EPA # 600/R-02/070
                                                                September 2002
Guidelines for the Application of SEM/EDX Analytical
        Techniques to Particulate Matter Samples
                                     by
                      Robert D. Willis and Fredrick T. Blanchard
                      ManTech Environmental Technology, Inc.
                         Research Triangle Park, NC 27709

                                    and

                                Teri L. Conner
                       National Exposure Research Laboratory
                       U.S. Environmental Protection Agency
                         Research Triangle Park, NC 27711
                             Contract 68-D-00-206
                                Project Officer
                                 Curtis Morris
                  Human Exposure and Atmospheric Sciences Division
                       National Exposure Research Laboratory

                            Work Assignment Manager
                                Teri L. Conner
                       National Exposure Research Laboratory
                        U.S. Environmental Protection Agency
                         Research Triangle Park, NC 27711
                       National Exposure Research Laboratory
                        Office of Research and Development
                       U.S. Environmental Protection Agency
                         Research Triangle Park, NC 27711

-------
                                         Notice
The information in this document has been funded wholly by the United States Environmental Protection
Agency under contract number 68D00206 to ManTech Environmental Technology, Inc. It has been subjected
to the Agency's peer and administrative review and has been approved for publication as an EPA document.

-------
                                         Abstract
Scanning Electron Microscopy (SEM) coupled with Energy-Dispersive X-ray analysis (EDX) is a powerful
tool in the characterization and source apportionment of environmental paniculate matter (PM), providing size,
chemistry, and morphology of particles as small as a few tenths of a micrometer. Such information can reveal
information about emission sources which cannot be determined through bulk chemical analysis. Automated
SEMs capable of routinely analyzing hundreds of particles per hour have dramatically increased the throughput
of SEM/EDX, making it feasible to conduct statistically meaningful analyses of PM samples and to generate
large data sets for source apportionment studies.
     The National Exposure Research Laboratory (NERL) of the U.S. EPA has been developing and evalu-
ating the use of SEM/EDX to characterize ambient and source-derived particles. The present document, which
evolved over several years as a product of research carried out in support of the U.S. EPA/NERL SEM/EDX
Laboratory, is intended to provide guidelines for researchers using SEM/EDX for aerosol characterization and
source apportionment. Topics include laboratory  procedures  for sample handling, sample preparation,
guidelines for successful manual and automated SEM/EDX analyses, data interpretation, issues relating to data
quality and method validation and case studies highlighting the use of SEM/EDX in PM research.

-------
                                         Foreword
The National Exposure ResearchLaboratory, Research Triangle Park, North Carolina, conducts intramural and
extramural researchinthe chemical, physical, and biological sciences. The researchis intended to characterize
and quantify ambient air pollutant levels and the resulting exposures of humans and ecosystems; to develop
and validate models to predict changes in air pollutant levels; to determine source-receptor relationships
affecting air quality and pollutant exposures; and to solve scientific problems relating to EPA's mission
through long-term investigation in the areas of atmospheric methods, quality assurance, biomarkers, spatial
statistics, exposure assessment, and modeling. The Laboratory provides support to program and regional
offices and state and local groups. This support includes technical advice, methods research and development,
quality assurance, field monitoring, instrument development, and modeling for quantitative risk assessment
and regulation. The Laboratory also collects, organizes, manages, and distributes data on air quality, human
and ecosystem  exposures and trends for the program and regional  offices,  the  Office of Research and
Development, the scientific community, and the public.

                                                Gary  J. Foley, Ph.D.
                                                Director
                                                National Exposure Research Laboratory
                                                Research Triangle Park, NC 27711
                                               in

-------
                                        Contents
Notice	i
Abstract  	 ii
Foreword  	iii
Figures	vii
Tables	viii
Acronyms and Abbreviations	ix
Acknowledgments	 x
Chapter 1:
Chapter 2:
Chapter:
Introduction 	  1
Overview of SEM/EDX Techniques and Instrumentation	  3
2.1  Overview of SEM/EDX  	  3
               2.1.1
            Overview of SEM	  3
                   Image Formation	  3
                   Electron Beam Production 	  4
                   Probe Size and Beam Current	  5
                   Accelerating Voltage 	  5
                   Guidelines for Imaging	  6
            Overview of EDX	  7
2.
2.
2.
2.
2.
0\
2.
2.
2.
2.
2.





e
.:
.:
.:
.:
.:
.1
.2
o
.J
.4
.5
rvi
>.l
1.2
1.3
IA
1.5
    2.1.2
                   Effective X-ray Volume 	  8
                   Choice of Primary Beam Energy  	  9
                   Choice of Spot Size or Probe Current	  9
                   Spectral Artifacts and X-ray Line Overlaps	  9
                   Guidelines for Qualitative EDX Analysis	  10
    2.1.3    Overview of CCSEM  	  10
2.2 Description of NERL Facilities	  10
    2.2.1    LEO/PGT System	  11
    2.2.2    Aspex Personal SEM System 	  11
    2.2.3    Vacuum Evaporators and Other Supporting Equipment	  12
Recommended Techniques and Procedures for SEM/EDX and CCSEM  	  13
3.1 Sample Receipt, Tracking, and Storage	  13
    3.1.1    Sample Submittal Form	  13
    3.1.2    Log-in Procedures	  13
    3.1.3    Sample Tracking	  13
    3.1.4    Sample Storage	  13
3.2 Sample Preparation	  14
    3.2.1    Choice of Filter Media  	  14
    3.2.2    Paniculate Loadings  	  15
    3.2.3    Mounting Filter Samples	  15
            3.2.3.1  Mounting Polycarbonate Filters	  16
            3.2.3.2. Mounting Teflon Filters 	  16
                                             IV

-------
                3.2.4   Preparing Bulk Dusts for SEM	  16
                       3.2.4.1  Procedure for Wet Preparation of Bulk Dusts  	  16
                       3.2.4.2  Procedure for Dry Preparation of Bulk Dusts	  17
                3.2.5   Coating Mounted Samples  	  18
            3.3  SEM/EDX Data Acquisition	  18
                3.3.1   Manual	  19
                3.3.2   CCSEM	  19
                3.3.3   Semi-Automated SEM/EDX	  21
                3.3.4   Size-Only	  21
                3.3.5   Dedicated Particle Searches  	  22
            3.4  Particle Classification 	  22
                3.4.1   Default Four-Element Typing	  22
                3.4.2   User-Defined Rules	  25
                3.4.3   Cluster Analysis 	  25
                3.4.4   Spectrum Matching	  27
                3.4.5   Neural Networks	  29
Chapter 4:   Data Quality and Validity	  31
            4.1  Instrument Calibration and Maintenance	  31
                4.1.1   Magnification Calibration	  31
                4.1.2   EDX Spectrometer Gain Calibration	  31
                4.1.3   Window Contamination Check	  32
            4.2  Precision and Accuracy of Particle Volume Estimates	  32
            4.3  Precision and Accuracy of Particle Mass Estimates 	  33
            4.4  Analysis of Ultrafine Particles	  34
                4.4.1   Limitations of the Present NERL SEM Facilities	  34
                4.4.2   Other Microscopic Techniques	  35
                4.4.3   Non-Microscopic Techniques	  36
                4.4.4   Ultrafine Summary  	  36
            4.5  Carbonceous and Submicron Particles  	  36
            4.6  CCSEM Data Quality and Validity	  38
                4.6.1   Precision of CCSEM (Repeat Analyses of Same Sample) 	  39
                4.6.2   Representativeness of Data	  40
                4.6.3   Errors Associated with CCSEM	  41
                4.6.4   EDX Acquisition Time	  42
Chapter 5:   Examples of Research Applications	  45
            5.1  Examples from the Literature  	  45
                5.1.1   Aerosol Characterization	  45
                       5.1.  .1  Fly Ash	  45
                       5.1.  .2  Carbonaceous Particles	  45
                       5.1.  .3  Sulfates	  46
                       5.1.  .4  Marine Aerosol	  46
                       5.1.  .5  Miscellaneous	  47
                5.1.2   Source Apportionment  	  47
                5.1.3   SEM/EDX Methodology	  47
            5.2  Phoenix PM2 5 Samples	  48
                5.2.1   Unmix Receptor Model Findings  	  48
                5.2.2   SEM/EDX Findings  	  48
                5.2.3   Summary of Results  	  50
            5.3  Fort Hall Source Apportionment Study	  50
            5.4  Baltimore Retirement Home Study	  53
                5.4.1   Methods  	  53
                5.4.2   Summary of Manual SEM/EDX Analysis Results  	  54
                5.4.3   Summary of CCSEM Analysis Results	  54
                5.4.4   Conclusions	  57

-------
           5.5 World Trade Center Study	  57
               5.5.1    Verification of XRF Results	  57
               5.5.2    Analysis of Bulk Sample	  57
               5.5.3    Analysis of Ambient Air Sample 	  58
           5.6 Source Particle SEM/EDX Data	  58
Chapter 6:  References  	  61
Appendix A  	  66
    Example of LEO/PGT Data Output  	  67
    Example of PSEM Data Output 	  69
Appendix B  	  71
    Submittal Formfor SEM/EDX Samples	  72
    Assignment of SEM IDs	  73
    Sample Log-In/Log-Out	  74
Appendix C  	  75
    Particle Classification Rules for PM10_2 5 (Coarse) Particles	  76
    Particle Classification Rules for PM2 5 (Fine) Particles	  77
Appendix D: Particles from Ambient Air Sample Collected in NYC Near WTC Site  	  78
Appendix E: Source Particle Atlas	  83
                                             VI

-------
                                         Figures
Number                                                                               Page

2-1  Photomicrograph of Ambient, Coarse-Fraction Air Sample Collected in Baltimore	  3
2-2  Photomicrograph of Lead-rich House Dust. Left: SE Image.  Right: BSE Image  	  4
2-3a Photomicrograph of Particulates from a Cement Kiln: Tungsten gun, 2.5 nA Probe Current,
     1000 cps  	  5
2-3b Identical Field Imaged withLaB6 Gun, 0.55 nA Probe Current, 1000 cps	  5
2-4  Photomicrograph of Pollen Grain Imaged with the LEO S440 at 30, 10, and 5 keV	  5
2-5  EDX Spectrum of Iron-Rich Aluminum Silicate Particle	7
2-6  EDXSpectraof 0.16, 0.3, and 1.0 /j,m Aluminum Silicate Particles	  8
2-7  Photo of the LEO S440 SEM withPGT IMIX EDX System	11
2-8  Photo of the Personal SEM (PSEM) System	  11
2-9a Photomicrograph of Salt Crystals (PSEM with Tungsten Gun) 	  12
2-9b Photomicrograph of Salt Crystals (LEO 440 with LaB6 Emitter)	  12

3-1  Photomicrographs of Four Filter Substrates Commonly Used for Aerosol Sampling	  14
3-2  Vacuum Filtration Apparatus Used for Wet Preparation of Bulk Dusts	  16
3-3  "Mini-cyclone" Apparatus Used for Dry Preparation of Bulk Dusts	  17
3-4  "Puffer" Apparatus Used for Dry Preparation of Bulk Dusts	  18
3-5  Photomicrographs and EDX Spectra of Mercury-Rich Particles Collected Downwind
     of Elemental Phosphorus Plant. Left: SE Image. Right: BSE Image  	  23, 24
3-6  Plot of Cluster Analysis "Stopping Rules" Calculated for a Simulated Data Set	  26
3-7  Plot of Hierarchical Cluster Analysis Populations of Simulated Data Set 	  26
3-8  Plot of Revised Cluster Populations After Non-Hierarchical Cluster Analysis  	  27
3-9a Plot of Source Apportionment Results Using Least-Squares Spectrum Matching	  29
3-9b Refined Source Apportionment Results Using Least-Squares Spectrum Matching	  29

4-la PGT Versus PSEM Volume Errors for Spherical Particle, as Function of Magnification  	  32
4-lb PGT Versus PSEM Volume Errors for Flat Particle, as Function of Magnification	  32
4-2  PSEM Particle Size Distributions for Two Measurement Thresholds	  34
4-3  LEO 440 Photomicrograph of Ultrafine Particle	  35
4-4a FE-SEM Photomicrograph of Gold Particles  	  35
4-4b FE-SEM Photomicrograph of TiO2 particles	  35
4-5  Superimposed EDX Spectra for Polycarbonate Filter and Formvar	  37
4-6  Photomicrograph of Submicron Diesel Soot Particle 	  37
4-7  Superimposed EDX spectra of Diesel Soot Particle and Blank Formvar	  37
4-8  Top: EDX Spectrum of a 0.22-um Aluminum Silicate Particle on Polycarbonate Substrate.
     Bottom: EDX Spectrum of a 0.24-um Aluminum Silicate Particle on Formvar	  37
4-9  Precision of Repeated CCSEM Analyses  	  39, 40
4-10 CCSEM Particle Classification Results for Repeat Analyses at Different X-ray Collection
     Times	  44
                                             vn

-------
5-1  Photomicrographs and Spectra for PM2 s Particles Collected in Phoenix Air Sample  	  49
5-2a Photomicrograph and Spectrum for Fine-Fraction Ambient Sample Collected Downwind
     of Elemental Phosphorus Plant	  51
5-2b Photomicrograph and Spectrum for Coarse-Fraction Ambient Sample Collected Downwind
     of Phosphorus Plant	  51
5-3  Photomicrograph and Spectrum of Fine-Fraction Ambient Sample Collected  Downwind
     of Phosphorus Plant During PM10 Exceedance	  51
5-4  Top: Photomicrograph and Spectrum of Fine-Fraction Sample Collected Downwind of
     Phosphorus Plant. Bottom: Photomicrograph and Spectrum of Ground Flare Plume Sample	52
5-5  Photomicrograph and Spectrum of Phos-Dock Source Sample	  52
5-6  Photomicrograph and Spectrum of Fly Ash Sphere from the Burden Level of the Furnace
     Building	  52
5-7  Comparison of Ambient, Coarse-Fraction Particle Classes at Residential Indoor Site, Residential
     Outdoor Site, and Community Site	  55
5-8  Comparison of Ambient, Fine-Fraction Particle Classes at Residential Indoor Site, Residential
     Outdoor Site, and Community Site	  56
5-9  Photomicrograph of Gold Particle in World Trade Center Dust Sample	  57
5-10 Photomicrographs of Paniculate Matter in Bulk Sample Collected Near the WTC Site  	  58
5-11 Photomicrographs and Spectra of Individual Fibers in Bulk Sample Collected Near the WTC Site   59
5-12 Photomicrographs and Spectra of Individual Particles (<10 ,wm) in Bulk Sample Collected
     Nearthe WTC Site  	  60
5-13 Low-Magnification Image of VAPS Coarse-Fraction Filter Showing "Honeycomb" Particle
     Distribution	  61

Appendix D: Photomicrographs and Spectra of Individual Particles in Ambient Air Samples Collected
     Nearthe WTC Site  	80
Appendix E: Selections from the NERL SEM/EDX Source Particle Atlas	  85
                                         Tables
Number                                                                              Page

4-1  LEO-PGT and PSEM Analysis of a 22-^g Lutetium Oxide Sample: Estimated Particle Mass Using
     Different Volume Formulas and Associated Errors 	  33
4-2  EDX Collection Time versus CCSEM Particle Composition (in weight percent concentration)  ...  43
                                            Vlll

-------
                         Acronyms and Abbreviations
aed      aerodynamic diameter
AEM    analytical electron microscope
AFM    atomic force microscopy
ANN    artificial neural network
BSE     backscattered electron
CCSEM  computer-controlled SEM
CMB    chemical mass balance
CPC     condensation particle counter
DMPS   differential mobility particle sizer
EDX    energy-dispersive X-ray
EF       enrichment factor
ELPI    Electrical Low Pressure Impactor
FE       field emission
FE-SEM  field-emission SEM
FMC    FMC Corporation
HgP     particle-bound mercury
LaB6    lanthanum hexaboride
MCE    mixed cellulose ester
MSEM   manual SEM
NAAQS  National Ambient Air Quality Standard
NERL   National Exposure Research Laboratory
NIST    National Institute for Science and Technology
PCB     particle class balance
PGT     Princeton Gamma-Tech
PM      paniculate matter
PSEM   Personal SEM
ROI     Region of Interest, or X-ray "energy window"
RSD     relative standard deviation
SAED   selected-area electron diffraction
SE       secondary electron
SEM    scanning electron microscopy
SMPS   scanning mobility particle sizer
SOP     standard operating procedure
STEM   TEM operated in the scanning mode
TEM    transmission electron microscopy
WAM   Work Assignment Manager
WD     working distance in mm
WDX    wavelength-dispersive X-ray
XRF     X-ray fluorescence
Z        atomic number
                                             IX

-------
                             Acknowledgments
The authors wish to thank Mr. Gary Casuccio and Mr. Steven Schlaegle of R.J. Lee Group, Inc., for many
useful discussions and suggestions.

-------
                                                   Chapter 1
                                                 Introduction
In July 1997, the U.S. EPA (U.S. EPA, 1997) promulgated a
new National  Ambient Air Quality Standard (NAAQS) for
paniculate matter (PM). The new standard is aimed at regulating
ambient concentrations of PM10 and PM2 5, particles with aero-
dynamic diameter (aed) <10 ,wmand <2.5 /j,m, respectively. The
new NAAQS was developed in response to studies that showed
a significant associationbetween human morbidity and mortality
and PM concentrations (Dockery  et al., 1993; Dockery and
Pope, 1994; Schwartz, 1994; Pope et al., 1995). Other studies
indicate that transition metals may play a significant role in the
health consequences of particle exposure (Ohio et al., 1992,
Ohio et al., 1996). These studies coupled with the new NAAQS
have  generated  increased interest in analytical techniques
capable of measuring the size, morphology, and chemical com-
position of individual aerosol particles. These data are essential
to the understanding of particle formation, transport and trans-
formation, and deposition mechanisms as well as the impact of
particles  deposited in the respiratory system. Furthermore,
chemical and physical characterization of individual particles
can reveal source information that cannot be determined through
bulk chemical characterization such as X-ray fluorescence
(XRF). Individual particle analysis is thus a complement to bulk
elemental analysis techniques as well as a potentially powerful
tool in source apportionment research.
     The National Exposure Research Laboratory (NERL) of
the U.S. EPA has been developing and evaluating the use of
scanning electron microscopy (SEM) to characterize coarse and
fine ambient and source-derived particles. Modern SEMs exhibit
a large depth-of-field and spatial resolution on the order of 30 A
or less. Typically, particles as small as 0.05 ,wm can be imaged
by SEM, providing information on the physical properties of
particles including size, shape, and surface morphology. SEMs
equipped with energy-dispersive (EDX) or wavelength-disper-
sive (WDX) X-ray spectrometers are, in addition, able to provide
information on the chemistry of individual particles: elemental
associations within individual particles  can  be revealed, and
particle chemistry as a function of particle size can be deter-
mined. The importance  of individual particle analysis as a
complement to bulk analysis has been highlighted in many
studies.
     The advent of computer-controlled SEM (CCSEM) in the
past two decades has enabled automated physical and chemical
characterization of filter-based aerosol samples at throughputs
that can approach 1000 particles per hour, making it possible to
examine a statistically significant sample of particles in a reason-
able time (Johnson et al., 1981; Casuccio et al., 1983a; Bernard
etal., 1986; Andersonetal., 1988; Saucy etal., 1987;  Germani,
1991; Van Borm et  al., 1989). Whereas bulk analytical tech-
niques such as XRF yield quantitative elemental concentrations
averaged over the entire filter, CCSEM typically provides qual-
itative composition data on individual particles comprising the
aerosol sample. Elemental composition data combined with
physical properties of the particle provide the basis for classi-
fying individual particles into different classes, using user-
defined classification rules or statistical methods, which can
often be related to known emission sources. To the extent that
particle morphology and chemistry can be used to define distinct
particle types,  CCSEM can quantitatively characterize the
composition of a sample in terms of the number percent (or mass
percent, if sufficient information is available) represented by
each particle type. Multivariate analysis of a large set of such
ambient CCSEM  classifications can provide information on
source-receptor relationships.
     The present  document  evolved over several years as  a
product of research  carried out  in support of the U.S. EPA/
NERL SEM/EDX laboratory. It is intended to provide guidelines
for researchers using SEM/EDX inboth a manual and computer-
controlled fashion for aerosol characterization and source appor-
tionment. It is not intended to be a comprehensive manual for
SEM analysis; a number of excellent, general-purpose textbooks
are available  for this purpose (e.g., Goldstein et al., 1992.)
Neither is it intended to serve  as a standard operating procedure
(SOP) for the analysis of samples by SEM/EDX or CCSEM.
Samples are submitted for SEM/EDX or CCSEM analysis in a
wide variety of forms to support a wide variety of projects and
research objectives. It is often necessary to develop customized

-------
sample preparation and analysis procedures to meet the specific
research objectives.  Moreover, different research laboratories
have different analytical and  support  equipment.  For these
reasons, it is not practical to develop a generic SOP for SEM/
EDX or CCSEM that is applicable in all situations.  While the
procedures and applications described in this document often are
specific to the instrumentation available in the NERL SEM/
EDX laboratory, readers can hopefully adapt some of these
recommendations to instrumentation available in their  own
laboratories.
     Chapter 2 of this document provides an introduction to the
SEM/EDX techniques, including basic concepts and hardware,
and a description of the NERL SEM/EDX facilities.  Chapter 3
presents recommended procedures for sample preparation and
analysis. Chapter 4 presents research related to SEM/EDX data
validity and quality  assurance and quality control  issues. In
Chapter 5, examples of research applications are presented.

-------
                                                  Chapter 2
                 Overview of SEM/EDX Techniques and  Instrumentation
2.1  Overview of SEM/EDX

2.1.1  Overview of SEM

2.1.1.1 Image Formation
The  SEM operates by scanning an energetic, finely focused
electron beam over an individual feature or a field of features.
This primary electron beam interacts with the specimen pro-
ducing a variety of secondary signals that can be monitored with
appropriate detectors. These signals can be collected in syn-
chronization with the position of the scanned electron beam to
generate high-resolution images providing detailed spatial and
composition information.  Figure 2-1 is a secondary electron
photomicrograph of a fairly typical  field of coarse-fraction
particles (2.5 /j.m aed < particle diameter < 10 /j.m aed) from an
urban air sample. The sample was collected in Baltimore on a
37-mm polycarbonate filter (0.4 ,wm pore size) over a 24-h
period, using a Versatile Air Pollutant sampler (VAPS). Particle
size  is indicated by the scale bar in the  lower left  corner.
The  particles display a variety of sizes, morphologies,  and
Figure 2-1. Photomicrograph of ambient, coarse-fraction air sample
collected in Baltimore, MD.
compositions reflecting different sources. The large, fluffy
particle in the upper left center is a soot agglomerate, perhaps
from diesel combustion. Two bright, micrometer-sized, iron-rich
spheres are observed: probable  products of combustion from
iron works or steel mills. The elliptical particle in the lower left
of the field is carbonaceous and appears to be a pollen or spore.
Most of  the remaining particles are various  crustal-related
quartz, aluminum silicates, and calcium-rich particles represent-
ing resuspended dust. The small, submicrometer holes observed
in the background are the pores of the polycarbonate filter.
     When the primary beam interacts with the specimen, the
electrons  undergo two types of collisions: (1) elastic collisions
in which the energy of the incident electrons is unchanged, and
(2)  inelastic collisions in which the primary  electrons lose
energy in a succession of collisions, leaving the electrons with
lower energy. Elastic collisions give rise to the backscattered
electron (BSE) signal,  which forms the BSE image. Inelastic
collisions give rise to the secondary electron (SE) signal, which
forms the SE image. The SE and BSE images provide different
but complementary information. Secondary electrons are emitted
from the atoms occupying the top surface and produce a readily
interpretable image of the  surface. The contrast in the image is
determined by the sample morphology. The SE image possesses
three-dimensional perspective, high  depth of field, and the
appearance of overhead illumination.  A high-resolution image
can be obtained because of the small diameter of the primary
electron beam. Backscattered electrons are primary beam elec-
trons, which are "reflected" from atoms in the solid. The contrast
in the BSE image is determined largely, though not exclusively,
by the atomic number of the elements in the sample. The image
can therefore show the distribution of different chemical phases
in the sample. Because backscattered electrons are emitted from
a depth in the sample, the resolution in the image is not  as good
as for secondary electrons.
    Figure 2-2 illustrates the difference in information provided
by the two types of images. The image  on the left is the SE
image and the image on the right is the BSE image of the same
particle. The particle is from a sample of lead-rich house dust

-------
Figure 2-2. Photomicrograph of lead-rich house dust particle. Left image:
secondary electron (SE) image. Right image:  backscattered electron
(BSE) image. Bright inclusions in BSE image are lead-rich. The bulk of
the particle is aluminum-silicate. Outlying particles that nearly disappear
in the BSE image are carbonaceous.

and has been collected on a polycarbonate filter. The SE image
is superior for displaying surface detail and particle morphology
but does not generally show chemical heterogeneity. Changes in
brightness  within the BSE image correspond to changes  in
effective atomic number: the small, very bright inclusions on the
order of one micrometer are lead-rich particles. The host matrix
(medium-gray areas) is an aluminum-silicate mineral. Areas that
are intermediate in brightness, such as the bottom left quadrant
of the particle, have a higher Si/Al ratio. (Although differences
in particle thickness can also account for brightness variations in
the BSE image, the higher Si/Al ratio was confirmed by EDX
analysis). Some smaller, neighboring particles that appear in the
SE image have approximately the same brightness level as the
polycarbonate substrate and hence are nearly invisible in the
BSE image. These particles are mostly carbonaceous.

2.1.1.2 Electron Beam Production
Primary electron beams for electron microscopy can be gen-
erated using different emission sources: the tungsten hairpin gun,
the lanthanum hexaboride (LaB6) gun, and the field-emission
(FE) electron gun—in order of increasing brightness, cost, and
complexity. Perhaps the most important measure of electron gun
performance is gun brightness.  Brightness is  defined as the
electron current density per solid angle of the electronbeam. The
microscopist can optimize the primary beam energy and/or the
beam spot size (beam current) depending on the analytical objec-
tives: highest resolution images are obtained with the smallest
possible probe  diameter (diameter of the  electron beam at the
particle surface), while maximum image brightness and efficient
X-ray microanalysis is obtained with the largest possible probe
current. The brighter the electron gun, the better these competing
requirements can be satisfied. Other performance factors impor-
tant in comparing electron guns are lifetime and stability. Com-
                                                             pared to tungsten, LaB6 offers longer life and higher current
                                                             density. Not only does LaB6 provide a factor of 10 increase in
                                                             brightness over tungsten, it also has a smaller effective source
                                                             size, hence smaller spot size. Disadvantages of LaB6 are its long
                                                             warm-up time (about 15 min) and its extreme chemical reactivity
                                                             when hot. In practical terms, this means that the vacuum in the
                                                             gun chamber must be approximately two orders of magnitude
                                                             better than the pressure required to operate tungsten. Typically,
                                                             the required low pressures are achieved by adding a dedicated
                                                             ion pump in the gun chamber.
                                                                  The improved image quality and image resolution obtained
                                                             with the LaB6 gun compared to the tungsten hairpin gun is
                                                             shown in Figures 2-3a and 2-3b. These figures show images of
                                                             the same particle field acquired with tungsten (Figure 2-3a) and
                                                             LaB6 (Figure 2-3b). Probe diameter was adjusted for each gun
Figure 2-3a. Photomicrograph acquired with tungsten filament. The probe
current was 2.5 nA, which yielded an EDX count rate of 1000 cps.
Figure 2-3b. Photomicrograph (identical field) acquired with LaB6 filament
at 0.55 nA probe current and 1000 cps.

-------
to yield the same X-ray count rate of 1000 cps, which approxi-
mates a typical EDX analysis count rate. The LaB6 gun, operat-
ing at a probe current of 0.55 nA, shows improved resolution
due to the smaller probe diameter compared to the tungsten
image acquired at 2.5 nA. Moreover, if X-ray microanalysis
were the primary interest, the LaB6 probe diameter could be
increased to match the image resolution of the tungsten gun,
with a sizable increase in X-ray count rate and decreased X-ray
analysis time. LaB6 crystals are more expensive to replace than
tungsten  filaments,  but the  lifetime of the LaB6 crystal  is
approximately 10 times that of tungsten.
     Conventional tungsten and LaB6 filaments are thermionic
sources in that the electrons are liberated from the filament by
heating the filament. InFE sources, the electrons are drawn from
a finely pointed filament tip (usually a tungsten crystal) by
means of an electric field. FE is 10-100x brighter than LaB6 and
-lOOOx brighter than tungsten. In addition, FE sources exhibit
a much greater depth of field than thermionic sources (tungsten
and LaB6). On the negative side, FE sources must be operated at
high vacuum (~10"9 torr) to stabilize electron emission and to
avoid contamination of the FE tip.

2.1.1.3 Probe Size and Beam  Current
Source brightness increases approximately proportionally to the
beam energy. Thus the probe size and beam current are inti-
mately related and are a function of beam energy in addition to
gun type. For high-resolution imaging it is desirable to maintain
a probe diameter of 10 nm or less. At 30-keVbeam energy, a 10-
nm spot  size corresponds to a beam current of -0.0InA for
tungsten and -0.2 nA for LaB6 (Goldstein et al., 1992). Note
that X-ray microanalysis typically requires about 0.1  nA  of
current. Thus, with tungsten it is difficult to carry out efficient
X-ray microanalysis simultaneous with high-resolution imaging.

2.1.1.4 Accelerating Voltage
The energy of the primary electron beam is typically adjustable
over a range from approximately 1 keV to  20 or 30 keV. This
enables the user to optimize  the instrument for image quality
objectives or for X-ray microanalysis.  The  highest feature
resolution, in theory, is provided by the most energetic electron
beam (higher energy = shorter wavelength = better resolution).
However, higher energy also means that the SE electrons that
form the particle image are generated over a much larger volume
of the specimen due to the deeper penetration of the electron
beam.  Lowering  the  beam  energy  shifts the production  of
detected secondary electrons toward the surface of the particle,
with a consequent enhancement of surface detail in the  SE
image. The effect of primary beam energy on the image of a
pollen grain is shown in Figures 2-4a, b, and c acquired at 30,
10, and 5 keV, respectively. The difference in resolution among
the three images is subtle, but the images acquired at 5 and 10
keV are clearly superior to the 30-keV image in terms of surface
detail.
Figure 2-4. Pollen grain imaged  with the LEO S440 at different
accelerating voltages. Top: 30 kV.  Middle: 10 kV. Bottom: 5 kV.

-------
2.1.1.5 Guidelines for Imaging
The following steps are recommended for obtaining the best
images and are summarized from Goldstein et al. (1992), which
has an excellent chapter on image formation and interpretation.
     •  Choose the accelerating voltage  appropriate for
        your objectives: high keV for best image resolution,
        low keV to enhance surface detail.
     •  Optimize working distance (WD). WD is the distance
        between the sample surface and the lower surface of
        the pole piece. Generally the best image resolution is
        obtained at the shortest working distances. Typical
        imaging WDs are on the order of 10 mm. However, for
        many instruments, the analytical WD (the optimum
        working distance for EDX analysis) is often larger than
        the optimal imaging WD due to EDX detector/sample
        geometry. The two SEMs in the NERL SEM labora-
        tory, for example, have analytical WDs of about 17 mm
        (Personal  SEM) and  24 mm (LEO S440).  This is a
        significant inconvenience requiring the  user to suc-
        cessively raise the stage for the best  particle image,
        thenlowerthe stage forthe EDX analysis. If automated
        analyses are being conducted in whichboth images and
        X-ray  spectra are  collected,  the working  distance
        typically must be set to the analytical WD, with some
        corresponding loss of image quality.
     •  Optimize final aperture size Dap. The depth of focus
        for a given magnification is roughly proportional to
        WD/Dap, where Dap is the diameter of the final aper-
        ture. If depth of field is of primary importance in the
        image, then increase the WD and/or decrease aperture
        size, if that is an option. (Some SEMs have a single,
        fixed-diameter final aperture.)
     •  Optimize probe size. Electron scattering within the
        sample (determined by the accelerating voltage and the
        effective atomic number of the sample) causes the
        effective signal-producing area on the specimen to be
        significantly larger than the physical spot size of the
        incident probe, especially for BSE signals, which com-
        prise more energetic electrons. Reducing probe current
        generally yields improved image  resolution, until at
        some  point,  image quality suffers due to decreasing
        brightness and contrast. For low-magnification work
        (<5000x), a larger probe diameter (current) will prob-
        ably work better than a finely focused, high-resolution
        probe. Typically one needs a minimum of approxi-
        mately 0.01 nA of beam current for imaging.
     •  Minimize specimen charging. Specimen charging oc-
        curs when one or more  areas of the specimen cannot
        dissipate the incident electronic charge which arrives at
        the surface via the incident or scattered electron beam.
        Charging is largely  a function of the local  surface
        conductivity of the specimen as well as the primary
        beam energy and probe current.  Charging causes a
        variety of undesirable effects in electron micrographs,
        including abrupt brightness changes within the image,
        image shift/image distortion, and particle displacement
        (the buildup of electrostatic forces can actually cause
        charging particles to move or completely leave the
        sample substrate). The conductivity  of insulating or
        nonconductive PM specimens has traditionally been
        made acceptable by  overcoating the sample with
        approximately 100-200 A  of conductive carbon or
        metal. This is accomplished in the NERL laboratory
        using a vacuum coater. If a coated specimen still ex-
        hibits charging  problems,  the  user can minimize
        charging by a combination of (1) reducing the primary
        beam energy, (2) reducing the probe current, and (3)
        acquiring images in the frame-averaging mode rather
        than the line-averaging mode (if this option is available
        on the SEM). Reducing beam energy and/or probe
        current, however, is likely to conflict with the needs for
        EDX analysis. Alternatively, "environmental" or "vari-
        able  pressure" SEMs,  which allow operation at high
        pressure (in the range of 0.1-1 kPa), effectively elimi-
        nate  specimen charging (hence also the need to coat
        samples) even for insulating samples.

The following tips are suggested for optimizing BSE images:

     •  Long WD yields best contrast but least topographic
        detail. Medium and short WD gives best BSE signal,
        topographic detail and good contrast.
     •  Beam energy must be high enough for  good Z
        contrast. Higher Z elements require  higher keVs. At
        lower keVs atomic number discrimination is reduced
        and at really low keVs one can get reversal of contrast
        of high Z elements.
     •  Signal level is very important. One cannot get good
        Z contrast if the BSE signal is too low! Lower signal
        equals a reduction in contrast.
     •  BSE amplifier gain also plays a role in contrast and
        noise in the final image.

A final word on the importance  of placing a reference scale bar
in the  image. Image magnification is defined as the  linear
dimension of the scan on the output device (monitor or printer
device) divided by the linear dimension of the scan on the
sample. Magnification is thus a hardware-dependent parameter.
The same feature image  acquired at the same nominal magni-
fication but with different output devices (polaroid film, thermal
printer, computer monitor, etc.), may have very different sizes so
that magnification does not immediately convey the size of the
features in the image. Rather than magnification, it is generally
recommended that a scale bar (micrometer marker)  be included
in each image to calibrate feature size.

-------
2.1.2  Overview of EDX
Interaction of the primary  electron beam with atoms in the
sample causes inner electron shell transitions, which result in the
emission of  X-rays.  Two  types of X-rays  are generated:
(1) Bremsstrahlung or continuous  X-rays, which generate a
broad and slowly-varying background over the entire X-ray
spectrum, and (2) characteristic X-rays, which are  narrow,
discrete peaks in the spectrum whose energies are characteristic
of specific elements present in the sample. A fraction of the X-
rays emitted  by the specimen are collected and analyzed by
means of an EDX and/or WDX analyzer. EDX provides a far
more rapid analysis and is therefore much better  suited for
CCSEM; WDX is  generally  more  sensitive  and accurate,
especially for the lighter elements. Our discussion will focus on
EDX analysis since this is the approach used in the NERL
laboratory and by the great majority of CCSEM users. Detection
and measurement of the characteristic X-ray energies yields an
analysis of the elements present in the feature. EDX spectro-
scopy can provide rapid, qualitative analysis, or, with adequate
standards, quantitative analysis of elemental composition with
a  sampling  depth of 1-2  um. X-rays  can  be collected in
synchronization with the image-forming electron beam to form
two-dimensional maps or one-dimensional line profiles showing
the distribution of elements in a sample surface. For an excellent
discussion of X-ray analysis, the reader is  referred to the text by
Goldstein etal. (1992).
     Figure 2-5  shows a representative  EDX spectrum. The
spectrum is that of an Fe-rich aluminum silicate particle from a
Phoenix air sample.  The spectrum was collected for 200 s at a
primary beam energy = 20 keV, probe current =100 pA, WD =
23 mm, and X-ray  count rate = 1000 cps. Note the  discrete
characteristic X-ray peaks superimposed on the Bremsstrahlung
background which varies slowly with X-ray energy.
Figure 2-5. EDX spectrum of an iron-rich aluminum silicate particle from
a Phoenix air sample. Spectrum was collected for 200 s at a beam energy
of 20 keV, probe current of 100  pA, count rate approximately 1000 cps,
and working distance of 23 mm.
     EDX detectors can be supplied with no window, a thin
window, or a thick window depending on the requirements of
the user. Thick window detectors have poor or zero transmission
of low-energy X-rays and are typically insensitive to elements
lighter than Na. Thin window detectors (which are perhaps most
commonly used in the general area of environmental analysis)
and windowless detectors provide limited detection of elements
below Na and poor or no detection of elements below beryllium
or boron.
     EDX detectors are characterized by their energy resolution.
Typical detector resolutions are 130-150 eV at 5.9 keV (Mn Ka
line). This resolution is much too broad for inferring chemical
states (e.g., elemental sulfur vs. sulfate vs. sulfite), but is gen-
erally adequate to separate characteristic X-ray lines from neigh-
boring elements. Nevertheless, interference between peaks of
different X-ray line families is a potentially serious problem for
EDX. This issue is discussed in more detail in Section 2.12 A.
     The EDX  detection limit for elements Na to U in bulk
materials and in individual particles larger than about 2 um is on
the order of 0.1 wt %  (Buseck and Bradley,  1982). In a 1-um
particle, the  detection limit increases to about 0.5 wt %. These
are only rough rules of thumb since the detection limit will also
depend on the X-ray collection time, which typically varies from
3 to 100 s. One needs to distinguish these EDX detection limits
from the ability of SEM  to  detect rare  or exotic particles.
Dedicated computer-controlled SEM/EDX searches, for exam-
ple, can detect and provide semi-quantitative mass concentration
estimates for exotic species, typically heavy metals such as Pb
or Hg,  which are present in the bulk sample  at sub-ppm
concentrations but which are present in a few isolated particles
at high wt % concentration. An example  of this capability is
presented in Section 3.3.5. The ability to conduct "needle-in-the-
haystack" analyses potentially  makes  computer-controlled
SEM/EDX an extremely sensitive technique and a powerful
complement to bulk analytical methods. In general, however, if
a species has a bulk mass concentration <0.1  wt %, and is
distributed homogeneously among all particles, it will probably
not be detected by EDX.
     EDX analysis can be performed at two levels of sophis-
tication: quantitative EDX  or qualitative EDX. Under optimal
conditions (appropriate samples, relevant standards, controlled
experimental setup,  and  sophisticated  data-reduction  pro-
cedures), EDX canyield elemental compositions of flat, polished
specimens with accuracies and precisions approaching  1%.
Quantitative analysis of individual particles, however, turns out
to be much more complicated than analyzing flat, polished
specimens. Armstrong and Buseck (1975)  and Buseck and
Bradley (1982) discuss the difficulties inherent  in quantitative
EDX analysis of individual particles. These difficulties largely
result from the need to correct X-ray yields from particles with
irregular surfaces and thicknesses less than the incident electron
range. For particles larger than a few micrometers, geometry-
dependent absorption effects become severe, while for smaller
particles, corrections for atomic number and thickness effects

-------
become critical. Using a correction procedure that accounts for
differences in atomic number, depth distribution of generated X-
rays, and absorption characteristics between the sample and a
standard, Armstrong and Buseck (1975)  and Grasserbauer
(1977) claim that errors can be reduced to 2-5% (relative) for
silicate particles > 0.5 um in diameter. In addition to requiring
a suitable standard, their  correction procedure requires an
estimate of particle thickness, which would seem to necessitate
time-intensive  manual  analysis. While this approach appears
feasible for manual EDX analysis of a few selected particles, to
our knowledge, quantitative EDX analysis of individual particles
is presently not a reality in automated SEM/EDX analysis.
     For the remainder of this document, we will understand
EDX analysis  to mean qualitative analysis since this  is the
approach used in most research efforts that employ SEM/EDX
for source apportionment. Thus, the same QA/QC considerations
that relate to precision and accuracy of quantitative techniques
such as XRF cannot be applied to SEM/EDX analysis. To the
extent that particle morphology and elemental chemistry can be
used  to define distinct particle types, then the SEM/EDX
technique can  quantitatively characterize a sample in terms of
number percent by particle type or chemical  class. However,
with respect to particle sizing or particle mass, SEM analysis
provides only semi-quantitative results since the technique uses
two-dimensional  information  to infer a  three-dimensional
quantity (size or volume).

2.1.2.1 Effective X-ray Volume
Proper interpretation of microanalysis results requires that users
be constantly aware of the difference between image resolution
and X-ray spatial resolution. Secondary electrons that form the
SE image typically come from only a very thin surface layer of
less than approximately 0.01 um in depth. (Secondary electrons
generated at deeper depths within  a particle do  not  have
sufficient energy to escape  the particle and hence do not con-
tribute to the SE image.) Higher energy backscattered electrons
can escape from much greater depths than secondary electrons,
explaining the relatively poorer surface resolution of BSE
images. The effective volume for X-ray production can be still
greater because X-rays penetrate matter far more readily than
electrons. The  user must understand that the SE or BSE image
of a particle and its accompanying X-ray spectrum are generated
from largely different volumes of the particle.  Features seen in
the image may, in fact,  have little correspondence to the X-ray
spectra collected simultaneously with the image. Because elec-
trons  scatter as the primary beam penetrates the target, the
effective interaction volume is  considerably  greater than the
simple product of the incident probe area and the range of a
20-keV electron.  The  effective interaction volume depends
strongly on  both the primary beam  energy  and the atomic
number of the specimen. High beam energies and low-Z targets
result in large interaction volumes, while low beam energies and
high-Z targets result in small interaction volumes. It is important
to note that despite the fact that the incident probe diameter is
well under 1 um in diameter, the interaction volume in a low-
density, low-atomic-number target has  dimensions of cubic
micrometers. Anderson and Hasler (1966) developed the fol-
lowing expression for the X-ray range:

        R = 0.064(E0168-Ec168)/p

where   R = X-ray range or X-ray spatial resolution (um)
        E0 = accelerating voltage (keV)
        Ec = critical absorption energy (keV)
        p = mean sample density (g/cc)

Note that in the above formulation, changing the probe size has
no effect on X-ray spatial resolution. The effective X-ray range
is determined by the beam energy and the sample density.
        The effective range of 20-keV electrons (equivalent to
the radius of a circle centered on the target surface at the beam's
impact point whose circumference defines the envelope of the
interaction volume) is approximately 5.3,4.2,1.5, and 0.9 umin
C, Al, Cu, and Au, respectively (equivalent semi-spherical
volumes ranging from 312 um3 to 1.5 um3). The lower size limit
for qualitative  X-ray analysis of a particle is ultimately deter-
mined by  signal statistics, signal to noise, and particle com-
position, but is on the order of 0.2 um for tungsten and LaB6
guns [Field emission SEMs (FE-SEMs) are capable of reaching
considerably lower size limits - see Section 4.4.2]. This limit is
illustrated in Figure 2-6 which shows EDX spectra acquired on
aluminum silicate particles of different sizes collected on a
polycarbonate  filter.  The instrument parameters  included  an
electron energy  of 20 keV, probe current of 200 pA, and
         1L
  Figure 2-6.  EDX spectra acquired on 0.16 //m (top), 0.30 //m
  (middle), and 1.0 ,um (bottom) aluminum silicate particles on a
  polycarbonate support. The signal (Al + Si peaks) to background (C
  peak) ratio improves with increasing particle size.

-------
counting time of 100 s. For a 0.16-^m diameter particle (top
spectrum), the Al and Si signals are barely above the noise level
and are swamped by the carbon signal generated by the poly-
carbonate filter substrate. Even with the 1.0-^m particle (bottom
spectrum), the polycarbonate substrate contributes almost half of
the total X-ray counts.

2.1.2.2 Choice of Primary Beam Energy
The choice of primary beam energy for X-ray microanalysis
represents a compromise between the need to efficiently excite
characteristic X-ray production from the specimen and the desire
to minimize specimen self-absorption of the X-rays. The latter
occurs because X-rays produced deep within a particle may be
absorbed within the particle before they can escape the particle
to be detected. The higher the primary energy, the greater the
penetration of the primary beam, and the greater the self-absorp-
tionlosses—especially for low-energy X-rays generated by low-
Z elements. A general rule of thumb is that characteristic X-ray
production is maximized for primary beam energy roughly 1.7
times the energy of the X-ray line being excited. This is called
the overvoltage factor. X-ray microanalysis is typically carried
out at 15-20 keV, which represents a compromise between the
need for adequate overvoltage to efficiently excite most elements
of interest and the  desire to minimize particle self-absorption
losses. With a beam energy of 15 keV or more, all possible X-
ray lines of an element in the range of 0.1-10 keV will be
efficiently excited (i.e., elements from Be to Ge, plus heavier
element L and M X-ray lines.).  Still, in larger particles, self-
absorption losses of light elements will be significant at primary
beam energies of 20 keV. If low-Z elements  are especially
important, the microscopist may want to work at a reduced beam
energy such as 5-10 keV to enhance sensitivity to the low-Z
elements while at the same time shifting the analysis towards
surface analysis and away from bulk analysis.

2.1.2.3 Choice of Spot Size or Probe Current
Probe current determines the X-ray count rate, which in turn
determines analysis time and sample throughput. Typical probe
currents vary greatly depending on the brightness of the gun.
More useful parameters are the counting times  (3-100 s) and
typical count rates  (1000-3000 s"1). These count rates can be
achieved with the PSEM (Section 2.2.2) using a spot size of
about 30% (corresponding to roughly 200 nm).  Increasing the
count rate significantly above a few thousand s"1 can result in
high detector dead time and degraded detector resolution, and
introduce artifacts  in the spectrum such as sum peaks. It  is
generally recommended to keep detector dead time below 30%.
The typical probe current for the LEO  S440  SEM (Section
2.2.1)  using a lanthanum  hexaborate  (LaB6) gun is 100 pA
yielding a count rate of approximately 1000 s"1, and corre-
sponding to a probe  diameter of about 10 nm. There  is a
tradeoff between image quality, analysis time, and spot size: the
best imaging is obtained with a small spot size. However, for
best backscatter imaging and X-ray acquisition, a large spot size
is needed. To perform automated analysis with reasonable speed,
imaging quality must be sacrificed in favor of less noisy BSE
images and higher EDX count rate: when no microimages are to
be collected, a very large spot size is acceptable. Microimages
of submicrometer particles, however may dictate that spot size,
and therefore analysis speed be reduced in order to improve
image quality. Note also that measurements  of particle size,
especially for submicrometer particles, is improved when spot
size is reduced.

2.1.2.4 Spectral Artifacts and X-ray Line Overlaps
Users must be on guard against misidentified peaks in the EDX
spectrum. Even with commercial EDX software, X-ray peaks
can be  misidentified due to spectrum artifacts (sum peaks and
escape peaks which are most prominently associated with high-
intensity peaks) and interelement interferences. In general, it is
very difficult to unravel two peaks separated by less than 50 e V.
Two types of interferences can be distinguished. For the first
series of transition metals, the K$ line of an element overlaps
with the Ka line of the next element in the series (K through Zn).
Second, the heavy metals have L- and M-family X-ray lines in
the 1-5 keV range that can interfere with the ^T-family lines of
Na to Ca. Some classic interferences relevant for PM research
include the difficult overlap between the Pb Ma (2.34 keV), S Ka
(2.31 keV), and Mo La (2.293  keV) lines. The lead La line
(10.549 keV) overlaps the As Ka line (10.532 keV). ZnZa (1.009
keV) interferes with  Na Ka  (1.041 keV); Ba La (4.467 keV)
interferes with Ti Ka (4.508 keV); and BrZa (1.480 keV) inter-
feres with Al Ka (1.487 keV).
     Different EDX systems may employ different methods for
handling X-ray line interferences. In most cases, spectral peak
overlaps canbe dealt with successfully by a combination of user
experience, peak deconvolution software provided with the EDX
system, and/or longer X-ray counting times. Improved counting
statistics  in the EDX spectrum can significantly improve  the
accuracy of peak deconvolution software, especially software
that fits the spectrum using stored libraries of X-ray lines  for
each element. In cases where the M- or Z-lines of a heavy metal
overlap with the ^T-lines of a lighter element (e.g., Pb Ma and S
Ka), the particle's brightness in the BSE image can often provide
definitive confirmation that  the heavier  element  is present.
Finally, off-line  data  review  software allows  the analyst to
review EDX spectra and to correct misidentified peaks. On rare
occasions, peak overlaps  in  the EDX  spectrum cannot be
resolved and the user must resort to WDX.
     Goldstein offers  specific "bookkeeping" suggestions to
facilitate  accurate qualitative EDX analysis: Identify peaks as
you work backwards from the high-energy end of the EDX
spectrum to the low-energy end; begin with the lines of highest
intensity  and  mark off all lines associated with an identified
element before proceeding, including sum and escape peaks;
low-energy L orMlines should be accompanied by K or L lines,
which should aid in identification; proceed with identification of
low-intensity lines associated with minor or trace elements.

-------
2.1.2.5 Guidelines for Qualitative EDX Analysis
The following is a brief summary of guidelines for qualitative
EDX  analysis.  For greater detail,  the reader is  referred to
Goldstein et al. (1992). For accurate peak identification, the
EDX spectrometer  should be calibrated periodically to ensure
that the spectral peak positions are within ± 10 eV of the
tabulated values. [Note: a piece of Cu attached to  any sample
holder provides a simple energy calibration check for two lines
(the Kp line at 8.907 keV and the La line at 0.928 keV) which
span most of the typical EDX spectral range]. A beam energy of
20 keV is generally considered a good compromise between
adequate overvoltage and minimumX-ray absorption. However,
at these energies there is still a potential for absorption of X-rays
below 2 keV  (P and lighter elements). X-ray acquisition time
should be  adequate  to  meet the  3o  criteria for statistical
significance, i.e., the minimum peak area P, after background
subtraction, should  be three times the standard deviation of the
background at the peak position.

2.1.3  Overview of CCS EM
Traditionally, SEM analyses have been conducted manually.
However, manual SEM (MSEM) is very time- and labor-inten-
sive and therefore limited in terms of particle throughput.
Furthermore, because microscopy is by nature a subjective tech-
nique, the results of MSEM are unavoidably operator-dependent
to some  degree. Advances in digital SEM technology have
spurred the development of CCSEM. Developed in the mid-
1970s and early 1980s (Lee et al.,  1979; Hanna et al.,  1980;
Kelly  et al., 1980; Lee and Kelly, 1980; Johnson et al.,  1981;
Casuccio et al., 1983a), CCSEM combines an SEM, an EDX
analyzer, and a digital scan generator under computer control.
CCSEM can  determine particle  size, shape parameters (e.g.,
aspect ratio, perimeter), and major elemental content of particles
larger than approximately 0.2 ,wm (depending on particle chem-
istry and other factors). Once the user has set up the sample for
analysis and initialized instrument parameters, the SEM auto-
matically locates features in the size range of interest; determines
the size, shape, and major elemental content  of individual
particles; acquires images of fields and individual particles; and
stores all this information electronically for later  review  and
interpretation.
    User-defined classification rules or statistical methods such
as cluster analysis, artificial  neural networks or multivariate
analysis can subsequently be employed to sort particles into
distinct particle classes based on size, shape, and chemistry. The
size and chemistry  for individual particles can be used (with
varying degrees of success—see Section4.3) to estimate particle
mass, which canbe summed, and scaled in order to estimate bulk
properties of the sample. However, the great strength of CCSEM
is the sheer quantity of information that can be obtained from the
analysis of hundreds or thousands of individual particles. With
current technology, the throughput of CCSEM can approach
1000 particles per hour on appropriately prepared samples, a
dramatic increase in throughput over conventional manual SEM
analysis. CCSEM is capable of collecting statistically meaning-
ful data sets from a large number of particles in a timely and
cost-effective manner. This breakthrough in particle throughput
opens up the possibility of using CCSEM in data-intensive
studies such as source apportionment studies.
     Numerous  applications of  CCSEM to  environmental
studies, aerosol characterization, and source apportionment are
reported in the literature. Interested readers are referred to papers
by Johnson and Twist (1982), Casuccio et al. (1983b, 1988),
Anderson et al. (1988, 1992), Kim and Hopke  (1988a,b),
Dzubay and Mamane (1989), Mamane (1990), Mamane et al.
(1995), Saucy et al. (1991), Vander Wood and Brown (1992),
Xhoffer et al. (1992), Katrinak et al. (1995), Johnson and Hunt
(1995), and Jambers and Van Grieken (1997).
     Despite substantial use of  CCSEM, the quality and accu-
racy of CCSEM data have received limited attention. Some data
quality studies have been conducted and are discussed in Section
4.6. Johnson et al. (1981) described a methodology to determine
a sample's bulk elemental composition from individual particle
data. Potential sources of error  associated with CCSEM were
discussed, including particle volume and density calculations,
assumptions about particle stoichiometry, and particle classifica-
tion. Watt (1990) discusses some of the difficulties and analyti-
cal pitfalls of CCSEM, including misidentification of particles
(especially organic particles or particles with low X-ray count
rate) arising from the use of "normalized"  X-ray data, and
analysis of aggregate particles.  Germani (1991) evaluated the
effects of critical instrumental parameters on  the time and
accuracy  of automated gunshot residue (GSR) analysis  (see
Section 4.6). GSR analysis is similar in many ways to CCSEM
analysis of aerosol samples. Mamane et al. (2001) examined a
number of issues affecting the quality and validity  of CCSEM
data: Stable  operation of the  SEM instrument during a multi-
hour CCSEM analysis of a sample is  obviously an essential
requirement. Sampling error is a concern for both manual SEM
and CCSEM since only a very small fraction (typically less than
0.1%) of the particles on an ambient filter is typically analyzed.
A sufficient number of particles must be analyzed to obtain a
representative sample (Section4.6.2). Finally, automated particle
recognition algorithms lack the sophistication of  the trained
human eye; errors can be made by CCSEM that are typically
avoided by an experienced operator performing manual analysis
(see Section 4.6.3). Such errors may include incorrect sizing of
complex particles; incorrect X-ray analysis, especially of aggre-
gate particles or organic and carbonaceous particles, which leads
to particle misclassification; missed particles (e.g., due to poor
contrast); and analysis of nonexistent features (contrast artifacts).

2.2 Description of  NERL  Facilities
The NERL SEM/EDX laboratory houses two digital automated
SEM/EDX  systems:  (1) the Personal SEM  (PSEM)  manu-
factured by Aspex Instruments (Delmont, PA, formerly R.J. Lee
Instruments, Ltd.), and (2) a LEO S440 SEM (LEO Electron
Microscopy, Inc., Thornwood, NY, formerly Leica) integrated
                                                         10

-------
with a Princeton Gamma-Tech (PGT) IMIX  EDX system
(Princeton Gamma-Tech Inc., Rocky Hill, NJ). In addition to
these two SEM systems, the laboratory is equipped with sup-
porting equipment. These facilities are described separately
below.

2.2.1  LEO/PGT System
The LEO/PGT system comprises a LEO S440 SEM integrated
with a PGT IMIX EDX system (Figure 2-7). The LEO S440 is
a digital SEM equipped with secondary and backscatter electron
detectors. The S440 operates with either a lanthanum hexaborate
(LaB6) electron gun or a tungsten  gun.  The IMIX system
employs a thin-window, large-area (50 mm2) EDX detector with
digital pulse processing. This enables X-ray detection of carbon
and heavier elements. The S440 is equipped with a computer-
controlled, eucentric sample  stage, and the IMIX EDX system
includes software for automated feature analysis (CCSEM). An
example  data output is shown in  Appendix  A. Both the
LEO/PGT and PSEM systems are capable of routinely charac-
terizing particles down to approximately 0.2  ,wm. The S440,
however, provides better imaging resolution (nominally 4.5 nm
at 30 kV) and is the instrument of choice for generating highest
quality photomicrographs.
Figure 2-7. The LEO S440 SEM with PGT IMIX EDX system.
2.2.2  Aspex Personal SEM System
Like the LEO S440 SEM, the Personal SEM (PSEM) (Figure
2-8) is  a  digital SEM/EDX system capable of automated
SEM/EDX analyses.  An  example data output is  shown  in
Appendix A. The two SEM systems are distinguished primarily
by imaging resolution, on-line and off-line software, and the
PSEM's variable pressure option.  The PSEM uses a tungsten
gun exclusively. The  X-ray detector employs a thin window
similar to the PGT detector.
Figure 2-8. The Personal SEM (PSEM) system.
     The PSEM is generally the instrument of choice for routine
work and for automated SEM/EDX  analyses when image
resolution is not critical. Much of the enhanced speed and ease
of use of the PSEM is due to the fact that it is a fully integrated
system with both SEM and EDX operations  controlled by a
single PC using one keyboard and one  mouse. In contrast, the
LEO/PGT system requires  two  computers  using  different
platforms (PC and UNIX), two keyboards, and two mice. The
serial-line communication between the  two computers signifi-
cantly slows CCSEM analyses compared to CCSEM analyses
performed on the PSEM.  Additionally,  the PSEM software for
automated feature analysis offers more  flexibility in setting up
CCSEM analyses, faster particle throughput, improved ease of
use, and better options for off-line data processing and analysis.
Overall, the PSEM is significantly easier for the novice to use
than the LEO/PGT system and for this reason has become a
multi-user facility.
     Particle relocation software (program XL ATE,  R.J. Lee
Group, Inc.) allows features of special interest in the PSEM to
be physically  relocated in the LEO S440.  This has  made it
possible to  combine the  superior CCSEM capabilities of the
PSEM and the superior imaging capabilities of the LEO S440.
The  PSEM, unlike the LEO S440, can be  operated in the
variable-pressure or low-vacuum mode in addition to the  con-
ventional high-vacuum mode. The variable-pressure capability
of the PSEM is provided by a microprocessor-controlled valve
and pumping configuration that allows user-selected pressure up
to 1 torr to be maintained  in the sample  chamber while keeping
the gun at high vacuum. SEM analysis of samples at elevated
pressure has the advantage that charging of insulating samples
is largely or completely eliminated, obviating the  need for
carbon-coating the samples, with no significant  loss of image
resolution. The advantages of reduced sample preparation time
are somewhat offset by the fact that only the backscatter mode
                                                       11

-------
of analysis can be employed in the variable-pressure mode.
Nevertheless, the variable-pressure  feature has proven to be
extremely beneficial for samples prone to charging and when the
user simply wants a rapid evaluation of a sample without the
need to first coat the sample.
     Figures 2-9a (PSEM) and 2-9b ( LEO S440) provide a
comparison of image  quality for the two instruments. Both
images are of the same feature at approximately the same magni-
fication. Instrument parameters for both SEMs were optimized
to produce the best possible image.

2.2.3  Vacuum Evaporators and Other
        Supporting Equipment
Samples that are to be examined by SEM in high vacuum (rather
than high pressure) typically require a thin conductive coating on
the order of 200 A to conduct charge  away from the sample and
prevent sample charging. The ideal coating would not contribute
X-rays to the X-ray spectrum of the coated particle. Although
the coatings  are very thin, conductive metal coatings  such as
chromium, gold, or palladium can generate non-negligible X-
ray peaks that interfere with particle-generated peaks in the EDX
spectrum. For this reason, carbon is generally the element of
choice for conductive coatings if the samples are to be analyzed
by EDX, since X-rays produced by the carbon film will inter-
fere significantly only with the carbon X-ray peak generated by
the sample. Unless one is analyzing a sample for carbon, the
carbon peak contributed by the coating is usually of no concern.
(Section 4.5 discusses the problems associated with the analysis
of carbonaceous particles).  If high-resolution imaging is of
primary importance, a thin coating of gold or Au/Pd evaporated
from  a tungsten basket can be applied to the sample.  This
enhances image contrast compared to carbon coating, but also
introduces unwanted X-ray lines in the EDX spectrum.
     Supporting facilities in the SEM/EDX laboratory currently
include two vacuum evaporators for coating SEM samples  with
thin carbon and/or metal conductive films. Both evaporator units
employ tilting, rotating specimen holders  to ensure uniform
coatings and to minimize shadowing, and both are capable of
applying carbon or metal coatings. Plans currently  include
upgrading to more  modern coating devices with dedicated
carbon and metal coating capabilities.
        Figure  2-9a.  Salt  crystals. Image acquired with
        PSEM at 20 kV, 7-mm working distance, 10% spot
        size, and tungsten filament.
Figure 2-9b. Same salt crystals imaged with LEO 440 SEM at 30 kV,
7-mm working distance, 3-pA probe current and LaB6 emitter.
                                                        12

-------
                                                Chapter 3
      Recommended Techniques and  Procedures  for SEM/EDX and  CCSEM
3.1  Sample Receipt, Tracking, and Storage
The following describes the procedures used by the NERL SEM/
EDX Laboratory to manage the flow of samples through the
laboratory. These procedures can be easily adapted to the needs
of other laboratories.

3.1.1   Sample Submittal Form
It is highly recommended that individuals interested in having
analysis done in the NERL SEM/EDX laboratory discuss their
needs with laboratory personnel as early into their project as
possible.  The submitting of actual samples to the laboratory is
formalized and tracked through the use of a Sample Submittal
Form, a copy of which is included in Appendix B. An electronic
version of  the form is made available to the requesting
researcher/client, who submits the completed form back to the
SEM laboratory contact. The client provides a project name and
contact person, a description of the samples, the analytical objec-
tives and expected outputs, and a target delivery date for results.
Duplicates of the submitted forms are kept in a binder in the
SEM laboratory and in the hard-copy file folder created for the
associated study. The Sample Submittal form initiates the sample
tracking process discussed below, and provides  a basis for
follow-up discussions.

3.1.2   Log-in Procedures
Samples received for analysis are assigned a project name and
a SEMID, a four-digit identification number used by the SEM
laboratory to track samples. ID numbers are assigned by the
SEM laboratory staff after consulting the "Assignment of SEM
IDs" notebook in the SEM laboratory. This notebook is a record
of all project names and the  ID numbers  assigned to all
substudies. An example page from this notebook is presented in
Appendix B. Analogous to the assignment of IDs in the NERL
XRF laboratory, the first three digits of the SEMID represent the
project name and the fourth digit is reserved for substudies
within the same project (maximum of 10 possible substudies for
each project). If the samples represent a  new project, a new
three-digit code followed by a zero (first substudy of the project)
is assigned to the samples.
     After assigning a SEMID, the staff enters the requested
information into the Sample Log-In/Log-Out notebook (Section
3.1.3) maintained in the laboratory. A file folder is created for
each SEMID and kept in a file cabinet in the SEM laboratory.
This folder contains all information relevant to that substudy,
including a copy of the Sample Submittal form, communications
between the client and the SEM laboratory, and hard copies of
data and outputs from the SEM analysis such  as photomicro-
graphs and reports. If the submitted samples were previously
analyzed in the NERL XRF facility, that should be noted on the
sample submittal form, and hard copies of the XRF data are in-
cluded in the file folder. The XRF data can be very useful in
guiding the SEM/EDX analysis of the same sample.

3.1.3  Sample Tracking
Samples are tracked by means of Sample Log-In/Log-Out note-
book maintained in the  SEM laboratory. The Sample Log-In/
Log-Out notebook is a record  of samples logged into  the
laboratory for analysis and logged out of the laboratory for any
purpose. All samples that are removed from the SEM laboratory
must be logged out, including those whose return is requested on
the SEM Sample Submittal form. "Sample" in the above discus-
sion refers both to the samples as submitted (filters, raw dusts or
powders, etc.) and to SEM sample stubs prepared from the as-
submitted samples. However, SEM  stubs are removed from the
laboratory only under exceptional circumstances since they are
useful only to people with access to a SEM. Information record-
ed in the Sample Log-In/Log-Out  notebook includes the fol-
lowing:  date the samples were received,  project name and
SEMID, number of samples, name of a contact person or client,
and number of samples returned and to whom returned.  An
example Sample  Log-In/Log-Out page   can be  found  in
Appendix B.

3.1.4  Sample Storage
Samples include as-submitted samples as well as the SEM stubs
prepared from the submitted samples. As-submitted samples may
include air filters, loose samples such as soil or dust samples, or
liquid suspensions. At the present time these samples are stored
in cabinets in the SEM laboratory unless the client requests their
                                                      13

-------
return on the Sample Submittal form. This policy may be revised
as storage space becomes limited. Half-inch SEM  stubs are
typically stored in plastic stub boxes with the project SEMID
and individual stub IDs (Section 3.2.3) marked on the box lid.
Recent SEM stubs or samples for active projects are typically
stored at room temperature in low-humidity Lucite sample stor-
age boxes in the SEM laboratory. Alternatively, sample stubs are
stored  in vertical filing cabinets labeled by project name.
Periodically, old stubs that have been determined to be non-
essential are destroyed and the aluminum stubs are cleaned and
reused.

3.2  Sample Preparation
Proper sample preparation is a prerequisite for successful SEM
analysis. The nature of the sample preparation will be governed
by the nature of the  sample and the analytical objectives.  Suit-
able  samples include most solids that are stable under vacuum
and exposure to an energetic electron beam (metals, ceramics,
polymers, minerals). Microscopists should be constantly aware
of the potential for loss of volatile particles in vacuum during
carbon coating and/or SEM analysis. For example, Parungo et al.
(1986) reported the  loss of sulfate particles due to sublimation
during EDX analysis.
     The range of motion of the  SEM stages limits the physical
size of samples to about 50 mm. Typically, only  a few square
mm of an air filter sample are sufficient. Sample preparation for
PM samples is relatively simple since we typically deal with dry,
non-biological specimens. Usually PM samples are analyzed as
is, with no additional preparation other than mounting on a stub
and overcoating with a thin carbon film to provide conductivity.
Quantitative X-ray microanalysis, which generally requires flat,
polished specimens  is rarely required in the analysis of aerosol
samples.

3.2.1 Choice of Filter Media
Several types of filters are commonly used in sampling aerosols,
but not all are compatible with SEM/EDX analysis. The  ideal
filter medium for SEM/EDX would have the following proper-
ties:  (1) a screen membrane (in contrast to a depth membrane),
providing uniform contrast, in which the  aerosol particles are
deposited on top of an optically  smooth, flat surface; (2) trans-
parent to 20-ke V electrons so that it contributes minimally to the
EDX spectrum of the sample; (3) thermally stable (insensitive to
localized heating from the electron beam)  and compatible with
high vacuum; (4) electrically conductive  to minimize sample
charging; (5) high purity/low blank to facilitate EDX analysis;
and (6) mechanically rugged.
     Figure 3-1 shows micrographs of four of the most com-
monly used filter types. Teflon, quartz, and MCE  (a mixture of
inert cellulose nitrate and cellulose acetate) are "depth" filters,
which trap particles throughout the depth of the filter rather than
on the surface. The three-dimensional fibrous structure of these
filters andtheirchemicalcompositiongreatly complicate the task
of identifying particles embedded in the filter. Depth filters have
greater sampling capacity than screen membrane filters but are
clearly less than ideal for microscopy purposes. In addition, the
EDX spectra generated by Teflon and quartz filters may interfere
significantly with the X-rays from the particles. (Teflon pro-
duces high F and C peaks, while quartz produces high Si and O
peaks.) Whenever possible, aerosol samples should be collected
on  polycarbonate  screen  membranes  (Poretics  Corporation,
Livermore, CA). Polycarbonate membranes, also called Nucle-
pore, are mechanically rugged and are an excellent substrate for
automated particle analysis. Polycarbonate filters are available
in diameters ranging from 13 mm to 47 mm and greater and with
pore sizes ranging from 0.01 ,wm up to 20 ,wm. The smooth two-
dimensional surface upon which the particles are collected has
a very low EDX blank, being composed of light elements (C, H,
and O). This composition provides excellent BSE image contrast
with particles of higher Z. A major drawback of polycarbonate
membranes  is  that they provide poor imaging contrast for
carbonaceous  particles, making  CCSEM analyses of small
carbonaceous particles especially difficult. Furthermore, these
filters are  nominally 10 ,wm thick and therefore not transparent
to 20-keV electrons. Thus, the filter generates carbon and oxy-
gen X-rays as well as bremsstrahlung radiation, which increases
background levels  and degrades signal to noise for EDX
analyses, especially of biological or carbonaceous particles.
     In many cases, the microscopist is presented with samples
that have  already been collected on non-polycarbonate filters,
most often Teflon or  quartz membranes.  For certain types of
analyses, it may be possible with considerable care and effort to
analyze aerosol samples on quartz or Teflon filters by automated
SEM/EDX (although the huge Si signal generated by the quartz
substrate will make it difficult, if not impossible, to determine
the chemistry of particles collected on quartz filters). For exam-
ple, automated searches for heavy metals can be carried out
Figure 3-1. Four filter substrates often used for aerosol sampling. Teflon
(upper left), quartz (lower left), and MCE (upper right) are depth filters,
while the polycarbonate (lower right) is a screen  membrane.  The
polycarbonate filter is ideal for SEM/EDX because of its low blank for
Z>12, and its optically flat, low-contrast surface.
                                                          14

-------
successfully,  since the "noisy" low-Z filter substrate can be
largely eliminated by use of a high-Z video threshold. For more
general applications involving particle sizing or classification of
low-Z as well as high-Z particles, one is usually faced with the
choice of either manual  analysis or transferring the aerosol
sample  onto  a polycarbonate  filter. Some laboratories have
reported success in redepositing particles from Teflon or quartz
onto polycarbonate filters (Casuccio, 2002). The redeposition
process involves  sonication of the original filter in filtered
acetone (depending on the sample, hexane or distilled water may
also be  used) to liberate the aerosol particles followed by re-
deposition of the particles onto a polycarbonate substrate using
vacuum filtration  apparatus. The major concerns with particle
transfer are the possibility that the redeposited particles have
been chemically or morphologically altered during the sonica-
tion and filtration steps, or that the redeposited sample is not
representative of the original sample. To ensure quality, it is a
good idea to mount a section of the original filter side by side
with a section of the redeposited filter and to manually compare
the two samples looking  for differences. Obviously, particles
that are soluble in the carrier liquid (e.g., salt crystals in distilled
water) will be lost during the sonication process.
     In addition to sample preparation procedures, some online
SEM analysis software has been developed which can provide
some help in the analysis of fibrous filters. On-line software for
the PSEM, for example,  includes a semi-automated mode of
analysis (I-P66 program, R.J. Lee Group, Inc., Monroeville, PA)
developed especially for Teflon or quartz filters in which the
features to be analyzed in each field are  selected by the user, but
the analysis otherwise proceeds automatically (Section 3.3.3).

3.2.2 Particulate Loadings
For aerosols in the submicrometer to 20-^m range, ideal mass
loadings for  SEM/EDX analysis are in the range of 5 to 20
^ig/crn2, though up to  30 ^g/cm2 may be acceptable in some
cases. As  loadings increase above 30  ^g/cm2,  more particles
come in contact with each other and it becomes increasingly
difficult to analyze individual particles by CCSEM. The average
distance between neighboring particles should be at least four
times the average diameter of the particles being analyzed. If the
loading is too light, the time required to  analyze a representative
number of particles may become impractical because of the time
required to move  the sample stage over many fields. Particle
loading is particularly critical for CCSEM  analyses for which
automated feature analysis software lacks the ability of the
human eye to resolve overlapping particles. Uniform distribution
of the particles is desirable for both manual and automated SEM
and is a prerequisite for estimating bulk properties of a sample
(e.g., the number or mass of particles per unit area or the overall
mass loading) based on a partial sampling of fields.

3.2.3 Mounting Filter Samples
Filter specimens are typically mounted on aluminum or carbon
sample stubs. Stubs are available in 1/2-in and  1-in diameters
with 1/8-in-diameter posts compatible with the PSEM and LEO
stub holders. In mounting samples, it  is critical that there be
good electrical conductivity between the sample and the spec-
imen stub in order to avoid charge build-up.  Filter sections
should lie flat on the stub with maximum surface area in contact
with the stub (or underlying carbon tab if a tab is employed).
Powder-free  gloves  are recommended whenever handling
anything that goes into the SEM in order to maintain the
cleanliness of the SEM column. (Oils  from the skin can ulti-
mately contaminate the SEM column due to volatilization.) The
following procedure is  recommended for  preparing sample
stubs:

     1.  Label the stub  on its underside with a six-digit ID
        number (the "Stub ID") using a permanent marking
        pen. The first four digits of the stub ID are the SEMID
        assigned to  the project. The  last two  digits (0-99)
        identify the individual sample mounted on the stub.
     2.  Label a plastic stub box (holding up to twelve  1/2-in
        stubs) with the project name and the date.  Label each
        stub position in the box with the  six-digit stub ID
        numbers.
     3.  Prepare the stub by  applying  conductive adhesive
        (silver paint, dag, or carbon tabs) for an approximately
        10-mm2 filter section.  Carbon  tabs are disks of carbon
        with adhesive on both sides. They are available  in the
        laboratory in 1/2-in and 1-in diameters. They provide
        good electrical conductivity between the filter and the
        stub. They are also excellent substrates for viewing
        particles by SEM because they provide a flat, artifact-
        free background; they withstand electron beam heating
        moderately well;  are  stable in vacuum; and have  a
        clean (carbon only) EDX spectrum. Carbon dag is  a
        dispersion of graphite  or carbon particles in a volatile
        carrier such as isopropyl alcohol. It provides excellent
        electrical conductivity and adequate adhesion between
        the stub and the mounted filter section. Apply  a thin
        layer of dag to the stub by brush. Wait until the dag has
        dried slightly but is still moist. For typically loaded
        aerosol samples, a 10-mm2  or larger section of filter
        should give adequate statistics for most purposes, but
        selecting a random subset of fields from a larger area
        may provide a more representative analysis  if the
        sample loading is somewhat inhomogeneous.
     4.  Lay the filter section onto the carbon tabs  or the wet
        dag. In our experience, the dag method seems to be
        most effective in minimizing charging problems.
     5.  Refer to Sections  3.2.3.1 and 3.2.3.2  for mounting
        polycarbonate and Teflon filters and Section 3.2.5 for
        coating the mounted samples.
     6.  Place the  mounted samples in plastic stub boxes to
        prevent contamination. From this point on, use special
        forceps designed for gripping the SEM stubs to handle
        the stubs.
                                                          15

-------
3.2.3.1 Mounting Polycarbonate Filters
Examine the filter to verify which side of the filter is the loaded
side. (The shiny side should be the loaded side, but occasionally
the dull side has been mistakenly loaded.)
     1. Place the filter on a clean, flat, glass surface.
     2. Using a clean stainless steel scalpel or razor blade, cut
        a square section (approximately 5 mm by 5 mm) from
        the interior of the filter.
     3. As an alternative to step 2, cut the section using clean
        scissors.
     4. Affix the section of filterto the stub with dag or carbon
        tabs.

3.2.3.2 Mounting Teflon Filters
     1. Verify that the loaded side of the filter is face up.
     2. Place the area of the filter to be  removed on top of a
        stub covered with a sticky carbon tab.
     3. With a clean scalpel or razor blade, cut a 5-mm by
        5-mm section of filter.
     4. Gently lift the filter from the stub leaving the desired
        section attached to the stub.
Figure 3-2. Vacuum filtration apparatus used for wet preparation of bulk
dusts. A fritted glass  base supporting a 37- or 47-mm filter extends into
the vacuum flask through the stopper. The flask is evacuated through its
side arm. The sample (in a liquid suspension) can be introduced to the
filter either through a funnel clamped on top of the filter, or by using a
glass nebulizer (foreground) to spray the exposed filter.
3.2.4  Preparing Bulk Dusts for SEM
In addition to characterizing filter samples, it is frequently
necessary to analyze bulk samples, such as dusts, soils, or pow-
ders. Both wet and dry methods have been developed in the
laboratory for the preparation of bulk particle samples.

3.2.4.1 Procedure for Wet Preparation of Bulk Dusts
We have used several options for wet preparation in the NERL
SEM laboratory. Particles in a liquid suspension can be depos-
ited onto a filter using glass microanalysis vacuum filtration
apparatus (Fisher Scientific, Pittsburgh, PA. Vacuum filtration
apparatus is available for 25-,  47-, and 90-mm filters).  This
method is  easy  to carry out and generally yields a uniform
particle distribution on the filter. In this approach, an aliquot of
bulk dust is suspended in reagent-grade acetone. A blank, tared
polycarbonate filter is clamped between a fritted glass support
base and a glass funnel. The base is coupled through a rubber
stopper to  a vacuum flask  whose side arm is connected to a
source of vacuum. (Figure 3-2 shows the vacuum  flask closed
with a rubber stopper attached to the fritted glass base. The glass
funnel is not shown.) The  liquid suspension is sonicated for
approximately 1  min and an aliquot is placed in the glass funnel.
Gentle vacuum is applied to the flask causing the liquid suspen-
sion to be sucked through the polycarbonate filter. As the liquid
level in the funnel drops, rinse the sides of the funnel with clean
acetone using a spray bottle. When the filter appears completely
dry, remove the clamp and the funnel while continuing to evac-
uate the flask. Then, shut off the vacuum and remove the filter
with clean tweezers.
     Alternatively, suspended particles can be sprayed onto a
clean stub  or filter using a glass nebulizer or a miniature air
brush.  (We have used the Model 250 Air Brush, Badger Air-
Brush Co., Franklin Park, IL.) The following details the air
brush and wet nebulizer methods:

     1. Suspend approximately 10-20 mg of sample (ground if
        preparing for XRF, or if altering particle size is not
        important.) in a 4-mL glass vial containing 2 mL  of
        reagent-grade 2-propanol or acetone.
     2. Sonicate for approximately 1 min to break up agglom-
        erates.
     3. Aerosolization is accomplished in a vented hood using
        a miniature air brush. Connect the air brush to 20 psi of
        dry nitrogen. Alternatively, put the  suspension  in a
        glass nebulizer attached to the hood nitrogen fitting.
     4. Mount a blank polycarbonate filter in an open-faced
        47-mm aluminum filter holder (Model 1220, Gelman
        Sciences Inc., Ann Arbor, MI) attached to a 500-mL
        Pyrex (Corning Labware and Equipment, Big Flats,
        NY) vacuum filtration flask.
     5. Insert air brush siphon fully into the sample vial.
     6. Apply vacuum to the backside of the filter assembly to
        accelerate the evaporation of the propanol or acetone
        carrier.
     7. Raster the air brush evenly over an area approximately
        10 times larger than the filter to improve deposit uni-
        formity. (A metal splashboard surrounding the filter
        catches the excess spray.)
     8. Turn off vacuum, unmount loaded polycarbonate filter,
        and mount section on SEM stub as described in Section
        3.2.3.

Note: The use of a wet technique for sample preparation raises
concerns that the 2-propanol or acetone carrier may alter particle
                                                          16

-------
chemistry, size, or morphology. Many salts and organic com-
pounds are soluble or slightly soluble in propanol, which leads
to a potential for biasing the results for such species, although
we have not investigated these effects to date. Likewise, we are
not aware of studies comparing size distributions for raw versus
wet-prepped aliquots of the same sample.

3.2.4.2 Procedure for Dry Preparation of Bulk Dusts
A dry sample preparation procedure was developed for creating
a uniform deposit of powders or dusts on Teflon or polycarb-
onate filters. In contrast to other preparation techniques for dry
powders,  this technique employs a miniature cyclone device
(John and Reischl, 1980), which, by adjusting the airflow rate,
allows one to select the size range of particles collected on the
filter. Figure 3-3  is a photo of the mini-cyclone device. The
motivation for developing the technique was to be able to pre-
pare powder samples suitable for quantitative XRF analysis,
which requires particles smaller than 10 ,wm, aed. But  the
method is also useful for preparing samples for SEM/EDX The
procedure is simple and easy to perform. Test samples examined
by SEM show very uniform particle distributions. The procedure
is outlined below.

     Materials needed:
     Procedure:

     1.  Label petri dish with sample ID number and date.
     2.  Equilibrate untared, clean filters in petri dishes for 24 h
        in climate-controlled weighing room.
     3.  Record filter tare weights and return filters to petri
        dishes.
     4.  Place filter in holder with backing filter.
     5.  Attach vacuum line from pump to filter holder.
     6.  Set flow rate on vacuum pump for desired particle size
        according to cyclone flowchart.
     7.  Use a spatula to feed sample into inlet tube.
     8.  Run vacuum for 1-2 minutes; turn off vacuum pump.
     9.  Remove backing filter and place loaded filter in labeled
        petri dish.
     10. Weigh loaded filter after 24-h equilibration in weighing
        room.
     11. Repeat procedure if necessary until desired loading is
        attained. (Note: For XRF analysis, desired loadings are
        approximately 0.5-1 mg for 37-mmfilters and 1-2 mg
        for 47-mm filters, which are up to 20 times the desired
        loadings for SEM.)
     12. Clean cyclone after each loading to prevent contami-
        nation.
        37- or 47-mm Teflo (2-^m pore size) or polycarbonate
        filters
        Quartz filter for backing
        Cyclone and cyclone flowchart
        Vacuum pump with flow gauge (1pm)
        Teflon  tweezers, spatula, plastic 47-mm petri dish,
        filter holder
The  NERL SEM Laboratory also prepares dry  dusts  and
powders using a custom-designed glass resuspension apparatus
(the  "puffer")  shown in Figure 3-4. The puffer  is a glass
chamber with a volume of 900 mL. The lid to the chamber has
a built-in mesh grid for supporting a 47-mm filter.  Air can be
pulled from the chamber through the filter by means of a
Figure 3-3. "Mini-cyclone" apparatus used to prepare filter samples of
bulk dusts and powders. Raw sample is sucked into the cyclone through
the horizontal section of tubing. The filter to be loaded is placed in the
filter holder on the top of the cyclone. A vacuum pump is shown in the
back.
Figure 3-4.  "Puffer" apparatus used to prepare filter samples of bulk
dusts and powders. The side arm below the shut-off valve is connected
to a vacuum pump.
                                                          17

-------
vacuum connection on the lid. The lid also has four air (or gas)
jets connected to a central inlet with a shut-off valve. The jets
extend via tubes into the resuspension chamber and are arranged
symmetrically about  the base of the chamber. Clean, filtered
air or inert gas can be introduced into the chamber via the four
jets. To operate the puffer, approximately 0.1 mg of sample is
placed in the bottom of the clean chamber. A blank, tared 37- or
47-mm filter is placed on the filter support grid and secured in
place by a Delrin cap. Vacuum is applied to the backside of the
filter  and the lid is then closed over the chamber so  that the
chamber is evacuated. The air inlet valve is then opened allow-
ing a burst of clean air or inert gas into the chamber via the four
jets. The burst of air resuspends the sample material, a fraction
of which is then pulled onto the vacuum-backed collection filter.
SEM examination of filters prepared with the puffer show very
uniform loadings. During the preparation of a sample with the
puffer, most of the resuspended sample dust redeposits on the
internal surfaces of the chamber; to avoid cross-contamination
of samples, the puffer must be thoroughly cleaned between uses.
It is possible that filter samples prepared in the puffer show
some size fractionation relative to the original sample material,
but size fractionation  effects  in the  puffer have not been
investigated.

3.2.5  Coating Mounted Samples
The mounted samples are coated with a thin conductive film
prior to SEM/EDX analysis to minimize sample charging prob-
lems. General guidance for coating samples is given below.
      1.  Use a vacuum evaporator device to overcoat specimens
        with carbon, or a vacuum evaporator or sputter coater
        to overcoat specimens with metal (typically gold). It is
        helpful to have a one-page instruction sheet for using
        each coating device kept with each unit.
     2.  Coat  specimens  to be  analyzed  by  EDX only with
        carbon, since metal coatings will complicate the X-ray
        spectrum of the particle. Metal coatings are often used
        to enhance image resolution on specimens that are not
        to be analyzed by EDX.
     3.  Produce carbon  films by evaporating 1/8-in  carbon
        rods. Thicknesses on the order of 200 A usually pro-
        vide adequate charge conductivity. If charging is still a
        problem, an additional C-coat may be required.
     4.  Metal films (typically Au, Pt, or Au/Pd)  are produced
        by  evaporating  a  length  of pure metal wire in a
        resistance-heated tungsten wire basket (vacuum evapo-
        rator) or sputtering metal targets  (sputter coater). A
        typical gold coating with the vacuum evaporator uses
        two  inches of 0.008-in-diameter wire placed in the
        tungsten basket.
     5.  Ensure that samples are not subject to the full power of
        the coating process for an extended period. The process
        of coating the sample subjects the sample to consider-
        able heat, which can alter the sample chemically and
        physically. Maximizing the distance between the evap-
        oration source and the sample can reduce the potential
        for thermal damage.
     6.  A rotary/planetary/tilting stage helps to ensure a uni-
        form coating and eliminate shadowing effects.
     7.  Protect  your  eyes.  Metal and carbon evaporations
        produce  extremely bright and potentially  damaging
        light visible through the glass bell jar or sample cham-
        ber. The user must wear eye protection (welding glas-
        ses)  or view the process through the  optically dense
        glass plates attached to the  outside of the bell jar or
        chamber.
Some coating devices are equipped with a thickness controller/
monitor. If the coater is not equipped with a thickness monitor,
a convenient technique to estimate the thickness of the carbon
coating is to place a polished brass stub in the coater along with
the samples to be coated. Thin-film interference  causes  the
polished brass to take on different colors, depending on  the
thickness of the carbon coating, according to the following color
scale:
 Color
                                 Carbon Thickness
                                   (Angstroms)
 Orange
 Indigo red
 Blue
 Bluish green
 Greenish blue
 Pale green
 Silver gold
150
200
250
300
350
400
450
The carbon coating is easily removed from the brass stubs by
using metal polish applied to a cotton swab.
     Gold by itself has relatively poor wetting properties. In
order to improve the quality of the gold thin film, a thin film of
carbon may first be evaporated onto the  sample. Or gold and
carbon may be evaporated  simultaneously by wrapping gold
wire around a sharpened carbon rod and doing a normal carbon
coat. A popular alternative to Au is Au/Pd which has improved
wetting properties.

3.3  SEM/EDX  Data Acquisition
SEM/EDX analyses can be conducted in fully manual mode,
fully automated mode, or interactive (semi-automated) mode.
For each mode of analysis, the  analytical objective may be
particle size distribution only, size and chemistry, images only,
or a targeted particle searches.  These modes are  discussed
below. These discussions are specific to the Aspex Instruments
(Denton, PA) Personal SEM (PSEM) and/or the LEO (Thorn-
wood, NY) SEM and PGT (Princeton, NJ) IMIX EDX system.
Users of other SEM/EDX instruments may need to adapt these
guidelines to their own instruments.
                                                         18

-------
3.3.1  Manual
Despite the impressive throughput of automated SEM, there will
always be a need for manual SEM (MSEM). Many samples are
unsuitable for CCSEM because of particle loading or inappro-
priate substrate, and therefore require MSEM. Moreover, MSEM
examination is strongly  recommended for all samples before
proceeding with either a more detailed manual analysis or a
CCSEM analysis. Preliminary manual examination of a sample
familiarizes the microscopist with the sample and provides vital
quality control. A preliminary examination can reveal  if the
sample is charging and needs another carbon coating or if the
sample has been seriously compromised (e.g., tears or holes in
filter-based samples, foreign objects or other evidence of con-
tamination, gross non-uniformity in particle loading, filter sam-
ple mounted wrong-side up). The sample should be scanned at
low  magnification to identify any obvious  non-uniform  or
irregular  particle deposits, sample  artifacts,  or areas of the
sample that should be avoided.
     Because of the low particle throughput of manual analysis,
sampling error is always a major concern. If the objective of the
manual analysis is to provide an accurate representation of the
sample, then the microscopist must examine enough fields and
enough particles to ensure representative sampling. Studies sug-
gest that this requires a minimum of several hundred particles
(Mamane et al. 2001). Because the eye is naturally attracted to
larger features or those  with distinctive morphology,  fields
should be selected at random (with eyes closed), and all particles
within each field should be analyzed to minimize the potential
for bias or operator subjectivity.  Although manual analysis is
very tedious, there is an opportunity in MSEM to bring one's
intelligence and experience to bear on the sample that is not so
direct in CCSEM. The microscopist can interact in real time with
the sample, making observations and drawing inferences about
the sample that cannot be done in the automated mode.
     Observations from a MSEM analysis session are typically
recorded in a lab notebook, accompanied by hardcopies of photo-
micrographs of fields  of particles as well as  micrographs  of
selected particles of interest. A common approach is to work from
a hardcopy photomicrograph of a field of particles which have
been numbered by the analyst. The analyst successively analyzes
each particle within the field, recording relevant data (e.g., date,
sample ID, field number, particle number, particle size, morph-
ology, chemistry, and particle class) in the notebook. Photomicro-
graphs of fields and individual particles are saved to disk.
     Although MSEM is indispensable for many samples, espe-
cially novel samples, it  is so labor intensive  that it generally
cannot provide the statistical depth needed for data-intensive
studies such as source apportionment where differences among
samples must be quantified.


3.3.2  CCSEM
Samples that are suitably prepared (proper loading, compatible
filter substrate, uniform particle distribution) may be candidates
for fully automated, computer-controlled SEM (CCSEM). The
PSEM uses vendor-supplied software (ZepRun, Aspex Instru-
ments, Delmont, PA) to perform a computer-controlled analysis
of the sample once the initial instrument conditions are set up.
These  parameters  include  the physical  boundaries for  the
analysis, video threshold, magnifications) at which the analysis
will  be conducted, number of particles  and/or fields to be
characterized,  method of field selection (random, manual, in
order), particle size range, aspect ratio range, X-ray acquisition
time, list of elements to be quantified,  and numerous other
parameters that affect data acquisition. The user may also create
customized rules to specify how often and for what types of
particles images should be acquired or to exclude particles that
are of no interest. Once the run-time parameters are initialized,
the CCSEM analysis can proceed without operator intervention.
Data acquired include particle size and shape parameters, parti-
cle location, particle and field images (if requested), video level,
total X-ray counts,  and element concentrations. The data are
stored in electronic format for later evaluation. Two software
packages (ZepView and ZepSum, Aspex Instruments, Delmont,
PA)  enable the user to review and summarize data off-line.
     During CCSEM analyses, the PSEM relies on a user-
selected grayscale video threshold (the detection threshold) to
discriminate between a particle and filter background. A care-
fully chosen threshold is essential for accurately identifying and
sizing particles. However, a complex aerosol sample may pre-
sent  a wide range of particle compositions, sizes, and morph-
ologies, which will complicate the strategy for threshold setting.
     CCSEM analyses are generally conducted using the BSE
mode rather than the SE mode. The BSE signal is better for
detecting particles on a filter substrate because of its higher
atomic number contrast, its relative lack of topographic detail,
and its lower susceptibility to electron beam charging artifacts,
which can occur even with carbon-coated samples. (The edges
of the holes in Nuclepore filters routinely exhibit charging
effects in the SE mode.) The BSE mode thus yields a much more
uniform and stable video  signal from the filter background,
enabling the user to precisely set the detection threshold. Particle
brightness in the BSE mode is governed by the effective atomic
number (Z) of the particle. Particles with high Z (e.g., Fe-rich)
have a highbackscattered electron yield and appear brighter on
the SEM image, while particles with low Z (carbon, sulfates,
organics, biological) will appear dim. Proper setting of the video
threshold is extremely important in CCSEM analyses. Ideally,
the video  threshold is set  such that all  particles having an
effective Z greater  than or equal to a threshold value Zt are
detected while all particles with Z
-------
pares the backscattered electron signal collected at each point in
a grid to the preset detection threshold. Once a signal above the
threshold is detected, the measurement mode is enabled. In this
mode, the distance between grid points is much smaller than in
the detection mode, and the preset measurement threshold has a
lower grayscale value than the detection threshold. The compu-
ter draws 16 chords through the particle's center of mass within
the confines of the measurement threshold in order to determine
the particle's maximum,  minimum, and average  diameters,
aspect ratio, and area. (An alternative approach to sizing, which
may yield somewhat better estimates of particle volume for large
particles, is to measure the particle perimeter. This "perimeter
walk" option is available in ZepRun's I-P66 mode of analysis,
Section 3.3.3).
     Very abbreviated guidelines  for CCSEM are presented
below. Users requiring more detailed guidelines are referred to
the PSEM ZepRun Users Manual (Aspex Instruments).

       How to Obtain Good Results from CCSEM

A.   Prepare a good sample
     1.  Label stub with 6-digit ID
     2.  Add fiducial marks for particle relocation
     3.  Ensure representative sample
     4.  Ensure uniform loading
     5.  Minimize particle overlap
     6.  Avoid contamination
     7.  Eliminate sample charging

B.   Setup SEM
     1.  Saturate filament
     2.  Align beam
     3.  Set working distance: 15-20 mmforbestEDX, 7-8 mm
        for best images
     4.  Set spot size: 30%-40% for EDX; for best imaging,
        use smallest spot size which still gives adequate bright-
        ness and contrast

C.   Setup stage
     1.  Degauss lens until focus is stable
     2.  Set up analysis area and focus for each sample
     3.  Set up origin and alignment reference points (for future
        particle relocation)

D.   Preview the sample manually
     1.  Check particle loading and deposit uniformity
     2.  Scan sample at low mag to identify areas to avoid
     3.  Particle size range determines choice of mags
     4.  Particle types determine choice of elements and on-line
        analysis rules
     5.  Optimize EDX countrate  and detector dead time on
        high Z particles and low Z particles by adjusting spot
        size.
E.   Setup CCSEM run parameters
     1.  Set up Run Parameter File (RPF):
        a.   Suggested mags:
            10- 100 ,wm: 200x    0.2-5 ,wm: I200x
            5-50 ij,m: 400x      0.2-2.5 ij.m: 2000x
            2.5-10 ,wm: 800x
        b.   Specify SEM run parameters:
            Maximum particles per mag; max fields per mag;
            max time per mag; particle diameter and aspect
            ratio limits (minimum and maximum) for search;
            search grid spacing; search size, search, measure,
            and image dwell times; guard band; microimage
            fill%; field image option; field selection method
            (random, in order, manual).
        c.   Specify EDX parameters:
            Minimum and target counts; normal and maximum
            EDX time; element threshold; X-ray mode (point,
            chord, raster, perimeter).
     2.  Set up Element Vector file (VEC)
        a.   Put C and O in the element list only if you want to
            analyze carbonaceous particles.
        b.   Be aware of possible peak overlaps:
            PbMa-SKa    AsLa-MgKa  BrLa-AlKa
            Zn L/3 - Na Ka   As Ka - Pb La  Ti Ka - Ba La
        c.   Require 3-5 sigma counts in secondary or tertiary
            ROIs (regions of interest or X-ray "energy win-
            dows") for potential overlaps: Pb, As, Br, Ba.
     3.  Set up Analysis Rule file (RUL)
        a.   Create rules based on particle size, shape, and/or
            chemistry for screening on-line data
        b.   Specify MAX EDX to increase the EDX time for
            elements of special importance: e.g., Pb, Br, Zn,
            As
        c.   For each particle  type defined by the rule file,
            specify whether to save microimages and/or spec-
            tra, and how frequently (e.g., save for all particles
            or save for every 10th, etc.)

F.   Final Checks
     1.  Ensure adequate disk space for data
     2.  Set up auto-threshold feature as desired
     3.  Move stage to first sample and check that focus is OK
     4.  Carefully set detection and measurement thresholds
     5.  Make a thermal print to record threshold settings

G.   Run the Analysis
Note: It is imperative that the microscopist monitor the start of
a CCSEM analysis for several minutes to verify that the analysis
is proceeding normally—i.e., particles are being correctly sized
and classified, images of selected particles or fields are being
collected as desired, and the estimated run time is adequate for
analyzing the desired number of particles or fields. It  is not
unusual that the operator may iterate between Steps C and D
                                                        20

-------
several times, refining the RPF file and restarting the analysis
until the CCSEM run parameters are optimized. It is also recom-
mended that the microscopist check on the analysis periodically
(if possible) to ensure that the analysis is proceeding as intended.

3.3.3  Semi-Automated SEM/EDX
The PSEM provides an interactive, semi-automatic mode of
analysis (referred to as the I-P66  mode), which operates as
follows. The run parameters are essentially the same as for a
CCSEM analysis. The  semi-automated mode differs from the
CCSEM mode in that each time a new field is located during the
analysis, the program pauses for user input. On each field, the
user manually optimizes the focus and the threshold setting. As
with CCSEM, the threshold setting automatically determines
which features are detected in the field, but with I-P66 the user
can manually select which subset of these features to include in
the analysis. Features are selected in one of three ways: (1) The
user selects a feature with the cursor, and the software auto-
matically sizes it and determines its shape parameters using the
current threshold setting (identical to CCSEM except that the
feature selection is done by the user). (2) The user selects a
feature and overlays it with a circle of adjustable diameter. The
feature shape is then assumed to be spherical with the diameter
of the overlaid circle. (3) Rather than define the particle as a
sphere, the user can define the particle as a fiber of adjustable
length. Once the user has selected the features to be analyzed in
the field, the program is restarted. The features are analyzed by
EDX and the results are automatically stored so that the output
data format is identical to  CCSEM.  The interactive analysis
mode is especially useful for the analysis of particles loaded onto
fibrous filters such as Teflon or quartz. The program makes use
of the superior ability of the human eye to discriminate aerosol
particles from the filter substrate. Although the I-P66 mode of
analysis has lower throughput than CCSEM and can also intro-
duce  user subjectivity  into the analysis, it can be  extremely
useful for samples when fully automated analysis is inappro-
priate (e.g., heavy loadings or "noisy" substrates).

3.3.4  Size-Only
Aerosol particle size not only reflects on the sources of the
particles, but also relates to their health effects. Some studies
conducted in the NERL SEM laboratory require measuring only
the size distribution of  an aerosol sample without the need for
chemical analysis. By size, we mean the two-dimensional pro-
jected image of the feature.  The Z-dimension or thickness of a
feature cannot easily be determined by SEM. Both the PSEM
and the LEO/PGT systems can perform automated analyses for
geometry in 2 dimensions only. In sizing particles, both systems
locate the projected particle's centroid and then draw a series of
chords at equiangular intervals through the centroid. Minimum,
maximum, and average diameters are then calculated. Obviously
it is critical that the particle loading be such that the probability
of two particles touching each other is insignificant. Care in
setting the threshold is critical for accurate sizing. Automated
size-only analyses are selected on the PSEM by specifying anX-
ray analysis time of zero seconds (Norm EDX = 0) when setting
up the ZepRun analysis.
     Even though the Z dimensions of particles are not deter-
mined, there is often interest in estimating the mass associated
with a particular class or classes of particles identified by SEM.
To do so requires both a volume and a density estimate. Given
the projected area or  diameters of a particle,  the PSEM and
LEO/PGT (IMIX) software make some assumptions in order to
estimate the volume of the particle. The default assumption used
by both the PSEM and the IMIX software is  that the particle is
a prolate spheroid whose volume is calculated as:
V = 77/6 * D
v   luv  -L-'in
                         D
                         '-'
where Dmax and Dmn are the maximum and minimum of 12
(IMIX) or 16 (PSEM) diameters through the particle's projected
area. Obviously, not all particles are prolate spheroids, and one
volume formula is not appropriate for all classes of particles. For
both the IMIX and the PSEM, the user may replace the  default
volume formula above with another formula of his or her choos-
ing. (The prolate spheroid assumption, however, is probably the
most reasonable choice for typical aerosol samples , given our
present state of knowledge). As discussed in Section 4.2, the
uncertainty in the Z-dimension can result in very large errors in
particle volume and particle mass. Users should be forewarned
that the accuracy of SEM for sizing and "weighing" particles
may diminish rapidly as particles deviate from spherical shape.
     It is important to distinguish between geometric size and
aerodynamic size. The geometric (physical) diameter (Dg) of a
particle is the physical diameter of the particle. The aerodynamic
diameter of a particle (Da) is defined as the diameter of a sphere
of unit density (1 g cm"3), which has the same terminal falling
speed in air as the particle of interest. Dg and Da are related to
each other through the particle density:

               Da = Dg • k • /(p/p,))

where pp is the particle density, p0 is unit density (1 g cm"3), and
k is a shape factor that is 1 for spherical particles.
     Aerosol samplers are typically designed to collect particles
within a certain aerodynamic size  range (e.g., PM25 and PM10
refer to aerodynamic diameters). If the densities of the collected
particles are known, as may be the case for test aerosols used in
wind tunnel studies, then the aerodynamic size distribution can
be determined directly from a size-only CCSEM analysis. If the
particle densities are not known a priori, they can be estimated
for each particle using the particle's X-ray spectrum. In this way,
the PSEM will generate aerodynamic size distributions for sam-
ples analyzed for both size and chemistry. The  IMIX  system
presently does not have an algorithm for estimating particle
density from the particle's X-ray spectrum and therefore cannot
calculate aerodynamic diameters.
                                                         21

-------
3.3.5  Dedicated Particle Searches
The PSEM and the IMIX systems allow the user to set up
computer-controlled dedicated searches for particles that meet
user-defined criteria for size, shape, and/or chemistry. Since the
instrument spends very little time on features that do not meet
the search criteria, large areas of a sample can  be  rapidly
analyzed.  This mode of analysis is ideal for "needle-in-the-
haystack" applications when one is interested in specific rare or
exotic particles, which may be present at number concentrations
approaching 1 ppm (or lower).
     An example of a dedicated particle search is a search for
particle-phase mercury in ambient air samples collected down-
wind of the Astaris elemental phosphorus plant near Pocatello,
ID (see Section 5.3).This was the first time  that mercury had
been observed in samples examined in the NERL  SEM/EDX
laboratory. The ambient sample was a fine-fraction 24-h dichot
sample collected on March 8, 1997, at the Primary monitoring
site immediately downwind of the Astaris facility. The sample
was collected on a Teflo (Gelman Sciences, Inc., AnnArbor, MI)
filter. X-ray  fluorescence  analysis of the sample showed a
mercury concentration of 51.3 ng/m3. Significant concentrations
of Se (49.4 ng/m3) and Cd (19.1 ng/m3) were also measured in
the sample. The total fine mass concentration was 21 ,wg/m3.
Mercury thus accounted for 0.24% of the total fine mass.
     An automated particle search was carried out on the PSEM
to identify individual Hg-bearing particles. Teflo filters typically
present a challenge for automated searches because of the "noisy"
image background generated by the filter. Mercury, however,
because of its high atomic number, is very bright in the BSE
mode compared to the Teflo substrate. By setting a high video
threshold for  the automated search, only particles with a high
effective Z are located and analyzed, dramatically improving the
efficiency of the  analysis. During a 48-h automated search
covering 10.3 mm2 of filter area (1.6% of the total area), 17 Hg-
bearing particles were found for which Hg was greater than 5%
of the particle's EDX spectrum. Identifications of the particles
were verified by returning to each particle at the conclusion of the
automated search using the PSEM relocation feature, manually
analyzing the  particle, and confirming the presence of Hg in the
EDX spectrum. A low-mag field image was  also collected for
each particle to facilitate relocating the particle in the LEO SEM.
The sample was then transferred to the LEO SEM for collecting
high-resolution images of the Hg-bearing particles. Particles were
relocated using XLATE, a relocation program developed by RJ
Lee Group (Monroeville, PA).
     Figures 3-5a, b, c, and d show images and EDX spectra for
four Hg-rich particles. Secondary (left  image) and backscatter
(right image)  electron images were collected for each particle.
Very bright areas in the backscatter image show the locations of
Hg. In all cases shown, submicrometer Hg-rich particles appear
to be embedded in or coated with phosphate, which typically
dominates the fine-fraction aerosol at this monitoring site. The
EDX spectra  for all four particles indicate that Hg  and Se are
present in most Hg-rich particles in a fairly constant ratio, per-
haps as mercury selenide (tiemmanite). In addition, several Hg-
rich particles also contained silver, although silver was below
XRF detection limits in the bulk filter analysis. The Hg-rich
particles display an interesting morphology, often appearing as
strands or filaments. Most of the Hg in the ambient samples
collected at the Primary site is in the fine-fraction, suggesting
combustion sources that include the ground and elevated CO
flares (no longer in operation), calciners, and furnace emissions.

3.4 Particle Classification
     CCSEM can generate a huge amount of data for a single
sample: it is not unusual to analyze  1000 particles per sample,
where each particle may be characterized by 50 or more size,
shape, location, and chemistry parameters. In addition, photo-
micrographs may be collected for some or all of the particles
and/or fields. In order  to facilitate  interpretation of CCSEM
results, researchers have explored various approaches to reduc-
ing the data to more manageable dimensions. Ideally, one would
like to be able to reduce a particle's description from 50+
parameters to only two or three parameters: particle size, particle
class or particle type, and possibly particle morphology. Meth-
ods investigated for particle classification include simple ele-
mental sorting (based on the highest three or four elements in the
particle), ruled-based sorting, spectral matching, cluster analysis,
and neural networks.
     Source apportionment studies have traditionally relied on
chemistry data obtained by bulk analysis of ambient and source
aerosol samples. However, source apportionment can also be
based on CCSEM data acquired on individual particles. In this
case, particles are sorted into distinct classes (Kim and Hopke,
1988a), and the masses of those classes are used to provide
quantitative  source apportionment (Kim and Hopke, 1988b).
Particle classes are most often based on particle composition (as
provided by EDX data), although, ideally,  shape could also be
included to refine the classification. A potentially significant
advantage of particle-based methods is that the additional infor-
mation obtained by analyzing hundreds of individual particles,
compared to a single bulk analysis of the same sample, may
minimize collinearity problems that frequently plague traditional
source apportionment approaches. SEM/EDX data have been
used in a number of source apportionment studies (see  Section
5.1.2).

3.4.1  Default Four-Element Typing
By default, the PSEM  offers a  choice of one-element, two-
element, three-element, or four-element particle classification
based simply on the dominant one, two, three, or four elements
in the individual particle spectra. The resulting particle types are
artificial in that they are not based on measurements  of known
standards. Furthermore, being sample-dependent they are not
directly transferrable from one sample to another. However, the
four-element typing approach can be useful as a first cut in
classifying particle data  and in highlighting differences between
samples.
                                                         22

-------
Figures 3-5a, b, c, d. Mercury inclusions in ambient air sample collected downwind of elemental phosphorus plant. Left
photos: SE images. Right photos: BSE images showing very bright Hg inclusions.
Figure 3-5b.
                                                               23

-------
Figure 3-5c.
 X-ray Display 1
Figure 3-5d.
                                                                 24

-------
3.4.2  User-Defined Rules
The PSEM's off-line analysis software includes a program for the
analyst to develop custom rules (ZepRule, Aspex Instruments,
Delmont, PA). These user-defined particle classification rules can
be based on size, shape (aspect ratio), elemental composition, X-
ray counts,  video level (grayscale brightness, reported as  a
numeric value) of the particle image, and/or any other parameter
measured and recorded during the CCSEM analysis. Examples of
user-defined rules are presented in Appendix C. The rules use
greater than, less than equal to, addition, and subtraction to put
boundaries on the values of each parameter used to define a rule.
More than one  parameter can be used to define a single rule.
Rules are applied in sequential order to each particle until a
particle class assignment is made; thus, each particle is assigned
to only one class, and the order in which rules are listed is critical.
In the examples in Appendix C, the first rule listed for both the
fine and coarse particle samples serves to remove particles which
were not in the aerodynamic size range of interest from further
classification (i.e., coarse particles are excluded from fine particle
classification, and vice versa).
     Typically, rules are applied to each sample analyzed, and
results evaluated by examining both measured parameters and
particle  images. Rules can then be changed or added based on
these evaluations, and the process continues in an iterative man-
ner until the particle classifications are judged by the analyst to
be satisfactory, based on the uniformity of chemical and physical
characteristics within a particle class. The rules can be applied
to other samples  to make minor  adjustments and to test the
robustness of the classification scheme, based on the uniformity
of particle characteristics within each particle class both within
a single sample and across all samples. It is recommended that
the X-ray spectrum be  reviewed to verify the identification of
elements,  such as trace metals, that are typically low in abun-
dance and/or subject to significant interferences.
     Particle classification in the PGTIMIX system is facilitated
by the "Chemical Classification" program, and can be carried
out on-line or off-line. In a similar fashion to the PSEM, parti-
cles are classified according to a set of user-defined rules. In
contrast to the PSEM, however, the IMIX classification rules are
based only on particle chemistry, typically expressed in terms of
EDX peak-to-background values and relative peak ratios for the
elements of interest. Thus, the IMIX class editor program does
not allow one to combine a physical criterion such as "circu-
larity", with EDX criteria (to define, for example, a class called
"Spherical Fly Ash"). An example of an DVIIX class file is that
of quartz which must meet the following four conditions: (1) Net
Si / Bkgnd Si >  10; (2) Net O / Bkgnd O > 3; (3) Net O / Net Si
> 0.05 and < 0.5, and (4) (Net O + Net Si) / (Net O + Net Al +
Net Si + Net S  + Net K + Net Ca + Net Fe)  > 0.7. The IMIX
particle classification program also differs from the PSEM rule-
based classification in that each IMIX feature is tested against
every class file, without regard to order. Thus, one particle can
be assigned to more than one class, if the defining rules are not
mutually exclusive.
     In addition to off-line classification, both the PSEM and the
PGT systems allow the user to define rules to screen or constrain
the particle data that is collected on-line. The on-line rules or
"selection formulas" can employ a comprehensive set of arith-
metic operators and functions, as well as basic logic operators
which can be applied to virtually any measurable feature quan-
tity. A common application of the  on-line rules is to exclude
particles from analysis that are either too small or too large. The
PSEM allows  on-line screening based on both physical and
chemistry parameters, while  the  IMIX  system "selection
formulas" only screen particles based on physical parameters.

3.4.3  Cluster Analysis
Cluster analysis (CA) is a multivariate statistical technique used
to reveal structure in large, multidimensional data sets. In the
area of aerosol analysis, CA provides a statistical method for
sorting particles into distinct groups separated from each other
by chemistry, size, and/or morphological differences. It has been
applied by researchers in  source apportionment studies to help
in the identification of emission sources (Section 5.1.3). The
results of CA can also be used to generate user-defined rules for
particle classification.
     One research group at the forefront of developing cluster
analysis techniques for use with electron probe or SEM/EDX
data is that of R. Van Grieken and colleagues at the University
of Antwerp. This group has  developed  a Windows-based
software package for cluster and factor analysis, called IDAS,
and has applied the tools  of IDAS in characterizing aerosols
by  electron probe  analysis and  identifying their  sources
(Bondarenko et al., 1996). IDAS provides an easy-to-use inter-
face to carry out  two main multivariate analysis techniques,
cluster analysis (CA)  and factor analysis (FA).  CA is imple-
mented as a three-step procedure: Hierarchical cluster analysis
(HCA) is the first step. Its results serve as an initial partition for
nonhierarchical cluster analysis (NHCA). Finally, the internal
structure of the obtained clusters and relationships between them
can be revealed with fuzzy clustering analysis (FCA). FCA is
implemented in IDAS as  a two-step process. First,  principal
factor analysis  (PFA) is used to estimate the "correct" number of
factors and to  calculate factor loadings and scores. The next
stage, meaningful interpretation, canbe done either with the help
of abstract factor rotation (APR) or with target transformation
factor analysis (TTFA).
     The feasibility of using IDAS software to identify sources
of particles was explored in studies conducted by the NERL
SEM/EDX Laboratory. The hypothesis was that if particle data
from known, chemically distinct emission sources were fed into
IDAS, the program should be able to  sort the  particles into
distinct clusters in elemental space, with each cluster repre-
senting one of the sources. If this proved successful, the next
step would be  to apply cluster analysis to  an ambient sample
impacted by these sources and attempt to apportion each ambient
particle to a specific source. For the preliminary tests discussed
below, the source particle database comprised seven sources of
                                                          25

-------
30 particles each. The 7 sources included six sources sampled in
the Czech Republic city of Ostrava (Willis et al, 1997): mobile
source emissions, stack emissions from two steel works, a coke
oven battery, a sintering plant, a coal-fired power plant, and a
coking plant. In addition, residual oil fly ash was characterized
and included in the source particle database. In this preliminary
attempt to use the cluster analysis software only particle chemis-
try was used. Future refinements of the technique may be able to
incorporate particle size  or shape parameters (e.g., aspect ratio
or circularity) in addition to chemistry in order to further resolve
clusters that may overlap in elemental space.
     The first step in cluster analysis with IDAS is HCA. The
results of HCA  are presented in the  form of a  dendrogram
starting with TV leaves (N = number of particles) and terminating
with a single group encompassing all particles. The user, relying
on his knowledge of the  air shed or on "stopping rules" gener-
ated within IDAS, must decide how many clusters are repre-
sented in the data, that is, where to terminate the dendrogram.
IDAS includes built-in graphical support to display the stopping
rules. Figure 3-6  shows a graph of stopping rules calculated for
the seven-source  data set. Four different tests can be applied to
the data  to predict the "true" number of clusters up to a maxi-
mum of  10. (For  each test, a local minimum in the plotted line,
or a strong deflection indicates the "best" number of clusters to
use in subsequent analysis.) Generally, the "CAIC" test is most
successful in determining the number of clusters. In Figure 3 -6,
the CAIC test (as well as the inflection in the DB plot) indicates
seven clusters. Both the WB  and SD tests would indicate eight
clusters.  Choosing seven clusters as the best number, IDAS
calculates the population of each hierarchical cluster and identi-
fies the location (cluster) of each individual particle. (HCA and
NHCA are both "hard" clustering techniques in that each particle
can be assigned to one and only one cluster).
     Figure 3-7 is a graph of the resulting cluster populations.
With our simulated data set,  one would ideally expect a popu-
                      Ckister populations
0.9 -
0.8 -
0.6 -
0.5 -
0.4 -
0.2 -
0.1 -
0.0
5





J
7
/
/


\
K
/ \
/
/





~~\
^-— '


\
\






N"-
V



X




^



\
\

^



•^ s





V
VJ
>VX^


V
> 	





-*•%.
^N



	 -^.







XX


•^,







S.

-O.fl
-0.8
-0.6
-0.5
-0.4
-0.2
-0.1
-0.0
                                                  — CAIC

                                                   - DB

                                                  — WB

                                                     SD
                   Number of clusters

Figure 3-6. A plot of the "stopping rules"calculated for a simulated data
set  comprising seven  emission  sources. The  plot  displays  the
dependence of four different clustering criteria on the number of clusters.
Local minima or inflections in the curves indicate the "best" number of
clusters.
Figure 3-7. Cluster populations resulting from hierarchical cluster
analysis. The data set comprised 30 particles from each of seven different
emission sources. One would expect each cluster to contain 30 particles
if all particles representing a given source were truly distinct from those
of all other sources.
lation of 30 particles in each of 7 clusters. As Figure 3-7 indi-
cates, the emission sources are not totally distinct, causing some
particles from different emission sources to be grouped together
in the same cluster. By examining the location of each particle,
we determined the following:

     Cluster 1 represents the  coal-fired power plant and in
addition includes some mobile source particles.
     Cluster 2 is a weak cluster containing particles from both
the  coke oven stack emissions as well as from mobile sources.
     Cluster 3 represents the steel works source.
     Cluster 4 is a combination of the coke oven stack particles
and residual oil fly ash.
     Cluster 5 represents the sintering plant.
     Cluster 6 is dominated by mobile source particles,  but
includes some oil fly ash  particles.
     Cluster 7 represents the coke battery emissions.

The results indicate that residual oil fly ash and the coke oven
stack emissions are somewhat overlapping clusters, and that the
mobile  source particles are scattered  among four different
clusters. The latter observation suggests that the mobile source
sample  may include particles emitted from other sources. (It is
very probable, in fact, that the mobile source signature, which
was collected in a busy traffic underpass, is actually a mixture
of mobile emissions diluted by ambient aerosol.)
     The clusters identified in HCA  serve as the initial "seed
points"  for NHCA. NHCA produces somewhat revised cluster
populations as shown in Figure 3-8.
     IDAS also calculates statistics for each cluster, including
the mean and standard deviations for each elemental species.
Ideally, the mean elemental  concentrations calculated for each
cluster canbe interpreted as source profiles. One could use these
values to create user-defined rules for the P SEM to facilitate off-
line sorting of particles into source-related clusters.
                                                            26

-------
                      Cluster population
Figure 3-8. Revised cluster populations after non-hierarchical cluster
analysis.
     The authors of IDAS recommend that any conclusions
about the structure of the data set should be based on comparison
of the results of all three clustering techniques: hierarchical, non-
hierarchical, and fuzzy. (The "fuzzy" technique was not ex-
plored in this preliminary study).
     In a second application of CA to particle classification, CA
was  used  to classify  particles collected downwind from an
industrial complex in Pocatello, Idaho. Chemistry and size data
were collected by SEM/EDX on 314 particles between 2 and 15
/-on in size. Hierarchical cluster analysis (HCA) was first carried
out on the data using Ward's method and assuming a Euclidean
distance metric (Statistica. Graphical tools estimated the "most
likely" number of clusters to be seven. Non-hierarchical cluster
analysis (NHCA) was then applied to the data using the seven
HCA cluster vectors as initial seed points for NHCA. The identi-
fication of the seven clusters (based on the mean elemental com-
position of each cluster) and their abundance (number %) were
as follows:  (1) K-P-rich, 16.2%;  (2) Ca-P-rich,  15.0%; (3)
aluminum silicates,  20.4%; (4) Ca-Si-P, 19.4%; (5) P-O-rich,
6.1%; (6) Carbonaceous, 8.0%; and (7) quartz,  15.0%. Analysis
of the particle  size distribution within each cluster of particles
could probably indicate which of these particle types is pre-
dominantly in the coarse-fraction,  and therefore likely to be
attributable to  resuspended dust sources, and  which types are
predominantly in the fine-fraction, and therefore likely to be
attributable to combustion sources.

3.4.4  Spectrum Matching
Another approach to particle classification has been developed
on the PGT IMIX system. The DVIIX system  differs from the
PSEM in that the IMIX system stores the complete EDX spec-
trum for a feature, while the PSEM typically stores only X-ray
counts in user-selected elemental regions of interest. In effect the
PSEM philosophy is to compress the raw spectral information
into a small set of numbers that are easily manipulated and take
up less disk space.
     In the PSEM, particles are sorted into distinct chemical
classes based on simple 4-element typing or more sophisticated
user-defined rules involving the particle size, shape, and X-ray
data. The process of generating these rules is typically an itera-
tive process that canbe very time-consuming and labor intensive
and ideally is best done by an experienced microscopist. The
ability of the IMIX system to save the raw spectra of particles
allows for a purely mechanical particle  classification scheme
based on the similarity between the sample spectrum and stored
standard spectra. The IMIX software includes a program called
Spectrum Match, which performs a least-squares fit of an un-
known spectrum against a library of (up to 200) standard spectra.
The program outputs chi-square for all matches  in order of
ascending chi-square.
     A new program built around the Spectrum Match program
shows promise as a method for creating a user-defined number
of particle classes from a collection of EDX spectra. The pro-
gram performs essentially three tasks: (1) Given a collection of
particle spectra, the script generates chemically distinct classes
based on a least-squares match of the current particle spectrum
against  an evolving library of standard spectra representing
different particle classes. (2) Given a library of standard spectra
representing different particle classes, the script determines the
best match for each unknown spectrum based on the chi-square
of the least-squares match. (3) Given a library of standards (or
particle types) the  script  performs a least-squares fit  of all
standards (or particle types) against each other in order to ident-
ify potential collinearities among the particle types.
     Given a collection of source samples, it is envisioned that
the new script will be able to generate source profiles each of
which consists of one or more particle types and their relative
abundance. The program will then matchparticles from ambient
samples against the library  of source types to apportion the
ambient sample to contributing sources. The source apportion-
ment  can be carried out using conventional  Chemical Mass
Balance (CMB), in which the usual elemental concentrations are
replaced with abundances of the different particle types.
     In order to generate particle classes "from scratch" (e.g., a
new source sample), the user inputs the chi-squared cut point,
which determines how  many different classes will be created
from a collection of particle spectra. A  small chi-square will
typically result in the creation of more classes, while a large chi-
square will produce fewer classes with broader chemical defini-
tion. The script operates as follows: Particle spectra are read into
the script successively and compared to standard spectra,  which
are continuously evolving. The very first particle  spectrum is
matched against itself to create the first standard (Type 1) since
no standard spectra exist initially. The second particle is com-
pared to the Type 1 spectrum. If chi-square of the least-squares
fit is greater than the user-selected cut point,  then the second
particle is classified as a new type, and its spectrumbecomes the
initial spectrum representing Type 2. If, on the other hand, chi-
square for the match is  less than the cut point, then particles 1
and 2 are both considered to be Type 1, and the Type 1 spectrum
                                                          27

-------
is subsequently  replaced by  the average of spectrum 1 and
spectrum 2. This continues until all particles have beenmatched.
     The  least-squares approach described above has  some
inherent flaws that might invalidate the approach. As the Type 1,
Type 2,... Type N spectra evolve, particles initially assigned to
a particular type may no longer fit that type by the time all
particles have been classified. This potential problem is dealt
with as follows: At the end of the initial sorting process, the
resulting particle types are frozen and the collection of particles
is matched against these  final, fixed types.  Another potential
problem with the classification scheme is that  the shift in the
"centroid" of a given type with the addition of the nth particle is
weighted by l/n where n is the number of particles in that class.
Although the final spectrum representing the class is the simple
arithmetic average of all  member spectra, the  membership in
each type  is largely determined by the first few particles in the
particle list coupled with the user's choice of chi-square. If the
particle list is shuffled and the script is re-run, different particle
classes will be generated.
     In order to  test the new script, approximately 50 particles
from each of 20 sources  were analyzed by  EDX. Spectra for
each source were individually run through the script to generate
particle types representing that source. A chi-square cutpoint of
1.5 was rather arbitrarily chosen. Some sources were represented
by only a single particle  type  (monodisperse), while  other
sources, especially soils, were characterized by 10 ormore types
(multidisperse). A total of 73 source types were generated, but
there were a number of collinearities (arbitrarily determined by
chi-square < 0.5) among the types. The same collection of 1150
spectra used to generate the  source profiles were then treated as
a proxy for ambient particle spectra and matched against the
resulting library of source types. The results of the particle class-
ification are shown in Figures 3-9a and 3-9b.
     The 20 different sources are shown along the x-axis of each
figure. The number in parentheses is the number of particles
used to generate the source types.  The sources are listed in more
detail below.
     Figure 3-9a shows  the  raw apportionment results after
matching the "ambient" particle spectra against the library of 73
particle types. The results are totally unconstrained by source
profiles, i.e., all  particle types are assumed  to be independent
from each other. It was hoped that by making use of the source
profiles to constrain the model fit that a more accurate source
apportionment could be obtained. The results, shown in Fig-
ure 3-9b, were obtained using the CMB model and the source
profiles obtained from the least-squares approach. The uncer-
tainties  in the "ambient" CMB file for each particle type were
largely determined by the number of other types judged to be
collinear with the selected  type.  Pairs of types whose spectra
matched each other with a chi-square <0.5 were assumed to be
collinear. In the worst case,  one soil type was collinear with 19
other particle types. The uncertainty associated with this type in
the CMB input file was therefore made to be very large. The raw
apportionment results in Figure 3-9a  and the refined CMB
results in Figure 3-9b are very similar, and it is not obvious that
the apportionment results improved with the CMB calculations.
      The test results above look promising. Ideally the calcu-
lated/true  concentration for all sources should be unity.  The
largest deviation from unity occurs for the "sticky carbon tab"
source, for which there was only a single particle represented in
the data set. Errors for the highly collinear sources were gen-
erally in the range of 20% to 35%. In one sense, the result
represents a best case scenario in that the particle types were
generated from the same spectra that were subsequently matched
against the  particle types. On the  other hand, the test case
presented an unusually difficult challenge for the methodology
because of the similarities between the Brownsville Soil, Czech
Soil, Arizona Road Dust, Mt. St. Helens Ash, and FMC Shale
samples. It must be emphasized as well that individual sources
were typically represented by only about 50 particles, whereas
studies described in Section 4.6.2 demonstrate that about 500
particles are needed to adequately represent ambient and some
multidisperse source  samples. Better statistics should improve
accuracy and confidence in the source profiles generated by the
method, which in turn may improve the CMB results. Finally,
for this initial test, the chi-square cut point for creating new
particle types was arbitrarily chosen as 1.5.  A lower value for the
cut point may yield some improvement in the results.
     Potential advantages of  the least-squares classification
scheme are as follows:

     •  Particle types are generated via a purely mechanical
        and objective process without the need for experienced
        human input (other than to select the chi-square cut
        point that determines the number of particle types).
     •  The PGT software allows up to 200 different particle
        types to be used in the classification process. This may
        permit more highly resolved and more accurate  source
        profiles than can be obtained by user-defined rules.
     •  The classification script is relatively quick: Particle
        types can be generated from a collection of  500
        particles in roughly  1.5 hours. Matching 500 particles
        against a library of 73 types takes approximately the
        same amount of computer time with the present PGT
        SparcS computer. With an upgraded computer, the time
        required for these tasks could probably be cut in half at
        least.
     •  The output of the classification script (source profiles
        expressed as percent  abundance of different particle
        types and the sorted ambient data) can be input directly
        into the CMB model for source apportionment calcula-
        tions. A statistically valid treatment of uncertainties in
        the source and ambient CMB data files, however, needs
        to be developed.
                                                          28

-------
 13 0.0
 =
                   SOURCE (number of particles represented)
                                                                I o.o
                                                                =
                                                                        (D  (£ OT_ —

                                                                        11 1 I
                                                                                          G S £.
                                                                                          
-------
information and data processing and have found impressive
success in areas such as pattern matching and classification and
data clustering. The theory behind ANNs is beyond the scope of
this document. However, descriptions of ANNs are numerous on
the web and in the literature.
     Wienke et al. (1994) and Xie et al. (1995) used a specific
type of ANN called adaptive resonance theory or ART-2a
(Carpenter et al. 1991) to rapidly identify particle shape from the
binary SEM particle image. The resulting shape parameter was
then combined with chemical composition data and input into a
second ART-2a net to yield a "more intelligent" classification of
CCSEM data. The authors investigated the potential of a ART-
2a for unsupervised classification (clustering) as well as super-
vised particle classification using training sets. An important
advantage of ART  networks, compared to other supervised
ANNs is the ability of ART to dynamically create new classes in
response to novel particle types which were not represented in
the training set. Applications of neural nets in the area of
environmental studies have been pioneered by Phil Hopke and
colleagues at Clarkson University. (Hopke and Song, 1997; Xie
et al., 1994; Song et al., 1999; Song and Hopke,  1996). Con-
ventional receptor modeling has difficulty in determining the
contributions of very similar composition source material. Song
et al. (1999) used a combination of the ART-2a neural network
and the back-propagation (BP) neural network to apportion the
mass contributions of three different crustal sources to ambient
particle samples. It is conceivable that  in the future, a neural
network could be trained to monitor a CCSEM analysis, pro-
viding data  quality control in real-time, identifying statistical
outliers in  the  data, and determining when to  terminate an
analysis (e.g., when additional data are changing the  results
negligibly).
                                                         30

-------
                                                   Chapter 4
                                        Data Quality and Validity
4.1  Instrument Calibration and Maintenance
Instrumental factors impacting the validity of the SEM/EDX
data include (1)  accuracy of the SEM magnification (which
affects the  accuracy of the particle size measurements) and
(2) the gain calibration of the EDX amplifier. The latter affects
the  accuracy of the X-ray peak  identification subroutine that
assigns element labels to the  spectral lines  according to the
channel (energy) locations of the peak centroids.

4.1.1  Magnification Calibration
The magnification of the SEM is primarily calibrated by using
the  MRS-3 magnification calibration standard (Geller Micro-
analytical Laboratory, Topsfield, MA). The MRS-3 comprises
a series of precisely etched geometric patterns that allow the
SEM mag calibration to be accurately checked at mags ranging
from lOx to 50,000x. The MRS-3 is available with or without
traceability to NIST. The standard used in the SEM laboratory
is not traceable; however, the manufacturer  has never seen a
pattern that deviated by more than ±0.5 ^m for the 50-^m and
500-^m pitch, or ±0.1 ,wm for the 2-^m pitch. The MRS-3 is
used to determine SEM error in measuring spacings on precisely
etched grid patterns ranging from 2-^m pitch to 500-^m pitch.
The maximum observed errors in either the LEO S440 or the
PSEM was have been on the order of +4%. As discussed below,
sources of error other than the magnification calibration appear
to be as important or more important in determining the SEM's
accuracy  in sizing particles. Thus,  magnification calibration
errors of up to a few percent are probably acceptable and do not
need to be corrected by the vendor.
     The use of poly styrene latex beads for calibrating or check-
ing  SEM magnification has been carefully explored  in our
laboratory with discouraging results.  For reasons that are not yet
understood, we have been unable to get agreement between the
MRS-3 standard and the NIST-certified PSL bead diameters on
either the PSEM or the LEO S440.  The discrepancy between
PSL beads  and the MRS-3 standard is greatest for submicro-
meter beads, where pixel resolution and particle size increase
due to gold or carbon coating can cause significant error, but
these  errors do not appear to explain all of the observed dis-
agreement.
     The magnification calibration for both SEMs appears to be
quite stable over time, however, a calibration check of the SEM
is recommended at least twice a year. Calibrations for both SEMs
are routinely checked during the annual preventive maintenance
visits  for each instrument. Using the MRS-3 standard we will
measure the magnification calibration errors of the two SEMs for
at  least  four  magnifications spanning  the range  of 200x to
20,000x. The  measured dimensions in both X and Y as well as
the ratio of the two must be within spec before the instrument is
used.  A record of the results should be  maintained so that any
drift in the magnification of either instrument over time will be
noted. If deviations from the MRS-3 standard exceed 5%, the
instrument's magnification calibration is adjusted by the field
engineer, or we will make the adjustments ourselves if possible.

4.1.2  EDX Spectrometer Gain Calibration
Both EDX systems use software to recalibrate the EDX spec-
trometer gain. The  spectrometer gain determines the keV per
channel for the EDX spectrum, which in turn provides the basis
for X-ray peak identification. EDX gain can be calibrated using
a gold/copper standard since X-ray lines from these two ele-
ments span almost the entire spectral range of the  detector. In
order to  recalibrate the spectrometer gain, it is necessary only to
acquire a spectrum of the Au/Cu standard and to run the auto-
calibration routine.  Recalibration ensures that the centroids of
the Au and Cu X-ray lines (and by extension, the lines of all
other  elements) coincide with the corresponding line markers
generated by  the EDX software so that  peaks in the EDX
spectrum are   assigned to the correct elements.  EDX peak
positions should be within 10 eV of the tabulated values.
Periodic checks on the PSEM and  IMIX  energy  calibrations
indicate that the calibrations are very stable. The energy cali-
brations are checked at least once per year as part of each instru-
ment's annual preventive maintenance.
                                                        31

-------
4.1.3  Window Contamination Check
Over time, deposits including ice or oil can build up on the ultra-
thin window of the EDX detector, resulting in a loss of detector
sensitivity. Changes in detector sensitivity can be monitored
fairly accurately by tracking Ka/La ratios for Cu or Ni over time.
Because X-ray absorption losses due to window deposits will be
greater for the La line than for the more energetic Ka line, the
Ka/La ratio will increase over time as deposits build up on the
detector window. We typically use a copper grid as a sample.
When  performing a detector  sensitivity check, one should
always use the same probe current, working distance, and X-ray
count rate whenever checking the Ka/La ratio. If loss of sensi-
tivity becomes significant, the detector window can be cleaned,
using extreme caution, either by the vendor or by the user.

4.2 Precision and Accuracy of Particle
     Volume  Estimates
     Particle sizing error was investigated in the PGT and
PSEM systems. In each system, we imaged and sized a roughly
spherical gold particle  (diameter ~1.3 ^m) and an irregular,
flattishparticle of lutetium oxide (average diameter- 1.6 ^m) at
five magnifications (1000,2000,4000,10,000, and 20,000). For
the PGT system, field images were acquired at 1024 by 1024
pixel resolution prior to particle sizing. This resolution yields
calibration scales of approximately 0.31, 0.156, 0.077, 0.031,
and 0.0156 ^an/pixel for the five magnifications above.  The
PSEM software is designed such that once a feature has been
located by searching the field along grid points, the spacing of
the grid points is reduced to provide higher pixel resolution for
sizing the feature. For the PSEM, the measurement grid spacing
was set to the highest possible resolution,  corresponding to
0.128,0.064,0.031,0.013, and 0.006 ^m/pixel at the five mags.
Thus, under the conditions of this test, the PSEM has approxi-
mately 2.5 times better pixel resolution than the PGT system.
The gold  and Lu2O3 particles selected for  sizing comprised
approximately 10-12 pixels in the PSEM at lOOOx, the minimum
recommended number of pixels needed to define a feature. The
magnification in the PGT system must be  greater than 2000x in
order for the minimum pixel size threshold to be exceeded. Even
at 4000x, a one-pixel error in the diameter of a 1.3-^m sphere
represents a volume error of approximately ±19% in the PGT
system and ±7%  in the PSEM.
     Forthe PGT system, particle volume was calculated at each
mag using the following four volume formulas:
     Vrjl = 7i/6 * Dmax  * Dmm2, where Dmax and Dmm are the
maximum and minimum of 12 diameters (PGT) or 16 diameters
(PSEM) through the particle center of mass;
     Vsphere = 4/3 * 7i"1/2 * Area3'2, where Area = projected area of
the particle; and
     Vprolate = (8/37i)* Area2 /Dmax
The four formulas reduce to the identical formula for perfect
spheres, but can yield significantly different volumes for a non-
spherical particle.  Vrjl is the default formula for  calculating
                                                            particle volume in the PSEM. Like Vprolate Vrjl assumes a prolate
                                                            spheroid rotated about the semi-major axis.  The difference
                                                            between the two formulas is that Vprolate is based on the projected
                                                            pixel area of the particle, while Vrjl uses the product of measured
                                                            diameters Dmax and Dmm. Errors in particle volume were com-
                                                            puted relative to the "true" volume which was assumed to be the
                                                            volume of the particle measured at 20,000x (least affected by
                                                            pixel resolution error) using the prolate sphere formula, Vprolate.
                                                            (This formula yielded the best agreement between the SEM-
                                                            determined mass and the known gravimetric mass of a lutetium
                                                            oxide sample - see Section 4.3).
                                                                 Volume calculation results for the PGT system and the
                                                            PSEM are  shown in Figures  4-la (gold particle) and 4-lb
                                                            (lutetium oxide particle). The PSEM and PGT measurements
                                                            were made on  different, but similarly sized, particles. Four
                                                            different formulas were used to calculate particle volume for the
                                                            PGT system while a single formula, Vrjl, was used for the PSEM.
                                                            Magnification calibration errors have been  corrected at each
                                                            magnification using a magnification calibration standard.
                                                                               Spherical Gold Particle
                                                                       1000   2000    4000    10000   20000
                                                                               SEM Magnification
                                                                                                          -O- SPHERE
                                                                                                          •-Q- PROLATE
                                                                                                             RJL_PROL
                                                                                                          -i- AVG_DIA
                                                                                                          -•»- PSEM
                                                            Figure 4-1a. Relative volume error for a 1.3-^m quasi-spherical gold
                                                            particle as a function of SEM magnification for PGT system (four curves)
                                                            and PSEM (one curve).
200
150
g
1 1°°
>
1 50
0
-50
j
'
<
I

! ! !
"^ I Lutetium Oxide Particle I
------,
i._
' 	 1
1 " ' - I

i
j
i
i
. 	 ^ 	
"-^.. '"*-** 	 T."^
• l(^_, ..•
a- - - 1 ... -I
' • - *
i
i
	
".lr....=i!


•=•-"•*-<
j, . . ,




t- - *•
i
1000 2000 4000 10000 20000
-O- SPHERE
-O- PROLATE
"-^- RJL_PROL
-A- AVG_DIA
-•• PSEM
                                                                               SEM Magnification
                                                             Figure 4-1b. Relative volume error for a 1.6-um lutetium oxide particle as
                                                             a function of SEM magnification for PGT system (four curves) and PSEM.
                                                         32

-------
     As expected, the PGT errors calculated from the four
volume formulas for the quasi-spherical gold particle tend to
converge at high magnification. True convergence is not ex-
pected, however, since the particle is not truly spherical. Com-
pared to Vrjl, Vavg_dia, and Vsphere, the area-based prolate formula,
Vproiate, and the PSEM volume calculation are less sensitive to
changes in the SEM magnification above 2000x.
     The four PGT volume estimates for the non-spherical lute-
tium oxide particle (Fig. 4-lb) show little convergence to the
"true" volume as magnification increases, but instead show that,
for non-spherical geometry, the four formulas calculate signifi-
cantly different volumes. Volumes calculated using Vavg ^ con-
sistently show the largest errors. Again, Vprolate, and the PSEM
volume calculation are least sensitive to changes in the SEM
magnification.
     If accurate particle volume is a concern in an automated
analysis, these results suggest that, for particles on the order of
2 jum or less, one should use a minimum magnification of 4000x
for the PGT system and 1400x for the PSEM. (The magnifi-
cations used in Figures 4-la,b assume the display monitor as the
output device. The PSEM  magnification is referenced to the
thermal printer  output and is approximately 2.9 times smaller
than the magnification referenced to the display monitor. A mag
of 1400x on the PSEM corresponds to about 4000x on Figures
4-la,b).

4.3  Precision and Accuracy of
     Particle Mass Estimates
In order to assess the precision and accuracy of mass estimates
from SEM data, a sample of lutetium oxide particles was sized
in the PSEM. The PSEM-calculated mass of this sample was a
factor of 4.33 higherthanthe gravimetric mass. This discrepancy
was attributed to errors associated with the assumption that all
particles are prolate spheroids. The same filter was  analyzed
using the LEO/PGT system in order to verify the PSEM results.
     Mass reconstructed from individual particle analysis is
critically dependant on volume estimates. Four different formu-
las (presented in Section 4.2), including the PSEM formula for
prolate spheroid volume, Vrjl, were used to calculate particle
volume and particle mass in data collected with both SEMs. One
objective of this exercise was to determine which of the above
volume formulas yields the best agreement with the gravimetric
mass  for  nonspherical particles. Results of the PSEM  and
LEO/PGT analyses are shown in Table 4-1. All SEM mass esti-
mates greatly overestimated the gravimetric mass of the sample
(22 /j,g). Best agreement with the gravimetric mass was obtained
using Vprolate for the LEO/PGT  system (236% error) and Vrjl for
the PSEM (323% error).  The maximum and minimum mass
estimates for the  LEO/PGT and PSEM systems differed by
factors of 2.1 and 1.4, respectively, indicating  considerable
sensitivity to the choice of volume formula.
     The large errors in Table 4-1 are attributed to the flat, plate-
like morphology of the Lu2O3 particles, which were poorly sized
Table 4-1. LEO-PGT and PSEM Analysis of a 22-^g Lutetium Oxide
Sample: Estimated Particle Mass Using Different Volume Formulas and
Associated Errors

LEO/PGT (//g)
LEO/PGT error (%)
PSEM (,ug)
PSEM error (%)
™ orolate
74
236
103
368
™ sohere
99
350
132
500
Vril
116
427
93
323
™ ava dia
157
613
125
468
using the volume formulas above. The shape of the  Lu2O3
particles is atypical of environmental aerosols which are typi-
cally more round than flat; thus the errors in Table 4-1 represent
an extreme case. Nevertheless, these results demonstrate that
particle shape is a major factor in the error associated with infer-
ring a three-dimensional particle volume from a two-dimensional
particle diameter or projected area. For  an unusual aerosol
sample with aytpical particle shape, one may need to examine
the sample first and choose a more appropriate volume formula,
before attempting to calculate the bulk mass of the sample.
     Similar analyses were performed on Arizona Road Dust
since the results would have a more direct bearing on SEM-
derived mass estimates for typical PM samples. Arizona Road
Dust is a test material with a bulk density of 2.65  g/cc and a
well-characterized particle size  distribution.  Examination by
SEM showed that particles were generally more round than flat.
Two samples were prepared from the "coarse" size fraction of
the Arizona Road Dust. Dust was suspended in propanol, aero-
solized using a glass bulb nebulizer, and manually sprayed onto
tared 25-mm Nuclepore filters placed on a glass frit and pumped
from the backside. After loading, the filters were equilibrated
and weighed to determine the mass loadings. The two filters had
nearly identical loadings (31  and 32 /j.g, respectively). Repeat
measurements were made on both samples in order to provide a
measure  of  precision in SEM-determined mass loadings via
PSEM analysis. Efforts were made to  maintain the same mea-
surement and detection thresholds from run to run, if necessary
via slight adjustments to the brightness.
     The precision of three repeat analyses of the same fields
(227 fields, 6000 particles) yielded a precision (relative standard
deviation) of 3.5%. Eight successive analyses of random fields
(6000 particles, -110 fields) yielded  a precision of 11%.
Because  of the small mass loadings on these samples, a few
anomalously large particles can give a significant positive bias
to the SEM mass estimate.  In one analysis, for example, the
largest particle (after normalizing for the fraction of sample
analyzed) accounted for 25% of the total estimated mass loading.
In order to eliminate particles that unduly bias the  results, all
analyses for a given sample were combined into a single analysis
and the resulting particle size distribution was plotted.  The
largest six particles for each of the two samples were identified
as outliers and eliminated from the data sets before estimating
the mass loadings.  The estimated mass loadings for our two
Arizona Road Dust samples were 24.4 ± 2.4 /j.g and 35.9 ±
4.1 /j,g, respectively, for the 31-,wg and 32-^g samples. These
                                                         33

-------
represent errors of -21% and +12%, respectively. Note that
without eliminating the six largest  particles, this last  mass
loading was estimated to be 41.1 /j,g. Given the ability of a few
large particles to severely bias the mass results, the accuracy of
these mass estimates probably cannot be improved except by
analyzing a much larger fraction of the sample.  (Each of the
individual analyses represented between 0.06% to 0.1% of the
total filter.)
     The  setting  of the  PSEM  measurement  threshold is
extremely critical in determining a  sample's size and  mass
distribution. This sensitivity is illustrated in Figure 4-2, which
shows  size distributions  for back-to-back  analyses (6000
particles in randomly selected fields) of the 31-^g sample.
                 Meas. Threshold Too Low
                Meas. Threshold OK
     0.0   0.5   1.0   1.5   2.0   2.5   3.0   3.5   4.0   4.5   5.0  5.5  6.0
                       Average Diameter, pm
Figure 4-2. Back-to-back particle size distributions using the PSEM with
different measurement threshold settings.

     The number of particles for both the line plot and the bar
plot were  normalized to the same analyzed area of the sample
and should, in principle, be very similar. The bar plot, however,
shows many more particles detected in nearly all size bins than
the line plot. Furthermore, the  estimated mass loadings were
59.4 fj,g versus 25.9 /j.g for the analyses  depicted in the bar and
line plots, respectively. The large discrepancy in the two  size
distributions  is a consequence of setting  the  measurement
threshold too low (too close to the noise level) in the former plot.
Although the threshold appeared to be set properly on the few
fields examined prior to the start of the analysis, a number of
fields analyzed during the  run  had slightly higher brightness
levels,  which elevated the noise  level to approximately the
measurement threshold. Particles that should have been excluded
for being below the minimum size threshold (0.4 ,wm) were then
oversized and counted. Particles already above the minimum size
threshold were pushed into larger size bins. These results  em-
phasize the importance of setting the analysis thresholds proper-
ly, collecting all field images, monitoring the SEM periodically
during a CCSEM analysis, and carefully examining the data on
a field by field basis after the analysis.
     It is important to keep in mind in this discussion that the
mission of SEM/EDX is not to replace bulk analysis, but to
complement bulk analysis by providing information not avail-
able via bulk techniques. SEM should not be considered  as a
tool for determining bulk, macroscopic properties  of a sample
such  as the sample  mass or the bulk  sample composition.
However, the "ball park" bulk mass estimates extrapolated from
microscopy data should be useful as a QC check on the sample
or on the quality of the SEM analysis. SEM-based mass esti-
mates that  differ greatly from the gravimetric mass may thus
indicate problems with the sample such as highly-non-uniform
particle loading, contamination of the sample by foreign parti-
cles, or gross overestimate or underestimate of carbonaceous
mass.

4.4 Analysis of Ultrafine Particles
In light of recent  studies that show an association  between
human morbidity and exposure to air particulates, especially
particles in the fine fraction (< 2.5 ,wm, aerodynamic), there is
interest at the U.S. EPA in identifying tools that can characterize
fine and ultrafine  particles  in the  air (size, concentrations,
chemical composition). Ultrafine particles are particles smaller
than 0.1 jum. The following paragraphs assess the capability of
current technologies to characterize ultrafine particles.

4.4.1  Limitations of the Present NERL SEM
        Facilities
The LEO S440 high-resolution SEM has a practical minimum
size limitation of approximately 0.1 /j,m, the upper size limit for
ultrafine particles. This limit is illustrated in the photomicrograph
in Figure 4-3, which was obtained with SEM operating conditions
optimized for image resolution: The working distance was set to
8 mm, the beam energy was set to 30 kV, and the probe current
was minimized at 1 picoamp (pA). The electron source  was a
lanthanum hexaboride crystal (LaEy operated near full satura-
tion. (The S440 can also be operated with a tungsten  gun, but
LaB6 is a brighter source and provides superiorimage resolution.)
Figure 4-3 shows the image of a particle approximately 0.1 ,wm
in size sitting on a polished carbon planchet. The feature is poorly
resolved and devoid of morphological information. Although the
feature could be sized manually, the accuracy wouldbe limited to
perhaps 20% due to the "fuzzy" edges. Contrast differences be-
tween particles and the background are critical to identifying and
sizing particles  by automated techniques, which would be
essential if such analyses  are to be carried out routinely.  But
feature contrast degrades as the feature size approaches the pract-
ical limit. The poor contrast observed in Figure 4-3 would not
allow for accurate counting and sizing of ultrafines in the auto-
mated mode with the present instrument.
                                                          34

-------
Figure 4-3. Photomicrograph taken with LEO S440 SEM. The particle
approximates the minimum feature size that can be analyzed with the
present SEM.
4.4.2  Other Microscopic Techniques
There are two microscopy alternatives available for sizing and
counting ultrafine particles: (1) field-emission SEM (FE-SEM),
and (2) transmission  electron microscopy  (TEM). The two
systems are similarly priced at about $330K. FE-SEMs have
excellent imaging resolution due to the field-emission source and
are capable of identifying and sizing particles down to approxi-
mately 0.02 ,wm; smaller particles should be imaged with a TEM.
Figures 4-4a and 4-4b are photomicrographs taken with different
FE-SEM instruments.  Figure 4-4a was acquired with a JEOL
Model JMS-6340F FE-SEM and shows gold particles on a
carbon substrate. The advertised resolution of this instrument is
1.2 nm at 15 kV. Figure 4-4b was acquired with a LEO 982 FE-
SEM and shows TiO2 particles (the smallest of which is approxi-
mately 0.07/-OI1). Note that excellent resolution is  obtained at
only 3 kV.
     TEM is probably a more robust imaging method for the
ultrafine size regime (Steel, 2001). The choice between TEM
and FE-SEM may depend on whether or not particle chemistry
is required. Typically, X-ray analysis requires electron probe
energies of at least 15  kV.  At these  energies, the  effective
excitation volume from which X-rays are detected is on the order
of 1 jwm3, a volume thousands of times larger than the volume of
an ultrafine particle. For FE-SEM, in which particles orair filters
are analyzed on stubs just as in conventional SEM, EDX analy-
sis of ultrafines would be meaningless since virtually all X-rays
would be  generated by  background. If particle chemistry is
required, then a TEM  operated in the scanning mode (STEM)
with an X-ray detector (typically called an analytical electron
microscope or AEM)  is the  best and perhaps only choice of
instrument. In the transmission mode, X-ray analysis can be
meaningful since very few X-rays are produced by background.
The major disadvantage  of TEM is that sample preparation is
Figure 4-4a. FE-SEM image of gold particles on a carbon substrate. The
image was acquired with a JEOL JMS-6340F  FE-SEM.  (Courtesy of
JEOL, Inc., Peabody, MA).
Figure 4-4b. TiO2 particles imaged by a LEO 982 FE-SEM operated at
3 kV. (Courtesy of Rick McGill, Eastman Chemical Co., Kingsport, TN).
considerably  more difficult and time-consuming. In order to
analyze a section of an air filter, for example, the filter substrate
would have to be dissolved and the particles  redeposited on a
thin film supported by a TEM grid. This raises concerns about
the potential for altering the particle chemistry or morphology.
TEMs and FE-SEMs  can have automated  feature analysis
capabilities, but feature contrast may be a problem, particularly
for FE-SEMs, depending on the particle and substrate type.
     Related to the analysis of ultrafines is the question of how
ultrafine samples should be collected for either FE-SEM or
TEM. Polycarbonate screen membranes with 0.1-^mpores have
essentially unity collection efficiency for particles at least down
                                                         35

-------
to 0.035 jum. One suggestion is to use a micro-orifice impactor
with Al-foil impactor substrates for FE-SEM; for TEM, collect
ultrafines directly onto a TEM grid mounted in the micro-orifice
impactor (Wight, 2001).
     The use of atomic force microscopy (AFM) to image
ultrafine particles has been investigated in several laboratories.
Researchers  have found that AFM is limited as a tool for
imaging ultrafines by artifacts due to convolution of the tip
shape with surface topography (Van Cleef et al, 1996). Using
both high-resolution SEM and AFM, Wurster and Ocker (1993)
analyzed 0.040-^m indium particles evaporated onto a vitreous
carbon substrate. The authors concluded that AFM images have
to be carefully interpreted because the unprocessed image rep-
resents an interaction of the scanning tip shape with the real
sample topography. At this time, it appears that AFM is not a
viable alternative to FE-SEM or TEM.

4.4.3  Non-Microscopic Techniques
Non-microscopic approaches for determining size distributions
and concentrations of ultrafine particles directly in real-time
include the differential mobility particle sizer (DMPS) and the
scanning mobility particle sizer (SMPS). Both are capable of
measuring airborne particle size distributions in the submicro-
meter range. The DMPS has a particle size range of approxi-
mately 0.01 to 1 ,um, while the SMPS has a size range of 0.005
to 1  jum. Both the DMPS and the SMPS systems include a
condensation particle counter (CPC) that determines particle
concentrations in each size bin. Unfortunately, the particle size
measured by such devices are given in terms of the "mobility
equivalent diameter" rather than aerodynamic diameter, which
is more relevant. The Electrical Low Pressure Impactor (ELPI,
Marjamcki et al., 1997) is a new instrument that combines a
conventional low-pressure cascade impactor for aerodynamic
particle size classification and an electrical detection system for
aerosol concentration determination. The size resolution of the
ELPI is, however, limited by the design of the cascade impactor.
The  impactor used by Marjamcki et al. has  12 stages evenly
spaced between 0.03 ^m and 10 ,wm.

4.4.4  Ultrafine Summary
The NERL SEM/EDX laboratory is presently equipped to rou-
tinely count and size particles down to a few tenths of a micro-
meter in size. More powerful tools are needed to characterize
ultrafine particles. If particle  chemistry is not  required and
particles smaller than approximately 0.02 ^m are not of primary
interest, then the FE-SEM is probably the instrument of choice.
Alternatively, if particle chemistry is  required in  addition to
counting and sizing and if particles smaller than 0.02 ^m are of
interest, then an AEM is recommended with the added caution
that there are concerns about sample preparation. If ultrafine size
distributions and particle concentrations are of primary interest,
these can be measured in real time using differential or scanning
mobility particle sizers.
4.5 Carbonaceous and Submicrometer
     Particles
Conventional wisdom holds that EDX analysis of submicrometer
particles is difficult and unreliable. The problem is that scat-
tering of the primary electron beam within the sample substrate
generates EDX signal from an effective volume that can be
substantially larger than the feature being analyzed. The result-
ing signal-to-background ratio is poor, with the signal generated
by  the particle often overwhelmed by background generated
from the underlying and surrounding substrate. Furthermore,
substrates  of choice for aerosol  field samples are  typically
carbon based such as  polycarbonate films. Thus, SEM/EDX
analysis of small carbonaceous particles is especially proble-
matic due to the inability to resolve the EDX spectra of the
particle and substrate, and also because of poor image contrast.
A recent journal article (Laskin and Cowin, 2001) describes
progress in automated single-particle SEM/EDX analysis of
particles as small as 0.1 ^m. The fundamental innovation used
by  the authors is to use extremely thin carbon films (com-
mercially available on TEM grids) as particle substrates. The
films are on the order of 25 nm thick and are almost transparent
to a 20-keV electron beam. The vast majority of the  electrons
penetrate the substrate with very little electron scatter, hence,
very little EDX background. In this way, Laskin and Cowin have
been able to obtain semi-quantitative EDX analysis of submicro-
meter, low-Z particles  in automated SEM/EDX analysis. This
work  has  potentially  important  implications for computer-
controlled SEM/EDX analysis of aerosols, especially  PM2 5.
     Experiments were conducted to verify the basic elements
of Laskin and Co win's work. Submicrometer  particles were
loaded onto TEM support films similar to those used by Laskin
and Cowin and carried out manual SEM/EDX analysis on the
particles. Figure 4-5 shows a comparison of the EDX signals
generated during 100-s analyses by our normal polycarbonate
filters (in black) and by a 25-nm carbon-coated formvar film (in
white). The latter is supported on a standard 3.08-mm copper
TEM grid. Note that the vertical scale is logarithmic so that the
true reduction in background afforded by the formvar film is
much greater than the apparent  reduction. For carbon, the
dominant species in both substrates, the reduction is  100-fold.
This low background  signal enables  semi-quantitative EDX
analysis of small and low-Z particles. The formvar film shows
small Cu and Al peaks in addition to C. (The Cu is due to
scattering of secondary electrons from the Cu support grid.) A
key factor  in the low  count rate generated by the formvar is
geometry:  After penetrating the formvar film, the probe beam
travels through open space for several inches before hitting the
bottom of the SEM sample chamber. There is a resulting small
solid angle for X-rays produced here to reach the secondary
detector,  and the  sample  wheel  that holds the TEM  grid
intercepts any line-of-sight X-rays produced at this location.
     Figure 4-6 is a photomicrograph of a submicrometer diesel
soot particle taken with the LEO S440 in the secondary electron
                                                        36

-------
Figure 4-5. EDX spectra for blank standard polycarbonate filter (black)
versus 25-nm carbon-coated formvar film (white).  Both spectra were
acquired for 100 s.
Figure 4-6. Photomicrograph of sub-micrometer diesel soot particle.


mode. The support substrate is a 25-nm carbon-coated formvar
film supported on a copper TEM grid. The carbon coating im-
proves the stability of the formvar to localized heating from the
electron beam. The formvar film provides uniform contrast and
a featureless, optically flat background ideal for imaging. (It
should be noted that the image was acquired under conditions
optimized for imaging, e.g., low probe current, short working
distance, and high pixel resolution. The image would be notice-
ably degraded  using SEM conditions  more appropriate for
computer-controlled SEM/EDX.)
     Figure 4-7 shows the EDX spectrum (in black) of the soot
particle shown in Figure 4-6. The spectrum is dominated by C,
as expected, but also shows  Al, Si, Cu, and O. Superimposed on
the spectrum (in white) is the approximate contribution from the
substrate, measured by moving the electron beam to a nearby
point off the particle. Both spectra were collected for 100 s at a
probe current of 200 pA. The carbon peak dominating the soot
particle spectrum represents a count rate of 17.3 cps, of which
Figure 4-7. EDX spectrum of diesel soot particle pi us formvar substrate
(black). White spectrum is the formvar blank for the same counting time.
the formvar substrate accounts for about 1.5 cps. For computer-
controlled SEM/EDX analyses, a more realistic X-ray acquisi-
tion time would be 10 s or less.
     Figure 4-8 compares EDX spectra acquired for 10 s on (1)
a 0.22-um aluminum silicate particle (Arizona Road Dust) on
standard polycarbonate  substrate  (top),  and  (2) a  0.24-um
aluminum silicate particle on the formvar film. In the top spec-
trum, carbon background from the polycarbonate filter, peaking
off-scale at 223  counts full scale, dwarfs the signal from the
mineral particle.  In the bottom spectrum, the signal is generated
predominantly by the particle; the small carbon and Cu peaks
originate primarily from the substrate and the TEM grid,
respectively. Compared to the polycarbonate filter, the formvar
film generates much less Bremsstrahlung radiation, which gives
rise  to the background  continuum underlying  the  discrete,
characteristic X-ray peaks. In the bottom spectrum, the Si and Al
peak areas are 197 and 96 counts, respectively. The net K and Fe
peak areas are only 6 and 25 counts, respectively, but could be
used in a qualitative way to assign particles to specific classes.
We believe that the probe current could probably be doubled or
        T
 Li
Figure 4-8. Top: 10-s EDX spectrum of a 0.22-um aluminum silicate
particle (Arizona Road Dust) on polycarbonate substrate. Bottom:: 10-s
EDX spectrum of a 0.24-um aluminum silicate particle on the formvar film.
                                                          37

-------
tripled to improve counting statistics without degrading signal-
to-noise too much.
     The analyses above were painstakingly carried out manually
on the LEO/PGT system. It is unlikely that the same data could be
acquired with this system in a computer-controlled analysis be-
cause of the  difficulty of reliably identifying submicrometer,
low-Z features automatically due to limited image contrast and
image resolution. Laskin and Cowin have the benefit of ad-
vanced  SEM  technology,  including a field-emission  SEM
(roughly lOOx brighter than our LaB6 gun, hence, potentially
better image resolution and X-ray count rate) and an in-line
secondary electron detector whose signal canbe mixed with that
of the backscatter detector to optimize feature-to-background
contrast. Evenif the analysis limitations could be overcome, one
must be able to deploy these fragile TEM substrates in the field
as active and/or passive aerosol collectors. Laskin and Cowin
report to have developed size-selective active and passive sam-
pling devices using TEM grids,  so that it appears that the major
obstacles canbe overcome, enabling the automated, semi-quan-
titative characterization of particles down to 0.1 /j,m.

4.6 CCSEM Data Quality and Validity
Studies have been conducted in the NERL SEM Laboratory and
elsewhere to assess the quality and validity of CCSEM data.
Issues relating to CCSEM data quality and validity include: the
stability of unattended CCSEM for multi-hour runs; the preci-
sion of CCSEM analyses; the number of particles that must be
analyzed in order to yield representative results; the sensitivity
of CCSEM results to instrument settings such as video threshold,
dwell time, grid point spacing, and X-ray counting time; and
errors generated by automated feature analysis software.
     Germani (1991) evaluated the effects  of critical instru-
mental parameters on the analysis time and  accuracy of auto-
mated gunshot residue (GSR) analysis. GSR analysis is similar
in many ways to CCSEM analysis of aerosol samples. Germani
carried out five experiments, in each of which, one parameter
was varied while the others remained the same. The instrumental
parameters were:  minimum particle size, video threshold, elec-
tron beam point spacing, video dwell time, X-ray counting time,
and X-ray analysis mode. For each experiment, successive auto-
mated analyses were run over the same area of the sample.
Results of these experiments are summarized below:

     1.  Changing the video threshold resulted  in different
        particle types, sizes, and morphologies being selected
        for analysis. A particle's BSE signal is a function not
        only of the effective Z  of the particle, but also the
        particle's size and morphology: smaller and flatter or
        thinner particles generate smaller BSE signals than
        larger, more spherically shaped particles of the  same
        effective Z.
     2.  The  number of true GSR particles found by wave-
        length-dispersive analysis (WDX) was always greater
        than that for EDX analysis, by as much as a factor of 5
        at the lowest threshold.
     3.  Dwell time is the amount of time that the electron beam
        is stationary at a point while the BSE signal is acquired
        and averaged. Varying dwell time from 2 to 256 /us
        with fixed video threshold had no significant effect on
        the number of GSR particles detected per field. How-
        ever, dwell time does directly affect the analysis time.
        Again, the number detected by WDX was  2-3 times
        greater than that for EDX.
     4.  Digital point spacing is the distance between  grid
        points in the digital raster used to search for particles.
        A "coarse" grid reduces analysis time, but at the risk of
        missing particles whose size is smaller or on the order
        of the digital point spacing. Germani found that a grid
        point spacing of about one half the minimum specified
        particle size was needed. Reducing the grid point spac-
        ing by another factor of 2 (to 0.25 /j,m) increased the
        number of detected GSR particles per field by 40%, but
        the analysis time increased by at least a factor of four.
        Interestingly, the ability of EDX to identify GSR parti-
        cles improved (using WDX results as the reference) as
        the grid point spacing increased, indicating that EDX
        performs better on larger particles.
     5.  Results of the "point" mode versus "raster" mode of
        analysis showed no significant difference between the
        two modes in the ability to detect GSR particles, con-
        sistent with a large X-ray emission volume for GSR
        particles at an accelerating voltage of 20 keV. How-
        ever, the point mode typically yielded three times the
        count rate of the raster mode, and so is the better choice
        for GSR analysis.
     6.  The effect of X-ray counting time was also tested. The
        number of GSR particles detected per field doubled
        with a nine-fold increase in counting time (from 3 s to
        27 s), due to the improved signal-to-noise.

Results from Germani's study indicate that trade-offs have to be
made between analysis time and accurate particle detection.
     Studies were also conducted in the NERL SEM Laboratory
to assess the following issues related to the quality and validity
of CCSEM data: (1) the stability of unattended CCSEM for
multi-hour runs, (2) the number of particles that must be ana-
lyzed in order to yield representative results, and  (3) errors
generated by automated feature analysis software. These studies
are described in detail in Mamane et al. (2001). A summary of
results is presented below.
     Stable operation of the SEM instrument during a multi-
hour CCSEM analysis of a sample is an essential requirement.
Instrumental drift in key parameters such as magnification,
electron probe current, threshold setting, or X-ray energy cali-
bration could seriously compromise the validity of the data.
                                                         38

-------
4.6.1  Precision of CCSEM (Repeat Analyses
        of Same Sample)
The precision and stability over time of computer controlled
SEM was evaluated using an outdoor ambient sample collected
from the 1998 Baltimore Retirement Home Study (Conner etal.,
2001). The sample, which was originally analyzed by CCSEM
on  3/26/99,  was  re-analyzed in triplicate 21  months  (on
12/26/00,  12/27/00, and 12/28/00) after the original  CCSEM
analysis was conducted. Every effort was made to reproduce the
original analysis conditions. The same parameter and other files
used in the original CCSEM analysis were used for the replicate
analyses. Other conditions were not as straightforward to dupli-
cate exactly. For instance, the filament condition cannot be
exactly duplicated. The brightness and contrast of the sample,
and thus the analysis thresholds, are dependent on the filament
brightness (beam current), which varies at the saturation current
depending on the age of the filament and other factors. Hard-
copy thermal print images of the original analysis threshold set-
up  were used as a guide for setting up the replicate analysis
thresholds.
     Results for each analysis were summarized using the same
set of user-defined rules that had been developed for the original
study. For the purposes of this analysis, it was assumed that the
analyses conducted on the three consecutive days (12/26-28/00)
were essentially concurrent. The results of these nearly concur-
rent analyses were used to calculate the repeatability  (precision)
of CCSEM for an urban ambient sample. For particle classes
assigned 100 or more particles, the CCSEM precision was 10%
or less. For most other categories, the precision was 26% or less.
However, precision deteriorated significantly for particle classes
containing an average of 15  or less particles (the so-called
"needle-in-a-haystack" particle types).
     Figures 4-9a-d, show the particle summaries forthese three
analyses plotted as the average, standard error, and  standard
deviation, compared with the results for the original analysis
conducted  21 months earlier. For  most particle  types, the
original analysis was within one standard deviation  of the later
   10000

    9000

 0  8000
 5T
 £  ™°°
 Q_
 _0  6000

 "ro  sooo
 Q.
 
-------
replicate analyses. For the higher abundance  particle types
(Figures 4-9a and b), those whose original analysis did not fall
within a standard deviation of the repeat  analysis were the
generalized particle types characterized only by number of X-ray
counts (not by chemistry.)
     Germani and Buseck (1991) evaluated the precision and
accuracy of mineral classifications by automated SEM. The
sample was prepared from particle dispersions of USGS stan-
dard rock powders.  Mineral types were  identified by cluster
analysis. Typical precisions of ± 10% for major mineral types
and ± 20%  for minor mineral types were determined from
replicate analyses.
     The results of a chemical mass balance between individual
particle analyses and bulk chemical compositions of the standard
rock powders indicate a precision of better than ± 15% for most
elements. The accuracy of the analyses,  when  normalized to
SiO2,  ranged from 1% to more than a factor  of 3. Large dis-
crepancies for some elements are due to error in particle volume
calculations  and problems associated with analyzing minor
mineral types.
     The precision of CCSEM-determined filter mass loadings
was  previously discussed in Section 4.3. The  sample was a
3 l,wg  loading of Arizona Road Dust on a polycarbonate filter.
Three repeat analyses of the same fields (227 fields, 6000 parti-
cles) yielded a relative  standard deviation (rsd) in the calculated
mass  loading of 3.5%. Eight successive  analyses of random
fields (6000 particles, ~ 110 fields) yielded an rsd of 11%.

4.6.2  Representativeness of Data
Sampling error  is also a concern for both manual  SEM and
CCSEM: typically, only a very small fraction  (less than 0.1%)
of the particles on an ambient filter is characterized by SEM or
CCSEM. This leads to  the question of how many particles must
be analyzed on a sample in order to obtain representative results.
     CCSEM was applied to the analysis of a coarse-fraction
(PM10_2 5) 24-h ambient particle sample collected in Baltimore
that was known to be impacted by natural and industrial sources.
The sample  was collected with a  dichotomous sampler on a
polycarbonate filter. The sample was fairly  typical of ambient
urban  aerosol samples in terms  of the variety of particle
types represented. Results show that the number of particles per
field of view for this sample stabilized within  a few percent of
its final value of 36.1 particles per field (based on the analysis of
all 78 fields) after the analysis of only 360 particles.  Similarly,
the cumulative averages for average particle  diameter and aver-
age particle mass loading show convergence to the final values
after about 360 particles.
     Cumulative averages for particle composition were exam-
ined to determine for what number of particles  the average
particle composition converges to the final composition of the
sample. Comparing  chemical properties  of particles is more
complex than comparing the physical properties discussed above
since  the composition of each particle is characterized by 20
elemental concentrations plus the total X-ray count. Results for
the major and minor elements stabilized only after analyzing
about 1000 particles. However, the average particle composition
determined from the first few hundred particles approximates the
final composition for most purposes.
     In the second step of the comparison, particles in each of
four nested subsets were independently sorted into 25 particle
types or  classes according to their elemental concentrations.
(The subsets  comprised particles 1-360, 1-734, 1-1456, and
1-2819; thus, each smaller subset of particles is contained within
the next larger subset.)
     Assuming that the full 2819-particle data set is representa-
tive of the sample, then  deviations in  particle classification
results of the  three smaller subsets from the full data set allow
one to estimate the minimum number of particles needed to
characterize an urban aerosol sample. For the five major cate-
gories, the average relative error decreased from 11.3% (N =
360) to 7.8% (N = 734) to 6.0% (N = 1456), showing a decrease
with increasing number of particles analyzed.For all but one of
the major categories the abundances converged to within a few
percent of the true values for N between 734 and 1456. For
many purposes, even 360 particles may adequately characterize
major class abundances in the sample.
     As the number of particles assigned to a class becomes
small, the relative errors become large and variable due to statist-
ical fluctuations. Abundances for three of the minor classes are
far from convergence to the true abundances even for N = 1456.
However, these classes are  populated by only  17, 22, and 3
particles, respectively, in this subset. In the full data set of 2819
particles, the minor classes are populated by only 42,36,33,20,
12, 11, 9, 4, and 2 particles, respectively. Here, the concept of
"true" class abundances becomes increasingly untenable, even
with 2819 particles. Results show that the average relative error
for the minor  classes improves from 66.6% (N = 360) to 31.5%
(N = 734) to  12.6% (N = 1456). For some PM studies, the
classes representing industrial particles (typically minor classes)
may be of most interest from the human health perspective.  If
one needs to accurately quantify the abundance of a minor class,
then one  must increase the total number of particles analyzed.
(Alternatively, the PSEM is able to conduct dedicated searches
for specific types of particles. In this mode, particles of no
interest are bypassed in the analysis. This greatly enhances the
efficiency of searching for rare or exotic particles.) The experi-
ments above show that the physical properties of the sample, as
represented by the number of particles per field, average particle
diameter  and particle mass loading per field, are well charac-
terized by analyzing approximately 360 particles. Little  addi-
tional information is gained by analyzing more particles. Chemi-
cal properties of the sample (average elemental composition and
major chemical class abundances) converged to within a few
percent of their final values after analyzing about 1000 particles.
Again, for many purposes several hundred particles may provide
adequate  characterization. Convergence of minor class (<1.5%
by number) abundances was limited by statistical fluctuations as
the number of particles populating a class became very small.
                                                         40

-------
4.6.3  Errors Associated with CCSEM
Automated particle recognition algorithms lack the sophisti-
cation of the trained human eye; errors can be made by CCSEM
that are typically avoided by an experienced operator performing
manual analysis. Such errors may include incorrect sizing of
complex particles; incorrect X-ray analysis, especially of aggre-
gate particles or organic and carbonaceous particles, leading to
particle misclassification; missed particles (e.g.,  due  to poor
contrast); and analysis of nonexistentfeatures (contrast artifacts).
Errors associated with CCSEM generally fall into two cate-
gories: those that result from the use of an automated threshold
and those associated with excessive particle loading on the filter.
The former are intrinsic to CCSEM and occur because current
automated particle recognition algorithms lack the human eye's
ability to interpret features in an image. Excessive particle load-
ing results in errors due to overlapping particles, which can be
minimized by restricting the mass loading to less than approxi-
mately 30 /j,g/cm2 for PM10_2 5. Specific types of errors  that can
be expected when using CCSEM include the following:

     •  Missed features. These are bonafide particles that are
        not identified by the CCSEM software and are there-
        fore not characterized. Most often these are low-con-
        trast, submicrometer organic or carbonaceous or other
        low-Z particles  such as sulfates that  can be identified
        by eye in the secondary electron mode but are below
        threshold in the backscatter mode used in most CCSEM
        analyses. Researchers should be aware that CCSEM
        analyses may underreport data  on small,  low-Z parti-
        cles  such as sulfates and carbonaceous particles. This
        problem is particularly troublesome forPM25 samples,
        which are typically dominated by such particles and
        may require that CCSEM be used in combination with
        other techniques for an accurate characterization of
        PM25.
     •  Overlapping particles. The computer is not able to dis-
        criminate touching, overlapping, or agglomerated parti-
        cles;  these will be analyzed as one large particle.
        Although the X-ray analysis may be valid for one of
        the overlapping features, that feature will be oversized
        while the other overlapping particle(s) will not be ana-
        lyzed at all. A key to minimizing the occurrence of
        overlapping particles is to avoid overloading the filter
        during sample collection.
     •  Contrast artifacts. Contiguous pixels whose video level
        exceeds the threshold setting are assumed to be real
        features by CCSEM. Occasionally, the substrate pro-
        duces contrast changes that mimic real particles, and
        these artifacts will be  analyzed as  real  particles by
        CCSEM. X-ray analyses of contrast artifacts just yield
        spectra  of the  filter substrate, and, in  the  case  of
        polycarbonate substrates, these may be indistinguish-
        able from the spectra produced by  carbonaceous or
        organic  particles. If images have been saved for all
        features, the  contrast artifacts can be  identified by
        reviewing the images of low-count features.
     •  Sizing errors. Particles can be incorrectly sized when
        particle contrast is at or near the video threshold, as is
        frequently the case, for example, for low-Z particles
        such as sulfates, pollens and spores, skin flakes, and
        carbonaceous particles. Poor contrast may also occur at
        the edges of a particle where it may be thinner. Sizing
        errors also occur for particles with complex morph-
        ology or with mixed composition. In the case of the
        latter, differences in composition are reflected in dif-
        ferent grayscale values in the backscatter image. If the
        grayscale of some portion of a feature drops below the
        threshold, the particle may be undersized.
     •  Heterogeneous particles. Particles of mixed composi-
        tion cannot be accurately characterized by acquiring a
        single-spot X-ray analysis at the particle's center of
        mass. Although the X-ray analysis of a single hetero-
        geneous particle may not be particularly meaningful,
        the results obtained from analyzing many such particles
        may  provide useful  information about the  average
        composition of the sample.

     Error rates associated with the Baltimore CCSEM analysis
were estimated by reviewing saved images, X-ray spectra, and
associated field images for 800 randomly selected particles from
the Baltimore data set. In the 22 field images reviewed, we
identified 47 particles that were present in the BSE field images
but were not processed in the CCSEM analysis. Most  of these
missing particles were associated with overlapping particles. The
remaining particles were inexplicably  bypassed during the
CCSEM run. This number does not include a small but unknown
number of low-Z particles that may have been visible in SE field
images but were below the detection threshold in the BSE field
images. Review of the particle images also revealed that approx-
imately 23 (2.7%)  of the identified "particles" were  actually
contrast artifacts, leaving a total of 824 real particles in the data
set. Of these, approximately  12% were judged during manual
review of the images to  have been sized incorrectly. Sizing
errors were generally the result of overlapping particles or poor
contrast (usually small, low-Z or carbonaceous particles). Sizing
errors, however, were both positive (oversized)  and negative
(undersized) and tended to cancel  each other out so  that the
average sizing error computed from all 96 particles was only
about  3%  (statistically indistinguishable  from  zero error).
Twenty-nine  features were analyzed and sized  as individual
particles, but were  actually part of an overlapping pair or mul-
tiplet of particles.  These contributed both to the sizing and
missing-particle errors already counted.
     The error rates reported above are sample dependent, being
in part a function  of particle loading and particle chemistry.
Nevertheless, they may reflect typical CCSEM  errors for an
aerosol sample that is moderately loaded on a polycarbonate
filter. Most of the  errors (those due to particle  sizing, over-
lapping particles, and contrast artifacts) can be corrected or
                                                         41

-------
eliminated during off-line review of the data. The ZepView
software allows the user to review all stored field and particle
images  and  spectra along with particle-by-particle  size and
chemistry data. Artifacts can be culled from the data set and
features can be resized provided that the particle images were
saved. As a  general practice, we strongly recommend off-line
review of CCSEM data both to improve data quality and also
because manual review enhances a researcher's understanding
of the sample.
     The above results need to be qualified by the following
remarks. Most air monitoring studies employ Teflon filters for
collecting coarse ambient aerosol rather than the polycarbonate
filter analyzed in this study. Teflon filters, however, are not well
suited for CCSEM. Also, the coarse mass concentration (5.6
Mg/m3)  for the chosen Baltimore sample is roughly half of
typical urban coarse mass concentrations; a more typical coarse-
fraction urban sample would be  expected to show increased
CCSEM error rates associated with overlapping particles com-
pared to our sample. Finally, this study was confined to a coarse
aerosol sample. Fine PM poses a number of analytical challenges
for CCSEM  that were beyond the scope of the present study.
     In conclusion, although CCSEM makes occasional errors
in recognizing features and in sizing and classifying particles,
these errors  are, in most cases, an acceptable trade-off for the
higher throughput, improved particle statistics, and improved
objectivity afforded by  CCSEM compared to manual SEM
analysis. Provided that  field and particle images have been
saved, most  errors can be eliminated during off-line review of
the data.

4.6.4  EDX Acquisition Time
The PSEM provides several options for setting the X-ray acqui-
sition time. The simplest  option is to terminate EDX acquisition
on all particles after a fixed number of seconds (Norm EDX
time). A typical value for the Norm EDX time is 3 s. A second
option is to terminate EDX acquisition when the net X-ray
counts for any single element reaches a target number of counts
(Trg Counts). According to the ZepRun Users Manual, "As a
general  principle,  Trg Counts is a better parameter to  use to
control X-ray time, because it will allow acquisition time to be
tailored to the information present in the  spectrum—'weak'
spectra  will  be acquired for longer than strong spectra. This
helps to effectively trade  off speed for precision; more time will
be spent where needed to establish spectral statistics; less time
where the spectrum is strong and well defined. It also helps by
making the statistical content of all spectra more uniform relative
to one another." Typical values for Trg Counts may be  on the
order of 3000 counts. Yet another option for controlling X-ray
acquisition is to extend the acquisition time (Max EDX time) for
particles of special interest while applying the Trg Counts and
Norm EDX limits to all other particles. Specific particle types,
defined by the user in a  ZepRun Rule File, can be selected to
receive the Max EDX time.
     The EDX acquisition time impacts the quality  of SEM/
EDX results. Although very short X-ray acquisition times can
boost CCSEM throughput to several thousand particles per hour,
the data quality may suffer. For example, repeated analyses of
the same  particle  could show significant variability  in  the
elemental percentages, small peaks could be missed, and noise
could be falsely identified as  a real peak. These can cause
particles to be misclassified when summarizing CCSEM results.
     A series of tests were conducted on the PSEM to determine
the effect of EDX X-ray collection time on sample composition
and particle classification. The  X-ray collection time is one of
the parameters the user sets in a computer-controlled SEM/EDX
analysis. The X-ray collection time is typically a major determi-
nant of the total sample analysis  time. Ideally, therefore, the user
will choose the minimum X-ray collection time that will still
allow accurate determination of each particle's composition. In
aerosol  sample analyses conducted at the  NERL  SEM/EDX
laboratory, EDX analyses are not meant to be quantitative on an
absolute concentration  basis. Instead, samples are generally
characterized in terms of number or weight percent of particle
classes (e.g., aluminum silicate,  quartz, etc.,), which are derived
from the qualitative EDX analysis of a representative number of
particles in the sample. We repeatedly analyzed identical fields
of particles using different EDX collection times in an attempt
to determine the minimum EDX time in which the resulting
particle  classes are not significantly different from the "true"
particle  classes of the sample. "Truth" in these studies  was
assumed to be determined by  the results based on an EDX
collection time of 25 seconds.
     The  test sample  for  these  tests  was  NIST Standard
Reference  Material 2710 (Montana Soil,  Highly Elevated
Traces). SRM material was suspended in acetone and deposited
by vacuum filtration onto a polycarbonate filter (12-^m pore).
Three repeat analyses of the same fields were conducted at each
of five different EDX collection times (Norm EDX = 25, 15, 8,
3, and 1 s). The Trg Counts parameter and the Max EDX para-
meter were not used in the following tests. Instrumental parame-
ters for the 15 CCSEM analyses were set up to ideally analyze
the same particles in each run, although this was only partially
successful. Only particles in the size range  of 10-40 ^m were
analyzed. X-ray analyses were carried out in the raster mode (X-
ray s collected by rastering the electron beam over the projected
area of each particle). In addition to X-ray analysis, photomicro-
graphs were  collected  for each particle. The  total CCSEM
analysis time is dominated by the X-ray counting time for collec-
tion times of 3 s or greater, but at shorter EDX times other tasks
such as  moving the stage and collecting micrographs, assume
competing importance in determining overall sample analysis
time. The number of particles analyzed in each of the 15 repeat
runs ranged from 522 to 590, and the total analysis times ranged
from 0.47  h to 4.49 h. The probe spot size  was set to  the
standard 30% used in EDX analysis, and the X-ray  count rates
were typically on the order of 600 cps.
     The 15 repeat analyses provide an opportunity to assess the
precision of CCSEM analysis  as a "bulk"  analytical tool for
determining sample mass. For  each run, the SEM-determined
masses of all particles analyzed were summed. The mean and
                                                         42

-------
standard deviation for the  sample mass was calculated from
these 15 sums. The one-sigma relative standard deviation for the
15 analyses was excellent: 5.8%, even though the particles ana-
lyzed in each run were not  entirely identical. These results re-
flect the precision of CCSEM in measuring particle size and
estimating particle density  (from the EDX analysis).  As other
studies have shown,  however, the accuracy  of  determining
sample mass by  CCSEM  is relatively poor  because of the
inability to accurately measure particle volume and density (see
Sections 4.2 and 4.3). Of the particles analyzed in each run, 372
were found in  all 15 analyses  (based on examination of the
individual micrographs). The precision of individual particle
mass measurements was determined for each of these 372 parti-
cles . The average relative standard deviation (RSD)  in individual
particle mass was 16%. Variability in particle mass is primarily
due to variability in the measured volume. Examination of the
micrographs of particles having very large standard
deviations (>100%) in mass do not reveal anything unusual.
     The effect of EDX collection time on particle chemistry is
shown in Table 4-2. Starting with the 372 particles found in all
analyses, N is the  number of particles for which the  given
element was  detected in all  three  25-s  analyses. These N
particles were subsequently analyzed (three repeat analyses
each) at EDX times of 15, 8, 3, and 1 s to generate the results
presented in Table  4-2. For example, 19 particles were Na-
bearing in all three 25-s analyses. The average measured Na
concentration in these 19  particles  is 2.6 ± 0.26 wt%. The
concentrations reported in Table 4-2 are not quantitative on an
absolute basis and  are not comparable to  the NIST-certified
"bulk" concentrations. For a single analysis, an element's wt%
value is calculated by estimating the element's mass in each of
the N particles, summing these individual particle contributions,
and dividing by the total mass of all N particles.  [The mass of
each particle is estimated by the product of the particle's volume
and density (density is calculated from the X-ray spectrum), and
the mass of a specific element in a particle is the particle mass
times the element' s fractional contribution to the particle' s X-ray
spectrum; no corrections are made for detector efficiency,  X-ray
fluorescence yields, or size-dependent X-ray absorption effects].
The Avg wt% values in Table 4-2 are the wt% values averaged
Table 4-2. EDX Collection Time versus CCSEM Particle Composition (in weight percent concentration)

Na
Mg
Al
Si
P
S
Cl
K
Ca
Ti
Cr
Mn
Fe
Cu
Zn
As
N
19
84
369
372
4
33
7
308
211
87
1
68
301
73
28
200
25s
Avg wt% Std. dev
2.6 0.26
5.0 0.08
22.6 0.39
53.6 0.42
1.8 0.03
1.8 0.07
12.5 0.63
10.1 0.08
2.7 0.11
2.8 0.07
21.0 0.00
3.2 0.09
8.1 0.08
1.7 0.11
1.3 0.01
1.5 0.07
15s
Avg wt% Std. dev
2.2 0.41
4.9 0.05
22.7 0.47
53.1 0.48
1.6 0.57
1.3 0.29
12.2 0.62
10.2 0.18
2.5 0.17
2.6 0.01
21.0 2.00
2.8 0.08
8.0 0.12
1.5 0.19
1.0 0.43
1.5 0.24
8s
Avg wt% Std. dev
2.0 0.38
4.7 0.38
22.8 0.42
53.1 0.04
1.6 0.48
1.5 0.13
12.2 1.50
9.7 0.20
2.6 0.13
2.8 0.13
21.7 0.58
2.9 0.21
7.9 0.20
1.5 0.16
1.4 0.38
1.7 0.11
3s
Avg wt% Std. dev
2.1 0.22
4.8 0.39
22.4 0.63
52.0 1 .00
1.2 0.75
1.2 0.04
11.9 0.82
9.9 0.46
2.4 0.15
2.7 0.34
21.0 2.65
2.8 0.34
7.8 0.33
1.2 0.17
1.2 0.11
1.6 0.35
1 s
Avg wt% Std. dev
2.2 0.07
4.6 0.69
21.8 1.48
50.8 1.04
1.5 1.25
2.0 0.41
11.2 1.66
9.3 0.32
2.6 0.17
2.9 0.26
19.0 4.00
2.7 0.20
7.9 0.55
1.6 0.20
1.0 0.53
2.1 0.40
 N = number of particles for which the given element was detected in all three 25-s analyses. Only these N particles were analyzed at shorter EDX
 times.


 Avg wt%: Elemental weight concentrations were calculated for each individual particle and averaged over N particles to get an average wt%. The
 average weight percents from three repeat analyses of the same N particles were averaged to get the Avg wt% shown in the table.

 Std. dev = percent standard deviation in the Avg wt% for 3 repeat analyses.
                                                           43

-------
over three repeat analyses of the same N particles. We assume
that the Avg wt% listed for the 25-s analysis represents the
"true" average concentration, and we are interested in looking
for significant deviations from the truth as the EDX collection
time is decreased. The results in Table 4-2 indicate that the effect
of reducing EDX collection time is minimal for most of the
listed species, even for X-ray analysis times as short as 1 s.
     In general, variability in the measured elemental concentra-
tions increases as the X-ray collection time is reduced, and the
variability among the three repeat  analyses at a given EDX
collection time approximately equals any changes in the average
measured concentration due to reducing the EDX time. For the
most robust species (e.g., K and Si), reducing the EDX time may
result in slightly lower measured concentrations. For example,
measured Si and  K concentrations  drop  by 5.4%  and  8.6%,
respectively, when the EDX time is reduced from 25 s to 1 s.
Table 4-2 suggests a possible increase in the "As" concentration
for the shortest EDX time. The arsenic signal is actually a noise
artifact due to inadequate noise suppression in the CCSEM setup
file. (A better element setup file for As would have included the
requirement that the secondary As Ka and Kp peaks be present
above noise as a condition for As to be detected). The As results
may indicate the need for careful noise suppression for species
that are typically present near the minimum detection limits.
     As  mentioned above,  samples are  typically not charac-
terized in terms of absolute elemental weight percents, but in
terms of particle classes. Particles in the 25-s data sets were
sorted, somewhat arbitrarily, into 13 particle classes, which were
suggested by cluster analysis of the particle EDX data. (The
Montana soil sample used for this test was not ideal for particle
classification because the sample did not show a large diversity
of particle classes. The great majority of the particles were either
aluminosilicates or quartz.) A set of rules were developed using
the Zeppelin Feature Rule Editor (Aspex Instruments) to sort the
particles  hierarchically into the  13 classes. The identical set of
372 particles was classified using this fixed set of rules for each
of the 5 EDX times (15 analyses in all). Figure 4-10 shows the

s
K 2-°
o
s
at ion Rele
3 C.
^
£




TT






^





^





^





ifij

T
_L
I
&





N
-L 1

                                                   n  15 sec
                                                   A  8 sec
                                                   o  3 sec
       AISi    Quartz   HiAI.K  Fe-Mg Al  Hi Si   NaAISi   CaAISi    °  1 sec
                      Particle Class

Figure 4-10. Particle classification results on repeat CCSEM analyses of
Montana Soil SRM at five different X-ray collection times. Each data point
represents the number of particles in the given particle class relative to
the class population for the 25-sec data.
resulting average class populations relative to the average "true"
populations measured at EDX times of 25 s. The figure displays
results for only those classes having "true" populations of seven
or more particles. Classes are arranged in order of decreasing
population, with the AISi class being the most populated (aver-
age "true" N = 153) and the CaAISi class being the least popu-
lated (average "true" N =  7). Each data bar shows the average ±
standard deviation of three repeat analyses. Classification results
at EDX times of 3 s and 1 s show increasing divergence from the
"true" populations, which is attributed to a systematic increase
in the population of the NaAISi class as EDX time is  reduced.
Because each particle in all 15 data sets must be sorted into one
and only one  of the same  13 classes, any growth in the popula-
tion of one class must be accompanied by a decrease in the
population of another  class or classes. The increase in the
NaAISi class  is thus accompanied by decreasing populations in
the High Al, K, Fe-Mg Al, and High Si classes. Examination of
the EDX data shows that the number of Na-bearing particles (Na
X-ray count >1% of total X-ray count) increases substantially as
EDX  time is reduced. We believe  that  noise in the EDX
spectrum is responsible for the apparent increase in the number
of Na-bearing particles at reduced EDX times. If an element
threshold of 3% (percentage of total X-ray counts needed for the
given element to be considered as present in the particle) had
been assumed in the run parameter file rather than  1%, the
classification results would be expected to be more consistent
across EDX times.
     The tests above need to be repeated for a sample that is
truly heterogeneous in particle types. These preliminary results,
however, suggest that if only the maj or elements are considered,
particle composition and classification  results are remarkably
constant and independent of X-ray counting time for EDX times
as short as  1 s. For elements whose concentrations in  particles
are near the EDX detection limit (roughly 1 wt %), the need for
careful noise suppressionbecomes increasingly important as the
X-ray counting time is  reduced. If a minor species is critically
important to the analytical results, the user should confirm the
results obtained at short EDX times by repeating the analyses of
a subset of particles at long X-ray counting times.
     In conclusion it should be mentioned that EDX detector
technology continues to improve. New pulse-processors with
significantly higher count rate capability combined with large
area detectors, have the potential to reduce X-ray counting times
and significantly boost particle throughput such that 10,000
particles per  hour may become routine with  state-of-the-art
technology.
                                                          44

-------
                                                   Chapter 5
                                Examples of Research Applications
5.1  Examples from the Literature
Numerous applications of SEM/EDX to environmental problems
have appeared in the literature. For convenience, these examples
are sorted into papers focusing on: (1) aerosol characterization,
(2) source apportionment, or (3) SEM/EDX methodology.

5.1.1  Aerosol Characterization
In response to growing concern over potential health effects and
air quality impacts of aerosols associated with anthropogenic
activities, numerous articles  have appeared in the literature
during the past three decades  focusing on characterization of
several important classes of anthropogenic aerosols:  fly ash,
soot, and sulfate. Buseck and Bradley (1982) discuss  electron
beam studies of natural and anthropogenic microparticles and
offer the following statement: "The range of particle sizes and
compositions in aerosols is extremely wide, as is the number of
possible mineral types and synthetic compounds. .  . .Although
the majority of airborne particles  are minerals, a significant
fraction in urban areas is anthropogenic. Furthermore, in these
urban regions even the natural particles carry surficial  deposits
of an anthropogenic character,  clearly the result of the  efficient
scavenging ability of fine-grained dusts."

5.1.1.1  Fly Ash
Numerous papers describe SEM characterization of fly ash from
coal- and/or oil-fired power plants. Studies consistently show
selective  accumulation  of volatile species on the surface  of
refractory metal oxide cores and enrichment of the fine fraction,
especially submicrometer particles, by toxic elements and com-
pounds (Pueschel, 1976;  Rothenberg et al., 1980; Mamane,
1984, Mamane et al. 1986; Hock and Lichtman, 1983). Linton
et al. (1977), using ion microprobe mass spectrometry and Auger
electron spectrometry demonstrated surface predominance of S,
Fe, K, Na, Li, Pb, Tl, Mn, Cr, and V in coal fly ash. This was
explained in terms of a volatilization-condensation mechanism.
In addition, the last five of these elements were found to be
highly leachable and therefore of particular environmental and
health significance. The general consensus appears to be that
particles formed at high temperatures act as substrates for the
condensation of other materials during cooling. Coal and oil fly
ash can be quite heterogeneous in both morphology and elemen-
tal content, greatly complicating the application of SEM/EDX to
quantitative receptor modeling. Morphological analysis of a coal
fly ash sample by Fisher et al. (1978) indicated 11 major classes
of fly ash particles. Mamane etal. (1986) analyzed fly ash from
both coal-fired and oil-fired power plants by SEM/EDX and
examined the potential for  SEM/EDX as a tool for  receptor
modeling. They found that  potentially useful bulk elemental
tracers such as As and Se in coal fly ash and V and Ni in oil fly
ash were not reliably detected in individual particles by EDX.
(As and Se were not detected while V and Ni were detected in
only 50-60% of the oil fly ash particles.) In addition, most fine-
fraction oil fly  ash particles resembled the mineral-rich spheres
that  are  typically  associated with coal-fired  power plants,
making it difficult for SEM/EDX to reliably distinguish these
two source types. Obrusnik et al. (1989) used SEM and neutron
activation analysis to examine differences  in trace  element
composition and  morphology between coal fly ash and oil fly
ash. These authors observed enrichment factors (EFs) in oil fly
ash > 500 for V, Ni, Cu, Zn,  Sb, Hg, Cl, Ta, and Co, and EFs in
coal fly ash > 60 for Se, Hg, As, Cu, and Sb. (The particle size
in these studies was apparently not directly determined, but the
authors suggest  that the geometric mean particle size was
between 1 and 3 0 ^m for the  coal fly ash and between 15 and 82
,wni for the oil fly ash). Buseck and Bradley (1982), Germani et
al. (1981), and Small et al. (1981) discuss bulk and individual
particle analyses of copper smelter fly ash. Buseck and Bradley
(1982) directly observed condensates of volatile elements on the
outer surfaces of smelter fly-ash spheres.

5.1.1.2 Carbonaceous Particles
Carbonaceous particles typically comprise one-third or more of
the total PM10 mass. However, these particles, including soot and
biological particles, are traditionally problematic for CCSEM
because of poor  contrast in the backscatter image, making it
difficult or impossible for feature analysis software to reliably
identify and  size  such particles, especially submicrometer
particles. Martello et al. (2001) and Casuccio et al. (2002) have
developed a procedure that optimizes  CCSEM analysis  of
carbonaceous particles. The  sample is collected on a Pd-coated
                                                         45

-------
polycarbonate filter. The Pd coating (-0.17 ,wm) attenuates the
carbon signal from the filter substrate, thereby improving the
carbon signal-to-noise for particles. No carbon or metal conduc-
tive coating is applied to  the sample in order to avoid con-
founding the EDX analysis. A section of the filter is fixed to a
stub with double-sided silver tape for maximum conductivity.
Sample charging problems were evidently minimal, enabling the
analysis to be carried out  in the high-resolution SE mode in
which carbonaceous  particles are more reliably detected and
sized. The sample was analyzed at 15 kV accelerating voltage,
which yields adequate image  and EDX  information  while
minimizing electron beam  penetration into  the filter substrate.
The C:Pd X-ray ratio for particle-free areas of the sample is used
as a baseline for quantifying the carbon in particles.
      Soot is primary combustion-generated carbonaceous aero-
sol. Chemically and structurally, soot is a complex mixture of
amorphous polymerized organic material plus graphitic elemen-
tal carbon. Soot emissions from gas turbines, diesel engines, and
other combustion sources are major contributors to air contami-
nation in industrial and urban environments, typically accounting
for 1-4  Mg/m3 of atmospheric  inhalable  paniculate  matter
(Palotas et al., 1998). Posfai et al. (1999) used TEM to show that
a significant fraction of sulfate particles occurs internally mixed
with soot. With a graphite-like structure possessing extremely
high surface area, soot is an efficient catalyst: Soot-catalyzed
reactions may be a major factor in the oxidation of SO2 to sulfate
in polluted atmospheres (Novakov et al., 1974). In addition, soot
may act as a carrier of carcinogenic polycyclic aromatic hydro-
carbons into the lungs (Palotas et al., 1998). Most studies of soot
have employed methods such as TEM to characterize the micro-
structure of soot (Katrinak et al. 1992). These studies show that
the structure and composition of soot particles vary from dif-
ferent sources because of  fuel compositions and combustion
conditions. The microstructure of soot can therefore potentially
be used to infer the emission source.

5.1.1.3 Sulfates
On a global scale, sulfur-containing particles are perhaps the
most  important class of particles in the submicrometer size
range, and the great majority of these particles are in the form of
sulfate. Several papers have reported evidence of S-enrichment
in minerals and spores. Mamane et al. (1992) showed that
ambient particles in the size range of 0.5-10 ^m were enriched
in sulfur and attributed the  S-enrichment to reactions that occur
in the atmosphere while particles are airborne. Microscopists
should be aware that volatile particles such as sulfates may be
sublimed during the carbon-coating process  and/or during SEM
analysis. Parungo et al. (1986) reported that sulfate particle
counts by EDX were  consistently lower than those obtained on
the same sample using the BaCl2 reaction test (discussed below),
and attributed the difference to  sublimation losses.  Thus, in
general, one should be aware that the particles characterized in
conventional, high-vacuum SEM/EDX analysis may represent
only the non-volatile  fraction of the sample.
     Sulfate particles are often difficult to identify in CCSEM
analyses because of their low contrast and are, therefore, easy to
undercount. A simple chemical test to quantify the fraction of
sulfate particles in a sample is described by Bigg et al. (1974),
Mamane and de Pena (1978), and Ayers (1978). For the detec-
tion of sulfates, a thin-film (-300 A) of BaCl2 may be applied to
the filter collection surface before or after the aerosol is sampled.
Particles containing sulfate react with the BaCl2 coating and
form distinctive reaction rings that can be examined  with an
electron microscope. This  method is applicable primarily for
particles smaller than 1 /j,m.

5.1.1.4 Marine Aerosol
The world's oceans are  a major source for both primary and
secondary aerosols. In our laboratory, sea salt has been observed
in ambient samples from Phoenix, AZ, Baltimore, MD, and
Brownsville, TX. It is important from the point of view of source
attribution to recognize the morphology and chemistry of marine
aerosols. Meszaros and Vissy (1974) employed SEM  in an
attempt to identify the chemical composition of marine aerosol
by particle morphology.  They identified four types of particles
(sodium chloride, sulfuric acid, ammonium sulfate, and mixed
sodium chloride and ammonium sulfate) and measured their size
distributions. Parungo et al. (1986) carried out SEM/EDX analy-
ses of marine aerosol collected over the Pacific Ocean and
showed that larger particles (d >  0.5 ^m) comprised a wide
variety  of morphologies and combinations of NaCl,  CaSO4,
MgSO4, and KC1. A large fraction of these particles contained
sulfate. Nitrate appeared to attach preferentially to  sea-salt
particles in this larger size range. The great majority of particles
< 0.5  jum were sulfate. This size range was more difficult to
characterize because many particles sublimed under the  electron
beam.  This study also reported excess  (non-sea salt) sulfate,
nitrate,  and ammonium  particles.  The authors attribute these
secondary particles to photochemically assisted, gas-to-particle
conversion of bio genie  precursor gases. Posfai et al. (1994)
studied marine aerosol from the Equatorial Pacific using TEM,
EDX, and selected-area  electron diffraction (SAED).  Sea-salt
aggregates consisted largely of NaCl and Na-Ca sulfate crystals.
Sulfates of Na and Ca with minor K and Mg form on NaCl
crystals. In addition, several species of submicrometer (typically
0.2 /j.m) S-bearing particles were observed, presumably includ-
ing ammonium sulfate and acid sulfates. The presumed ammon-
ium sulfate particles were rapidly volatilized in the  electron
beam. Diatom fragments were common, and there was  a sparse
crustal component consisting mostly of kaolinite and rutile. The
authors report that volatile elements were lost during  EDX
analysis. Results provided evidence of gas-phase HC1 released
from sea-salt. Parungo et al. (1990) also reported a deficiency of
chloride ion relative to seawater composition in aerosol collected
in the Gulf of Mexico and hypothesized the release  of HC1 as
SO2 and NOX replace chloride ion in sea salt particles. Using
laser microprobe mass analysis, Bruynseels and Van  Grieken
(1985) proved the existence of sulfate and nitrate layers on the
                                                         46

-------
surface of sampled marine  aerosol, although it could not be
established whether the surface layers existed in the atmosphere
or were created after collection by differential crystallization
from homogeneous liquid-phase particles. Andreae et al. (1986)
examined individual aerosol particles from the remote marine
atmosphere and found that a very large fraction of the silicate
mineral component of the aerosol was internally mixed with sea-
salt particles. They proposed cloud processing as a mechanism
and suggested that this could also explain excess sulfate (non-sea
salt) enrichment of sea-salt particles.

5.1.1.5 Miscellaneous
CCSEM has been used to characterize and identify sources of
lead-rich particles in the environment (Mamane et al., 1995;
Johnson and Hunt, 1995; and Vander Wood and Brown, 1992).
Mamane  (1988)  used  manual SEM to  characterize  particles
emitted from a municipal waste incinerator. The results were then
used in combination with conventional bulk analysis of ambient
filters to apportion the fraction of Philadelphia aerosol contributed
by municipal waste incineration (Mamane,  1990). Germani and
Zoller(1994) characterized paniculate emissions from a municipal
waste incinerator using SEM/EDX. They found that small incin-
erator particles are composed of a rather homogeneous mixture of
chloride salts of Na, K, Zn, and Pb, and that, like coal-fired power
plants, incinerators enrich volatile elements via a vaporization-
condensation mechanism. Their data suggest that small incinerator
particles are coated with a thick, soluble chloride salt coating.
Waste incineration appears to be the major source of Zn, Cd, Sb,
and possibly Ag, Sn, In, and Pb in urban atmospheres.

5.1.2  Source Apportionment
Individual particle analysis, combined with bulk chemical analy-
sis or as an alternative to bulk techniques, can potentially im-
prove source resolution in source apportionment studies (Kim
and Hopke, 1988b; Mamane,  1990). Dzubay  and Mamane
(1989) explored the feasibility of increasing the source reso-
lution of the CMB method by using SEM/EDX data with bulk
XRF results. In this way, the authors were able to determine two
additional components (coal fly-ash and botanical matter) that
could not be determined by XRF data alone. SEM provided
estimates for other components as well including municipal
incinerators. SEM generally confirmed CMB source contribution
estimates, but in some cases CMB results predicted a contri-
bution from municipal incinerators, but SEM  results were not
able to support a contribution.
     Casuccio et al. (1988) compared upwind,  stack, diluted
stack, and plume samples from a coal-fired power plant, demon-
strating the ability of CCSEM to provide information on source/
receptor relationships.  Kim and Hopke (1988b)  used CCSEM
data in a particle class balance (PCB)  analysis to apportion
sources of PM in El Paso. The authors conclude that the large
number of particle  classes typically identified by CCSEM
reduces collinearity problems and enables greater resolution of
sources compared with conventional bulk analysis. Hopke and
Mi (1990) classified CCSEM particle data from a power plant
and were able to distinguish in-plume particles from in-stack and
ambient particles. Anderson et al. (1992) determined the size
distribution of ~ 30,000 fine  (0.1 to ~ 2 /j.m) arctic aerosol
particles and attempted to identify their sources. The ability to
reliably detect low-Z particles as small as 0.1 /j.m distinguishes
this work from many others. Katrinak et al. (1995) used CCSEM
to identify individual particle  types in the  Phoenix aerosol.
Jambers and Van Grieken (1997) combined CCSEM analysis of
riverine suspension particles with hierarchical cluster analysis to
shed light on sources of pollution in Lake Baikal. Xhoffer et al.
(1992) apportioned particles collected in the North Sea surface
microlayer and in underlying bulk waters to atmospheric or
riverine suspension sources.

5.1.3  SEM/EDX Methodology
Researchers have had mixed success in efforts to establish a
quantitative link between individual particle  analysis and bulk
chemical analysis of a sample. Johnson et al. (1981) compared
bulk XRF results and (summed) individual particle results on the
same filter samples. The authors were unable to resolve  dif-
ferences between the elemental mass accounted for by the two
approaches and concluded  that accurate calculation of bulk
elemental weight percents based on individual particle analysis
was not quite feasible. Nevertheless, CCSEM can provide an
internally consistent bulk chemical analysis, which is useful in
comparing samples. Potential sources of error associated with
CCSEM are discussed including particle volume and density
calculations, assumptions about particle stoichiometry, and parti-
cle classification. Casuccio et al. (1983b)  compared bulk and
microscopic analyses of ambient High Volume (Hi Vol) filters.
Mamane (1988 and 1990) showed excellent agreement between
mass concentrations derived from bulk and manual individual
particle analysis (regression line slope of almost 1.0 and cor-
relation coefficient of 0.9).
     A major issue  with  CCSEM  data  is how to classify
particles in order to facilitate interpretation of the data.  In the
decades since the introduction of CCSEM, particle classification
schemes have grown in number and sophistication. Johnson and
Twist  (1982) discuss approaches to sorting and classifying
individual particle data for use  in microscopy-based receptor
models. Johnson et al. (1981) classified particles into environ-
mentally recognizable classes, which were based on a variety of
source standards (coal and oil fly ash, wood combustion, high
temperature industrial emissions,  clay mineral classes, etc).
Johnson et al. (1984b) and Dzubay et al. (1984) describe the use
of SAX (scanning electron microscopy with automated  image
analysis and X-ray energy spectroscopy) to classify particles in
Houston aerosol. Aerosol samples were described in terms of 25
fixed particle types making up the mass on each filter. The 25
particle classes were developed from a combined set  of 18
ambient and 30 source signatures subjected to factor analysis.
     The statistical technique of cluster analysis (including hier-
archical and  non-hierarchical  cluster analysis) has proven
                                                         47

-------
successful in discerning particle groupings in large data sets with
no prior knowledge of the group characteristics (Kim et al,
1987; Kim and Hopke, 1988a; Hopke and Mi, 1990; Anderson
et al., 1992; Saucy et al., 1987,  1991; Shattuck et al., 1991;
Germani andBuseck, 1991; VanBormetal.,  1989;Katrinaket
al., 1995). Shattuck etal. (1991)applied non-hierarchical cluster
analysis to a complex Phoenix aerosol sample that was about
75% crustal in origin. The authors concluded that cluster analy-
sis was effective  for determining the types of particles that
occurred in the Phoenix aerosol. This paper presents a particu-
larly detailed multistep recipe for applying cluster analysis,
assessing the validity of the clusters, and interpreting relation-
ships and trends among multiple samples. There appears to be
some disagreement among researchers  as  to the need for
pretreating CCSEM data prior to performing cluster analysis.
However, Shattuck et al. (1991) argue that data normalization
gives artificially low values for the standard deviations of such
variables and correspondingly too great a weight after normal-
ization.  Hopke and Mi (1990), on the other hand, found that
cluster analysis results were more reasonable when the raw
CCSEM data were pretreated as follows: (1)  Any X-ray peaks
containing fewer than Nc counts, where Nc = 2NT"2 and NT is the
total X-ray count in the spectrum, were considered to be  noise
and set to zero. (2) The noise-reduced data were log-transformed
after adding 1 to all  X-ray intensities; thus, XtJ =logig(l+xv),
where x^ is the ith variable of the jth particle and Xy is the
transformed value of x^. (3) Finally, the data are transformed to
have zero means and unit variance (z-transformation).
     Routine quantitative analysis of individual particles has
generally been considered to be an intractable problem because
of the difficulty of relating  X-ray intensities  to a particle's
element concentrations. Armstrong and Buseck (1975)  derived
theoretical correction factors specific  for a variety of  particle
shapes. The correction procedures were applied to particles of
known composition, and the results demonstrated that routine,
manual quantitative analysis of microparticles  is both feasible
and straightforward. Successful application of these correction
procedures, however, assumes that each particle's thickness and
shape can be approximated during the time of analysis;  it is not
clear whether the particle shape parameters typically acquired
during a CCSEM analysis are sufficient for approximating parti-
cle shape and thickness.

5.2 Phoenix PM25 Samples
A 3-yr  campaign of PM25 air sampling  beginning in  1995
provided a database of sample chemical analyses of sufficient
size to be well-suited to the large data quantity requirements of
multivariate analysis.  A study was conducted to demonstrate the
multivariate  receptor model Unmix (Henry,  1997, 2001) on a
typical urban aerosol data set. Using only ambient concentration
data for PM2 5 mass and various  chemical species the Unmix
software estimates the number of contributing sources, their
compositions (withuncertainties), their PM25 mass contributions
to each sample, and their average PM2 5  mass contributions (with
uncertainties). Unmix is a type of factor analysis, but geo-
metrically constrained to generate source contributions and pro-
files with the physically meaningful attribute of non-negativity.
The Unmix analysis was supplemented with SEM/EDX of a
limited number of filter samples. These results provided infor-
mation about additional low-strength sources that Unmix did not
quantify. Details of this study are  reported in Lewis et al. (in
press). The application of SEM/EDX in the study is featured
here as a demonstration of an SEM/EDX application.

5.2.1  Unmix Receptor Model Findings
The multivariate receptor model Unmix has been used to analyze
a 3-yr PM25 ambient aerosol data set collected in Phoenix AZ
beginning in 1995. The analysis generated source profiles and
overall percentage source contribution estimates (SCE) for five
source categories:  gasoline engines (33±4%), diesel engines
(16±2%), secondary sulfate (19±2%), crustal/soil (22±2%), and
vegetative burning (10±2%). Unmix results show  the element
Mn playing an unexpectedly large role in the diesel source.
Except for the diesel engine source category the Unmix SCE's
were generally consistent with an earlier multivariate receptor
analysis of essentially the same data using the Positive Matrix
Factorization (PMF) model (Paatero and Tapper, 1994; Paatero,
1997).

5.2.2  SEM/EDX Findings
SEM/EDX was used to obtain information about additional low-
strength sources that Unmix did not quantify, and to investigate
the association of Mn with the diesel source. The Personal SEM®
(PSEM) (formerly R. J. Lee  Instruments, Ltd., now Aspex
Instruments, Trafford, PA) was used to conduct manual single-
particle analyses of a limited number of samples.

Sea salt. Because of the inland location of Phoenix it is not
intuitive  that its  aerosol would have a  measurable marine
component. In fact, however, particles  such as the one shown in
Photo 1 of Figure 5-1, consisting only  of Na and Cl and having
a cubic structure, were quite easy  to find in Phoenix samples
having measurable amounts of Na. The Hybrid Single-Particle
Lagrangian Integrated Trajectory Model version 4.3 (Draxler,
1999) was used to compute back-trajectories for eleven samples
having the highest Na concentrations. Virtually all of the trajec-
tories originated from a westerly direction, with most having
passed over the Pacific Ocean less than 48  h earlier, clearly
suggesting a marine impact. Previous SEM work (Saucy et al.,
1991; Katrinak etal., 1995) has identified possible sea salt parti-
cles in the Phoenix area.

Fly Ash particles. Fly ash particles (spherical alumino-silicates
with minor amounts  of Fe and Ca)  were observed in most
samples examined by SEM/EDX.  A typical fly ash particle is
shown in Photo 2 of Figure 5-1. A spherical  particle shape is
indicative of any high-temperature process, including com-
bustion  or smelting.  Fly  ash particles have  been previously
                                                         48

-------
         Photo 1
                                                                            Photo 2
         Photo 3
                                                                            Photo 4
         Photo 5
                                                                            Photo 6
Figure 5-1. Examples of particles identified by SEM/EDX (Photos 1- 6). The upper-left quadrant of each photomicrograph shows a field of view with
the particle of interest within the smaller square in that field. The upper-right quadrant shows a zoomed-in view of the feature (the area from within the
square in the upper-left quadrant), and the lower half shows the elemental spectrum acquired with the electron beam centered on the small square in
the zoomed-in view. A 1 -urn scale appears just below the upper-right quadrant.
                                                                49

-------
observed in Phoenix and attributed to coal-fired power plant
emissions (Post and Buseck, 1984).

Cu-, Pb-, and Zn-bearing particles. Various  studies of the
Phoenix aerosol have identified (Saucy et al, 1991) and sought
to quantify (Chow et al., 1991; Ramadan et al., 2000) emissions
from copper smelters, the nearest of which are located about 100
km to the southeast. Source measurements have shown (Small
et al., 1981) that copper smelter plumes in this region are highly
enriched in the elements Cu, Zn, and Pb compared with crustal
abundances. Other sources such as municipal  incinerators or
industries making or using bronze may also be strongly enriched
in one or more of these elements. Particles from a sample asso-
ciated with a southeasterly back-trajectory (the general direction
of the copper smelters in the area) were examined by SEM/EDX.
On an individual particle basis, Cu-bearing particles were most
often seen as spheres in their pure oxide form (Photo 3 of Figure
5-1), while Pb-bearing particles were most often seen combined
with Cl (Photo 4 of Figure 5-1). These two elements were not
generally found in combination with each other (although Photo
4 does show a small amount of Cu in the Pb-Cl particle). Zn-
bearing particles were observed less frequently than Pb- or Cu-
bearing particles and were typically combined with Pb, Cu, or
other metals. Collectively, this evidence suggests that any high-
temperature or other process employing Cu, Pb, and/or Zn may
be contributing to particles composed of these elements mea-
sured in the Phoenix area.

Spherical Feparticles. Numerous iron oxide or Fe-bearing parti-
cles were observed on the samples examined by SEM/EDX.
Among these were spherical iron oxide particles (Photo 5 of
Figure 5-1), again indicating a high-temperature process was
involved in their formation.  Researchers have previously re-
ported iron oxides as being one of the main types of particles
produced by iron foundries (Post and Buseck, 1984). Two such
facilities were operating near Phoenix during the time of this
field study (U.S. EPA, 1996). Additional SEM/EDX work has
identified iron microspheres in foundry emissions (Michaud et
al,  1993). Iron foundry operations were not identified by the
Unmix calculation, most likely because of their low impact on
the overall fine mass loading.

Mn-bearingparticles. These particles were investigatedby SEM
because  of their unexpected association with  the  Unmix-
generated source that was attributed to diesel emissions. Photo
6 of Figure 5-1 shows a typical Mn-bearing fine particle located
on  a Teflon® membrane sample filter identified as having a
relatively high diesel source contribution. The particle shown is
composed principally of Fe, Mn, Cr and Si, with Fe being the
largest component. Its  spherical shape is indicative of a high-
temperature process, such as combustion or smelting.
     Researchers using SEM (Zayed et al., 1999) have shown
that for automobiles using the gasoline additive methylcyclo-
pentadienyl manganese tricarbonyl (MMT), Mn-bearing parti-
cles sampled from the tailpipe also contained phosphorus and
sulfur. Iron or other metals were not typically found. These do
not appear to be similar to the Mn-bearing particles observed on
the samples reported here, which are primarily iron-oxides, with
lesser amounts of Mn and other metals but no phosphorus and/or
sulfur. While this  argues against the possibility that  Mn in
Phoenix aerosol may originate from the combustion of gasoline
containing MMT,  as has been speculated (Ramadan et al.), it
does not clearly demonstrate that Mn is associated with diesel
emissions.
     An earlier study conducted in Phoenix (Post and Buseck,
1984) identified particles containing oxides of Fe or Zn, with
smaller amounts of Mn, Cr, Si, Al, Cl, K, or Ca and suggested
that these particles originate from iron foundries located SE of
Phoenix. The Mn-bearing particles identified in the present study
are consistent with the type of particles that may be emitted by
iron foundries.

5.2.3  Summary of Results
Application of the Unmix receptor model to the 1995-1998
Phoenix PM2 5 data set has resulted in a 5-source apportionment.
Limited SEM examination of filter samples from the field study
indicated the presence of additional sources (sea salt, copper
smelter, iron foundry, fly ash), but their presumably small im-
pacts on PM2 5 were not quantified by Unmix.

5.3  Fort Hall Source Apportionment Study
The  objective of the  Fort Hall Source Apportionment Study
(Willis et al., 2001) was to identify and quantify individual
sources within the Astaris (formerly FMC) elemental phos-
phorus plant that contribute to exceedances of the NAAQS for
24-h PM10.  The Astaris plant is located on the Fort Hall Indian
Reservation near Pocatello, ID. The source apportionment effort
relied heavily  on  XRF analysis of ambient 24-h PM25 and
PM10_2.5 data collected at three monitoring sites. Wind-directional
analysis was also critical in locating major emission sources. The
major conclusion of the study was that fine-fraction phosphate
was  the dominant species contributing to  PM10 exceedances,
though in general, resuspended coarse dusts from raw and
processed materials at the plant were also needed to create an
exceedance. Major sources that were identified included the
calciners, the CO flares, process-related dust,  and electric-arc
furnace operations.
     SEM/EDX analyses of source and ambient samples played
an important supporting role in the study by complementing and
confirming results obtained by bulk XRF analysis, by providing
visual and chemical confirmation that combustion-related phos-
phate particles dominate the ambient fine fraction downwind of
the Astaris facility, and by helping in the identification of
potential sources. Examples are discussed below.
     Figure 5-2a is a micrograph of a fine-fraction dichot filter
collected at the Primary site on August 26,1997. The 24-hPM10
concentration was  86 ^g/m3. Particles are on the order of 1 ,wm
in size and nearly  all are P-rich. Many of the P-rich particles
have a "wet" appearance and cling to the filter's Teflon fibers
                                                        50

-------
like dewdrops. During exceedances, the concentration of P-rich
particles on the filter can be so great that the particles begin to
coalesce on the filter, forming P-rich islands as shown in Figure
5-3, collected during a PM10 exceedance when the 24-h PM10
concentration was 206 ,wg/m3. Manual SEM/EDX analyses of
ambient filters like these provided evidence early in the study for
a major combustion source or sources emitting fine-fraction P-
rich particles. A comparison of ambient particles with source
particles collected in the ground flare (see Figure 5-4) and at the
phos dock (see Figure 5-5) showed that the particle chemistry
and morphology were very similar. Receptor model results, in
fact, attribute on average  about one-third  of the PM10 during
exceedances  to flaring operations (the  ground flare  and the
elevated flare) and fugitive emissions from the furnace building.
     Particles in the coarse fraction show a distinctly different
chemistry and morphology. Figure 5-2b shows the coarse-frac-
tion mate to Figure 5-2a. Most coarse particles have the rough,
irregular surface morphology characteristic of crustal dust or soil
particles. Particle chemistry is dominated by Ca and Si, which
are the major species  in  the phosphate ore used  at  Astaris.
Ambient particle morphology and particle chemistry indicate a
strong contribution to the coarse fraction from re-entrained dust
related to raw or processed phosphate ore. Occasional fly ash
spheres from combustion  processes are  observed such as the
large Ca-Si-rich sphere in the upper center of the field. Similar
combustion spheres  were observed  in dust and  air samples
collected in the furnace building (see Figure 5-6), suggesting the
furnace as a probable source. The  fine-fraction component
collected on this coarse filter can be seen in the background as
small, submicrometer phosphate particles clinging to the filter
fibers.
     SEM/EDX analyses can elucidate elemental associations in
particle chemistry. For example, particle-bound mercury (HgP)
was detected by bulk XRF in a few ambient fine-fraction sam-
ples at the level of about 0.2% of the total fine mass. A dedicated
CCSEM search for Hg-rich particles  yielded a number of Hg-
richparticles embedded in P-rich matrix (see Sections.3.5). The
EDX spectra for these particles indicate that Hg and Se are
present in  most Hg-rich particles in a fairly constant ratio,
perhaps as  mercury selenide (tiemmanite). In addition, several
Hg-rich particles also contained silver,  although silver  was
below XRF detection limits in  the bulk filter analysis.
     Particles originating from different sources or processes
often have distinctive chemistry and/or morphologies that canbe
used to identify and quantify  the impact  of these sources in
ambient samples. Examples include the droplet-like P-rich parti-
cles from the flares or emissions from the furnace building; the
Fe-P-V-rich ferrophos particles,  which exhibit chonchoidal,
glass-like fractures; and the P-K-rich particles from the hot slag
tapping area.
Figure 5-2a. Fine dichot sample collected at the Primary site (8/26/97).
The superimposed X-ray spectrum shows that nearly all particles are
phosphorus-rich. (The fluorine and carbon peaks are generated by the
Teflon filter).
Figure 5-2b. Coarse dichot sample collected at the Primary site (8/26/97).
Coarse particle chemistry is dominated by Ca and Si. Spherical particles
such as the  large Ca-Si-rich sphere in the upper center of the field are
produced in combustion processes.
                                                          51

-------

Figure 5-3. Ambient sample 57f, Primary site, 3/11/97. PM10 concen-
tration for this day was 206 |jg/m3. Liquid-like phosphorus-rich particles
coalesce on the heavily-loaded filter to form a quasi-continuous layer of
phosphorus-rich  material. The X-ray  spectrum was acquired from the
point shown by the cross hairs.
                                                                     Figure 5-4. Top photo: Fine dichot sample collected at the Primary site
                                                                     (8/26/97). Bottom photo: Personal air sample collected from the ground
                                                                     flare plume during a miniflush. Particles are approximately 1  micrometer
                                                                     in size and cling to fine Teflon fibers comprising the filter. Nearly all
                                                                     particles are droplet-like P-rich particles similarto the one centered in the
                                                                     magnified  image on  the right,  and similar in size,  morphology and
                                                                     chemistry to the particles dominating the ambient fine-fraction sample in
                                                                     the top photo.
                                                                  52

-------
Figure 5-5. Personal air sample collected on the phos-dock. P-rich
particles dominate the sample.
Figure 5-6. Calcium silicate fly ash from the burden level of the furnace
building.
5.4 Baltimore Retirement Home Study
The United States Environmental Protection Agency (U.S. EPA)
recently conducted the 1998 Baltimore Paniculate Matter (PM)
Epidemiology-Exposure Study of the Elderly. The primary goal
of that study was to establish the relationship between outdoor
PM concentrations and actual human PM exposures within a
susceptible (elderly) sub-population. The overall study design
has been described by Williams et al., 2000a, and Williams et
al.,  2000b. The study was conducted over a four-week period
during July and August  1998.  Personal,  indoor, and outdoor
sampling of paniculate matter was conducted at a retirement
center in the Towson area of northern Baltimore County. Con-
current sampling was  conducted at a central community site
10  km from the  retirement facility  and four  km from the
Baltimore harbor area. The main objective of this work was to
use computer-controlled scanning electron microscopy with
individual-particle X-ray  analysis (CCSEM) to measure the
chemical and physical characteristics  of  geological and trace
element particles collected at the various sampling locations in
and around the retirement facility. The CCSEM work is de-
scribed in detail in Conner et al. (2001) and is synopsized here.
This work provides a preliminary model for how to conduct
CCSEM analyses of ambient PM samples and how to present the
results.
5.4.1  Methods
Three sets of samples were selected for CCSEM analysis based
on mass loading on the filter and meteorological air mass trans-
port during the sampling period (Sampson and Moody, 1981,
Draxler, 1999). Each set consisted of the concurrently collected
community, residential outdoor, and residential indoor coarse-
fraction VAPS samples. Day 13 (8/7/98) of the sampling cam-
paign was influenced by easterly transport and was selected to
represent a primarily marine air mass. Day 20 (8/14/98) was
influenced by southeasterly transport and represents industrial
emissions, but was also impacted by marine air. Day 18 (8/12/
98) was influenced by northerly transport and was selected to
represent the background air with minimal influence of Balti-
more  industrial emissions and marine air masses. These sam-
pling  days will be referred  to as "marine", "industrial", and
"background", respectively,  in  subsequent  text, tables, and
figures.
     The  PERSONAL SEM®  (PSEM) (formerly R.  J. Lee
Instruments, Ltd., now Aspex Instruments, Trafford, PA) was
used to conduct the manual SEM/EDX and computer-controlled
(CCSEM) analyses.  The PSEM features software that enables
the user to review and summarize the data off-line. Particle
classes canbe developed by the analyst and are based on particle
size, shape, image brightness, X-ray counts and/or chemical
composition criteria. Particle classes provide a convenient way
of summarizing  the very large data  sets  acquired through
computer-controlled analysis, and help to interpret the possible
sources.
     For the  computer-controlled analyses, the PSEM was
operated with a 20-kV, 30% spot size electron beam. (For the
PSEM, spot size corresponds to beam current.) The working
distance was set at 18-19 mm, which is the optimum distance for
X-ray acquisition for the PSEM. A lens degaussing procedure
was performed prior to each analysis to correct for lens hyster-
esis which may occur when operating conditions are changed
between analyses. The backscattered electron mode was used for
particle location, measurement, and analysis. A magnification of
lOOOx was used for the coarse particle analysis; fine particles
were analyzed at  a magnification of 1800x. Secondary electron
images for each particle and analysis field were acquired. The
                                                        53

-------
magnification for the individual particle images varied based on
the size of the particle. Acquired images played a key role in the
development of rules  for particle classification.
     The minimum physical diameter for the coarse particles
was set at 1 ,wm in anticipation of some of those particles being
greater than 2.5 ,wm aerodynamic diameter. The maximum phys-
ical diameter was set at 12 ,um. The fine particle analysis was set
up to look for particles with a physical diameter between 0.1 and
2.5 ,wni. The overlap was intentional to ensure that no particles
were missed. The post-analysis classification rules eliminated
any duplication.
     X-ray counts for 26 elements (coarse fraction) ranging in
atomic number from Na to Bi were acquired for each particle
and saved along with  other measured parameters. Bromine and
Sr were also included in the fine particle analysis for a total of
28 elements. Afew elements included are nottypically measured
in outdoor ambient particles but were found in the preliminary
examination of the filters. Some might be found in the indoor
environment (e.g., Zr, Bi), while others could be found in the
industrial emissions in the Baltimore airshed (e.g.,  Sb, Cd).
     A minimum number of X-ray counts was specified to
adequately  characterize each particle and  to  minimize the
amount of data acquired for non-particle features  (e.g., filter
pores), or incomplete particle features (e.g., small portions of
large, light-element particles such as skin  flakes). The post-
analysis summary applied an additional low counts rule to all
samples to separate the low count particles from the rest of the
data set.
     Approximately 500 to  1000 particles need to be charac-
terized to get a representative sample (Mamane et al., 2001),
depending on the complexity of the sample and the  overall
research objectives. The Baltimore airshed contains a complex
mixture of particles from a variety  of natural and industrial
sources, with the coarse fraction having more variety than the
fine fraction. An adequate number of particles must be specified
in the analysis routine to compensate for the later exclusion of
some of the particles which fall outside  of the size range of
interest, or which will be excluded  based on chemical com-
position (e.g., salt). Thus, up to 2000 particles were measured for
each analysis, and two analyses (one for coarse particles and one
for fine particles) were conducted per sample. The actual filter
area covered in any given analysis depends on the loading of the
particles meeting the analysis specifications.
     The threshold is the brightness (grayscale) value in the
digital imaging system which is chosen by the  operator to
discriminate between particles  and filter  background.  The
thresholds for particle detection and measurement were selected
to acquire data for as  many features as possible while avoiding
excess detection of non-particle features. The threshold was self-
correcting to account for the slight variation in image brightness
from one analysis field to another. In addition, the threshold
stability was manually verified at least once per analysis.
     The CCSEM analysis results shouldbe considered a lower-
limit estimate of the numbers of carbonaceous, sulfate, ammon-
ium salts, or other light-element particles. Such particles do not
produce a significant backscattered electron signal, resulting in
poor contrast with the filter medium, and are typically smaller
than 1 jum.

5.4.2  Sum maty of Manual SEM/EDX Analysis
        Results
Each sample was surveyed manually by SEM/EDX to assess
their  suitability  for computer-controlled  analyses.  Particle
loading must be light enough to have adequate separation of the
particles for the computer-controlled analysis. The preliminary
manual survey also provides information on particle  size and
chemistry, which is used to  set up the CCSEM analysis parame-
ter files and familiarize the  analyst with the sample.
     Manual examinations indicated that the selected samples
had adequately spaced particles. Both residential outdoor and
community samples were loaded with particles of geological and
industrial origin. The indoor samples were lightly loaded. Most
samples collected at the outdoor and community sites on the
marine air  and industrial  air days were covered with large
amounts of both fine and coarse salt (NaCl). The overwhelming
numbers of salt particles on samples impacted by a marine air
mass would have  required a prohibitive amount of time to
acquire data for a representative number of non-salt particles, the
particles of primary interest. Thus, an analysis rule was written
to reject most of the salt particles. In addition,  a post-analysis
rule was developed as part  of the particle classification system
(see Section 5.4.3)  to separate the remaining salt particles that
were not screened out by the analysis rule.

5.4.3  Summary of  CCSEM Analysis Results
Following the analysis, particle classification rules (see Appen-
dix C) were developed to summarize the CCSEM data. These
particle classification rules were designed to classify particles
based on size, shape (aspect ratio), elemental composition, X-ray
counts, and/or video (grayscale brightness) level (reported as a
numeric value) of the particle image. The rules were applied to
each sample, and results evaluated by examining both measured
parameters  and particle images. Rules were changed or added
based on these evaluations, and the process continued in an
iterative manner until the particle classifications were judged to
be satisfactory, based on the uniformity of chemical and physical
characteristics within a particle class. The rules were applied to
other samples to make minor adjustments and to test the robust-
ness  of the classification scheme, based on the uniformity of
particle characteristics within each particle class both within a
single sample and across all samples. For all particles reported
to contain Pb or other trace  metals, the X-ray spectrum was
reviewed to verify the identification of such elements.
     The CCSEM coarse particle classifications are compared
in Figures 5-7. The CCSEM fine particle  classifications are
compared in Figures 5-8. Data are reported as number of parti-
cles per unit filter area for each particle class.  Unit filter area
was used as a surrogate for volume of air sampled because
approximately the same volume of air (maximum and minimum
                                                         54

-------
  E
  o

  &
  t>
  .2  1E+2
  ro
  Q.
ri| Community Marine A Outdoor Marine Q Indoor Marine
rj Community Background A Outdoor Background O Indoor Background
QJ Community Industrial A Outdoor Industrial O Indoor Industrial
o
N
0 p
Q
o
o
                                 Particle Class
  E
  o
  5
  r
  Q. 1E+2
                                 Particle Class
1E+5
tM
£ 1E+4
O
£ 1M
73
1 1E+2
1E+1

An n H Q n n ^
|B|| 0A I|A *
g*l^« 0 °D iM
O O
o o
^o^S^^5pS^>^5o;^ ^^^S
•c i- -c co -c > ,_ .c ro-co =3717" c a -c: ro
5fS|
-------
E
O  1E+3

<5
Q.
   (
   Q.
                     Community Marine     A Outdoor Marine     O Indoor Marine

                     Community Background A Outdoor Background O Indoor Background

                     Community Industrial   A Outdoor Industrial   (9 Indoor Industrial
                                        D
                                      Particle Class
  CM  1E+6


  E
  O  1E+5

  <5
  Q.

    1E+4
  0)

  o
  K  1E+3

  ro
  Q.

  «  1E+2
                                      Particle Class
J1  1E+5

E
o

5  1E+4
Q.
   ro
   Q.
                     14T
             D
                         -®-
                                                   A


                                      Particle Class
Figure  5-8.  Comparison of CCSEM particle classifications for  PM25  (fine)  particle

classes (a) dominant at the residential indoor site, (b) dominant at the residential outdoor

site, (c) dominant at the ambient community site.
                                        56

-------
volumes differed by  less than 9%) was drawn through each
filter. Reporting data in this way provided a satisfactory means
of comparing particle number concentrations among samples.
     The following are the principal findings of this analysis:
     1.  At both the community and outdoor sites, particle types
        associated with industrial activities (e.g., Al rich, Cr
        rich, Ni and/or V, Fly Ash) were measured at their
        highest concentrations on the industrial air mass trajec-
        tory day. These were found at much lower concentra-
        tions, or not at  all, at the indoor site  on this  day, as
        expected for coarse particles of industrial origin.
     2.  Coarse particles characterized by high concentrations
        of Ca, Mg, and K (components of sea salt) were mea-
        sured at their highest concentrations on the marine air
        mass trajectory day at the community site. The outdoor
        site appeared to be less impacted by marine air on that
        day.
     3.  Several particle classes were observed predominantly
        or exclusively at the indoor sampling location.  The
        most prominent is the Al-Zr-Cl particle class, found
        exclusively indoors in both the fine and coarse frac-
        tions. Source images and spectra (Conner et al., 2001)
        support the hypothesis that Al-Zr-Cl  particles origi-
        nated from a personal antiperspirant product.
     4.  Evidence was found to support the hypothesis that cos-
        metic products may be a source of paniculate matter
        indoors among this unique study population (Conner et
        al.,  2001). However, this population may not be con-
        sidered fully representative of the general  elderly
        population.
     5.  Review of acquired images demonstrated the capability
        of the CCSEM technique to identify spherical particles,
        which are generally indicative of combustion or other
        high temperature processes characteristic of industrial
        activities. These spherical particles were found in the
        coarse fraction almost exclusively at the outdoor and
        community sites.
     6.  Review of acquired images demonstrated the capability
        of the CCSEM technique to identify pollens and spores.
        The highest concentration of these particles was found
        at the outdoor site on the background air day.

5.4.4  Conclusions
The CCSEM results show that the relative abundances of some
geological and trace element particle classes identified at the
outdoor and community locations  differ from  each other and
from the indoor location. Particle images acquired during the
computer-controlled analyses played a key role in the identifi-
cation of certain particle types. Review  of these images was
particularly useful in distinguishing spherical particles (usually
indicative of combustion) from non-spherical particles of similar
chemical composition. Pollens and spores were also identified
through a  manual review of the particle  images. Overall, this
work shows that CCSEM and manual SEM/EDX can contribute
to the understanding of the sources of paniculate matter in dif-
ferent micro-environments in a way that is complementary to
bulk elemental analysis techniques.

5.5 World Trade Center Study
The potential health implications of exposure to dust from the
collapse of the World Trade Center (WTC) motivated a study of
samples collected on and near the WTC site. SEM/EDX was
used to obtain chemical and physical characteristics of particles
and fibers found in hand-collected bulk material and in filter-
based air monitoring samples.

5.5.1  Verification  of XRF Results
Figure 5-9 shows a gold particle identified in a hand-collected
dust sample. X-ray fluorescence analysis of this dust had shown
detectable levels of gold which is unusual. The size of the gold
particle (~1 ,wm) and the roughly spherical shape suggest that the
particle was  formed in a high-temperature combustion process.
Figure 5-9. Micrometer-sized gold particle in World Trade Center dust
sample.
5.5.2  Analysis of Bulk Sample
Bulk material was collected from a car windshield located in a
parking garage about 3 blocks from the WTC site. This sample
is thought to represent the initial light-colored cloud that ema-
nated from the tower collapse. The covered garage location pro-
vided some protection from rains which had fallen since the
initial deposit of the material.
     A small amount of the bulk material was sprinkled onto
sticky carbon tabs affixed to aluminum SEM stubs. Images and
X-ray spectra of representative features were manually acquired.
Images were created using the backscattered electron mode to
avoid distortion of the image due to charging of the specimen
surface. Each image and spectrum was saved electronically to a
Zip Disk.
     In general, this bulk material was quite fibrous and tended
to aggregate into large clumps. Evidence of the fibrous nature of
                                                         57

-------
the particles is presented in Figure 5-10,  which shows three
randomly selected fields of view taken at low magnification.
Examples of individual fibers are shown in Figure 5-11. Note
that though SEM is not the definitive or sole method for identi-
fying asbestos, it is unlikely that the fibers found in this par-
        Figure 5-10. Three randomly selected fields of view
        of a fibrous bulk  material  collected  in a parking
        garage near the WTC site.
ticular sample are asbestos fibers, based on their variable size
and chemical composition. Most of the individual particles or
fibers had a Ca-Si-Al-Mg-S composition and ranged in diameter
between approximately 1 and 10 ,wm. The size and composition
of these fibers suggests insulation materials as the source. Some
carbonaceous fibers were  also found, as  depicted in one of
the micrographs in Figure  5-11. There are numerous possible
sources for carbonaceous fibers, including any type of fabric or
textile which may have been used in the buildings' interior
structure  or furnishings.
     Examples of particles less than 10 ,wm in diameter are
presented in Figure 5-12. Most  of these were  composed
primarily of Ca and S (gypsum) with wallboard and plaster
materials being the likely source. Some metals including Fe and
Zn, were also found.

5.5.3  Analysis of Ambient Air Sample
A coarse fraction VAPS sample was collected in NYC in Sep-
tember 2001.  The coarse fraction of a VAPS filter also contains
a small fraction of the fine particles, enabling simultaneous
examination of both the coarse and fine paniculate matter.
     A small (less than 1 cm2) piece of filter material was
affixed to aluminum SEM stubs using a carbonaceous suspen-
sion. Images and X-ray spectra of representative features were
manually acquired. Images were created using both the back-
scattered and secondary electron modes. Each image and spec-
trum was saved electronically.
     SEM/EDX data  is best presented in image  format. A
randomly selected field of view is depicted in Figure 5-13. This
figure clearly shows the "honeycomb" particle distribution typi-
cal of samples collected on polycarbonate filters with no addi-
tional filter backing to act as a diffuser.
     Images of 44 representative particles, with emphasis on
metal-containing particles,  were acquired and are presented in
Appendix D.  Most of the particles were either Ca-S, Fe or Fe-
containing Pb-Cl or other Pb-containing, Zn-S, Na-Cl, carbona-
ceous, aluminosilicates, carbonaceaous,  or spheres of various
compositions. Such a variety of composition is typical of urban
ambient air samples. The prevalence of the Ca-S particles prob-
ably indicates some impact from the WTC collapse (see previous
section on analysis of bulk  sample).

5.6  Source Particle SEM/EDX Data
The NERL  SEM Laboratory has an on-going project to create an
atlas of various source particulates. The  goal of the NERL
Particle Atlas is to provide a reference tool to assist micro-
scopists in identifying particles and possible sources in ambient
air samples. Each entry in the atlas includes a photomicrograph
and an X-ray  spectrum of a particle (or particles) considered to
be representative of the source type. A description of the sample
and details about the sample' s collection and preparation are also
provided. Selections from the  particle atlas are presented in
Appendix E.
                                                         58

-------
Figure 5-11.  Examples of individual fibers from a fibrous bulk material collected in a parking garage near the WTC site. The
upper-left quadrant of each photomicrograph shows a field of view with the particle of interest within the smaller square in that
field. The upper-right quadrant shows a zoomed-in view of the feature (the area from within the square in the upper-left
quadrant), and the lower half shows the elemental spectrum acquired with the electron beam centered on the small square in
the zoomed-in view.
                                                      59

-------
Figure 5-12. Examples of individual particles (<10 urn) from bulk material collected in a parking garage near the WTC site. The upper-
left quadrant of each photomicrograph shows a field of view with the particle of interest within the smaller square in that field. The
upper-right quadrant shows a zoomed-in view of the feature (the area from within the square in the upper-left quadrant), and the lower
half shows the elemental spectrum acquired with the  electron beam centered on the small square in the zoomed-in view.
                          Figure 5-13.  Randomly selected field of view of VAPS-collected coarse
                          fraction filter showing "honeycomb" particle distribution.
                                                         60

-------
                                                  Chapter 6
                                                 References
Anderson, J.R.; Buseck, P.R.; Saucy, D.A. (1992). Characteri-
zation of Individual Fine-Fraction Particles from the Arctic
Aerosol  at Spitsbergen, May-June  1987.  Atmos. Environ.
26A: 1747-1762.

Anderson, C.A.; Hasler, M.F. (1966). InProc. 4thIntl. Conf. On
X-ray Optics andMicroanalysis (R. Castaing, P. Deschamps and
J. Philibert, eds.) Hermann: Paris, p. 310.

Anderson, J.R.;  Aggett, F.J.;  Buseck, P.R.; Germani, M.S.;
Shattuck, T.W.  (1988).  Chemistry  of  Individual Aerosol
Particles from Chandler, Arizona, an Arid Urban Environment.
Environ. ScL  Technol. 22:811-818.

Andreae, M.O.; Charlson, R.J.; Bruynseels, F.; Storms, H.; Van
Grieken, R.; Maenhut, W. (1986). Internal Mixture of Sea Salt,
Silicates, and Excess  Sulfate in Marine Aerosols. Science
232:1620-1624.

Armstrong, J.T.; Buseck, P.R. (1975). Quantitative Chemical
Analysis of  Individual  Microparticles Using  the  Electron
Microprobe. Theoretical Anal Chem. 47: 2178-2192.

Ayers, G.P. (1978). Quantitative Determination of Sulphate in
Individual Aerosol Particles. Atmos. Environ. 12:1613-1621.

Bernard,  P.C.;  Van  Grieken,  R.E.;  Eisma,  D.  (1986).
Classification of Estuarine Particles Using Electron Microprobe
Analysis and Multivariate  Methods. Environ. ScL Technol.
20:467-473.

Bigg, E.K.; Ono, A.; Williams, J.A. (1974). Chemical Tests for
Individual  Submicron  Aerosol  Particles. Atmos.  Environ.
8:1-13.

Bruynseels, O.F.; Van Grieken, R. (1987). Study of Inorganic
Ammonium Compounds in individual Marine Aerosol Particles
by Laser Microprobe Mass Spectrometry. Analytica Chimica
Acta 195:117-124.
Buseck, P.R.; Bradley, J.P. (1982). Electron Beam Studies of
Individual Natural and Anthropogenic Microparticles: Com-
positions, Structures, and Surface Reactions. In Heterogeneous
Atmospheric Chemistry, Geophysical Monograph 26,  D.R.
Schryer, ed., American Geophysical Union, Washington, DC, pp
57-76.

Buseck, P.R.; Bradley, J.P. (1982). Electron Beam Studies of
Individual Natural and Anthropogenic Microparticles: Compo-
sitions, Structures, and Surface Reactions. In Heterogeneous
Atmospheric Chemistry, Geophysical Monograph 26,  D.R.
Schryer, ed., American Geophysical Union, Washington, DC, pp
57-76.

Carpenter, G.A.; Grossberg, S.; Rosen, D.B. (1991). ART2-A:
An Adaptive Resonance Algorithm for Rapid Category Learning
and Recognition. Neural Networks, 4:493-504.

Casuccio, G.S.; Janocko, P.B.; Lee, R.J.; Kelly, J.F.; Dattner,
S.L.; Mgebroff, J.S. (1983a). The Use of Computer Controlled
Scanning Electron Microscopy in Environmental Studies. /. Air
Pollut. ControlAssoc. 33:937.

Casuccio, G.S.; Janocko, P.B.; Dattner,  S.L.; Mgebroff,  J.S.;
Zalar, J.L. (1983b). Measurement of Ambient Hi-Vol Filters by
Microscopic and Bulk Chemistry Methods. Presented at the 76th
Annual  Meeting of the Air Pollution  Control Association,
Atlanta, GA, June 1983.

Casuccio, G.S.; Schwoeble, A.J.; Henderson, B.C.; Lee, R.J.;
Hopke, P.K.; Sverdrup, G.M. (1988). The Use of CCSEM and
Microimaging  to  Study Source/Receptor  Relationships. In
Receptor Models in Air Resources Management, J.G. Watson,
ed., Air Pollution Control Association Specialty Conference, San
Francisco, CA, February 1988.

Casuccio, G.S. (2002). Personal communication.
                                                        61

-------
Casuccio, G.S.; Lersch, T.L.; Schlaegle, S.F.; Martello, D.V.
(2002). Characterization of Ambient Carbonaceous Particles
Using Electron Microscopy Techniques. American Chemical
Society, Fuel Chemistry Division Preprints 47(1).

Chow,  J.C.;  Watson, J.G.;  Richards, L.W.; Haase, D.L.;
McDade, C.; Dietrich, D.L.; Moon, D.; Sloane, C. The 1989-
1990 Phoenix PM10 Study, Volume II: Source Apportionment;
DRI Document No. 8931.6F1, Desert Research Institute, Reno,
NV, 1991.

Conner, T.L., Norris, G.A., Landis, M.S., and Williams, R.W.
Individual particle analysis of indoor, outdoor,  and community
samples from the  1998 Baltimore paniculate matter study.
Atmospheric Environment 35 (23):3935-3946 (2001).

Costantino,  J.P.;  Janocko,  P.B.; Casuccio,  G.S.  (1983).
Assessment of Thoracic Paniculate Levels at Surface-Mining
Operations. Bituminous Coal Research, Report No. BCR L-
1355, March.

Dockery, D.W.; Pope, C.A.; Xu, X.; Spengler, J.D.; Ware, J.H.;
Fay, M.E.; Ferris, E.G.; Speizer, F.E. (1993). An Association
between Air Pollution and Mortality in Six U.S. Cities. TV. Engl.
J.Med. 329:1753-1759.

Dockery, D.W.; Pope  III,  C.A.  (1994). Acute  Respiratory
Effects of Particulate Air Pollution. Ann. Rev. Public Health
15:107-132.

Draxler, R.D.,  1999. Hysplit_4 User's Guide, Draft  NOAA
Technical  Memorandum, Air Resources Laboratory, Silver
Springs, MD.

Dzubay, T.G.; Mamane, Y. (1989). Use of Electron Microscopy
Data in  Receptor  Models  for  PM-10.  Atmos.  Environ.
23:467-476.

Dzubay, T.G.; Stevens, R.K.; Balfour, W.D.; Williamson, H.J.;
Cooper, J.A.;  Core, J.E.;  De  Cesar, R.T.; Crutcher, E.R.;
Dattner, S.L.; Davis, B.L.;Heisler, S.L.; Shah, JJ.;Hopke,P.K.;
Johnson, D.L. (1984). Atmos. Environ. 18:1555-1566.

Fisher,  G.L.; Prentice, B.A.; Iberman, D.S.; Ondov, J.M.;
Bierman, A.H.; Ragaini, R.C.; McFarland, A.R. (1978). Physical
and Morphological Studies of Size-Classified Coal Fly-Ash.
Environ. Sci. Technol. 12:447-451.

Germani, M.S.; Zoller, W.H. (1994). Solubilities of In-Stack
Suspended Particles from a Municipal Incinerator.  Atmos.
Environ. 28:1393-1400.

Germani, M.S.; Small, M.; Zoller, W.H.; Moyers, J.L.  (1981).
Fractionation of Elements During Copper Smelting. Environ.
Sci. Technol 15:299-305.
Germani, M.S.; Buseck,  P.R.  (1991). Automated Scanning
Electron Microscopy for Atmospheric Particle Analysis. Anal.
Chem. 63:2232-2237.

Germani, M.S. (1991). Evaluation of Instrumental Parameters
for Automated ScanningElectronMicroscopy/GunshotResidue
Particles Analysis. /. Forensic Sci, 331-342.

Ohio, A. I; Kennedy, T.P.; Whorton, R.; Crumbliss, A.L.; Hatch,
G.E.; Hoidal, J.R.  (1992). Role of Surface Complexed Iron in
Oxidant  Generation  and  Lung  Inflammation  Induced  by
Silicates. Am. J. Physiology. 263:L511-L518.

Ohio, A.J.;  Stonehuerner,  J.; Prtichard, R.J.; Piantadosi, C.A.;
Quigley, D.A.; Dreher, K.L.; Costa, D.L. (1996). Humic-like
Substances   in  Air  Pollution Particulates  Correlate  with
Concentrations of Transition Metals and Oxidant Generation.
Inhalation  Toxicology 8:479-494.

Goldstein, J.I.; Newbury, D.E.; Echlin, P.; Joy, D.C.; Romig Jr.,
A.D.; Lyman, C.E.; Fiori, C.; Lifshin, E. (1992). Scanning
Electron Microscopy andX-rayMicroanalysis, Second Edition.
Plenum Press, New York.

Grasserbauer, M. (1977). The Present State of Local Analysis:
Analysis of Individual Small Particles. Mikrochimica Acta
1:329-350.

Hanna, R.B.; Karcich, K.J.; Johnson, D.L. (1980). Determination
of Particle  Identities  Via a Computer Assisted SEM-EDXA
System Scanning Electron Microsc. 1:323-328.

Henry, R.C. Unmix Version 2.3 Manual; 2001, available with
Unmix software (rhenry@usc.edu).

Hock, J.L.;  Lichtman, D. (1983). A Comparative  Study of In-
Plume and In-Stack Collected Individual Coal Fly AshParticles.
Atmos. Environ. 17:849-852.

Henry, R.C. History and Fundamentals  of Multivariate Air
Quality Receptor Models;  Chemom. Intell. Lab. Syst. 1997, 37,
37-42.

Hopke, P.K.; Mi,  Y. (1990). Use of a Rule-Building Expert
System for  Classifying Particles Based on SEM  Analysis. In
Scientific Computing and Automation, E.J. Karjalainen, Ed.,
Elsevier Science Publishers B.V., Amsterdam, pp  179-198.

Hopke, P.K.; Song,  X.-H. (1997).  Classification  of Single
Particles by Neural Networks Based on the Computer-Controlled
Scanning Electron Microscopy Data. Analytica Chimica Acta
348:375-388.
                                                        62

-------
Jambers,  W.; Van  Grieken,  R.  (1997).  Single  Particle
Characterization  of Inorganic  Suspension in  Lake Baikal,
Siberia. Environ. Sci. Technol. 31:1525-1533.

John, W.; Reischl, G. (1980). A Cyclone for Size-Selective
Sampling of Ambient Air. /. Air Poll Control Assoc. 30:872-
876.

Johnson, D.; Hunt, A. (1995). Analysis of Lead in Urban Soils
by Computer Assisted SEM/EDX—Method Development and
Early Results. In Lead in Paint, Soil and Dust: Health Risks,
Exposure Studies, Control Measures, Measurement Methods,
and Quality Assurance, Special Technical Publication 1226, M.
Beard and S. Allen Iske, Eds., American Society for Testing and
Materials, Philadelphia, PA, pp 283-299.

Johnson, D.L.; Mclntyre, B.L.; Stevens, R.K.; Fortmann, R.C.;
Hanna,  R.B.  (1981). A Chemical  Element Comparison of
Individual Particle  Analysis and Bulk  Chemical Method.
Scanning Electron Microsc. 1:469-476.

Johnson, D.L.; Twist, J.P.  (1982).  Statistical Considerations in
the Employment  of SAX Results for Receptor Models. In
Receptor Models Applied to Contemporary Pollution Problems,
APCA Specialty  Conference Proceedings SP-48, S.L. Dattner
and  P.K.  Hopke, Eds.,  Air Pollution Control Association,
Danvers, MA, pp 224-237.

Johnson, D.L.; Davis, B.L.; Dzubay, T.G.; Hasan, H.; Crutcher,
E. R.; Courtney, W.J.; Jaklevic, J.M.; Thompson, A.C.; Hopke.
P.K. (1984b). Chemical and Physical Analyses of Houston
Aerosol for Interlaboratory Comparison of Source Apportion-
ment Procedures. Atmos. Environ. 18:1539-1553.

Johnson, D.L.; Mclntyre, B; Fortmann, R; Stevens, R.K.; Hanna,
R.B. (1984a). Particle Analysis ->• Bulk Analysis.  Chemtech
November:678-683.

Katrinak,  K.A.;  Rez, P.; Buseck,  P.R.  (1992).  Structural
Variations in Individual Carbonaceous Particles from an Urban
Aerosol. Environ. Sci. Technol 26:1967-1976.

Katrinak, K.A.; Anderson, J.R.; Buseck, P.R. (1995). Individual
Particle Types in the Aerosol of Phoenix, Arizona. Environ. Sci.
Technol. 29:321-329.

Kelly, J.F.; Lee R.J.; Lentz, S. (1980). Automated Characteriza-
tion of Fine Particulates. Scanning Electron Microsc. 1:3111.

Kim, D.S.; Hopke, P.K.; Massart, D.L.; Kaufman, L.; Casuccio,
G.S. (1987). Multivariate Analysis of CCSEM Auto Emission
Data. The Science of the Total Environment 59:141-155.
Kim, D.S.; Hopke, P.K. (1988b). Source Apportionment of the
El Paso Aerosol by Particle Class Balance Analysis. Aerosol Sci.
Technol. 9:221-235.

Kim, D.S.;  Hopke, P.K. (1988a).  Classification of Individual
Particles Based on Computer-Controlled Scanning Electron
Microscopy Data. Aerosol Sci. Technol. 9:133-151.

Lee, R.J.; Fasiska, E.J.; Janocko, P.; McFarland, D.; Penkala, S.
(1979).  Electron-Beam Particulate Analysis. Ind. Res.  Dev.
June:25-28.

Lee, R.J.;  Kelly,  J.F. (1980). Applications of  SEM-Based
Automated  Image  Analysis. In Microbeam Analysis,  D.B.
Wittry, ed., San Francisco Press, San Francisco, CA.

Lewis,  C.;  Norris,  G.;  Henry,  R.;  Conner,  T.  Source
Apportionment of  Phoenix  PM25 Aerosol  with the  Unmix
Receptor Model; Journal of the Air and Waste Association (in
press.)

Linton, R.W.;  Williams. P.;  Evans, C.A. Jr.; Natusch, D.F.S.
(1977). Determination of the Surface Predominance of Toxic
Elements in Airborne Particles  by Ion Microprobe Mass
Spectrometry and Auger Electron Spectrometry. Anal. Ghent.
49:1514-1521.

Mamane, Y. (1988). Estimate of Municipal Refuse Incinerator
Contribution to Philadelphia Aerosol Using Single Particle
Analysis—I. Source Analysis. Atmos. Environ. 22:2411-2418.

Mamane, Y.; Willis, R.;  Conner, T. (2001). Evaluation of
Computer-Controlled ScanningElectronMicroscopy Applied to
an Ambient Urban Aerosol  Sample. Aerosol Sci. Technol.
34:97-107.

Mamane, Y.;dePena,R. (1978). A Quantitative Method for the
Detection of Individual Submicrometer Size Sulfate Particles.
Atmos. Environ. 12:69-82.

Mamane, Y. (1990). Estimate of Municipal Refuse Incinerator
Contribution to Philadelphia Aerosol Using Single Particle
Analysis—II.  Ambient   Measurements.  Atmos.  Environ.
246:127-135.

Mamane, Y. (1984). Characterization of Individual Particles
Collected During the Deep Creek Lake Experiment. Paper 84-
58.6, presented at the 77th Annual Meeting of the Air Pollution
Control Association, San Francisco, 24-29 June.

Mamane, Y.; Miller, J.L.; Dzubay, T.G. (1986). Characterization
of Individual Fly AshParticles Emitted from Coal- and Oil-Fired
Power Plants. Atmos. Environ. 20:2125-2135.
                                                        63

-------
Mamane, Y.; Dzubay, T.G.; Ward, R. (1992). Sulfur Enrichment
of  Atmospheric  Minerals  and Spores.  Atmos.  Environ.
26A:1113-1120.
Parungo, F.; Nagamoto, C.; Hoyt, S.; Bravo, H. (1990). The
Investigation of Air Quality and Acid Rain Over the Gulf of
Mexico. Atmos. Environ. 24A:109-123.
Mamane, Y.; Willis, R.; Stevens, R.; Miller, J. (1995). Scanning
Electron Microscopy/X-ray Fluorescence Characterization of
Lead-Rich Post-Abatement Dust. In Lead in Paint, Soil and
Dust: Health  Risks,  Exposure Studies,  Control Measures,
Measurement  Methods,  and  Quality Assurance,  Special
Technical Publication 1226, M. Beard and S. Allen Iske, eds.,
American Society for Testing and Materials, Philadelphia, PA,
pp 268-282.

Marjamcki, M. J.; Keskinen, D. Chen; Pui, D. (1997). Calibration
of the Electrical Low Pressure Impactor (ELPI). Abstracts of the
16th Annual Conference  of the  American Association  for
Aerosol Research, 63.

Martello, D.V.; Anderson, R.R.; White, C.M.; Casuccio, G.D.;
Schlaegle,  S.F.   (2001).  Quantitative  Scanning  Electron
Microscopy  Methods  to Characterize Ambient  Air PM25.
American Chemical Society, Fuel Chemistry Division Preprints
46(1).

Meszaros, A.; Vissy,K. (1974). Concentration, Size Distribution
and  Chemical Nature of Atmospheric Aerosol Particles in
Remote Oceanic Areas. /. Aerosol Sci. 5:101-109.

Michaud,  D.;  Baril, M.;  Perrault, G.  Characterization of
Airborne Dust from Cast Iron Foundries by Physico-Chemical
Methods and Multivariate Statistical Analyses; J. Air & Waste
Manage. Assoc. 1993, 43, 729-735.

Novakov, T.; Chang, S.G.; Marker, A.B. (1974).  Sulfates as
Pollution Particulates:  Catalytic Formation on Carbon (Soot)
Particles. Science 186:259-261.

Obrusnik, I;  Starkova, B.; Blazek, J. (1989). Composition and
Morphology of Stack Emission from  Coal and Oil Fuelled
Boilers.  /.   Radioanalytical  and  Nuclear  Chemistry
133:377-390.

Paatero, P; Tapper, U. (1994). Positive Matrix Factorization: A
Non-Negative Factor Model with Optimal Utilization of Error
Estimates of Data Values. Environmetrics 5:111-126.

Paatero, P. (1997).  Least Squares Formulation of Robust Non-
Negative Factor  Analysis. Chemometrics  and  Intelligent
Laboratory Systems 37: 23-35.

Palotas, A.B.; Rainey, L.C.; Sarofim, A.F.; Vander Sande, J.B.;
Flagan,  R.C.  (1998).  Where  Did  That  Soot Come  From?
Chemtech July:24-30.
Pope III, C.A.; Dockery, D.W.; Schwartz, J. (1995). Review of
Epidemiological Evidence of Health Effects of Paniculate Air
Pollution. Inhal. Toxicol. 7(1): 1-18.

Parungo, F.;  Nagamoto, C.; Rosinski,  J.;. Haagenson,  P.L.
(1986). A Study of Marine Aerosols over the Pacific Ocean. /.
Atmos. Chem. 4:199-226.

Posfai, M.; Anderson, J.R.; Buseck, P.R; Sievering, H. (1999).
Soot and Sulfate Particles in the Remote Marine Troposphere. /.
Geophys. Res. 104:21,685-21,693.

Posfai, M.; Anderson, J.R.; Busick, P.R.;  Shattuck, T.W.;
Tindale, N.W. (1994). Constituents of a Remote Pacific Marine
Aerosol:  a TEM Study. Atmos. Environ. 28:1747-1756.

Post, J.E.; Buseck, P.R. (1984). Characterization of Individual
Particles in the Phoenix urban Aerosol Using Electron Beam
Instruments. Environ. Sci. Technol. 18:35-42.

Pueschel,  R.F. (1976).  Aerosol  Formation During Coal
Combustion: Condensation of Sulfates and Chlorides onFlyash.
Geophys. Res. Lett. 3:651-653.

Ramadan,  Z.; Song,  X.-H.;  Hopke, P.K. Identification  of
Sources of Phoenix Aerosol by Positive Matrix Factorization; J.
Air& Waste Manage. Assoc. 2000, 50, 1308-1320.

Rothenberg,  S.J.;  Denee, P.; Holloway, P. (1980). Coal
Combustion Fly Ash Characterization: Electron Spectroscopy
for Chemical Analysis, Energy Dispersive X-ray Analysis, and
Scanning ElectionMicmscopy.Appl. Spectroscopy 34:549-555.

Sampson,  P.J., Moody,  J.L.,  1981.  Trajectories  as two-
dimensional probability fields In: Air Pollution Modeling and its
Application, Plenum Press, New York, pp.  43-54.

Saucy, D.A.; Anderson, J.R.; Buseck, P.R.  (1991). Aerosol
Particle Characteristics Determined by Combined Cluster and
Principal Component Analysis./. Geophys. Res. 96:7407-7414.

Saucy, D.A.; Anderson, J.R.; Buseck,  P.R.  (1987). Cluster
Analysis Applied  to Atmospheric Aerosol Samples  from the
Norwegian Arctic. Atmos. Environ. 21:2191-2206.

Schwartz, J.  (1994). Air Pollution and Daily Mortality:  A
Review and Meta-Analysis. Environ. Res. 64:36-52.

Shattuck,  T.W.;  Germani,   M.S.; Buseck,  P.R.   (1991).
Multivariate Statistics for  Large Data Sets:  Applications  to
Individual Aerosol Particles. Anal. Chem. 63:2646-2656.
                                                        64

-------
Small, M; German!, M.S.; Small, A.M.; Zoller, W.H.; Moyers,
J.L.  (1981). Airborne Plume  Study of Emissions from the
Processing of Copper Ores in Southeastern Arizona. Environ.
Sci. Technol 15:293-299.

Song, X.-H.; Hopke, P.K.. Solving the Chemical Mass Balance
Problem Using an Artificial Neural Network.  Environ. Sci.
Technol. 30:531-535.

Song,  X.-H.;  Hadjiiski,  L.;  Hopke, P.K.; Ashbaugh, L.;
Carvacho, O.; Casuccio, G.S.; Schlaegle, S. (1999). Source
Apportionment of Soil Samples by the Combination of Two
Neural Networks Based  on Computer-Controlled Scanning
Electron Microscopy.  /. Air &  Waste Manage. Assoc. 49:773-
783.

U.S. Environmental Protection Agency.  National Emissions
Trends Air Pollution  Point Sources Database  (1996).  http://
www. epa. gov/air/data/index. html

Van Cleef, M.; Holt,  S.A.; Watson, G.S.; Myhra, S. (1996).
Polystyrene Spheres on Mica Substrates: AFM Calibration, Tip
Parameters and Scan Artefacts. /. Microsc. 181:2-9.

Van Borm,  W.A.; Adams, F.C;  Maenhaut,  W.A. (1989).
Characterization of Individual Particles in the Antwerp Aerosol.
Atmos. Environ. 23:1139-1151.

Vander Wood, T.; Brown, R. (1992).  The Application of
Automated ScanningElectronMicroscopy/Energy Dispersive X-
ray Spectrometry to the Identification of Sources of Lead-Rich
Particles in Soil and Dust. Environmental Choices Technical
Supplement. 1 July/August.

Watt, J. (1990). Automated Feature Analysis in the Scanning
Electron Microscope. Microscopy and Analysis, January.

Wienke,  D;  Xie,  Y; Hopke, P.K.  (1994).  An Adaptive
Resonance Theory Based Artificial Nueral Network (ART-2a)
for Rapid Identification of Airborne Particle Shapes from Hteir
Scanning  Electron Microscopy Images.  Chemometrics and
Intelligent Laboratory Systems 25:367-387.

Williams, R., Suggs, J., Zweidinger, R., Evans, G., Creason, J.,
Kwok, R.,  Rodes, C., Lawless, P., Sheldon, L., 2000a. The 1998
Baltimore paniculate matter epidemiology-exposure study: part
1 - comparison of ambient, residential outdoor, indoor and
apartment  paniculate  matter monitoring. Journal of Exposure
Analysis and Environmental Epidemiology, 10: 518-532.
Williams, R., Suggs, J.,  Creason, J., Rodes, R., Lawless, P.,
Kwok, R.,  Zweidinger,  R., Sheldon, L.,  2000b.  The  1998
Baltimore paniculate matter epidemiology-exposure study: part
2 - personal exposure assessment associated with an elderly
study  population.  Journal   of  Exposure  Analysis  and
Environmental Epidemiology, 10: 533-543.

Willis, R.D.; Ellenson, W.D.; Conner, T.L. (2001). Monitoring
and Source Apportionment of Paniculate Matter near a Large
Phosphorus Production Facility. /. Air Waste Manage. Assoc.
51:1142-1166.

Wurster, R.; Ocker, B. (1993). Investigation of Nanoparticles by
Atomic and Lateral Force Microscopy. Scanning 15:130-135.

Xhoffer,   C.;  Wouters,  L.;   Van  Grieken,  R.  (1992).
Characterization of Individual Particles in the North Sea Surface
Microlayer  and   Underlying  Seawater:   Comparison  with
Atmospheric Particles. Environ. Sci. Technol. 26:2151-2162.

Xie, Y.; Hopke, P.K.; Wienke, D. (1994).  Airborne Particle
Classification with a Combination of Chemical Composition and
Shape Index Utilizing an Adaptive Resonance Artificial Neural
Network. Environ. Sci. Technol 28:1921-1926.

Zayed, J.;  Hong, B.; L'Esperance, G. Characterization  of
Manganese-Containing Particles Collected from the Exhaust
Emissions of Automobiles Running with MMT  Additive;
Environ. Sci. Technol. 1999, 33 (19), 3341-3346.
                                                        65

-------
        Appendix A




Example of LEO/PGT Output



Example of PSEM Output
             66

-------
                         Example of LEO/PGT Data Output
Sample: Brownsville 159609    Units:  micrometers

Number of features:
   Total:                         323
   Excluded by selection formula:   99* (See note at bottom)
   Deleted manually:                 0
   Remaining in analysis:          224

Area of all fields:
   Total    :    7.2256e+04
   Remaining:    7.0033e+04

Area of all features:
   Total    :    6.5718e+03 (   9.10% of field)
   Remaining:    4.3491e+03 (   6.02% of field)
                (   6.21% of the remaining field)

Area of all features  filled:
   Total    :    6.6053e+03 (   9.14% of field)
   Remaining:    4.3615e+03 (   6.04% of field)
                (   6.23% of the remaining field)


Feature classification report:
   Class name    Total    % of Analyzed  % of All
   Al_Si:          42           18.75         18.75
   Quartz:         16            7.14          7.14
   Ca_rich:        25           11.16         11.16
   Ca_S:           19            8.48          8.48
   Organic:        22            9.82          9.82
   Metal_rich:      2            0.89          0.89
   trace_metals:     1            0.45          0.45
   Na_Cl:          11            4.91          4.91
   Mixed_mineral:   45           20.09         20.09
   Na_enriched:    29           12.95         12.95
   Mg_enriched:    17            7.59          7.59
   S_enriched:     18            8.04          8.04
   Cl_enriched:    50           22.32         22.32
   No class match: 47           20.98         20.98

* Note added by authors: The  "selection formulas" are user-defined rules  which
determine whether a given feature  will be included or excluded in the  feature analysis
report above.  Selection formulas use  a comprehensive set of  arithmetic operators  and
functions and basic logic operators and can be applied to virtually any measurable
physical  (not chemistry) particle  parameter (e.g.,  feature number,  x and  y coordinates,
feature diameters, area, perimeter, volume, aspect ratio,  circularity,  roughness).  The
conditions in the selection formulas  are tested and if they  are false  for a given
feature, then the feature is  excluded. If they are true,  then the feature is included
in the feature analysis report. A common application of the  selection  formula is  to
exclude features  that  are either too  small or too large.
                                          67

-------
Sample: Brownsville 159609   Units: micrometers

Measurement    Average    Median    Minimum    Maximum   1 Std. Deviation

Avg Diameter     5.61      4.71       3.01       14.56        2.61

Sample: Brownsville 159609    Units: micrometers

Field  Area   #Features   Area of Features   ft/Area   Area Fraction   urn/Pixel    1
72256             323            6605        0.00447     0.0910        0.5940



Sample: Brownsville 159609    Units: micrometers

Feature#*  Field   Avg Diameter   Circularity   Volume  Classes

  1          1        3.39            1.77        9.0    Organic
                                                         Mixed mineral
2
5
6
7
10
11
13
14
16
17
18
1
1
1
1
1
1
1
1
1
1
1
3
7
3
3
8
3
4
7
8
4
10
.01
.58
.14
.02
.01
.21
.99
.22
.16
.24
.76
2
2
2
1
1
2
1
2
1
1
2
.05
.08
.24
.67
.6
.12
.84
.86
.90
.89
.33
5.
90.
5.
7.
143.
6.
27.
47.
119.
18.
189.
16
24
52
81
64
29
24
75
57
83
79
Na enriched
Cl enriched
not classified
not classified
not classified
Ca rich
S enriched
not classified
Organic
Organic
Ca rih
Cl enriched
Mixed mineral
                                                         Cl_enriched

 21          1        4.52            2.43       17.45   Na_Cl

*Note added by authors: "Feature number" is simply the number, in sequential order,
assigned to each particle as it is identified by the image-processing software. Each
particle identified in a field image is assigned a unique number, but only those
particles which meet the criteria set in the selection formulas are included in the
feature report. Thus, particles 3, 4, 8, 9, 12, 15, 19, and 20 were identified by the
image processing software but excluded by the selection formula.
                                           68

-------
                            Example of PSEM Data Output
Client Name mukerjee
Client Number
Project Number 0001
Sample Number 159601
Analysis Date
Operator
Instrument

11/22/96
rdw


Personal SEM
Mag Fields
600 4.
2000 39
Classes
Calcite
Ca-silicate
Silicate
Si/CaCl
CaCl
Quartz
Misc.
Ca-sulf ate
Salt/CaCl
MgSi
Si/Salt
Na-silicate
Salt
Fe-rich
Cl-rich
Na-rich
Sulfate
Salt/CaS04
AISi
Barite
Fe-silicate
Si/CaS04
Dolomite
Fe-spheres
Totals
3584
Grid
0.579


.1068 0.174
#
366
103
150
65
250
58
49
256
39
4
18
15
289
4
141
29
115
42
2
1
1
1
1
1
2000
Number %
16.76
4.63
6.83
2.94
12.71
2.59
2.42
13.92
1.87
0.19
0.81
0.68
15.65
0.18
7.61
1.47
6.21
2.20
0.09
0.04
0.04
0.04
0.06
0.06
100.00
Area %
37.12
14.13
14.77
7.38
7.75
6.10
3.21
1.28
1.64
0.52
1.35
0.88
1.35
0.29
0.85
0.36
0.53
0.31
0.14
0.02
0.01
0.01
0.01
0.00
100.00
                                   Vol %
                                   38.80
                                   13.69
                                   15.07
                                    6.92
                                    6.31
                                    8.07
                                    3.31
                                      74
                                      40
                                      24
                                      07
                                    0.69
                                    0.47
                                    0.31
                                    0.29
                                    0.24
                                    0.21
                                    0.11
                                    0.05
                                    0.01
                                    0.00
                                    0.01
                                    0.00
                                    0.00
                                  100.00
  Wt %
 41.87
 13.38
 13.31
  6.91
  6.68
  6.42
    59
    54
    49
    12
    07
  0.66
  0.51
  0.50
  0.33
  0.25
  0.19
  0.11
  0.05
  0.01
  0.01
  0.00
  0.00
  0.00
100.00
Counts
  2687
  3833
  3516
  3789
  2239
  2797
  2923
  2110
  3628
  2952
  3500
  3989
  2783
  3012
  1953
  2536
  1928
  2245
  1411
  2508
  3620
  3270
  1855
  2329
  2666
 Area Fr.
0.1358812
0.0517038
0.0540556
0.0270002
0.0283688
0.0223359
0.0117516
0.0047001
0.0060083
0.0018956
0.0049276
0.0032083
0.0049404
0.0010538
0.0031066
0.0013047
0.0019496
0.0011334
0.0005012
0.0000840
0.0000385
0.0000499
0.0000190
0.0000012
0.3660194
  Num/cmA2*
3.920E+005
1.084E+005
1.598E+005
 .868E+004
 .974E+005
 .062E+004
 .669E+004
3.256E+005
4.375E+004
4.430E+003
1.906E+004
1.593E+004
3.661E+005
4.181E+003
1.780E+005
3.429E+004
1.453E+005
5.137E+004
 .090E+003
 .045E+003
 .045E+003
 .045E+003
 .294E+003
1.294E+003
2.339E+006
* Note added by authors: "Num  /  cmA2",  last column  of data, is the  number of particles
per square  cm  of sample for  the  listed class.  Typically, only a  small fraction of  a
square cm of a sample is analyzed.  Therefore  the  values in this  column are simple
extrapolations based on the  number of particles in  the listed class which were
identified  in  the area of  sample actually analyzed.  The extrapolated Num/cm2 therefore
assumes a uniform particle loading over the entire  sample.
                                             69

-------
Client_Name     mukerjee
Client Number
Project_Number  0001
Sample_Number   159601
AnalysisJDate   11/22/96
Operator        rdw
Instrument      Personal SEM

Aerodynamic Mass Distribution by Aerodynamic Diameter  (microns)
                           0.0   0.7   1.3   2.0   2.7   3.3    4.0    4.7    5.3    6.0
6.7   7.3   8.0   8.7   9.3
Classes
7.3 8.0
Calcite
2.0 1.5
Ca-silicate
2.0 3.4
Silicate
2.6 3.4
Si/CaCl
0.9 6.5
CaCl
1.9 0.7
Quartz
1.8 1.1
Misc .
2.8 1.2
Ca-sulfate
0.0 4.4
Salt/CaCl
11.9 11.3
MgSi
0.0 0.0
Si/Salt
0.0 5.2
Na-silicate
0.0 0.0
Salt
0.0 0.0
Fe-rich
4.7 0.0
Cl-rich
0.0 0.0
Na-rich
22.8 0.0
Sulfate
0.0 0.0
Salt/CaS04
0.0 50.0
Mass
8.7
41
2

3

3

3

3

1

5

0



0

18

9

0

0

0



43

0
.1
13
.3
13
.8
6
.1
6
.8
6
.2
3
.3
1
.0
1
0.0
1
.0
1
.1
0
.9
0
.0
0
.0
0
.0
0
0.0
0
.6
0
.0
9
.9
1
.4
1
.3
3
.9
5
.7
3
.4
1
.6
0
.5
0
.5

.1
0
.1
0
.7
0
.5
0
.5
0
.3
0
.3

.2
0
.1
0
«
.3
0.
.7
0.
.9
0.
.9
0.
.0
0.
.6
0.
.1
0.
.0
0.
.0
0.
4.6
0.
.0
0.
.0
0.
.0
0.
.0
0.
.0
0.
.0
0.
0.0
0.
.0
0.
.0
<
10
0
1
0
3
0
7
0
2
0
2
0
5
0
0
0
8
0

0
0
0
0
0
0
0
0
0
0
0
0
0

0
0
0
0
0.
.0
0.
.9
0.
.0
0.
.4
0.
.7
0.
.3
0.
.6
0.
.0
0.
.9
0.
6.8
0.
.0
0.
.0
0.
.0
0.
.0
0.
.0
0.
.0
0.
0.0
0.
.0
0.
.0
7
»
0
86.
0
83.
0
70.
0
75.
0
76.
0
81.
0
86.
1
79.
0
54
0
98.
0
68.
0
71.
0
62.
0
87.
0
64.
0
52
3
0.
0
0.
1
>
0
8
0
0
0
6
0
4
0
4
0
7
0
2
1
5
0
.6
0
9
0
4
0
2
5
6
0
3
4
2
0
.4
5
0
3
0
.3
.0

.0

.0

.0

.2

.0

.0

.7

.1

.0

.0

.0

.6

.0

.6

.8

.8

.2

2
0

0

0

0

0

0

0

1

0

0

0

0

10

0

7

1

7

5

.0
.0

.0

.1

.0

.4

.0

.2

.3

.1

.0

.1

.0

.7

.0

.4

.1

.6

.7

2
0

0

0

0

0

0

0

1

0

0

0

0

6

0

4

1

3

3

.7
.1

.0

.0

.1

.5

.2

.1

.1

.1

.0

.0

.0

.3

.0

.5

.7

.9

.7

3
0

0

0

0

0

0

0

1

0

0

0

1

4

0

2

3

1

14

.3
.1

.1

.2

.0

.4

.5

.1

.4

.4

.0

.4

.3

.5

.0

.1

.7

.3

.8

4
0

0

0

0

1

0

0

0

0

1

0

1

5

0

3

3

9

13

.0
.4

.3

.8

.2

.0

.2

.0

.3

.4

.0

.8

.6

.4

.6

.4

.0

.8

.4

4
0

0

0

0

1

1

0

0

1

0

0

4

0

0

0

4

8

9

.7
.6

.2

.7

.1

.2

.1

.4

.9

.7

.0

.9

.4

.0

.0

.0

.0

.5

.2

5
0

0

1

0

2

2

0

0

3

0

3

2

3

0

0

2

8

0

.3
.8

.6

.2

.5

.1

.5

.7

.4

.2

.0

.1

.6

.8

.0

.0

.8

.7

.0

6
0

1

2

2

1

1

2

0

2

0

0

6

1

0

13

7

10

0

.0
.7

.2

.3

.8

.8

.2

.3

.0

.6

.0

.6

.3

.2

.0

.9

.8

.6

.0

6.7
1.3

0.9

3.1

2.7

3.8

1.8

0.6

0.0

2.1

0.0

2.4

2.6

0.0

7.4

0.0

0.0

0.0

0.0

                                           70

-------
         Appendix B




Submittal Form for SEM/EDX Samples



Assignment of SEM IDs



Sample Log-In/Log-Out
              71

-------
                         Submittal Form for SEM/EDX Samples

Project Name (8 characters or less): 	Date:	
6-digit XRFID (if assigned by the XRF Lab):

O/?Your 6-digit Sample ID:	
Description (what are they, how collected, particle size, chemistry, etc. If samples have been prepped,
sieved, or deposited on filters or substrates, please describe): Respond in the expanding boxes below.
NOTE: Samples must be vacuum-compatible. Loose samples not collected on appropriate substrates
will require additional prepping by us — SUGGESTIONS?
What information do you want SEM/EDX to provide?
             	 Quick evaluation?
             	 Particle size distribution (specify upper and lower size limits')
             	 Chemical class distribution
             	 Pretty pictures (Polaroids and/or . TIF files')
             	 Search for specific type of particle
             	 Please call me to discuss
             	 Special requests or requirements (describe):
How do you want the results? (Specify desired electronic format).
       	Hard copy                 	Electronic copy
       	Text report                 	Spreadsheet          	Photomicrographs
       	Size distribution histograms 	Oral debriefing        	Other (specify)
Contact person:
Tel: 	  Fax: 	  E-mail:

When do you need results?	
Sample disposition:  return to client	      archive in SEM lab	         discard

SUBMIT ELECTRONIC FORM TO TERI CONNER, SEM WORK ASSIGNMENT MANAGER
Tel: (919) 541-3157   email: Conner.Teri@epamail.epa.gov
                                               72

-------
                 Assignment of SEM Ids
   o»
   GO
   
u.
O



|1

H
OJ (0 co
   
-------
                         Sample Log-In/Log-Out
                              SAMPLE
                                      OG-IN
        OG-OUT
Date
Sample
                              Number
Delivered
                                             SEMIDs
SEMIDs
                                                Sample
Samples,
Samples
Description
                              Of
                                     To Lab
                .EO/PGT
        lumbers
                                                                    Returned
Received
                              Samples
iJL
                                                            Returned
                                                                    o
                                               S
        fw 3,5"  T
                                               y
3-j-oo
                                      r
                                               s
3-lV-oo
          -^
                               /9
                                       r.
                                                \/
        o/rc
         6v rc
         Po i-y C«S
                                     74

-------
                 Appendix C




• Particle classification rules for PM10_2 5 (coarse) particles



• Particle classification rules for PM2 5 (fine) particles
                       75

-------
                     Particle classification rules for PM10.25 (coarse) particles
 Particle Class
Classification Rule
 aerodiam
 Fly Ash

 Pb containing
 Crrich
 Zn rich
 Cu rich
 Mn rich
 Other Mn
 High Ti
 Ti rich
 Ti rich al-silicates
 Other Ti
 Ni and/or V
 Ba containing
 High Fe
 Fe rich
 Fe rich al-silicates
 Other Fe
 Ca and Mg rich
 Ca and S rich

 High Ca
 Ca rich
 Ca rich al-silicates
 Other Ca
 K and S  rich
 K and P rich
 Quartz
 Al-Zr-Cl
 Al rich
 Talc

 Mixed al-silicates
AeroDiaa < 2.495
Aspectb < 1.5 and Al > 20 and Al < 50 and Si > 40 and Si < 60 and K >= 2 and K < 8 and Mg <= 5
and Na < 3 and Cl< 3 and Fe >= 1.15 and counts >= 1000
Pb >= 20 and counts >= 1000 or Pb >= 20 and video0 >= 120
Cr >= 50 and counts >= 1000 or Cr >=  50 and video >= 120
Zn >= 50 and counts >= 1000 or Zn >= 50 and video >= 120
Cu >= 50 and counts >= 1000 or Cu >= 50 and video >= 120
Mn >= 50 and counts >= 1000 or Mn >= 50 and video >= 120
Mn >= 20 and counts >= 1000 or Mn >= 20 and video >= 120
Ti >= 70 and counts >= 1000
Ti >= 50 and counts >= 1000 or Ti >= 50 and video >= 120
Al >= 10 and Si >= 20 and Ti >= 10 and Si >= Ti and counts >= 800 and video >=80
Ti >= 20 and Ti > Ca and Ti > Fe and counts >= 800 and video >= 80
Ni + V >= 20 and counts >= 1000 or Ni + V >= 20 and video >= 120
Ba >= 10 and Ba > Fe and counts >= 1000 or Ba >= 10 and Ba > Fe and video >= 120
Fe >= 70 and counts >= 1000
Fe >= 50 and Ca < 25 and S < 25 and Cl < 25 and Si <  10 and Al < 10 and Ba < 10 and counts >=
1000 or Fe >= 50 and Ca < 25 and S < 25 and Cl< 25 and Si < 10 and Al < 10 and Ba < 10 and video
>= 120
Al >= 10 and Si >= 20 and Fe >= 10 and Si >= Fe and Fe > Ca and counts >= 800 and video >=80
Fe >= 20 and counts >= 800 or Fe >= 20 and video >= 80
Mg >= 10 and Ca >= 25 and Mg > S and Mg > Si and Ca > Si and counts >= 800 and video >= 80
Ca + S >= 50 and Ca >= 20 and S >= 20 and Ca > Si and counts >= 800 and video >= 80 or S >= 30
and Ca >= 30 and Ca > Si and counts >= 800  and video >= 80
Ca >= 70 and counts >= 800 or Ca >= 70 and video >= 80
Ca >= 50 and counts >= 800 or Ca >= 50 and video >= 80
Al >= 10 and Si >= 20 and Ca >= 10 and Si >= Ca and counts >= 800 and video >=80
Ca >= 20 and counts >= 800 and video  >= 80
S >= 10 and S > P and K >= 20  and K + S >= 40 and counts >= 800 and video >= 80
P >= 40 and K >= 10 and K + P >= 60 and counts >= 800 and video >= 80
Si >= 75 and Al < 10 and counts >= 1000
Al >= 30 and Zr >= 20 and Cl >= 3 and video >= 100 and counts >= 1000
Al >= 50 and counts >= 800 or Al >= 50 and video >= 80
Mg >= 10 and Al < 10 and Si >= 40 and Ca < 10 and Fe < 10 and counts >= 800 and video >= 80 or
Mg >= 15 and Si >= 40 and counts >= 500
Al >= 10 and Si >= 20 and counts >= 800 and video >= 80 or Si >= 40 and counts >= 800 and video
>=80
Na + Cl >= 40 and counts >= 1000 or Na + Cl >= 40 and video >= 120
Counts >= 1000
Counts < 1000 and counts >= 500
Counts < 500
 Salt
 Other - High Counts
 Other - Mid Counts
 Other - Low Counts
Aerodynamic diameter
baspect ratio (ratio of maximum diameter to perpendicular diameter)
cgrayscale brightness value
                                                    76

-------
                        Particle classification rules for PM25 (fine) particles
 Particle Class
Classification Rule
 aerodiam
 Salt/Marine

 Sulfate

 Other - Low Counts
 Fly Ash

 Pbrich
 Other Pb
 Bi rich
 Other Bi
 Crrich
 Other Cr
 Zn rich
 Other Zn
 Cu rich
 Other Cu
 Mn rich
 Other Mn
 High Ti

 Ti rich
 Other Ti
 High Fe

 Fe rich
 Ca and Mg rich
 Ca and S rich
 High Ca
 Ca rich
 K-S-P rich

 Quartz
 Fe rich al-silicates

 Other Fe
 Ba containing
 Ni and/or V
 Al-Zr-Cl
 Al rich
 Talc
 Other Ca
 Mixed al-silicates
 Other - High Counts
 Other - Mid Counts
aaerodynamic diameter
baverage diameter
AeroDiaa >= 2.495
Na >= 5 and Cl >= 20 and Cl > Si and Cl > Al orNa + Cl >= 30 and Cl > Si and Cl > Al or S >= 10
and Cl >= 10 and Ca >= 10 and Cl > Si and Cl > Al or Na + Mg + S + Cl + Br >= 70 and Ca < 20
S >= 40 and counts >= 250 and Daveb >= 0.3 and Na < 10 and Mg < 10 and Al + Br < 10 and Si < 10
andCl 10andK< 10andCa< 10andFe< 10
Counts < 500
Aspect0 < 1.5 and Al + Br > 20 and Al + Br< 50 and Si > 40 and Si < 60 and K >= 2 and K < 8 and
Mg <= 5 and Na < 3  and Cl< 3 and Fe >= 1.15 and Fe < 10 and Ca <  10
Pb >= 50 and videod  >= 100 and counts >= 1000
Pb >= 20 and counts  >= 750 and  video >= 90
Bi >= 50 and video >= 100 and counts >= 1000
Bi >= 20 and counts  >= 750 and  video >= 90
Cr >= 50 and counts  >= 1000 or Cr >= 50 and video >= 80
Cr >= 20
Zn >= 50 and counts >= 1000 or Zn >= 50 and video >= 80
Zn>=20
Cu >= 50 and counts >= 1000 or  Cu >= 50 and video >= 80
Cu >= 20
Mn >= 50 and counts >= 1000 or Mn >= 50 and video >= 80
Mn>=20
Ti >= 70 and counts >= 750 and video >= 90 or Ti >= 70 and counts >= 1000 or Ti >= 70 and video
>= 100
Ti >= 50 and counts >= 750 or Ti >= 50 and video >= 80
Ti >= 20
Fe >= 70 and counts  >= 750 and video >= 90 or Fe >= 70 and counts >= 1000 or Fe >= 70 and video
>= 100
Fe >= 50 and counts  >= 750 or Fe >= 50 and video >= 80
Mg >= 10 and Ca >= 25 and Mg > S and Mg > Si
Ca + S >= 50 and Ca >= 20 and S >= 20 or S >= 30 and Ca >= 30
Ca >= 70 and counts >= 1000 or Ca >= 70 and video >= 80
Ca >= 50 and counts >= 1000 or Fe >= 50 and video >= 80
K + S + P >= 60 and K >= 10 and K > Ca or K + S >= 40 and K >= 10 and K > Ca or K + P >= 40 and
K>=10andK>Ca
Si >= 75 and Al + Br < 10 and counts >= 1000 or Si >= 75 and Al + Br < 10 and video >= 80
Si >= 10 and Al + Br >= 10 and Fe >= 10 and Si > S and Fe > Ti and video >= 80 or Si >= 10 and Al
+ Br >=10 and Fe >= 10 and Si > S and Fe > Ti  and counts >= 1000
Fe >= 20
Ba >= 10 and Ba > Fe and counts >= 1000 or Be >= 10 and Ba > Fe and video >= 80
Ni + V >= 20 and counts >= 1000 or Ni + V >=  20 and video >= 80
Al + Br >= 30 and Zr >= 20 and Cl >= 3
Al + Br >= 50 and counts >= 750 or Al + Br >= 50 and video >= 90
Mg >= 10 and Al + Br < 10 and Si >= 40 and Ca < 10 and Fe <
Ca >= 20
Al + Br + Si >= 30 and Al + Br >= 10 and Si >= 20 or Si >= 40
Counts >= 1000
Counts < 1000 and counts >= 500
                             °aspect ratio  (ratio of maximum diameter to perpendicular diameter)
                                                          dgrayscale brightness value
                                                    77

-------
                     Appendix D



Particles from ambient air sample collected in NYC near WTC site
                          78

-------

-------

-------

-------

-------
  Appendix E



Source particle atlas
        83

-------
                                      Tire Wear Debris
Sample Collection: Sample was collected following a routine auto emissions test in the EPA dynamometer
facility. The 1-hr test simulated stop and go driving (brakes were applied during the test), and highway
driving. The sample was collected from the front-axle roller which was wiped clean before the
dynamometer test. At the conclusion of the test, the portion of the roller which had been under the front
passenger wheel was wiped with a 47-mm nuclepore filter. A section of the filter was removed for
SEM/EDX analysis.

Particle Description: Particles tend to be elongated, elliptical or cigar-shaped. Bright inclusions in BSE
image are Fe-rich aluminum silicates or pure iron.
Submitted by: Bob Willis
Date: August 27, 2001
                                               84

-------
                                Tire Wear Debris, Page 2
Bright inclusions are mostly rich in Fe or
Fe-Si. Occasional Ti-rich inclusions
                                             85

-------
                                    Tire Wear Debris, Page 3
              ,• lira!    Sp*ctrun Collection
                       (, 0       fl.0
1-0      3.0       40       g 0
                               S.fl      10.0
                                                   86

-------
                   Mushroom Spores (Chloropyllum molybdites)
Sample Collection: Contributed by Herb Jacumin.

Particle Description: Spores seem to hold up well to beam heating and vacuum.

Submitted by: Bob Willis
Date: September 7, 2001
                                          87

-------
                                       Ragweed Pollen
                       Ragweed
0.0
           2.0
                       4.0
                                   6.0
                                     keV

                                               8.0
                                                           10.0

                                                                  EHT = 20.00 kV
                                                                                 Detector = SE1
                                                                            13.0
Sample Collection: Southern ragweed. Collected by John Miller.

Particle Description: EDX spectrum is mostly C and O with minor P,S, C, K, Ca. The Al peak is generated
by the underlying Al stub.
Submitted by: Bob Willis
Date: September 7, 2001

-------
                                     Calcite Powder
Sample Collection:

Particle Description: Aluminum is
contaminant in calcite powder. Particles
show rectangular cleavage.

Submitted by: Bob Willis
Date: September 13, 2001
                                            89

-------
                                      Dandelion Pollen
                      ,/ dandelion  Spect
0.0
           2.0
                       4.0
                                   6.0
                                                8.0
                                     keV
                                                           .**
                                                                 rl

                                                                  EHT = 20.00 kV
                                                                                 Detector = SE1
                                                          —-r-
                                                           10.0
                                                                            13.0
Sample Collection:



Particle Description: EDX spectrum is mostly C and O with minor P, S, Cl, K, and Ca. The Al peak is

generated by the stub.



Submitted by: Bob Willis

Date: September 13, 2001
                                               90

-------
                                Flyash #8730 (Superfine)
Sample Collection: Sample provided by John Miller. Source unknown.

Particle Description: EDX spectrum is dominated by Aluminum. BSE image shows that spheres have
different composition. Brightest spheres are very Fe-rich. Medium bright spheres approach pure Al. Dull
spheres are Aluminum-silicates with composition similar to spectrum above.

Submitted by: Bob Willis
Date: September 13, 2001
                                             91

-------