                 UNITED STATES ENVIRONMENTAL PROTECTION AGENCY
                                  REGION III
                               1650 Arch Street
                       Philadelphia, Pennsylvania  19103


DATE:	September 22, 2011

SUBJECT:	Technical Support Document for Determinations of Attainment of the 1997 Fine Particle Standard for the Metropolitan Washington, District of Columbia-Maryland-Virginia (DC-MD-VA) and Martinsburg-Hagerstown, West Virginia-Maryland (WV-MD) Areas
			/s/
FROM:	Irene Shandruk, Physical Scientist 
            Office of Air Program Planning 

TO:	File
			/s/	
THRU: 	Cristina Fernandez, Associate Director 
            Office of Air Program Planning


A.  BACKGROUND 

In 1997, EPA revised the health-based National Ambient Air Quality Standard (NAAQS) for fine particulate matter (PM2.5) , setting it at 15.0 micrograms per cubic meter (ug/m[3]) based on a 3-year average of annual mean PM2.5 concentrations, and a 24-hour standard of 65 ug/m[3] based on a 3-year average of the 98th percentile of 24-hour concentrations.  EPA established the standards based on significant evidence and numerous health studies demonstrating that serious health effects are associated with exposures to particulate matter.  
      
On January 5, 2005 (70 FR 944), EPA finalized its air quality designations and classifications across the country with respect to the 1997 PM2.5 standard.  These designations became effective on April 5, 2005.  The Metropolitan Washington, DC-MD-VA and the Martinsburg-Hagerstown, WV-MD nonattainment areas were designated nonattainment for the 1997 annual PM2.5 NAAQS.  See, 40 CFR part 81.309 (the District), 40 CFR 81.321 (Maryland), 40 CFR 81.347 (Virginia), and 40 CFR 81.349 (West Virginia).  

The Metropolitan Washington 1997 annual PM2.5  nonattainment area consists of the District of Columbia (the District), a Northern Virginia portion (Arlington, Fairfax, Loudoun, and Prince William Counties and the cities of Alexandria, Falls Church, Fairfax, Manassas, and Manassas Park), and Charles, Frederick, Montgomery, and Prince George's Counties in Maryland.  The Martinsburg-Hagerstown 1997 annual PM2.5 nonattainment area consists of Washington County in Maryland and Berkley County in West Virginia.  
      
On October 17, 2006 (71 FR 61144), EPA retained the 1997 annual PM2.5 NAAQS at 15 ug/m[3] based on a 3-year average of annual mean PM2.5 concentrations, and promulgated a 24-hour standard of 35 ug/m[3] based on a 3-year average of the 98[th] percentile of 24-hour concentrations (the "2006 24-hour standard").  On November 13, 2009, EPA designated the Martinsburg-Hagerstown, WV-MD and Metropolitan Washington, DC-MD-VA areas as attainment for the 2006 24-hour standard (74 FR 58688).  In that action, EPA also clarified the designations for the NAAQS promulgated in 1997, stating that these geographical Areas were designated as nonattainment for the annual standard, but attainment for the 1997 24-hour standard (40 CFR part 81.309 for the District, 40 CFR part 81.321 for Maryland, 40 CFR part 81.347 for Virginia, and 40 CFR part 81.349 for West Virginia). Today's action, however, does not address attainment designations of either the 1997 or the 2006 24-hour standard.

In response to legal challenges of the annual standard promulgated in 2006, the U.S. Court of Appeals for the District of Columbia Circuit (D.C. Circuit) remanded this standard to EPA for further consideration.  See American Farm Bureau Federation and National Pork Producers Council, et al. v. EPA, 559 F.3d 512 (D.C. Cir. 2009).  However, given that the 1997 and 2006 annual standards are essentially identical, attainment of the 1997 annual standard would also indicate attainment of the remanded 2006 annual standard.

EPA previously made clean data determinations related to the 1997 annual PM2.5 NAAQS for each of these Areas pursuant to 40 CFR 51.1004(c).  Determinations were made for the Metropolitan Washington Area on January 12, 2009 (74 FR 1146) and for the Martinsburg-Hagerstown Area on November 20, 2009 (74 FR 60199).  These clean data determinations remain in effect.

Under CAA section 179(c), EPA is required to make a determination that a nonattainment area has attained by its attainment date, and publish that determination in the Federal Register.   The determination of attainment is not equivalent to a redesignation, and the states must still meet the statutory requirements for redesignation in order for the Areas to be redesignated to attainment.  Complete, quality-assured, and certified PM2.5 air quality monitoring data recorded in the EPA Air Quality System (AQS) database for 2007 through 2009, show that the Metropolitan Washington, DC-MD-VA and the Martinsburg-Hagerstown, WV-MD nonattainment areas attained the 1997 annual PM2.5 NAAQS by their applicable attainment date of April 5, 2010.
      

B.  EPA EVALUATION

EPA has reviewed the ambient air monitoring data for PM2.5, consistent with the requirements contained in 40 CFR part 50 and recorded in the data in the EPA AQS database for the Metropolitan Washington, DC-MD-VA and the Martinsburg-Hagerstown, WV-MD nonattainment areas for the monitoring period from 2007 through 2009.  On the basis of that review, EPA has concluded that the area attained the 1997 annual PM2.5 NAAQS based on data for the 2007-2009 monitoring period. 

Under EPA regulations at 40 CFR 50.7, the annual primary and secondary PM2.5 standards are met when the annual arithmetic mean concentrations, as determined in accordance with 40 CFR part 50, appendix N, is less than or equal to 15.0 ug/m[3], at all relevant monitoring sites.  The values calculated in accordance with 40 CFR part 50, appendix N, are referred to as design values, and these values are used to determine if an area is attaining the PM2.5 NAAQS.  According to the PM2.5 implementation rule, the attainment date for these Areas is April 5, 2010 and the monitoring data from 2007 through 2009 is used to determine if the Areas attained by April 5, 2010.  

Data Completeness

A complete year of air quality data comprises four calendar quarters, with each quarter containing data from at least 75% capture of the scheduled sampling days.  Data that does not meet the 75% capture has been analyzed by EPA with substitution of conservative values to determine that some sites not meeting capture rates are still sufficient to show attainment.  The Metropolitan Washington Area contained one such site which had incomplete data:  Fairfax County Virginia Annandale Site.

As discussed in greater detail below, the Annandale monitor (AQS ID 510591005) was the only monitor in the Metropolitan Washington Area without a valid attaining design value for 2007-2009 due to incomplete data.  This monitor has never had a violating design value (DV) but was incomplete for 2007-2009 because it had to be removed temporarily from the end of April 2008 to the middle of June 2008.  This temporary removal was due to the roof on which the monitor was stationed needing replacement.
	
EPA has reviewed the ambient air monitoring data for PM2.5, consistent with the requirements contained in 40 CFR part 50 and recorded in the EPA Air Quality System database for the Metropolitan Washington Area from 2007 through the present time. Currently all monitors are measuring concentrations averaging well below the annual standard of 15.0 ug/m3. The highest DV for the most recent 3 years, 2007-2009, for the Metropolitan Washington Area is 12.4 ug/m3, based on monitoring data collected in Maryland at the Bladensburg monitor. All 3-year averages of the annual mean PM2.5 concentrations are below the NAAQS from 2007-2009.

In order to determine whether the NAAQS has been attained for the Annandale monitor, AQS ID 510591005, EPA has applied a recently developed procedure to address the missing data, and determine valid design values for the monitor for the annual standard for 2007-2009. EPA's procedure is described in the following section.

Addressing Missing Data

An objective statistical method was developed to determine if monitors that lack complete data would have met the NAAQS in subsequent time periods.  The Annandale monitor in the Metropolitan Washington nonattainment area lacked complete data due to a siting issue.

The general concept behind the statistical method to determine whether an incomplete monitor (target) would have met the NAAQS is to establish a linear regression relationship between the target monitor and another monitor in the nonattainment area (candidate).  The linear regression would then be used to fill in the missing data for the target monitor.  The results are checked with an additional statistical technique, known as bootstrapping, to verify if the conclusion of attainment is correct.  The Metropolitan Washington nonattainment area has 16 monitoring locations which provide an ample number of monitors to compare to the target monitor.

The monitor of interest in this analysis is the Annandale monitor. This site was temporarily removed from April 2008 to June 2008 due to a roof replacement causing an incomplete design value of 11.3 μg/m3 for 2007-2009. The goal of this methodology is to appropriately fill in the missing data from the second quarter of 2008 to judge whether this location would likely have monitored attainment of the NAAQS had the monitor had complete data. The Metroplitan Washington Area has a monitoring location with complete data for 2007-2009, which is similar enough to warrant comparison to the Annandale monitor.

Concept for Determining Design Values

The concept behind the method for determining a design value for a monitor with incomplete data is to establish a linear regression relationship between a target site with incomplete data and another site in the Washington, D.C. PM2.5 nonattainment area which has more complete data in the period in which the target site is data deficient.  The linear regression relationship would be based on time periods in which both the target and candidate monitor was operating.  The linear regression equation developed from the relationship between the target and candidate monitor was used to fill in missing data for the target site in any quarter in the most recent three-year period that did not have 75 percent completeness based on actual observations, so that the normal data completeness requirement of 75 percent of data in each quarter of the three years is met.  After the missing data for the sites was filled in, the results were verified through an additional statistical test.  The additional statistical test used, referred to as bootstrapping, examines the differences between the predicted fill in concentrations and the actual observed concentrations to determine if the design values for the monitors with incomplete data meet the annual NAAQS for PM2.5. 

EPA established criteria to assure that the DV for the Annandale monitor in the Metropolitan Washington Area, as estimated by the method, was robust enough to allow consideration of the results in a weight of evidence decision.  EPA established criteria both for the operating period of the comparison monitor and for the number of days of samples that the comparison monitor and the Annandale monitor have in common.  EPA has not established bright lines or specific requirements for determining DVs for monitors with less than complete data.

The candidate monitor would have operated during the 3-year period (2007-2009) in the Metropolitan Washington PM2.5  nonattainment area in which the target monitor has incomplete data in order to establish a significant relationship between the monitors.  The candidate monitor does not necessarily have to have complete data in all 12 quarters of the 3-year period, but the candidate monitor would have enough observations to allow it to be used to fill in enough missing days for the target monitor so that the target monitor will end up with at least 75 percent completeness in all 12 quarters.  

Pairs of sample data for the target monitor and candidate monitor from the most recent 20-quarter period of operation of the target monitor was used.  The "period of operation" of the target monitor was based on the date of the last reported sample from the monitor.  The candidate monitor had at least 100 historical sample pairs in common with target monitor during this period.  When there were more than 100 pairs in the 20-quarter period all available pairs were used.  In addition to the 100 pairs over the 20-quarters, candidate monitors had at least 20 paired days of data in common with the target monitor for each calendar quarter (adding together the number of days in a given calendar quarter [Q1, Q2, Q3, or Q4] across the 20-quarter period).

Several candidate sites met the eligibility criteria of 100 pairs over the 20 quarters, and at least 20 paired days of data in common with the target monitors for each calendar quarter. 

Method for Determining Design Values

Step 1:  Organize data and perform linear regression(s) between target monitor and candidate monitors.

All the available paired data within the period limit described above was used to develop a linear regression to predict target concentrations from candidate concentrations.  A regression was performed for each possible pairing that met the criteria utilized for 100 data points and 20 data points per calendar quarter (across all years) in the most recent 20 quarters for each candidate monitor in the Metropolitan Washington PM2.5 nonattainment area. 

Step 2:  Select Candidate site

The Metropolitan Washington PM2.5 nonattainment area has a large number of monitors which could have been used as candidate sites.  Only one candidate monitor (Mclean, AQS ID 51059500) was selected for the Annandale monitor.  The follow steps were used to select the candidate monitor.  

      1) Candidate monitors were placed into order of highest correlation (highest r-squared in the linear regression) with the Target.  

      2) Using the regression equation relating concentrations at the Target monitor and the Candidate monitor, estimates of the 24-hour average concentrations were made at the Target monitor for all missing scheduled days in the initially incomplete quarter, in the most recent three-year period at the Target. Estimates were only made for days on which a concentration is available at the Candidate monitor.  Estimates were not made for any days in the quarter for which the Target monitor had 75 percent or more complete data initially.  The estimates were combined with any reported data at the Target monitor in the recent 3-year period and the data capture rates were computed for the originally incomplete quarter.  The computation of data capture rates required an assumption that the required sampling frequency for the Target monitor was recorded.   If the Candidate monitor had a lower sampling frequency than the Target monitor, some missing scheduled days for the Target were not substituted for the regression-based estimates.  When the Candidate monitor had a lower sampling frequency than the Target monitor, the Target monitor sampling schedule was assumed to be a combination of the Target and Candidate monitors schedules, weighted by the number of originally present and originally missing scheduled days at the Target monitor.  

      3) If after performing the process in step 2, the results produce all 12 quarters at the Target monitor having at least 75 percent of scheduled values, the Candidate monitor was selected as the Candidate for the remaining steps.  If 75 percent of the scheduled values for the Target monitor were not obtained, step 2 was repeated for the next best correlated site.  This process was repeated until the Target monitor had 75 percent of its scheduled values.

Step 3:  Design Value Computed

The three-year design value was calculated using the actual observations of the Target monitor and the regression-based estimates of daily concentrations.  The design value for the Annandale monitor in the Metropolitan Washington PM2.5 nonattainment area was below the NAAQS.

Step 4:  Statistical confidence of a design value below the NAAQS checked

The results of Step 3 showed the design value for the Annandale monitor in the Metropolitan Washington PM2.5 nonattainment area to be below the NAAQS.  In order to show that there is at most a 10 percent probability that the actual design value (had the Annandale monitor operated more days) was above the NAAQS, further analysis was performed as follows:

      (a)  Residuals were calculated from the regression.
      (b)  A bootstrap analysis was performed, repeating the regression-based substitution in Step 2 but adding to each estimated 24-hour value that substitutes for a missing scheduled monitored value a plus or minus residual that has been randomly drawn (with replacement) from the pool of regression residuals.  This step randomly applies real residuals from the linear regression to the imputed current-period Target values.
      (c)  Repeat (b) for a total of 1000 runs.  
      (d)  For each of the 1000 bootstrap trials, the 3-year design value was calculated.  This will provide a pseudo confidence interval for the partially imputed current-period design value for the Target.
      (e)  Since none of the 1000 runs resulted in a design value above the NAAQS, the Annandale was considered to have met the NAAQS.   
 
Determining the design value for Annandale monitor AQS ID: 510591005

Steps 1 and 2 were performed with Annandale (AQS ID: 510591005) as the Target monitor.  The results of the regressions can be found in Figure 2.  On Figure 2, the Target monitor is listed as Site A.  The Candidate site that was used is listed as Site B.  The monitor Mclean, (AQS ID: 510595001), had the highest correlation with Annandale and met the operating period and paired data eligibility criteria for being a Candidate site.  The results of the regression between the Annandale and Mclean monitors is shown in Figure 3.  Estimates for the missing data for second quarter 2008, were made.  As a result of filling in the missing data, the Annandale monitor now had 30 days of data for second quarter 2008.  A design value of 11.4 g/m[3] was calculated for the Annandale monitor when the missing data was filled in Figure 4.  The statistical confidence of the design value was checked using the method outlined in Step 4.  The results of the 1000 bootstrap trials provided the following results:  21 DVs of 11.2 ug/m[3], 716 DVs of 11.3 ug/m[3], 260 DVs of 11.4 ug/m[3], and 3 DVs of 11.5 ug/m[3].  None of the 1000 bootstrapping trials resulted in a violating DV, which provides statistical confidence that the Annandale monitor is attaining the annual PM2.5 NAAQS.





	






After accounting for the data completeness issue in the Metropolitan Washington Area, Table 1 shows the annual PM2.5 design values for each monitor in the Metropolitan Washington, DC-MD-VA and the Martinsburg-Hagerstown, WV-MD areas for the years 2007-2009.  All 2007-2009 design values are below 15.0 ug/m[3].	











Table 1.  Design Values for the Metropolitan Washington, DC-MD-VA and Martinsburg-Hagerstown, WV-MD Areas*
                                     State
                                    County
                                  Monitor ID
                               2007 Annual Mean
                               2008 Annual Mean
                               2009 Annual Mean
                            Certified Design Value
                             2007-2009 (ug/m[3])
                       Metropolitan Washington, DC-VA-MD
DC
District of Columbia
110010041
                                                                           13.6
                                                                           12.0
                                                                           10.5
                                     12.0
 
District of Columbia
110010042
                                                                           13.7
                                                                           12.3
                                                                           10.1
                                     12.1
 
District of Columbia
110010043
                                                                           13.0
                                                                           11.6
                                                                           10.2
                                     11.6
VA
Alexandria
No monitor
                                       
                                       
                                       
                                       
 
Arlington
510130020
                                                                           13.8
                                                                           12.0
                                                                           10.1
                                     11.9
 
Fairfax
510590030
                                                                           12.5
                                                                           11.1
                                                                            9.8
                                     11.1
 
Farifax County
510591005
                                                                           13.3
                                                                           11.2
                                                                            9.5
                                     11.3
 
Fairfax
510595001
                                                                           13.5
                                                                           11.8
                                                                            9.7
                                     11.7
 
Falls Church
No monitor
 
 
 
                                       
 
Loudon
511071005
                                                                           12.8
                                                                           11.5
                                                                            9.2
                                     11.2
 
Manassas
No monitor
                                       
                                       
                                       
                                       
 
Manassas Park
No monitor
                                       
                                       
                                       
                                       
MD
Charles
No monitor
 
 
 
 
 
Frederick
No monitor
 
 
 
 
 
Montgomery
240313001
                                                                           11.7
                                                                           10.8
                                                                            9.4
                                     10.7
 
Prince George
240330025
                                                                           14.1
                                                                           12.4
                                                                           10.7
                                     12.4
 
Prince George
240330030
                                                                           11.8
                                                                           10.9
                                                                            8.7
                                     10.5
 
Prince George
240338003
                                                                           12.4
                                                                           11.2
                                                                            8.8
                                     10.8
                         Martinsburg-Hagerstown, WV-MD
WV
Berkley 
240430009
                                                                           12.9
                                                                           11.8
                                                                            9.7
                                     11.5
MD
Washington 
540030003
                                                                           15.6
                                                                           14.2
                                                                           12.1
                                     14.0
* The data presented in Table 1 are available at http://www.epa.gov/air/airtrends/values.html.

C.  CONCLUSIONS AND RECOMMENDED AGENCY ACTION

Based on data for the 2007-2009 monitoring period, the Metropolitan Washington, DC-MD-VA and the Martinsburg-Hagerstown, WV-MD areas have attained the 1997 annual PM2.5 NAAQS by the attainment date of April 5, 2010. 


