MEMORANDUM

SUBJECT:	PM CEMS Random Error Contribution by Emission Limit

DATE:	March 22, 2023

FROM:	Barrett Parker
		Measurement Policy Group

TO:		Docket ID No. EPA-HQ-OAR-2018-0794

Measuring source emissions through approaches such as Agency-developed test methods or instruments such as PM CEMS, is a function of the approach's detection level, or minimum amount of material that is needed to determine whether or not a pollutant can be observed.  Measurements made at detection levels have a high degree of uncertainty, on the order of 100 percent; about all that can be said is that the pollutant exists or does not exist in the sample.  Fortunately, longer sampling time durations can provide lower detection levels, so, in many cases, one can lengthen the sampling time to obtain more material.  However, other factors should be considered when longer sampling times are viewed, including additional cost, general safety concerns associated with stack testing, and increased risks related to longer sampling durations.  In order to reduce measurement level uncertainty at the detection level, the Agency increases the average detection level of the best-performing  -  that is, the lowest emitting  -  sources by a factor of three; such a value is identified as the 3xRDL value.  Measurements made at or above the 3xRDL have an uncertainty of less than twenty percent, which is consistent with the uncertainties found by the American Society of Mechanical Engineers ReMAP study for selected Agency test methods.  The Agency's practice is to establish air emission limits at or above the 3xRDL value, as determined by collected data. 

Uncertainties of emissions as measured by instruments like PM CEMS also exist.  Those uncertainties are due to three types of error:  random, systematic, and measurement.  Ways to minimize these uncertainties include use of longer sampling times, of standard practices, and of more precise instrumentation.  Instrumental monitors typically include a tolerance that accounts for the sum of these errors; for PM CEMS, the tolerance is twenty-five percent of the emission limit.  

PM CEMS require a site-specific correlation between PM CEMS values and reference testing results over three fractions of the emission limit (low, medium, and high levels) occurring over 15 individual test runs.  Prudent EGU owners or operators generally choose to conduct 20 runs, as up to 5 run values can be discarded, as well as choose to ensure that their PM CEMS measure accurately at one-half the emission limit.  

Relying on the method detection level of 2 mg/dscm for a one-hour test run (which is equivalent to a mass per heat input value of 0.0082 lb/mmBTU); on three hour run durations, which have been selected when considering cost and personnel safety concerns; and on an operational target limit of 0.003 lb/mmBTU (one-half of the emission limit of 0.006 lb/mmBTU), one can calculate the range of the contribution of random error to the total PM CEMS tolerance to be between thirty-six and seventy-three percent; the average contribution of the random error is fifty-five percent.  While it is possible that the random error contributions of some PM CEMS will be at the lower end of the range, the average random error contribution for a 3-hour run sampling time, as shown in Table 1, exceeds those of other emission limits  -  including those in the current rule for existing and new EGUs as well as the other proposed emission limit of 0.010 lb/mmBTU; those other limits have an average random error contribution of less than forty-three percent.  


                                       
                                       
                           Emission limit descriptor
                                       
                                       
                             Run sampling time, hr
                                       
                                       
                           Emission limit, lb/mmBTU
                                       
                                       
                            Emission limit, mg/dscm
                                       
                                       
                       Target compliance level, mg/dscm
                                       
                                       
                         Sampling time 3xRDL, mg/dscm
                                       
               Average measurement uncertainty of 3xRDL, mg/dscm
        Target compliance level percent uncertainty due to random error
         Average random error contribution of total tolerance, percent
                                       
                                       
                              Stack Testing Days
Existing
                                                                            0.8
                                                                          0.030
                                                                           21.9
                                                                           11.0
                                                                           7.69
                                                                            1.2
                                                                           10.5
                                                                           42.1
                                                                              3
New
                                                                            3.1
                                                                          0.009
                                                                            6.6
                                                                            3.3
                                                                           1.91
                                                                            0.3
                                                                            8.7
                                                                           34.8
                                                                             10
Proposal 1
                                                                              3
                                                                          0.010
                                                                            7.3
                                                                            3.7
                                                                           2.00
                                                                            0.3
                                                                            8.2
                                                                           32.9
                                                                             10
Proposal 2
                                                                              3
                                                                          0.006
                                                                            4.4
                                                                            2.2
                                                                           2.00
                                                                            0.3
                                                                           13.6
                                                                           54.5
                                                                             10
Example 3
                                                                            3.8
                                                                          0.006
                                                                            4.4
                                                                            2.2
                                                                           1.57
                                                                           0.24
                                                                           10.7
                                                                           42.8
                                                                             10
Example 4
                                                                              6
                                                                          0.006
                                                                            4.4
                                                                            2.2
                                                                           1.00
                                                                            0.2
                                                                            6.8
                                                                           27.3
                                                                             20
Example 5
                                                                              8
                                                                          0.003
                                                                            2.2
                                                                            1.1
                                                                           0.75
                                                                            0.1
                                                                           10.2
                                                                           40.9
                                                                             20

Table 1.  PM CEMS Random Error Contribution by Emission Limit


Example 3 in the table shows that in order to obtain an average random error contribution of less than forty-three percent while maintaining a proposed limit of 0.006 lb/mmBTU, the run sampling time would need to increase to 3 hours, 49 minutes (3.8 hours) per run, increasing total sampling time to 76.2 hours, but keeping the total stack sampling period to 10 days.  Such an increased testing run duration should have a minimal increase in cost.  As shown in Example 4 in the table, the average random error contribution for this proposed limit could be decreased to twenty-seven percent by relying on 6-hour testing run durations; however, the total sampling time would increase to 120 hours, the total stack sampling period would double to 20 days, and the correlation testing costs would increase by fifty percent over those relying on 3 hour run durations.

While no emission limit lower than 0.006 lb/mmBTU coupled with a 3-hour run duration can provide an average random error contribution of less than forty-three percent, lower emission limits coupled with longer duration runs could provide average random error contributions less than forty-three percent; as shown in Example 5, an emission limit of 0.0030 lb/mmBTU with an 8-hour run duration would provide an average random error contribution of forty-one percent but total stack sampling period would remain at 20 days, the total sampling time would increase to 160 hours, and the correlation testing costs would double over those relying on 3-hour run durations.

