SEPA
Environmental Protection
Agency
Comparison of Integrated Filter and Semi-
Continuous Measurements of PM2 5 Nitrate,
Sulfate, and Carbon Aerosols in the Speciation
Trends Network (STN)
-------
EPA 454/R-05-004
December 2005
Comparison of Integrated Filter and Semi-continuous Measurements of PM2.5
Nitrate, Sulfate, and Carbon Aerosols in the Speciation Trends Network (STN)
By:
David Vaughn
Theresa O'Brien
Paul T. Roberts
Sonoma Technology, Inc.
1360 Redwood Way, Suite C
Petaluma, CA 94954-1169
Joann Rice
U.S. Environmental Protection Agency
Office of Air Quality Planning and Standards
Emissions, Monitoring and Analysis Division
Research Triangle Park, NC 27711
Contract No. EP-D-05-004
Task Order 1-04
U.S. Environmental Protection Agency
Office of Air Quality Planning and Standards
Emissions, Monitoring and Analysis Division
Research Triangle Park, NC 27711
-------
ACKNOWLEDGMENTS
The authors would like to thank Solomon Ricks and Kevin Cavender at the U.S.
Environmental Protection Agency (EPA) for their guidance during this project. Obviously this
work could not have been done without the hard work of staff at the various states operating the
semi-continuous speciation monitors: Darcy Anderson and Brant Englund at the Arizona
Department of Environmental Quality; Mark Bohlin of Cook County Department of
Environmental Control, Illinois, and Bob Swinford at the Illinois Environmental Protection
Agency; John Wicker and Don Klotz at the Indiana Department of Environmental Management;
Ed Michel at the Texas Commission of Environmental Quality and Earle Wright, the site
contractor in Texas; and Jim Frost and J.B. Dennison at the Washington Department of Ecology.
In addition, Jim Homolya of EPA was the original field measurements and special study
coordinator. At Sonoma Technology, Inc., Anna Frankel provided data processing and statistics
support, and Liz Simon and Nicole Hyslop supported the Arizona field site early in the study.
Sandy Smethurst of the STI publications staff provided editing support.
EPA DISCLAIMER
The information in this document has been funded wholly or in part by the United States
Environmental Protection Agency (EPA) under Contract EP-D-05-004 to Sonoma Technology,
Incorporated. Mention of trade manes or commercial products does not constitute EPA
endorsement or recommendation for use.
IV
-------
TABLE OF CONTENTS
Section Page
ACKNOWLEDGMENTS iv
EPA DISCLAIMER iv
LIST OF FIGURES vii
LIST OF TABLES x
1. INTRODUCTION 1-1
2. OBJECTIVES 2-1
3. CONCLUSIONS AND RECOMMENDATIONS 3-1
3.1 R&P 8400NNitrate Monitor 3-2
3.2 R&P 8400s Sulfate Monitor 3-2
3.3 R&P 5400 Carbon Monitor 3-3
3.4 Sunset Carbon Monitor 3-3
4. SITE DESCRIPTIONS 4-1
4.1 Chicago, Illinois 4-1
4.2 Phoenix, Arizona 4-1
4.3 Seattle, Washington 4-1
4.4 Houston, Texas 4-1
4.5 Indianapolis, Indiana 4-2
5. INSTRUMENTATION 5-1
5.1 STN Filter-Based Instruments 5-1
5.1.1 MetOneSASS 5-1
5.1.2 URGMASS400andURGMASS450 5-1
5.2 Semi-Continuous Monitors 5-1
5.2.1 R&P 8400N Nitrate Monitor 5-1
5.2.2 R&P8400S Sulfate Monitor 5-2
5.2.3 Carbon 5-2
5.3 Data Set Summary 5-4
5.4 Statistical Approach 5-6
5.4.1 Comparability 5-6
5.4.2 Predictability 5-7
6. PREVIOUS PRECISION AND COMPARISON RESULTS FOR STN SAMPLERS .... 6-1
6.1 Precision Estimates for Data From Collocated STN Samples 6-1
6.2 Results for Data From Collocated STN and IMPROVE 6-1
7. RESULTS 7-1
7.1 Precision Estimates 7-1
7.1.1 Single- (Within-) Sampler Precision Estimates 7-1
7.1.2 Between Sampler Precision Estimates 7-5
-------
7.2 R&P 8400N Nitrate Versus STN Filter-Based Nitrate 7-8
7.2.1 Nitrate Data Corrections 7-8
7.3 R&P 8400S Sulfate Versus STN Filter-Based Sulfate 7-13
7.4 Carbon 7-15
7.4.1 R&P 5400 Semi-Continuous Versus STN Filter-Based Carbon 7-16
7.4.2 Sunset Semi-Continuous Versus STN Filter-Based Carbon 7-18
8. OPERATIONAL ISSUES 8-1
8.1 R&P 8400N and 8400S 8-1
8.1.1 Distribution of Annual Hours 8-1
8.1.2 Flash Strips 8-1
8.1.3 Aqueous Standards 8-3
8.1.4 Water Reservoir andHydration Tube 8-3
8.1.5 Orifice Cleaning 8-3
8.1.6 Molybdenum Converter 8-3
8.1.7 Data Screen Failures 8-4
8.1.8 Data Acquisition and Reduction 8-4
8.2 R&P 5400 8-4
8.3 Sunset Carbon Monitor 8-6
9. REFERENCES 9-1
APPENDIX A: SCATTER PLOTS AND REGRESSIONS FOR PHOENIX, SEATTLE,
AND INDIANAPOLIS A-1
VI
-------
LIST OF FIGURES
Figure Page
Figure 5-1. Range of deployment dates for each of the monitors in at the five sites in the
study. Vertical lines denote yearly increments 5-5
Figure 7-1. Twenty-four-hour average R&P 8400N PM2.5 nitrate versus 24-hr integrated
STN filter nitrate at Chicago, from May 2, 2002, to May 10,2005 7-11
Figure 7-2. Twenty-four-hour average R&P 8400N PM2 5 nitrate versus 24-hr integrated
STN filter nitrate at Houston, from January 27, 2003, to April 13,2005 7-11
Figure 7-3. Average diurnal variability in R&P 8400N nitrate concentrations in Phoenix
during the winter (November through February) and summer (March through
October). Adjustments to the R&P 8400N data were made based on the regression
of 8400N data with the filter data 7-12
Figure 7-4. Nitrate data from Phoenix, Chicago, Seattle, and Indianapolis illustrating the
typical non-linearity of the semi-continuous data relative to the STN filter-based
data. Houston data are atypical and excluded 7-13
Figure 7-5. Twenty-four-hour average R&P 8400S PM2.5 sulfate versus 24-hr integrated
STN filter sulfate at Chicago, May 2, 2002 to May 10, 2005 7-14
Figure 7-6. Twenty-four-hour average R&P 8400S PM2.5 sulfate versus 24-hr integrated
STN filter sulfate at Seattle, May 2, 2002 to February 3, 2005 7-15
Figure 7-7. Twenty-four-hour average R&P 5400 PM2.5 total carbon versus 24-hr
integrated STN filter total carbon at Chicago, from May 2, 2002 to May 7, 2005 7-16
Figure 7-8. Twenty-four-hour average R&P 5400 PM2.5 organic carbon versus 24-hr
integrated STN filter organic carbon at Chicago, May 2, 2002 to May 7, 2005 7-17
Figure 7-9. Twenty-four-hour average R&P 5400 PM2.5 elemental carbon versus 24-hr
integrated STN filter elemental carbon at Chicago, May 2, 2002 to May 7, 2005 7-18
Figure 7-10. Twenty-four-hour average Sunset PM2.5 total carbon versus 24-hr integrated
STN filter total carbon at Chicago, August 1, 2004 to May 7, 2005 7-19
Figure 7-11. Twenty-four-hour average Sunset PM2.5 organic carbon versus 24-hr
integrated STN filter organic carbon at Chicago, August 1, 2004 to May 7, 2005 7-20
Figure 7-12. Twenty-four-hour average Sunset PM2.5 elemental carbon versus 24-hr
integrated STN filter elemental carbon at Chicago, August 1, 2004 to May 7, 2005.... 7-20
vn
-------
LIST OF FIGURES
Figure Page
Figure 7-13. Twenty-four-hour average Sunset PM2.5 total carbon versus 24-hr integrated
STN filter total carbon at Seattle, August 13, 2004 to April 10, 2005 7-21
Figure 7-14. Twenty-four-hour average Sunset PM2.5 organic carbon versus 24-hr
integrated STN filter organic carbon at Seattle, August 13, 2004 to April 10, 2005. ... 7-21
Figure 7-15. Twenty-four-hour average Sunset PM2.5 elemental carbon versus 24-hr
integrated STN filter elemental carbon at Seattle, August 13, 2004 to April 10,
2005 7-22
Figure 7-16. Twenty-four-hour average Sunset PM2.5 total carbon versus 24-hr integrated
STN filter total carbon at Phoenix, July 2, 2004 to May 13, 2005 7-22
Figure 7-17. Twenty-four-hour average Sunset PM2.5 organic carbon versus 24-hr
integrated STN filter organic carbon at Phoenix, July 2, 2004 to May 13, 2005 7-23
Figure 7-18. Twenty-four-hour average Sunset PM2.5 elemental carbon versus 24-hr
integrated STN filter elemental carbon at Phoenix, July 2, 2004 to May 13, 2005 7-23
APPENDIX A
Figure A-l. Twenty-four-hour average R&P 8400N PM2.5 nitrate versus 24-hr integrated
STN filter nitrate at Phoenix, from September 26, 2002, to May 13, 2005 A-2
Figure A-2. Twenty-four-hour average R&P 8400N PM2 5 nitrate versus 24-hr integrated
STN filter nitrate at Seattle, from May 2, 2002, to April 14, 2005 A-2
Figure A-3. Twenty-four-hour average R&P 8400N PM2.5 nitrate versus 24-hr integrated
STN filter nitrate at Indianapolis, from June 13, 2002, to March 11, 2005 A-3
Figure A-4. Twenty-four-hour average R&P 8400S PM2.5 sulfate versus 24-hr integrated
STN filter sulfate at Phoenix, from September 30, 2003, to May 12, 2004 A-3
Figure A-5. Twenty-four-hour average R&P 8400S PM2 5 sulfate versus 24-hr integrated
STN filter sulfate at Indianapolis, from June 19, 2002, to July 8, 2004 A-4
Figure A-6. Twenty-four-hour average R&P 8400S PM2 5 sulfate versus 24-hr integrated
STN filter sulfate at Houston, from January 27, 2003, to April 13, 2005 A-4
Figure A-7. Twenty-four-hour average R&P 5400 PM2.5 total carbon versus
24-hr integrated STN filter total carbon at Phoenix, from December 13, 2002, to
May 13,2005 A-5
Vlll
-------
LIST OF FIGURES
Figure Page
Figure A-8. Twenty-four-hour average R&P 5400 PM2.5 total carbon versus 24-hr
integrated STN filter total carbon at Seattle, from May 2, 2002, to March 29, 2005 A-5
Figure A-9. Twenty-four-hour average R&P 5400 PM2.5 total carbon versus 24-hr
integrated STN filter total carbon at Indianapolis, from June 26, 2002, to April 9,
2004 A-6
Figure A-10. Twenty-four-hour average R&P 5400 PM2 5 total carbon versus 24-hr
integrated STN filter total carbon at Houston, from January 24, 2003, to March 26,
2005 A-6
Figure A-l 1. Twenty-four-hour average R&P 5400 PM2.5 organic carbon versus 24-hr
integrated STN filter organic carbon at Phoenix, from December 13, 2002, to May
13,2005 A-7
Figure A-12. Twenty-four-hour average R&P 5400 PM2 5 organic carbon versus 24-hr
integrated STN filter organic carbon at Seattle, from May 2, 2002, to March 29,
2005 A-7
Figure A-13. Twenty-four-hour average R&P 5400 PM2.5 organic carbon versus 24-hr
integrated STN filter organic carbon at Indianapolis, from June 26, 2002, to April
9,2004 A-8
Figure A-14. Twenty-four-hour average R&P 5400 PM2 5 organic carbon versus 24-hr
integrated STN filter organic carbon at Houston, from January 24, 2003, to March
26,2005 A-8
Figure A-l5. Twenty-four-hour average R&P 5400 PM2.5 elemental carbon versus 24-hr
integrated STN filter elemental carbon at Phoenix, from December 13, 2002, to
May 13,2005 A-9
Figure A-16. Twenty-four-hour average R&P 5400 PM2 5 elemental carbon versus 24-hr
integrated STN filter elemental carbon at Seattle, from May 2, 2002, to March 29,
2005 A-9
Figure A-17. Twenty-four-hour average R&P 5400 PM2.5 elemental carbon versus 24-hr
integrated STN filter elemental carbon at Indianapolis, from June 26, 2002, to April
9,2004 A-10
Figure A-18. Twenty-four-hour average R&P 5400 PM2 5 elemental carbon versus 24-hr
integrated STN filter elemental carbon at Houston, from January 24, 2003, to
March 26, 2005 A-10
IX
-------
LIST OF TABLES
Table Page
3-1. Summary of comparisons of semi-continuous nitrate, sulfate, and carbon monitor
data with collocated filter-based STN data at the Phoenix, Chicago, Indianapolis,
Deer Park, and Seattle sites 3-1
5-1. Effect on regression statistics with two methods of data completeness evaluations
for the 10-minute nitrate records 5-5
6-1. Collocated precision estimates for speciated components in the STN network 6-1
6-2. Summary of regression results between IMPROVE and STN filter-based data 6-2
7-1. Coefficients of variation of measured nitrate mass for the 8400N blind performance
evaluations #4 and #5 of nitrate aqueous standards 7-3
7-2. Coefficients of variation of measured mass for the 8400S blind performance
evaluations #4 and #5 of sulfate aqueous standards 7-4
7-3. Coefficients of variation for the Sunset blind PE of EC, OC, and TC conducted in
spring 2005 7-5
7-4. Summary statistics for comparison of semi-continuous speciated PM2.5 data with
filter-based STN data 7-6
7-5. Theoretical conversion efficiencies from aqueous standards tests and gas analyzer
efficiencies from routine span audit 7-9
8-1. Percent of annual hours spent on operational and maintenance tasks for the R&P
8400 series monitors 8-2
8-2. Percent of annual hours spent on operational and maintenance tasks for the R&P
5400 OCEC monitor 8-5
-------
1. INTRODUCTION
The U.S. Environmental Protection Agency (EPA) established the National Ambient Air
Quality Standard (NAAQS) for PM2.5 gravimetric mass in 1997. A chemical Speciation Trends
Network (STN) of 54 sites was established across the United States in support of the NAAQS to
evaluate the chemical components of PM2.5. Major components of PM2.5 include nitrate (NOs),
sulfate (804), elemental carbon (EC), and organic carbon (OC). At STN sites, approved
speciation samplers are used to collect 24-hr filter-based measurements, nominally on every third
day. Nitrate, sulfate, and carbon fractions are routinely analyzed, along with a suite of elements
and PM2.5 mass. Research Triangle Institute (RTI) in North Carolina performs all filter analyses
for the STN. PM2 5 mass is determined from Teflon® filters using a gravimetric method that is
similar to the method used in the PM2 5 mass attainment network. Elemental analysis is
performed on the Teflon® filters (subsequent to gravimetric mass) by Energy-Dispersive X-Ray
Fluorescence (EDXRF). The major inorganic ions of filter-based PM2 5 STN samples are
determined from nylon filters by ion chromatography (1C) following extraction in water. EC and
OC fractions are collected using QMA-quartz filters that are analyzed by a thermal-optical
transmittance (TOT) method that is a modified version of NIOSH 5040 (NIOSH, 1999).
Semi-continuous instruments have been developed to measure ambient aerosol
concentrations of nitrate, sulfate, EC, and OC on a near real-time basis. At this time, no semi-
continuous or continuous monitors have been developed for determination of elemental
composition of PM2 5 in ambient air. Collocated comparisons of the instruments with the
standard methods used in the STN are needed to demonstrate their bias and precision. In the
spring 2001, the EPA initiated a study to evaluate these monitors and their potential use in the
routine speciation monitoring network. The semi-continuous monitors were collocated with
STN samplers at five monitoring sites. Ideally, the quality of data from the semi-continuous
monitors will be adequate to supplement speciation data collection; the data will have higher
time resolution and be acquired more rapidly; and the monitors will require less maintenance.
Proven continuous monitoring would eliminate transportation and laboratory analysis costs
associated with filter-based samples and would benefit public outreach and air quality
forecasting by increasing the timeliness of the data. Continuous-monitoring data may support
more accurate source apportionment and help elucidate the processes leading to the occurrence
of nitrate, sulfate, EC, and OC in PM2.5.
Because the spatial requirements of the comparison study are geographically diverse, a
wide range of environments was included in the study design. The participants in the study were
the Arizona Department of Environmental Quality (Phoenix site); Illinois Environmental
Protection Agency (Chicago site, operated by Cook County); Washington Department of
Ecology (Seattle site); Indiana Department of Environmental Management (Indianapolis site);
and Texas Council on Environmental Quality (Houston site). Although data were available from
two sites in Illinois, only data from the Chicago site were analyzed here. Monitors have been
collecting data for nearly three years at most of these sites.
1-1
-------
2. OBJECTIVES
EPA contracted with Sonoma Technology, Inc. (STI) to conduct an analysis of the
collocated filter-based and semi-continuous data collected during this study. Beyond the 54 STN
sites, the EPA speciation network has an additional component of state and local-directed
supplemental monitoring sites called SLAMS (State and Local Air Monitoring Stations). The
SLAMS comprise a very dynamic network that is currently comprised of about 200 sites placed
to meet state and local air monitoring needs. The goal is to determine if the commercially-
available semi-continuous carbon, nitrate, and sulfate monitors used in this study are sufficiently
robust to allow routine application in the speciation monitoring network.
Using filter-based samplers in the STN as the benchmark, the semi-continuous monitors
were evaluated for comparability and predictability (see Section 4.4 for definitions). Monitors
meeting comparability criteria would be useful for detecting spatial or temporal differences and
would be acceptable for use in supplementing the data collected by the routine STN network.
Comparable monitors would also be acceptable for use in the SLAMS network. Comparable
monitors would not require and corrections or adjustments to the data. Criteria for predictability
are the less stringent, but would allow calibration of the high time-resolution, semi-continuous
data with the filter-based data, leading to more accurate representations of the diurnal variation
that is known to exist for the major components of PM2.5.
2-1
-------
3.
CONCLUSIONS AND RECOMMENDATIONS
The results of the comparisons between the semi-continuous nitrate, sulfate, and carbon
data and collocated filter-based nitrate, sulfate, and carbon data from the STN network were
reviewed. The comparisons were based on 24-hr average data from the semi-continuous
monitors and 24-hr filter sampler data from the STN network. The results were then compared
with the criteria for comparability and predictability, as discussed in Section 5.4. Table 3-1
summarizes whether the data from the monitors, operated at each of the five sites, met the
comparability and predictability criteria.
Table 3-1. Summary of comparisons of semi-continuous nitrate, sulfate, and carbon
monitor data with collocated filter-based STN data at the Phoenix (AZ), Chicago (IL),
Indianapolis (IN), Deer Park (TX), and Seattle (WA) sites. Blank cells indicate that
the comparison did not meet the criteria.
Instrument
R&P 8400Na
Pre-molyb
Post-molyc
Comparability
(R2, slope and intercept)
AZ
NA
NA
NA
IL
NA
NA
NA
IN
NA
NA
NA
TX
NA
NA
NA
WA
NA
NA
NA
Comparability
(R2, ratio of means)
AZ
•/
IL
/
S
•/
IN
>/
S
•/
TX
S
WA
d
d
R&P 8400S
NA
NA
NA
NA
NA
S*
S*
S
S
R&P 5400 EC
R&P 5400 OC
R&P 5400 TC
g
g
A
•/
s
Sunset EC
Sunset OC
Sunset TC
,/
,/
e
e
e
e
e
e
d
S
g
g
g
g
g
e
e
e
e
e
e
g
>/
S
,/
,/
S
S
S
S
e
e
e
e
e
e
s
s
s
f
Entire data set; both prior to and after converter replacement
Only the R&P 8400N data subset prior to converter replacement
Only the R&P 8400N data subset after converter replacement
Close to meeting criteria
No monitor installed
Met ratio-of-means criteria, but not correlation criteria
g Met correlation, but not ratio-of-means criteria
NA Criteria did not apply
When evaluating comparability, neither the 8400N nor the 8400S provided results that
consistently met both the ratio-of-means and the correlation comparability criteria suggested for
this study. Only one 8400N monitor in Phoenix met both comparability criteria after converter
3-1
-------
replacement. It is clear that more improvements and enhancements are required to improve
comparability of these monitors with filter-based measurements. However, both the 8400N and
8400S did a better job at meeting the predictability (correlation) criterion. The 8400N monitors
at four of five sites met the predictability criterion; and the fifth site was close to meeting it. The
8400S met predictability at two sites (Indianapolis and Seattle). These monitors largely do not
provide data that are comparable to filter-based STN measurements; however, where monitors
met the predictability criterion, the data may be adequate for adjustment using the filter-based
STN measurements to more accurately represent diurnal patterns at these sites.
The comparability of the 5400 OCEC monitor, as compared to the filter-based STN
carbon measurements, was poor. Predictability was also poor. The predictability criterion for
OC was met at only one site (Seattle).
The Sunset carbon monitor did a better job at meeting, or came close to meeting, the
ratio-of-means and correlation criteria for carbon. However, it met the slope and intercept
comparability criteria on only one site. The Sunset carbon monitor met the predictability
criterion at all five sites, indicating that adjustment of the data against STN filter-based
measurements may be appropriate to more adequately represent diurnal patterns in speciation
monitoring network.
3.1 R&P 8400N NITRATE MONITOR
Data from the R&P 8400N met the correlation and ratio-of-the-means comparability
criteria at the Phoenix site, only for data after the converter was replaced. Data from
Indianapolis and Chicago met the ratio-of-means criteria, but not the correlation criterion. Data
from the other sites did not meet either of the comparability criteria. Data from the R&P 8400N
met the predictability criteria at four of the five sites and were close at the fifth site.
Compared with the STN nitrate data, there was significant bias in the R&P 8400N data at
all five sites; the bias was negative at four sites and positive at the Houston site. The nitrate
response appears to be non-linear, with a stronger bias evident at higher ambient nitrate
concentrations. The nitrate monitors seemed to have had consistently high conversion
efficiencies and gas analyzer efficiencies, although there were some indications that problems
with the gas analyzer converter efficiency could cause different results. The most time-
consuming activity for the nitrate monitor was data acquisition and data processing; running
aqueous standards and dealing with the flash strip failures were the next most time-consuming
activities.
3.2 R&P 8400S SULFATE MONITOR
Data from the R&P 8400S at the Phoenix and Houston sites met the ratio-of-means
criteria, but not the correlation criterion. The predictability criteria were met at only two of the
five sites. When compared with the STN sulfate data, there was significant bias in the R&P
8400S data at all five sites; the bias was negative at four sites and positive at the Seattle site. The
3-2
-------
sulfate monitors had significantly different conversion efficiencies at the various sites, although
the gas analyzer efficiencies were consistently high. The most time-consuming activity for the
sulfate monitor was data acquisition and data processing; running aqueous standards and dealing
with the flash strip failures were the next most time-consuming activities.
3.3 R&P 5400 CARBON MONITOR
Of the EC, OC, and TC data from the R&P 5400, only the OC and TC data at the Seattle
site met the predictability criteria while the TC data at the Indianapolis site almost met the
predictability criteria. R&P 5400 EC, OC, and TC data did not meet the comparability criteria at
any site. When compared with the STN carbon data, there was significant negative bias in the
R&P 5400 TC, OC, and EC data at all five sites; regression slopes typically averaged 0.4 to 0.6.
The most time-consuming activity for the 5400 carbon monitor was data acquisition and data
processing; flow problems, along with oven and afterburner problems, were the most frequent
and recurring operational issues.
3.4 SUNSET CARBON MONITOR
The results for the Sunset carbon monitor operated at three (Phoenix, Chicago, and
Seattle) of the sites were significantly better. The EC data at Phoenix met the slope and intercept
comparability criteria; the EC, OC, and TC data from all three sites either met or almost met the
ratio-of-means and correlation comparability criteria; and EC, OC, and TC at all three sites met
the predictability criteria. Compared with the STN TC and OC data, there was noticeable
positive bias in the Sunset data at low concentrations, but little bias at higher concentrations.
The most significant operational problems with the Sunset carbon monitor were failures with the
associated laptop computer and differences in the version of analysis software used to calculate
final concentrations.
3-3
-------
4. SITE DESCRIPTIONS
4.1 CHICAGO, ILLINOIS
The Chicago Com Ed site is located in a trailer at the Commonwealth Edison (Com Ed)
Maintenance facility, 7801 Lawndale Avenue, Chicago, Illinois. The AIRS Site ID is
170310076. The locational coordinates are latitude 41.885864 (41 deg 45 min 4.9 secN) and
longitude -87.625729 (87 deg 42 min 49.5 sec W). The site elevation is 186 m above mean sea
level.
The Com Ed facility covers an area of several square blocks to the east of Lawndale
Avenue. Residential areas are located immediately east, south, and west. Industrial areas,
including large rail yards, are located 1 km to the northwest and 1.5 km to the east. Chicago
Midway Airport is located 4 km to the northwest.
4.2 PHOENIX, ARIZONA
Phoenix is located in the central Arizona desert where many sources contribute to
observed PM2.5. It differs from the other locations in the study because most of its PM is
generally in the coarse fraction (PMio-2.s). The Phoenix data included in the this study were
collected at the Phoenix JLG Supersite, AIRS Site ID 04-013-9997, located at
4530 N. 17th Avenue (latitude 33.502959, 33 deg 30 min 13.0 sec N and longitude-112.095785,
112 deg 5 min 42.0 sec W). The site is in a residential neighborhood about 1.5 km east of
Interstate 17. The elevation is 346 m above mean sea level.
4.3 SEATTLE, WASHINGTON
The Seattle Beacon Hill site is located in a sampling trailer at the Beacon Hill Reservoir
at Charleston Street and 15th Avenue South. The AIRS Site ID is 53-033-0080. The locational
coordinates are latitude 47.569722 (47 deg 34 min 11 sec N) and longitude -122.312500
(122 degrees 18 min 45 sec W). The site elevation is 91 m above mean sea level. The Beacon
Hill reservoir covers an area of approximately 50 acres and is located to the south and east of
downtown Seattle. The site is adjacent to Jefferson Park and a golf course and is surrounded by
mixed residential and light commercial development. The Duwamish industrial area and
Interstate 5 are located approximately 1 km to the west of the site.
4.4 HOUSTON, TEXAS
The Houston Deer Park site used in this study is located at 4514-1/2 Durant St., Houston,
Texas. The site is in a residential neighborhood, but the Houston Ship Channel and associated
petrochemical and refining industries are a short distance to the north. The AIRS Site ID is
48-201-1039. The coordinates of the site are latitude 29.669722 (29 deg 40 min 11 sec N) and
longitude -95.128611 (95 deg 07 min 43 sec W). The site is 6 m above mean sea level.
4-1
-------
4.5 INDIANAPOLIS, INDIANA
The Indianapolis monitoring site is located in Washington Park, a public park adjacent to
a police station parking lot to the west. It is surrounded by both commercial and residential
properties, with mild traffic volume on the main road. The AIRS Site ID is 180970078. The
locational coordinates are latitude 39.811097 (39 deg 48 min 39.9 sec N) and longitude
-86.114469 (86 deg 6 min 52.1 sec W). The site elevation is 235 m above mean sea level.
4-2
-------
5. INSTRUMENTATION
5.1 STN FILTER-BASED INSTRUMENTS
The every-third-day filter-based measurements at the five sites were made with
speciation sampler types. The Spiral Ambient Speciation Sampler (SASS™; MetOne, Grants
Pass, Oregon) were used at Phoenix and Indianapolis, and the Mass Aerosol Speciation Sampler
(MASS 400 or MASS 450, URG, Chapel Hill, North Carolina) were used at Chicago, Houston,
and Seattle.
5.1.1 MetOne SASS
Sites with MetOne SASS samplers hosted either a MetOne SASS or a MetOne
SuperSASS. There are no significant operational differences between the two versions (the
SuperSASS has additional channels for multiple-event sampling). Parallel channels allow
simultaneous sampling, and each channel has a removable cartridge with a sharp-cut cyclone.
One cartridge contains a Teflon filter for gravimetric analysis. A second cartridge contains a
denuder to capture ammonia and nitric acid followed by a nylon filter. This sample filter is
extracted and analyzed by 1C for nitrate and sulfate. A third cartridge contains a quartz filter
which is analyzed by thermal oxidation to determine OC and EC concentrations.
5.1.2 URG MASS400 and URG MASS450
The URGMASS400 (Mass Aerosol Speciation Sampler for Anions, Cations, and Trace
Metals) is the filter-based sampler (with Teflon and Nylon filters) used for nitrate and sulfate
measurements. The URG MASS450 (Mass Aerosol Speciation Sampler for Organic and
Elemental Carbon and Semi-Volatile Organic Compounds) is the filter-based sampler (with a
quartz filter) used for aerosol carbon measurements.
5.2 SEMI-CONTINUOUS MONITORS
Speciation monitors manufactured by Rupprecht and Patashnik (R&P, Inc., Albany, New
York) deployed in this study include the R&P 5400 (carbon), R&P 8400N (nitrate), and R&P
8400S (sulfate) instruments. In mid-2004, a semi-continuous carbon monitor from Sunset
Laboratory, Inc., Tigard, Oregon, was added at three of the five sites.
5.2.1 R&P 8400N Nitrate Monitor
A two-step process is used by the R&P 8400N Nitrate Monitor (8400N) to collect and
analyze the aerosol sample. In this study, a 10-minute cycle time was used, with the first 8.75
minutes allotted to sample collection and the remaining time to analysis. In the collection step,
ambient air is drawn through a system that includes a PM2.s sharp-cut cyclone, a carbon
honeycomb denuder that removes interfering gases, and a Nafion humidifier that provides for
5-1
-------
uptake of water by the particles and minimizes particle bounce from the NiChrome® collection
strip. The air sample is maintained near ambient temperature by a sheath of ambient air
surrounding the sample inlet line. During the analysis period, purified nitrogen is used to purge
the collection cell; the purified nitrogen also acts as a carrier gas to the chemiluminescence
oxides of nitrogen (NOX) analyzer. A baseline reading is followed by the flash volatilization that
catalytically reduces the nitrate to NOX, which is carried to the high sensitivity gas analyzer for
analysis. The difference between the integrated areas under the baseline and sample gas stream
curves (adjusted for theoretical conversion efficiency) is combined with sampled volume to yield
an ambient nitrate concentration.
Previous field evaluations of data from the prototype instrument and the commercialized
version against various filter nitrate data have yielded inconsistent results. Intercomparisons
between data from the prototype instrument and filter-based data in initial field tests gave
favorable results (Stolzenburg and Hering, 2000; Liu et al., 2000; Hering and Stolzenburg,
1998). Following commercialization of the R&P instrument, data from field comparisons
against filter-based data suggest that significant bias in nitrate concentrations exists compared to
data from STN monitors (Hogrefe et al., 2004; Wittig et al., 2004; Harrison et al., 2004; Reid et
al., 2005). Some researchers have found good data precision with the 8400N, allowing
calibration of the high time-resolution data by the collocated filter data (Harrison et al., 2004;
Wittig et al., 2004). Calibration of the data provides potential for additional temporal analyses
but does not alleviate the equivalency requirement that would allow use of the semi-continuous
monitors in place of STN monitors.
5.2.2 R&P 8400S Sulfate Monitor
The R&P 8400S Sulfate Monitor (8400S) follows a sampling protocol that parallels the
R&P 8400N. The Pulse Generator component—the instrument that collects and flash volatilizes
the sample—is identical to the 8400N except that platinum flash strips are used, a shorter flash
duration (but a higher flash temperature) is used, and purified air is used as the carrier gas. A
high sensitivity SO2-pulsed UV fluorescence analyzer is used to measure the evolved gas.
Comparisons with filter-based sulfate data during a one-month study conducted at the
Atlanta Supersite revealed that data from the 8400S and from measurements were "fairly well
correlated" (Weber et al., 2003). Data from the 8400S were lumped with data from four other
semi-continuous instruments and then regressed against the 24-hr filter-based data, yielding a
slope of 1.15 and a correlation (r) of 0.92. Sample size for the 8400S was small (13 samples).
Drewnick et al. (2003) found high correlation (r = 0.99) between 8400S and STN filter-
based measurements of sulfate, with a slope of 0.935. Sample size was small (7 samples).
5.2.3 Carbon
Thermo Optical Analysis (TOA) is routinely used for analysis of OC and EC from filter-
based samples. It utilizes combustion in controlled atmospheres at selected temperatures,
combined with light transmittance or reflectance to allow correction for pyrolyzed (charred) OC.
5-2
-------
The gases produced in TOA are either converted to carbon dioxide (CO2) and detected directly
or converted to methane (CH4) and detected by flame ionization.
OC and EC fractions are defined operationally by the method used. Changes in
temperature regimes and dwell times, as well as the nature of the sample, can have significant
effects on the OC/EC split (Schauer et al., 2003). Different thermal evolution protocols will
yield different results; consistency in those protocols is needed to compare data sets. Thermal
Optical Reflectance (TOR) is the method used in the Interagency Monitoring of Protected Visual
Environments (IMPROVE) network, while Thermal Optical Transmission (TOT) is the method
used in the STN network. The IMPROVE TOR protocol uses a lower temperature protocol
compared to STN TOT. Many other EC and OC analysis protocols are widely used, but
generally the methods are insufficiently documented, making evaluations of their equivalence
difficult.
R&P 5400 Monitor
The R&P 5400 carbon monitor (5400) measures OC and total carbon (TC) mass in PM2.5.
As used in this study, the monitor collects a sample for three hours, and analyzes the collected
sample in about 20 minutes. The instrument has two separate sampling trains, allowing the
collection and analysis of separate samples to occur in parallel. When sample collection ends on
one train, sample collection begins on the second. The sample stream passes through a PM2.5
size-selective inlet, and the PM collects on an impaction surface. At the end of the 3-hr sampling
period, collection stops. In a closed system of ambient air, the chamber containing the sample is
heated to two temperature plateaus in succession—340°C and 750°C. In the first plateau of
340°C (8 minutes long), the assumption is that all of the OC combusts in the presence of oxygen
to form CO2 and that no EC combusts. In the second temperature plateau of 750°C, all EC
combusts to CO2. A non-dispersive infrared sensor measures the CO2 concentrations during
these periods and based on the cumulative CO2 concentration and sampled air volume, the
ambient concentrations of OC and TC are calculated. The EC concentration is then determined
by difference. There is no correction for pyrolysis in the 5400.
Measurements from the 5400 monitor have been compared with filter-based
measurements in a few field studies. R&P 5400 TC measurements were not comparable to or
predictable from collocated filter-based measurements, analyzed by TOR methods, at the Fresno
Supersite (Watson and Chow, 2002), where frequent instrument problems led to low data
recovery for the 5400. Average TC was 40-60% higher than filter-based TC. Correlation
coefficients (r) between the 5400 and filter-based measurements for TC, OC, and EC were 0.46,
0.38, and 0.61, respectively.
In a comparison with samples collected with a filter-based Reference Ambient Aerosol
Sampler (RAAS), analyzed by TOT at RTI (Rice, 2004), the 5400 underestimated TC and OC by
64% and 78%, respectively, while the EC component was overestimated by 89%. Correlation
coefficients (r) for TC and OC were 0.64 and 0.67, respectively, and the correlation for EC was
0.37. Aside from the lack of pyrolysis correction in the 5400, a positive artifact associated with
the filter-based measurements made without a denuder may have exacerbated the observed
differences.
5-3
-------
Comparisons conducted at the Atlanta Supersite showed very good agreement between
data from the 5400 and from an in situ TOT method for TC and OC, with regression slopes and
correlation coefficients close to 1.0 (Lim et al., 2003).
Sunset Carbon Monitor
In mid-2004, a Sunset Laboratory, Inc., Carbon Aerosol Analysis Field Instrument
(Sunset) employing TOT was installed at three of the five sites: Chicago, Phoenix, and Seattle.
This instrument is a field version of the Sunset laboratory-based analyzer, with the analysis
section of the analyzer built and operated similarly to the laboratory-based analyzer described by
Turpin et al. (1990). The monitor reports hourly concentrations of OC and EC, based on 47
minutes of sampling at a nominal flow rate of 8 1pm. The monitors at the three sites utilize a
modified NIOSH 5040 protocol. Following sampling, the filter is heated in an oxygen-free,
high-purity helium atmosphere to 600°C and then to 850°C, oxidizing OC to CO2 that is then
measured by a non-dispersive infrared detector. A red-light laser is used to monitor pyrolytic
conversion of OC to EC. Following the oxygen-free heating, the oven is switched to a 2%
oxygen/helium mixture and heated to 600°C and then to 850°C. During this phase, both original
EC and pyrolyzed OC burn in the presence of oxygen to form CO2. The point during the second
heating cycle when the laser transmittance equals its beginning value is considered the split
point—carbon measurements prior to this point are assigned to OC and those after to EC.
Bae et al. (2004) operated two collocated Sunset carbon monitors concurrently with
integrated a 24-hr filter-based sampler at the St. Louis-Midwest Supersite throughout 2002; the
sampler was built by Jamie Schauer at the University of Wisconsin, Madison, based on the
Caltech organics sampler. The semi-continuous analyzers operated on alternate hours,
employing a full hour of sampling instead of just 47 minutes so a truly continuous OCEC
measurement was obtained. The slopes of the linear regressions of Sunset TC and OC against
the filter-based measurements were 0.97 and 0.93, respectively, and exhibited correlation
coefficients (r) of 0.94 and 0.95, respectively. EC exhibited a slope of 0.95 but a poorer
correlation coefficients (r=0.60) attributed to very low EC levels at the sampling site (annual
average 0.70 ug/m3), with a large fraction of EC measurements near the detection limit.
5.3 DATA SET SUMMARY
Figure 5-1 graphically illustrates the deployment dates for the monitors at each of the
five sites. In constructing 24-hr average data from the semi-continuous monitors, at least
75% data completeness was required. Hourly reported data for the Sunset carbon monitor
required a minimum of 18 valid hourly averages per day. The R&P 5400 monitor reported 3-hr
averages in this study, and five of the six daily samples (83%) served as the completeness
criteria.
The 8400N and the 8400S report data as 10-minute averages, so determination of data
completeness is less straightforward. The method for assessing completeness of these data
differed between sites. For the Phoenix data, the nitrate measurements were assessed for data
completeness by the daily averages based on the validity of 108 of the possible 144 daily
10-minute averages. In all other cases, hourly averages were calculated as an initial step and
5-4
-------
accepted as valid if five of the six (83%) of the 10-minute periods were valid. This procedure
was followed with a secondary calculation of the 24-hr average, requiring 18 (75%) or more
valid hourly values.
Using the 10-minute nitrate data from the Phoenix site, a brief analysis was undertaken to
determine the effect on regression statistics by these two alternative completeness methods.
Results indicate minimal effects on the slopes, intercepts, and coefficients of determination for
this particular data set (Table 5-1).
AAAAAAAAAAAAAAAAAAAAAAAAk^AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
tfitftfiww^ . _ ...; ;
/XmYVYmWiWVWWV
YV\ftflftVW(VyWVVVS!SA/W¥WVV
-TeXias:;
vwywvwttMfflVfflwvgft^^ -....-'...
'i •'rv'rv'rv'rv'rv'rv'rv'rv'rv'rv'r I v-;Vi '.• fv'rv'rv'rv'rv'rv'rv'rv'rv'rv'rv'rv'rv'
. /wvvvvw^flivvvwwvwaw
Indiana
4/15/2002
4/15/2003
4/14/2004
4/14/2005
Figure 5-1. Range of deployment dates for each of the monitors in at the five sites
in the study. Vertical lines denote yearly increments.
Table 5-1. Effect on regression statistics with two methods of data completeness
evaluations for the 10-minute nitrate records.
Method
Hourly (greater than or equal to 5 of
6 10-min averages), then daily
(greater than or equal to 1 8 hourly
averages)
10-minute (greater than or equal to
108 of 144)
Number of
Records
895
903
Slope
0.7853
0.7846
Intercept
0.3449
0.343
R2
0.8742
0.8735
5-5
-------
5.4 STATISTICAL APPROACH
Formal evaluations of comparability and predictability between the semi-continuous
monitors and the STN monitors are based on ordinary least squares linear regressions. Average
y/x ratios and y-minus-x differences, with associated standard deviations, are presented to
supplement the regression statistics. The criteria used to decide whether the semi-continuous
monitors are sufficiently robust to be used to supplement STN and SLAMS monitors is based on
comparability and predictability. Below, we present criteria for comparability and predictability;
the results for the speciated monitors in this study are compared against these criteria in Sections
5 and 6.
5.4.1 Comparability
The criteria for comparability are less stringent than those, for example, used in PM2.5
mass equivalence, but the precisions involved are still sufficient to discern concentration
differences in space and time. There are two sources of criteria for comparability, the criteria
defined for STN samplers by the Expert Panel (Koutrakis, 1999), and the results obtained from
evaluation of several speciation samplers in the Four-City Study (Solomon et al., 2000).
Comparability is achieved when data from a monitor collocated with an STN monitor meet the
following criteria:
Nitrate criteria
o Squared correlation (r2) greater than or equal to 0.90
o The ratio of the means of 1 ± 0.1 for nitrate (Koutrakis, 1999). The Four-City Study
(Solomon et al., 2000), where a number of commercially-available speciation samplers
were field-tested, suggested that the ratio-of-means criteria be 1 ± 0.15 for nitrate. This is
the criteria used in this study to determine comparability for nitrate, since it reflects
method performance in a field application.
Sulfate criteria
o Squared correlation (r2) greater than or equal to 0.95
o The ratio of the means of 1 ± 0.05 for sulfate.
o The Four-City Study report did not suggest an alternative criterion for sulfate.
Carbon criteria
o The Expert Panel did not provide criteria for carbon; however, the Four-City Study report
suggested that the ratio-of-means criteria be 1 ± 0.15 for organic and elemental carbon.
o Definitions for comparability used for evaluation of in situ carbon measurements have
been suggested by Watson and Chow, 2002. The criteria for comparability were 1) a
slope that was 1 ± 3 standard errors; and 2) intercept equal to 0 ± 3 standard errors; and
3) a correlation (r) > 0.9. Both the ratio-of-means criteria and those proposed by Watson
and Chow, 2002 will be used to assess comparability for carbon.
There is no federal reference method (FRM) for speciation sampling; therefore, the STN
samplers are used as the reference for evaluating collocated semi-continuous speciation
samplers. Note that the collocated STN data for the most commonly used sampler (the
MetOne SASS) meet both the Expert Panel and Four-City Study criteria above (see Section
5-6
-------
5.1). In addition, data from collocated STN and IMPROVE samplers are similar for nitrate
and sulfate; thus, these criteria seem reasonable for an assessment of comparability.
5.4.2 Predictability
Predictability criteria are typically based on a minimum correlation coefficient (r)
between two measured variables. Watson and Chow suggest a criterion for correlation of 0.9.
The Data Quality Objectives (DQOs) for assessing the relationship between PM2 5 FRM and
continuous PM2.5 measurements provided a range of squared correlations (R2) from 0.73 to 0.84,
depending on the acceptable amount of decision error (U.S. EPA, 2002). Therefore, a
correlation of 0.9 seems to be a reasonable criterion for predictability in this study. The slope
may deviate substantially from unity and the intercept may also deviate from zero. Predictability
is of interest here, particularly with respect to accurately portraying highly time-resolved diurnal
variability. Nitrate, sulfate, and carbon fractions in ambient air are susceptible to substantial
diurnal variability from local sources. In cases where the high time-resolution, semi-continuous
monitors exhibit substantial bias but good precision, benchmark filter-based data may be used to
calibrate and adjust the high time-resolution data to more accurately portray these diurnal
patterns.
5-7
-------
6.
PREVIOUS PRECISION AND COMPARISON RESULTS FOR STN SAMPLERS
In this section, we present STN collocated results and comparison results of data from the
STN urban and IMPROVE (Interagency Monitoring of PROtected Visual Environments) rural
networks. The purpose of presenting these results is to provide a context and examples of how
well speciation data from filter-based STN samplers might be expected to compare with data
from collocated STN filter-based samplers where the sampling and analysis protocols are the
same and collocated IMPROVE samplers that use different sampling procedures, and in the case
of carbon, different analysis methods.
6.1 PRECISION ESTIMATES FOR DATA FROM COLLOCATED STN SAMPLES
The EPA has performed collocated sampling within the STN network (Rice, 2005).
Statistics for collocated samples for nitrate, sulfate, and carbon species are shown in Table 6-1.
Detailed statistics for the MetOne sampler, which is the most-used, are also shown in Table 6-1;
the slope and intercept results are from Deming regressions. The regression coefficients are
quite high (r) greater than 0.92 for all these species, with slopes very near one and intercepts very
near zero.
Table 6-1. Collocated precision and regression estimates for component species
in the STN network for 2002-2004.
Sampler
Type
MetOne
Parameter
Nitrate
Sulfate
OC
EC
TC
Number
1048
1047
1042
1041
1041
Regression
Slope
.985 ±
.01
1.019±
.01
.970 ±
.01
1.009±
.01
0.981 ±
.01
Intercept
Slope
-.052
±.05
-.094
±.02
.016±
.09
-.016
±.01
-.039
±.09
Ratio of
Mean
.977
.990
.973
.994
.984
Correlation
(r)
.99
.99
.92
.95
.93
STN/STN
%cv
12.3
10.6
15.5
17.0
16.2
6.2 RESULTS FOR DATA FROM COLLOCATED STN AND IMPROVE
STN filter-based data are being used to evaluate the composition of PM2.5 throughout the
United States. Publicly available STN and IMPROVE data from collocated samplers at three
locations (Seattle, Washington; Phoenix, Arizona; and Washington, DC) were obtained from
EPA's Air Quality System (AQS), and regressions of IMPROVE data on STN data were
conducted for nitrate, sulfate, EC, OC, and TC. A summary of the regression results is given in
Table 6-2.
In general, we used data for January 2002 through May 2004. However, sometimes
either the STN or IMPROVE data are missing; thus, the records are shorter. For the
6-1
-------
Washington, DC, site, we used carbon data from January 2, 2002 through May 30, 2004, and ion
data from January 2, 2002, through November 14, 2003. For the Seattle site, we used carbon and
ion data from May 2, 2002, through May 30, 2004. For the Phoenix site, we used carbon data
from December 13, 2002, through May 30, 2004, and ion data from August 12, 2002, through
May 30, 2004 (see Hyslop et al., 2004).
Correlation coefficients (r) of the nitrate and sulfate data from the two networks exceeded
0.97 at all three sites; the Seattle location showed a slight positive bias for both particle-phase
ions, and the Washington location a negative bias for nitrate. These results for nitrate using the
MetOne sampler and for sulfate using the MetOne and Anderson samplers are quite similar to
the collocated results shown in Table 6-1.
Carbon analysis methods differ between the two networks; thus, these comparisons are
not as good as those for collocated STN comparisons shown in Table 6-1. IMPROVE uses a
lower temperature protocol than STN. IMPROVE uses TOR for char correction, while STN uses
TOT. The Seattle comparison results indicate less OC and TC bias than do the results from the
other two sites. The URG sampler at the Seattle site has a nominal flow rate of 16.7 LPM, while
the MetOne and Andersen samplers have flow rates of 6.7 and 7.3 LPM respectively. The
IMPROVE sampler has a nominal flow rate of 22.8 LPM. The bias at the Phoenix and
Washington sites is likely due to the increased magnitude of the difference in sampling flow
rates for the MetOne and Andersen samplers and resulting difference in filter face velocity.
Correlations for the EC component were substantially lower than for OC and TC.
Table 6-2. Summary of regression results between IMPROVE and STN filter-
based data. Data are from the AQS.
Observable
Nitrate
Sulfate
rc
OC
EC
Location
Phoenix, AZ
Seattle, WA
Washington, DC
Phoenix, AZ
Seattle, WA
Washington, DC
Phoenix, AZ
Seattle, WA
Washington, DC
Phoenix, AZ
Seattle, WA
Washington, DC
Phoenix, AZ
Seattle, WA
Washington, DC
Sampler
y
Improve
Improve
Improve
Improve
Improve
Improve
Improve
Improve
Improve
Improve
Improve
Improve
Improve
Improve
Improve
X
MetOne
URG
Anderson
MetOne
URG
Anderson
MetOne
URG
Anderson
MetOne
URG
Anderson
MetOne
URG
Anderson
Ordinary Least Squares
Regression
Slope ±
Standard Error
0.965 ±0.01
1.34 ±0.02
0.873 ±0.02
0.979 ±0.02
1.101 ±0.01
0.999 ±0.02
0.86 ±0.02
1.1 ±0.02
0.69 ±0.02
0.872 ±0.03
1.15 ±0.03
0.657 ±0.02
0.751 ±0.04
0.756 ±0.05
0.665 ± 0.06
Intercept ±
Standard Error
0.02 ± 0.02
-0.05 ± 0.02
-0.01 ±0.04
0.04 ± 0.02
-0.04 ± 0.02
0.20 ±0.11
-1.36 ±0.16
-0.32 ±0.09
-0.14 ±0.12
-1.5 ±0.2
-0.41 ±0.1
-0.23 ±0.11
0.18 ±0.05
0.15 ±0.04
0.25 ± 0.04
Correlation
(r)
(Pearsons)
0.99
0.98
0.97
0.98
0.99
0.98
0.96
0.96
0.90
0.94
0.94
0.90
0.85
0.70
0.63
Number
of Pairs
196
206
175
82
205
175
158
215
220
158
215
220
158
215
220
6-2
-------
7. RESULTS
The EPA's NAREL (National Air and Radiation Environmental Laboratory) provides
quality assurance support to the EPA speciation networks. One of NAREL's tasks is to support
the semi-continuous speciation study by preparing performance evaluation (PE) samples, which
provide an estimate of the monitor bias. The PE samples are single-blind, meaning the operator
knows that the sample is a PE, but does not have the known concentration. When the PE
samples are analyzed in replicate, an estimate of within-monitor precision can obtained. This
section provides the precision results from the PE samples and summarizes the comparisons
between the semi-continuous and filter-based results (Section 7.1); detailed discussions of the
comparison results for nitrate (Section 7.2), sulfate (Section 7.3), and carbon (Section 7.4) are
also provided.
7.1 PRECISION ESTIMATES
This section provides the results of replicate, blind PE samples for the R&P nitrate and
sulfate monitors and for the Sunset carbon monitor (Section 7.1.1); no precision estimates are
provided for the R&P carbon monitor because this monitor's collection and analysis system is
closed; therefore, no PE sample can be introduced into the system and there is no way to easily
introduce a sample through the inlet. Section 7.1.2 provides precision estimates and detailed
statistics for the collocated semi-continuous and STN comparisons.
7.1.1 Single- (Within-) Sampler Precision Estimates
Collocation of semi-continuous instruments would provide the most robust method of
determining measurement precision because it includes the error contribution of all sampling
(e.g., flow rate) and analytical system components. No collocated semi-continuous instruments
were operated at the STN sites during this study. However, replicate blind PE samples can
provide estimates of single-sampler precision; these results are presented as coefficient of
variation (CV). Such estimates do not completely characterize the total measurement system
precision because they omit field errors, such as those associated with flow rate. In addition,
error unrelated to the sampling system may be introduced through differences in site-operator
techniques in administering the PE.
To date, there have been five PE tests for the nitrate and sulfate monitors, and two PE
tests for the Sunset carbon monitor. Replicate measurements were made at multiple levels of
deposited nitrate and sulfate mass. Similarly, for the Sunset carbon monitor, replicate samples of
filter punches were taken from sucrose-spiked filters and filters from STN ambient samplers.
NAREL provided blind aqueous standards for nitrate and sulfate and carbon filter punches for
the PE tests. The results are used here only for estimates of precision. Evaluation of the PE
results, including a discussion of bias as well as precision, is presented elsewhere (Taylor, 2005).
Within-sampler precision estimates (CV) for the R&P 8400N (Table 7-1) and 8400S
(Table 7-2) are based on available data from PE4 (fall 2004) and PE5 (spring 2005) that were
conducted at all five sites (Taylor, 2005). Precision estimates for the Sunset carbon monitors
7-1
-------
(Table 7-3) are based on the PE1 of fall 2004 and PE2 of spring 2005 at three sites (Taylor,
2005). No precision estimates are provided for the R&P 5400 carbon monitor.
For the Sunset monitor, CVs are presented from ambient samples only—sucrose-spiked
filters have very high CVs for EC because of the absence of any EC in these spiked samples.
Measurements of low-level chemical species such as EC are more likely to show large relative
variability compared to PM2.5 measurements. The Sunset within-sampler precisions averaged 2%
for OC and TC (Table 7-3). The EC precision of 8% was inflated by the high CV (28%) in
Phoenix for PE 2. If the suspect data underlying the 28% CV are eliminated, the CV for EC
drops to 4%.
The average CV for nitrate, including all sites over all levels of deposited nitrate mass,
was 5.2% in both PE4 and PE5 (Table 7-1). The average CV for sulfate was 10.7% in PE4 and
5.1% in PE5 (Table 6-2); overall average is 8.3%. There were no Phoenix PE data for sulfate,
and a sulfate PE was not conducted at the Indianapolis site in spring 2005.
7-2
-------
Table 7-1. Coefficients of variation of measured nitrate mass for the 8400N PE4 (fall 2004) and PE5 (spring 2005) of
nitrate aqueous standards. The overall mean CV is 5.2%. (Data courtesy of NAREL.)
Site
Phoenix PE 4
Chicago PE 4
Indianapolis PE 4
Houston PE 4
Seattle PE 4
Average PE 4
Phoenix PE 5
Chicago PE 5
Indianapolis PE 5
Houston PE 5
Seattle PE 5
Average PE 5
Mass Deposited (ng)
10
20
30
100
200
250
300
400
CV (%)
1.4%
4.2%
9.7%
5.0%
1.8%
4.4%
—
—
—
—
—
_
—
—
-
—
—
—
6.3%
3.2%
10.4%
2.8%
7.8%
6.1%
2.9%
1.0%
5.1%
4.8%
7.8%
4.3%
—
—
—
—
—
_
13.8%
2.6%
5.5%
3.2%
1.9%
5.4%
3.2%
1.3%
1.2%
6.1%
5.6%
3.5%
—
—
-
—
—
—
7.8%
0.8%
11.3%
12.6%
9.3%
8.3%
1.1%
0.6%
8.2%
7.0%
4.4%
4.3%
—
—
—
—
—
_
—
—
-
—
—
—
3.4%
1.7%
2.0%
10.3%
6.9%
4.9%
9.0%
1.3%
6.3%
6.3%
15.3%
7.6%
5.6%
2.2%
3.8%
1.8%
2.1%
3.1%
Overall Average
Average
5.6%
1.9%
6.9%
5.3%
6.2%
5.2%
5.3%
1.8%
5.8%
6.7%
6.3%
5.2%
5.2%
-------
Table 7-2. Coefficients of variation of measured mass for the 8400S PE4 (fall 2004) and PE5 (spring 2005) of sulfate
aqueous standards. The overall average CV is 8.3%. (Data courtesy of NAREL.)
Site
Phoenix PE 4
Chicago PE 4
Indianapolis PE 4
Houston PE 4
Seattle PE 4
Average PE 4
8400S PE 5
Phoenix PE 5
Chicago PE 5
Indianapolis PE 5
Houston PE 5
Seattle PE 5
Average PE 5
Mass Deposited (ng)
30
60
100
150
240
300
600
900
1200
CV (%)
—
3.8%
11.6%
15.5%
12.1%
10.7%
—
—
—
—
—
—
—
-
—
—
—
4%
—
13%
6%
7.8%
—
8.1%
29.9%
21.4%
1.2%
15.1%
—
—
—
—
—
—
—
2.8%
13.9%
20.3%
11.8%
12.2%
—
—
—
—
—
—
—
—
-
—
—
—
2%
—
8%
14%
8.4%
—
5.6%
30.6%
13.1%
7.0%
14.1%
—
—
—
—
—
—
—
—
-
—
—
—
2%
—
4%
3%
3.2%
—
4.9%
13.0%
8.4%
5.5%
8.0%
—
1%
—
2%
7%
3.4%
—
2.5%
4.1%
2.4%
7.9%
4.2%
—
2%
—
4%
2%
2.6%
Overall Average
Average
4.6%
17.2%
13.5%
7.6%
10.7%
2.3%
—
6.2%
6.7%
5.1%
8.3%
-------
Table 7-3. Coefficients of variation for the Sunset blind PE1 (fall 2004) and PE2
(spring 2005) of EC, OC, and TC. The overall mean CV is 4%. The average CV
for EC drops from 8% to 4% if the Phoenix PE 2 value is eliminated. (Data
courtesy of NAREL.)
Sunset
Site
Phoenix PE 1
Phoenix PE 2
Chicago PE 1
Chicago PE 2
Seattle PE 1
Seattle PE 2
Average CV
CV (%)
EC
0%
28%
5%
2%
6%
6%
8%
OC
1%
6%
2%
1%
1%
2%
2%
TC
1%
4%
1%
1%
0%
2%
2%
7.1.2 Between Sampler Precision Estimates
Regression statistics and precision estimates (a), as the standard deviation of the paired
differences, for the nitrate, sulfate, and carbon comparisons are shown in Table 7-4. The slopes
and intercepts, and their standard errors, from the ordinary least squares regression are presented
for each set of collocated measurements. The regressions utilized the semi-continuous monitor
data as the y variable and the filter-based data as the x variable. Also, Pearsons correlation
coefficients (r), the number of pairs in the comparison, the average y/x ratios and their standard
deviations, the average of the paired differences (y - x), and the standard deviation of the paired
differences are given. The distribution of the data pairs whose difference is less than la,
between 1 and 2a, between 2 and 3a, and greater than 3a are also provided; the standard
deviation of the paired differences is represented by a. Representative scatter plots and
regressions are presented in the discussions of the nitrate results (Section 7.2), sulfate results
(Section 7.3), and carbon results (Section 7.4); scatter plots not shown in the text are shown in
Appendix A. Also discussed in Sections 7.2 and 7.3 are the results of nitrate and sulfate aqueous
standards tests and of gas analyzer tests; these tests are important components of quality control
activities for the semi-continuous speciation monitors.
In the discussions in Sections 7.2 through 7.4, we compare the results shown in Table 7-4
with the criteria discussed in Section 5.4.
7-5
-------
Table 7-4. Summary statistics for comparison of semi-continuous speciated PM2.5 data with filter-based STN data.
Page I of 2
Observable
NO3 (all data)
NO3 ("Pre-moly")
NO3 ("Post-moly")
S04
Location
Chicago, IL
Phoenix, AZ
Indianapolis, IN
Houston, TX
Seattle, WA
Chicago, IL
Phoenix, AZ
Indianapolis, IN
Houston, TX
Seattle, WA
Chicago, IL
Phoenix, AZ
Indianapolis, IN
Houston, TX
Seattle, WA
Chicago, IL
Phoenix, AZ
Indianapolis, IN
Houston, TX
Seattle, WA
Sampler
y
R&P 8400N
R&P 8400N
R&P 8400N
R&P 8400N
R&P 8400N
R&P 8400N
R&P 8400N
R&P 8400N
R&P 8400N
R&P 8400N
R&P 8400N
R&P 8400N
R&P 8400N
R&P 8400N
R&P 8400N
R&P 8400S
R&P 8400S
R&P 8400S
R&P 8400S
R&P 8400S
X
URG
MetOne
MetOne
URG
URG
URG
MetOne
MetOne
URG
URG
URG
MetOne
MetOne
URG
URG
URG
MetOne
MetOne
URG
URG
Ordinary Least Squares
Regression
Slope ±
Standard
Error
0.54 ±0.01
0.79 ±0.02
0.59 ±0.02
1.97 ±0.07
0.85 ±0.03
0.50 ±0.02
0.85 ±0.02
0.62 ±0.03
2.02 ±0.09
0.81 ±0.03
0.67 ±0.03
0.65 ±0.02
0.52 ±0.02
1.90 ±0.12
0.96 ±0.05
0.59 ±0.02
0.76 ±0.10
0.66 ±0.03
0.81 ±0.06
1.25 ±0.03
Intercept ±
Standard
Error
0.53 ±0.05
0.34 ±0.04
0.41 ±0.08
0.34 ±0.09
0.24 ±0.03
0.54 ±0.06
0.34 ±0.04
0.39 ±0.10
0.32 ±0.12
0.29 ±0.03
0.43 ±0.10
0.27 ± 0.04
0.40 ±0.10
0.37 ±0.16
0.11 ±0.05
0.78 ±0.10
0.18±0.10
0.50 ±0.18
0.40 ±0.23
0.24 ± 0.04
Correlation
(r)
(Pearsons)
0.91
0.94
0.90
0.89
0.87
0.92
0.94
0.90
0.90
0.87
0.94
0.98
0.95
0.87
0.89
0.85
0.71
0.93
0.74
0.94
Number
of Pairs
277
270
178
194
297
197
208
126
122
203
80
62
52
72
94
263
62
94
175
245
Average
Ratio of y/x
± Standard
Deviation
0.88 ±0.27
1.30 ±0.80
0.82 ±0.26
2.65 ±1.75
1.26 ±0.53
0.85 ±0.25
1.37 ±0.86
0.85 ±0.28
2.51 ±0.95
1.31 ±0.56
0.97 ±0.28
1.05 ±0.52
0.73 ±0.16
2.89 ±2.59
1.16 ±0.43
0.94 ±0.31
0.99 ±0.33
0.81 ±0.19
0.96 ±0.45
1.52 ±0.27
Avg
Difference
of y-x
(ug/m3)
-0.64
0.05
-0.74
1.28
0.12
-0.69
0.16
-0.61
1.33
0.14
-0.48
-0.34
-1.05
1.20
0.08
-0.58
-0.04
-1.33
-0.31
0.55
S.D. of
Avg
Difference
ofy -x
(ug/m3)
1.33
0.63
1.21
1.19
0.34
1.43
0.56
1.15
1.19
0.36
1.01
0.68
1.31
1.19
0.29
1.59
0.31
1.77
1.41
0.41
Distribution
3a
9
5
6
13
6
8
6
6
9
4
3
1
2
4
1
6
0
o
3
3
19
-------
Table 7-4. Summary statistics for comparison of semi-continuous speciated PM2.5 data with filter-based STN data.
Page 2 of 2
Observable
rc
OC
EC
Location
Chicago, IL
Phoenix, AZ
Indianapolis, IN
Houston, TX
Seattle, WA
Chicago, IL
Phoenix, AZ
Seattle, WA
Chicago, IL
Phoenix, AZ
Indianapolis, IN
Houston, TX
Seattle, WA
Chicago, IL
Phoenix, AZ
Seattle, WA
Chicago, IL
Phoenix, AZ
Indianapolis, IN
Houston, TX
Seattle, WA
Chicago, IL
Phoenix, AZ
Seattle, WA
Sampler
y
R&P 5400
R&P 5400
R&P 5400
R&P 5400
R&P 5400
Sunset Lab
Sunset Lab
Sunset Lab
R&P 5400
R&P 5400
R&P 5400
R&P 5400
R&P 5400
Sunset Lab
Sunset Lab
Sunset Lab
R&P 5400
R&P 5400
R&P 5400
R&P 5400
R&P 5400
Sunset Lab
Sunset Lab
Sunset Lab
X
URG
MetOne
MetOne
URG
URG
URG
MetOne
URG
URG
MetOne
MetOne
URG
URG
URG
MetOne
URG
URG
MetOne
MetOne
URG
URG
URG
MetOne
URG
Ordinary Least Squares
Regression
Slope ±
Standard
Error
0.53 ±0.03
0.35 ±0.02
0.51 ±0.02
0.46 ±0.03
0.58 ±0.01
0.94 ±0.04
0.76 ±0.03
0.91 ±0.02
0.54 ±0.02
0.33 ±0.02
0.50 ±0.03
0.47 ±0.03
0.53 ±0.02
0.84 ±0.04
0.68 ±0.03
0.79 ±0.03
0.31 ±0.03
0.40 ±0.02
0.39 ±0.04
0.29 ±0.14
0.63 ± 0.02
1.11 ±0.04
1.04 ±0.03
1.23 ±0.06
Intercept ±
Standard
Error
0.27 ±0.11
0.95 ±0.14
0.49 ±0.13
0.81 ±0.11
0.47 ± 0.06
0.80 ±0.15
0.22 ±0.19
0.58 ±0.09
0.35 ±0.08
0.72 ±0.13
0.60 ±0.13
0.65 ±0.10
0.57±0.05
0.87 ±0.13
0.39 ±0.20
0.73 ±0.11
0.03 ±0.03
0.28 ±0.03
0.03 ± 0.03
0.21 ±0.06
0.02 ± 0.02
0.09 ±0.04
0.005 ±0.03
-0.03 ± 0.06
Correlation
(r)
(Pearsons)
0.78
0.85
0.89
0.75
0.92
0.95
0.95
0.98
0.80
0.80
0.86
0.73
0.90
0.93
0.91
0.96
0.49
0.80
0.65
0.16
0.83
0.95
0.97
0.92
Number
of Pairs
270
170
142
155
325
70
91
74
270
170
142
155
325
70
91
74
270
170
142
155
325
70
91
74
Average
Ratio of y/x
± Standard
Deviation
0.61 ±0.18
0.52 ±0.18
0.63 ±0.13
0.83 ±0.36
0.76 ± 0.24
1.20 ±0.23
0.81 ±0.20
1.15±0.32
0.67 ±0.18
0.48 ±0.17
0.66 ±0.15
0.81 ±0.35
0.80 ±0.27
1.19 ±0.26
0.78 ±0.23
1.14 ±0.33
0.37 ±0.34
0.83 ±0.75
0.45 ±0.27
0.99 ±1.29
0.68 ±0.30
1.26 ±0.29
1.05 ±0.23
1.24 ±0.68
Avg
Difference
ofy-x
(ug/m3)
-1.48
-3.89
-2.07
-0.71
-1.04
0.56
-1.23
0.21
-1.05
-3.47
-1.71
-0.65
-0.81
0.39
-1.26
0.06
-0.42
-0.41
-0.36
-0.07
-0.22
0.17
0.04
0.15
S.D. of
Avg
Difference
ofy-x
(ug/m3)
0.99
2.82
1.34
0.94
1.07
0.54
1.19
0.47
0.80
2.38
1.27
0.88
0.97
0.51
1.23
0.66
0.31
0.57
0.25
0.35
0.25
0.17
0.17
0.31
Distribution
3a
21
15
12
2
12
2
3
2
21
17
9
o
5
13
1
3
1
17
5
9
3
10
3
2
3
-------
7.2 R&P 8400N NITRATE VERSUS STN FILTER-BASED NITRATE
7.2.1 Nitrate Data Corrections
Aqueous standard salt solutions of potassium nitrate (KNOs) were used to evaluate the
conversion efficiency of the 8400N at each site. Frequency of the tests ranged from monthly to
quarterly. Replicate measurements were made at zero nitrate and four additional nitrate levels
(nominally 20, 40, 60, and 80 ng applied nitrate). Slopes of the measured mass versus deposited
mass were applied as corrections to the data, either as inputs in the 8400N setup under
"Theoretical Conversion Factor", or during post-processing of the data. Zero and span audits of
the NO analyzer were undertaken every few days (e.g., every three days at the Phoenix site).
Average results for each site are shown in Table 7-5. The average theoretical conversion
efficiency is the percent of the nitrate placed via the aqueous standard on the strip that is
measured in the gas analyzer; we expect these efficiencies to be consistent across monitors (i.e.,
from site to site in this study). We also expect the efficiency to be close to 100%, thus
demonstrating that all nitrate placed via the aqueous standard on the strip was volatilized and
subsequently measured in the gas analyzer. The efficiencies for nitrate are generally high (87%
to 96%) and fairly consistent among the sites.
Other corrections could potentially be applied to the semi-continuous 8400N nitrate data.
Although none of the additional corrections discussed below have been applied to the data from
this study, their potential to affect the summary statistics should be noted.
In the Pittsburgh Air Quality Study (PAQS), Wittig et al., 2004 applied a correction for
instrument offset (18% on average), determined by sampling HEPA-filtered air (dynamic zero)
during a series of 10-minute cycles, that was subtracted from the 8400N measurements.
Bimonthly dynamic zero measurements were made during particulate nitrate measurements at
the St. Louis-Midwest Supersite (Reid et al., 2005), but the data were not used to correct for
instrument offset because of high variability in the blank values and uncertainty as to whether the
dynamic zero values reflected a true blank measurement.
Corrections for sample flow drift may also be applied, based on measured sample flow
rate and the flow rate indicated by the instrument. In PAQS (Wittig et al., 2004), flow drift
corrections averaged -3% for nitrate. It is also possible to construct correction curves to account
for reaction cell vacuum drift. The 8400N reaction cell vacuum set point is 5.0 in. Hg.
Deviations from this set point can affect nitrate measurements through the instrument offset,
conversion efficiency, and gas analyzer efficiency. During PAQS (Wittig et al., 2004),
corrections for reaction cell vacuum drift averaged -1%.
Note: None of the additional corrections described above, were applied to the data from
this five site study.
7-8
-------
Table 7-5. Theoretical conversion efficiency from aqueous standards tests and gas
analyzer efficiency from routine span audits.
Location
Phoenix, AZ
Chicago, IL
Seattle, WA
Houston, TX
Indianapolis, IN
Nitrate
Aqueous Standards Tests
Average
Theoretical
Conversion
Efficiency
± St Dev (%)
96.1 ± 14.2
94.4 ±4.0
96.1 ±4.7
95.6 ±13.1
87.5 ± 10.9
Average
Correlation
Coefficient
(r)
0.998
0.995
0.992
0.996
0.958
Gas Analyzer
Audits
Average
Analyzer
Efficiency
± St Dev (%)
96.3 ±2.9
99.2 ± 1.3
98.1 ±2.1
N/A
N/A
Sulfate
Aqueous Standards Tests
Average
Theoretical
Conversion
Efficiency
± St Dev (%)
215.2 ±95 .2
63.3 ±7.7
50.3 ±8.9
64.9 ±15.4
83.4 ±9.9
Average
Correlation
Coefficient
(r)
0.978
0.979
0.981
0.992
0.955
Gas Analyzer
Audits
Average
Analyzer
Efficiency
± St Dev (%)
94.6 ± 16.8
99.7 ±0.8
100.1 ±3.5
N/A
N/A
N/A = Data not available.
Applying all the corrections above during PAQS (Wittig et al., 2004) resulted in good
correlations with filter-based measurements, but significant bias in 8400N measurements still
existed.
The molybdenum catalyst in the NC>2 converters has a fixed lifetime (nominally one year)
and exhibits decreased efficiency with time. Operators at each of the five sites returned the NOX
analyzer of the 8400N monitors to R&P during 2004 to have the converters evaluated and
replaced. These converter changes occurred at different times of the year for each site, and the
instruments were inoperable at differing time periods. As a consequence, the 8400N nitrate data
were evaluated in three ways: pre-converter replacement, post-converter replacement, and all
data combined. Replacement of the molybdenum converter increased the regression slopes for
the data from the Chicago and Seattle monitors, while the slopes for the data from the Phoenix,
Indianapolis, and Houston monitors decreased following converter replacement (Table 7-4).
From filter-based data, nitrate concentrations ranged up to 5 |j.g/m3 in Seattle and
Houston, 12 |ig/m3 in Indianapolis, 16ug/m3 in Chicago, and to 18 ug/m3 in Phoenix (see
Figures 7-1 to 7-4, for example).
Data from the R&P 8400N met the correlation and ratio-of-means comparability criteria
at the Phoenix site after converter replacement; data from the Illinois site met the ratio-of-means
criteria in all cases, but not the correlation criteria; and data from the Indiana site met the ratio-
of-means criteria after converter replacement, but not the correlation criteria. The other sites did
not meet either of the comparability criteria. Data from the R&P 8400N met the predictability
criteria at four of the five sites and were close at the fifth site. With the exception of the Houston
8400N, the semi-continuous monitors underestimated nitrate concentrations, as measured by the
STN filter-based samplers, except at very low concentrations. Regression slopes ranged from
7-9
-------
0.50 to 0.96, with associated standard errors of 0.01 to 0.05. All the data sets showed positive
intercepts, suggesting a positive sampling artifact. The intercepts ranged from 0.11 to
0.54 H-g/m3, reflecting the manufacturer's stated instrument error of 0.4 ug/m3. No dynamic ze
data were available for correcting the nitrate data for instrument offset.
In Chicago (Figure 7-1), the 8400N averaged only 54% of the filter-based concentrations
using all data. The average ratio of 8400N nitrate to filter-based nitrate was 0.88 ± 0.27, and the
average difference (y - x) in concentration was -0.64 ug/m3. The Chicago nitrate data is typical
of the response observed at all sites except Houston.
In Phoenix and Seattle, the low slopes were coupled with high 8400N/STN (y/x) ratios.
This is due to the preponderance of data at low concentrations and the positive intercepts. In
Phoenix, even though the slope of 8400N versus STN indicated that the semi-continuous monitor
was capturing only about 79% of the nitrate collected by the filters, the ratio of 8400N nitrate to
filter-based nitrate was 1.3 ± 0.80, and the average concentration difference (y - x) was only
0.05 M-g/m3. Approximately 75% of the concentrations in Phoenix were less than 2 ug/m3.
These low concentrations combined with the positive intercept to yield a high ratio and low
average difference.
In Seattle, about 80% of the filter-based nitrate concentrations were less than 1 ug/m3,
yielding an average y/x ratio of 1.26 ± 0.53 and an average concentration difference of
0.12ug/m3.
There was a substantial break in the time series of the Indianapolis nitrate data
(Figure 5-1) so there are about one hundred fewer sample pairs than at the other sites.
The response of the Houston 8400N was atypical among the five sites (Figure 7-2), with
results in the opposite direction. Slopes approached 2 for all data subsets, indicating a
substantial overestimate of STN filter-based nitrate. Average y/x ratios were between 2.5 and
3.0 for the data subsets, and the average difference (y - x) exceeded 1.2 ug/m3. This large
discrepancy between the Houston site and the others is not explained, but raises additional
concern about the true bias of the monitor compared to filter-based measurements.
The Seattle and Houston sites use URG samplers. For nitrate, the URG sampler uses a
Teflon® and backup nylon filter for the collection and analysis of total nitrate. Non-volatile
nitrate is collected on the front Teflon® filter and volatilized nitrate is collected on the backup
nylon filter. Teflon filters are also used for gravimetric mass determinations and Energy-
Dispersive X-Ray Fluorescence (EDXRF) for elemental analysis. The Teflon® filter is weighed
first for mass, then subject to vacuum EDXRF, and finally extracted for ion analysis (including
nitrate). The 4-City Study report (Solomon et al., 2002) noted that up to 40% of the nitrate
collected on the Teflon filter can be lost due to vacuum XRF. Any losses on the Teflon® filter
would affect the bias between the STN URG and 8400N at these sites.
7-10
-------
One-to-one
Pre-moly
O Post-moly
D all data
estfit (post-moly)
estfit (pre-moly)
est fit (all data)
All Data
y= 0.5406X +0.534
ost-moly
y = 0.6693X + 0.4345
R2 = 0.8743
Pre-moly
y = 0.5627X + 0.543S
URG Filter-based Nitrate
Figure 7-1. Twenty-four-hour average R&P 8400N PM2.5 nitrate versus 24-hr
integrated STN filter nitrate at Chicago, from May 2, 2002, to May 10, 2005.
"a
3.12
One-to-one
A Pre-moly
O Post-moly
D all data
• • Best fit (pre-moly)
~ • Best fit (post-moly)
Best fit (all data)
All Data
y= 1.9716X +0.3411
R2 = 0.7885
Pre-moly
y=2.0198x +0.3201
R2 = 0.804
URG Filter-based Nitrate (|jg/m3)
Figure 7-2. Twenty-four-hour average R&P 8400N PM2.5 nitrate versus 24-hr
integrated STN filter nitrate at Houston, from January 27, 2003, to April 13, 2005.
Although data from the R&P 8400N met both the correlation and ratio-of-means
comparability criteria at only the Phoenix site, the R&P 8400N data did meet the predictability
criteria at four of the five sites and were close at the fifth site. One benefit of this predictability
is that the high time-resolution data may be calibrated by the filter data, yielding improved
accuracy for examination of diurnal variability in nitrate concentrations. An example of this,
using the nitrate data from Phoenix, is given in Figure 7-3, where the average diurnal nitrate
7-11
-------
concentrations for wintertime (November through February) and summertime (March through
October) nitrate data are adjusted, based on the filter data. Wintertime concentrations increased
following adjustment while summertime concentrations decreased, reflecting the overestimation
of nitrate by the 8400N at low ambient concentrations and the underestimation at higher
concentrations at the Phoenix monitor.
Scatter plots and regressions for Phoenix, Seattle, and Indianapolis are included in
Appendix A. With the exception of the atypical Houston data, the semi-continuous nitrate data
reflect a non-linear component with respect to the filter-based data (Figure 7-4), where at low
concentrations, the semi-continuous data exhibit minimal scatter and adhere more closely to the
1:1 line. At higher concentrations the semi-continuous data are noticeably lower than the filter-
based nitrate. This pattern is similar to those reported by other researchers (Reid et al., 2005;
Wittig et al., 2004). One of the postulated explanations for this pattern is a matrix effect between
8400N nitrate recovery and aerosol composition, where a deficiency of electron donors prevents
complete reduction of nitrate to NO and NO2 (Reid et al., 2005). An experiment utilizing
CO-doped purge gas (increasing available electron donors) yielded a sustained 20% increase in
nitrate concentration for the 8400N, although 8400N nitrate remained substantially lower than
the filter-based measurements (Reid et al., 2005). Continued evaluations may lead to further
performance enhancements and the potential for future comparability of the 8400N with STN
filter-based methods.
3.5 T..-.- ,,
12:00 AM
6:00 AM
12:00 PM
6:00 PM
12:00 AM
Figure 7-3. Average diurnal variability in R&P 8400N nitrate concentrations in
Phoenix during the winter (November through February) and summer (March
through October). Adjustments to the R&P 8400N data were made based on the
regression of 8400N data with the filter data.
7-12
-------
20.00
15.00
E
ro
10.00
5.00
L :^'yT .-• •-;••. ... i
_ ..^,r-:v *;-•• *
*V*:" *- • /
0.00
10
STN Nitrate (ng rrf3
15
20
Figure 7-4. Nitrate data from Phoenix, Chicago, Seattle, and Indianapolis
illustrating the typical non-linearity of the semi-continuous data relative to the
STN filter-based data. Houston data are atypical and excluded.
7.3 R&P 8400S SULFATE VERSUS STN FILTER-BASED SULFATE
Aqueous standard salt solutions of ammonium sulfate (NFLjSO^ were used to evaluate
the conversion efficiency of the 8400S at each site. Frequency of the tests ranged from monthly
to quarterly. Replicate measurements were made at zero nitrate and four additional nitrate levels
(nominally 60, 120, 180, and 240 ng applied sulfate). Slopes of the measured mass versus
deposited mass were applied as corrections to the data, either as inputs in the 8400S setup under
"Theoretical Conversion Factor", or during post-processing of the data. Zero and span audits of
the SC>2 analyzer were undertaken every few days (e.g., every three days at the Phoenix site).
Average results for each site are shown in Table 7-5. These results for sulfate are significantly
worse than those for nitrate; the theoretical conversion efficiencies are neither near 100%, nor
consistent among the sites, varying from 50% to over 200%. However, the SC>2 gas analyzer
efficiencies, as determined by aqueous standards, are quite high and close to 100%, as is
expected.
The Phoenix and Seattle sites had substantially lower sulfate concentrations than did the
more easterly cities of Houston, Chicago, and Indianapolis.
7-13
-------
Only the Phoenix and Houston sites met the ratio-of-means criterion, but none of the data
from the 8400S monitors met both comparability criteria at any of the five sites. Predictability
criteria for sulfate were met at two of the five sites (Indianapolis and Seattle). The Phoenix
sulfate monitor was operationally problematic. This analyzer was moved from Phoenix to
Houston in June 2004, so the data comparison here with the Phoenix STN data involved only 62
data pairs covering the time period from September 2003 through May 2004 (Figure 5-1). Once
the analyzer was moved to Houston, collocated 8400S measurements were made; however, at the
time of this report, insufficient collocated data were available to analyze for precision estimates.
With the exception of data from the Seattle 8400S monitor, all sulfate regression slopes
were less than or equal to 0.8 with positive intercepts (Table 7-4). No dynamic zero data were
available to apply corrections for instrument offset. Although slopes were comparable between
Chicago, Phoenix, and Indianapolis, the average difference between 8400S and STN sulfate data
(y - x) was more negative for the more eastern cities because of the higher concentrations
measured there. The scatter plot and regression for Chicago (Figure 7-5) were similar to those
for Phoenix and Indianapolis.
20.00 -,
y = 0.5938X + 0.7763
R2 = 0.7285
0.00
0.00
4.00
8.00 12.00
URG Filter-based (ug/m3)
16.00
20.00
Figure 7-5. Twenty-four-hour average R&P 8400S PM2.5 sulfate versus 24-hr
integrated STN filter sulfate at Chicago, May 2, 2002 to May 10, 2005.
In contrast to the monitors at other sites, the Seattle 8400S sulfate monitor measured
more sulfate than the STN filter-based sampler for all but 6 of the 245 data pairs, with a slope for
the comparison of 1.25 ± 0.03 (Figure 7-6). All but a few measured concentrations in Seattle
7-14
-------
were below 3 ug/m3, and the linear fit at these low concentrations was very good. Both the
Seattle (r = 0.94) and the Indianapolis (r = 0.92) sites met the predictability criteria.
URG Filter-based Sulfate (ug/m3)
Figure 7-6. Twenty-four-hour average R&P 8400 S PM2.5 sulfate versus 24-hr
integrated STN filter sulfate at Seattle, May 2, 2002 to February 3, 2005.
7.4 CARBON
That operationally defined measurements of aerosol TC, OC, and EC can yield highly
variable results has been well documented. In a summary of comparison studies of filter
methods to measure OC and EC, EC concentrations were found to differ by up to a factor of 7
among filter-based methods; factor of 2 differences were common (Watson et al., 2005). In the
comparisons discussed in this report, the analytical methods differ substantially in such factors as
the analysis atmosphere (e.g., ambient air versus pure helium), temperature regimes and dwell
times, and correction for pyrolysis of OC. The STN method utilizing TOT is used as the
benchmark, but the criteria for comparability or predictability must be viewed with the
understanding that one analysis method is not necessarily better than another.
7-15
-------
7.4.1 R&P 5400 Semi-Continuous Versus STN Filter-Based Carbon
There are substantial differences in analysis methods between the 5400 and the STN TOT
procedure. The 5400 carbon analysis is undertaken in ambient air, while the TOT procedure
uses pure helium for the OC step and a helium/oxygen mixture for the EC step. There is no
correction for pyrolyzed OC (POC) in the 5400. Thus, any POC that is generated is potentially
added to the EC fraction. Maximum temperatures differ (750°C in the 5400 and 920°C in STN
TOT), and the OC versus EC split temperature is 600°C in TOT versus 275°C in the 5400. The
TOT method converts the generated CO2 to methane and measures it with a flame ionization
detector (FID) while the 5400 employs a non-dispersive infrared (NDIR) detector. That results
are dissimilar is not unexpected.
Of the EC, OC, and TC data from the R&P 5400 at the five sites, only OC and TC data
from the Seattle site met the predictability criteria while the TC data from the Indianapolis site
almost met the predictability criteria. No R&P 5400 EC, OC, and TC data met the comparability
criteria at any site (see Table 7-4).
Slopes for TC ranged from 0.35 to 0.58, with correlation coefficients from 0.78 to 0.92.
TC was underestimated over all concentration ranges, yielding average y/x ratios from 0.52 to
0.83, and average differences (y - x) from -0.71 to -3.89 ug/m"3. Figure 7-7 shows a typical
regression for TC. Scatter plots of the 5400 versus STN TC for the other four sites are shown in
Appendix A.
12.00
0.00
0.00
2.00
4.00 6.00 8.00
URG Filter-based Total Carbon (ug/m3)
10.00
12.00
Figure 7-7. Twenty-four-hour average R&P 5400 PM2 5 total carbon versus 24-hr
integrated STN filter total carbon at Chicago, from May 2, 2002 to May 7, 2005.
7-16
-------
Slopes for OC ranged from 0.33 to 0.54, with correlation coefficients from 0.73 to 0.90.
OC was underestimated over all concentration ranges, yielding average y/x ratios from 0.48 to
0.81, and average differences (y - x) from -0.65 to -3.47 ug/m3. Figure 7-8 depicts the
5400-to-STN relationship for OC in Chicago. Plots for the other locations are shown in
Appendix A.
12.00
y = 0.5447X + 0.3502
R2 = 0.6398
0.00
0.00
2.00
4.00 6.00 8.00
URG Filter-based Organic Carbon (ug/m3)
10.00
12.00
Figure 7-8. Twenty-four-hour average R&P 5400 PM2.5 organic carbon versus 24-
hr integrated STN filter organic carbon at Chicago, May 2, 2002 to May 7, 2005.
Slopes for EC ranged from 0.29 to 0.63, with correlation coefficients from 0.16 to 0.83.
EC was underestimated over all concentration ranges, yielding average y/x ratios from 0.37 to
0.99, and average differences (y - x) from -0.07 to -0.42 ug/m3. A typical regression plot, from
Chicago, is shown in Figure 7-9.
In previous studies comparing data from the 5400 with filter-based aerosol carbon data,
EC was overestimated by the 5400 (Watson and Chow, 2002; Rice, 2004). At the five STN sites
included in the current analysis, the 5400 consistently underestimated the EC fraction, as
reflected in the low slopes, the average y/x ratios all less than 1, and the consistently negative y-x
differences.
7-17
-------
4.00
S- 3.00
"&
c
o
.Q
to
O
I 2.00
HI
01
LU
S
1.00
y = 0.3088X + 0.0344
R2 = 0.2384
0.00
0.00
1.00 2.00 3.00
URG Filter-based elemental Carbon (ug/m3)
4.00
Figure 7-9. Twenty-four-hour average R&P 5400 PM2.5 elemental carbon versus 24-hr
integrated STN filter elemental carbon at Chicago, May 2, 2002 to May 7, 2005.
7.4.2 Sunset Semi-Continuous Versus STN Filter-Based Carbon
A Sunset Laboratory thermal-optical carbon monitor was added to the suite of
instruments at the Chicago, Phoenix, and Seattle sites during 2004. The analysis method is
similar to the STN TOT method—it uses a pure helium atmosphere for the OC step, followed by
a helium/oxygen mixture for the EC step; maximum temperatures are similar (850°C for the
Sunset carbon monitor and 920°C for STN TOT), and both methods have a light transmittance-
based pyrolysis correction. The Sunset carbon monitor used in this comparison study use an
NDIR to directly detect evolved CO2. This method is similar to the laboratory-based STN
analysis method.
The data sets for the Sunset comparisons are considerably smaller than the other data sets
being evaluated here (70 to 90 pairs versus 150 to 300 pairs, see Table 7-1), but still sufficient
for statistical analysis.
The results for the Sunset carbon monitor operated at three of the sites were significantly
better than the results for the R&P 5400—the EC data at one site (Phoenix) met the slope and
intercept comparability criteria; the EC, OC, and TC data from all three sites either met or almost
met the ratio-of-means and correlation comparability criteria; and EC, OC, and TC at all three
sites met the predictability criteria. The comparability evaluations between the Sunset and filter-
7-18
-------
based OC and TC data offer some interesting results and point out the difficulty in assessing
statistical equivalence and comparability.
Initial examination of the regression plots for Chicago (Figures 7-10 through 7-12) and
Seattle (Figures 7-13 through 7-15) shows good agreement between the semi-continuous and
filter-based data. The correlations are all greater than 0.90, and the regression slopes equal unity
within 3 standard errors and/or average y/x ratios equal unity within 1 standard deviation (Table
7-1). However, the TC and OC intercepts differ from zero by more than 3 standard errors,
making the Sunset data not comparable to filter-based data when intercept is a criterion (Table
7-1). How important is the intercept criterion? In the case of the Chicago Sunset TC, the slope
is 0.91 and r = 0.98, and, as the regression plot (Figure 7-10) clearly demonstrates, the
relationship over the range of measured concentrations is qualitatively good. However, under
low ambient concentrations, an intercept of 0.80 ±0.15 imparts significant bias. This may
indicate the need to determine a dynamic zero to adjust for background for the Sunset monitor.
In addition, the Sunset monitor has an organic carbon gas-phase denuder and the STN does not.
It is unclear how the denuder influences these results.
The Phoenix Sunset monitor met the correlation and ratio-of-means comparability criteria
for OC and TC even though the low slopes for the Phoenix TC (Figure 7-16) and OC (Figure
7-17) were considerably outside the acceptable magnitude. Since OC is a major component of
TC, OC concentrations dominated the TC slope, even though EC showed excellent agreement
with the filter-based data (Figure 7-18).
12.00
10.00
CO
o
o
8.00
6.00
4.00
2.00
0.00
y = 0.9358X + 0.8049
R2 = 0.9084
0.00
2.00
4.00 6.00 8.00
URG Filter-based Total carbon (ug/m3)
10.00
12.00
Figure 7-10. Twenty-four-hour average Sunset PM2.5 total carbon versus 24-hr
integrated STN filter total carbon at Chicago, August 1, 2004 to May 7, 2005.
7-19
-------
12.00
y = 0.8432X + 0.8696
R2 = 0.8663
0.00
0.00
2.00
4.00 6.00 8.00
URG Filter-based total carbon (ug/m3)
10.00
12.00
Figure 7-11. Twenty-four-hour average Sunset PM2.5 organic carbon versus 24-hr
integrated STN filter organic carbon at Chicago, August 1, 2004 to May 7, 2005.
4.00
y = 1.1051 x +0.0876
R2 = 0.904
1.00
2.00
URG Filter-based (ug/m3)
3.00
4.00
Figure 7-12. Twenty-four-hour average Sunset PM2.5 elemental carbon versus 24-hr
integrated STN filter elemental carbon at Chicago, August 1, 2004 to May 7, 2005.
7-20
-------
URG Filter-based (ug/m3)
Figure 7-13. Twenty-four-hour average Sunset PM2 5 total carbon versus 24-hr
integrated STN filter total carbon at Seattle, August 13, 2004 to April 10, 2005.
12
10
S 6
I 4
O
y = 0.7911x + 0.7321
R2 = 0.918
468
URG Filter-based organic carbon (ug/m3)
10
12
Figure 7-14. Twenty-four-hour average Sunset PM2.5 organic carbon versus 24-hr
integrated STN filter organic carbon at Seattle, August 13, 2004 to April 10, 2005.
7-21
-------
y = 1.2272x- 0.0274
R2 = 0.8491
1 2
URG Filter-based organic carbon (ug/m3)
Figure 7-15. Twenty-four-hour average Sunset PM2 5 elemental carbon versus 24-hr
integrated STN filter elemental carbon at Seattle, August 13, 2004 to April 10, 2005.
25.00
y = 0.7603X + 0.2221
R2 = 0.8964
0.00
0.00
5.00
10.00 15.00
MetOne Filter-based (ug/m3)
20.00
25.00
Figure 7-16. Twenty-four-hour average Sunset PM2.5 total carbon versus 24-hr
integrated STN filter total carbon at Phoenix, July 2, 2004 to May 13, 2005.
7-22
-------
20.00
16.00
1
1
5
•E
,1 12.00
S>
o
8.00
4.00
0.00
0.00
y = 0.677X + 0.391
R2 = 0.8196
4.00 8.00 12.00
MetOne Filter-based organic carbon (ug/m3)
16.00
20.00
Figure 7-17. Twenty-four-hour average Sunset PM2 5 organic carbon versus 24-hr
integrated STN filter organic carbon at Phoenix, July 2, 2004 to May 13, 2005.
5.00
4.00
c
o
£2
^ 3.00
o 2.00
c
o
o
= 1.00
y=1.0385x + 0.0045
R2 = 0.9488
0.00
0.00
1.00 2.00 3.00
MetOne Filter-based elemental carbon (ug/m3)
4.00
5.00
Figure 7-18. Twenty-four-hour average Sunset PM2.5 elemental carbon versus 24-hr
integrated STN filter elemental carbon at Phoenix, July 2, 2004 to May 13, 2005.
7-23
-------
8. OPERATIONAL ISSUES
A desirable characteristic of semi-continuous speciation monitors is ease of maintenance.
Site operators at the five locations were polled, and their comments about operational issues are
qualitatively summarized here. Prior to the study, Standard Operating Procedures (SOPs),
quality control (QC) checklists, and other assistance documents were jointly prepared by the
study participants.
8.1 R&P 8400N AND 8400S
The 8400N and 8400S instruments have few operational differences beyond those
mentioned in Section 5.2.2, so their operational characteristics are discussed together.
The general impression of the 8400 series operating manuals was that they were poorly
written, lacked coherence, and, in some cases, contained erroneous information. Fortunately, the
SOP and assistance documents were generally complete and well thought out. Site operators
noted that the QC checklists should be continually updated as field experiences were discussed
and evaluated during the field study. One example would be updating acceptable values for the
nitrate flash duration range. Also mentioned was the importance of maintaining consistency in
units between the nitrate and sulfate instrument QC checklists to avoid confusion for staff.
8.1.1 Distribution of Annual Hours
Site operators agreed that, as a very general estimate, about 400 hours were spent
annually on each 8400 series monitor. Data acquisition and reduction alone are estimated to
account for about 44% of all annual hours expended on these instruments, followed by aqueous
standards tests at 14% of total hours. Most of these hours were within weekly or monthly visits,
while PE tests dominated the hours attributed to quarterly tasks. Table 8-1 gives estimates of the
proportion of annual hours allotted to specific tasks. Note that these estimates are for operations,
maintenance, and data processing, and do not include reporting tasks.
8.1.2 Flash Strips
The 8400S exhibited more flash-strip-related problems than did the 8400N, apparently
related to the washers that are used to hold the flash strip in place and the fact that the 8400S
platinum strips are more delicate than the 8400N NiChrome® strips. The 8400S was originally
supplied with round washers meant to hold the flash strips, but R&P replaced these round
washers with square washers. Users reported that the jagged edges of the square washers
apparently caused more strip failures. Both round and square washers are currently in use at the
different sites. Some operators found no difference in the washers, while others strongly disliked
the square ones. R&P was supposed to have distributed a flash strip installation guide tool to all
operators. This tool was intended to hold the square washers in place while retaining nuts were
tightened to avoid stressing the flash strip. One operator who received the tool stated that it was
very helpful, but operators at some sites did not receive the tool. Site operators changed the
8-1
-------
strips nominally once a month and reported that it was important to check the washers at every
flash strip change for black marks or burning, and to replace them if they looked or felt bad. One
site operator stated that constant strip-breaking on the 8400S (every 2-3 days) was a significant
problem, and the instrument was eventually shut down because of this problem.
The posts on which the flash strips are mounted also tend to develop burrs or pits that can
contribute to flash-strip failure, and these posts need to be sanded with a fine emery cloth.
If the flash duration of the 8400S is too short (less than 8 ms), the strips are prone to
"burning up". A flash duration of between 10 and 16 ms will prolong the life of the 8400S flash
strip.
Table 8-1. Percent of annual hours spent on operational and maintenance tasks
for the R&P 8400 series monitors. The annual total for all tasks is estimated at
400 hours for each 8400N or 8400 S monitor.
Maintenance Item or Task
Carrier Gas
Calibration Gas
Water Reservoir
Flash strip
Flash duration
Carbon denuder maintenance
Sample flow rate
Orifice cleaning
Make-up flow and filter check
R-cell pressure adjustments
Zero and span drift
Manual analyzer audits
Cyclone maintenance
Field blanks
Leak checks
Flow checks
Ambient T and P verification
Aqueous standards
Performance evaluations (blind tests)
Data acquisition and reduction issues
Pump Maintenance/ Repair
Sample Line Replacement
Percent of Annual Hours
4.8
4.8
4.0
1.5
3.3
0.1
3.3
1.5
0.3
1.5
4.5
3.0
1.5
0.8
1.5
0.8
0.4
13.5
4.0
43.5
1.3
0.5
8-2
-------
8.1.3 Aqueous Standards
Testing against aqueous standards is time-consuming but necessary. The primary issues
concerning aqueous standards for the 8400 series instruments are related to using varying
volumes of the standard solution. One consequence is that for the larger volumes, there may not
be adequate time for complete evaporation to occur, especially at lower temperatures. This issue
was apparently a larger problem for the 8400S than for the 8400N. One suggested resolution of
this problem was to use multiple concentrations of standards at one low volume (e.g.,
0.2 ul SO4, and 0.5 ul NO3). R&P promised to provide standard solutions in multiple
concentrations, but none of the site operators actually received these standards. PE solutions
provided by NAREL for routine aqueous standards tests are being used at at least one site.
Another suggestion for ensuring complete evaporation of the syringe-applied standard was to
provide a heated flash cell (e.g., 35 to 50° C).
8.1.4 Water Reservoir and Hydration Tube
Some participants commented on problems with discoloration of the hydration tube. One
participant attributed the discoloration to bacteria growing in the humidifier because the water
was not bacteria-free. This operator has been using ultra-pure bacteria-free water to fill the
hydration bottle. Others used distilled water. The true cause of this discoloration is
undetermined. Another possibility is a reaction with the metal fittings, since the discoloration
tends to be on either end of the hydration tube, near the metal "T" fittings. Discoloration of the
hydration tube was commonly reported; but according to R&P, the ability of the permeation
tubing to hydrate the sample air is not affected.
8.1.5 Orifice Cleaning
Although the 8400 series manuals recommend monthly intervals between sample orifice
cleaning, site operators reported that weekly schedules (or at each site visit) are more apt to
minimize flow rate problems.
8.1.6 Molybdenum Converter
The molybdenum converter in the 8400N instruments was evaluated and replaced during
2004. The efficiency of the "moly chips" that convert the nitrogen species to NO under high
temperatures is reduced over time, and it is important to periodically check the conversion
efficiency. There is some uncertainty about which nitrogen species should be used to check the
converter efficiency. As discussed above, the effect of the converter replacement on the
monitors' efficiencies, as measured by the periodic aqueous standards tests, varied among sites.
8-3
-------
8.1.7 Data Screen Failures
Some sites experienced "data screen failures" due to failures in the backlight controller.
R&P delivered an upgrade kit that involves the installation of a "piggy-back" board on the
existing backlight controller.
8.1.8 Data Acquisition and Reduction
Methods of data acquisition and reduction for the semi-continuous monitors differed
among sites. Some sites logged analog data and other sites logged digital data. All site operators
indicated that viewing the data every day, often via remote access, was a valuable activity. The
operators viewed the data to determine if the instrument was operating, and operating properly,
to evaluate various instrument operating parameters that are provided on these new instruments
using serial outputs and to identify major problems. The review and evaluation of important
operational data from these monitors, such as flash duration, flow rate, and R-cell pressures, was
valuable.
Several different systems were used to view the data on a daily basis. The Indiana site
operator did a quick, daily review of the data through a special MeteoStar LEADS web site. The
Texas operator performed his daily review via the Texas Commission on Environmental Quality
(TCEQ) web site, which also uses MeteoStar LEADS, but the operator used an on-site PC to
collect the data that were validated and used for reporting. The Illinois operators do not use the
MeteoStar system. At the Phoenix site, an STI data acquisition system was used to collect the
serial data. The data were then automatically screened against minimum validation criteria,
plotted and made available on a web site and archived.
8.2 R&P 5400
Site operators agreed that, as a very general estimate, about 150 hours were spent
annually on the R&P 5400 OCEC monitor. Data acquisition and reduction alone accounted for
more than 50% of all annual hours spent on this instrument. Monthly and semi-annual checks
and repairs of furnace lamps, afterburner lamps, and oven fuses combined to account for an
additional 20% of annual hours. Table 8-2 shows estimates of the proportion of annual hours
allotted to specific tasks. Note that these estimates are for operations, maintenance, and data
processing but do not include subsequent reporting tasks.
8-4
-------
Table 8-2. Percent of annual hours spent on operational and maintenance tasks
for the R&P 5400 OCEC monitor. The annual total for all tasks are estimated at
150 hours for each 5400 monitor.
Maintenance Item or Task
Flow audit
Collection path leak test
Analysis loop leak test
Licor calibrations
Furnace burners/afterburners fuses (monthly)
Inlet cleaning
Intake filters (cooling fans)
In-line filters
Data acquisition and reduction issues
Collector replacement
Furnace Lamp Checks/ Replacement (semi-annually)
Pump Maintenance/ Replacement
Percent of Annual Hours
3.9
3.9
2.0
3.3
9.8
11.8
0.7
0.7
52.3
5.2
10.5
2.0
Flow problems, along with oven and afterburner problems, were the most frequent and
recurring operational issues with the R&P 5400. It had been known that the sample flow through
the collectors in the 5400 monitor tend to degrade over time, requiring replacement of the
collectors, and this degradation was observed during the course of the Study. The collectors are
expensive (~ $1,200 per pair), and it is recommended that both collectors be replaced whenever
one of them starts to become clogged. Failure to reach set point temperatures was most
frequently caused by blown fuses or burnt out oven lamps or afterburner lamps. The contact
points for the lamps were also reported to have melted on occasion.
The 5400 monitor in Phoenix experienced leak problems. Although regular leak checks
yielded apparently acceptable results, evidence exists that a leak may have existed and gone
undetected for an extended period of time. The hypothesis is that the pressure sensor in the
optical bench has a maximum range and that this range was being exceeded during the manual
leak tests. The R&P recommendation is to achieve a pressure of about 1200 mb for this test.
Empirically, it appears that the pressure sensor tops out at 1206 mb. If this pressure were to be
exceeded, a pressure drop could be occurring during the leak test but would not be detected. On
February 4, 2005, after both sides of the analyzer failed an analysis loop leak check while a
manual leak check was passed, this hypothesis was tested. A manual leak check using a
maximum pressure of 1180 mb was initiated, a pressure drop was noted immediately, and the
leak was isolated at the septum/buffer tank by pinching off the line above the buffer tank. This
action stopped the pressure drop, and the septum fittings were replaced. It is recommended that
all manual leak checks use a maximum pressure of 1180 mb. The rated maximum pressure of
the optical bench pressure sensor is not known.
In Phoenix, the replacement of some hardware in the Licor NDIR component in
September 2004 caused false status codes during the auto-calibration procedure. R&P's
8-5
-------
recommendation was to disable the auto-calibration and auto-leak check functions. This step
increased the duties of the site operators because these checks had to be performed manually.
Some site operators felt that the data from the R&P 5400 were so disappointing that they
did not want to waste their time discussing operational issues. Most of the above issues are
based on comments in the Phoenix site log.
8.3 SUNSET CARBON MONITOR
Generally, site operator experiences with the Sunset OCEC analyzer have been good. The
Sunset analyzer has no analog output, so users without serial datalogging capability are at a
disadvantage. This characteristic combined with a failure in the laptop PC control system to
cause massive (one month) data loss at the Chicago site, which apparently had only analog
capabilities. A hard-drive failure in the laptop occurred one day before the month's data were
due to be downloaded. A loss of six days' data occurred at the same site when the laptop locked
up just after a Thursday visit by the site operator and was not rebooted until the next visit the
following Tuesday. All site operators reported problems with the laptop computer at some point.
The burner coils in the Sunset have to be replaced periodically. The first report of burner
failures came from Phoenix in August 2005, when both front and rear burners failed within a few
weeks of each other.
Over the past year, the three Sunset monitors in the evaluation differed in their analysis
software. All comparisons presented here have been based on a recalculation of the
concentrations from the raw data, employing the most recent version of the analysis software. It
will be important for the manufacturer to ensure that the deployed instruments have up-to-date
versions of software and employ identical procedures.
Although the audit punches in the PE were easy to complete, the manual integration of
the peaks was tedious. Modification of the software to enable the field results file to duplicate
the post-calculated values would be helpful.
A few comments noted that the Sunset manual was inadequate to acquaint an operator
with operation of the instrument. While covering the instrument setup routines fairly well, the
manual offers minimal information on the theory of operation and data output interpretation.
There are no schematics or illustrations. More clarification of and details about filter
replacement procedures would be helpful. The inclusion of the NIOSH laboratory SOP in the
manual is confusing because of differences between the laboratory and field instruments (FID
versus NDIR, for example).
No sucrose spikes were ever applied by the site operators, except for those provided by
NAREL for the PEs. This action would provide useful checks on the monitors' performance.
8-6
-------
9. REFERENCES
Bae M.S., Schauer J.J., DeMinter J.T., Turner J.R., Smith D., and Gary R.A. (2004) Validation
of a semi-continuous instrument for elemental carbon and organic carbon using a
thermal-optical method. Atmos. Environ. 38 (18), 2885-2893.
Drewnick F., Schwab J. J., Hogrefe O., Peters S., Husain L., Diamond D., Weber R., and
Demerjian K.L. (2003) Intercomparison and evaluation of four semi-continuous PM2.5
sulfate instruments. Atmos. Environ. 37 (24), 3335-3350.
Harrison D., Park S.S., Ondov J., Buckley T., Kim S.R., and Jayanty R.K.M. (2004) Highly time
resolved fine particle nitrate measurements at the Baltimore Supersite. Atmos. Environ.
38, 5321-5332.
Hering S.V. and Stolzenburg M.R. (1998) A new method for the automated high-time resolution
measurement of PM2.5 nitrate. In PM2.s: a fine particle standard, J. Chow and P.
Koutrakis eds., Air and Waste Management Association, 312-317.
Hogrefe O., Schwab J., Drewnick F., Lala G., Peters S., and Demerjiam K. (2004)
Semicontinuous PM2.5 sulfate and nitrate measurements at an urban and a rural location
in New York: PMTACS-NY summer 2001 and 2002 campaigns. J. Air & Waste Manag.
Assoc. (54), 1040-1060.
Hyslop N.P., Simon E.M., and Roberts P.T. (2004) Operation of Rupprecht and Patashnick
(R&P) continuous PM2.5 speciation samplers at the Phoenix Supersite. Final report
prepared for the Arizona Department of Environmental Quality, Phoenix, AZ, by Sonoma
Technology, Inc., Petaluma, CA, STI-903603-2481-FR, February.
Koutrakis P. (1999) Recommendations of the expert panel on the EPA speciation network, final
summary. Available on the Internet at .
Lim H.-J., Turpin B.J., Edgerton E., Hering S.V., Allen G.A., Maring H., and Solomon P. (2003)
Semicontinuous aerosol carbon measurements: comparison of Atlanta supersite
measurements. J. Geophys. Res. 108 (D7), 8419.
Liu D.-Y., Prather K.A., and Hering S.V. (2000) Variations in the size and chemical composition
of nitrate-containing particles in Riverside, CA. Aerosol Sci. Technol. 33, 71-86.
NIOSH (1999) Method 5040: Elemental Carbon (Diesel Exhaust). In NIOSHManual of
Analytical Methods, 4th Ed., Issue 3, National Institute of Occupational Safety and
Health, Cincinnati, OH.
Reid C.S., Turner J.R., and Hering S.V. (2005) Fine particulate matter nitrate measurements by
flash volatilization: results from the St. Louis-Midwest Supersite. From the Air & Waste
Manage Association Symposium on air quality measurement methods and technology,
San Francisco, CA, April 19-21 (Paper No. 37).
Rice J. (2004) Comparison of integrated filter and automated carbon aerosol measurements at
Research Triangle Park, North Carolina. Aerosol Science & Technology (38), 23-36.
Rice J. (2005) U.S. Environmental Protection Agency, Research Triangle Park, NC. Personal
communication.
9-1
-------
Schauer J.J., Mader B.T., Deminter J.T., Heidemann G., Bae M.S., Seinfeld J.H., Flagan R.C.,
Gary R.A., Smith D., Huebert B.J., Bertram T., Howell S., Kline J.T., Quinn P., Bates T.,
Turpin B., Lim H.J., Yu J.Z., Yang H., and Keywood M.D. (2003) ACE-Asia
intercomparison of a thermal-optical method for the determination of particle-phase
organic and elemental carbon. Environ. Sci. Technol. 37 (5), 993-1001.
Solomon P.A., Mitchell W., Tolocka M., Norris G., Gemmill D., Wiener R., Vanderpool R.,
Murdoch R., Natarajan S., and Hardison E. (2000) Evaluation of PM2.5 chemical
speciation samplers for use in the EPA national PM2 5 chemical speciation network, final
report. Available on the Internet at .
Stolzenburg M.R. and Hering S.V. (2000) Method for the automated measurement of fine
particle nitrate in the atmosphere. Environ. Sci. Technol. 34 (5), 907-914.
Taylor S. (2005) Performance evaluation of R&P 8400 and Sunset Labs ambient air monitors.
Personal communication, National Air and Radiation Environmental Laboratory, U.S.
Environmental Protection Agency, Research Triangle Park, NC (unpublished
memorandum), August.
Turpin B.J., CaryR.A., and Huntzicker J.J. (1990) Aerosol Sci. Technol. 12, 161-171.
U.S. EPA (2002) Data Quality Objectives (DQOs) for Relating Federal Reference (FRM) and
Continuous PM2.5 Measurements to Report and Air Quality Index (AQI). Office of Air
Quality Planning and Standards, Research Triangle Park, NC; November 2002.
Watson J.G. and Chow J.C. (2002) Comparison and evaluation of in situ and filter carbon
measurements at the Fresno Supersite. J. Geophys. Res. 107 (D21) (ICC3-1-ICC3-15).
Watson J.G., Chow J.C., and Chen L.-W.A. (2005) Summary of organic and elemental
carbon/black carbon analysis methods and intercomparisons. Aerosol and Air Quality 5
(1), 65-102.
Weber R., Orsini D., Duan Y., Baumann K., Kiang C.S., Chameides W., Lee Y.N., Brechtel F.,
Klotz P., Jongejan P., ten Brink H., Slanina J., Boring C.B., Genfa Z., Dasgupta P.,
Hering S., Stolzenburg M., Butcher D.D., Edgerton E., Hartsell B., Solomon P., and
Tanner R. (2003) Intercomparison of near real time monitors of PM2.5 nitrate and sulfate
at the U.S. Environmental Protection Agency Atlanta Supersite. Journal of Geophysical
Research-Atmospheres 108 (D7).
Wittig A.E., Takahama S., Khlystov A., Pandis S.N., Hering S., Kirby B., and Davidson C.
(2004) Semi-continuous PM2 5 inorganic composition measurements during the
Pittsburgh Air Quality Study. Atmos. Environ. 38, 3201-3213.
9-2
-------
APPENDIX A
SCATTER PLOTS AND REGRESSIONS
FOR PHOENIX, SEATTLE, AND INDIANAPOLIS
A-l
-------
A.1 NITRATE
20
0)16
•z.
O
co
-------
20 -,
"3)
. 16
O
O
One-to-one
A Pre-moly
O Post-moly
D all data
— ™ Best fit (post-moly)
™ ™ Best fit (pre-moly)
•Best fit all data)
Pre-moly
y = 0.6221 x + 0.3928
R2 = 0.8036
All data
y = 0.5851 x +0.4065
R2 = 0.8137
Post-moly
y = 0.5203X + 0.3974
R2 = 0.9002
12
16
20
MetOne Filter-based Nitrate (|jg/m3)
Figure A-3. Twenty-four-hour average R&P 8400N PM2.5 nitrate versus 24-hr
integrated STN filter nitrate at Indianapolis, from June 13, 2002, to March 11, 2005.
A.2 SULFATE
4.00
y = 0.7622x +0.1785
R2 = 0.5028
0.00
0.00
1.00 2.00
Met One Filter-based Sulfate (ug/m3)
3.00
4.00
Figure A-4. Twenty-four-hour average R&P 8400S PM2.5 sulfate versus 24-hr
integrated STN filter sulfate at Phoenix, from September 30, 2003, to May 12, 2004.
A-3
-------
30
25
20
15
y = 0.6604X + 0.4971
R2 = 0.866
10
10 15 20
Met One Filter-based Sulfate (ug/m3)
25
30
Figure A-5. Twenty-four-hour average R&P 8400S PM2.5 sulfate versus 24-hr
integrated STN filter sulfate at Indianapolis, from June 19, 2002, to July 8, 2004.
y = O.SOSSx + 0.4029
R2 = 0.5485
URG Filter-based Sulfate (ug/m3)
Figure A-6. Twenty-four-hour average R&P 8400S PM2 5 sulfate versus 24-hr
integrated STN filter sulfate at Houston, from January 27, 2003, to April 13, 2005.
A-4
-------
A.3 R&P 5400 VERSUS STN CARBON
A.3.1 Total Carbon
25.00
0.00
0.00
5.00
10.00 15.00
Met One Filter-based Total Carbon (ug/m3)
20.00
25.00
Figure A-7. Twenty-four-hour average R&P 5400 PM2 5 total carbon versus 24-hr integrated
STN filter total carbon at Phoenix, from December 13, 2002, to May 13, 2005.
20 i
URG Filter-based Total Carbon (ug/m3)
Figure A-8. Twenty-four-hour average R&P 5400 PM2.5 total carbon versus 24-hr
integrated STN filter total carbon at Seattle, from May 2, 2002, to March 29, 2005.
A-5
-------
Met One Filter-based Total Carbon (ug/m3)
Figure A-9. Twenty-four-hour average R&P 5400 PM2 5 total carbon versus
24-hr integrated STN filter total carbon at Indianapolis, from June 26, 2002, to
April 9, 2004.
16 -i
y = 0.4623X + 0.8104
R2 = 0.5562
URG Filter-based Total Carbon (ug/m3)
Figure A-10. Twenty-four-hour average R&P 5400 PM2.5 total carbon versus
24-hr integrated STN filter total carbon at Houston, from January 24, 2003, to
March 26, 2005.
A-6
-------
A.3.2 Organic Carbon
20.00
16.00
12.00
o
a.
8.00
4.00
0.00
y = 0.3292x + 0.7216
R2 = 0.6442
0.00
4.00 8.00 12.00 16.00
Met One Filter-based Organic Carbon (ug/m3)
20.00
Figure A-l 1. Twenty-four-hour average R&P 5400 PM2.5 organic carbon versus
24-hr integrated STN filter organic carbon at Phoenix, from December 13, 2002, to
May 13, 2005.
y = 0.5333X + 0.569
R2 = 0.8039
4 6
URG Filter-based Organic Carbon (ug/m3)
Figure A-12. Twenty-four-hour average R&P 5400 PM2.5 organic carbon versus
24-hr integrated STN filter organic carbon at Seattle, from May 2, 2002, to March 29,
2005.
A-7
-------
20 -,
y = 0.5001 x +0.5906
R2 = 0.7315
4 8 12
Met One Filter-based Organic Carbon (ug/m3)
16
20
Figure A-13. Twenty-four-hour average R&P 5400 PM2.5 organic carbon versus
24-hr integrated STN filter organic carbon at Indianapolis, from June 26, 2002, to
April 9, 2004.
12 i
g
s
S1
o
a.
*
y = 0.4665X + 0.6502
R2 = 0.5386
URG Filter-based Organic Carbon (ug/m3)
Figure A-14. Twenty-four-hour average R&P 5400 PM2.5 organic carbon versus
24-hr integrated STN filter organic carbon at Houston, from January 24, 2003, to
March 26, 2005.
A-8
-------
A.3.3 Elemental Carbon
5.00
4.00
3.00
2.00
m
a.
2
1.00
0.00
y = 0.4005X + 0.2763
R2 = 0.6357
0.00
1.00 2.00 3.00 4.00
Met One Filter-based Elemental Carbon (ug/m3)
5.00
Figure A-15. Twenty-four-hour average R&P 5400 PM2.5 elemental carbon versus
24-hr integrated STN filter elemental carbon at Phoenix, from December 13, 2002, to
May 13, 2005.
1 2 3
URG Filter-based Elemental Carbon (ug/m3)
Figure A-16. Twenty-four-hour average R&P 5400 PM2.5 elemental carbon versus
24-hr integrated STN filter elemental carbon at Seattle, from May 2, 2002, to
March 29, 2005.
A-9
-------
3 i
m
g
S 1
a.
y = 0.3941 x +0.0278
R2 = 0.4255
1 2
Met One Filter-based Elemental Carbon (ug/m3)
Figure A-17. Twenty-four-hour average R&P 5400 PM2.5 elemental carbon versus
24-hr integrated STN filter elemental carbon at Indianapolis, from June 26, 2002, to
April 9, 2004.
n
~u>
c
o
re
O
LU
o
I1
Q.
y = 0.2851 x +0.2054
R2 = 0.0267
1 2
URG Filter-based Elemental Carbon (ug/m3)
Figure A-18. Twenty-four-hour average R&P 5400 PM2.5 elemental carbon versus
24-hr integrated STN filter elemental carbon at Houston, from January 24, 2003, to
March 26, 2005.
A-10
------- |