent. 

As mentioned earlier, the emission rate at the median position in the
data set of minimum manual stack test averages for the forty sources
that composed the MACT floor data set was 1.97 x 10-8 lbs Hg/MMBtu. 
Applying Equation 1 to this emission rate yields the following estimate
of the 30-day average that could be achieved by a unit with a minimum
six-hour stack test averaging 1.97 x 10-8 lbs Hg/MMBtu:

APR predicted at median of MACT floor (without variability) =
0.0018(1.97 x 10-8)-0.569  =  43

Predicted maximum 30-day average at median of MACT floor (without
variability) = APR x 1.97 x 10-8 

Predicted max. 30-day avg at median of MACT floor (without variability)
= 8.52 x 10-7 lbs Hg/MMBtu + 30%

Additional work to be conducted

Additional work planned includes:

1. Additional QA of the data set and calculations, and 

2. Variability analysis of the equation

3. Calculate UPL using predicted maximum 30-day averages

Figure 1 

Figure 1a 

Figure 2

Table 1 – Summary of Hg CEMS Data

  1.97 x 10-8 lbs Hg/MMBtu, which is the average of Chambers
Cogeneration LP, Unit 1 (ORIS 10566) and Reid Gardner, Unit 1 (ORIS
2324)

DRAFT – DELIBERATIVE – NOT FOR DISTRIBUTION

PRELIMINARY RESULTS, QA CHECKS NOT COMPLETED

DRAFT – DELIBERATIVE – NOT FOR DISTRIBUTION

PRELIMINARY RESULTS, QA CHECKS NOT COMPLETED

Table 1 – Summary of Hg CEMS Data

