     Supplemental Assessment of the Air Quality Consequences of Applying 
                  Adjusted Long Term Average Emission Limits
	
	EPA's 2014 guidance on sulfur dioxide (SO2) nonattainment planning, entitled "Guidance for 1-Hour SO2 Nonattainment Area SIP Submissions," dated April 2014, provided recommendations regarding the establishment of SO2 emission limits based on averaging times of up to 30 days, notably recommending the downward adjustment of the level of the limit such that the limit has comparable stringency to the 1-hour average emission limit that would otherwise be set.  EPA explained that limits satisfying these recommendations in some cases could be found to assure attainment of the 1-hour SO2 air quality standard.  
	Appendix B to this guidance described an assessment that supported this proposition.   Subsequently, in response to proposed rulemaking involving application of 30-day average emission limits, a commenter expressed concern that the assessment described in Appendix B does not appropriately compare air quality with a 1-hour emission limit against air quality with a longer term average emission limit.  The assessment described in Appendix B analyzed the air quality expected to result from an emissions distribution in which the maximum 30-day average emission level matched the allowable 30-day average emissions but the remainder of the inventory was distributed across a range of lower level, in proportion to the distribution of actual emissions in the original inventory.  Thus, the assessment in Appendix B may be considered to reflect analyses of the air quality impact of actual emission inventories that might be expected to result from compliance with a 30-day average emission limit.  The commenter's concern was that the results of these analyses, in which maximum emissions match the limit but most emissions (even on a 30-day average basis) are substantially lower, cannot validly be compared to the results of modeling of emissions constantly at a 1-hour emission limit.
	To address this concern, EPA conducted additional assessment of the impact of emissions that are allowed under a 30-day average emission limit.  A key component of this assessment is the formulation of emission profiles that may be considered to represent allowable emissions under a 30-day average limit.  By its very nature, a 30-day average limit allows variations in emissions, within the constraints of the 30-day average limit.  The intent of this assessment was not to consider all possible combinations of emissions that would comply with such a limit, but instead to apply realistic emissions distributions at levels that challenge sources' ability to comply with 30-day average limits.  The particular approach followed in this assessment was to analyze the selected emissions databases in accordance with Appendix C of the guidance, to identify the 99[th] percentile 30-day average emission level, and then to identify the emissions distribution of the 30-day period that most closely matches that 99[th] percentile value.  These sets of 30 days of varying emissions, scaled to match the 30-day average limit, were used to represent allowable emissions for the 30-day average limit.  The remainder of the assessment assessed the air quality impact of these emission distributions using the methods described in Appendix B.
	The first section that follows provides a synopsis of the assessment described in Appendix B, many features of which are also features of this assessment.  The second section that follows describes the derivation of the emission distributions used to represent allowable emissions in this assessment.  The third section describes the results of the air quality analysis based on these emission distributions.  The final section offers conclusions from this assessment.
Synopsis of Assessment in Appendix B
	In support of the 2014 SO2 nonattainment area planning guidance, which offered the option in some cases to establish emission limits averaged on up to a 30-day average basis, EPA conducted an assessment, described in Appendix B of that guidance, of the air quality expected from compliance with a 30-day average limit in two sample scenarios.  This assessment relied on air quality modeling for a previously existing power plant in South Carolina known as Canadys.  This modeling indicated that the critical emission value, i.e. the constant emission rate which would provide for attainment of the 2010 SO2 NAAQS, is 1,831 pounds of SO2 per hour.  EPA then formulated a variety of hourly varying emission data sets, which EPA used in conjunction with the hourly modeled impact per pound of emissions to assess the overall air quality expected to result from each of these emission data sets.
	EPA review of emissions variations at power plants indicates that power plants that operate flue gas desulfurization equipment tend to have different emission profiles than plants that do not.  Therefore, EPA compiled two primary emission profiles, one representing plants with flue gas desulfurization and one without such control (labeled a low sulfur fuel plant).  Since no suitable profiles were available for Canadys, these profiles reflected emissions data for Unit 4 and Unit 3, respectively, at the Weston Plant near Wausau, Wisconsin.
	These emission data sets, analyzed in accordance with Appendix C of the guidance, indicated adjustment factors for the scrubber and the non-scrubber distributions of 68% and 69%, respectively.  These adjustment factors, multiplied times the 1,831 pounds per hour critical emission value, yielded recommended 30-day average limits of 1,253 pounds per hour and 1,254 pounds per hour, respectively.  A scaling factor was calculated by dividing the pertinent 30-day average emission limit by the maximum 30-day average emissions in the respective Weston emissions data set.  This scaling factor was multiplied times each hour's emissions in the respective data set.  The resulting emissions distributions thus have the same relative distribution of emissions as is found in the actual Weston emissions data set, with the maximum 30-day average emissions equaling the 30-day average limit.  That is, the inventory shows compliance with the limit at all times and overcompliance most of the time.  The annual average emissions in these scaled data sets were 66 percent and 77 percent of the allowable emission levels for the scrubber and non-scrubber data sets, respectively.
	The assessment described in Appendix B also included additional analyses evaluating air quality on the basis of randomly redistributed emissions and evaluating air quality applying the different emission profiles that occurred in different years.  Further details are provided in Appendix B.  No such analyses were conducted as part of this supplemental assessment.  The supplemental assessment for this rulemaking also used different emissions data sets, as described in the following section.  Otherwise, the supplemental assessment used the same analytical approach as the assessment described in Appendix B, most notably using the same modeling information (in particular the same hourly information on concentrations at the 10 highest concentration receptors per pound of emissions) as the assessment described in Appendix B.
Emission Profiles Used in This Assessment
	A key challenge in evaluating whether long term average emission limits assure attainment is to evaluate the emissions that such limits may be considered to allow.  By their very nature, such limits may be met with a wide range of emission distributions.  For example, some compliant distributions involve relatively steady emission rates with emissions deviating little from the allowable average emission rate, and other compliant distributions involve emission spikes accompanied by other times with emissions well below the allowable average emission rate.  EPA guidance advises that long term average limits are most effective at assuring attainment if the frequency and magnitude of emissions exceeding the critical emission value are minimized.  However, even with that constraint, a broad range of scenarios qualify as alternative definitions of allowable emissions.  The challenge for this assessment, then, is to identify emission profiles that reflect realistic representations of the range of routinely occurring emission distributions that comply with a long term average emission rate.
	The approach taken in this assessment is to use actual emission distributions for the periods that have long term average emissions as close as possible to the 99[th] percentile long term average emissions for the respective plant/unit.  This assessment especially seeks to evaluate 30-day average limits, with the assumption that similar findings would apply to limits with intermediate (e.g., 24-hour) averaging times.  Therefore, the first step in this assessment was to compute 30-day emission averages for each of the selected data bases according to the recommendations of Appendix C.  The second step was to compute the 99[th] percentile among these 30-day emission averages.  The third step was to identify the 30-day period with average emissions that most closely approximated the 99[th] percentile average.  The 720 hourly emissions values in this 30-day period served as the basis for the emission profile.
	Two further adjustments were made to these 720-hour emission data sets.  	The first adjustment was to scale these profiles to the associated allowable emissions level.  As noted above, the critical emissions value was 1,831 pounds per hour.  For each data set, a 30-day average allowable emissions level was determined by determining an adjustment factor in accordance with the methods of Appendix C using the full multi-year data set and multiplying that adjustment factor times 1,831 pounds per hour.  A scaling factor was computed as the ratio between this limit and the average of the actual emissions during the 30-day sample.  This scaling factor was multiplied times the emissions for each hour in the adjusted sample.  
	The second adjustment was to address non-operating times in the selected 30-day sample.  Although EPA guidance is to average emissions only during operating time, the intent here was to assess air quality under conditions of full-time operation.  Therefore, for any time during the 30-day period of non-operation, the blank value in the actual emissions data base was replaced with a value equal to the allowable emissions level.  The result in each case was a 30-day emissions profile with variability reflective of the actual variability in the 99[th] percentile among 30-day average periods (modified to reflect full-time operation) with 30-day average emissions exactly equal to the allowable emissions level.  
	For the air quality analyses, the emission profile was assumed to occur repeatedly, i.e. that after each series of the 720 hours of the emissions profile being applied, emissions would begin again at the Profile Hour 1 value.  Since every 30-day period included the same 720 emission values (albeit starting at 30 different points in the profile), this approach assured that every 30-day period in each analysis had average emissions equal to the allowable 30-day average level.
	An important additional consideration is whether the profiles derived by the above means could realistically recur repeatedly.  In some cases, the 99[th] percentile profile will include a short period when emissions control is malfunctioning and emissions are unusually high.  However, for several reasons, such periods occur only rarely.  From an engineering perspective, many units are designed with a system involving consistent operation of their scrubbers, and deviating from this system erratically will cause damage. Units also are often subject to other emissions limits that require operating their controls (such as limits on wastewater discharge or on hazardous air pollutants).  There is little economic incentive to fail to operate installed flue gas desulfurization equipment properly, especially after accounting for the disincentive against high emissions that arises from trading program requirements.  A review of power plant operating data reported to EPA (available at https://ampd.epa.gov/ampd/) confirms that power plants with flue gas desulfurization systems operate these systems efficiently at almost all times, especially during times with normal to high operating rates.  
	The intent of this assessment is to formulate a profile of representative emissions variations within a period where a facility was just meeting a 30-day average emission limit, then to assess the estimated air quality that would result from applying that profile on a recurring basis.  In particular, this assessment sought to identify a 30-day emissions pattern that is representative of the emission patterns that can be expected when facilities are emitting at levels allowed by a 30-day average limit, and then to assess the estimated air quality that would result from that 30-day emissions profile occurring a little over 12 times a year.  Occasions with exceptionally high SO2 emissions, generally reflecting malfunction of the SO2 emission control equipment, are rare events that do not recur as frequently as every 30 days.  Therefore, an additional step in the development of profiles in this assessment was to screen out any data sets with hours with exceptionally high SO - 2 emission levels, to avoid any profiles that would not occur repeatedly.
	Five emission profiles were developed for this assessment.  A pair of profiles was developed for Unit 1 of the Cardinal plant in Ohio, and three profiles were developed for Unit 3 of the Petersburg plant in Indiana.  Both of these plants have become subject to emission limits pursuant to nonattainment planning requirements, and in both cases the data used in this assessment were from a period when the facility was meeting the pertinent limits.  Specifically, the Cardinal profiles were derived from data for 2013 to 2017, and the Petersburg profiles were derived from data for 2017 to 2019.  For each plant, separate profiles were developed on a pound per hour (#/hour) basis versus on a pound per million British Thermal Unit (#/MMBTU) basis, since the 30-day period with the 99[th] percentile #/hour average was a different period with a different profile than the 30-day period with the 99[th] percentile #/MMBTU value.  For reasons described below, the profiles were developed from the Petersburg data for both the 99[th] percentile and the 98[th] percentile #/MMBTU value.
	Conceptually, for periods with lower heat input, a limit on #/MMBTU yields a lower allowable emission level than would apply with a #/hour limit.  However, for purposes of this assessment, all emissions were scaled up to the 30-day average #/hour value that corresponds to the critical emission value (maximum heat input times a critical #/MMBTU value) as multiplied times the appropriate (#/MMBTU-based) adjustment factor.  By this means, the profile serves to approximate the maximum emissions (with maximum heat input) that would be allowed with this limit.
	Table 1 shows features of the emission profiles.  For each profile, the first entry in this table is the adjustment factor calculated according to Appendix C for the associated multi-year data base.  The second entry is the resulting calculated 30-day average emission limit, i.e. the critical emission value times the adjustment factor.  The third entry is the average emissions during the 30-day period that best approximates the 99[th] (or 98[th]) percentile value.  The fourth entry is the scaling factor (reflecting the ratio of the allowable emissions level to the average emissions in the actual emissions profile) that was multiplied times each hour's value in the actual emissions profile.  The fifth entry is a standard deviation among the 720 hourly emission values in the profile, included as a measure of the variability of the emissions within the profile.
Table 1. Characteristics of Analyzed Emission Profiles and Associated Data Sets
                                    Profile
                             Adjustment Factor (%)
                      Allowable Average Emissions (#/hour
                    Average Emissions in Raw Data (#/hour)
                                Scaling Factor
                      Standard Deviation as % of Average
                                Cardinal #/hour
                                     73.6%
                                     1,348
                                     1,799
                                     0.75
                                      24%
                               Cardinal #/MMBTU
                                     77.5%
                                     1,418
                                     1,805
                                     0.79
                                      23%
                               Petersburg #/hour
                                     64.0%
                                     1,171
                                     1,018
                                     1.15
                                      36%
                              Petersburg #/MMBTU
                                     73.7%
                                     1,350
                                      896
                                     1.51
                                     114%
                              Petersburg 98[th] 
                              Percentile #/MMBTU
                                     73.7%
                                     1,350
                                      979
                                     1.38
                                      47%
 
      These profiles are also displayed graphically in Figures 1-5.
      
Figure 1. Emission Profile for Cardinal with Approximately the 99[th] Percentile Pound per Hour Value 
      
      
      
Figure 2. Emission Profile for Cardinal with Approximately the 99[th] Percentile Pound per Million BTU 
      
      
Figure 3. Emission Profile for Petersburg with Approximately the 99[th] Percentile Pound per Hour Value 
      
      
      
Figure 4. Emission Profile for Petersburg with Approximately the 99[th] Percentile Pound per Million BTU Value 
      
      
      
      
Figure 5. Emission Profile for Petersburg with Approximately the 98[th] Percentile Pound per Million BTU Value 
      
      
      
      
      This table and these figures highlight the degree of variability of emissions within these five emission profiles.  For the two Cardinal profiles (which occur in nearly identical 30-day periods) and for the Petersburg #/hour profile, the standard deviation among the data in the profile is between 23 and 36 percent of the average value, suggesting significant but not dramatic variability in hourly emissions.  In contrast, the Petersburg 99[th] percentile #/MMBTU profile shows much more dramatic variability, with a standard deviation among the data that is 14 percent larger than the average value, as a result of a modest number of hours with high emissions in which flue gas desulfurization was apparently at best minimally effective.  
      
      As discussed above, the Petersburg 99[th] percentile #/MMBTU scenario involves hours in which flue gas desulfurization was at best minimally effective.  This is not a scenario that would realistically occur multiple times a year.  Therefore, this scenario does not fit the design of this assessment, which is to assess the air quality impact of allowable emission profiles that could reasonably be expected to recur a little over 12 times a year.  Consequently, this assessment does not include analysis of the air quality impact of this scenario.  Instead, the unrealism of repeated occurrence of this profile prompted the development of a replacement profile that would more realistically recur.  This replacement profile is the fifth profile shown above, reflecting the emission profile for the 30-day period with an average #/MMBTU that most closely approximates the 98[th] percentile value at Petersburg Unit 3.
      
      One feature that distinguishes these profiles from the profiles in the Appendix B assessment concerns emission categories.  For Appendix B, emission categories were identified, and the analysis used the category level rather than a direct, unrounded emission quantity.  For this assessment, the analysis used the direct emission level (as scaled).  However, the categories in the Appendix B assessment were relatively narrow, and so the use of this approach likely had little effect on estimated air quality.
      
Air Quality Analysis

      The next step in this assessment was to assess the air quality impact of these emission profiles, using the same modeling information (i.e., the same hourly emissions-air quality relationship) as was used in the assessment described in Appendix B.  Thus, hourly emissions of the applicable profile were matched with hourly information on concentrations per pound of emissions to estimate hourly concentrations at each of 10 critical receptors.  For each receptor, these data were then tallied to compute 5-year average receptor-specific design values.  The highest of these design values, i.e. the overall design values for each of the four emission profiles, are shown in Table 2 below.
      
Table 2. Design values estimated for each emission profile
      
                                    Profile
                                 Design Value 
                               (ug/m[3] (ppb))
                                Cardinal #/hour
                                 181.2 (69.2)
                               Cardinal #/MMBTU
                                 190.6 (72.8)
                               Petersburg #/hour
                                 156.3 (59.7)
                        Petersburg 98[th] %-ile #/MMBTU
                                 190.5 (72.7)
      
Conclusions
      
      The assessment described in Appendix B to EPA's 2014 guidance on SO2 nonattainment area planning was intended to evaluate whether areas with SO2 emission limits averaged over 30 days, developed in accordance with that guidance (reflecting an adjustment determined in accordance with the procedure in Appendix C), can reasonably be expected to attain the SO2 air quality standard.  The emission data sets evaluated in the Appendix B assessment had maximum 30-day average emission rates matching the presumptive 30-day average emission limit but used lower emission rates for most of the year, consistent with the distribution of actual emissions at the plants from which the emission profiles were derived.  The air quality analyses based on these emission profiles found that resulting air quality can be expected to be well below the standard. 
      
      That assessment begged the question as to whether emission profiles that more closely matched allowable emissions under a long term average emission limit would also result in attainment.  Longstanding policy embodied in 40 CFR 51 Appendix W and Appendix A to EPA's 2014 guidance on SO2 nonattainment area planning specifies that emission limits are to be set in accordance with the results of modeling maximum allowable emissions.  Thus, the question here is whether emission profiles that more closely match the average emissions allowed by a long term average emission limit can be expected to result in attainment of the standard.  This is the question that the assessment described here seeks to address.  By evaluating the air quality impact of emission profiles that better reflect allowable emissions under a 30-day average emission limit, the results here are also more comparable to modeling of allowable emissions under a 1-hour emission limit, to address whether setting a long term limit will have comparable stringency to a corresponding 1-hour limit (and assuring that emission spikes are limited in frequency and magnitude) and can be expected to yield comparable air quality.
      
      Some caveats warrant noting.  While the profiles addressed here reflect a range of emission distributions, this assessment must be considered only a sampling of plausible emission distributions permissible under long term emission limits and their resulting air quality.  EPA's guidance recommends that states take steps to assure that occasions with elevated emissions (i.e., emissions above the critical emissions value) are limited in frequency and magnitude.  In addition, as noted above, engineering and other constraints help assure that the occurrence of exceptionally high emissions is rare or nonexistent.  Nevertheless, this assessment did not include any analysis of the impact of occasional exceptionally high emissions values (or the compensating reduction of emissions at other times to maintain compliance with a 30-day average emission limit).  
      
      A second set of caveats relates to policy on demonstrations that long term average limits provide for attainment.  EPA's 2014 guidance envisions that most states seeking to establish long term emission limits will conduct modeling aimed at determining suitable 1-hour emission limits and then will determine comparably stringency long term limits by using the procedures recommended in Appendix C of the guidance, without conducting any additional modeling aimed at assessing the air quality impact of the patterns of allowable variable emissions.  The guidance includes a general statement that "The EPA is not precluding states from using other approaches to determine appropriate longer term average limits."  However, the guidance does not include any recommendations as to how such other approaches might be designed.  Appendix B in particular provides no guidance on how to formulate emission profiles reflective of allowable emission levels; indeed, the emission distributions used in the Appendix B assessment were more reflective of an actual emissions distribution (constrained to have a maximum 30-day average emission level equal to the calculated limit) than of an allowable emission level.  Similarly, this document should not be construed as guidance on how to evaluate the air quality impact of the variable emissions allowed by longer term limits.
      
      Nevertheless, this assessment has analyzed the air quality impact of several emission profiles with variable emissions in compliance with 30-day average emission limits.  These profiles are scaled in order that every 30-day period has average emissions exactly equal to the 30-day limit (or, for the #/MMBTU limits, 30-day average emissions in #/hour that exactly equal the maximum emissions expected in compliance with the limit).  For all five profiles, the resulting air quality is below the air quality standard, with design values ranging from 156.3 to 190.6 ug/m[3] (59.7 to 72.8 ppb).  This may be compared to the results for the corresponding 1-hour limit, yielding a design value of 196.4 ug/m[3] (75 ppb).  Thus, based on modeling allowable emissions, this assessment supports the finding described in Appendix B that establishment of a 30-day average limit in accordance with EPA guidance can commonly result in comparable air quality as establishment of the corresponding 1-hour limit.
      
      
      
