MEMORANDUM 

SUBJECT:	The Impact of Emissions Averaging Time on the Stringency of an
Emission Standard



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

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

DATE:	December 9, 2011



Purpose

These analyses were conducted on the Part II CEMS data to evaluate the
impact of averaging time on variability and to “predict” the UPL
value for different averaging times for the MACT floor facilities. This
predictive tool has not been previously used and is intended to
“ground truth” the variability estimated by the UPL calculation
methodology.  These analyses are not the result of additional
statistical analyses of the stack testing data collected under Part II
or any data received under Part III, directly.

Note: These results have not been through an extensive QA.

 General Approach

Part II data from 87 units were evaluated for data completeness and
hours flagged as invalid were removed prior to conducting any
calculations.

For each unit, each hour of data was evaluated to determine the
operating load (gross MW) and to determine if zero emissions were
reported for the hour. Hours of zero reported emissions were excluded
from the calculation of daily averages. 

Hours were classified as startup or shutdown periods if the gross
megawatts recorded for the hour were less than 5 percent of the maximum
hourly generating rate recorded in the data set.  These hours were not
included in the calculation of daily averages. Daily averages were
calculated in accordance with Equation 19-19 from Method 19.

Contiguous six-hour periods of valid CEMS data at operating conditions
equivalent to typical stack testing conditions (i.e., steady-state flow
rate and steady load at greater than 90 percent of the maximum recorded
operating load) were used to calculate all the short-term emission rate
averages occurring in the data set that were analogous to short-term
stack test averages; these periods are referred to as “surrogate stack
tests” in this memo.

All averaging periods calculated (30-day, 60-day, 90-day, 180-day, and
360-day) excluded startup and shutdown periods. 

For each averaging period analyzed, the variability ratio of [the 99th
percentile long-term average] to [the 3rd percentile surrogate stack
test value] was calculated for each operating unit with a sufficient
number of hourly records to calculate at least one long-term period.
These percentiles were used to eliminate outliers in the data sets.

The variability ratios for all units within a given averaging period
data set were used to develop a general equation defining the
relationship between the 99th percentile long-term average and the
near-minimum achieved surrogate stack test. The general form of the
equation is presented as Equation 1:

[Equation 1]		y = Cx-z

Where 	y  =  the estimated long term emissions average 

	x  =  the near-minimum short-term emissions average

	z  =  was a calculated exponent derived from all available CEMS hourly
data.

	C =  was a calculated coefficient derived from all available CEMS
hourly data.

For each averaging period analyzed, the general equation, developed in
step 6, was used to transform short-term emissions data for each unit
floor into an estimate of the 99th percentile long-term performance for
the floor unit. The transformed estimates were calculated by inserting
the minimum actual stack test data from each of the 47 units into the
general equation from step 6. Appendix A of this memorandum only
includes the two plots utilized to determine the equation for
transforming MACT floor stack test averages to 30-day averages. The
attached spreadsheets contain the plots of all other averaging periods.

The upper predictive limit for each long-term average Hg emissions rate
was calculated for the floor units using the equation from step 6. Only
the minimum test value (i.e., row 5 in the UPL spreadsheet) was used;
all other stack test rows previously used to determine unit-specific
variability were excluded.

Conclusion

Table 1 presents the results of the analysis of what the Hg MACT floor
for existing coal units would look like if EPA decides to promulgate a
compliance period longer than 30 days. As shown in Table 1, the amount
of data available for each averaging period analysis varied. For
example, only 23 units had sufficient data to calculate at least one
360-day average. Since the objective of this analysis was to compare
achievable emissions rates based on the duration of the averaging
period, the analysis used only the 23 units with sufficient data to
calculate all averaging periods as a “control” to assess the impact
of comparing the different data sets.

Table 1 – UPL’s Calculated with All Available Data

Average Period	Available CEMS	Calculated UPL

(days)	Data Sets	With All Available CEMS Data

(lb Hg/MMBtu)

30	87	1.11E-06

60	83	1.01E-06

90	77	9.13E-07

180	66	8.04E-07

360	23	7.60E-07



To quantify the effect of the reduction in the count of available
datasets with sufficient data as the averaging periods increase, the
analysis also includes a series of UPL calculations using only the
“control” data set. Table 2 presents a comparison of the calculated
UPLs for each averaging period without the analytical artifacts caused
by the limited availability of very long-term (i.e., semi-annual and
annual) CEMS data.

Table 2 – UPL’s Calculated with Control Data (23 Units Only)

Average Period	Available CEMS	Calculated UPL

(days)	Data Sets	With Control CEMS Data

(lb Hg/MMBtu)

30	23	1.32E-06

60	23	1.13E-06

90	23	1.03E-06

180	23	9.17E-07

360	23	7.60E-07



The data in Table 2 confirms that an appropriate ratio between a 30-day
compliance period and a 360-day compliance period is approximately 60
percent. This is consistent with the ratio achieved by San Juan Unit 4
based on the maximum averages recorded by CEMS with no statistical
variability added.

Figure   SEQ Figure \* ARABIC  1 .-San Juan Unit 4 Actual CEMS Data
Averages



Appendix A

Typical Data Sets Used to Derive Average Data Sets

 Note that the data set of 87 units utilized to derive the UPL of
1.11E-06 is presented in Appendix A, Figure A-1.

 Note that the data set of 23 units utilized to derive the UPL of
1.32E-06 is presented in Appendix A, Figure A-2.

DRAFT – DELIBERATIVE – NOT FOR DISTRIBUTION

PRELIMINARY RESULTS, QA CHECKS NOT COMPLETED

  PAGE   \* MERGEFORMAT  4 

