United States
Environmental Protection
Agency
Office of Air Quality
Planning and Standards
Research Triangle Park, NC 27711
EPA-454/R-01-008
DATE June 2001
Air
vvEPA
Final Report
Evaluation of PM2 5 Speciation Sampler
Performance and Related Sample
Collection and Stability Issues
-------
FINAL REPORT
Evaluation of PM2 5 Speciation Sampler Performance and
Related Sample Collection and Stability Issues
for
U.S. ENVIRONMENTAL PROTECTION AGENCY
Office of Air Quality Planning and Standards
Emissions, Monitoring, and Analysis Division (MD-14)
Research Triangle Park, North Carolina 27711
Jim Homolya, EPA Work Assignment Manager
Vickie Presnell, EPA Project Officer
Contract No. 68-D-98-O3O
Work Assignment 4-O6
April 27, 2OO1
Prepared by
Basil Coutant and Shannon Stetzer
BATTELLE
505 King Avenue
Columbus, Ohio 43201-2693
April 27,
2001
-------
EPA DISCLAIMER
The information in this document has been funded wholly or in part by the United States
Environmental Protection Agency under Contract 68-D-98-030 to Battelle Memorial
Institute. It has been subject to the Agency's peer and administrative review, and it has
been approved for publication as an EPA document. Mention of trade manes or
commercial products does not constitute endorsement or recommendation for use.
BATTELLE DISCLAIMER
This report is a work prepared for the United States Environmental Protection Agency
by Battelle Memorial Institute. In no event shall either the United States Environmental
Protection Agency or Battelle Memorial Institute have any responsibility or liability for
any consequences of any use, misuse, inability to use, or reliance upon the information
contained herein, nor does either warrant or otherwise represent in any way the
accuracy, adequacy, efficacy, or applicability of the contents hereof.
April 27,
2001
-------
TABLE OF CONTENTS
EXECUTIVE SUMMARY vi
1 .0 INTRODUCTION 1
2.0 OVERVIEW OF THE DATA VALIDATION AND EXPLORATORY ANALYSES 4
2.1 DATA VALIDATION AND FLAGGING 4
2.2 EXPLORATORY ANALYSES 5
3.0 ANALYSES OF THE ROUTINE DATA 11
3.1 THE MEASURED PM2 5 MASS CONCENTRATION 13
3.2 ROUTINE PARAMETER MEASUREMENTS 2O
3.3 ROUTINE PARAMETER MEASUREMENTS RELATIVE THE MEASURED
MASS 27
3.4 RELATIVE PRECISIONS OF THE SAMPLER TYPES 28
4.0 ANALYSIS OF THE BLANKS 26
5.0 EXPERIMENTS TO SIMULATE AND TEST POTENTIAL SAMPLE
INTEGRITY ISSUES WHEN USING SEQUENTIAL SPECIATION SAMPLERS 32
5.1 COLLECTION OF VOLATILE ORGANIC COMPOUNDS ON BLANK
QUARTZ
FILTERS 32
5.2 COLLECTION OF VOLATILE ORGANIC COMPOUNDS ON EXPOSED
QUARTZ FILTERS 34
5.3 TESTING THE EFFECTS OF FACE VELOCITY ON THE COLLECTION
OF VOLATILE ORGANIC COMPOUNDS ON QUARTZ FILTERS 37
5.4 COLLECTION OF AMBIENT NITRATE AND SULFATE ON BLANK NYLON
FILTERS 42
5.5 COLLECTION OF AMBIENT NITRATE AND SULFATE ON EXPOSED
NYLON
FILTERS 44
5.6 DISCUSSION 47
6.0 SYNTHESIS 5O
APPENDIX A: GUIDE TO THE GRAPHICAL OUTPUT FROM TASKS 1 AND 2 A-1
A.1 EXPLORATORY PLOTS A-2
A.2 DATABASE DICTIONARY A-1 5
A.3 STEPS TAKEN TO PRODUCE GRAPHS A-1 7
Outline for the m2mc graphs containing the Routine and FRM data . . . A-1 7
Outline for the m2mc graphs containing the FIELD and TRI P BLANK
data A-19
Outline for the boxplots containing the Routine and FRM data A-2O
Outline for the log_FRM plots containing the Routine and FRM data . . . A-21
Outline for the cddvspress and cddvstemp plots containing the Routine
and FRM data A-22
Outline for the cdpress and cdtemp plots containing the Routine and
FRM data A-23
April 27,
iii 2001
-------
Outline for the parmbp and parmbps plots containing the Routine and
FRMdata A-24
Outline for the mec plots containing the Routine and FRM data A-25
Outline for the vmvstmc plots containing the Routine and FRM data . . . A-26
Outline for the smvstmc plots containing the Routine and FRM data . . . A-27
APPENDIX B: TABLES OF SUMMARY STATISTICS B-1
B.1 Median Relative Differences between Samplers for each Site and
Parameter B-2
B.2 Mean Differences Between Sampler Types for Each Site and Parameter . . B-9
B.3 Sampler Type Means for all Sites and Parameters B-1 7
B.4 Significance of the Difference in Relative Composition of the Mass
Constituents by Site B-27
LIST OF TABLES
Table 1.1 Sampling Sites and Samplers 2
Table 1.2 Task 1 and 2 Study Parameters 2
Table 3.1 FRM to Speciation Sampler Regression Results, Sorted by the Ratio of
the
Mean Mass to the Mean FRM Mass 16
Table 3.2 FRM to Speciation Sampler Regression Results, Sorted by the R-Squared
Value 17
Table 3.3 P-values for Tests of Significant Differences Among Sites, Sampler
Types, and All Site-Sampler Type Combinations 21
Table 3.4 Probabilities of Sampler Type a Yielding Values Greater than Sampler
Type
B P-values for the Significance of Sampler Type and Site 26
Table 3.5 Significance of the differences in the relative amounts of Organic Carbon
at each site 28
Table 3.6 Estimated Variance Components for Each Parameter 29
Table 4.1 Summary of the Blank Data 28
Table 4.2 Modeling Results for the Blank Data 3O
Table 5.1 Mean Difference Between Standard Collection of Samples and Those
Left in the Sampler 35
Table 5.2 Regression Results for Modeling the High Volume Concentrations
Against the Low Volume Concentrations 39
Table 5.3 Regression Results for Modeling the High Volume Concentrations
Against the Low Volume Concentrations on the Log Scale 41
Table 5.4 Mean Nitrate and Sulfate on the Experiment Blanks 44
Table 5.5 Mean Difference Between Standard Collection of Samples and Those
Left in the Sampler 45
LIST OF FIGURES
Figure 3.1 Ratio of Mean Measured Mass to Mean Mass From a Co-located FRM ... 18
Figure 3.2 The Relative Deviations of the Ratio of the Measured Mass to the Mean
FRM Mass (the plotted values show the ratios minus the site mean) 18
Figure 3.3 The Linear Regression R-squared with the Co-located FRM 19
Figure 3.4 Speciation Sampler Measured Mass Versus a Co-located FRM
Measurement 19
IV
April 27,
2001
-------
Figure 3.5 Log OC Concentrations at Bismarck, ND Plotted Against the Mean of the
Logs of the Concentration 22
Figure 3.6 Log OC Concentrations at Tampa, FL Plotted Against the Mean of the
Logs of the Concentrations 23
Figure 3.7 Box Plots of the Daily Range in the Logs of the OC Concentrations 24
Figure 4.1 The Proportion of Field and Trip Blanks with Silicon Measurements
Greater than Q3 by Sampler Type 31
Figure 4.2 The Proportion of Field and Trip Blanks with Chlorine Measurements
Greater than Q3 by Sampler Type 31
Figure 5.1 Effects of Leaving Blank Filters in the Sampler for a Week on the
Measured OC 33
Figure 5.2 Effects of Leaving Blank Filters in the Sampler for a Week on the
Measured EC 33
Figure 5.3 The Effects of Leaving a Filter in a Sampler for a Week on the Observed
EC Concentration. Note That the Two Unusual Values Are in Opposite
Directions 36
Figure 5.4 The Effects of Leaving a Filter in a Sampler for a Week on the Observed
OC Concentration 36
Figure 5.5 The Effects of Leaving a Filter in a Sampler for a Week on the Observed
Ratio of the EC Concentration to the OC Concentration 37
Figure 5.6 High Volume Elemental Carbon Concentrations Versus Low Volume
Elemental Carbon Concentrations with a 1-1 line 38
Figure 5.7 High Volume Organic Carbon Concentrations Versus Low Volume
Organic Carbon Concentrations with a 1 -1 line 38
Figure 5.8 High Volume EC Data Versus Low Volume EC Data on the Log Scale 4O
Figure 5.9 High Volume OC Data Versus Low Volume OC Data on the Log Scale .... 4O
Figure 5.1O Log OC Concentrations from High and Low Volume Samples Plotted
Against the Mean of the Logs of the Concentrations 42
Figure 5.11 Effects of Leaving Blank Filters in the Sampler for a Week on the
Measured Nitrate 43
Figure 5.12 Effects of Leaving Blank Filters in the Sampler for a Week on the
Measured Sulfate 43
Figure 5.13 The Effects of Leaving a Filter in a Sampler for a Week on the Observed
Nitrate Concentration 46
Figure 5.14 The Effects of Leaving a Filter in a Sampler for a Week on the Observed
Sulfate Concentration 46
Figure 5.15 The Effects of Leaving a Filter in a Sampler for a Week on the Observed
Ratio of the Nitrate Concentration to Sulfate Concentration 47
April 27,
2001
-------
EXECUTIVE SUMMARY
PM2.5 chemical speciation sampling is included in the monitoring requirements promulgated as
part of the PM2 5 National Ambient Air Quality Standards. Under this requirement EPA will establish a
PM2 5 chemical speciation network of approximately 50 core NAME for routine speciation monitoring.
This network will be used to provide a nationally consistent set of data for the assessment of trends and
provide long term characterization of PM constituents. This network will also be used as a model for
up to 250 additional speciation sites established by the States for information that may aid the
development of State Implementation Plans. A vital consideration for these PM speciation monitoring
sites is data comparability. EPA initially chose three samplers for consideration under the National
Sampler Contract. The data under consideration in this study were collected from February 2000
through July 2000 at 13 sites across the nation with co-located samplers, in an effort to evaluate the
consistency of these three samplers. Further data were collected in August 2000 to examine some
special issues.
The analysis results detailed in this report are the end result of three important efforts. First, the
data underwent a careful screening for outliers, or unusual data, so that results would not be skewed by
these values. Next, considerable effort was put into graphical analysis of the data to determine what
factors should be considered in the assessment of data comparability. (The results of these first two
efforts are detailed in the appendices.) The third effort detailed in this report was statistical modeling
based on the outcomes of the first two efforts.
The following are some of the major findings:
Within a site the data show fairly consistent biases between co-located samplers on a
log scale.
April 27,
vi 2001
-------
There are significant site to site differences in: the number of days with outliers, the
variability of parameters, the relationship between samplers, etc. These first two points
are assumed in the remaining items.
The measured PM2.5 mass was compared with co-located FRM measurements.
Seventeen out of 24 of the samplers met the Expert Panel data objective of an R2 value
of at least 0.9 in a linear regression of the mass values against the FRM measurement.
Deviations from this criteria appear to be caused by site influences that affect all the
monitors at a site, rather than differences among sampler types.
Only half of the samplers met the Expert Panel data objective that the ratio of the FRM
mass mean to the speciation sampler mass mean be at least 0.9 and at most 1.1. The
ratios tested strongly dependent on both site and sampler type. In all cases with co-
located FRMs, the means for the mass followed the following ordering: URG <
Andersen < MetOne. Six of the seven URG means were less than the corresponding
FRM mean, all eight MetOne means were greater than the corresponding FRM mean,
and six of the nine Andersen means were greater than the corresponding FRM mean.
The concentration ordering noted for the mass applies to most of the species, namely
URG < Andersen < MetOne. Moreover, while parameter specific, the percent of the
time that this relation holds is consistent across sites. The exceptions to this ordering
are chlorine, zinc, ammonium, and sulfate. For each of these exceptions, the percent of
the time that the sampler types have one relationship or another varies by site. Of the
species that do follow the general ordering above, only the nitrate data showed site to
site differences in the percent of the time one sampler type is above another.
April 27,
vii 2001
-------
For all species the magnitude of the biases between sampler types is strongly site
dependent from a statistical point of view. The magnitudes are summarized in the
appendix by site and species so that the practical significance can be assessed.
The variability found in the sampling precision across sampler types is probably due to
site influences, but is probably not generally of any practical concern.
The blanks generally do not show site to site differences. The trip blanks and field
blanks are generally about the same. The URG blanks tend to be the cleanest except
for nitrate. Nearly all of the "dirtiest" 25 percent of the blanks were from the URG
samplers. The practical difference among the sampler types needs to be assessed
separately.
The five special experiments all suffer from a lack of data. As such, modeling results
are not robust against the inclusion or exclusion of outliers. Assuming that the outliers
have been properly identified, then there is little or no significant effect on sulfate,
nitrate, elemental carbon, or organic carbon concentrations found with leaving filters in
the sampler for an extended period either before or after sampling. The only statistically
significant difference found was that blanks left for a week in the sampler collected on
average an extra 3.25 micrograms of organic carbon (before sampling).
Qualitatively, both the face velocity experiment and the sampler to sampler comparisons
suggest that measurements of carbon from low volume sampling yield higher
concentrations than high volume sampling. The data from the sampler to sampler
comparisons showed more consistency than the data from the face velocity experiment.
(Carbon is measured from the Quartz filter. In the sampler to sampler comparisons the
quartz filter flow rates were 16.7 1pm for the URG, 7.3 1pm for the Andersen, and 6.7
April 27,
viii 2OO1
-------
1pm for the MetOne. The face velocity experiment used Andersen samplers with flow
rates of 7.3 1pm and 16.7 1pm.)
April 27,
ix 2001
-------
1.O INTRODUCTION
Chemical speciation of PM2 5 is included in the monitoring requirements set forth in the Federal
Register (62 FR 38763). As a result EPA will establish a PM2 5 chemical speciation network of
approximately 50 NAMS, the "Trends" network, for routine speciation monitoring to provide nationally
consistent data for the assessment of trends and long-term characterization of the metals, ions, and
carbon constituents of PM2 5. This network will be a model for the chemical speciation efforts of the
States, up to 250 SLAMS, to be used for State Implementation Plans (SIPS).
Data comparability is a primary concern for the Trends network. A multi-sampler field
evaluation study of co-located samplers was conducted from February 2000 through July 2000, with
additional special studies in August of 2000, to evaluate consistency (referred to as the mini-Trends
network). This report details the analyses to assess comparability of the various methods. The work
was divided into three tasks corresponding to the data involved.
The first task worked with the routine monitoring data collected from February 2000 through
July 2000 at 13 sites with co-located speciation samplers. Table 1.1 shows the site locations and the
number of each sampler type at those locations. Table 1.2 shows the 14 species examined in the
study.
April 27,
2001
-------
Table 1.1 Tasks 1 and 2 Sampling Sites and Samplers
Region
1
2
3
4
5
6
7
8
9
10
State
MA
NY
PA
FL
IL
TX
MO
ND
UT
AZ
CA
OR
WA
Site
Boston (Roxbury)
Bronx Garden
Philadelphia
Tampa-St. Petersburg
Chicago
Houston
St. Louis (Blair St.)
Bismarck
Salt Lake City
Phoenix
Fresno
Portland-Vancouver
Seattle (Beacon Hill)
Number of Samplers
MetOne
2
1
1
1
1
1
1
1
1
Andersen
2
1
1
1
1
1
1
1
1
URG
1
1
1
1
1
1
2
1
Start Date
2/9/2000
2/9/2000
2/9/2000
2/9/2000
2/9/2000
2/9/2000
2/9/2000
2/9/2000
2/9/2000
2/9/2000
2/9/2000
2/9/2000
2/9/2000
Table 1.2 Task 1 and 2 Study Parameters
Study Parameters
1.
3.
5.
7.
9.
11.
13.
PM25 Mass
Aluminum
Calcium
Chlorine
Iron
Lead
Tin
2.
4.
6.
Silicon
Zinc
Ammonium
8. Organic Carbon
10.
12.
14.
Nitrate
Elemental Carbon
Sulfate
The second task examined the field and trip blanks associated with these data and was
also restricted to the parameters listed in Table 1.2.
April 27,
2001
-------
Finally, the third task examined the results from five experiments run in August 2000 to
investigate issues related to sequential sampling and sample collection methods. These special studies
were restricted to either organic and elemental carbon or sulfate and nitrate.
The analyses generally followed three phases: data validation and outlier detection, mostly
graphical exploratory analyses, and statistical modeling and testing. The methods and results of the first
two phases as applied to Tasks 1 and 2 are briefly discussed in Section 2.0, with additional details
given in the appendices. (See the accompanying CD for the collection of the main graphical output.)
Section 3 examines the results of the modeling phase for Task 1 data, the routine data from the 13 sites.
Section 4 examines the modeling results from the Task 2 data, the field and trip blanks collected with
the routine data. Section 5 examines each of the five special experiments and ends with an overall
assessment based on those experiments. Section 6 gives some guidance toward combining all of the
results into a coherent picture.
April 27,
2001
-------
2.O OVERVIEW OF THE DATA VALIDATION AND
EXPLORATORY ANALYSES
The first two phases of the analysis involved data validation and exploratory analysis. The
primary output from this activity was to update the database with a series of flags that indicated unusual
or questionable data and guided the modeling phase. A list of the flagged data has already been
provided to EPA in electronic form, so that information is not duplicated here. Rather, procedures
used to flag the data are described here.
2.1 DATA VALIDATION AND FLAGGING
While Battelle was carrying out this analysis, Research Triangle Institute (RTI managed the data
collection) was also completing a review and flagging of the data. They also produced a list of data that
they felt should not be included in any analysis. These data were removed before any modeling, but
were included in most of the exploratory analyses. Their flagging was based on the historical record of
the data. They flagged data with extremely low mass values (approximately the bottom 1 percent),
data with inconsistent reconstructed mass balance, and data with an extreme anion to cation ratio
(approximately the top and bottom 1 percent). The RTI flagged data is intended for flagging as "null" in
AIRS.
Battelle flags were based on either a self consistency check of the data from a given sampler on
a given day, or on a consistency check between co-located samplers. The self consistency checks
were based on comparisons of the species mass or a partial reconstructed mass to the measured mass.
If either a species mass or the partial reconstructed mass were significantly larger than the measured
mass then a flag was generated (and applied to all the data for the particular sampler and day). Initially
another flag was generated based on the ratio of a parameter mass to the measured mass that was
extreme in terms of the number of inter quartile ranges from the median. This flag always coincided
with one or more of the others and was not included in the final list of flags.
April 27,
4 2001
-------
The sampler to sampler consistency flags were based on comparing the difference between the
high and low value for co-located measurements of a parameter (including FRM mass). For each site
these differences might all be very small or large with considerable variation. The flags indicate the
extremes for the site. Hence, a difference that was flagged at one site may not be flagged at another.
The flags were combined into an alphanumeric code that indicated which parameters) caused the flag.
As in the above, if a flag was generated, then all the associated data for that day was flagged. After the
RTI flagged data was removed, there were 45 out of 1,072 site-day combinations with data that were
flagged with discrepancies in at least two species.
In most of the modeling analyses the modeling was done both with and without the flagged data.
Some modeling was also done with certain sites removed, since the outliers tended to cluster by site.
For the most part there was not an appreciable difference in the outcomes, because there is generally
enough data to average out the outliers. All of the results and summaries referenced in this report are
based on using data from all sites. As noted, the results and summaries are either based all non RTI
flagged data (i.e., all valid data) or data that were not flagged by either Battelle or RTI.
2.2 EXPLORATORY ANALYSES
Many of the graphs generated during the exploratory analysis for Tasks 1 and 2 served multiple
purposes. The main examples are in the appendix. The primary criteria for including the specific
examples chosen for Appendix A was either to demonstrate outliers or trends within the data. In the
latter case, the main concern was influences on the differences in the measured concentrations of the 14
study parameters. So while temperature certainly is a factor related to the magnitude of several of the
species, it was not found to have any appreciable relationship to the difference between one sampler
reading and another. Nor was the difference between temperatures reported for co-located samplers
related to the difference in the concentrations. See the appendix for a list of examples and descriptions
for how the graphs were generated.
April 27,
2001
-------
One of the key outcomes of the exploratory analysis was the decision to do the main modeling
on the logarithms of the concentrations rather than the raw concentrations. There are two reasons for
using a log-scale. The primary reason is that for many of the parameters and sites the biases between
co-located samplers are constant on the log-scale. Hence, the log-scale is more appropriate for
describing the typical differences observed. The second reason for using the log-scale is that deviations
from the mean are more symmetric on this scale. The statistical tests of significance are derived from
normality assumptions. The tests are more robust to deviations from the normality assumptions when
the random errors are symmetrically distributed.
April 27,
2001
-------
3.O ANALYSES OF THE ROUTINE DATA
There were three main types of analyses performed on the routine data. (Additional techniques
were applied to the measured mass. See Section 3.1.) First, the concentrations for each parameter
were studied with analysis of variance (ANOVA) models. The ANOVA models were applied to log
transformed data, so deviations from the mean indicate multiplicative differences. The results of this
investigation were site dependent. The second set of analyses examined an indicator of which sampler
type gave the higher value. The results from this investigation were generally independent of the site,
with the same qualitative result for most parameters. The third set of analyses examined the relative
amounts of constituent parameters with an ANOVA model. The results of these analyses show that
generally there are statistically significant differences in the relative composition among the sampler
types.
The first ANOVA model started with modeling the concentration value for a parameter at a
given site and day from a particular sampler as having an overall mean with deviations from the mean,
which could be attributed to random daily shifts, either of the MET variables, temperature and
barometric pressure, or the sampler's deviations from the daily means in temperature or pressure. As
expected from the exploratory analyses, these were statistically insignificant. In the further analyses the
MET variables were dropped from the model.
The next set of ANOVA analyses modeled the concentration value for a parameter at a given
site and day from a particular sampler as having an overall mean with deviations from the mean, which
could be attributed to site dependent random daily shifts, an overall site mean, a sampler type bias, and
sampler type-site interaction. The sampler type-site interaction allows the model to assume that the
bias of the sampler type is dependent on the site. These were modeled both assuming a common
variance for all three sampler types, and assuming that the residual error (measurement error) for the
three sampler types could be different.
7 April 27, 2OO1
-------
While the statistical test for any difference between the relative sizes of the residual errors was
significant (i.e., the statistical tests indicate that there is a difference in the CVs among the sampler
types), this should be tempered with two observations. First, the differences in the CVs may not be of
any practical significance, but, more importantly, the difference from site to site is generally much
greater than that of sampler to sampler. The difference in the precision may be more of a reflection of
the differences among the sites than that of the samplers, where a given sampler type may have by
chance been more frequently situated at more variable sites. Unfortunately, the statistical models
assuming different error variances by site did not converge, so it was not possible to test which was the
more important factor in the precision estimates.
The model(s) described above was run using all valid data (data not flagged by RTI) above the
MDL, using all the valid data except the data from Boston and Phoenix1, and finally with all of the valid
data excluding the Boston and Phoenix data and any data that were flagged in the outlier analysis. The
results for each case were similar, only the results from the first case are reported.
The sampler type-site interactions were statistically significant. This indicates, for example, that
how a URG compares to a MetOne depends on the site. The exploratory analysis and a close look at
the estimates from the models indicates a strong pattern in how the sampler types compared with each
other. While the scale of the deviations was strongly site dependent, the qualitative direction was quite
uniform. To investigate the strength of this observation without comparing the size of any relative bias,
the concentrations were modeled in a different manner in addition to the ANOVA analyses.
For each pair of data points, an indicator function was modeled. This yielded estimates of how
often a URG measurement will be less than an Andersen measurement in a side by side comparison.
To do this for each site with two samplers, the samplers were labeled A or B. If the site had a MetOne
sampler, the MetOne was labeled A and the other was labeled B. If the site only had an Andersen and
These two sites had more data with multiple flags for large sampler to sampler inconsistencies than the other
sites.
8 April 27, 2OO1
-------
a URG sampler, then the Andersen was labeled A and the URG was labeled B. (So the URGs are
always labeled B.) The indicator used was 1 if sampler B had a concentration value greater than the
value from A. If there was no bias between the samplers, then on average the indicator should be 1
about half of the time. The probability that the indicator is 1 was modeled to have a different
probability for each site (with logistic regression and the standard link for the binomial relationship), and
this model was compared with a model that had separate probabilities for each sampler type pair (the
three possibilities were MetOne-Andersen, MetOne-URG, and Andersen-URG). The statistical test
for the fit of the more general model (site dependent probabilities) versus the sampler type specific
probabilities shows that for most parameters, the same probabilities for sampler type pair are observed
across sites. However, the probabilities associated with a sampler type pair are usually not 0.5.
Together, these analyses indicate that there is a clear relative bias between the sampler types,
with URG < Andersen < MetOne for most parameters. They also indicate that the precision of the
samplers is more dependent on site than sampler type. (If not, then the sites with a greater mean
difference between the samplers would have had significantly different probabilities for concentration A
< concentration B. Since this was not the case, it must be that the variability increases with the relative
bias.) Hence, the results in Section 3.4 should be taken as being an indication of the expected site to
site variability in precision rather than differences in sampler precision due to sampler type.
3.1 THE MEASURED PM2.5 MASS CONCENTRATION
In the case of mass, there are co-located Federal Reference Method (FRM) measurements to
compare with the speciation sampler measurements. This section has comparisons of the speciation
sampler measurements to the FRM measurements based on the Expert Panel recommendations.
Section 3.2 has the results for the analyses that were done for each of the parameters in this study.
April 27, 2001
-------
The Expert Panel for the EPA Speciation Network set performance criteria for the mass
concentration based on a linear regression with co-located FRM measurements. The criteria for a
candidate method was: (1) the linear regression should have an R2 value at least 0.9 and (2) the ratio of
the candidate sampler concentration mean to the FRM mean should be between 0.9 and 1.1.
Moreover, these criteria should be based on at least twenty 24-hour samples. Table 3.1 shows the
results for the linear regression and the ratio of the sample means for each of the twenty-four samplers
studied. In each case, at least twenty-one 24-hour samples were used to obtain the results. The results
are not independent from each other in that the same set of FRM values are used for all the samplers at
a site. The table shows that essentially all but seven of the samplers met the R2 criteria (in one case the
R2 rounds up to 0.9). It also shows that the Andersen samplers faired both the best and the worst, and
that the ratio for all of the URG samplers is strictly less than the ratio for all of the MetOne samplers.
The ratio and the R2 values were tested for statistically significant site differences, sampler type
differences, and differences that depend jointly on the site and sampler type. To better meet statistical
assumptions the R2 values were transformed2 into the value Y = 1A ln((l+R)/(l-R)). For both the ratio
of the site means and the Ys, any differences due to the site and sampler type tested independent of
each other.
The R2 values have a significant site dependency (p-value = 0.0190) and a marginally
significant sampler type dependancy (p-value = 0.0769). The site dependancy is certainly due in part
to the fact that the same FRM results are used for all of the sampler comparisons at a site. If the FRM
values are incorrect, then one expects the correlation to degrade. This may be part of the problem at
the Boston site. Consider Figure 3.4, note that there are pairs of values for which the two Andersen
samplers have the same mass value, but are different from the FRM value for the day. However, this
would have the same effect on all of the samplers at a site and does not explain the discrepancies seen.
2 Hogg, R., and Tanis, E. (1977). Probability and Statistical Inference, Macmillan Publishing Company, Inc., New
York, New York.
10 April 27, 2OO1
-------
Hence, there must be other site effects that generally impact all of the samplers at a site. (See Figure
3.3.)
The ratio of the mean mass measured by the sampler versus the mean FRM mass is strongly
dependent on both the site and the sampler type. Again, the common FRM values for a site contribute
to the site effect, but there may be other site related effects as well. In this case, there is a very clear
sampler type effect as well. All of the MetOne samplers have a ratio above 1. All except one of the
URG samplers have ratios less than 1. Both of the above statements are somewhat confounded by the
site effect. However, at all sites the URG means are less than the Andersen means, which in turn are
less than the MetOne means. (See Figure 3.1.)
11 April 27, 2001
-------
Table 3.1 FRM to Speciation Sampler Regression Results, Sorted
by the Ratio of the Mean Mass to the Mean FRM Mass.
Site
Boston
Boston
Bismark
Chicago
Boston
Chicago
Philadelphia
Seattle
Tampa
Fresno
Houston
New York
Tampa
Philadelphia
New York
St. Louis
St. Louis
Fresno
Houston
Bismark
Salt Lake
New York
Seattle
Salt Lake
POC
7
5
6
6
6
5
6
6
6
6
6
5
5
5
6
6
5
5
5
5
6
7
5
5
n
24
21
25
37
23
37
24
39
32
26
23
26
32
24
36
37
37
26
23
25
23
37
39
23
Intercept
0.228(.056)
1.712(.109)
0.226(.046)
-0.150(.035)
3.896(.115)
-0.272(.020)
0.907(.031)
0.314(.022)
-1.668(.067)
0.463(.052)
0.845(.034)
0.707(.024)
1.163(.061)
1.074(.028)
1.368(.023)
2.927(.071)
3.582(.091)
2.052(.102)
3.248(.115)
1.700(.080)
1.025(.030)
1.539(.027)
1.329(.050)
1.916(.066)
Slope
0.854(.056)
0.750(.109)
0.862(.046)
0.928(.035)
0.594(.115)
0.969(.020)
0.916(.031)
0.944(.022)
1.116(.067)
0.957(.052)
0.959(.034)
0.984(.024)
0.973(.061)
0.990(.028)
1.014(.023)
0.923(.071)
0.878(.091)
0.916(.102)
0.891(.115)
0.838(.080)
1.014(.030)
1.062(.027)
1.042(.050)
0.963(.066)
R-squared
0.914
0.713
0.938
0.953
0.559
0.986
0.975
0.981
0.903
0.934
0.975
0.985
0.895
0.983
0.983
0.830
0.727
0.769
0.742
0.826
0.982
0.978
0.920
0.910
Ratio
0.874
0.890
0.904
0.920
0.927
0.953
0.984
0.987
0.990
1.010
1.027
1.047
1.061
1.071
1.130
1.137
1.139
1.146
1.150
1.154
1.156
1.186
1.223
1.229
Sampler
URG
Andersen
URG
URG
Andersen
Andersen
URG
URG
URG
Andersen
URG
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
MetOne
Andersen
MetOne
Andersen
MetOne
MetOne
MetOne
12
April 27, 2001
-------
Table 3.2 FRM to Speciation Sampler Regression Results, Sorted
by the R-Squared Value.
Site
Boston
Boston
St.Louis
Houston
Fresno
Bismark
St.Louis
Tampa
Tampa
Salt Lake
Boston
Seattle
Fresno
Bismark
Chicago
Houston
Philadelphia
New York
Seattle
Salt Lake
New York
Philadelphia
New York
Chicago
POC
6
5
5
5
5
5
6
5
6
5
7
5
6
6
6
6
6
7
6
6
6
5
5
5
n
23
21
37
23
26
25
37
32
32
23
24
39
26
25
37
23
24
37
39
23
36
24
26
37
Intercept
3.896(.115)
1.712(.109)
3.582(.091)
3.248(.115)
2.052(.102)
1.700(.080)
2.927(.071)
1.163(.061)
-1.668(.067)
1.916(.066)
0.228(.056)
1.329(.050)
0.463(.052)
0.226(.046)
-0.150(.035)
0.845(.034)
0.907(.031)
1.539(.027)
0.314(.022)
1.025(.030)
1.368(.023)
1.074(.028)
0.707(.024)
-0.272(.020)
Slope
0.594(.115)
0.750(.109)
0.878(.091)
0.891(.115)
0.916(.102)
0.838(.080)
0.923(.071)
0.973(.061)
1.116(.067)
0.963(.066)
0.854(.056)
1.042(.050)
0.957(.052)
0.862(.046)
0.928(.035)
0.959(.034)
0.916(.031)
1.062(.027)
0.944(.022)
1.014(.030)
1.014(.023)
0.990(.028)
0.984(.024)
0.969(.020)
R-squared
0.559
0.713
0.727
0.742
0.769
0.826
0.830
0.895
0.903
0.910
0.914
0.920
0.934
0.938
0.953
0.975
0.975
0.978
0.981
0.982
0.983
0.983
0.985
0.986
Ratio
0.927
0.890
1.139
1.150
1.146
1.154
1.137
1.061
0.990
1.229
0.874
1.223
1.010
0.904
0.920
1.027
0.984
1.186
0.987
1.156
1.130
1.071
1.047
0.953
Sampler
Andersen
Andersen
MetOne
Andersen
MetOne
MetOne
Andersen
MetOne
URG
MetOne
URG
MetOne
Andersen
URG
URG
URG
URG
MetOne
URG
Andersen
MetOne
Andersen
Andersen
Andersen
13
April 27, 2001
-------
M> of Mgm MM** Mm to MOT FflM Mm
1*1-
-I 1 1 1 h-
Figure 3.1
Ratio of Mean Measured Mass to Mean Mass From
a Co-located FRM.
Mto tt User kfcausd MM t> Me* m» Mow
OHM'
fl HUB1
OMB'
!"•
^g;
din.
-i 1 1 1 1 1 1 1-
Figure 3.2 The Relative Deviations of the Ratio of the Measured
Mass to the Mean FRM Mass (the plotted values show
the ratios minus the site mean).*
*Note that all of the MetOne values are above 0 and all of the URG values are below 0.
14
April 27, 2OO1
-------
MplMfen H^CJUBId BlMlMI ttW MMIUId IMN Vld RM MMt
•
I
B
UB
WtO
J. ••*•
*
Sk»
Figure 3.3
The Linear Regression R-squared with the Co-
located FRM.
SpMM&n Memo MMMnd Mm \nmrn FHM MM*
M
1/1
IV
M
(V
i « ^^1 • ••
I sf r IT
OB 0.7 OB
14
fi=AraJH
Figure 3.4 Speciation Sampler Measured Mass Versus a Co-
located FRM Measurement.
15
April 27, 2OO1
-------
3.2 ROUTINE PARAMETER MEASUREMENTS
For each parameter the log-concentrations were modeled as follows. An individual value was
treated as the sum of an overall mean, deviations from that mean based on the site, site specific
deviations from the (site) mean due to the vendor, random site specific day-to-day variations from the
site mean, and random measurement error. The site specific deviations due to a sampler type were
tested to see if these were the same across all sites. The "average" sampler type and site deviations
were tested to see if they were statistically different from zero.
Table 3.3 shows the p-values for these tests. A p-value less than 0.05 is generally considered
significant. This general rule of thumb is often modified when many tests are being considered so that
the overall error rate stays at 5 percent. In this case it does not matter, almost everything is highly
significant. The conclusion of these tests is that individual samplers have statistically different relative
biases among them and these biases are not consistent across sites or sampler types.
These results only indicate statistically significant differences. Hence, the differences are too
large to be attributed to random chance alone. They do not indicate whether or not the differences are
of practical significance. That requires expert judgement, typically in the form of DQOs . In the case of
mass, the Expert Panel Recommendations can be used as a guide for assessing the practical significance
of the differences. More generally, there are several tables given in Appendix B that summarize the
findings from different points of view that should be used to evaluate the practical differences among the
sampler types.
16 April 27, 2OO1
-------
Table 3.3 P-values for Tests of Significant Differences Among
Sites, Sampler Types, and All Site-Sampler Type
Combinations.
Parameter
PM2.5 Mass
Aluminum
Calcium
Chlorine
Iron
Lead
Tin
Silicon
Zinc
Ammonium
Organic Carbon
Nitrate
Elemental Carbon
Sulfate
Estimated p-value*
Site
<0001
<0001
<0001
<0001
<0001
<0001
0.0031
<0001
<0001
<0001
<0001
<0001
<0001
<.0001
Vendor
<.0001
<.0001
<0001
<.0001
<.0001
<.0001
<.0001
<0001
<.0001
<.0001
<.0001
<.0001
<0001
<.0001
Site*Vendor
0.0001
<.0001
<0001
<.0001
<.0001
0.0002
0.3095
0.0005
<0001
<.0001
0.0037
0.0087
0.0596
<0001
* Results are shown using the all valid data.
Individual pairs of samplers are consistent (on a log scale), see Figures 3.5 and 3.6. However,
at the Bismark site a typical difference (on the natural log scale) is for the MetOne to be 0.3 higher than
the URG. At the Tampa site the difference typically is about 0.15 in the same direction. Figure 3.7
shows boxplots of the daily differences for organic carbon measurements for all sites on the log base 10
scale. The medians are clearly different from site to site, even between sites with samplers of the same
type. (There is no visual difference between using the natural log versus base 10, only the units
change.)
17
April 27, 2OO1
-------
Comparisons Between CoHbcated Monitors
PARM=OrBarft> Carbon SffTE=B|sniarklND (Bflsmarck Residential!)
OJB40
0^462
0.264
=0.112
=0,300
-OJ050
OJQ7S 0206 0.334
Dafly Average Iog(0oncierrtral0on}
0.462
0,590
Figure 3.5 Log OC Concentrations at Bismarck, ND Plotted
Against the Mean of the Logs of the Concentration.
18
April 27, 2OO1
-------
Comparisons Between CoHkxzted Monitors
RWM=Organfc Carbon SffTE=Tainpa-StP9lBreburg-Cbarwat9r,FL (Lawfe)
0320
0.754
OjSBB
0.422
0.256
OJ090
0.110
0266 0.422 0578 0.734
Daly Average DagfCorrantnatibn)
OJ90
Figure 3.6 Log OC Concentrations at Tampa, FL Plotted Against the
Mean of the Logs of the Concentrations.
19
April 27, 2OO1
-------
Dait/ Ran?* Eto>. Pots
IW1M = Ccganio Csrtson
'J.V
D.4-
*Z fN.JI IS. f*J>l P. f*,U L»U1 Nx.l-.nlUl HO IM<.1 NYM.il MC'N.JI OH (WAI r* tt.Ul IX I*. U IT* A' V* fWJUl
Figure 3.7 Box Plots of the Daily Range in the Logs of the OC
Concentrations.
There is a consistency across sites (and even species). As seen with the mass data, the
directions of the relative biases are frequently the same. To further explore this the sites with two
sampler types were tested as follows. All the MetOne samplers were labeled A and all of the URG
samplers were labeled B. If an Andersen was paired with a MetOne, then the Andersen was labeled
B. If an Andersen was paired with a URG, then the Andersen was labeled A. In this way there are
three different pairings, and the order is always fixed. Then, for each pair of data points an indicator
was created. If the concentration value from an "A" sampler was higher than the concentration from
"B", then the indicator was 1. Otherwise the indicator was 0. The percentage of the time that the
indicator is 1 (for a given parameter) can be tested (using a technique called logistic regression) to see if
this is a function of the site, or of just the sampler type, or if it is independent of both.
20
April 27, 2OO1
-------
Table 3.4 shows the results for the tests of the null hypothesis that the probabilities are
dependent on both site and sampler type versus the alternative that they depend only on sampler type.
In most cases this test indicates that the probabilities are not dependent on the site. The exceptions are
chlorine, ammonium, nitrate and sulfate. Since most species showed no site dependence the model was
refit with only a sampler type pair dependency, and this was tested against no sampler type
dependence. The p-values for these tests are strongly significant. Hence, generally there is a significant
difference between the sampler types. However, since sampler pair and site are confounded the p-
value for the significance of the sampler pair should be ignored whenever the site effect is significant.
The overall conclusion is that there are significant vendor specific relative biases and the percent of the
time that one vendor type is greater than another is not dependent on the site for most species. These
results mimic the findings noted in the comparisons with the FRM mass measurements (Section 3.1). In
fact, the ordering is the same for most parameters. See Table 3.4.
21 April 27, 2OO1
-------
Table 3.4 Probabilities of Sampler Type A Yielding Values Greater
than Sampler Type B P-values for the Significance of
Sampler Type and Site.
Parameter
PM2.5 Mass
Aluminum
Calcium
Chlorine
Iron
Lead
Tin
Silicon
Zinc
Ammonium
Organic Carbon
Nitrate
Elemental Carbon
Sulfate
Estimated Probability of
Vendor A > Vendor B
Andersen-
URG
0.90
0.73
0.96
0.70
0.96
0.63
0.48
0.93
0.76
0.23
0.95
0.79
0.85
0.29
MetOne-
Andersen
0.72
0.56
0.50
0.29
0.42
0.75
0.98
0.48
0.12
0.61
0.70
0.75
0.54
0.70
MetOne-
URG
0.96
0.94
0.94
0.81
0.96
0.94
1.00
0.90
0.27
0.05
0.99
0.82
0.80
0.47
Estimated
p-value
Pair
<0001
<0001
<0001
<0001
<0001
0.0011
<0001
<0001
<0001
<0001
<0001
0.4360
<0001
<0001
Site
0.0659
0.3519
0.5119
0.0005
0.1192
0.2697
0.8930
0.3871
0.0878
0.0008
0.0988
0.0026
0.8722
0.0004
Implied ordering to
the vendor
measurements*
U
-------
3.3 ROUTINE PARAMETER MEASUREMENTS RELATIVE THE
MEASURED MASS
Given that there are differences among the samplers that show a consistent ordering with the
ordering for the measured mass, a logical next question is whether or not the samplers yield the same
relative compositions. To test this for each parameter a new variable was created that was equal to the
logarithm of the ratio of the parameter concentration to the concentration of the measured mass. This
was modeled to have a random mean for site and day with fixed mean deviations for each site and
sampler type within site.
For each parameter, there are significant deviations in the relative compositions between co-
located samplers. Table 3.5 shows the significance of difference in relative composition of organic
carbon at each site. Among the sites with Andersen and MetOne samplers, three of the five show no
significant difference in the relative composition while relative amounts at Fresno and Portland are
significantly different. Table B.4 has the complete list for all constituent parameters studied. Also see
Tables B.I and B.2 for the magnitude of the differences.
23 April 27, 2OO1
-------
Table 3.5 Significance of the differences in the relative amounts of
Organic Carbon at each site.
Sampler Pair Type
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Motu ipr;
Site
Fresno
St. Louis
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Significance of Sampler
Differences for the Site
0.0113
0.4731
0.7536
0.004
0.5642
<.0001
<.0001
<.0001
0.0136
0.2732
<.0001
0.0004
< nnm
3.4 RELATIVE PRECISIONS OF THE SAMPLER TYPES
In addition to looking for relative biases among the sampler types, the precisions were also
studied. The estimates of the variability were obtained by modeling two different sources of variability,
temporal variability and measurement error, along with the estimates of mean behavior (all done on the
log scale). The estimates in Table 3.6 are based on two different types of models with the data for each
parameter modeled separately. The temporal and the aggregate measurement error estimates are
based on models that assumed that the measurement error was independent of the sampler type. The
sampler specific measurement errors come from statistical models for the data that assumed a temporal
component to the variability and three distinct measurement errors. For both cases Table 3.6 shows
the results using all non-RTI flagged data.
Generally, the statistical test for the significance of using three distinct measurement errors rather
than a single aggregate measurement error was positive. Hence, the data showed that the precisions
are generally distinct. An inspection of Table 3.6 shows that for many parameters, the MetOne
precision was lower than the other two. However, there is a strong possibility that the differences
noted here are in fact due to site to site differences, not sampler to sampler differences. To test this
24 April 27, 2OO1
-------
directly, more complex statistical models were tried, but the models failed to converge. The modeling
based on indicators of which sampler had the higher concentration for a given site and day gives indirect
evidence that the differences in precision noted here are due more to site differences, rather than the
sampler type (See Section 3.2).
Where the models converged they show that the temporal component of the variability was
often two to three times higher than the measurement error and many times larger than any difference
among the samplers. Hence, for the purpose of establishing long term averages or annual trends,
sampling frequency would be more important than measurement error and much more important than
sampler type. Hence, from a practical point of view, there is extremely little difference in the precisions.
Table 3.6 Estimated Variance Components for Each Parameter
Parameter
DM2.5 Mass
Muminum
Dalcium
Chlorine
ron
_ead
Tin1
Silicon
line
^monium
Drganic Carbon
\litrate
Elemental Carbon
Bulfate
Temporal
Variation (CV)
0.470
0.979
0.540
1.673
0.615
0.477
0.000 1
0.706
0.794
0.862
0.381
0.862
0.499
0.652
Aggregate
Measurement
Error (CV)
0.134
0.440
0.302
0.547
0.263
0.266
0.187 1
0.323
0.269
0.267
0.181
0.221
0.250
0.141
Andersen
Precision (CV)
0.152
0.622
0.379
0.739
0.302
0.344
0.206 1
0.409
0.254
0.278
0.111
0.219
0.218
0.150
MetOne
Precision
(CV)
0.140
0.226
0.195
0.263
0.147
0.208
0.1671
0.149
0.349
0.138
0.077
0.059
0.266
0.063
URG Precision
(CV)
0.099
0.414
0.297
0.578
0.304
0.231
0.184 1
0.335
0.180
0.355
0.291
0.318
0.253
0.187
The statistical model did not converge to a self-consistent state.
25
April 27, 2OO1
-------
4.O ANALYSIS OF THE BLANKS
In addition to the monitoring, filters are stepped through part or all of the sampling process
except for the drawing of air through them. Ordinarily, analyzing these filters is used as a check for any
contamination on the filter due to the handling. The purpose here is slightly different. In this case the
blanks, both trip blanks and filter blanks, were used to test whether or not there is any tendency for one
sampler type's filters to be contaminated significantly more than another. As usual this question is first
asked in a statistical sense, and then if there are differences, there is a different question of whether
there is a practical difference. However, the second question is further complicated by the fact that
different volumes of air are drawn through the routine filters. For some this may change the point of
view of what is a practical difference. For others it may not since any evidence of contamination casts
some doubt on the measurements. For the purpose of this study, the masses found on the filters were
not "corrected" for an average volume.
The statistical analysis of the blanks is complicated by the fact that many of the measurements
are below the MDL, and frequently 0. Unlike the routine data there is not a transformation of scale that
is suitable for ANOVA-like techniques. Instead, the data was converted to a binary form. This can be
thought of as treating the individual values as either "negligible" contamination or "non-negligible." The
goal was to use a practical definition that would separate the data by sampler type if there is any
"difference". However, this a difficult item to quantify in a satisfactory manner. The MDL is not a good
cutoff. For some compounds, essentially all of the data are on one side of the MDL or the other, so
there is no basis for deciding whether one sampler type is "cleaner" than another. Further, the
quantities labeled as the MDLs may or may not be the true detection limits. It was decided to use the
data themselves as the basis for a practical cutoff, namely the parameter specific third quartile of all the
blanks. (The third quartile is denoted Q3, and equals the value such that it is greater than or equal to
75 percent of the data and less than or equal to 25 percent of the data.) This assures that there will be
April 27,
26 2001
-------
sufficient data both above and below the cutoff to be useful (unless as in the case of zinc nearly 100
percent of the data is equal to 0). Also by its very nature Q3 is a practical (achievable) bound for the
contamination levels with the current technology.
Using an indicator function treats very large values the same as values just over the cutoff. For
instance, there were three cases where the mass was over 100 micrograms. Such extremes may or
may not be real and certainly would influence an analysis that considered the scale. It was decided not
to keep such extreme cases in the analysis even though they would not have an undue influence on the
analysis. It should also be pointed out that such values appear to be "replicatable." That is, the
experiments discussed in Section 5 included multiple blanks, and there were cases where unusually high
values were replicated (See Sections 5.1 and 5.4).
Table 4.1 summarizes the data as used in the analysis. For each compound the MDL, the
overall median, and overall Q3 is listed. These give an indication for the spread of the data and the
relationship between typical values and the MDL. Also shown in Table 4.1 is the percent of the time
that data from a given sampler type is above the overall Q3. For mass, the typical values are many
times greater than the MDL. All of the actual proportions are less than 25 percent. This is possible
when there are many values that are equal to Q3. Still there is a clear difference in percentages with
Andersen > MetOne > URG. The modeling of the data checks to see if this is more likely due to site
effects or if it is a true difference between the sampler types.
April 27,
27 2001
-------
Table 4.1 Summary of the Blank Data.
Parameter
PM2.5 Mass
Aluminum
Calcium
Chlorine
Iron
Lead
Tin
Silicon
Zinc
Ammonium
Organic Carbon
Nitrate
Elemental Carbon
Sulfate
avg MDL
(ug)
0.976
0.105
0.033
0.056
0.019
0.053
0.172
0.073
0.014
0.163
1.412
0.078
1.412
0.117
Median
(ug)
13.000
0.003
0.041
0.000
0.031
0.026
0.184
0.015
0.000
0.000
12.199
0.739
0.689
0.864
Q3
(ug)
18.000
0.052
0.056
0.018
0.045
0.041
0.227
0.042
0.000
0.000
15.380
1.124
0.984
1.534
Actual Proportion > Q3
Andersen
(out of 93), %
23.7
20.4
14.0
16.1
17.2
14.0
9.7
17.2
4.3
10.8
20.4
4.31
17.2
21.5
MetOne
(out of 83), %
16.9
20.5
21.7
22.9
19.3
16.9
20.5
20.5
2.4
0
28.9
8.4
20.5
27.7
URG
(out of 81),%
9.9
11.1
17.3
12.3
13.6
18.5
22.2
13.6
0
3.7
3.7
42.52
11.1
3.7
1 92 obs
2 80 obs
Table 4.2 shows the modeling results. In each case, the cutoff Q3 is listed for reference. The
next three columns are p-values for the three tests of interest. The last three columns give confidence
intervals for the probability of observing a value greater than Q3. These estimates are based on a
model without a site effect and are "averaged" over blank type.
The column "Type" is for a test of any significant difference between field blanks and trip
blanks. Trip blanks are taken and opened at the site, and then resealed for analysis. Field blanks are
additionally placed in the sampler for a moment or two (sometimes with the sampler turned on to
"shake loose" anything in the sampler). It would be natural to assume that for nonvolatile compounds
28
April 27,
2001
-------
the field blanks would naturally be higher than the trip blanks. However, field blanks and trip blanks
are equally likely to have values greater than Q3. Lead is a notable exception to this observation with
about 10 percent of the trip blanks greater than Q3 versus 28 percent of the field blanks. Ammonium,
nitrate, and sulfate also show differences, but in the opposite direction since trip blanks are higher. For
ammonium 5 percent of the field blanks are greater than 0 versus 12.5 percent of the trip blanks. For
nitrate and sulfate, approximately 21-22 percent of the field blanks are above Q3 versus 33-35 percent
of the trip blanks.
The "Sampler" column in Table 4.2 is a test of whether or not there are significant differences
among the sampler types. The effects are averaged over blank type (see below). The effect of sampler
type appears generally insignificant for the mass and the metals except tin. There are significant
differences between the sampler types for ammonium, OC, nitrate, and sulfate.
The "site" column is a test of site dependent effects. The models with a site effect did not
always converge because there were insufficient non-zero data to test. Where the models did
converge, the site effect is negligible overall.
April 27,
29 2001
-------
Table 4.2
Modeling Results for the Blank Data.
Parameter
DM2.5 Mass
Muminum
Dalcium
Chlorine
ron
_ead
Tin
Silicon
line
^monium
Drganic Carbon
\litrate
Elemental Carbon
Bulfate
Q3
18.000
0.052
0.056
0.018
0.045
0.041
0.227
0.042
0.000
0.000
15.380
1.124
0.984
1.534
Significance of
Type
0.3030
0.3690
0.4442
0.4371
0.3270
0.0065
0.9602
0.2784
0.8938
0.0558
0.6974
0.0441
0.4932
0.0138
Sampler
0.1374
0.2560
0.1924
0.1583
0.5867
0.5573
0.0160
0.4483
0.2156
0.0044
<0001
<0001
0.2379
<0001
Site
0.5849
0.1507
0.4446
0.3644
0.0403
0.2500
0.3586
0.1528
*
*
0.0599
0.5491
0.1929
0.1843
Estimated Probability of Being > Q3
Andersen
(0.23,0.47)
(0.19,0.42)
(0.09,0.28)
(0.13,0.34)
(0.12,0.33)
(0.07,0.26)
(0.07,0.24)
(0.12,0.33)
(0.01,0.13)
(0.07,0.27)
(0.17,0.40)
(0.01,0.13)
(0.13,0.35)
(0.21,0.46)
MetOne
(0.15,0.39)
(0.19,0.44)
(0.19,0.44)
(0.21,0.46)
(0.16,0.40)
(0.10,0.32)
(0.19,0.44)
(0.17,0.42)
(0.01,0.14)
**
(0.28,0.54)
(0.07,0.27)
(0.20,0.46)
(0.31,0.59)
URG
(0.09,0.31)
(0.09,0.31)
(0.15,0.38)
(0.09,0.30)
(0.10,0.32)
(0.12,0.35)
(0.22,0.48)
(0.10,0.32)
**
(0.02,0.18)
(0.02,0.16)
(0.54,0.80)
(0.09,0.31)
(0.02,0.18)
Model does not converge
Estimate not reliable
The above analysis is misleading for chlorine and silicon because the effect of sampler type is
blank type specific. As a result, grouping the data by either factor tends to obscure the effects and
leads to null conclusions. For these two compounds statistically significant sampler type-blank type
differences were noted. See Figures 4.1 and 4.2. For all other compounds, a model with a sampler
type-blank type interaction tested insignificant.
30
April 27,
2001
-------
^- na
> us
UHQ
Figure 4.1
The Proportion of Field and Trip Blanks with Silicon
Measurements Greater than Q3 by Sampler Type.
Ftqwtfcn > 08
UHQ
Figure 4.2 The Proportion of Field and Trip Blanks with
Chlorine Measurements Greater than Q3 by
Sampler Type.
31
April 27,
2OO1
-------
5.0 EXPERIMENTS TO SIMULATE AND TEST POTENTIAL SAMPLE
INTEGRITY ISSUES WHEN USING SEQUENTIAL SPECIATION
SAMPLERS
Five experiments were undertaken to test various issues that are associated with
sequential samplers and sampling integrity. Four of the experiments are directly concerned with
the fact that, in a sequential sampler the filters can collect material by passive sampling while
sitting unsealed in the sampler. Also, if the filters have already collected material as part of a
sample, then there is the possibility that some of the sample material may volatilize.
Sections 5.1 and 5.2 examine the results of two experiments that simulate the sequential
sampling, specifically looking at organic and elemental carbon. Sections 5.4 and 5.5 examine the
results for corresponding experiments that look at the effects on nitrate and sulfate. Section 5.3
examines the results of an additional experiment targeting a concern about the collection or loss
of carbon compounds due to the face velocity at the quartz filter.
The experimental design called for the collection of more data than was collected for
these five experiments. All of the available (and directly relevant) data is at least plotted in each
of the following sections. The lack of data can affect the ability to detect small differences. This
is especially true for data that have a relatively high variability. As will be seen, the results of the
face velocity test are not quantitatively consistent from the results with the routine samples. The
problem may be the lack of data.
5.1 COLLECTION OF VOLATILE ORGANIC COMPOUNDS ON BLANK
QUARTZ FILTERS
In this experiment, a MetOne sampler was loaded with five quartz modules. The five
modules were then left in the MetOne sampler for six to nine days while leaving the sampler idle.
In this way the filters were exposed to ambient air for a week and the effects of filters sitting in
the sampler for an extended period before sampling were simulated.
Figures 5.1 and 5.2 show scatter plots of the raw data for the experiment.
32 April 27, 2OO1
-------
Effects of Leaving the Filter in the Monitor
Phoenix Fresno Tampa St.Louis New York Bismark Portland Salt Lake Seattle
Site
X=Experiment Blank F=Field Blank R=Freezer Blank T=Trip Blank
Figure 5.1 Effects of Leaving Blank Filters in the Sampler for a Week on the
Measured OC.
Effects of Leaving the Filter in the Monitor
Phoenix Fresno Tampa St.Louis New York Bismark Portland Salt Lake Seattle
X-Experiment Blank F-Field Blank R-Freezer Blank T-Trp Blank
Figure 5.2 Effects of Leaving Blank Filters in the Sampler for a Week on the
Measured EC.
33
April 27, 2OO1
-------
Clearly the Phoenix data in this experiment has a different nature than the other sites.
(See Section 5.6.) Hence, to begin with, the Phoenix data are treated as outliers and only the
other data points are modeled. The mean response was modeled as an overall mean for each
blank type with random variations due to the site. For OC the overall mean for the experiment
blanks was estimated to be 11.53 micrograms with a standard error of 0.685 micrograms while
the field blanks had an overall mean of 8.26 micrograms with a standard error of
0.894 micrograms. The p-value for the test of whether the true means are statistically different
is <0.0001. Hence, with respect to OC, there is a significant statistical difference between the
field blanks and the blanks that are left in the sampler for at least a week.
The EC data were modeled similarly (with the Phoenix data removed). For EC the
overall mean for the experiment blanks was estimated to be 0.541 micrograms with a standard
error of 0.108 micrograms while the field blanks had an overall mean of 0.295 micrograms with
a standard error of 0.171 micrograms. The p-value for the test of whether the true means are
statistically different is 0.129. Hence, with respect to EC, there is no significant statistical
difference between the field blanks and the blanks that are left in the sampler for at least a week.
5.2 COLLECTION OF VOLATILE ORGANIC COMPOUNDS ON EXPOSED
QUARTZ FILTERS
In this experiment five modules were loaded with quartz filters for a MetOne sampler.
Two of these were recovered according to standard procedures (within 48 hours of sampling).
The remaining filters were recovered at least six days after sampling. This simulated the
condition of a sequential sampler where a sample is left in the sampler for an extended period.
Figures 5.3 and 5.4 show scatter plots of the raw data for the experiment. Figure 5.5 shows the
ratio of the EC concentration to the OC concentration.
Both Figures 5.3 and 5.4 show unusual values in opposite directions. The Phoenix data
may not seem to matter because there were not any standard recovery measurements in the data.
However, it could be useful in estimating the sampling error size, if this represents real data. To
help decide, the ratio of the concentrations is also plotted in Figure 5.5.
34 April 27, 2OO1
-------
The modeling was based on removing the Salt Lake and Phoenix data. The values were
assumed to have a different mean for each site and day. The effect of leaving the filter in the
sampler was modeled as producing a shift in the site mean (where the same shift is used for all
sites). Table 5.1 shows the mean difference between the samples that were collected within
48 hours and those that were left in the sampler for at least 6 days and the associated p-value. In
both cases there is no significant difference between the collection methods.
Table 5.1 Mean Difference Between Standard Collection of Samples and
Those Left in the Sampler.
Compound
Organic carbon
Elemental carbon
Mean difference
0.234
0.039
Standard error
0.330
0.042
p-value
0.4841
0.3565
35
April 27, 2OO1
-------
Effects of Leaving Sample in Monitor
1.2
1.1-
1.0
0.9
as
0.7-
0.6-
0.5
0.4
03-
02
0.1
0.0
Phoenix fiesno Tampa St.Louis New York Bismark Portland Salt Uke
Site
0-standard recovery 1-recovery after six days
Figure 5.3 The Effects of Leaving a Filter in a Sampler for a Week on the
Observed EC Concentration. Note That the Two Unusual
Values Are in Opposite Directions.
Effects of Leaving Sample In Monitor
Phoenix Fresno Tampa St.Louis New York Bismark Portland Salt Lake Seattle
Site
0=standard recovery 1=recovery after six days
Figure 5.4 The Effects of Leaving a Filter in a Sampler for a Week on the
Observed OC Concentration.
Note That There Are Unusual Values in Opposite Directions.
36
April 27, 2OO1
-------
Effects of Leaving Sample in Monitor
029
0.28
0.27
OSS
025
024
023
022
021
020
0.19
0.18-
0.17
0.16
0.16
0.14
0.13-
0.12
0.11
0.10
0.09-
0.08-
0.07
0.06
0.05
0.04-
0.03
0.02
0.01
0.00-
1
Phoenk Fresno Tampa St.Louis New York Blsmark Portland Salt Lake Seattle
Site
0=standard recovery 1=recovery after six days
Figure 5.5 The Effects of Leaving a Filter in a Sampler for a
Week on the Observed Ratio of the EC Concentration
to the OC Concentration.
* *Note that the Seattle ratios now all appear "normal." The two unusual points, one in
Phoenix and one in Salt Lake, are both low compared to the other values.
5.3 TESTING THE EFFECTS OF FACE VELOCITY ON THE COLLECTION OF
VOLATILE ORGANIC COMPOUNDS ON QUARTZ FILTERS
In this experiment, channels 1 and 2 of Andersen samplers were loaded with a quartz
filter. The two channels have flow rates of 7.3 1pm and 16.7 1pm under normal operations. In
this way simultaneous samples were collected under the two different conditions on two separate
quartz filters (with fewer differences between the sampling methods compared to using two
different samplers from different vendors). Figures 5.6 and 5.7 below show the raw data with the
concentration from the high volume channel plotted against the concentration from the low
volume channel.
37
April 27, 2OO1
-------
2
High Vblume EC Data Versus Low Volume EC Data
P B
1 2
EC Cone (Low Volume)
B=Boston C=Chicago P=Philadelphia
Figure 5.6 High Volume Elemental Carbon Concentrations Versus Low
Volume Elemental Carbon Concentrations with a 1-1 line.
High Volume OC Data Versus Low Volume OC Data
P C
456
OC Cone (Low Volume)
B-Boston C-Chkago P-Philadelphia
8 9
Figure 5.7 High Volume Organic Carbon Concentrations Versus Low
Volume Organic Carbon Concentrations with a 1 -1 line.
38
April 27, 2OO1
-------
For both cases, the high volume concentrations were regressed against the low volume
concentrations (the standard for an Andersen quartz filter). The regression results are shown
below in Table 5.2. The regression procedure treats the low volume measurements as error free.
Table 5.2 Regression Results for Modeling the High Volume
Concentrations Against the Low Volume Concentrations.
Compound
Elemental Carbon
Organic Carbon
Intercept (SE)
0.451 (0.280)
4.809(1.534)
Slope (SE)
0.252 (0.184)
-0.194 (0.244)
R2
0.136
0.050
To be more comparable with the results shown in Chapter 3, the above was repeated on
the log scale. They seem to show much more disagreement than would be expected. (Compare
Figure 5.10 with Figures 3.5 and 3.6. These compare the OC concentrations from co-located
pairs of a URG sampler and a MetOne sampler. The flow rate for the MetOne is 6.7 1pm and the
flow rate for the URG is 16.7 1pm.)
39
April 27, 2OO1
-------
Log Transformed High Volume EC Data Versus Log Transformed Low Volume EC Data
1
-1
-2-
-3
-4
-0.8 -0.6 -0.4 -0.2 0.0 02 OA 0.6 0.8 1.0
h(EC Cone) (Low Volume)
B-Boston C-Chlcago P-Philadelphia
Figure 5.8 High Volume EC Data Versus Low Volume EC Data on the Log
Scale.
Log Transformed High Volume OC Data Versus Log Transformed Low Volume OC Data
2-
-1
0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2.0 2.1 2.2
ki(OC Cone) (Low Volume)
B=Boston C=Chfcago P=Philadelphia
Figure 5.9 High Volume OC Data Versus Low Volume OC Data on the Log
Scale.
4O
April 27, 2OO1
-------
Table 5.3 Regression Results for Modeling the High Volume
Concentrations Against the Low Volume Concentrations on the
Log Scale.
Compound
Elemental Carbon
Organic Carbon
Intercept (SE)
-0.700 (0.265)
0.350(1.0158)
Slope (SE)
0.985 (0.484)
0.451 (0.572)
R2
0.257
0.049
While there are less data for this experiment than was planned, there should be enough to
detect some trend or correlation between the two sets of concentrations. On both scales there are
three points with high low volume concentrations and low high-volume concentrations. As a
result in both cases the slope is not significantly different from 0. (Hence, from a statistical
perspective the low volume concentration provides no information about the high volume
concentration.)
The routine data may have showed a bias between vendors, but at least they correlated
well with each other. These data show no significant correlation. This inconsistency should be
considered before drawing any conclusions based on these data.
41
April 27, 2OO1
-------
2.5
2.0
1.5
1.0
0.5
0.0
-0.5
-1.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8
Mean ln(OC Cone)
HB=Boston High Volume LB=Boston Low Volume HC=Chicago High Volume LC=Chfcago Low Volume
HP=Philadelphia High Volume LP=Philadelphia Low Volume
Figure 5.1O
Log OC Concentrations from High and Low Volume
Samples Plotted Against the Mean of the Logs of the
Concentrations.
5.4 COLLECTION OF AMBIENT NITRATE AND SULFATE ON BLANK
NYLON FILTERS
In this experiment in addition to field, trip, and "freezer" blanks (blanks stored in a
freezer in the field), a MetOne sampler was loaded with 5 nylon modules. The five modules
were then left in the MetOne sampler for six to nine days while leaving the sampler idle. In this
way the filters were exposed to ambient air for a week and the effects of filters sitting in the
sampler for an extended period before sampling was simulated.
Figures 5.11 and 5.12 show scatter plots of the raw data for the experiment.
42
April 27, 2OO1
-------
Effects of Leaving the Filter in the Monitor
Phoenix Fresno Tampa New Vbrk Bfemark Portland Salt Lake
X-Experiment Blank F-HekJ Blank R-Freezer Blank T-Trip Blank
Figure 5.11
Effects of Leaving Blank Filters in the Sampler for a Week
on the Measured Nitrate.
Effects of Leaving the Filter in the Monitor
Phoenix Fresno Tampa New York Bismark Portland Salt Lake
Site
X=Experiment Blank F=Field Blank R=Freezer Blank T=Trip Blank
Figure 5.12
Effects of Leaving Blank Filters in the Sampler for a Week
on the Measured Sulfate.
43
April 27, 2OO1
-------
Clearly the Tampa data in this experiment has a different nature than the other sites. (See
Section 5.6.) Hence, to begin with, the Tampa data are treated as outliers and only the other data
points are modeled. The mean response was modeled as an overall mean for each blank type
with random variations due to the site. The means and standard errors for the various blank types
are provided in Table 5.4.
Table 5.4 Mean Nitrate and Sulfate on the Experiment Blanks.
Blank type
Experiment blank
Field blank
Freezer blank
Trip blank
Nitrate Mean
(micrograms)
0.5962
0.5490
0.1653
0.6063
Nitrate
Standard error
0.1163
0.1804
0.2696
0.2690
Sulfate Mean
(micrograms)
0.8497
0.4862
0.04528
0.4524
Sulfate Standard
error
0.1351
0.2687
0.4214
0.4210
Two statistical tests were conducted for both the nitrate data and the sulfate data. The
first test was a combined test that the freezer blanks were 0 and that the field and trip blanks were
the same. This is consistent with the data in both cases with p-values of 0.82 for the nitrate data
and 0.99 for the sulfate data. The second test was a test of whether or not the experiment blanks
differed significantly from the mean of the field and trip blanks. Again, this is consistent with
the data (i.e., there is no significant difference). The p-values for these tests were 0.91 and 0.17
for the nitrate data and sulfate data, respectively.
5.5 COLLECTION OF AMBIENT NITRATE AND SULFATE ON EXPOSED
NYLON FILTERS
In this experiment five modules were loaded with quartz filters for a MetOne sampler.
Two of these were recovered according to standard procedures (within 48 hours of sampling).
The remaining filters were recovered at least six days after sampling. This simulated the
condition of a sequential sampler where a sample is left in the sampler for an extended period.
44
April 27, 2OO1
-------
Figures 5.13 and 5.14 show scatter plots of the raw data for the experiment. Figure 5.15 shows
the ratio of the nitrate concentration to the sulfate concentration.
The modeling was based on all the data in this case since there are only four sites. (The
only unusual point was from one of the filters that was collected within 48 hours.) As in
experiment IB, the values were assumed to have a different mean for each site and day. The
effect of leaving the filter in the sampler was modeled as producing a shift in the site mean
(where the same shift is used for all sites). Table 5.5 shows the mean difference between the
samples that were collected within 48 hours and those that were left in the sampler for at least
six days, and the associated p-value. In both cases, there is no significant difference between the
collection methods. However, note that the negative values in Table 5.5 indicate that the samples
tended to lose mass over time. The lack of statistical significance may be due to the lack of data
for this experiment.
Table 5.5 Mean Difference Between Standard Collection of Samples and
Those Left in the Sampler.
Species
Nitrate
Sulfate
Mean difference
-1.347
-0.153
Standard error
0.653
0.176
p-value
0.0568
0.9320
45
April 27, 2OO1
-------
Effects of Leaving Sample In Monitor
Fresno
Tampa
New York
Portland
Site
0=standand recovery 1=recovery after six days
Figure 5.13
The Effects of Leaving a Filter in a Sampler for a Week on
the Observed Nitrate Concentration.
Effects of Leaving Sample In Monitor
Fresno
Tampa
New York
Site
0=standard recovery 1=recovery after sk days
Portland
Figure 5.14
The Effects of Leaving a Filter in a Sampler for a Week on
the Observed Sulfate Concentration.
46
April 27, 2OO1
-------
Effects of Leaving Sample in Monitor
Fresno
Tampa
New York
Portland
Site
0=standard recovery 1=recovery after six days
Figure 5.15
The Effects of Leaving a Filter in a Sampler for a Week on
the Observed Ratio of the Nitrate Concentration to Sulfate
Concentration.
5.6 DISCUSSION
Consider Figure 5.11. Generally speaking, the spread and the magnitude of the
experiment blanks (the ones left in the sampler for a week) are consistent with the spread and
magnitude associated with the other blanks. (Keep in mind the Os are really non-detects and
could be anything up to the MDL or even slightly larger.) There is a notable exception. There
are three experiment blanks for Tampa with approximately five micrograms of nitrate compared
to a value of about one microgram for all of the other blanks.
Is this an anomalous result (i.e., an outlier) that should be thrown out before modeling?
There are two arguments against this. First, the result appears to be replicated three times here.
Moreover, both the nitrate and sulfate values are high. In fact, similar results occur in each of the
four experiments that mimic sequential sampling conditions. Also, there a few data points in the
47 April 27, 2OO1
-------
blanks from the Task 2 data that have unusually high data. Hence, it would seem that, although
rare, these represent real values. The second argument is that the whole point of this experiment
is to guard against significant amounts of contamination.
Also, a note of caution is needed about the apparent replication. A true replicate would
ideally have been from a filter loaded in a separate sampler. In this case, there are five filters all
loaded into a common sampler. Since they are physically isolated from each other, then there is
not as much of a problem treating these as true replicates. The usual problem with this is that the
statistical model will end up estimating analytical error instead of sampling error. Since the
analytical can be much less than sampling error, the statistical basis for trying to determine what
represents a significant difference is drastically under estimated. Looking at the data for both
Sections 5.1 and 5.4 it would appear that the measurements can be used as replicates, because not
all of the experiment blanks from the associated site and day are unusually high.
As a result of the problems noted above, the measurements were modeled in several
ways. First, they were modeled with all of the data, treating everything as true replicates. Next,
they were modeled with all of the data, but with first replacing the experiment blanks from a site
and day with their average. (So instead of having five measurements, there is only one
experiment blank data point per site and day.) The third and fourth models are just as above with
the anomalous sites removed. (The model version that excluded the sites with extreme data and
treated the values as replicates was reported in the earlier sections.) The results are essentially
the same, and are exactly as you would expect from looking at the plots. If the large values are
removed, then there is little or no significant difference between the trip blanks, field blanks, and
the experiment blanks. (The freezer blanks are usually significantly less.) Otherwise the
experiment blanks can be significantly higher.
The blanks associated with the routine data showed only extremely rare occurrences of
these high values. Yet unusual values occurred more than once within the five experiments.
Hence ,it would seem that the correct answer is not to treat the unusual values here as outliers.
The correct conclusion should be that there is significantly more sampling variability than would
be indicated by the modeling results because these "outliers" should be included. (The results in
48 April 27, 2OO1
-------
Section 3.3 include these values.) This leads to the same answers that there is little or no
significant difference in the mean response, except for OC.
49 April 27, 2OO1
-------
6.O SYNTHESIS
We begin by summarizing the major findings. First, there are significant site to site differences
that affect all parts of the assessment. At the Phoenix site, 13 of the 45 dates with data were flagged
for inconsistency, while at the Fresno site only 3 of the 48 days with data were flagged. The evidence
points to site-to-site differences that do not depend on the combination of samplers. Any of the results
that show a significant site-to-site dependence need to be taken with caution because the sampler types
are not distributed evenly across sites. Second, not all parameters behave the same. They do,
however, frequently group as one would expect: soil components tend to be similar; sulfate, nitrate,
and ammonium sometimes show similar results; and EC and OC frequently have similar results.
Given that there are fairly consistent biases between co-located samplers (on a log scale) the
most important findings are:
• The measured PM2.5 mass was compared with co-located FRM measurements.
Seventeen out of 24 of the samplers met the Expert Panel data objective of an R2 value
of at least 0.9 in a linear regression of the mass values against the FRM measurement.
Deviations from this criteria appear to be caused by site influences that affect all the
monitors at a site, rather than differences among sampler types.
Only half of the samplers met the Expert Panel data objective that the ratio of the FRM
mass mean to the speciation sampler mass mean be at least 0.9 and at most 1.1. The
ratios tested strongly dependent on both site and sampler type. In all cases with co-
located FRMs, the means for the mass followed the following ordering: URG <
Andersen < MetOne. Six of the seven URG means were less than the corresponding
50 April 27, 2OO1
-------
FRM mean, all eight MetOne means were greater than the corresponding FRM mean,
and six of the nine Andersen means were greater than the corresponding FRM mean.
The concentration ordering noted for the mass applies to most of the species, namely
URG < Andersen < MetOne. Moreover, while parameter specific, the percent of the
time that this relation holds is consistent across sites. The exceptions to this ordering
are chlorine, zinc, ammonium, and sulfate. For each of these exceptions, the percent of
the time that the sampler types have one relationship or another varies by site. Of the
species that do follow the general ordering above, only the nitrate data showed site to
site differences in the percent of the time one sampler type is above another.
For all species, the magnitude of the biases between sampler types is strongly site
dependent from a statistical point of view. The magnitudes are summarized in the
appendix by site and species so that the practical significance can be assessed.
The variability found in the sampling precision across sampler types is probably due to
site influences, but is probably not generally of any practical concern.
The blanks generally do not show site to site differences. The trip blanks and field
blanks are generally about the same. The URG blanks tend to be the cleanest except
for nitrate. Nearly all of the "dirtiest" 25 percent of the blanks were from the URG
samplers. The practical difference among the sampler types needs to be assessed
separately.
The five special experiments all suffer from a lack of data. As such, modeling results
are not robust against the inclusion or exclusion of outliers. Assuming that the outliers
have been properly identified, then there is little or no significant effect on sulfate,
nitrate, elemental carbon, or organic carbon concentrations found with leaving filters in
51 April 27, 2OO1
-------
the sampler for an extended period either before or after sampling. The only statistically
significant difference found was that blanks left for a week in the sampler collected on
average an extra 3.25 micrograms of organic carbon.
• Qualitatively, both the face velocity experiment and the sampler to sampler comparisons
suggest that measurements of carbon from low volume sampling yield higher
concentrations than high volume sampling. The data from the sampler to sampler
comparisons showed more consistency than the data from the face velocity experiment.
All of these findings need to be assessed for their practical significance. The tables in
Appendix B provide estimates that show the magnitude of the differences observed. It may be that
because there were over 1,000 days worth of data to work with the differences detected by the
statistical techniques are not of practical significance. The standard errors shown indicate the level of
sensitivity for the statistical tests. If the standard errors are an order of magnitude less than "practical"
differences, then the declared differences may not have any practical meaning. On the other hand, if the
standard errors are approximately equal to or greater than what is considered a significant practical
difference, then the findings above have practical implications.
Both absolute and relative differences should be considered. However, it may be easier to
eliminate the cases where the site medians (Table B.I) are less than about one-tenth of the MDLs, or
some other nominal value, so that only relative differences need to be considered. For example, the
differences noted for aluminum, lead, and tin probably do not have any practical implications, since the
data are all very close to or below the MDL. On the other hand, the sulfate data are well above the
MDL. However, the relative differences are mostly below 10 percent, so these may not be of practical
significance either.
52 April 27, 2OO1
-------
Finally, it may also be helpful to rank the species by data user needs. The species have very
different impacts on visibility and very different relative risks. Such characteristics may guide the level
of acceptable differences between samplers.
53 April 27, 2OO1
-------
APPENDIX A:
GUIDE TO THE GRAPHICAL OUTPUT
FROM TASKS 1 AND 2
April 27, 2001
-------
Table of Contents
APPENDIX A: GUIDE TO THE GRAPHICAL OUTPUT FROM TASKS 1 AND 2
A.1 EXPLORATORY PLOTS
A.2 DATABASE DICTIONARY
A.3 STEPS TAKEN TO PRODUCE GRAPHS
Outline for the m2mc graphs containing the Routine and FRM data . . .
Outline for the m2mc graphs containing the FIELD and TRI P BLANK
data
Outline for the boxplots containing the Routine and FRM data
Outline for the log_FRM plots containing the Routine and FRM data . . .
Outline for the cddvspress and cddvstemp plots containing the Routine
and FRM data
Outline for the cdpress and cdtemp plots containing the Routine and
FRM data
Outline for the parmbp and parmbps plots containing the Routine and
FRM data
Outline for the mec plots containing the Routine and FRM data
Outline for the vmvstmc plots containing the Routine and FRM data . . .
Outline for the smvstmc plots containing the Routine and FRM data . . .
A-19
A-20
A-21
A-22
A-23
A-24
A-25
A-26
A-27
List of Figures
Graph 1: Centered Daily Differences Versus Pressure. These plot the deviation
from the daily mean of a monitor's parameter value versus the average
pressure from all the monitors at the site A-2
Graph 2: Centered Daily Differences Versus Temperature. These plot the
deviation from the daily mean of a monitor's parameter value versus the
average temperature from all the monitors at the site A-3
Graph 3. Centered Daily Pressure Versus Date. These show a time series of the
daily average pressure and the deviations from the mean by each
monitor A-4
Graph 4. Centered Daily Temperature Versus Date. These show a time series of
the daily average temperature and the deviations from the mean by each
monitor A-5
Graph 5. Daily Range Boxplots. Boxplots of the difference between the daily
maximums and the minimums of a parameter for each site A-6
Graph 6. Comparisons Between Co-located Monitor Blanks. These plot the daily
average parameter mass of the blanks against the minimum and
maximum values observed (connected by a vertical line). The letter
indicates either the first letter of the vendor or a "T" for a trip blank.
The daily means are connected by the diagonal line. The MDL is
indicated with dashed rectangular boxes A-7
Graph 7. Speciation Monitor Measured Mass Versus FRM mass. These are
scatter plots of the total mass as measured by a monitor versus the co-
located FRM measurement. The POC number is indicated on the graph . . A-8
Graph 8. Comparisons Between Co-located Monitor Blanks. These plot the daily
average parameter mass of the routine data against the minimum and
maximum values observed (connected by a vertical line). The daily
means are connected by the diagonal line. The letter indicates either the
first letter of the vendor. The MDL is indicated with dashed rectangular
boxes A-9
A-ii
April 27, 2001
-------
Graph 9. Checks of Measurement Error Correlation. These are scatter plots of a
monitor's deviation from the site mean on one versus the difference
measured 3 days later A-10
Graph 1 O. Boxplots of Parameter Values by Monitor ID. These are side-by-side
notched boxplots of the log-concentration for a parameter. The notch is
an approximate 95 Percent confidence interval for the median. (In the
example shown, the notches for two monitors at the Tampa site do not
overlap. Hence, these have significantly different medians.) A-11
Graph 11. Boxplots of Parameter Values by Site. These are side-by-side notched
boxplots of the log-concentration for a parameter. The notch is an
approximate 95 Percent confidence interval for the median. (In the
example shown, the notches for Seattle and Salt Lake City do not
overlap. Hence, these have significantly different medians.) A-1 2
Graph 1 2. Partial Reconstructed Mass Versus Total Mass. These are boxplots of
the log of the partial reconstructed mass (based on the 13 study
parameters) over the measured mass for each monitor and site A-1 3
Graph 13. Partial Mass Versus Total Mass. These are boxplots of the log of the
parameter mass over the measured mass for each site A-1 4
A-iii
April 27, 2001
-------
APPENDIX A: GUIDE TO THE GRAPHICAL OUTPUT FROM
TASKS 1 AND 2
This appendix contains examples of the major graphs considered throughout the project to
assess the comparability of the sampler types. The three sections are the exploratory plots (the
examples), a data dictionary, and the steps taken to produce the graphs. The graphs described will
accompany the final report on CD. Until then, they are available at the Mini Trends website,
www.sdas.battelle.org/minitrends.
A-1 April 27, 2OO1
-------
A.I EXPLORATORY PLOTS
c
55
ME
C
TJ
-------
C.5-
C.4-
C.3-
J>
5
Jta
Q
Q
~"j
3>
O
01-
.3-
.4-
on
entered Daiy Cifere'css Ve'SJS Tsmpsralre
SITE= Chbags L (SE Poles Sfetion)
ac.B
Temoeraure
-Ft-
•/ ^,-j
x«o
Graph 2: Centered Daily Differences Versus Temperature. These
plot the deviation from the daily mean of a monitor's
parameter value versus the average temperature from all
the monitors at the site.
A-3
April 27, 2001
-------
Certered 2ait/ ^essute Versus Date
SrE=Chbagc,!L (SE Poloe Slat on)
i
5
•ji
OL
^
'1
C
a>
5
5
o
1000-
9 00 -
8 00 -
700-
6 00 -
5 00 -
4 00 -
3 00 -
2 00 -
1 00 -
0 00 -
-1 00 -
-2 00 -
-300-
-400-
-500-
/w^*A A* ** « '• •* ••*/***
' M m \ , i \ f 'i . * *
* "* * k^ 4. #
i B
S
5 88 i i 6
5 E E i
5 5 E5 i B B SB f B 66 E B ii
3 3! 1 3Blt ! ISI5B 1 IlEE'tlttftfES
E E EB i 5 5 15 C 5 55 5 5 SS
B 5 E i
B 55 ] 55
1
i 5
- 803.00
- 793.00
- 783.00
- 773.00
- 763.00
- 753.00 t
- 743.00 |
- 733.00 i
- 723.00 *
- 713.00 D
ft)
- 703.00 ,jf
- 693.00 ^
- 683.00 5
- 673.00 8
C
- 6S3.00 ;
- 653.00 "
- 643.00
- 633.00
- 623.00
- 813.00
- 603.00
Date
5= Ancerscn 8= URG
Graph 3. Centered Daily Pressure Versus Date. These show a time
series of the daily average pressure and the deviations
from the mean by each monitor.
A-4
April 27, 2001
-------
Centred Daily Tempa-aire Versus C&te
SrE=Chicagc,IL (SE Poke Steton)
»
a)
5
$
5
a
1
>^
1
c
a>
S
H
O
10 00 -
9 00 •
8 00 -
7 00 -
6 00 -
5 00 -
4 00 -
3 00 -
2 00 -
1 00 -
0 00 -
-1 00 •
-2 00 -
-3 00 -
-4 00 -
-5 00 -
- 4100
- 35.00
- 3D.OO £
f cf
* 9 l S
\. « *,, f r\ , f *•, ft (( - 25.00 (Q
/ <. ' ^ ¥ V ' ^ ; V ^ i \ ' 1 ' * * ! ^ ^*
. ' >! f\ i' , / '.,' ' I if' \i D
« .;; / \ ) U i \ \ - 2100 e.
' I i ' ^ ' l! ' '^
1 >/ \! J - 15.00 ^
*^ * ~"
I i S S 1 5 Slf.i'HSrj i icS i jg, i|,Sj 5. t n rtfl -
1 ' i ' ' B I ' jBiSB5J,ifEB" S'E'S sSllSSBSBjlllSSjSliliiiiSS SilBIH 1J.UU g
f"1', .' \ / '• i •! $
' ' I V *•! * - 500 »
f \ ! 5
' l i
4 \'m - 100
- -5.00
ill
tqovzoDO OW/SODC op 12x0 optecoo uroracoo osjoraooo ostDraooo
Date
5=Ancersoi S= URG
Graph 4. Centered Daily Temperature Versus Date. These show a
time series of the daily average temperature and the
deviations from the mean by each monitor.
A-5
April 27, 2001
-------
3-
DC •?
Daily Farge Doi =1cte
-ARM=ALmnhu~)
I I I I I I
i;(N,«3 c* IM^J n (M,J: IL (*,u) ma (i,wi NO (HA)
I I
-JD (N.uj en
Oite
n.uj UT tuf-i
Graph 5. Daily Range Boxplots. Boxplots of the difference
between the daily maximums and the minimums of a
parameter for each site.
A-6
April 27, 2001
-------
'iscns Deti,«en Co— ocatsc Hcnitcj' Dfenks
parrr= AU'nrun SITE= ChbagcJL (3E Pofco Stoton"
J.13-
!. 10 -
0 0-' -
0.01 -
-002-
i- <-
-0 020
C.038
O.CSS
0/20
Graph 6. Comparisons Between Co-located Monitor Blanks. These
plot the daily average parameter mass of the blanks
against the minimum and maximum values observed
(connected by a vertical line). The letter indicates either
the first letter of the vendor or a "T" for a trip blank.
The daily means are connected by the diagonal line. The
MDL is indicated with dashed rectangular boxes.
A-7
April 27, 2001
-------
1,0-|
1.5
1.4-1
Special ion Vic niter Mea sired Mass Versus FRM Mass
(Sf Pdicf= fihrto-)
_
0.4 0.5 O.fi
1 i I I i
.7 Of: 0.9 1.0 1.' 1-
1.4 15 1.0
5— Ancerson 6— UP3
Graph 7. Speciation Monitor Measured Mass Versus FRM mass.
These are scatter plots of the total mass as measured by
a monitor versus the co-located FRM measurement. The
POC number is indicated on the graph.
A-8
April 27, 2001
-------
Comparisons Getwser Co—heated Monto-s
fi\3M=ALminum SITE=Chor,go,IL iSE Pol bo Stalion)
s
t
55
"^
o
-o /t- -
— 1.17 -
—i t.y -
-200-
;> ,j j .
—283 -
—
1
— Or — H
V^
I
<
1
r ,*.«
J.oJ
'
r-f*
t
^
^
^
x^
I
— £ ^4
j
^
\.
^
J-"""
>^
, , ix* J
( J*^T
,4 XC0
Jr^ «-
nj
3
111)
1 33 1 £""> —1 t^ rt
^^ 1 . ^?^f ^^ I ,-JL. ^^ \ , U ^^ V
Graph 8. Comparisons Between Co-located Monitor Blanks. These
plot the daily average parameter mass of the routine
data against the minimum and maximum values
observed (connected by a vertical line). The daily means
are connected by the diagonal line. The letter indicates
the first letter of the vendor. The MDL is indicated with
dashed rectangular boxes.
A-9
April 27, 2001
-------
D
o
o
Q
LJ
Checks of MeEsurerrert zrror Correlation
-AlU-Alu-niniim finr-Ohcsgo.ll (fif Pd os
05-"
04-
03-
02-
3,1-
00
-3.1-
•02-
-03-
-04-
_0.4 -n.3 -i
.? -n.1 o.o o.i n.?
Previous nFej-HicF; frnrn ft^ih/
5- Anderson
o.a
04 0.5
Graph 9. Checks of Measurement Error Correlation. These are
scatter plots of a monitor's deviation from the site mean
on one day versus the difference measured 3 days later.
A-10
April 27, 2001
-------
B«>: F1»te *f Parameter v'dtes by M->nit*f D
= Aluminum
C -
r -- -
—4 ~-
,1
r
L
MCf: HUH
j* ik, AZ
Mo- i-.H Nrf
=ita iu, OA Ta i
I pr; ArH MPfi
=L Ct iunju, IL
Busiui, MA
ArH
tuis MC
Graph 1O. Boxplots of Parameter Values by Monitor ID. These are
side-by-side notched boxplots of the log-concentration
for a parameter. The notch is an approximate 95
Percent confidence interval for the median. (In the
example shown, the notches for two monitors at the
Tampa site do not overlap. Hence, these have
significantly different medians.)
A-11
April 27, 2001
-------
Plo-ts of Parameter Valjss k>y
=Aluminun
r
c
-2 -
-6 --
- I >=<
D J_
Ji: Fretn.Ol Tints Fl Chia.L Eati.Hl 3tL«j.MO Nw^hV Bnml.ND Psri.OK PHIs.PA loiH.TX Sstt.UT S.-rt.Vtt
Site
Graph 11. Boxplots of Parameter Values by Site. These are side-
by-side notched boxplots of the log-concentration for a
parameter. The notch is an approximate 95 Percent
confidence interval for the median. (In the example
shown, the notches for Seattle and Salt Lake City do not
overlap. Hence, these have significantly different
medians.)
A-12
April 27, 2001
-------
Partial Reconstructed Mass Versus Total Mass
'tf''
u.
tu
i
S C'~
LL
'51s
&
E
r
_i
III
f * \
o
0 8 .^ n...| MrtOlr L DQ JUi^Dvn U HC j.l -.^jn..| U|ljnl..,l UHQ M^.Cll- AnJllD.. I
l-hrpfpy, J!^ — Fsnn, (Ui larnp;, l-l ia>irar|i II Histm, MA Stlnjis, MD
Graph 12. Partial Reconstructed Mass Versus Total Mass. These
are boxplots of the log of the partial reconstructed mass
(based on the 13 study parameters) over the measured
mass for each monitor and site.
A-13
April 27, 2001
-------
Partial Mass Yarsus Tcta Mass
Ch cc.go IL (SE Pel bo Stedon'i — Vendor = Anckstocr
Moiitcr = 5
-2
.0
1
I
i
-4
It
il
Or$m
I I I
[••so Ctafejjm *^!t.mr
I I
Graph 13. Partial Mass Versus Total Mass. These are boxplots of
the log of the parameter mass over the measured mass
for each site.
A-14
April 27, 2001
-------
A.2 DATABASE DICTIONARY
Database Dictionary
Variable Name
Type
Farm
Site
Zdate
Meth
Vendor
POC
Value
Units
MDL
Brief Description of Variable Contents
Measurement Type
Routine or FRM
5-digit AIRS Parameter Code
Total of 66 different parameters
9-digit AIRS Site ID
• (ST-CTY-SITE#)
Recorded within the database without hyphens
Total of 13 sites
o Each site has co-located monitors
Actual Sample Date
Formatted mm/dd/yyyy
3-digit Collection/Analysis Method Code
Distinct AIRS method code for each of the collection/ analysis methods
Total of 1 6 methods
Speciation Sampler Design
Andersen
• URG
MetOne
1 -digit Parameter Occurrence Code
Distinguishes co-located instruments
5, 6, or 7
Sample Values
Units of measure
Minimum Detection Limits for each target analyte
Specific to species/vendor
Total of 60 parameters with an associated MDL
o Parameter 88309 and 883 10 only have an associated URG MDL
o Total of 6 temperature and pressure parameters do not have an
MDL
A-15
April 27, 2001
-------
Variable Name
Brief Description of Variable Contents
BATTELLE FLAG1
Indicates monitor to monitor inconsistency
• If a specific site, date, POC, and parameter was flagged, then all
corresponding observations were also flagged
o The alphanumeric flag contains letters corresponding to each
parameter that was responsible for causing the site and date to
be flagged
• Alphanumeric flag that represents parameters:
A = PM2.5Mass (88101)
B = Aluminum (88104)
C = Ammonium (88301)
D = Calcium (88111)
E = Chlorine (88115)
F = Elemental Carbon (88307)
G = Iron (88126)
H = Lead (88128)
I = Nitrate (88306)
J = Organic Carbon (88305)
K= Silicon (88165)
L = Sulfate (88403)
M=Tin (88160)
N = Zinc (88167)
BATTELLE FLAG2*
Indicates that the reported mass concentration of one of the species was
significantly greater than the mass concentration of the sample
• If a specific site, date, POC, and parameter were flagged, then all
corresponding observations were also flagged
o The alphanumeric flag contains letters corresponding to each
parameter that was responsible for causing the site, date, and
POC to be nagged
o (the same alphanumeric parameter associations as for
BATTELLE_FLAG1)
BATTELLE FLAG3*
Indicates that the sum of the masses of the 13 components that we were
considering was significantly greater than the total reported mass
• Takes on a value of 1 or • (missing)
Care needs to be taken in using these flags in areas that have high nitrate concentrations (e.g. in the
Southwest). The reconstructed mass is based on multiple filters that can account for the volatilization of the
nitrate.
A-16
April 27, 2001
-------
A.3 STEPS TAKEN TO PRODUCE GRAPHS
Outline for the m2mc graphs containing the Routine and FRM data
(title: Comparisons Between Co-located Monitors)
General Overview: plots of each day's observations where the y axis corresponds to the Iog10
transformation of value and the x axis corresponds to the day's mean Iog10 transformation of value; on
any given day there will be either 2 or 3 points observed (depending on how many pocs were recording
at that site) and labeled with the corresponding vendor's initial; a vertical line connects the min and max
values for the day and a 45 degree line connects the means across all days; MDL lines are represented
on each graph as a vertical and a horizontal line that meet at a point on the 45 degree line - each MDL
line is also labeled with the corresponding vendor's initial.
Step 1.
Read in Routine and FRM data from the original dataset while eliminating all null observations
(which are obvious outliers), keeping only the 14 parms of interest, keeping only days/sites with
at least two vendors present, and only keeping observations where value>0
o Eliminated all RTT flagged observations (known outliers)
o Assigned a vendor name and an MDL value to each observation depending on the site
and poc #
o Get only the parm*site*zdate combinations where there were more than one vendor
(thus to compare vendors) - 13 sites remain
o Transformed the concentration value to the Iog10 scale; the resulting variable:
Iog10_value
A-17 April 27, 2OO1
-------
Step 2.
Find specific parm*site mins, maxs, and means (of Iog10_value) to get:
1. Range for plot axes
The y axis for each parm*site differs
i. The axis goes from the minimum Iog10_value over all days to the
maximum Iog10_value over all days (for that particular parm*site
combination)
The x axis for each parm*site differs
i. First, the mean Iog10 value was found for each parm*site*zdate (i.e.,
within a given parm*site combination, the mean value was computed
for each day)
ii. Of these daily means, the minimum mean value and the maximum
mean value were found and used as the axis boundaries
2. Reference lines for daily means
Creates a straight 45 degree reference line on the graph connecting the daily
means (i.e.,where y=x)
3. Vertical bars for the daily range
Connects the daily minimum and maximum values with a straight vertical line
4. MDL vertical and horizontal lines
The MDL lines for each vendor*parm differs
i. Plots the MDL line vertically from the x axis and horizontally from the
A-18 April 27, 2OO1
-------
y axis and connects at the reference line
ii. Each MDL line is labeled with its corresponding vendor's initial
Step 3.
Each y*x=group is plotted and labeled with the corresponding vendor
o The variable y is the Iog10 transformation of the original value while the variable x is the
mean Iog10_value, or the mean of all Iog10_values for a particular day
o Each vertical connecting line represents one day of recording - a "summary" of each
days' observations
Outline for the m2mc graphs containing the FIELD and TRIP BLANK data
(title: Comparisons Between Co-located Monitor Blanks)
General Overview: Basically the same plots as for the m2mc with Routine and FRM data, but with a
few minor changes
Changes:
Changed the MDL value to the MDL_blank value with the formula:
MDL_blank = (cone. MDL)(flow rate in L/min)(1.44 to convert to micrograms) = MDL in
micrograms
Trip blanks were labeled as their own "group" in the plots, similar to the vendor labeling except
with a "T"
A-19 April 27, 2OO1
-------
The axes were set on the plot to extend above or below the minimum and maximum points,
whether they be the MDL lines or the actual values
Outline for the boxplots containing the Routine and FRM data
(title: Daily Range Boxplots)
General Overview: Daily range boxplots were created for each parm to see site-to-site comparisons
of the (maximum Iog10 value - minimum Iog10 value) differences.
Using the dataset (d) that only contains observations for the 14 parms of interest, only days/sites with at
least two vendors present, observations where value>0, no null observations, the Iog10 transformation
of the original value, and only non-RTI flagged data:
Found the minimum and maximum observations of Iog10 value (using the proc means
procedure) for each parm*site*zdate combination
• From these, a new variable called 'diff was created by taking the maximum Iog10_value and
subtracting the minimum Iog10_value
For each parm separately, 'diff was plotted against site
o Boxplots with whiskers and endlines were created for each set of points (parm*site
specific); observations more than 1.5 iqr away from this box were represented by a star
o Sites were labeled with the site's state abbreviation, as well as the first initial of the
vendors that were present at that site
A-20 April 27, 2OO1
-------
Outline for the log_FRM plots containing the Routine and FRM data
(title: Speciation Monitor Measured Mass Versus FRM Mass)
General Overview: Scatter plots of daily values for each site's Iog10 transformed speciation mass vs
Iog10 transformed frm mass.
Using the dataset (d) that only contains observations for the 14 parms of interest, only days/sites with at
least two vendors present, observations where value>0, no null observations, the Iog10 transformation
of the original value, and only the non-RTI flagged data:
Plotted the Iog10 transformed speciation mass against Iog10 transformed frm mass
Added a 45° line to the graph
o Found the minimum and maximum Iog10 transformed FRM values, by site
o In order to plot these two points, set the x variable (Iog10 FRM) was set to equal the
minimum Iog10 FRM value and then set the y variable (Iog10 value) equal the minimum
Iog10 FRM value and designated it as a different poc number to allow for connecting the
points later with Interpol=join; this was repeated similarly for the maximum Iog10 FRM
value
o Each plotted point on the graph was labeled with its corresponding poc number
o The footnote denoted the vendor-to-poc mapping information for that particular site
A-21 April 27, 2OO1
-------
Outline for the cddvspress and cddvstemp plots containing the Routine and FRM data
(title: Centered Daily Differences Versus Pressure/ Centered Daily Differences Versus
Temperature)
General Overview: Plot, by parm and site, the centered daily differences for a given parm versus the
average daily pressure or temperature.
• Read in Routine and FRM data from the original dataset while eliminating all null observations
(which are obvious outliers), keeping only the 14 parms of interest, keeping only days/sites with
at least two vendors present, and only keeping observations where value>0
o Eliminated all RTT flagged observations (known outliers)
o Assigned a vendor name and an MDL value to each observation depending on the site
and poc #
o Get only the parm*site*zdate combinations where there were more than one vendor
(thus to compare vendors) - 13 sites remain
o Transformed the concentration value to the Iog10 scale; the resulting variable:
Iog10_value
o Created a separate dataset with only 2 parms, daily temperature and barometric
pressure
• With the temp/press dataset, found average temperature and pressure by site and date; created
separate datasets for temp and pressure
A-22 April 27, 2OO1
-------
Found daily mean (by parm and site) for the concentration values and created the centered
daily values (i.e., Iog10_value -Iog10_value_mean)
Merged the centered daily values data with the mean temp and pressure data (separately)
Plot, by parm and site, the centered daily differences vs the average daily temp or pressure
o Labeled each point with its poc number and then connected each point
o The footnote denoted the vendor-to-poc mapping information
Outline for the cdpress and cdtemp plots containing the Routine and FRM data
(title: Centered Daily Pressure Versus Date / Centered Daily Temperature Versus Date)
General Overview: Plot, by site, the centered daily pressure or temperature values versus date.
• Read in Routine and FRM data from the original dataset while eliminating all null observations
(which are obvious outliers), keeping only sites with at least two vendors present, and keeping
only the 2 parms of interest
o Eliminated all RTI flagged observations (known outliers)
• Found the mean temp or pressure for each parm*site*zdate*poc
• Found the daily mean temp or pressure for each parm*site*zdate
A-23 April 27, 2OO1
-------
Merged these two datasets together and calculated the centered daily difference for temp and
pressure separately
Plotted the centered value versus the date
o Labeled each point with the poc number
o The footnote denoted the vendor-to-poc mapping information
Plotted, on the same graph, the daily average temp or pressure versus the date and connected
the points
o This allowed us to see how the centered values were behaving while having an idea of
what the actual averages were doing
Outline for the parmbp and parmbps plots containing the Routine and FRM data
(title: Boxplots of Parameter Values by Monitor ID/ Boxplots of Parameter Values by Site)
General Overview: Boxplots were created for each parameter, of the concentration values, by either
monitor id or site.
Using the dataset (d) that only contains observations for the 14 parms of interest, only days/sites with at
least two vendors present, observations where value>0, no null observations, the Iog10 transformation
of the original value, and only the non-RTI flagged data:
Created a monitor id number for each observation using the formula mon_id=(site* 10)+poc2
A-24 April 27, 2OO1
-------
o Monitor id is a unique number specifying a specific site and poc
o With this monitor id, an index was created to more simply specify the monitor id number,
thus assigning it a number 1 — 40
Boxplots were created for each parameter by plotting the Iog10 transformation of value against the
index (or monitor id)
o Note that the plot for each parm spanned two pages (i.e., two separate jpegs) since there
were numerous index values
Boxplots were also created for each parameter by plotting the Iog10 transformation of value against
the site (i.e., over all monitors at a given site)
Outline for the mec plots containing the Routine and FRM data
(title: Checks of Measurement Error Correlation)
General Overview: Scatter plots of measurement error for a given parm*site*poc2; looking at the
day's centered Iog10 transformed value versus the centered Iog10 transformed value from three days
ago; there is logically a correlation between the values from day to day since it will take more than a
day to increase or decrease a concentration, but there should not be a correlation between the centered
values
Using the dataset (d) that only contains observations for the 14 parms of interest, only days/sites with at
least two vendors present, observations where value>0, no null observations, the Iog10 transformation
of the original value, and only the non-RTI flagged data:
A-25 April 27, 2OO1
-------
• Found the daily mean Iog10 transformed value for each parm*site*zdate
Merged these two datasets together (d and the daily mean dataset) and calculated the centered
daily difference
From this merged dataset, each observation was set into a separate data set according to POC
type (i.e., all POC=5, POC=6, POC=7, and POC=9 (FRM))
o Looking at each parm*site*zdate separately, the zdate was compared to the previous
observed date; if this observed date was 3 days previous, then the centered difference
for that day was labeled 'pre_val'; thus for each observation, there is a centered value
and a 'pre_val' if the previous date is only 3 days earlier (else, there is no prejval
associated with this observation)
o These four separate datasets were then set back together to form one complete dataset
• Plotted the centered value versus the previous centered value (if there was one) for each
parm*site*poc
o Labeled each point with the poc number
o The footnote denoted the vendor-to-poc mapping information
Outline for the vmvstmc plots containing the Routine and FRM data
(title: Partial Mass Versus Total Mass)
General Overview: Boxplots were created of the variable mass divided by total PM mass for each
parameter by site, date, and monitor
A-26 April 27, 2OO1
-------
Read in Routine data from the original dataset while eliminating all null observations (which are obvious
outliers), keeping only the 14 parms of interest, and only keeping observations where value>0
Eliminated all RTT flagged observations (known outliers)
Assigned a vendor name and an MDL value to each observation depending on the site and
monitor
• While looking at PM mass separately, found each site and date combination where there was
more than one monitor present (i.e., for a given site and date, at least two monitors were
recording) and only kept those sites and dates
o These observations were merged with the remaining 13 variables by site, zdate, and
monitor
P The variable Log_Val was created by dividing each parameter's value by the
PM mass value and taking the Iog10 transformation of the quotient; this was
done for each site*zdate*monitor separately
Plotted this variable, Log_Val, versus the associated parameter, for each
parameter* site*vendor*monitor combination
Outline for the smvstmc plots containing the Routine and FRM data
(title: Partial Reconstructed Mass Versus Total Mass)
General Overview: Boxplots were created of the summed parameter values divided by the total PM
mass, by site and vendor
A-27 April 27, 2OO1
-------
Took the dataset previously created for the vmvstmc plots which had site*zdate*monitor combinations
with more than one monitor observed
The variable Sum_Val was created by summing the values of all parameters for a given
site*zdate*monitor combination
Another variable, Log_CM, was created by dividing the summed value by the PM mass value
and taking the Iog10 transformation of the quotient; this was done for each site*monitor
separately
Plotted this variable, Log_CM, versus a variable called Count (a number unique to a particular
site*vendor*monitor)
A-28 April 27, 2OO1
-------
APPENDIX B:
TABLES OF SUMMARY STATISTICS
April 27, 2001
-------
Table of Contents
APPENDIX B: TABLES OF SUMMARY STATISTICS B-1
B.1 Median Relative Differences between Samplers for each Site and
Parameter B-2
B.2 Mean Differences Between Sampler Types for Each Site and Parameter . . . B-9
B.3 Sampler Type Means for all Sites and Parameters B-1 7
B.4 Significance of the Difference in Relative Composition of the Mass
Constituents by Site B-27
B-ii April 27, 2OO1
-------
APPENDIX B: TABLES OF SUMMARY STATISTICS
This appendix consists of four tables. Table B.I shows the site median concentrations and
maximum sampler MDL for each parameter. This is followed by the median relative differences between
the samplers. This is defined as the median difference in concentration divided by the median for the site.
Table B.2 shows mean differences calculated for the statistical models (for this table, all flagged data was
removed). Table B.3 has the site means for each sampler type. Finally, Table B.4 extends Table 3.5 to
all parameters.
B-1 April 27, 2OO1
-------
Table B.I Median Relative Differences between Samplers for each Site and Parameter
Sampler type for
Parameter
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Calcium
Site
Fresno
St.Louis
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Fresno
St.Louis
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Fresno
Pair
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
Samplers
MetOne
MetOne
Andersen
Andersen
MetOne
Andersen
Andersen
Andersen
Andersen
URG
MetOne
MetOne
MetOne
MetOne
MetOne
Andersen
Andersen
MetOne
Andersen
Andersen
Andersen
Andersen
URG
MetOne
MetOne
MetOne
MetOne
Site Maximum
Sampler6 Sampler? Median MDL
Andersen
Andersen
MetOne MetOne
MetOne
Andersen
URG
Andersen URG
URG
URG
URG MetOne
URG
URG
URG
Andersen
Andersen
MetOne MetOne
MetOne
Andersen
URG
Andersen URG
URG
URG
URG MetOne
URG
URG
URG
Andersen
10.7749
15.2749
11.9236
9.1515
6.4351
14.0000
9.7000
12.5083
12.3024
8.0869
12.0103
5.5000
7.2000
0.0328
0.0258
0.0100
0.0139
0.0546
0.0153
0.0065
0.0109
0.0357
0.0977
0.0117
0.0104
0.0104
0.0526
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
Relative median difference between samplers
5 to 6 5 to 7 6 to 7 StoFRM 6toFRM 7 to FRM
1040
1040
1040
1040
1040
1040
1040
1040
1040
1040
1040
1040
1040
0109
0109
0109
0109
0109
0109
0109
0109
0109
0109
0109
0109
0109
0035
12
1
-8
-9
4
4
3
11
10
-1
10
22
24
-8
-10
-21
-6
11
19
54
70
68
-9
67
83
28
-0
3% .
6% .
0% -13.1% -2.3%
7% .
6% .
6% .
1% 3.8% 3.8%
1% .
6% .
0% -24.5% -20.8% .
4% .
9% .
6% .
4% .
7% .
6% -36.7% -2.7% .
9% .
7% .
6% .
9% 207.9% 78.1% .
0% .
9% .
4% -55.3% -18.2%.
6% .
6% .
9% .
8% .
9.3%
11.3%
2.1%
21.6%
-6.2%
-8.3%
8.0%
14.1%
9.5%
12.0%
19.6%
-1.7% .
2.7% .
11.6% 17.1%
12.9% .
-9.3% .
-3.2% -12.0%
0.7% .
0.7% .
-1.0% .
-8.9% .
-3.4% .
B-2
April 27, 2001
-------
Sampler type for
Parameter
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Iron
Iron
Iron
Iron
Site
St.Louis
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Fresno
St.Louis
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Fresno
St.Louis
New York
Portland
Pair
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
Samplers
MetOne
Andersen
Andersen
MetOne
Andersen
Andersen
Andersen
Andersen
URG
MetOne
MetOne
MetOne
MetOne
MetOne
Andersen
Andersen
MetOne
Andersen
Andersen
Andersen
Andersen
URG
MetOne
MetOne
MetOne
MetOne
MetOne
Andersen
Andersen
Sampler6 Sampler?
Andersen
MetOne MetOne
MetOne
Andersen
URG
Andersen URG
URG
URG
URG MetOne
URG
URG
URG
Andersen
Andersen
MetOne MetOne
MetOne
Andersen
URG
Andersen URG
URG
URG
URG MetOne
URG
URG
URG
Andersen
Andersen
MetOne MetOne
MetOne
Site Maximum
Median MDL
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1127
0382
0311
1461
0801
0327
0359
0620
1310
0581
0283
0340
0038
0077
0044
0113
0100
0087
0072
0047
0397
0449
0087
0007
0111
0885
1336
0834
0519
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0035
0035
0035
0035
0035
0035
0035
0035
0035
0035
0035
0035
0058
0058
0058
0058
0058
0058
0058
0058
0058
0058
0058
0058
0058
0020
0020
0020
0020
Relative median difference between samplers
5 to 6 5 to 7 6 to 7 StoFRM 6toFRM 7 to FRM
-4
-10
-2
4
19
-4
45
50
-3
29
60
27
30
-51
-6
30
-45
7
-1
37
38
-1
141
259
46
-1
-9
7%
0%
7%
8%
5%
1%
5%
5%
0%
2%
0%
9%
7%
3%
4%
1%
5%
5%
4%
3%
4%
0%
0%
8%
2%
6%
7%
-9.4%
2
5%
-12.0% -0.4% .
32.9% 28.4% .
-38.1% -31.6%.
30.5% 21.5%.
48.5% 51.1%.
-10.3% -6.4%.
-4.6% 1.2%.
B-3
April 27, 2001
-------
Sampler type for
Parameter
Iron
Iron
Iron
Iron
Iron
Iron
Iron
Iron
Iron
Lead
Lead
Lead
Lead
Lead
Lead
Lead
Lead
Lead
Lead
Lead
Lead
Lead
Tin
Tin
Tin
Tin
Tin
Tin
Tin
Site
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Fresno
St.Louis
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Fresno
St.Louis
New York
Portland
Salt Lake
Chicago
Boston
Pair
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
Samplers
MetOne
Andersen
Andersen
Andersen
Andersen
URG
MetOne
MetOne
MetOne
MetOne
MetOne
Andersen
Andersen
MetOne
Andersen
Andersen
Andersen
Andersen
URG
MetOne
MetOne
MetOne
MetOne
MetOne
Andersen
Andersen
MetOne
Andersen
Andersen
Sampler6 Sampler?
Andersen
URG
Andersen URG
URG
URG
URG MetOne
URG
URG
URG
Andersen
Andersen
MetOne MetOne
MetOne
Andersen
URG
Andersen URG
URG
URG
URG MetOne
URG
URG
URG
Andersen
Andersen
MetOne MetOne
MetOne
Andersen
URG
Andersen URG
Site Maximum
Median MDL
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1204
1257
0580
0740
0536
1335
0469
0304
0498
0030
0119
0049
0057
0046
0070
0033
0050
0026
0034
0041
0027
0046
0134
0131
0174
0117
0122
0093
0076
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
Relative median difference between samplers
5 to 6 5 to 7 6 to 7 StoFRM 6toFRM 7 to FRM
0020
0020
0020
0020
0020
0020
0020
0020
0020
0055
0055
0055
0055
0055
0055
0055
0055
0055
0055
0055
0055
0055
0179
0179
0179
0179
0179
0179
0179
3
12
-3
38
31
-3
18
43
23
35
5
-34
-35
35
6
4
7
-0
-15
29
63
17
87
74
-61
-94
85
-4
-9
8%
8%
5%
8%
4%
5%
2%
3%
2%
2%
9%
1%
6%
9%
4%
4%
2%
3%
1%
5%
6%
4%
7%
0%
2%
2%
6%
6%
4%
24.6% 40.4% .
-33.1% -25.9% .
-36.8% -13.9%.
13.5% 5.2% .
-40.8% -48.1% .
-83.8% -23.6% .
-3.0% 11.8%.
B-4
April 27, 2001
-------
Sampler type for
Parameter
Tin
Tin
Tin
Tin
Tin
Tin
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Site
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Fresno
St.Louis
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Fresno
St.Louis
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Pair
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Samplers
Andersen
Andersen
URG
MetOne
MetOne
MetOne
MetOne
MetOne
Andersen
Andersen
MetOne
Andersen
Andersen
Andersen
Andersen
URG
MetOne
MetOne
MetOne
MetOne
MetOne
Andersen
Andersen
MetOne
Andersen
Andersen
Andersen
Andersen
URG
Sampler6 Sampler?
URG
URG
URG MetOne
URG
URG
URG
Andersen
Andersen
MetOne MetOne
MetOne
Andersen
URG
Andersen URG
URG
URG
URG MetOne
URG
URG
URG
Andersen
Andersen
MetOne MetOne
MetOne
Andersen
URG
Andersen URG
URG
URG
URG MetOne
Site Maximum
Median MDL
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0087
0079
0087
0115
0124
0129
1723
1230
0645
0562
1888
0990
0547
0727
1407
2859
1191
0699
0451
0112
0244
0197
0062
0061
0366
0089
0124
0054
0047
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0179
0179
0179
0179
0179
0179
0075
0075
0075
0075
0075
0075
0075
0075
0075
0075
0075
0075
0075
0014
0014
0014
0014
0014
0014
0014
0014
0014
0014
Relative median difference between samplers
5 to 6
-1
7
-14
88
80
102
-1
-11
-6
-0
3
18
3
38
22
-6
18
35
30
-166
-36
8
36
-24
2
-1
8
16
-6
1%
7%
3%
3%
2%
7%
1%
2%
5%
4%
5%
0%
3%
2%
1%
5%
1%
5%
5%
8%
7%
2%
3%
3%
4%
2%
0%
6%
0%
5 to 7 6 to 7 StoFRM 6toFRM 7 to FRM
-111.9% -103.2% .
-13.2% -1.4%.
27.9% 31.0%.
-46.5% -18.9%.
7.7% 1.0%.
2.9% 5.2% .
18.8% 27.4% .
B-5
April 27, 2001
-------
Sampler type for
Parameter
Zinc
Zinc
Zinc
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Site
Tampa
Bismarck
Seattle
Fresno
St.Louis
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Fresno
St.Louis
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Pair
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Samplers
MetOne
MetOne
MetOne
MetOne
MetOne
Andersen
Andersen
MetOne
Andersen
Andersen
Andersen
Andersen
URG
MetOne
MetOne
MetOne
MetOne
MetOne
Andersen
Andersen
MetOne
Andersen
Andersen
Andersen
Andersen
URG
MetOne
MetOne
MetOne
Sampler6 Sampler?
URG
URG
URG
Andersen
Andersen
MetOne MetOne
MetOne
Andersen
URG
Andersen URG
URG
URG
URG MetOne
URG
URG
URG
Andersen
Andersen
MetOne MetOne
MetOne
Andersen
URG
Andersen URG
URG
URG
URG MetOne
URG
URG
URG
Site Maximum
Median MDL
0
0
0
0
1
1
0
0
1
0
1
0
0
1
0
0
4
4
3
4
4
3
4
3
2
3
2
1
2
0053
0021
0060
6452
6273
1238
3066
2622
2792
6538
4074
9261
3603
2229
4721
4806
3780
3132
6228
0846
0413
4811
0557
3165
1605
4351
8770
8676
8822
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
Relative median difference between samplers
5 to 6 5 to 7 6 to 7 StoFRM 6toFRM 7 to FRM
0014
0014
0014
0170
0170
0170
0170
0170
0170
0170
0170
0170
0170
0170
0170
0170
1460
1460
1460
1460
1460
1460
1460
1460
1460
1460
1460
1460
1460
-14
-23
-14
20
-2
-5
-1
6
-2
9
-15
-70
0
-55
-32
-25
1
4
2
-4
9
33
2
30
18
-8
29
49
41
3%
1%
1%
4%
3%
0%
3%
0%
8%
2%
7%
2%
2%
9%
1%
5%
6%
3%
4%
3%
4%
8%
7%
7%
7%
0%
3%
8%
1%
1.5% 1.5%.
-48.5% -42.6% .
-3.1% -1.7%.
-3.2% 0.3% .
23.2% 27.3% .
-43.7% -28.2% .
B-6
April 27, 2001
-------
Sampler type for
Parameter
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Elemental
Elemental
Elemental
Elemental
Elemental
Elemental
Elemental
Elemental
Elemental
Elemental
Elemental
Elemental
Elemental
Sulfate
Sulfate
Carbon
Carbon
Carbon
Carbon
Carbon
Carbon
Carbon
Carbon
Carbon
Carbon
Carbon
Carbon
Carbon
Site
Fresno
St.Louis
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Fresno
St.Louis
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Fresno
St.Louis
Pair
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
Samplers
MetOne
MetOne
Andersen
Andersen
MetOne
Andersen
Andersen
Andersen
Andersen
URG
MetOne
MetOne
MetOne
MetOne
MetOne
Andersen
Andersen
MetOne
Andersen
Andersen
Andersen
Andersen
URG
MetOne
MetOne
MetOne
MetOne
MetOne
Sampler6 Sampler?
Andersen
Andersen
MetOne MetOne
MetOne
Andersen
URG
Andersen URG
URG
URG
URG MetOne
URG
URG
URG
Andersen
Andersen
MetOne MetOne
MetOne
Andersen
URG
Andersen URG
URG
URG
URG MetOne
URG
URG
URG
Andersen
Andersen
Site Maximum
Median MDL
1.1004
1.0490
0.7440
0.6488
0.3669
1.2154
0.5088
1.3094
0.6023
0.3847
0.5128
0.2734
0.6071
0.4945
0.7649
1.1654
0.6413
0.6382
0.9903
0.9590
0.6573
0.3397
0.6048
0.4757
0.1959
0.6486
1.6899
3.5810
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
Relative median difference between samplers
5 to 6 5 to 7 6 to 7 StoFRM 6toFRM 7 to FRM
0080
0080
0080
0080
0080
0080
0080
0080
0080
0080
0080
0080
0080
1460
1460
1460
1460
1460
1460
1460
1460
1460
1460
1460
1460
1460
0120
0120
10
-0
-1
-6
11
15
3
12
23
4
7
33
16
-9
2
3
-5
5
22
19
21
12
-2
22
-15
34
12
-1
1%
2%
9%
0%
0%
5%
7%
2%
4%
6%
9%
5%
2%
1%
0%
2%
3%
4%
9%
5%
9%
6%
5%
1%
3%
8%
5%
1%
-5.3% -1.3%.
8.9% 6.2% .
-20.2% -20.6% .
8.1% -1.5%.
-9.0% -2.7% .
-10.8% -15.3%.
B-7
April 27, 2001
-------
Sampler type for
Parameter
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Site
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Pair
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Samplers
Andersen
Andersen
MetOne
Andersen
Andersen
Andersen
Andersen
URG
MetOne
MetOne
MetOne
Sampler6 Sampler?
MetOne MetOne
MetOne
Andersen
URG
Andersen URG
URG
URG
URG MetOne
URG
URG
URG
Site Maximum
Median MDL
2
1
0
3
2
3
3
1
3
1
1
9367
1490
8744
0957
1997
1937
6743
0603
9425
2935
4512
0
0
0
0
0
0
0
0
0
0
0
Relative median difference between samplers
5 to 6 5 to 7 6 to 7 StoFRM 6toFRM 7 to FRM
0120
0120
0120
0120
0120
0120
0120
0120
0120
0120
0120
-1
-3
6
-2
0
-2
-9
-1
-4
0
3
7%
9%
5%
3%
5%
0%
0%
5%
8%
4%
1%
3.9% 0.5% .
-15.7% -10.4%.
-8.5% -8.0% .
B-8
April 27, 2001
-------
Table B.2 Mean Differences Between Sampler Types for Each Site and Parameter
Parameter
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Sampler
Pair Type
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Site
Fresno
St. Louis
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Fresno
St. Louis
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Mean Difference
Between
Sampler Types
(migrograms/m3)
-1.3052
-0.5698
-1.4748
-1.2551
-0.7965
0.4394
0.3000
1.3219
1.9952
1.8153
1.1467
1.2272
1.8529
0.0055
-0.0001
-0.0059
-0.0005
-0.0133
0.0043
0.0104
0.0140
0.0833
0.0471
Standard Error
0.3441
0.3090
0.2696
0.3023
0.2733
0.3027
0.3198
0.2837
0.3030
0.2621
0.3250
0.4092
0.2434
0.0101
0.0150
0.0137
0.0115
0.0079
0.0121
0.0193
0.0134
0.0108
0.0075
Relative
Difference
-12.7%
-2.8%
-10.6%
-15.1%
-9.9%
3.0%
2.7%
9.5%
16.9%
23.6%
9.3%
25.6%
27.0%
19.3%
-12.8%
-34.2%
-19.6%
-12.7%
22.5%
116.9%
54.9%
78.1%
44.8%
Standard Error
2.7%
2.7%
2.2%
2.3%
2.2%
2.8%
3.0%
2.8%
3.2%
2.9%
3.2%
4.6%
2.8%
13.7%
14.7%
10.2%
10.5%
8.0%
17.0%
47.7%
23.6%
22.1%
12.4%
B-9
April 27, 2001
-------
Parameter
Aluminum
Aluminum
Aluminum
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Sampler
Pair Type
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
Site
Tampa
Bismarck
Seattle
Fresno
St. Louis
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Fresno
St. Louis
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Mean Difference
Between
Sampler Types
(migrograms/m3)
0.0166
0.0268
0.0089
-0.0006
0.0162
-0.0068
-0.0002
-0.0185
0.0254
0.0114
0.0210
0.0402
0.0551
0.0266
0.0231
0.0105
-0.0030
0.0108
0.0102
0.0152
0.0097
-0.0054
0.0079
0.0006
Standard Error
0.0141
0.0231
0.0126
0.0075
0.0067
0.0059
0.0066
0.0060
0.0066
0.0069
0.0062
0.0066
0.0057
0.0070
0.0089
0.0053
0.0403
0.0273
0.0266
0.0172
0.0150
0.0188
0.0190
0.0297
Relative
Difference
164.5%
122.8%
62.2%
0.4%
9.8%
-14.4%
-4.4%
-4.5%
22.1%
31.9%
57.2%
55.9%
45.4%
49.2%
76.7%
34.0%
-34.5%
41.7%
-2.2%
59.1%
28.2%
-4.2%
64.4%
26.8%
Standard Error
42.2%
58.5%
23.1%
5.7%
5.6%
3.8%
4.7%
4.3%
6.1%
6.9%
7.3%
7.8%
6.3%
7.9%
11.9%
5.4%
27.0%
39.2%
26.5%
27.7%
19.5%
18.2%
31.5%
38.3%
B-10
April 27, 2001
-------
Parameter
Chlorine
Chlorine
Chlorine
Chlorine
Iron
Iron
Iron
Iron
Iron
Iron
Iron
Iron
Iron
Iron
Iron
Iron
Iron
Lead
Lead
Lead
Lead
Lead
Lead
Lead
Sampler
Pair Type
And-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
Site
Houston
Phoenix
Tampa
Seattle
Fresno
St. Louis
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Fresno
St. Louis
New York
Portland
Salt Lake
Chicago
Boston
Mean Difference
Between
Sampler Types
(migrograms/m3)
0.0906
0.0052
0.0666
0.0038
0.0001
0.0074
-0.0065
0.0032
-0.0127
0.0194
0.0212
0.0312
0.0415
0.0464
0.0093
0.0172
0.0142
-0.0040
-0.0016
-0.0031
-0.0041
-0.0019
0.0006
0.0004
Standard Error
0.0149
0.0125
0.0293
0.0148
0.0064
0.0058
0.0050
0.0057
0.0051
0.0057
0.0060
0.0053
0.0057
0.0049
0.0061
0.0076
0.0046
0.0011
0.0005
0.0005
0.0006
0.0006
0.0005
0.0006
Relative
Difference
73.0%
16.8%
465.6%
35.5%
2.2%
5.7%
-7.5%
5.3%
-6.8%
13.4%
39.9%
44.0%
41.4%
36.6%
21.5%
64.9%
29.8%
-57.8%
-12.2%
-40.4%
-39.7%
-34.9%
7.2%
9.6%
Standard Error
25.9%
14.8%
169.0%
20.3%
4.3%
4.0%
3.1%
3.9%
3.1%
4.2%
5.5%
5.0%
5.3%
4.4%
4.9%
8.3%
3.9%
6.2%
6.0%
4.4%
5.1%
5.4%
7.1%
8.7%
B-11
April 27, 2001
-------
Parameter
Lead
Lead
Lead
Lead
Lead
Lead
Tin
Tin
Tin
Tin
Tin
Tin
Tin
Tin
Tin
Tin
Tin
Tin
Tin
Silicon
Silicon
Silicon
Silicon
Silicon
Sampler
Pair Type
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
Site
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Fresno
St. Louis
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Fresno
St. Louis
New York
Portland
Salt Lake
Mean Difference
Between
Sampler Types
(migrograms/m3)
0.0003
-0.0001
0.0033
0.0025
0.0038
0.0032
-0.0130
-0.0136
-0.0149
-0.0130
-0.0151
0.0002
-0.0005
0.0009
0.0003
0.0141
0.0125
0.0143
0.0151
-0.0015
0.0153
-0.0089
0.0049
-0.0201
Standard Error
0.0005
0.0006
0.0009
0.0007
0.0020
0.0006
0.0012
0.0010
0.0010
0.0010
0.0010
0.0010
0.0011
0.0009
0.0012
0.0011
0.0011
0.0015
0.0009
0.0159
0.0143
0.0125
0.0140
0.0127
Relative
Difference
4.7%
-6.8%
74.1%
68.4%
136.8%
76.8%
-55.7%
-54.4%
-61.2%
-56.8%
-61.1%
1.0%
-4.8%
7.4%
3.0%
153.4%
129.0%
163.3%
154.6%
1.3%
6.7%
-11.8%
5.2%
-3.6%
Standard Error
7.0%
8.3%
21.6%
15.8%
58.9%
14.9%
3.2%
2.9%
2.4%
2.8%
2.4%
6.2%
6.6%
5.9%
7.7%
17.2%
16.1%
24.1%
14.2%
5.1%
4.8%
3.5%
4.7%
3.9%
B-12
April 27, 2001
-------
Parameter
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Ammonium
Ammonium
Ammonium
Sampler
Pair Type
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
Site
Chicago
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Fresno
St. Louis
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Fresno
St. Louis
New York
Mean Difference
Between
Sampler Types
(migrograms/m3)
0.0228
0.0215
0.0373
0.1137
0.1249
0.0284
0.0510
0.0149
0.0202
0.0107
0.0013
0.0016
0.0016
0.0016
0.0007
0.0016
0.0008
-0.0008
-0.0008
-0.0012
-0.0008
-0.1561
-0.1150
-0.0802
Standard Error
0.0140
0.0148
0.0131
0.0140
0.0121
0.0150
0.0189
0.0113
0.0010
0.0008
0.0007
0.0008
0.0007
0.0008
0.0008
0.0007
0.0008
0.0007
0.0009
0.0020
0.0007
0.0624
0.0544
0.0474
Relative
Difference
18.8%
33.8%
39.3%
35.2%
37.1%
27.8%
67.2%
31.3%
512.6%
60.9%
6.7%
27.5%
29.5%
-0.7%
11.0%
12.2%
23.2%
-13.5%
-15.1%
-24.0%
-15.4%
-18.1%
-3.8%
-3.6%
Standard Error
5.3%
6.3%
5.8%
6.0%
5.3%
6.1%
10.0%
4.7%
52.0%
10.6%
6.0%
8.6%
7.4%
6.3%
7.4%
6.6%
8.2%
5.2%
6.1%
12.6%
4.6%
7.0%
7.1%
6.2%
B-13
April 27, 2001
-------
Parameter
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Nitrate
Sampler
Pair Type
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
Site
Portland
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Fresno
St. Louis
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Fresno
Mean Difference
Between
Sampler Types
(migrograms/m3)
0.0229
-0.0240
-0.1300
-0.3268
-0.2749
-0.5601
0.0044
-0.7566
-0.1448
-0.1648
-0.1283
-0.2487
0.0078
-0.2105
-0.4120
1.1204
0.9913
0.8070
0.5445
0.9815
0.7263
0.7600
1.2147
-0.1386
Standard Error
0.0588
0.0521
0.0532
0.0572
0.0499
0.0559
0.0461
0.0571
0.0720
0.0490
0.1248
0.1245
0.2075
0.1123
0.1126
0.1099
0.1716
0.1029
0.1332
0.0961
0.1179
0.1481
0.0884
0.0526
Relative
Difference
15.1%
-3.3%
-4.8%
-40.7%
-13.8%
-50.7%
1.1%
-53.2%
-23.7%
-43.2%
-2.6%
-5.4%
-1.5%
-4.7%
-9.7%
36.0%
28.8%
27.5%
35.3%
28.2%
29.3%
53.4%
52.2%
-7.2%
Standard Error
9.3%
6.9%
7.0%
4.6%
5.9%
3.8%
6.4%
3.6%
7.5%
3.8%
3.4%
3.3%
5.6%
3.0%
2.8%
4.1%
6.1%
3.6%
5.0%
3.4%
4.2%
6.3%
3.7%
5.0%
B-14
April 27, 2001
-------
Parameter
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Sampler
Pair Type
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Site
St. Louis
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Fresno
St. Louis
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Mean Difference
Between
Sampler Types
(migrograms/m3)
0.0001
-0.0544
-0.0363
-0.0802
0.1894
0.1442
0.1810
0.1657
0.0824
0.0770
0.0998
0.1508
0.0029
0.0474
0.0636
-0.0599
-0.0196
0.2406
0.0109
0.1637
0.1042
0.0965
0.0984
0.0766
0.2762
Standard Error
0.0459
0.0400
0.0459
0.0430
0.0449
0.0483
0.0421
0.0471
0.0389
0.0482
0.0607
0.0367
0.0463
0.0460
0.0760
0.0425
0.0418
0.0408
0.0635
0.0381
0.0511
0.0362
0.0436
0.0628
0.0332
Relative
Difference
1.5%
-6.6%
-5.2%
-11.0%
12.4%
21.3%
18.2%
40.1%
26.2%
19.4%
30.7%
27.8%
-3.3%
7.0%
7.6%
-4.6%
-4.4%
23.8%
-1.3%
22.0%
29.7%
16.7%
23.1%
36.5%
47.4%
Standard Error
4.7%
3.8%
4.4%
3.9%
5.1%
5.9%
5.1%
6.7%
5.0%
5.8%
8.1%
4.8%
7.4%
8.1%
13.3%
6.7%
6.6%
8.4%
10.4%
7.7%
11.0%
7.0%
8.8%
14.1%
8.1%
B-15
April 27, 2001
-------
Parameter
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sampler
Pair Type
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Site
Fresno
St. Louis
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Mean Difference
Between
Sampler Types
(migrograms/m3)
-0.1848
-0.0316
-0.0209
-0.0637
-0.0440
-0.2514
-0.3957
-0.1281
-0.2489
0.1026
-0.3445
-0.0275
0.0051
Standard Error
0.0779
0.0679
0.0592
0.0680
0.0636
0.0664
0.0715
0.0623
0.0697
0.0575
0.0714
0.0899
0.0542
Relative
Difference
-10.7%
0.2%
-0.3%
-4.5%
-5.2%
-5.3%
-14.0%
-4.2%
-6.9%
9.2%
-9.0%
-1.7%
2.5%
Standard Error
2.7%
2.6%
2.3%
2.5%
2.3%
2.4%
2.4%
2.3%
2.5%
2.4%
2.5%
3.4%
2.1%
B-16
April 27, 2001
-------
Table B.3
Sampler Type Means for all Sites and Parameters
Parameter
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
PM2.5 Mass
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Samplertype
pair
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
Site
Fresno
Fresno
St. Louis
St. Louis
New York
New York
Portland
Portland
Salt Lake
Salt Lake
Chicago
Chicago
Boston
Boston
Philadelphia
Philadelphia
Houston
Houston
Phoenix
Phoenix
Tampa
Tampa
Bismarck
Bismarck
Seattle
Seattle
Fresno
Fresno
St. Louis
St. Louis
New York
New York
Portland
Portland
Salt Lake
Salt Lake
Chicago
Sampler
Type
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
URG
Andersen
URG
Andersen
URG
Andersen
URG
MetOne
URG
MetOne
URG
MetOne
URG
MetOne
URG
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
Mean
(migrograms/m3)
9.027
10.332
15.451
16.020
14.093
15.568
8.468
9.724
8.004
8.800
17.518
17.079
11.993
11.693
14.991
13.669
14.333
12.338
9.659
7.844
15.024
13.878
6.777
5.550
9.482
7.629
0.042
0.037
0.037
0.037
0.015
0.021
0.019
0.020
0.076
0.089
0.026
Standard
Error
1.271
1.267
1.049
1.046
0.992
0.975
1.100
1.100
1.078
1.078
1.162
1.161
1.208
1.213
1.032
1.026
1.236
1.236
1.018
1.004
1.123
1.120
1.456
1.456
0.930
0.931
0.022
0.022
0.027
0.027
0.025
0.024
0.021
0.022
0.019
0.019
0.026
B-17
April 27, 2001
-------
Parameter
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Samplertype
pair
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Site
Chicago
Boston
Boston
Philadelphia
Philadelphia
Houston
Houston
Phoenix
Phoenix
Tampa
Tampa
Bismarck
Bismarck
Seattle
Seattle
Fresno
Fresno
St. Louis
St. Louis
New York
New York
Portland
Portland
Salt Lake
Salt Lake
Chicago
Chicago
Boston
Boston
Philadelphia
Philadelphia
Houston
Houston
Phoenix
Phoenix
Tampa
Tampa
Bismarck
Bismarck
Sampler
Type
URG
Andersen
URG
Andersen
URG
Andersen
URG
MetOne
URG
MetOne
URG
MetOne
URG
MetOne
URG
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
URG
Andersen
URG
Andersen
URG
Andersen
URG
MetOne
URG
MetOne
URG
MetOne
URG
Mean
(migrograms/m3)
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
022
017
006
025
011
256
173
173
126
035
018
101
075
035
026
054
054
128
112
044
050
036
036
163
181
123
098
044
033
057
036
109
069
188
133
078
052
063
040
Standard
Error
0.026
0.034
0.037
0.027
0.028
0.026
0.026
0.018
0.017
0.027
0.026
0.046
0.046
0.024
0.023
0.012
0.012
0.010
0.010
0.010
0.009
0.010
0.010
0.010
0.010
0.011
0.011
0.011
0.011
0.010
0.009
0.011
0.011
0.010
0.009
0.011
0.010
0.014
0.014
B-18
April 27, 2001
-------
Parameter
Calcium
Calcium
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Iron
Iron
Iron
Iron
Iron
Iron
Iron
Iron
Iron
Iron
Iron
Iron
Samplertype
pair
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
Site
Seattle
Seattle
Fresno
Fresno
St. Louis
St. Louis
New York
New York
Portland
Portland
Salt Lake
Salt Lake
Chicago
Chicago
Boston
Boston
Philadelphia
Philadelphia
Houston
Houston
Phoenix
Phoenix
Tampa
Tampa
Seattle
Seattle
Fresno
Fresno
St. Louis
St. Louis
New York
New York
Portland
Portland
Salt Lake
Salt Lake
Chicago
Chicago
Sampler
Type
MetOne
URG
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
URG
Andersen
URG
Andersen
URG
Andersen
URG
MetOne
URG
MetOne
URG
MetOne
URG
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
URG
Mean
(migrograms/m3)
0.044
0.033
0.018
0.021
0.024
0.013
0.023
0.012
0.044
0.028
0.019
0.009
0.042
0.047
0.020
0.012
0.008
0.008
0.200
0.109
0.065
0.060
0.085
0.019
0.061
0.057
0.088
0.088
0.170
0.162
0.103
0.109
0.061
0.058
0.140
0.153
0.180
0.161
Standard
Error
0.009
0.009
0.042
0.045
0.035
0.038
0.034
0.029
0.025
0.026
0.020
0.022
0.028
0.029
0.027
0.028
0.037
0.034
0.024
0.024
0.020
0.018
0.035
0.034
0.020
0.020
0.016
0.016
0.014
0.014
0.013
0.012
0.014
0.014
0.014
0.014
0.015
0.015
B-19
April 27, 2001
-------
Parameter
Iron
Iron
Iron
Iron
Iron
Iron
Iron
Iron
Iron
Iron
Iron
Iron
Iron
Iron
Lead
Lead
Lead
Lead
Lead
Lead
Lead
Lead
Lead
Lead
Lead
Lead
Lead
Lead
Lead
Lead
Lead
Lead
Lead
Lead
Lead
Lead
Lead
Lead
Lead
Samplertype
pair
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Site
Boston
Boston
Philadelphia
Philadelphia
Houston
Houston
Phoenix
Phoenix
Tampa
Tampa
Bismarck
Bismarck
Seattle
Seattle
Fresno
Fresno
St. Louis
St. Louis
New York
New York
Portland
Portland
Salt Lake
Salt Lake
Chicago
Chicago
Boston
Boston
Philadelphia
Philadelphia
Houston
Houston
Phoenix
Phoenix
Tampa
Tampa
Bismarck
Bismarck
Seattle
Sampler
Type
Andersen
URG
Andersen
URG
Andersen
URG
MetOne
URG
MetOne
URG
MetOne
URG
MetOne
URG
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
URG
Andersen
URG
Andersen
URG
Andersen
URG
MetOne
URG
MetOne
URG
MetOne
URG
MetOne
Mean
(migrograms/m3)
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
073
052
101
070
133
091
184
138
055
045
060
043
073
059
003
007
015
016
005
008
005
009
005
007
010
009
004
004
006
005
004
004
007
004
007
005
008
004
010
Standard
Error
0.016
0.016
0.013
0.013
0.016
0.016
0.013
0.013
0.015
0.014
0.019
0.019
0.012
0.012
0.002
0.002
0.001
0.001
0.001
0.001
0.001
0.001
0.001
0.001
0.001
0.001
0.001
0.001
0.001
0.001
0.002
0.002
0.001
0.001
0.001
0.001
0.003
0.003
0.001
B-20
April 27, 2001
-------
Parameter
Lead
Tin
Tin
Tin
Tin
Tin
Tin
Tin
Tin
Tin
Tin
Tin
Tin
Tin
Tin
Tin
Tin
Tin
Tin
Tin
Tin
Tin
Tin
Tin
Tin
Tin
Tin
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Samplertype
pair
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
Site
Seattle
Fresno
Fresno
St. Louis
St. Louis
New York
New York
Portland
Portland
Salt Lake
Salt Lake
Chicago
Chicago
Boston
Boston
Philadelphia
Philadelphia
Houston
Houston
Phoenix
Phoenix
Tampa
Tampa
Bismarck
Bismarck
Seattle
Seattle
Fresno
Fresno
St. Louis
St. Louis
New York
New York
Portland
Portland
Salt Lake
Salt Lake
Chicago
Sampler
Type
URG
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
URG
Andersen
URG
Andersen
URG
Andersen
URG
MetOne
URG
MetOne
URG
MetOne
URG
MetOne
URG
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
Mean
(migrograms/m3)
0.007
0.010
0.024
0.012
0.025
0.009
0.024
0.010
0.023
0.010
0.025
0.011
0.011
0.009
0.010
0.010
0.010
0.010
0.009
0.023
0.009
0.022
0.010
0.023
0.009
0.025
0.010
0.181
0.183
0.151
0.136
0.093
0.102
0.076
0.071
0.237
0.257
0.134
Standard
Error
0.001
0.001
0.001
0.001
0.001
0.001
0.000
0.001
0.001
0.001
0.001
0.001
0.001
0.001
0.001
0.001
0.001
0.001
0.001
0.001
0.001
0.001
0.001
0.001
0.001
0.001
0.001
0.038
0.038
0.031
0.031
0.030
0.029
0.033
0.033
0.032
0.032
0.035
B-21
April 27, 2001
-------
Parameter
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Samplertype
pair
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Site
Chicago
Boston
Boston
Philadelphia
Philadelphia
Houston
Houston
Phoenix
Phoenix
Tampa
Tampa
Bismarck
Bismarck
Seattle
Seattle
Fresno
Fresno
St. Louis
St. Louis
New York
New York
Portland
Portland
Salt Lake
Salt Lake
Chicago
Chicago
Boston
Boston
Philadelphia
Philadelphia
Houston
Houston
Phoenix
Phoenix
Tampa
Tampa
Bismarck
Bismarck
Sampler
Type
URG
Andersen
URG
Andersen
URG
Andersen
URG
MetOne
URG
MetOne
URG
MetOne
URG
MetOne
URG
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
URG
Andersen
URG
Andersen
URG
Andersen
URG
MetOne
URG
MetOne
URG
MetOne
URG
Mean
(migrograms/m3)
0
0
0
0
0
0
0
111
077
056
121
084
402
289
0.481
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
356
136
107
167
116
066
051
022
002
034
024
024
023
009
007
008
007
051
049
010
009
016
015
006
006
005
006
006
007
002
004
Standard
Error
0.035
0.036
0.036
0.031
0.031
0.037
0.037
0.031
0.030
0.034
0.033
0.044
0.044
0.028
0.028
0.003
0.003
0.002
0.002
0.002
0.002
0.002
0.002
0.002
0.002
0.003
0.003
0.003
0.003
0.002
0.002
0.003
0.003
0.002
0.002
0.003
0.002
0.005
0.005
B-22
April 27, 2001
-------
Parameter
Zinc
Zinc
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Samplertype
pair
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
Site
Seattle
Seattle
Fresno
Fresno
St. Louis
St. Louis
New York
New York
Portland
Portland
Salt Lake
Salt Lake
Chicago
Chicago
Boston
Boston
Philadelphia
Philadelphia
Houston
Houston
Phoenix
Phoenix
Tampa
Tampa
Bismarck
Bismarck
Seattle
Seattle
Fresno
Fresno
St. Louis
St. Louis
New York
New York
Portland
Portland
Salt Lake
Salt Lake
Sampler
Type
MetOne
URG
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
URG
Andersen
URG
Andersen
URG
Andersen
URG
MetOne
URG
MetOne
URG
MetOne
URG
MetOne
URG
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Mean
(migrograms/m3)
0.007
0.008
0.583
0.739
1.664
1.779
1.491
1.572
0.414
0.391
0.352
0.376
2.218
2.348
1.055
1.382
1.685
1.960
0.806
1.366
0.396
0.391
1.210
1.967
0.463
0.608
0.431
0.596
4.569
4.698
4.209
4.458
4.394
4.386
4.293
4.503
4.145
4.557
Standard
Error
0.002
0.002
0.216
0.215
0.172
0.172
0.163
0.160
0.200
0.200
0.187
0.187
0.191
0.191
0.195
0.196
0.167
0.166
0.209
0.209
0.165
0.162
0.182
0.181
0.235
0.235
0.171
0.172
0.327
0.323
0.310
0.311
0.353
0.305
0.290
0.292
0.316
0.316
B-23
April 27, 2001
-------
Parameter
Organic
Organic
Organic
Organic
Organic
Organic
Organic
Organic
Organic
Organic
Organic
Organic
Organic
Organic
Organic
Organic
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Carbon
Carbon
Carbon
Carbon
Carbon
Carbon
Carbon
Carbon
Carbon
Carbon
Carbon
Carbon
Carbon
Carbon
Carbon
Carbon
Samplertype
pair
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Site
Chicago
Chicago
Boston
Boston
Philadelphia
Philadelphia
Houston
Houston
Phoenix
Phoenix
Tampa
Tampa
Bismarck
Bismarck
Seattle
Seattle
Fresno
Fresno
St. Louis
St. Louis
New York
New York
Portland
Portland
Salt Lake
Salt Lake
Chicago
Chicago
Boston
Boston
Philadelphia
Philadelphia
Houston
Houston
Phoenix
Phoenix
Tampa
Tampa
Bismarck
Sampler
Type
Andersen
URG
Andersen
URG
Andersen
URG
Andersen
URG
MetOne
URG
MetOne
URG
MetOne
URG
MetOne
URG
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
URG
Andersen
URG
Andersen
URG
Andersen
URG
MetOne
URG
MetOne
URG
MetOne
Mean
(migrograms/m3)
4
3
4
3
3
3
2
1
4
3
3
2
2
1
3
2
1
1
1
1
1
1
0
0
0
0
2
2
1
0
1
1
0
0
0
0
0
0
0
507
387
632
640
983
176
494
949
306
325
526
799
360
600
755
540
254
392
560
560
081
135
775
811
758
838
563
374
102
958
499
318
868
702
542
459
613
536
589
Standard
Error
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
302
302
482
482
266
264
373
373
260
253
297
298
365
365
238
238
236
236
189
188
178
176
198
198
202
202
210
210
214
215
183
182
230
230
180
178
199
198
258
B-24
April 27, 2001
-------
Parameter
Nitrate
Nitrate
Nitrate
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Samplertype
pair
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
Site
Bismarck
Seattle
Seattle
Fresno
Fresno
St. Louis
St. Louis
New York
New York
Portland
Portland
Salt Lake
Salt Lake
Chicago
Chicago
Boston
Boston
Philadelphia
Philadelphia
Houston
Houston
Phoenix
Phoenix
Tampa
Tampa
Bismarck
Bismarck
Seattle
Seattle
Fresno
Fresno
St. Louis
St. Louis
New York
New York
Portland
Portland
Salt Lake
Sampler
Type
URG
MetOne
URG
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
URG
Andersen
URG
Andersen
URG
Andersen
URG
MetOne
URG
MetOne
URG
MetOne
URG
MetOne
URG
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
MetOne
Andersen
Mean
(migrograms/m3)
0.489
0.847
0.697
0.560
0.557
0.878
0.831
1.398
1.334
0.656
0.716
0.688
0.708
1.301
1.061
0.910
0.899
0.823
0.659
0.412
0.307
0.704
0.608
0.553
0.454
0.279
0.202
0.890
0.614
1.552
1.737
4.334
4.366
3.941
3.962
1.308
1.372
0.890
Standard
Error
0.258
0.167
0.167
0.091
0.090
0.084
0.085
0.104
0.082
0.078
0.080
0.085
0.085
0.082
0.081
0.130
0.130
0.073
0.072
0.102
0.101
0.072
0.068
0.081
0.081
0.107
0.102
0.065
0.065
0.491
0.491
0.392
0.391
0.368
0.366
0.412
0.412
0.420
B-25
April 27, 2001
-------
Parameter
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Samplertype
pair
And-Met1
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Site
Salt Lake
Chicago
Chicago
Boston
Boston
Philadelphia
Philadelphia
Houston
Houston
Phoenix
Phoenix
Tampa
Tampa
Bismarck
Bismarck
Seattle
Seattle
Sampler
Type
MetOne
Andersen
URG
Andersen
URG
Andersen
URG
Andersen
URG
MetOne
URG
MetOne
URG
MetOne
URG
MetOne
URG
Mean
(migrograms/m3)
0.934
4.587
4.839
3.062
3.458
4.469
4.597
3.803
4.052
1.280
1.177
4.732
5.076
1.388
1.415
1.470
1.465
Standard
Error
0.420
0.437
0.437
0.446
0.446
0.379
0.378
0.478
0.478
0.374
0.372
0.413
0.412
0.535
0.535
0.347
0.347
B-26
April 27, 2001
-------
Table B.4
Significance of the Difference in Relative Composition
of the Mass Constituents by Site
Parameter
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Aluminum
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Calcium
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Chlorine
Sampler pair
type
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
Site
Fresno
St. Louis
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Fresno
St. Louis
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Fresno
St. Louis
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Houston
Significance of the differences in
relative composition
0.0051
0.5504
0.013
0.4937
0.7091
0.1275
0.0008
0.0123
0.0017
0.0584
<0001
0.0136
0.0617
0.0102
0.0127
0.32
0.0118
0.1777
0.0004
<0001
<0001
<0001
0.0001
<0001
<0001
0.1866
0.4513
0.1245
0.8266
0.0003
0.0146
0.6376
0.0086
0.8171
0.0085
B-27
April 27, 2001
-------
Parameter
Chlorine
Chlorine
Chlorine
Iron
Iron
Iron
Iron
Iron
Iron
Iron
Iron
Iron
Iron
Iron
Iron
Iron
Lead
Lead
Lead
Lead
Lead
Lead
Lead
Lead
Lead
Lead
Lead
Lead
Lead
Tin
Tin
Tin
Tin
Tin
Tin
Tin
Tin
Sampler pair
type
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
Site
Phoenix
Tampa
Seattle
Fresno
St. Louis
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Fresno
St. Louis
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Fresno
St. Louis
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Significance of the differences in
relative composition
0.4424
<0001
0.669
0.0005
0.0337
0.3221
<0001
0.3384
0.0144
<0001
<0001
<0001
0.0035
0.0158
<0001
0.4817
<0001
0.2303
<0001
<0001
0.0014
0.4892
0.6344
0.3106
0.045
0.0108
<0001
0.008
0.0011
<0001
<0001
<0001
<0001
<0001
0.4986
0.1768
0.4493
B-28
April 27, 2001
-------
Parameter
Tin
Tin
Tin
Tin
Tin
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Silicon
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Zinc
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Sampler pair
type
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
Site
Houston
Phoenix
Tampa
Bismarck
Seattle
Fresno
St. Louis
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Fresno
St. Louis
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Fresno
St. Louis
New York
Portland
Salt Lake
Chicago
Significance of the differences in
relative composition
0.0455
<0001
<0001
<0001
<0001
0.0029
0.0383
0.7213
<0001
0.0862
0.0012
<0001
<0001
0.001
0.0061
0.0009
<0001
0.3542
<0001
<0001
0.0049
<0001
<0001
0.6211
0.3427
0.6825
0.3348
<0001
0.0018
0.0041
<0001
0.5197
0.9005
0.3157
0.0009
0.5352
0.3109
B-29
April 27, 2001
-------
Parameter
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Ammonium
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Organic Carbon
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Nitrate
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Sampler pair
type
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
Site
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Fresno
St. Louis
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Fresno
St. Louis
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Fresno
St. Louis
New York
Portland
Significance of the differences in
relative composition
<0001
0.0008
<0001
0.0022
<0001
<0001
<0001
0.0113
0.4731
0.7536
0.004
0.5642
<0001
<0001
<0001
0.0136
0.2732
<0001
0.0004
<0001
0.3286
0.3575
0.3184
0.0282
0.3755
0.0762
0.0019
0.0861
0.0002
0.6289
0.0789
0.5923
0.7199
0.1999
0.2362
0.5705
0.0896
B-30
April 27, 2001
-------
Parameter
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Elemental Carbon
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sulfate
Sampler pair
type
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
And-Met1
And-Met1
And-Met1
And-Met1
And-Met1
And-URG
And-URG
And-URG
And-URG
Met1-URG
Met1-URG
Met1-URG
Met1-URG
Site
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Fresno
St. Louis
New York
Portland
Salt Lake
Chicago
Boston
Philadelphia
Houston
Phoenix
Tampa
Bismarck
Seattle
Significance of the differences in
relative composition
0.291
0.0108
0.9003
0.1773
0.399
0.4172
0.1473
0.8575
0.0112
0.6654
0.4026
0.0006
0.0008
0.4998
0.0164
<0001
<0001
<0001
<0001
<0001
<0001
<0001
B-31
April 27, 2001
-------
TECHNICAL REPORT DATA
(Please read Instructions on reverse before completing)
1. REPORT NO.
EPA-EPA454/R-01-008
3. RECIPIENT'S ACCESSION NO.
4. TITLE AND SUBTITLE
Evaluation of PM2.5 Speciation Sampler Performance and Related
Sample Collection and Stability Issues
5. REPORT DATE
April 21, 2001
6. PERFORMING ORGANIZATION CODE
7. AUTHOR(S)
Basil Coutant and Shannon Stetzer
8. PERFORMING ORGANIZATION REPORT NO.
N/A
9. PERFORMING ORGANIZATION NAME AND ADDRESS
U.S. Environmental Protection Agency
Office of Air Quality Planning and Standards
Research Triangle Park, NC 27711
10. PROGRAM ELEMENT NO.
11. CONTRACT/GRANT NO.
68-D-98-030, Work Assignment 4-06
12. SPONSORING AGENCY NAME AND ADDRESS
Director
Office of Air Quality Planning and Standards
Office of Air and Radiation
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
13. TYPE OF REPORT AND PERIOD COVERED
Final, January-April, 2001
14. SPONSORING AGENCY CODE
EPA/200/04
15. SUPPLEMENTARY NOTES
16. ABSTRACT
EPA has established a PM2.5 chemical speciation trends network of 54 monitoring sites for routine operation. This
network is used to provide a nationally consistent set of data for the assessment of trends and provide long term
characterization of PM constituents. A vital consideration for these sites is data comparability. EPA initially chose
three samplers for potential use in this network. The samplers were operated between February and July, 2000, at
thirteen sites as part of an comparative assessment study. Further data were collected in August, 2000, to examine
some special issues dealing with sample collection media stability. The analysis results detailed in this report are the
end result of three important efforts. First, data underwent a careful screening for outliers, or unusual data, so that
results would not be skewed by these values. Next, considerable effort was put into graphical analysis of the data to
determine what factors should be considered in the assessment of data comparability. The third effort detailed in
theis report was statistical modeling based on the outcomes of the first two efforts. The measured PM2.5 mass was
compared with co-located Federal Reference Method sampler measurements. Seventeen out of 24 of the samplers
met study data obj ectives of an R2 value of at least 0.9 in a linear regression of the mass values against the FRM
measurement. Deviations from this criteria appear to be caused by site influences that affect all the monitors at a site,
rather than differences among sampler types. There are significant site to site differences in: the number of days with
outliers, the variability of parameters, the relationship between samplers, etc. The variability found in the sampling
precision across sampler types is probably due to site influences, but is probably not generally of any practical
concern.
-------
17 KEY WORDS AND DOCUMENT ANALYSIS
a. DESCRIPTORS
PM2.5, air monitoring, chemical speciation, air
monitoring network, atmospheric particle sampling.
18. DISTRIBUTION STATEMENT
Release Unlimited
b. IDENTIFIERS/OPEN ENDED TERMS
Air Pollution control
19. SECURITY CLASS (Report)
Unclassified
20. SECURITY CLASS (Page)
Unclassified
c. COSATI Field/Group
21. NO. OF PAGES
94
22. PRICE
EPA Form 2220-1 (Rev. 4-77) PREVIOUS EDITION IS OBSOLETE
------- |