Document jyoV1eq3pp6Z7Jy1JoRG8RvR9

ANALYSIS OF THE UPPER TOLERANCE LIMIT PROCESS PROPOSED BY EPA FOR DETERMINING THE "MAXIMUM ACHIEVED HOURLY EMISSIONS RATE" FOR ELECTRIC GENERATING UNITS PREPARED FOR UTILITY AIR REGULATORY GROUP PREPARED BY Lowell L Smith LLS Consulting Laguna Hills, California Michael C. Hein Travelers Rest, South Carolina 2024-EPA-05254 August 6, 2007 1 Sierra Club FOIA 2024-EPA-05254 ATTACHMENT 1 ED_017426_00001808-00001 SC_EVERSPLIT0001011 Table of Contents Section Page 1 Summary 4 2 Introduction 6 3 Analysis Methodology 9 3.1 Unit Selection 9 3.2 Evaluation of EPA UTL Process 9 4 EPA UTL Process Results 11 4.1 Evaluation of Unfiltered Datasets 11 4.2 Evaluations of Datasets with Outliers Removed 12 5 Evaluation of Adjustments to the EPA UTL Methodology 13 5.1 Results for UTL Method Sorted by Pollutant Emission Rate -- Unfiltered 13 5.2 Results for UTL Method Sorted by Pollutant -- Outliers Removed 13 5.3 Results for UTL Methodology with Equivalent Highest 10% Hourly Heat Input Post- Change Dataset 14 5.4 Evaluation of the Impact of Independent Variables 14 5.5 Evaluation of Pre-Change Five-Year Operating Period 16 5.5.1 Evaluation of Calendar Years Against Highest Heat Input Year 16 5.5.2 Evaluation of Highest Heat Input Year Against 5-Year Hourly Emissions.. 16 5.6 Evaluation of a Revised UTL Calculation Process 18 6 Conclusions 20 6.1 Evaluation of EPA UTL Process 20 6.2 Evaluation of UTL Process Sorted by Pollutant Emission Rate 20 6.3 Evaluation of UTL Process with Equivalent Post-Change Heat Input Sort Dataset 20 6.4 Evaluation of Adjusting the UTL Independent Variables 21 6.5 Evaluation of UTL for 5-Year Long Operating Period 21 6.5.1 Comparing 5-Calendar Years UTL Emissions 21 6.5.2 Comparing 5 Calendar Years Hourly Emissions Against UTL Emissions 22 6.5.3 168-Hour Revision of the UTL Process 22 7 References 24 APPENDIX A Normality Tests 36 APPENDIX B Five Year EPA UTL Analysis Data Plots 55 APPENDIX C Five Year Revised UTL Analysis Data 64 2024-EPA-05254 2 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00002 SC_EVERSPLIT0001012 List of Figures Number Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Impact of Confidence Limits on Standard Deviation Multiplier Impact of Fraction of 365 Day Period on Standard Deviation Multiplier Unit El Five Year Hourly NO Average Unit El Five Year Hourly SO2 Averages Unit 1 168 Hour NO UTL Results Unit E 168 Hour SO2 UTL Results Page 27 27 30 30 34 35 List of Tables Number Table 1 Table 2 Table 3 Table 4 Table 5 Table 6 Table 7 Table 8 Table 9 Table 10 Table 11 Table 12 Table 13 Table 14 Table 15 Electric Utility Units Selected for UTL Analysis Results for EPA UTL Method for 2006 Results for EPA UTL Method for 2006, Potential Outliers Removed UTL Results for NO and SO2 Sorted Datasets for 2006, Unfiltered Dataset UTL Results for NO and SO2 Sorted Datasets for 2006, Outliers Removed Comparison of UTL False Positive Hours for Various Modifications to Datasets, Year 2006 Five Year Comparison of UTL Results for Heat Input Sorting, Years 2002 through 2006 Results for EPA Heat Input Sorting UTL Method for 5 Years, Comparing All Hourly Emissions Post-Change, Unfiltered Results for EPA Heat Input Sorting UTL Method for 5 Years, Comparing All Hourly Emissions Post-Change, Outliers Removed Results for EPA NO),/SO2 Sorting UTL Method for 5 Years, Comparing All Hourly Emissions Post-Change, Unfiltered Results for EPA NO),/SO2 Sorting UTL Method for 5 Years, Comparing All Hourly Emissions Post-Change, Outliers Removed Results for EPA NO),/SO2 Rate Sorting UTL Method for 5 Years, Comparing All Hourly Emissions Post-Change, Outliers Removed Results for EPA NO),/SO2 Rate Sorting UTL Method for 5 Years, Comparing All Hourly Emissions Post-Change -- Unfiltered 1% Population @ 99.98CI Results for EPA NO),/SO2 Rate Sorting UTL Method for 5 Years, Comparing All Hourly Emissions Post-Change -- Filtered 1% Population @ 99.98CI Comparison of Proposed UTL and Revised UTL Methods, EPA Proposed Statistics (99.9 and 90.0) Page 24 24 25 25 26 26 28 29 30 31 32 32 33 33 34 3 2024-EPA-05254 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00003 SC_EVERSPLIT0001013 1 Summary EPA has proposed a process for determining if an electric generating unit (EGU) exceeds its pre-change "maximum achieved hourly emissions rate" by utilizing an upper tolerance limit process developed by the National Bureau of Standards. This process is governed by the following equation UTL = x + s * Z 4 + i =p (Z 1 = p 2)- 2 Zi q 1 2 * (n -- 1) * z 1= p 2 2 1 Zisci 2 * (n -- 1) 2 z 1= q n The variables in this equation are the mean emission level, its standard deviation and the population of the dataset. EPA has proposed utilizing the following process to estimate whether an EGU has increased its maximum hourly emissions after a change. The process is as follows: 1. Determine the highest heat input year from 5 years of CEMS or PEMS data prior to a proposed change; 2. Sort the data based on heat input rates and eliminate the lower 90 percent of the data; 3. Calculate the UTL given by the equation above and establish this as the maximum emission rate; 4. Compare this emission rate to each hourly emission rate post-change; and 5. An emission increase would result if the hourly post-change emissions rate exceeded the UTL emission level. EPA hypothesized that this process captured the natural variability of EGUs emissions and would insure that the pre-change maximum achieved hourly emissions rate would not be exceeded simply by random variability of the system. Analyses were performed to determine if this process was viable for realistically comparing pre-change and postchange hourly emissions rates and to test EPA's hypotheses. These analyses utilized actual CEMS data from 9 coal-fired EGUs. The analyses tested the usefulness of various variations to the UTL process, including some proposed by EPA, to attempt to improve the process. These analyses evaluated the following variations to the proposed UTL process: 1. Data sorting by heat input prior to applying the UTL process (the method proposed in the Federal Register notice); 2. Data sorting by pollutant species (NOX and SO2 emission rates) prior to applying the UTL process; 3. Variation of the statistical parameters; 4. Variation of the size of the filtered dataset; 4 2024-EPA-05254 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00004 SC_EVERSPLIT0001014 5. Evaluation of an equivalent post-change dataset population (highest 10 percent of data, based on heat-input sort); and 6. Elimination of potential anomalous readings through statistical removal of outliers. The basic finding of these evaluations was that EPA's proposed UTL process does not support EPA's hypotheses. The underlying cause for this failure to support the hypotheses is that the UTL process uses a statistical process with a given level of confidence and a relatively large pre-change dataset to compare against one hour emissions data postchange. In order for this process to be viable, either the pre- and post-change datasets must be similar or the confidence level must be extremely high (close to 100 percent). The highest post-change hourly emissions data point essentially represents the far right portion of the normal distribution tail. Any confidence level less than 100 percent would potentially cause an emission exceedance. All of the analyses performed herein showed that EPA's proposed UTL process cannot be made to satisfy the EPA hypotheses based on actual EGU CEMS data. For any of the variations to the process listed above for the 2006 dataset, the best that could be achieved was that only 5 of the 9 units would not exceed the UTL utilizing the same set of data before and after change. This result was for a post-change dataset equal to the prechange dataset, i.e., 10 percent highest heat-input points of the highest 365 day emissions. Utilizing post-change 2006 hourly emissions data (as proposed by EPA) resulted in all 9 units exceeding the UTL. These results clearly show that the EPA proposed UTL process for this application is flawed and would ensure that virtually all units subject to this process would have false positives, i.e., show emission increases where no increases have actually occurred. Analysis of the EPA UTL process for the five year datasets further confirmed that the process would show that the majority of units would have increases in emissions (false positives) by this method. A slight revision to EPA's proposed UTL process results in fewer EGUs having false positives than that for EPA's original proposal. The revision more closely satisfies EPA's hypotheses stated in the Preamble. The revised UTL process follows the following steps: 1. Perform a 168-hour rolling average of the 5-year pre-change emission data, 2. Calculate the UTL based upon these 168-hour rolling averages, 3. Utilize the highest UTL to compare against the hourly emission data post- change. This revised UTL process results in only 2 of the nine units having false positives for NOx and 3 for SO2. This compares to 7 units for NOx and all nine units for SO2 having false positives for the EPA proposed UTL process. But because even this revised UTL process does not eliminate false positives altogether, additional steps should be used to account for these false positives. For example, the following two steps can be added: 4. Test the pre-change hourly emissions for false positives, and 5. In determining whether there has been a significant increase in emissions after a change, allow as a safe harbor the number of hourly false positives that occurred during the pre-change period. 5 2024-EPA-05254 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00005 SC_EVERSPLIT0001015 2 Introduction On May 8, 2007, EPA published a supplemental proposal for a revised emissions increase test for electric generating units (EGUs) (72 Fed. Reg. 26202-227). This supplemental proposal includes a new statistical alternative that calculates a maximum "achieved" hourly emission rate on either an input- (lb/hr) or output- (lb/MW-hr) basis that cannot be exceeded for even a single hour during the five years after a project is undertaken. This calculated maximum achieved emission rate is based on actual emission data for the five years prior to a proposed change at the EGU. The statistical process is based on the upper tolerance limit statistical process (UTL) developed by the National Bureau of Standards (Ref. 1). This proposed UTL statistical process is represented by the equation below: UTL = x + s * Z + 1- p z 2' 1-p , 2 z1-q 2 1 2 * 'n - 1 * Z1-p Z 2 1-q 1 2 * (n - 1) z 2 1-q n Equation 1 Where: X bar = s= Zi_p = Zi_ci = Mean value of the dataset, Standard Deviation of the dataset, 3.090, Z score for the 99.9 percentage of the interval, 2.326, Z score for the 99.0 percent confidence interval. EPA believes "the statistical approach properly accounts for the variability inherent in EGU operations and air pollution control technology." 72 Fed. Reg. 26215, col. 3. Further, EPA believes that this statistical process "helps to ensure that the emissions from an EGU will not exceed its pre-change maximum achieved hourly emissions rate simply through the random variability of the system . . . ." 72 Fed. Reg. 26215, col. 3. EPA has proposed the values of the Zl _p parameter and the Zi _q parameter above but requests comment on whether it would be appropriate to set these two parameters at lower levels of 90.0 percentage of the interval (Z1_p) and either a 90.0 or 95.0 confidence interval (Zi _q). This paper concentrates on the original proposed values of both Zi_p (99.9%) and Zi_q (99.0%) since these represent UTL values that would be higher then those for the other intervals. The impact of using higher values for these two parameters was also evaluated. EPA's statistical process can only be applied to EGUs that are equipped with CEMS or PEMS. This paper only evaluates datasets from CEMS since most utility units above 25 6 2024-EPA-05254 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00006 SC_EVERSPLIT0001016 MW have installed CEMS for NO and SO2. EPA's proposed UTL process begins by evaluating five years of CEMS data prior to the proposed change and selecting 365 contiguous days of data (not necessarily in the same calendar year) representing the highest heat input period for the EGU during the previous five years. The dataset is filtered to delete emissions data for periods of start-up and shut-down, periods of malfunction and periods of non-compliance with the appropriate emission limitations. Next these data are sorted by heat input and the highest 10 percent of the dataset are selected. For the rest of the analysis, under the proposal, the NOx and SO2 emissions rates are utilized to determine the mean and standard deviation of the datasets and each is substituted into Equation 1 along with the number of data points used in the analysis. The UTL values of NOx and SO2 derived from this process establish the pre-change maximum achieved hourly emission rate for these pollutants. These levels will later be compared to the highest hourly emission rates during the five years after the project. As with the prechange data, periods of start-up and shut-down, periods of malfunction and periods of noncompliance with the applicable emission limitations are eliminated from the hourly data before determining the maximum post-change hourly emission rate. According to the proposal, any remaining hourly rate above the UTL value during the five years after the project would be deemed to be an increase in the hourly emissions rate, potentially triggering new source review (NSR). EPA has also requested comments on the advantages of using the UTL process by first sorting on the emission rate of the actual specie (NOX and SO2 in this case) rather than sorting on the heat input to obtain the highest 10 percent of the hourly data. This paper addresses the advantages and disadvantages of this alternative sorting approach. There are a number of uncertainties and limitations relative to how certain types of emissions data are handled by the proposed EPA process. These include, but are not limited to, the following: 1. Data for EGUs whose emission controls operate seasonally, e.g., only during the ozone season; 2. Data for common stack EGUs; 3. Elimination of periods of non-compliance: a. Applicable emission limits may be based on averaging periods greater than one hour; b. An hourly emission level higher than the applicable limit may not cause an actual non-compliance event due to trading, averaging or other compliance approaches; 4. How a malfunction is determined. This paper provides insight into the efficacy of the UTL approach for single stack utility coal-fired EGUsl unit that operate with the same pollution control equipment throughout the entire year. Therefore, issues 1 and 2 do not come into play. The issue of elimination of non-compliance periods is not addressed here since the units selected for analysis comply 1 EPA's proposed rule would apply to all fossil-fired EGUs, including electric utility steam generating units and combustion turbines. The analysis in this paper focuses on 9 coal-fired EGUs. 7 2024-EPA-05254 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00007 SC_EVERSPLIT0001017 with their emission limitations, which are typically set for averaging periods greater than one hour. It is assumed in this analysis that none of the units had periods of noncompliance. As reported to EPA, none of the data used here is identified as resulting from a malfunction of pollution control equipment, the EGU or CEMS. Nevertheless, it may be that some of the outliers are due to such malfunctions. Accordingly, the effect of removing from the datasets potential malfunctions is evaluated by analyzing "filtered" datasets in which statistical outliers are eliminated. 2024-EPA-05254 8 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00008 SC_EVERSPLIT0001018 3 Analysis Methodology The following paragraphs discuss the unit selection process for the UTL analyses, the UTL analysis process, and potential variations of this process. 3.1 Unit Selection CEMS data from nine coal-fired EGUs were obtained from the EPA Electronic Data Reporting (EDR) site. DOE data were used to ascertain the basic attributes of the population of coal-fired EGUs for those units operating during the year 2005 (2006 was not available). From this 2005 DOE database, units were selected that had the following attributes: 1. High utilization or hours of operation in 2005; 2. NO emissions representative of two control technologies: a. Selective Catalytic Reduction (SCR), below 0.1 lb/MMBtu, and b. Combustion controls, between 0.5 and 0.4 lb/MMBtu; 3. Units with both wall- and tangential-firing unit designs; 4. Units with single stack CEMS; 5. Different coal types (bituminous and sub-bituminous); and 6. Different SO2 control processes (spray and venture scrubbers). Using these criteria, nine coal-fired EGUs described in Table 1 below were selected. These units are generally representative of coal-fired EGUs commonly used in the United States. The combustion-controlled and the SCR-controlled units have very different NO), characteristics. Furthermore, combustion-controlled, wall- and tangential-fired units have different characteristic NO curves relative to load. In addition, the bituminous and subbituminous-fired units generally have very different levels of NO emissions. The SO2 controls include spray and venturi sulfur oxide removal processes. The purpose of using the units listed in Table 1 was to analyze a cross-section of utility units to gain insight into whether the UTL process was sensitive to these parameters, i.e., coal type, unit type, NOx control type and, to the extent possible, sulfur oxide control type. In addition, other parameters such as sample size and confidence intervals were also analyzed. 3.2 Evaluation ofEPA UTL Process The purpose of the analysis of the nine selected units listed in Table 1 was to test the hypothesis that EPA established in its proposal, i.e., that the UTL method provides a reasonable maximum "achieved" hourly emission rate that accounts for the normal variability of EGU emission data. It tests EPA's statement that the UTL process "helps to ensure that the emissions from an EGU will not exceed its pre-change maximum emission rate simply through random variability of the system." Initially, to test these hypotheses, 9 2024-EPA-05254 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00009 SC_EVERSPLIT0001019 EDR data from calendar year 2006 was downloaded from the EPA EDR website for each unit. These data were then subjected to the process delineated by EPA to determine the UTL based on Equation 1. Next, the assumption was made that, in the future after a hypothetical physical or operational change that occurs at the unit after 2006, these units would operate identically (loads, emissions, hours, etc) as they did in 2006. The hypotheses were tested by assuming that the UTL process accurately established the maximum achieved hourly emission rate during the 365 day baseline period and that during the subsequent five years, the unit will be operated identically to the five years before the project. That is, these units were assumed to have undertaken a change in the interim that, in fact, had no effect whatsoever on the units' emissions profile for the entire year. The analysis seeks to determine whether there are any occurrences of hours where the maximum achieved emission rate is exceeded. Any such false positives would indicate a flaw in the statistical procedure, because NSR would be triggered even though there had been no change whatsoever in the emissions profile at the unit. The CEMS data analyzed are for NOx Method Code 1 or 2 EDR data and all SO2 data which represent only actual emissions -- no substituted data were used. The UTL analyses were performed exactly as described in the proposal for the EPA UTL methodology together with several other variations of this process. These variations were for various sample sizes (1 to 10 percent), tolerance levels (Zi _q = 99.0 to 99.9) and various percentages of the interval (Z1_p = 99.0 to 99.98 percent). Also, since it was not possible to extract any potential unreported malfunctions of the unit or the pollution control equipment from the CEMS data, a parallel analysis was done to eliminate potential outliers. This outlier process eliminated any emissions from the 365 day dataset that were above three standard deviations from the mean. These variations to the EPA-proposed UTL process were undertaken to evaluate the impact of the independent variables in the UTL procedure. All of the following analyses were performed to determine if the proposed UTL process was viable for evaluating postchange maximum hourly emission rates and to investigate whether adjustments to the process would improve this evaluation so that no false positives would result. Subsequent to this initial assessment, five years of CEMS data from 2002 through 2006 were also analyzed (this additional analysis is discussed in section 5.4). 2024-EPA-05254 10 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00010 SC_EVERSPLIT0001020 4 EPA UTL Process Results Two sets of analyses were performed utilizing the procedure delineated in EPA's supplemental proposal. The raw datasets from the nine units were analyzed first. A second set of analyses was performed by removing potential outliers. The outlier criterion assumed that any emissions data above three standard deviations from the mean emission level were anomalous and were removed before sorting the data by heat input. It is believed that these two sets of data bound the solution. The upper bound being the unfiltered dataset and the lower bound being for the dataset with potential outliers removed. 4.1 Evaluation of Unfiltered Datasets Table 2 below shows the results of the analysis for the unfiltered dataset for both NO and SO2. The column labeled "False Positives" represents the number of times the full 365 day dataset had hourly emission values above the UTL value. As can be seen in Table 2, only one unit (Unit B) shows no false positives. Most of the units had varying amounts of false positives. For NON, this ranged from 1 to 129 hourly NO false positives. For SO2, this ranged from 1 to 687 SO2 false positives. Potentially, some of these exceedances could have been the result of outliers that may not be representative of normal variations in unit performance and control and measurement processes. The results for this unfiltered case should be considered as the upper bound of the potential exceedances. The National Bureau of Standards UTL equation was developed for datasets that are normally distributed (Ref. 1). The Shapiro-Wilks (W) coefficient is a measure of the normality of a distribution -- 1.0 being normal and those below potentially being nonnormal. Datasets with W close to 1.0 could reasonably be considered normally distributed. The SO2 rate datasets for the units analyzed have Shapiro-Wilks (W) coefficients which are generally around 0.97, however three units have a W coefficients of 0.438, 0.766 and 0.862. Similarly, the NON-rate W coefficients are generally around 0.97, but three are 0.438, 0.871 and 0.904. These low-coefficient units may have datasets that depart significantly from a normal distribution. In a perfectly normally distributed emission dataset, there would not be a finite maximum emission rate. However, there is a practical maximum emission rate dependent on the uncontrolled emission level and the amount of coal burned. Whether the actual distribution is normally distributed or not does not assure that the EPA hypotheses are valid. At the 99.9 percent confidence interval, the potential for over 400 false positives are possible in a five year period. Therefore whether the distribution is normal or not cannot significantly change the results since the tolerance limits establish the potential for exceeding the UTL. The only way to insure that the UTL process will satisfy the EPA hypotheses is to set the tolerance level extremely high (99.9977 % for one exceedance in 5 years). This will be illustrated in the analyses provided in Section 5. The normality of the emissions dataset may have less impact on 11 2024-EPA-05254 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00011 SC_EVERSPLIT0001021 the attainability of the UTL method than the actual confidence levels used to set the UTL. The statistics for these 18 emission datasets are included in Appendix A. 4.2 Evaluations ofDatasets with Outliers Removed To set a lower bound for the potential failures of the UTL process, an outlier removal process was used on the datasets for the nine units analyzed above. As a surrogate for removing potential anomalous outliers, all emissions data above three standard deviations were removed from the datasets for the nine units. Table 3 shows the results for this lower bound analyses. Even with the potential outliers removed, eight of the nine units still have many false positives after a hypothetical change at the unit. These filtered results (outliers removed) should be considered the lower bound of the potential false positives. 2024-EPA-05254 12 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00012 SC_EVERSPLIT0001022 5 Evaluation of Adjustments to the EPA UTL Methodology There are a number of adjustments to the proposed UTL methodology that might reduce the number of false positives found in Tables 2 and 3 above. EPA has suggested two possibilities: 1. Utilize the UTL methodology by sorting data by the pollutant (NOX or SO2 rate) rather than the heat input; or 2. Change the statistical parameters related to the confidence interval and percentage of data. The following sections provide analyses related to changes that may improve the UTL process. 5.1 Results for UTL Method Sorted by Pollutant Emission Rate Unfiltered Analyses similar to that in Section 4.0 were performed to evaluate the effect of sorting the initial datasets by NO and SO2 rates (instead of heat input) before eliminating the lower 90 percent of the data. From a phenomenological point of view, this makes much more sense than sorting by heat input. It is not necessarily true that high heat input translates into measured high emissions rates. It would make much more sense to sort by pollutant emission rate since that is what is used in the "After" change period to assess whether or not the UTL was exceeded. Table 4 shows the results of applying the UTL process on the datasets sorted by NOx and SO2 rates. The major difference between these results and those shown in Table 2 are: 1. The single unit that did not have a false positive now had some; and 2. The overall, total number of false positives decreased dramatically. While there were as many as 134 NOx false positives for the methodology as proposed (i.e., sorting by hourly heat input rates), the maximum number of false positives for this adjustment resulted in a maximum of 28. Similarly, there were 682 SO2 false positives for the methodology as proposed and only a maximum of 42 for the adjusted methodology. It appears that this methodology would be closer to satisfying EPA hypotheses than sorting by heat input. Nevertheless, none of the units were able to operate without triggering false positives for either NOx or SO2. 5.2 Results for UTL Method Sorted by Pollutant - Outliers Removed To bound the potential false positives, the data for the pollutant-sorted analyses were performed for datasets with the outliers removed. All emissions data for NOx and SO2 13 2024-EPA-05254 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00013 SC_EVERSPLIT0001023 emissions rates above 3 standard deviations from the dataset mean were eliminated. The UTL process was then performed on the remaining dataset. Table 5 shows the results for this filtered dataset. As with the unfiltered data, none of the units were able to avoid false positives for both NO), and SO2. Again, it appears that this methodology would be closer to satisfying EPA's hypotheses than sorting by heat input. The maximum number of false positives for any unit for NOx went from 28 to 18 and the maximum for SO2 went from 42 to 11 by eliminating the outliers. But none could simultaneously avoid triggering false positives for both NOx and SO2. 5.3 Results for UTL Methodology with Equivalent Highest 10% Hourly Heat Input Post- Change Dataset Currently, the process proposed by EPA calculates a UTL value based on 10 percent of the available data (sorted on the basis of heat input) in a 365 day period before a change and compares it to hourly data for the entire 365 day period after the change. It seems inappropriate to compare different populations of data before and after a change, since the UTL is determined for emission rates, not heat input rates. An adjustment that could make the UTL process more reasonable would be to compare the same type of data "before" and "after" the change. This adjustment would compare the hourly emission data from the highest 10 percent (sorted on the basis of heat input) of the dataset "after" the change to the UTL calculated on the basis of the highest 10 percent heat input dataset before the change. This would ensure that the before and after change datasets were based on the same sorting criteria. Table 6 summarizes the results of this analysis. Columns 5, 6, 9 and 10 show the results for the UTL process using only the highest 10 percent of the heat input dataset, both for calculating the UTL and in the "after-change" dataset, for both unfiltered and filtered (outliers removed) datasets. For comparison purposes, columns 3, 4, 7 and 8 show the results for the UTL process using all of the emissions data "after" the change for both unfiltered and filtered datasets (i.e., using EPA's proposed sorting and comparison methods, as described earlier in section 4). These results show that limiting the datasets before and after the change to the same highest 10 percent of heat input data improves the possibility of avoiding false positives. This is an improvement in the process, but some units still trigger false positives. 5.4 Evaluation of the Impact ofIndependent Variables The independent variables that can be changed in the EPA UTL formula are: 1. Zi_p, the percentage of the interval; 2. Zi-q, the percent confidence interval; and 3. N, the percentage of the 365 day period used. 14 2024-EPA-05254 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00014 SC_EVERSPLIT0001024 Since the analyses performed to this point have shown that setting the parameters at those proposed by EPA result in false positives at eight of the nine units analyzed, it is evident that the Zi _p and the Zi _q values must be increased, rather than decreased as suggested by EPA, to obtain more reasonable results. For the sake of illustration, the maximum for these parameters will be set at 99.98 percent which represents approximately 2 potential false positives per year. The following shows the impact of a range of these values: 1. Zl _p from 99.0 to 99.98 percent 2. Zi_q from 90 to 99.98 percent Figure 1 shows the impact of these parameters on the standard deviation in the UTL equation. The middle (red) line represents the EPA-proposed value for the percentage interval. The bottom (blue) line represents an alternate that EPA has suggested. Obviously, since the UTL as proposed results in many false positives, the lower values of these parameters are not appropriate. Comparing the low end factor for the data interval (99.0 %) with the high end value of the data interval (99.98 %) shows that the maximum increase in the standard deviation coefficient is approximately 18 percent. For some of the units, this is not sufficient to avoid false positives. Therefore, the only other independent variable to change would be the size of the dataset. Currently it is set at 10 percent of the 365 day period. Based upon the UTL equation, decreasing this parameter would increase the standard deviation multiplier and possibly allow all units to avoid false positives. Decreasing the percentage of the 365 day data used to calculate the UTL also changes the value of the standard deviation. The results for simply decreasing the amount of data analyzed are shown below in Figure 2. This figure shows that as the size of the dataset is decreased, the standard deviation multiplier increases. For the sake of setting a lower bound for these parameters, one could analyze only one percent of the data and employ a 99.98 confidence interval. This would likely decrease the number of false positives provided that the standard deviation itself did not decrease drastically as well. Table 7 shows the results of these analyses for filtered datasets comparing the NOx and SO2 UTL before change to the highest hourly emissions rate after change and also to the highest emission rate for 1 percent of the data after change. The number of false positives was determined for the nine units for the case where only one percent of the data was used in the UTL determination. For the analysis comparing all of the hourly emission data after change, 3 units had NOx false positives and 4 units had SO2 false positives. For the analysis for 1 percent of the highest heat input data after change, 1 unit had NOx false positives and the same unit had a SO2 false positive. This latter analysis method, using 1 percent of the data before and 1 percent after, comes very close to satisfying the EPA hypotheses. 2024-EPA-05254 15 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00015 SC_EVERSPLIT0001025 5.5 Evaluation of Pre-Change Five-Year Operating Period The following analyses utilize five years (2002 to 2006) of emission data for the 9 units. Previous sections concentrated on the statistical nature of the process. The following provides insight into the year-to-year variation of the UTL and also provides analyses of the UTL process for the original proposed methodology (i.e., using 5 years worth of data, instead of just one year -- 2006). 5.5.1 Evaluation of Calendar Years Against Highest Heat Input Year The previous analyses compared the 2006 pre-change UTL against the same dataset post-change. The following analysis compares the UTL for the highest heat input calendar year against the 4 other calendar years and against the highest heat input calendar year as well. Table 8 shows the results of these analyses. This table illustrates several shortcomings of EPA's proposed method using the 10 percent highest heat input for a 365 calendar day period against hourly emissions for the calendar year ("All Data After" columns). In addition, results are shown in Table 8 for an analysis in which the UTL process is modified by using the same 10 percent of the highest heat input both pre- and post-change data ("10% Data After" columns). The data shown in Table 8 is for the unfiltered emission data (no outliers removed). In addition to the fact that it was previously shown that the process may not be viable due to basic statistical issues, the major shortcomings that become evident as a result of the five-year period analyses are briefly summarized below: 1. The highest UTL year for NO is not necessarily the highest year for SO2 for 4 of the 9 units as illustrated by the green shaded data. 2. Maximum emissions in calendar years other than the highest UTL calendar year cause false positives for 4 of the 9 units as illustrated by the yellow shaded data, i.e., the highest UTL year did not capture the highest emission rate. 3. The UTL varies significantly for year to year. Utilizing a sort by pollutant (EPA's alternate approach) would potentially solve item 1 above. It would not solve the issues related to other years with maximum emissions greater than the highest UTL year during the five year period. 5.5.2 Evaluation of Highest Heat Input Year Against 5-Year Hourly Emissions Analyses were performed for the exact UTL process proposed by EPA for the five year period from 2002 to 2006 (i.e., assuming a change that is undertaken at the end of 2006). The highest heat input year was determined from this five year dataset irrespective of the calendar year. The UTL calculated from this highest year established the maximum NOx and SO2 emission rates. Figures 3 and 4 illustrate this process for Unit E1. Appendix B contains similar plots for the other 8 units analyzed. Tables 9 and 10 show the results for 16 2024-EPA-05254 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00016 SC_EVERSPLIT0001026 the proposed methodology sorting on heat input for the dataset with all EDR emission data (no exclusions) and for the same dataset with potential outliers removed, respectively. For this analysis the five-year period was assumed to be exactly the same for pre-change and post-change. That is, each unit was assumed to operate identically in all respects in the 5 years after a hypothetical change undertaken at the end of 2006 to the way the unit operated before that change. Table 9 shows that 7 out of the 9 units had false positives for NOx and that all had false positives for SO2. Table 10 illustrates the same results for the emission dataset filtered to eliminate potential outliers. For this filtered dataset, 5 out of the 9 units had false positives for NOx and 7 out of 9 had false positives for SO2. Two units (B5 and D4) had no false positives for either NOx or SO2. Consequently, 7 out of the 9 units had false positives for either NOx or SO2. While this shows that some units could meet the test, the majority would not. For the alternate methodology suggested by EPA, the emission data were sorted by species (NOX or SO2 emission rates) and the highest 365 day period for these were used to set the UTL maximum emission level. Tables 11 and 12 show the results for unfiltered and filtered emission datasets, respectively. Similar to the heat-input sorted results (Tables 9 and 10), the majority of the units had false positives. For the unfiltered dataset (Table 11), all units had false positives. For the filtered dataset (Table 12), all but one unit had false positives. As with the heat-input sort methodology, the emission rate sort alternative methodology resulted in the majority of the units having false positives. Shortening the sorting period to one percent and changing the basic statistics to 99.98 percent confidence interval potentially would be a method for improving the UTL process. EPA appears to have selected the ten percent sorting criterion as a way to characterize the emissions rate of a unit at maximum output. If that is the case, one percent would be even more characteristic of the emissions rate of the unit at maximum output. The NSR consequences of an EGU failing the NSR test are much greater than those for rulemakings in which an emission rate standard (or limit) is set, such as NSPS. Therefore, if EPA uses a statistical method, such as the UTL process, to determine whether there has been an increase in the hourly emissions rate after a project, it is extremely important to assure that the UTL process does not show false positives for the population of EGUs. For the nine units analyzed, EPA's proposed statistics for the NSR test may result in as many as 124 and 425 false positives in five years for either NOx or SO2, respectively. Unlike the exceedance of an emissions limit, such as an NSPS, failure of the NSR test for a significant emission increase has potential monumental costs associated with retrofits of costly control equipment. One failure of the test could result in the EGU being required to retrofit controls costing many hundreds of millions of dollars. Since the current statistics (99.9 and 90.0 %) have the potential to fail the test numerous times, the statistics should justifiably be increased to allow fewer potential exceedances. For one exceedance in five years, the confidence interval would have to be set at 99.998 percent. EPA has in the past established a 1 violation of the NSPS in 10 years criterion. For 24 hour block days, this amounts to a confidence interval of 99.9726 percent which was used in the 1978 NSPS and subsequent rulemakings (References 2, 3, 4 & 5). Had the 1978 NSPS been on an 17 2024-EPA-05254 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00017 SC_EVERSPLIT0001027 hourly basis, the confidence interval would have been set at 99.9989 percent for one hourly violation in 10 years. Table 13 and 14 show the results for the data sorted by species for a one-percent sorting criterion and a confidence interval of 99.98 for both unfiltered and filtered data, respectively. While this decreases the number of units that have false positives, still only two units can have no false positives for both SO2 and NOx. The process analyzed above is exactly that proposed by EPA. The results show that a majority of the units analyzed could not operate without false positives regardless of the sorting method, statistics or population size. 5.6 Evaluation of a Revised UTL Calculation Process EPA proposed to utilize a two step process for calculating the upper tolerance limit for EGU emissions. The first step was to identify the highest 365 day period for either heat input or NOx and SO2 The second step was to determine the highest 10 percent of the values for these parameters and calculate the UTL based on this 10 percent population. By selecting only the highest values of the parameters within the 365 day period, the process effectively distorts the natural variability of the data during that period. It essentially minimizes the standard deviation of the data (variability) and causes the UTL to be lower than that for the actual 365 day dataset. This process makes the UTL artificially low particularly when it is to be tested against hourly data after the change. An improvement to eliminate this artificially low UTL would be to calculate it using all of the data in the highest period. In the case of the EPA proposed methodology, this would be for the entire 365 contiguous days. The disadvantage of utilizing this large population (8760 hours) is that it masks the true normal response of a utility EGU. Utility EGUs have both diurnal and weekly variation depending upon the time of year. These variations within a day or a week can be much greater than those represented by an annual average and significantly greater than the highest 10 percent of the annual average. Therefore, a more realistic time period for selecting the highest emissions would be in the order of a week (168 hours). This captures both the diurnal and weekly variations that typical utility EGUs experience during the year. Utilizing the UTL calculated on this basis likely would more closely fit the hypotheses stated by EPA in the preamble. The methodology for this revised UTL process would be to first determine the rolling average 168-hour periods for both NOx and SO2 during the 5 years prior to the change. Then determine their UTLs based upon the statistics for these periods. And finally, select the highest UTL for the two species and utilize this value to compare against the hourly NOx and SO2 emissions levels after the change. Since this revised process and the UTL process proposed by EPA both have the potential for resulting in false positives of the NSR test, a test for internal consistency should be utilized to prevent false positives that might have actually occurred during the before change period. As a check against internal consistency, the number of false positives would be determined for the five year period prior to the change. The resulting number of false positives determined in the before 18 2024-EPA-05254 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00018 SC_EVERSPLIT0001028 change period would then be allowed for the 5-year period after the change. If the number of measured hourly emissions rates that exceed the UTL in the 5 years after the project is not more than the allowed number of exeedances, there is no increase in the maximum achieved hourly emissions rate. Figures 5 and 6 show the 168-hour UTL process described above for Unit E using the statistics proposed by EPA , i.e., Zl _p = 3.090 for the 99.9 percentage of the interval and Zi _q = 2.326 for the 99.0 percent confidence interval. As can be seen from the figures, the maximum UTL values for NOx and SO2 for this process were 6007 and 1658, respectively. Utilizing these UTL values resulted in no false positives for NOx and 15 false positives for SO2 This compares with 41 false positives for NOx and 425 false positives for SO2 for the proposed EPA UTL process. Table 15 shows the results for this revised 168-hour UTL process for the nine units analyzed previously. Appendix C provides plots of the 168 hour rolling averages for remaining 8 units analyzed. Clearly this revised 168-hour UTL process minimizes the number of units with false positives compared to the original process proposed by EPA. In addition, it minimizes the number of allowed false positives. Increasing the NSR statistics to something greater than presently proposed would result in fewer potential failures for both the EPA proposed process and the alternate 168-hour UTL process For the alternate 168-hour process, it would likely result in only one of the nine units failing the NSR test without the use of an allowed number of exceedances. 2024-EPA-05254 19 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00019 SC_EVERSPLIT0001029 6 Conclusions The analyses described in the previous sections were performed to evaluate EPA's proposed Upper Tolerance Limit (UTL) method for establishing an EGU's pre-change maximum "achieved" hourly emission rate. Specifically, the analyses were performed to test EPA's hypotheses regarding whether the UTL process adequately captures the variability of EGU emissions and whether the UTL level is properly set so that no "false positives" will result. The analyses considered data from 9 coal-fired EGUs. The following briefly summarizes the conclusions: 6.1 Evaluation ofEPA UTL Process X The UTL process applied to raw EDR data for 9 units showed that 8 had false positives. X The UTL process applied to filtered (outliers removed simulating unreported malfunctions) EDR data for 9 units also showed that 8 had false positives. The conclusion based on these analyses is that the UTL process for the majority of utility units would likely result in false positives. 6.2 Evaluation of UTL Process Sorted by Pollutant Emission Rate X The UTL pollutant sort process applied to unfiltered raw EDR data for 9 units showed that all would have false positives. X The UTL pollutant sort process applied to filtered EDR data for 9 units again showed that all would have false positives. The conclusion based on these analyses is that, while sorting on the basis of pollutant emission rates improves the process (i.e., shows less overall false positives), all of the units analyzed would have false positives. 6.3 Evaluation of UTL Process with Equivalent Post-Change Heat Input Sort Dataset X The UTL process applied to unfiltered EDR data for 9 units showed that 5 had false positives. X The UTL process applied to filtered EDR data for 9 units showed that 4 had false positives. 2024-EPA-05254 20 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00020 SC_EVERSPLIT0001030 The conclusion based on these analyses is that utilizing the same population (highest 10 percent heat input rates) post-change decreases the number of false positives. Nevertheless, many units still would have false positives. 6.4 Evaluation of Adjusting the UTL Independent Variables X Decreasing the Zi_p and Zi_q parameters below those proposed would result in more false positives. X Increasing the Zi _p and Zi _q parameters to 99.98 percent (representative of - 2 potential failures per year) would likely result in significantly fewer units having false positives. X Decreasing the heat input sample size N below 10 percent would likely result in significantly fewer units having false positives provided the standard deviation was approximately the same. X The UTL process applied to filtered EDR data for 9 units for the parameter extreme values of Zi _p= 99.98, Zi _q + 99.98 and N = 1 percent for the entire 365 day period shows that 3 units had NOx false positives and 4 units had SO2 false positives. X The UTL process applied to filtered EDR data for 9 units for the parameter extreme values of Zi _p= 99.98, Zi _q + 99.98 and N = 1 percent for the same postchange population (i.e., highest 10 percent heat input dataset) shows that 2 units had NOx false positives and 1 unit had SO2 false positives. Based on these analyses, even if the UTL parameters were set at the extremes, many units would still have false positives. 6.5 Evaluation of UTL for 5-Year Long Operating Period The following analysis shows the results of applying the UTL process to five years of data for the 9 units. The first analysis compared the UTL calculated for each calendar year to that for other calendar years. The second analysis compared the UTL calculated based on the 365-day period with the highest parameter (heat input, NOx or SO2 emissions rates) to the hourly emission rate for the five year period. This latter analysis simulates the proposed and alternate UTL processes described in the supplemental proposal. 6.5.1 Comparing 5-Calendar Years UTL Emissions X Comparing 5 calendar years of emission data against the other 4 years using the heat input sort UTL process showed that the highest UTL year for NOx was not necessarily the highest year for SO2 for 4 of the 9 units for unfiltered emission datasets. X Comparing 5 calendar years of emission data against the other 4 years using the heat input sort UTL process showed that the maximum emissions in calendar years other than the highest UTL calendar year caused false positives 21 2024-EPA-05254 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00021 SC_EVERSPLIT0001031 for 4 of the 9 units for unfiltered emission datasets, i.e., the maximum emissions did not occur in the highest UTL calendar year. X The UTL varies significantly from year to year. 6.5.2 Comparing 5 Calendar Years Hourly Emissions Against UTL Emissions X Comparing 5 years of emission data using the heat input sort UTL process proposed by EPA resulted in all 9 units having false positives for the unfiltered dataset. X Comparing 5 years of emission data using the heat input sort UTL process proposed by EPA resulted in 7 of the 9 units having false positives for the unfiltered dataset. X Comparing 5 years of emission data using the alternate NOx and SO2 emissionrate sort UTL process proposed by EPA resulted in all 9 units having false positives for the unfiltered dataset. X Comparing 5 years of emission data using the alternate NOx and SO2 emissionrate sort UTL process proposed by EPA resulted in 8 of the 9 units having false positives for the unfiltered dataset. 6.5.3 168-Hour Revision of the UTL Process X Revising the UTL period from 365 days to 168 hours and utilizing the statistics from this 168-hour period improves the UTL process. X The revision more closely satisfies EPA's original hypotheses, X The number of units showing false positives under the UTL test for NOx decreased from 6 to 2 in the revised process, X The number of units showing false positives under the UTL test for SO2 decreased from 9 to 3 in the revised process, X For units showing false positives under the UTL test, the maximum number of NOx false positives decreased from 124 to 8 in the revised process, X For units showing false positives under the UTL test, the maximum number of SO2 false positives decreased from 490 to 15 in the revised process. The general conclusion based on these analyses for CEMS data from 9 coal-fired EGUs is that the hypothesis that EPA used to establish its proposed UTL process was not proven out by applying it to typical EGU CEMS datasets. The datasets utilized avoided the potential complications brought into the process by CEMS data for units that do not operate with the same emission requirements during the entire year (Ozone Season units). Since the analyses were performed for units with high capacity factors, it also did not address the complications arising from cyclic operation of EGUs with relatively small capacity factors. As a consequence, the analyses performed herein are felt to be the best that could be expected. Therefore, under the best conditions, EPA's proposed UTL process does not support the EPA hypotheses. 22 2024-EPA-05254 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00022 SC_EVERSPLIT0001032 There were two analyses discussed in this paper that resulted in a reasonably small number of positives: (1) the proposed UTL process, revised to use a one-percent sorting criterion and a confidence interval of 99.98; and (2) the 168-hour UTL process. These analyses minimized the number of false positives, but they did not eliminate them. Therefore, additional steps should be used to provide internal consistency and account for the remaining false positives. For example, the following two steps can be added: X Test the pre-change hourly emissions for false positives, and X In determining whether there has been a significant increase in emissions after a change, allow as a safe harbor the number of hourly false positives that occurred during the pre-change period. 2024-EPA-05254 23 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00023 SC_EVERSPLIT0001033 7 References 1. Mary Gibbons Natrella, "Experimental Statistics," NBS Handbook 91, U.S. Department of Commerce, 1963. 2. Standards of Performance for Electric Utility Steam Generating Units for which construction commenced after September 18, 1978 40 CFR 60, Subpart Da, promulgated June 11, 1979. 3. Proposed New Source Performance Standards for Industrial-Commercial -- Institutional Generating Units, 40 CFR, Subpart Db, proposed June 19, 1986. 4. Statistical Analysis of Wet Flue Gas Desulfurization Systems and Coal Sulfur Content, Volume 1: Statistical Analysis, Radian Corporation, August 18, 1983 EPA Report No. 68-20-3816. 5. Fed. Reg. Vol. 40, No. 194 -- Monday, October 6, 1975, 60.45. 2024-EPA-05254 24 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00024 SC_EVERSPLIT0001034 Plant TABLE 1 Electric Utility Units Selected for UTL Analysis Unit Unit WSize, Boiler Type NOx Control SO2 Control M Fuel E 1 780 Venturi Sub-Bituminous A 2 775 Tangential Combustion Spray Bituminous C 3 510 Spray Bituminous B 5 770 Spray Bituminous F 1 700 Dry Bottom Spray Bituminous Wall Combustion G 1 790 Spray Bituminous D 4 800 Spray Sub-Bituminous H 1 635 Dry Bottom SCR Spray Bituminous Wall I 8 650 Tangential SCR Venturi Sub-Bituminous Plant TABLE 2 Results for EPA UTL Method For 2006 Unit NOx SO2 N UTL NSR Max NOx - Number UTL NSR Max SO2- All Number NOx Before All Data False SO2 Before Data After False After Positive Hrs Positive Hrs E 1 795 4361 4829 11 1,096 2,117 49 A 2 775 2985 2854 0 9,766 10,257 2 C 3 506 544 561 1 2,538 2,561 1 B 5 769 4848 4686 0 9,443 9,328 0 F 1 759 4337 4319 0 6,092 6,170 1 G 1 788 3886 4614 33 5,388 6,705 1 D 4 793 2525 2603 4 2,332 3,595 687 H 1 783 1309 2801 129 1,825 15,366 400 I 8 857 311 603 36 1,325 1,551 123 25 2024-EPA-05254 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00025 SC_EVERSPLIT0001035 TABLE 3 Results for EPA UTL Method for 2006 Potential Outliers Removed Plant Unit UTL NSR NOx Before NOx SO2 Max NOx All Data After Number False Positive Hrs UTL NSR Max SO2- All Number SO2 Before Data After False Positive Hrs E 1 4361 4829 11 1023 1061 19 A 2 2985 2854 0 9766 10257 2 C 3 544 561 1 2538 2561 1 B 5 4848 4686 0 9443 9328 0 F 1 4337 4319 0 6092 6170 1 G 1 3810 4254 10 5350 4214 0 D 4 2525 2603 4 2333 3055 682 H 1 785 1123 134 1829 4084 317 I 8 311 331 7 1325 1431 98 2024-EPA-05254 TABLE 4 UTL Results for NOx and SO2 Rate Sorted Datasets for 2006 Rate Sorted Datasets for 2006 Unfiltered Dataset Plant Unit UTL NSR NOx Before NOx Max NOx All Data After SO2 Number UTL NSR Max SO2- All Number False SO2 Before Data After False Positive Hrs Positive Hrs E 1 4,272 4,829 15 1,356 2,117 17 A 2 2,862 2,854 0 9,311 10,257 10 C 3 539 561 1 2,512 2,561 2 B 5 4,421 4,686 5 8,720 9,328 11 F 1 4,137 4,319 5 5,210 6,170 9 G 1 4,123 4,614 28 4,439 6,705 4 D 4 2,479 2,603 8 3,108 3,595 4 H 1 2,357 2,801 5 11,969 15,366 42 I 8 412 603 23 1,504 1,551 5 26 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00026 SC_EVERSPLIT0001036 TABLE 5 UTL Results for NOx and SO2 Rate Sorted Datasets for 2006 Outliers Removed Plant NOx S02 Unit UTL NSR NOx Before Max NOxAll Data After Number False Positive Hrs UTL NSR Max SO2-All Number SO2 Before Data After False Positive Hrs E 1 4,272 4,829 15 1,064 1,061 0 A 2 2,862 2,854 0 9,311 10,257 10 C 3 539 561 1 2,512 2,561 2 B 5 4,421 4,686 5 8,720 9,328 11 F 1 4,137 4,319 5 5,210 6,170 9 G 1 3,772 4,254 11 4,275 4,214 0 D 4 2,479 2,603 8 3,073 3,055 0 H 1 1,143 1,123 0 3,508 4,084 10 I 8 303 331 18 1,461 1,431 0 TABLE 6 Comparison of UTL False Positive Hours for Various Alterations to Datasets Year 2006 Plant 1 E A C B F NOx SO2 EPA Method - EPA Method - EPA Method - EPA Method - Unit EPA UTL EPA Method - Unfiltered Filtered using EPA UTL EPA Method - Unfiltered Filtered using Method - Filtered using 10% 10% Highest Method - Filtered using 10% 10% Highest Unfiltered Highest HI HI Data After Unfiltered Highest HI HI Data After Data After Data After 2 3 4 5 6 7 8 9 10 1 11 11 8 8 47 16 8 2 2 0 0 0 0 2 2 0 0 3 1 1 0 0 1 1 0 0 5 0 0 0 0 0 0 0 0 1 0 0 0 0 1 1 0 0 G 1 33 10 4 0 1 0 0 0 D 4 4 4 4 4 662 658 17 17 H 1 129 134 17 3 331 327 8 7 I 8 36 7 2 2 125 99 11 11 27 2024-EPA-05254 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00027 SC_EVERSPLIT0001037 Standard Deviation Multiplier 4.0 3.9 3.8 3.7 3.6 3.5 3.4 3.3 3.2 3.1 3.0 2.9 2.8 2.7 2.6 252.4 99.0% FIGURE 1 Impact of Confidence Limits on Standard Deviation Multiplier -Z(1-p)=99% -Z(1-p)=99.9% -Z(1-p)=99.98% 99.1% 99.2% 99.3% 99.4% 99.5% 99.6% Confidence Level (Z ") 99.7% 99.8% 99.9% 100.0% UTL Standard Deviation Multiplier Figure 2 Impact of Fraction of 365 Day Period on Standard Deviation Multiplier -Z(1-p)=99.9% 5.0 4.8 - 4.6 4.4 4.2 4.0 3.8 3.6 3.4 3.2 3.0 2.8 2.6 2.4 0.0% 1.0% Z(1-q)=99% -z(1-p)=99.9% Z(1-q)=99.9% -z(1-p)=99.98% Z(1-q)=99.98% 2.0% 3.0% 4.0% 5.0% 6.0% 7.0% Percent of 365 Day Data Used in UTL Equation 8.0% 9.0% 10.0% 28 2024-EPA-05254 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00028 SC_EVERSPLIT0001038 TABLE 7 Comparison of UTL False Positive Hours for One Percent of Dataset Year 2006 Sorted on Highest Heat Input for Filtered Data Z1-p and Z1-q = 99.98 for 1 Percent of Data NOx SO2 Plant Unit EPA Method - All Data EPA Method - 1 % Of EPA Method - All Data EPA Method - 1 % Of After Change Data After Change After Change Data After Change E 1 0 A 2 0 C 3 0 B 5 0 F 1 0 G 1 25 D 4 0 H 1 5 I 8 27 0 16 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 21 0 2 135 1 0 0 0 2024-EPA-05254 29 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00029 SC_EVERSPLIT0001039 Table 8 Five Year Comparison of UTL Results for Heat Input Sorting Years 2002 Through 2006 Plant and Unit Operating Year NOx All Data After UTL rT,4", Max NOx- Number BOIL All Data PBt)osloitricve Flcuri i After E-teeding 10 % Data After Max NOx- Number False 10% Data After Hrs UTL 1.'6k `. SO2 All Data After Max_.._i., - Number All Data Hours After Ex:ea.:ling 10 % Data After Max SO2 10% Data After Number False Positive Hrs 2002 4334 4485 5 4485 5 1045 1020 '320 2003 3:317 3943 10 3943 10 1005 10115 6 'T5 6 El 2004 4 - - 4354 10 4354 10 1097 1084 '984 2005 4068 4138 6 4138 6 1014 10 07 1007 2006 4272 4829 4829 1064 1061 1061 2002 3162 3730 3230 2 9775 10173 4 10123 2003 %906 3 f-1 3 1 3063 1 9737 Ydlf-, 5 9916 A2 2004 :e55 2760 6 2760 6 96.39 10/ 17 19 10717 19 2005 2775 2776 1 2776 1 91~7 9455 9466 2006 2H2 2854 2854 9311 10257 1 I 11257 10 2002 L27 7 653 7 272:8 26,31 2561 2003 699 787 9 787 9 :3178 31 14 3114 C3 2004 640 671 671 9 2624 2605 2505 2005 552 571 571 2 2739 2710 27'0 2006 539 561 1 561 1 2512 2 2561 2 2002 4393 4762 9 4762 9 8296 8 3Y 8 8 2003 4302 4337 4 4337 4 8130 15 3355 15 B5 2004 4Ligi 4460 = 4460 1 1 2005 46 2 50&1 4 5064 4 9532 10062 4 111232 4 2006 4421 4686 4686 8720 9328 11 9328 11 2002 4090 4907 6 4907 6 6607 0485 6485 2003 4047 43'6 10 4376 10 6189 x;484 5181 Fl 2004 4144 4325 4 4325 4 5126 5375 11 5375 11 2005 4305 4424 4 4424 4 5521 5750 7 5750 7 2006 4137 4319 5 4319 5 5710 6170 9 3'70 9 2002 5193 4905 4905 3796 3878 7 3878 7 2003 5374 5195 5196 3803 3922 4 3372 4 G1 2004 4507 4- 4538 1 3902 4077 11 4977 11 2005 4055 41'.=1 2 4139 2 4527 4679 5 4679 5 2006 3772 4 )54 11 4254 11 4275 4214 4214 2002 2951 3070 6 3070 6 2534 244=1 2449 2003 2908 'il3i1 7 3080 7 27co 23211 2 2320 2 D4 2004 2270 )4'1 16 2471 16 2185 2219 3 2210 3 2005 2340 2:304 3 2:384 3 2296 2327 4 2327 4 2006 2479 211:0x3 0 760:3 3 2005 2325 18 2301 13 2002 2360 2706 2786 2u4:2 1978 19Th 2003 5692 4633 4633 2990 2903 2903 H1 2004 4132 3388 3388 3982 4146 3 4146 3 2005 1715 1402 1402 3425 4177 21 4177 21 2006 1143 1123 1123 3508 4084 13 41)64 13 2004 537 586 15 566 15 1165 1413 1413 18 2005 314 353 12 353 12 1366 1337 1337 2006 303 331 18 331 18 1161 1431 1431 2024-EPA-05254 30 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00030 SC_EVERSPLIT0001040 Figure 3 Unit El Five Year Hourly NOx Averages UTL for NOx Calculated for Highest 365 Day Heat Input 6,000 UTL for NOx = 4330 No. of Pts. Above UTL = 41 5,000 - 4,000 :6 0 7) 3,000 E w x z 2,000 - 1,000 - 0 1/1/2002 1/1/2003 1/1/2004 1/1/2005 1/1/2006 1/1/2007 Figure 4 Unit El Five Year Hourly SO2 Averages UTL for SO2 Calculated for Highest 365 Day Heat Input 2,500 UTL for SO2 = 951 No. of Pts Above UTL = 425 2,000 Highest Heat Input Year 4I 1,500 0 6 500 0 1/1/2002 1/1/2003 1/1/2004 1/1/2005 1/1/2006 1/1/2007 31 2024-EPA-05254 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00031 SC_EVERSPLIT0001041 TABLE 9 Results for EPA Heat Input Sorting UTL Method for 5 Years Comparing All Hourly Emissions Post-Change Unfiltered Plant NOx SO2 Unit UTL NSR Max NOx - Number UTL NSR Max SO2- All Number Mean NOx NOx Before All Data False Mean SO2 SO2 Before Data After False After Positive Hrs Positive Hrs E 1 3537 4330 4838 41 665 951 3,616 425 A 2 2586 3111 5749 47 8,216 10,132 10,717 12 C 3 541 653 787 35 2,178 3,073 3,114 12 B 5 4195 5449 5084 0 8,124 10,723 11,841 2 F 1 3400 4181 4907 124 4,361 6,260 7,154 21 G 1 3352 5103 7712 41 2,773 4,204 7,719 61 D 4 2652 3061 3080 3 1,652 2,919 7,630 77 H 1 1697 6083 5819 0 1,289 2,922 21,546 490 TABLE 10 Results for EPA Heat Input Sorting UTL Method for 5 Years Comparing All Hourly Emissions Post-Change - Oultilers Removed Plant Unit UTL NSR NOx Before NOx Max NOx After Number False Positive Hrs UTL NSR SO2 Before SO2 Max SO2 After Number False Positive Hrs E 1 4282 4542 32 946 1,048 277 A 2 3111 3230 10 10,132 10,717 12 C 3 653 715 31 3,073 3,114 12 B 5 5449 5084 0 10,685 10,082 0 F 1 4181 4658 120 6,260 6,380 10 G 1 5063 4835 0 4,194 4,370 32 D 4 3061 2948 0 2,873 2,630 0 H 1 5136 3473 0 2,748 3,754 298 2024-EPA-05254 32 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00032 SC_EVERSPLIT0001042 TABLE 11 Results for EPA NOx/SO2 Rate Sorting UTL Method for 5 Years Comparing All Hourly Emissions Post-Change - Unfiltered Plant Unit NOx UTL NSR Mean NOx NOx Before Max NOx After Number False Positive Hrs Mean SO2 SO2 UTL NSR SO2 Before Max SO2 After Number False Positive Hrs E 1 3784 4219 4838 64 839 1,250 3,616 46 A 2 2777 3162 3162 2 8,925 9,684 10,717 33 C 3 562 669 787 26 2,394 2,888 3,114 51 B 5 4575 4841 5084 4 8,780 9,611 11,841 6 F 1 3660 4127 4907 172 4,902 5,542 7,154 295 G 1 4030 4831 7712 125 3,437 4,068 7,719 102 D 4 2697 2950 3080 10 1,954 2,595 7,630 364 H 1 2640 3512 5819 811 2,867 9,969 21,546 101 TABLE 12 Results for EPA NOx/SO2 Rate Sorting UTL Method for 5 Years Comparing All Hourly Emissions Post-Change - Outliers Removed Plant Unit NOx Mean NOx UTL NSR NOx Before Max NOx After Number False Positive Hrs Mean SO2 SO2 UTL NSR SO2 Before Max SO2 After Number False Positive Hrs E 1 3531 4282 4542 32 712 950 1,048 260 A 2 2587 3115 3162 0 8,216 10,132 10,717 12 C 3 541 653 715 31 2,178 3,073 3,114 12 B 5 4195 5449 5084 0 8,120 10,685 10,082 0 F 1 3451 4302 4658 50 4,518 5,980 6,380 83 G 1 3369 5014 4835 0 3,013 3,934 4,370 149 D 4 2652 3061 2948 0 1,571 2,379 2,630 546 H 1 2408 3083 3473 232 1,285 3,078 3,754 145 33 2024-EPA-05254 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00033 SC_EVERSPLIT0001043 TABLE 13 Results for EPA NOx/SO2 Rate Sorting UTL Method for 5 Years Comparing All Hourly Emissions Post-Change - Unfiltered 1% Population @ 99.98 CI Plant NOx SO2 Unit UTL NSR Max NOx Number UTL NSR Max SO2 Number Mean NOx NOx Before After False Mean SO2 SO2 Before After False Positive Hrs Positive Hrs E 1 4072 5030 4838 0 1,040 2,627 3,616 2 A 2 3020 3332 5749 37 9,423 10,239 10,717 10 C 3 637 754 787 3 2,753 3,353 3,114 0 B 5 4745 5065 5084 1 9,282 11,450 11,841 2 F 1 3980 4478 4907 17 5,331 6,215 7,154 26 G 1 4573 5304 7712 24 3,837 5,356 7,719 9 D 4 2866 3110 3080 0 2,349 4,370 7,630 38 H 1 3270 4725 5819 89 8,068 27,684 21,546 0 TABLE 14 Results for EPA NOx/SO2 Rate Sorting UTL Method for 5 Years Comparing All Hourly Emissions Post-Change - Filtered 1% Population @ 99.98 CI Plant NOx SO2 Unit Mean NOx UTL NSR Max NOx Number False Mean SO2 UTL NSR Max SO2 Number False NOx Before After Positive Hrs SO2 Before After Positive Hrs E 1 3998 4418 4542 22 940 1,076 1,048 0 A 2 3020 3332 3230 0 9,423 10,239 10,717 10 C 3 637 754 715 0 2,753 3,353 3,114 0 B 5 4745 5065 5084 1 9,249 10,898 10,082 0 F 1 3980 4478 4658 13 5,331 6,215 6,380 15 G 1 4503 5064 4835 0 3,774 4,284 4,370 10 D 4 2866 3110 2948 0 2,194 2,566 2,626 25 H 1 3082 3573 3473 0 3,271 4,874 3,754 0 34 2024-EPA-05254 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00034 SC_EVERSPLIT0001044 Table 15 Comparison of Proposed UTL and Revised UTL Methods EPA Proposed Statistics (99.9 and 90.0) Plant Unit NOx False Positives SO2 False Positives EPA Method 168 Hr Method EPA Method 168 Hr Method E 3 35 0 12 0 A 3 41 C 2 47 0 425 15 0 12 0 B 5 0 0 2 0 F 1 44 0 145 0 G 4 3 0 77 2 D 1 124 0 21 0 H 1 41 I 8 87 1 490 3 8 315 0 7000 6000 5000 _i H X 2 4000 I' (o0p 1; 3000 c = 0 ce 2000 1000 0 1/1/02 2024-EPA-05254 Figure 5 Unit El 168 Hour NOx UTL Results Max UTL = 6007 1/1/03 1/1/04 1/1/05 35 1/1/06 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00035 SC_EVERSPLIT0001045 1800 1600 1400 1200 c8n 1000 ro ;:n 800 c K 600 400 200 0 1/1/02 Figure 6 Unit El 168 Hour SO2 UTL Results Max UTL = 1658 1/1/03 1/1/04 1/1/05 1/1/06 2024-EPA-05254 36 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00036 SC_EVERSPLIT0001046 APPENDIX A Normality Tests 2024-EPA-05254 1 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00037 SC_EVERSPLIT0001047 Unit C3 NOx Normality Test Std Deviation Sum Weights 495 Uncorrected Coeff Variation 1338002.65 2.33917338 Shapiro-Wilk Tests for Normality p Value W 0.965804 Pr < W Kolmogoro - 0.101974 Cramer-von Mises W-Sq 1.174798 Anderson- ar ing 6.637608 2024-EPA-05254 2 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00038 SC_EVERSPLIT0001048 Unit C3 SO2 Normality Test oments 495 urn Weights 779.235152 Sum Observations Std Deviation 161.04918 ariance Skewness -1.5592683 495 385721.4 25936.8385 5.39238441 Uncorrected SS Coeff Variation 313380472 20.667597 12812798.2 7.23862235 Test Shapiro-Wilk Statistic p Value 0.87108 Pr < W <0.0001 Kolmogorov-Smirnov Cramer-Non Mises Anderson-Darling 0.132511 Pr> W-Sq P 2.219216 r > W-Sq 244,d-1, A-S 14.14721 <0.0100 <0.0050 <0.0050 2024-EPA-05254 3 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00039 SC_EVERSPLIT0001049 Unit El NOx Normality Test Mean Moments 809 um Wei 1774.19197 um Observations 809 1435321.3 Std Deviatio Skewness 484.345308 Variance fi 0.54147479 Kurtosis 234590.378 0.60875723 Uncorrected SS 2736084555 189549025 Coeff Variation 27.2994871 d Error Mean 17.0286742 Shapiro-Wilk Cramer-von Mises Anderson-Darling Tests for Normality Statistic 0.978899 p Value < W <0.0001 0.088927 <0.0100 W-Sq 1.009841 q <0.0050 A-Sq 4.954598 r > A-Sq <0.0050 2024-EPA-05254 4 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00040 SC_EVERSPLIT0001050 Unit El SO2 Normality Test N Mean oments 795 um Weights 471.098365 um Observations Std Deviation 209.983818 Skewness, 0.56555629 Uncorrected SS 211447271 Coeff Variation 44.5732428 795 374523.2 44093.2038 0.498716 35010003.8 7.44735856 Test Shapiro-Wilk Tests for No Statistic p Value W 0.974141 Pr< W <0.0001 Kolmogorov-Smirnov 0.0669 Pr > D <0.0100 0.639596 Pr > W-Sq <0.0050 4.210138 Pr > A-Sq <0.0050 2024-EPA-05254 5 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00041 SC_EVERSPLIT0001051 Unit H1 NOx Normality Test Mean oments 783 Sum Weights 440.915014 Sum Observations 345236.456 Std Deviation 271.622058 73778.5421 Skewness 1.89939919 Kurtosis 3.42654247 Uncorrected SS 209914757 rrec ed SS 57694819.9 Coeff Varia o 61.6041751 9.70698042 Test Shapiro-Wilk Tests for Normality p Value Pr < W <0.0001 ogo ov-Smirnov Cramer-von Mises W-Sq Pr > W-Sq <0.0050 2024-EPA-05254 6 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00042 SC_EVERSPLIT0001052 Unit H1 SO2 Normality Test N Mean oments 783 Sum Weights 688.703959 um Observations 783 539255.2 Sid DeviaitO 1262.26472 1593312.22 Skewness 4.69787168 urtosi 25.646491 Uncorrected SS - 1617357344 Corrected SS 1245970153 Coeff VariatiO 183.281176 d Error Meali 45.1096609 Shapiro-Wilk 0.437925 ' r < W 1 <0.0001 Kolmogorov-Smirno, 0.31442 <0.0100 Cramer-von Mises W-S 28.77426 <0.0050 Anderson-Darling -S i 144.3258 <0.0050 2024-EPA-05254 7 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00043 SC_EVERSPLIT0001053 Unit A2 NOx Normality Test N .. Mean -- .. .... Moments 775 Sum Weights .... 812.832318 Sum Observation , - Std Deviation 135.638381 Varianc .... Skewness .--. ga I , emur' Uncorrected SS ... .E- , W... . -0.3076543 Kurtosis 526279567 ,Corrested SS 775 629945.047 18397.7703 1.13710541 r* 14239874.2 Coeff Variation 16.6871294 4.87227446 Test Shapiro-Wilk Tests for Normality Statistic p Value 0.976013 Pr < W <0.0001 Kol o orov-Sm rnov Cramer-Non Mises <0.0050 Anderson-Darling 4.820102 <0.0050 2024-EPA-05254 8 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00044 SC_EVERSPLIT0001054 Unit A2 SO2 Normality Test N Mean Moments 775 um Weights 2365.60645 um Observations 775 1833345 Std Devia io 258.372181 Variance 66756.1841 Skewnes -0.7004401 Uncorrected SS a 4388642047 orrected SS 2.61806914 51669286.5 Coeff Variation 10.9220273 Std Error Mean 9.28100273 Shapiro-Wilk Kolmog C -v n Mises Anderson-Darling Tests for Normality Statistic 0.962366 Pr < W 0.057033 Pr > D 0.50806 A-Sq I 3.890565 <0.0001 <0.0100 <0.0050 <0.0050 2024-EPA-05254 9 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00045 SC_EVERSPLIT0001055 Unit B5 NOx Normality Test Std Deviation Skewness 1328.13423 Observations 321.663135 ariance -0.5181826 Uncorrected SS 41 1435933052 Corrected SS Coeff Variation 24.2191737 kolmogor Cramer-von Mises 0.065287 Pr > D W-Sq 1.076395 A-S 8.014388 <0.0100 <0.0050 <0.0050 2024-EPA-05254 10 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00046 SC_EVERSPLIT0001056 Unit B5 SO2 Normality Test N Mean Std Deviation Moments 769 Sum Weights - -, 3462.85748 hum Observations ' i . ,.. . 576.735806 769 2662937.4 332624.19 Skewness -0.4337334 -0.5043948 U c rected, 9476828065 Coeff Variation k 16.6549103 255455378 20.7976251 Test Shapiro-Wilk Tests for Normality Statistic p Value W 0.974875 Pr < W <0.0001 Kolmogorov-Smirnov Cramer-von Mises 0.057843 Pr > D Ate-,-AC-,0 <0.0100 W-Sq 0.810927 <0.0050 2024-EPA-05254 11 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00047 SC_EVERSPLIT0001057 Unit 18 NOx Normality Test Mean Moments 857 um Weights 196.123721 m Observations Std Deviatio 52.0784426 Skewness 1.47640305 Uncorrected SS 35285701.1 Corrected SS Coeff Variation 26.5538723 td Error ea 857 168078.029 2712.16418 10.9861023 2321612.54 1.77896571 Shapiro-Wilk Tests for N y Statistic p Value 0.861744 Pr < W <0.0001 Kolmogorov-Smirnov 0.107319 Pr> D <0.0100 a e -vori ses 2.631251 <0.0050 Anderson-Darling 17.32757 > A-Sq <0.0050 2024-EPA-05254 12 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00048 SC_EVERSPLIT0001058 Unit 18 SO2 Normality Test N Mean Moments 857 Sum Weights 722.87853 Sum Observations Sf evr o ,, Skewness 205.502307 -0.1698377 857 619506.9 42231.1983 -0.5152756 Uncorrected SS 483978143 Corrected SS Coeff Variation 28.4283318 td Error Mean 36149905.7 7.01982509 Tes Shapiro-Wilk Tests for Normali W 0.982579 <0.0001 Kolmogorov-Smirnov 0.065399 <0.0100 Cramer-von Mises 1.122564 <0.0050 Anderson-Darling 6.182175 Pr > A-Sq <0.0050 2024-EPA-05254 13 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00049 SC_EVERSPLIT0001059 Unit D4 NOx Normality Test Mean oments um Weights 1566.91496 um Observations 1255098.88 Std Dem o 313.687393 Variance 98399.7804 Skewness -2.1447492 Kurtosi 4.9275006 2045353040 Corrected SS 78719824.3 Coeff Varia ion 20.0194268 11.083599 Test Shapiro-Wilk Tests for Normality Statistic p Value W 0.765549 Pr < W <0.0001 Kolmogorov-Smirnov 0.183144 Pr> D <0.0100 Cramer-von W-Sq 9.566471 Pr > W-Sq <0.0050 Anderson-Darling 54.96793 <0.0050 2024-EPA-05254 14 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00050 SC_EVERSPLIT0001060 Unit D4 SO2 Normality Test ean Std Deviation Skewness 793 r, urn Weights 1851.8454 hum Observa -0.3349436 Uncorrected SS 3145712332 -1.3199864 2024-EPA-05254 15 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00051 SC_EVERSPLIT0001061 Unit Fl NOx Normality Test Shapiro-Wilk Tests for Norma ay Statistic p Value W 0.96053 Pr <W <0.0001 Kolmo orov-Smirnov 0.061201 <0.0100 Cramer-Non Mises Anderson-Darling 611N1111111111111111111111111111111111111L 2024-EPA-05254 16 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00052 SC_EVERSPLIT0001062 Unit F1 SO2 Normality Test Mean oments 758 Sum Weights 2039.48496 m Observations Mtn 758 1545929.6 Std Deviation 819.166081 Variance --- 671033.068 Skewness -0.3110148 -0.6091456 Uncorrected Coeff Variatio 3660872201 Corrected SS 40.1653406 Std Error Mean 507972032 3 29.7534519 Test Shapiro-Wilk Tests for Normality Statistic 0.977796 p Value <0.0001 Kolmogoroy- 0.047927 Pr > D <0.0100 er-vo 0.521363 Pr > W-Sq <0.0050 Anderson-Darling A-Sq 4.103342 0.0050 2024-EPA-05254 17 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00053 SC_EVERSPLIT0001063 Unit G1 NOx Normality Test Mean Moments 786 Sum Weigh 1355.80702 um Observation 786 1065664.32 Std Devia a Skewness 396.797275 , 0.08074266 Ku 9 157448.078 0.92305045 Uncorrected SS 1568431911 Corrected SS 123596741 Coeff Variation 29.2665009 Std Error Mea 14.1532901 Shapiro-Wilk Kolmogorov-Smirnov 4 4*. Cramer-von Mises 0.037563 W-Sq 0.207288 Anderson-Darling A-Sqa 1.668778 <0.0100 <0.0050 <0.0050 2024-EPA-05254 18 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00054 SC_EVERSPLIT0001064 Unit G1 SO2 Normality Test Uncorrected Shapiro-Wilk Tests for Normality Statistic p Value W 0.984928 Pr < W <0.0001 Kolmogo ov-Smirnov <0.0100 Cramer-von Mises <0.0050 Anderson-Darling <0.0050 2024-EPA-05254 19 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00055 SC_EVERSPLIT0001065 APPENDIX B Five Year EPA UTL Analysis Data Plots 2024-EPA-05254 1 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00056 SC_EVERSPLIT0001066 900 800 700 .c 600 tn _a in 500 O E 400 w 300 200 100 0 1/1/2002 Unit C Hourly NOx 2002 to 2006 Unfiltered Data 1/1/2003 1/1/2004 1/1/2005 1/1/2006 1/1/2007 SO2 Emissions (Ibs/hr) Unit C Hourly SO2 2002 to 2006 Unfiltered Data 5,000 4,500 4,000 0 0 Highest HI Year 3,500 3,000 SO2 UTL 2,500 2,000 . . ,:7 1,500 1,000 I or 500 030t 0 1/1/2002 eb O oo 8 gjoo is o o g c SEp o O:8 0 god 0 1/1/2003 o e O 1/1/2004 o o: % s8. O 0 0 : 0CP% 2, 0 Cb O o g o "D O 4 200 o . 6, 0 00 O 09, 0000 0 o oo 1/1/2005 1/1/2006 1/1/2007 2024-EPA-05254 2 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00057 SC_EVERSPLIT0001067 NOx Emissions (Ibs/hr) 7,000 Unit A Hourly NOx 2002 to 2006 Unfiltered Data 6,000 5,000 4,000 Highest Heat Input Year 3,000 2,000 1,000 0 1/1/2002 1/1/2003 1/1/2004 1/1/2005 1/1/2006 25,000 20,000 Highest Heat Input Year Unit A Hourly SO2 2002 to 2006 Unfiltered Data 15,000 10,000 SO, UTL 5,000 0 1/1/2002 1/1/2003 1/1/2005 1/1/2006 1/1/2007 1/1/2007 SO2 Emissions (Ibs/hr) 2024-EPA-05254 3 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00058 SC_EVERSPLIT0001068 NOx Emissions (Ibs/hr) 8,000 7,000 6,000 5,000 4,000 3,000 2,000 1,000 0 1/1/2002 Unit B Hourly NOx 2002 To 2006 Unfilterd Data Highest Heat Input Year 1/1/2003 1/1/2004 1/1/2005 1/1/2006 25,000 20,000 Unit B Hourly SO2 2002 To 2006 Unfiltered Data IHighest Heat Inpui Year a 15,000 0 W 10,000 O fA 5,000 SO2 UTL a 1/1/2002 1/1/2003 1/1/2004 1/1/2005 1/1/2006 1/1/2007 1/1/2007 2024-EPA-05254 4 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00059 SC_EVERSPLIT0001069 Unit H Hourly NOx 2002 To 2006 Unfiltered Data 1/1/2002 25,000 20,000 tn -2 15,000 tn C 0 7) ,4 E w 10,000 N O (i) 5,000 1/1/2003 1/1/2004 1/1/2005 1/1/2006 Unit H Hourly SO2 2002 To 2006 Unfiltered Data 1/1/2007 1/1/2003 1/1/2004 1/1/2005 1/1/2006 1/1/2007 2024-EPA-05254 5 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00060 SC_EVERSPLIT0001070 4,500 4,000 3,500 Unit D Hourly NOx 2002 To 2006 Unfiltered Data tn 2,500 E 2,000 O z 1,500 1,000 500 0 1/1/2002 1/1/2003 1/1/2004 1/1/2005 1/1/2006 9,000 8,000 - 7,000 - 2E- 6,000 tn 5,000 c 0 E4` 4,000 LI/ 3,000 2,000 1,000 I 1/1/2002 Unit D Hourly SO2 2002 To 2006 Unfiltered Data SO2 UTL Soil Rai* 1/1/2003 Si e 1/1/2004 1/1/2005 1/1/2006 1/1/2007 1/1/2007 2024-EPA-05254 6 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00061 SC_EVERSPLIT0001071 6,000 5,000 Unit F Hourly NOx 2002 To 2006 Unfilterd Data Highest Heat Input Year In 0 *L7) 3,000 E w 2,000 1,000 0 1/1/2002 1/1/2003 1/1/2004 1/1/2005 1/1/2006 1/1/2007 SO2 Emissions (lbs/hr) 10,000 9,000 8,000 7,000 6,000 5,000 4,000 3,000 2,000 1,000 0 1/1/2002 Unit F Hourly SO2 2002 To 2006 Unfiltered Data Highest Heat Input Year 1/1/2003 1/1/2004 1/1/2006 1/1/2007 2024-EPA-05254 7 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00062 SC_EVERSPLIT0001072 9,000 8,000 7,000 Unit G Hourly NOx 2002 To 2006 Unfiltered Data o Highest He: t Input Yea o in 5,000 O E 4,000 1 w x O z 3,000 e 2,000 a O i ,- - o o 801) 1,000 r O. O 8, o .8 O o 8 0 OO % O o 0 s o co 1/1/2002 1/1/2003 1/1/2004 8 o 2, 0 Iof 0 8 10 o 93 o O 1/1/2005 NOx UTL - lel O O o ep., o oos O 1/1/2006 1/1/2007 Unit G Hourly SO2 2002 To 2006 Unfiltered Data 9,000 8,000 - 7,000 - 2E- 6,000 A in 5,000 c O 4E` 4,000 -- cNi 3,000 2,000 O O 0 0 0 1,000 0 8 1/1/2002 I ,e 1/1/2003 O 1/1/2005 1/1/2006 2024-EPA-05254 8 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00063 SC_EVERSPLIT0001073 NOx Emissions, lb/Hr Unit I Hourly NOx 3/1/04 to 2006 1400 0 1200 1000 800 600 400 0 0 0 o$ co 0 03 8 NOx UTL 0 0 0 Q I Highest Heat Input Year o o 0 0 o8 5 00 200 0 1/1/2004 1/1/2005 1/2/2006 2000 1800 1600 1400 1200 1000 800 600 400 200 0 1/1/2004 Unit I Hourly SO2 2004 to 2006 1/1/2005 1/2/2006 SO2 Emissions, lb/Hr 2024-EPA-05254 9 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00064 SC_EVERSPLIT0001074 APPENDIX C Five Year Revised UTL Analysis Data 2024-EPA-05254 1 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00065 SC_EVERSPLIT0001075 Unit C 168 Hour NOx UTL Results Rolling 168 Hr NOx UTL 01/01/02 4500 4000 3500 - I 3000 I7 N O (i) 2500 I ccoo - 2000 co C = Z w 1500 1000 500 0 01/01/02 01/01/03 01/01/04 01/01/05 01/01/06 Unit C 168 Hour SO2 UTL Results Max UTL = 4101 01/01/03 01/01/04 01/01/05 01/01/06 2024-EPA-05254 2 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00066 SC_EVERSPLIT0001076 Unit A 168 Hour NOx UTL Results 5000 4500 Max UTL = 4532 4000 - 3500 1- 14 3000 Ox I 2500 CO CD r2 2000 O 1500 - 1000 - A vomtiorki) 500 0 1/1/02 1/1/03 1/1/04 1/1/05 1/1/06 Rolling 168 Hr SO2 UTL 14000 12000 10000 8000 6000 4000 2000 0 1/1/02 Unit A 168 Hour SO2 UTL Results Max UTL = 12201 1/1/03 1/1/04 1/1/05 1/1/06 2024-EPA-05254 3 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00067 SC_EVERSPLIT0001077 Rolling 168 Hr SO2 UTL 8000 7000 6000 -i I5000 x 0 Z I 4000 CO CD co c E 3000 O Ce 2000 1000 0 1/1/02 14000 12000 10000 8000 6000 4000 2000 1/1/02 Unit B 168 Hour NOx UTL Results 1/1/03 1/1/04 1/1/05 1/1/06 Unit B 168 Hour SO2 UTL Results 1/1/03 1/1/04 1/1/05 1/1/06 2024-EPA-05254 4 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00068 SC_EVERSPLIT0001078 Rolling 168 Hr NOx UTL 12000 10000 8000 6000 4000 2000 0 01/01/02 Unit H 168 Hour NOx UTL Results Max UTL = 10322 t iri !Itim00#644A 01/01/03 01/01/04 01/01/05 01/01/06 25000 20000 Unit H 168 Hour SO2 UTL Results Max UTL = 6881 15000 10000 .,lik-Ad L 5000 0 01/01/02 01/01/03 01/01/04 01/01/05 01/01/06 Rolling 168 Hr SO2 UTL 2024-EPA-05254 5 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00069 SC_EVERSPLIT0001079 4000 3500 3000 -i F 2500 x 0 z I 2000 CO Co 0) c E 1500 O Ce 1000 500 0 1/1/02 Unit D 168 Hour NOx UTL Results 1/1/03 1/1/04 1/1/05 1/1/06 7000 6000 5000 -i iD N O 4000 C/) I ccoo co 3000 c = Z W 2000 1000 0 1/1/02 Unit D 168 Hour SO2 UTL Results I 1/1/03 1/1/04 1/1/05 1/1/06 2024-EPA-05254 6 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00070 SC_EVERSPLIT0001080 Rolling 168 Hr SO2 UTL 7000 6000 - 5000 -i I7 X O 4000 Z I 00 CD -co 3000 c O cc 2000 1000 - 0 1/1/02 10000 9000 8000 7000 6000 5000 4000 3000 2000 1000 0 1/1/02 Unit F 168 Hour NOx UTL Results 1/1/03 1/1/04 1/1/05 1/1/06 Unit F 168 Hour SO2 UTL Results I 1/1/03 1/1/04 1/1/05 1 1/1/06 2024-EPA-05254 7 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00071 SC_EVERSPLIT0001081 8000 7000 6000 -i I- 5000 x 0 Z I 4000 CO CD 0) c E 3000 0 Ce 2000 1000 0 1/1/02 8000 7000 6000 -I ID 5000 N 0 (i) I 4000 CO CD 0) c = 3000 7), CL 2000 1000 0 1/1/02 Unit G 168 Hour NOx UTL Results 1/1/03 1/1/04 1/1/05 1/1/06 Unit G 168 Hour SO2 UTL Results Max UTL = 6818 1/1/03 1/1/04 1/1/05 ) 1/1/06 2024-EPA-05254 8 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00072 SC_EVERSPLIT0001082 700 600 500 -I I7 X z 400 I cw o ; 300 c 0 r[ 200 100 0 1/1/02 2000 1800 1600 -I 1400 I7 Csi 1200 O U) I 1000 co co F 800 = 0 W 600 400 200 0 1/1/02 Unit 1168 Hour NOx UTL Results 1 1/1/03 1/1/04 1/1/05 1/1/06 Unit 1168 Hour SO2 UTL Results ti,i,,t14A 1/1/03 1/1/04 1/1/05 1/1/06 2024-EPA-05254 9 Sierra Club FOIA 2024-EPA-05254 ED_017426_00001808-00073 SC_EVERSPLIT0001083