United States
            Environmental Protection
          Environmental Monitoring Systems
          Research Triangle Park NC 2771 1
EPA-GOO 9-84-019
November 1 984
            Research and Development

National Symposium on
Recent Advances in
Pollutant Monitoring of
Ambient Air and
Stationary Sources

                                             November 1984


              Radison Plaza Raleigh Hotel

                     May 8-10, 1984
             U.S. Environmental Protection Agency
             Region 5, library 

This document has been reviewed in accordance with
U.S. Environmental Protection Agency policy and
approved for publication.  Mention of trade names
or commercial products does not constitute endorse-
ment or recommendation for use.


     Measurement and monitoring research efforts are designed to anticipate
potential environmental problems, to support regulatory actions by develop-
ing an in-depth understanding of the nature and processes that impact health
and the ecology, to provide innovative means of monitoring compliance
with regulations and to evaluate the effectiveness of health and environ-
mental protection efforts through the monitoring of long-term trends.
The Environmental Monitoring Systems Laboratory, Research Triangle Park,
North Carolina, has the responsibility for:  assessment of environmental
monitoring technology and systems; implementation of agency-wide quality
assurance programs for air pollution measurement systems; and supplying
technical support to other groups in the Agency including the Office of
Air, Noise and Radiation, the Office of Pesticides and Toxic Substances
and the Office of Solid Waste and Emergency Response.

     This symposium is part of a continuing effort to explore recent advances
in pollutant monitoring of ambient air and stationary sources.  It serves as
a forum for exchange of ideas and scientific information.  In response to
the Agency regulatory needs, this symposium focused on acid deposition,
personal exposure and toxic substances.  This publication is intended to
assist those researchers interested in furthering the science of air
                            Thomas R. Mauser, Ph.D.
                                   Di rector
                  Environmental Monitoring Systems Laboratory
                     Research Triangle Park, North Carolina

                             TABLE OF CONTENTS

Foreword 	 m
Introduction 	 vii
PM-10 Instruments: A Manufacturer's Perspective	1
Generation and Use of Large, Solid Calibration Aerosols	8

A Size Classifying Isokinetic Aerosol Sampler Designed for
  Application at Remote Sites	9
Particle and Substrate Losses During Shipment of Teflon and
  Quartz Filters 	 12
Pollutant Losses in Dichotomous Samplers 	 24
Mass Distribution of Large Ambient Aerosols and Their Effect on
  PM-10 Measurement Methods	28
Rotary Impactor for Coarse Particle Measurement - Mass and
  Chemical Analysis	33
Individual Micrometer-Size Aerosol Compounds 	 36
Human Exposure Assessment:  A New Methodology for Determining the
  Risk of Environmental Pollution to Public Health  	 52
Results of the Carbon Monoxide Study in Washington, D.C., and
  Denver, Colorado, in the Winter of 1982-83  	 57
A Review of  Indoor Air Quality Research at Oak Ridge National
  Laboratory	61
Passive Sampling Devices with Reversible Adsorption:  Mechanics
  of Sampling	67
Portable  Instrument for the Detection and  Identification of Air
  Pollutants	73
Problems  and  Pitfalls of Trace Ambient Organic Vapor Sampling
  at Uncontrolled Hazardous Waste Sites	82
New Continuous Monitoring Systems for Measurement of
  Hazardous  Pollutants 	 91
Reagent Impregnated Film Badges for  Passive Pollutant Sampling  	 96
A Cryogenic  Preconcentration-Direct  Flame  lonization Method
  for Measuring Ambient NMOC	104
Mobile Air Monitoring by MS/MS -  A Study of the TAGAR 6000 System.  .  .  .109
Development  of Surface-Enhanced Raman Spectroscopy  for
  Monitoring  Toxic Organic Pollutants	113

Thermal Desorption Techniques for the Gas Chromatographic
  Analysis of Particulate Matter	115
A Method to Specify Measurements for Receptor Models	127
The Application of SIMCA Pattern Recognition to Complex Chemical
  Data	131
Description of a Continuous Sulfuric Acid/Sulfate Monitor	140
Automated Sampling and Analysis of Flue Gases from Plasma Pyrolizer. .  .152
The Ratio of Benzo(a)pyrene to Particulate Matter in Smoke from
  Prescribed Burning	161
Volatile Organic Sampling Train (VOST) Development at MRI	171
An Evaluation of Instrumental Methods for the Analysis of Vinyl
  Chloride in Gaseous Process Streams	180
Overview of Semiconducting Gas Sensing Devices	193
Examination of Calibration Precision Calculations and Protocols for
  Air Monitoring Data	198

     The fourth annual national symposium sponsored by EPA'e Environmental
Monitoring Systems Laboratory was held May 8-10, 1984 in Raleigh, North
Carolina.  In seven sessions over three days, papers and discussions
focused on state-of-the-art systems for monitoring source emissions,
ambient air, acid deposition, hazardous emissions and personal monitoring.
The sessions were categorized as follows:

     SESSION I     Particulate Pollutants
     SESSION II    Personal Monitoring
     SESSION III   Hazardous Waste Monitoring
     SESSION IV    Organic Pollutants
     SESSION V     Analysis of Complex Chemical Data
     SESSION VI    Acid Deposition
     SESSION VII   General and Source Oriented Monitoring
     The papers are in the same order as presented by the speakers.
Several papers are omitted because the speakers did not submit them
in time for the agency's peer review.

                By:   Michael L.  Smith,  Andersen Samplers,  Inc.
     Andersen Samplers, Inc. and its subsidiaries, General Metal Works (GMW)
and Sierra-Andersen (S-A),  have been manufacturing and marketing size specific
particulate samplers since the mid-1970's.   The U. S.  Environmental Protection
Agency recently proposed revisions to the National Ambient Air Quality
Standard (NAAQS) for particulate matter     which would base the primary,
health-related standard on only those particles smaller than 10 micrometers
aerodynamic diameter (PM-10).
    Both GMW and S-A manufacture and market a complete line of PM-10 instru-
ments, including Medium Flow Samplers, Dichotomous  Samplers and Size Selective
High Volume Samplers.  Although the instruments sold by each company are
designed to measure the same particulate pollutants, the collection mechanisms
are different.  The GMW instruments are based upon  cyclonic collection whereas
the S-A instruments are based upon impaction.
    Each of the three types of instruments have specific features which give
them certain advantages in certain applications.  Both the Medium Flow and
the Dichotomous Samplers utilize high vacuum pumps  and therefore can use
Teflon membrane filters to collect the PM-10 particles.  The use of Teflon
filters allows subsequent chemical analysis using x-ray flourescent analysis.
The Dichotomous operates at a low flowrate (16.7 1pm)  but separates "fine"
mode particles smaller than 2.5 micrometers from "coarse" mode particles in
the range of 2.5 to 10 micrometers.  The Medium Flow Sampler operates at
4 CFM and collects all PM-10 particles on one 102 mm filter.
    Because of the separation of coarse and fine mode particles, the Dicho-
tomous Sampler provides the most information in areas where difficult
compliance strategies are required.  An automated version of the Dichotomous
Sampler which allows up to 15 samples without operator intervention provides
a method to sample "episode" events and to study short term (e.g., day
versus night) fluctuations or cycles.  The Medium Flow Sampler collects
larger samples and provides the basis for developing compliance strategies
for noncompliance areas.  The Medium Flow Sampler is easier to use than the
Dichotomous Sampler.

     For routine monitoring stations,  the Size Selective High Volume Sampler
will probably be the instrument of choice because the operating procedures
are similar to the current TSP High Volume  Sampler  and  it  is  easy  to  use.
Existing TSP Hi-Vols can be easily converted to  PM-10 Hi-Vols  by  adding  a  size
selective inlet,  a flow controller,  a  flow recorder  and  a filter  paper cart-
ridge.   Glass fiber filters will  not be  allowed  because  of  artifact formation,
and the quartz filters  require a  filter  paper cartridge  because they are more
     The proposed Federal Reference Mothod  (FRM) performance specifications
for PM-10 Samplers are shown in Table  1  and the "sampling effectiveness"
(penetration) curves for the GMW and S-A Size Selective High Volume Samplers
are shown in Figure 1.   Both inlets exhibit sharp sampling  effectiveness
curves which meet the FRM performance  specifications.  The  cutpoint of the
S-A Model 321-A Two Stage SSI is closer to the 10 micrometer cutpoint
desired by EPA (10 micrometers versus  9.0 micrometers for the GMW Model
9000 inlet).  Table 2 summarizes the cutpoints of the two inlets  at wind-
speeds of 2, 8 and 24 km/h.
     As part of the FRM performance specifications,  candidate samplers must
collect within ±10% of the mass that an "Ideal Sampler" would collect  if
both sampled a hypothetical ambient mass distribution.  A summary of the
performance of the GMW and S-A inlets  compared to the "Ideal Sampler"  is
shown  in Table 3.  For  the hypothetical  mass distribution specified in the FRM,
the S-A Model 321-A would read 0.4% low whereas the GMW Model 9000 would
read 3.6% low.
     There has been some question as to whether the hypothetical  ambient
mass distribution specified in the FRM  is truly representative.   The  FRM
distribution is representative of urban environments with relatively high
fine particle concentrations, but may not be representative of rural or
fugitive emissions distributions.  As a further test of the PM-10 samplers,
we have compared their expected performance to the "Ideal Sampler" for the
three  additional hypothetical ambient mass  distributions shown in Figure 2.
Table  4  summarizes the variances of the S-A and GMW  samplers from the
"Ideal Sampler" for each different mass distribution.  Even for the Case
III distribution  (fugitive  emission, high large-particle concentration),

the S-A Model 321-A would read 0.7% high while the GMW Model 9000 would
read 4.5% low.
     Commercial PM-10 instruments are now available which meet or exceed all
of the proposed Federal Reference Method performance specifications.  These
inlets have been tested in the wind tunnel and in collocated field inter-
comparison studies.  PM-10 concentrations measured with different commercial
instruments should all be well within ±10% of 'the concentration that an "Ideal
Sampler" would be expected to measure.
1.  Federal Register, Vol. 49, No. 55, pages 10408-10462, March 20, 1984.

                               PM-10 SAMPLERS
Sampling Effectiveness
  A.  Liquid Particles
  B.  Solid Particles
Cutpoint (50%)


Flow Rate Stability
  %       Within ±10% of "Ideal Sampler"

  %       <57o higher than results for
           liquid particles for 20pm

  ym      10± 1 ym D

  %       _
Expected Mass
 (at 8 km/h)
Relative Error
 Compared to
"Ideal Sampler"
S-A 321 A (2 Stage)

GMW - 9000

"Ideal Sampler"



         Size Distribution
               Expected Mass Collected By
                321-A     Wedding    Ideal
              Two Stage   Hi-Vol    Sampler
                 SSI       Inlet
                yg/m3      yg/m3     yg/m3
Case I;
Total Aerosol Concentration 27.2 pg/m3
Variance from Ideal

Case II:
Total Aerosol Concentration 71.5 yg/m3
Variance from Ideal

Case III:
Total Aerosol Concentration 179.9 yg/m3
Variance from Ideal
  15.8      16.6
  -4.8%     .  .  .





           	 GMW- 9000
                  4      6    8   10         20

Figure  1:  Aerosol Sampling Characteristics of  PM-10 Inlets for the
          Hi-Vol Sampler.  Wind  Speed = 8 km/h.  Flow Rate =  1.13 m3/min.

   160  -
             CASE  I   	
             CASE  I	
             CASE  HI	
< 60
                                             /      x       !
                                             /  _  \     \
   0.01           O.I            1.0           10
                 PARTICLE  DIAMETER ,   Dp  ,
          FIGURE  2:  Three Hypothetical Ambient Mass Distributions


                       R.¥. Vanderpool  and D.A. Lundgren
                             University  of Florida
                             Gainesville, Florida

     The  calibration  of  four,   large-particle  impactors  developed  at  the
University  of  Florida  required  the  development  of  a  technique  for  the
generation of large calibration aerosols.  Slight modifications  to a vibrating
orifice aerosol generator  (Model  3050,  TSI  Inc.,  St. Paul,  Minn.) enabled the
successful  generation of  solid   ammonium  fluorescein  particles  up  to 70  urn
aerodynamic  diameter.   When generated  under  the  proper test conditions,  the
particles were found  to be spherical  and of uniform size.

     The developed  generation  procedure does  not  involve the  somewhat  awkward
inversion of  the  aerosol generator.  The dilution  flowrate of  the generator,
however, is  inadequate  to  suspend generated droplets larger  than about 65 urn.
As a result, large  droplets will  normally settle out  and be lost before having
sufficient  time  to  dry  to  the  desired   particle   size.    The  successful
generation of large particles, therefore, requires  the  use  of liquid solutions
of  high  volume  concentrations.   This  allows  production  of  droplets  of
suspendable  size  which  dry to  form particles of  the  desired  diameter.   By
dissolving  fluorescein powder  in  aqueous  ammonia,  volume concentrations  as
high  as 30%  were  produced.   Although  the  use of high volume  concentration
solutions requires  more  patience  to start and  maintain  the liquid jet through
the  orifice,  the  overall  generation process   is   considered  to  be  more
convenient than inversion of the  generator.

     The  generation   of  particles  larger   than   20  urn   requires   careful
optimization  of  the  operating parameters  of  the  aerosol  generator including
orifice  diameter,   liquid   feedrate,  vibrational  frequency,   and  dilution
flowrate.   Guidelines  for  the  proper  selection  of  these  parameters  are
outlined.   The  results  of  the   impactor  calibrations using   the  described
aerosol generation  technique are  briefly discussed.


           _C._F_._ Rogers and J.G. Watson, Desert Research Institute,
        Atmospheric "Sciences Center, P.O. Box 60220, Reno, NV  89506;
          and C.V. Mathai, AeroVironment, Inc., Pasadena, CA  91107

    Research quality  aerosol  monitoring  projects  require an aerosol  filter
sampling device with the following characteristics:

    •   Measurement of inhalable  (0 to 10  or 15  ym) and fine (0 to 2.5 ym)
        size-classified particulate matter, with acceptable sampling effec-

    •   Simultaneous  sampling  on  two  different  substrates,  one amenable to
        elemental and the other amenable to carbon analysis.

    •   Sequential sampling, without operator intervention, at greater than
        75 1/min  flow  rates to obtain  continous  samples over 4- to 24-hour
        sample durations with sufficient deposits for chemical analysis.

    •   Simple and reliable field and laboratory operation at an affordable

    A Size  Classifying  Isokinetic Sequential  Aerosol  Sampler (SCISAS)  com-
bines the  best  features  of the SURE/ERAQS  (Mueller and Hidy et al., 1983)
sequential  filter sampler and  the WRAQS  (Allard  et al.,  1982)  and Henry
(1977) isokinetic sampling manifolds to meet these requirements.

    Ambient  air  is  continuously  drawn at  a rate of 1100 1/min  into a ten
inch  diameter  PVC  stack  through  a  10  or  15 ym  McFarland size-selective
inlet (SSI).  Particles are then drawn from this stack, at a velocity close
to that flowing through the main stack to sample 0 to 10 or 15 ym particles
on two different filter media.  Each of the two 0 to 2.5 ym aerosol samples
is  withdrawn from the  main  stack at  a  flow  rate of  113  1/min  through a
single two-inch  internal diameter  tube which leads to a cyclone for exclu-
sion of particles  larger  than 2.5 ym  diameter.   The outlet of the cyclone
leads  to   a  simple   rectangular  supply  manifold  into  which  six  47  mm
Nuclepore   filter holders  are  mounted.   A general  view of the  SCISAS is
shown in Figure 1.

    The basic configuration of the SCISAS includes fourteen two-inch inter-
nal diameter supply  tubes clustered  inside the main ten-inch  stack.   The
flow rate   of   approximately  80  1/min within each inhalable particle two-
inch supply  tube  was  chosen empirically to  provide very nearly isokinetic
matching between the average velocities in these tubes  and. that in the main
stack with 1100 1/min flow.   An  alternative design  draws the  0 to 10 or
15 ym sample  at  isokinetic   velocities  through  a four inch diameter tube
into a 46" long sampling plenum.  Filters along the side of the plenum draw
the sample from it.

    The following particle loss mechanisms were theoretically evaluated and
predicted to be negligible (less than 1%):

    •   Electrostatic capture in the main PVC stack.

    •   Turbulent diffusion in the main stack and two inch sampling tubes.

    •   Brownian diffusion losses.

    •   Inertial impaction losses in the two inch tubes.

Sedimentation  losses  in the  inhalable  particle sampling tubes  are calcu-
lated to be a maximum of 1% for 15 pm particles.  The maximum bias in the 0
to 10 or 15 um mass estimation is much less than 5%.

    In ten  tests  of  a 15 pm  cut-point  SCISAS  prototype*  aerosol  mass con-
centrations  measured  in Reno,  NV,  ranged  from 6 yg/m   to 32 ug/m  ;  the
maximum difference between  any  two  of the  four SCISAS  0  to 15  ym particle
sampling tubes operated simultaneously in each of these tests was 4%.   More
typically,  any  two  sampling tubes  agreed  to  better than 3%.   At  the same
sampling site,  three  comparisons of  the  SCISAS prototype  to  a collocated
hivol outfitted  with  an  identical  15  ym  size selective  inlet  were con-
ducted.   Ratios of the mass concentrations measured by the SCISAS, to those
measured by the hivol/SSI, were 0.98,  0.95, and 0.96 for these three preli-
minary tests.

    Further tests of  the  SCISAS are now scheduled  and  will include exten-
sive  comparisons  with  other  sampling  devices.    Other  tests  include
1) measurement  of passive  deposition  inside  the  SCISAS  sampling  tubes,
2) measurements  of  re-entrainment  of  large particles  inside  the  SCISAS,
3) evaluation of virtual impaction into non-operating sampling tubes in the
SCISAS tube cluster,  4)  quantitative  evaluation  of the effects  of non-
isokinetic mismatches at the  entrance to the  tube cluster,  5)  and measure-
ment  of  the effects  of flow rate  variations   in  the  main  stack  and SSI.
Velocities  inside the  main  stack at the approach  to  the  tube cluster will
be mapped  with  a  hot-wire anemometer.   A  theoretical   evaluation of  the
possible  effects  of  aerosol  particle  deliquesence or  shrinking, due  to
heat transfer in the SCISAS, will also be performed.


Allard,  D.W., Tombach, I.H., Mayrsohn, H., and Mathai, C.V., 1982. "Aerosol
    Measurements:  Western  Regional   Air  Quality Studies"  Air  Pollution
    Control  Association Annual Meeting,  New Orleans, LA.

Henry,  R.,   1977.    "A Factor  Model  of  Urban  Aerosol   Pollution,"  Ph.D.
    Dissertation Oregon Graduate Center, Beaverton, Oregon.

Mueller, P.K., Hidy, G.M., Baskett, R.L.,  Fung, K.K.,  Henry, R.C., Lavery,
    T.F., Warren,  K.K.  and  Watson,  J.G.,  1983.    "The  Sulfate  Regional
    Experiment:  Report of Finding  Volume  1  Report EA-1901, Electric  Power
    Research Institute, Palo Alto, CA.

                                        SIZE SELECTIVE
                                        INLET = I0fj.m or
                                    TEN INCH STACK



V — '


    TUBES (12)
         HI VOL FAN
                                 ATYPICAL VACUUM MANIFOLD
                                   (3 MORE NOT SHOWN)
FIGURE 1:  General view of  SCISAS,  not showing  Nuclepore filter
          holders, three of four vacuum manifolds, four suction
          pumps, connecting tubing  and stand.

                             AND QUARTZ FILTERS
          V. Ross Highsmith, U.S. Environmental Protection Agency
            Andrew E. Bond, U.S. Environmental Protection Agency
               James E. Howes, Battelle Columbus Laboratories

     A special study was conducted to evaluate particle and filter substrate
losses resulting from routine handling of particulate samples collected on
quartz and teflon filters.  Filters were weighed at pre-determined stages of
the filter handling process to estimate changes in mass corresponding to the
various filter handling operations.  Control filters, both field blanks and
sampled filters, were used to estimate passive artifact formation and particu-
late matter volatilization.  The remaining filters were shipped to the labora-
tory for observation and returned for final weighing.  Changes in shipped
filter mass could be contributed to both a loss of large particles during
shipment and a loss of particulate matter from volatilization.  A comparison
of control filter weight changes with shipped filter weight changes would
provide an estimate of the overriding mechanism responsible for any observed
particle and substrate losses following sample collection.

          The data presented in this report suggest no significant weight
loss from routine high volume sampling of particulate matter using quartz
filter media, as long as the final filter weights are performed without
archiving or shipping the filter to the laboratory.  A reduction in filter
mass was observed after the shipment of total suspended particulate quartz
filters to the laboratory.  Particulate matter loss from volatilization was
also noted with the high volume samples collected in Phoenix, Az.   No signifi-
cant weight change was observed in the routine handling of dichotomous teflon

     This is an abstract for proposed publication and does not necessarily
reflect EPA policy.

     The Environmental Protection Agency recently conducted a field evaluation
of commercially available nominal 10 micrometer (10pm) inlets for particulate
samplers.   Total suspended particulate (TSP) and 10pm size selective inlet
(SSI) high volume samplers as well as 10pm dichotomous samplers were operated
in four cities using established Inhalable Particulate Network (IPN) operating
procedures.   TSP and SSI samples were collected using 8" x 10" quartz filters.
Dichotomous samples were collected using 37 mm teflon filters media identical
to those employed in the IPN.
     A special filter evaluation study was conducted at two of the four
cities, Phoenix, A2 and St. Louis, MO.  The purpose of this special study was
to evaluate various aspects of the filter handling operation in order to
estimate the magnitude of particle and filter substrate losses from quartz
and teflon filters during the particulate matter (PM) measurement process.
The initial consideration was to determine the magnitude of particle and
substrate losses resulting from the folding of quartz high volume sampler
filters.  Commercially available quartz filters have a tensile strength equal
to about 25-40% of the tensile strength cited for glass fiber filters.   When
folded, quartz filters have a greater tendency to crack and fray along the
crease.  Glass fiber filters typically do not crack or fray when folded.
Monitoring both particulate-loaded and blank quartz filter weights before and
after folding would provide information to assess the combined particle and
substrate losses resulting from the folding process.  The second aspect to be
considered was the loss of particulate matter resulting from the shipment of
sampled filters to the laboratory for final weighing.  Filter weights were
monitored both before and after shipment to document mass losses resulting
from shipment.  Loss of PM mass during filter shipment is thought to be a
large particle phenomenon; i.e., with increased large particle concentration,
PM loss resulting from shipment is expected to increase.  Compared to St.
Louis, Phoenix's particle size distribution data indicates a significantly
larger coarse particle fraction.  Therefore, the Phoenix samples would be
expected to be more adversely affected by large particle losses during ship-
ment than the St. Louis samples.

     The third aspect to be evaluated in this study was PM losses resulting
from volatilization over long periods of time.    For a typical PM monitoring
network, a lapse of up to 30 days occurs from the date the sample is collected
in the field to the date the gross weight is obtained in the laboratory.
This study data would provide a means to estimate particulate mass loss
resulting from volatilization.  Estimating passive artifact formation on
quartz filters was the last aspect considered.   Sulfate and nitrate artifact
formation on glass fiber filters, routinely used in PM monitoring networks
such as the IPN, has been documented.   Unlike glass fiber filters which have
a pH of ca. 9.5, quartz filters have a pH of ca. 7.0.  Consequently, passive
sulfate and nitrate artifact formation on the quartz filters used in this
study should be minimal.
     The results from this study are considered to be "best case".  Extra
precautions were taken by the field operators in both the handling of samples
in the field and the operation of the samplers in accordance with the
manufacturer's specifications.  In addition, the study quality assurance
protocol, especially with regards to acceptable weighing procedures, was
strictly enforced.  Most important, both TSP and SSI sampling was conducted
using filter cassettes, minimizing filter handling in the field and reducing
the potential for voided filters.
     Five sampling days were scheduled in both Phoenix, AZ and St. Louis, MO
to conduct the special study.  Figure 1 diagrams the high volume filter
handling process.  Prior to sampling, groups of quartz and teflon filters
were placed opened in racks inside the mobile weighing laboratory with controlled
chamber conditions of 40±3% relative humidity and 20±2° Centigrade.  Following
24 hours of conditioning, filter tare weights were recorded.  Quartz filters
were then loaded into filter cassettes, with lids, while the teflon filters
were first placed in Lexan jigs and then into petri dishes.  On each sampling
day, two TSP and four SSI high volume samplers were operated.  Four dichotomous
samplers were also operated for each 24-hour sampling period. Additionally,
on each sampling day, two quartz filter field blanks and two teflon filter
field blanks were identified.  These field blanks, quartz filters loaded in
cassettes with lids and teflon filters in Lexan jigs inside petri dishes,
were removed from the conditioning chamber and placed in the sampling
environment during the 24-hour sample period.

     At the completion of the sampling period, each filter was then returned
to the controlled conditioning chamber.  Filters were removed from the cassettes
and Lexan jigs and placed opened in the conditioning chamber racks for 24
hours.  The gross weight was then recorded for each filter.  After weighing,
one TSP filter sample, one SSI filter sample, one quartz filter blank, one
dichotomous fine sample, one dichotomous coarse sample and one blank teflon
filter were designated as controls and returned to the conditioning shelf.
The remaining TSP, three SSI and one blank quartz filters were individually
folded, placed in separate cardboard supports and enclosed inside manilla
envelopes.  These folded quartz filters were immediately removed from the
envelopes and reweighed to obtain a weight after folding.  After reweighing,
the filters were returned to their appropriate cardboard supports and envelopes.
After weighing, the three coarse, three fine and one blank dichotomous samples
were reloaded into their Lexan jigs and placed in their petri dishes.  The
folded quartz filters and dichotomous filters were then mailed to the laboratory
using Jiffy  bags, with three or four samples per bag.
     Upon receipt of the filters at the laboratory, each filter was visually
inspected for obvious PM loss, cracks, tears or any other unusual physical
change.  Following this visual inspection, the filters were repackaged and
mailed back to the sampling site.  Upon arrival at the sampling site, the
filters were opened and returned to the conditioning shelf for 24 hours.
After reconditioning, the shipped and corresponding control samples were
again weighed and final weights obtained.   For both Phoenix and St. Louis,
the final filter weighing occurred within 30 days after sample collection.
     The resulting mass data was summarized and analyzed using standard
statistical tests for both paired and non-paired data  at the 5% significance
level as described below:

For paired data

     Calculated T Statistic   T =   / (    , .—
                                       ad ' 1 n;
          where d  = mean difference between pairs
               a, = standard deviation of the mean difference
                n = number of pairs

     Paired Test Interval    t          ± T

For non-paired data
     where d.. - d?  = difference between population means
           n^n^    = number of observations for groups 1 and 2
            p              (nj + n2 - 2)
             2   2
           Sl  'S2   = variance f°r groups 1 and 2

For paired data sets, if the test interval, calculated test statistic (T)
plus or minus the tabulated Student-t (t           ) value contains zero,
                                        n."" i j i *" of/ £,
then the two sets of data being compared are determined to be indistinquishable.
For non-paired data sets, if the test interval defined above includes zero,
then the two sets of non-paired data are considered to be indistinquishable.
Otherwise, the two data sets are considered distinquishable, i.e., the
difference in the two sets are statistically signficant.  In instances where
the statistical analysis of the data indicate the two data sets to be
distinquishable, but the magnitude of the real difference between the two
data sets is determined to fall within the experimental error associated with
the weighing process, the difference is determined to be marginal and not of
practical significance.
     Three paired statistical comparisons were performed on the shipped
quartz filter weight data for each of the three sample types (TSP, SSI and
blank) for each sampling city.  First, the 24-hour weights were statistically
compared to the corresponding folded filter weights to determine mass loss
resulting from folding.  Secondly, the folded filter weights were compared to
the 30-day weights to estimate the combined losses resulting from PM
volatilization and shipment of the filters to the laboratory.   Finally,  the
24-hour weights were compared to the 30-day weights and a total mass loss
since sample collection was calculated.   Likewise, the 24-hour control filter
weights were compared to the corresponding 30-day weights.  As these filters
were neither shipped nor folded, any significant loss in mass  could be attributed
to volatilization.  Any significant gain in filter weight observed with the

control filters would be attributed to passive artifact formation.  Using the
non-paired test statistic, the average weight loss for each shipped filter
type was then compared to the average weight loss for the corresponding
control filter type to determine if any statistical difference in filter mass
observed with the shipped filters was also observed with the control filters.
The results of both paired and non-paired statistical tests were then used to
determine whether the filter folding process, shipment of the filter or
volatilization was the overriding mechanism contributing to any observed
weight change.
     The dichotomous filter weight data was statistically tested following
the same procedures outlined above for the quartz high volume samples with
one exception.  The shipped dichotomous filters were only weighed at two
intervals, 24 hours following sample collection and at the 30-day interval.
Therefore, only one paired comparison of the shipped dichotomous sample data
was conducted for each city.
     When received at the laboratory, the shipped filters were opened and
examined for tears, cracks, loss of large particles or any other usual physical
change.  More that 80% of the shipped TSP and SSI samples received from both
Phoenix and St. Louis were cracked and/or frayed along the folded crease.
However, less than 10% of the shipped blank filters experienced cracking
along the crease.  No explanation can be given at this time regarding why the
sampled filters cracked and the blank filters did not crack upon folding and
shipping.  Examination of the St. Louis dichotomous filters revealed no
obvious physical changes in the filters.  The Phoenix coarse dichotomous
filters, however, did show some loss of large particles both to the Lexan jig
and the petri dish.  No particle loss or other physical change was noted with
the Phoenix fine dichotomous samples.
     Table 1 summarizes the results of the paired t-test statistics performed
on the Phoenix and St. Louis quartz high volume filter weights.  For all
three sample types (TSP, SSI and blank samples), at both cities, comparisons
of the 24-hour filter weights to the corresponding folded filter weights
yielded no significant difference in mass.  This implies that neither
particulate matter nor filter substrate material was lost as a result of the
folding process.  A statistically significant loss in filter mass was observed
for the Phoenix TSP (12.7 (Jg/m3), Phoenix SSI (4.7 M8/m3) and St. Louis TSP

(4.6 |Jg/m3) samples shipped to the laboratory.  This statistical test suggests
that this change in filter mass is directly attributed to both a large particle
loss during shipment and to a loss of PM due to volatilization.  The shipped
St. Louis SSI filter weight change, although statistically significant, was
considered marginal as the average SSI filter mass loss falls within the
experimental error established for the filter weighing process.  An analysis
of the shipped blank quartz filter data shows no significant weight change.
This suggests that the blank quartz filters used in this special study neither
loss substrate due to folding or shipping nor did they undergo significant
passive artifact formation.  For both St. Louis and Phoenix shipped quartz
filters, the overall mass change observed in the 24-hour versus 30-day
comparisons corresponded to the summation of the mass change calculated for
both the 24-hour versus folded comparison and the folded versus the 30-day
comparison.  The control quartz high volume sample data reveals that the
Phoenix control TSP and SSI filters experienced a significant loss in filter
mass equivalent to ca. 3 (Jg/m3, over the 30-day period.  Since the control
filters were neither folded nor shipped, this mass loss is thought to correspond
to volatilization.  The St. Louis control quartz filter data, as well as the
blank quartz filter data, indicate no significant mass change over the 30-day
period and therefore no loss of PM due to volatilization.  The overall
differences in the St. Louis particle size distribution, the chemical
constituency of the particles and the 24-hour mass loadings as compared to
Phoenix are considered the primary reasons for this observation.
     Using standard statistics for non-paired data, the overall weight change
for each shipped sample type was compared with the weight change for the
corresponding control sample type.  Testing the Phoenix shipped TSP filters
against the Phoenix control TSP filters yielded a significant difference in
weight loss.  Recalling the earlier paired filter comparisons, this indicates
that the observed weight change for the shipped Phoenix TSP filters represents
the combined effects of shipment and volatilization.  A similar comparison of
the Phoenix SSI shipped and controlled filters yielded no statistically
significant difference in observed shipped SSI weight loss.  Therefore, the
shipped Phoenix SSI sample weight loss is solely attributable to volatilization
and not large particle loss during shipment.  The St. Louis shipped versus
controlled TSP filter data suggests that the observed weight changes results
solely from large particle losses occurring during shipment of the sample to
the laboratory.

     Table 2 summarizes the analysis results for the dichotomous filter data.
A significant loss in filter mass was noted with the shipped Phoenix fine and
coarse dichotomous samples.  As was noted previously for the shipped quartz
filters, this data indicates a loss of large particles during shipment as
well as a loss of PM due to volatilization.  No other significant changes in
shipped filter mass was observed for either the Phoenix or St. Louis shipped
dichotomous filters.  The control dichotomous sample statistics show no
distinquishable differences between the 24-hour and 30-day weights and indicates
that volatilization did not significantly affect the collected dichotomous
samples.  Therefore, the mass loss observed with the shipped Phoenix dichotomous
samples is contributed to a loss of large particles during shipment.  Since
the Phoenix size distribution contains an extremely large coarse fraction,
the loss in filter mass seen in the shipped Phoenix dichotomous filters is
attributed to this abnormal large particle loading.  Although significant for
Phoenix or any other arid environment heavily laden loaded with coarse particles,
a loss in filter mass resulting from shipping dichotomous filters to the
laboratory is considered unlikely for most routine sampling sites.
     Folding quartz high volume filters does not significantly affect filter
mass.  Both Phoenix and St. Louis shipped TSP samples experienced large
particle losses equivalent to approximately 5% of the filter mass loading as
a result of shipping the filter to the laboratory.  Based on the blank quartz
filter data, no filter substrate material was lost nor did passive artifact
formation significantly affect the mass determinations.  The Phoenix TSP and
SSI sample data showed significant weight loss corresponding to volatization.
Except in areas with extremely high coarse particle loadings, you would not
expect particle loss from shipping dichotomous filters to the laboratory.
     This special study suggests that quartz filters can be used in routine
PM monitoring.  However, this was a "best case" study.  Field operators and
laboratory personnel exercised caution in the handling of these quartz filters.
The sampling equipment was routinely monitored to insure compliance with the
manufacturer's specifications.  And most important, filter cassettes were
used for high volume sampling, minimizing the handling of quartz filters in
the field.  Additionally, cracked or frayed quartz filters were not voided
but were considered as acceptable samples in this study.


1.   Rodes, C.E., R.M.  Burton,  L.J.  Purdue and K.E.  Rehme.   Protocol for PM
     Inlet Evaluation and Comparison with the Wide Range Aerosol  Classifier,
     April 1983, U.S. Environmental  Protection Agency,  Research Triangle
     Park, N.C.   27711.

2.   Inhalable Particulate Network Operations and Quality Assurance Manual,
     March 1983, U.S. Environmental  Protection Agency,  Research Triangle
     Park, N.C.   27711.

3.   Whatman, Inc.  9 Bridwell  Place,  Clifton, N.J., 1984.

4.   Clement, R.E. and F.W. Kurasek.   Sample Composition Changes  in Sampling
     and Analysis of Organic Compounds in Aerosols.   Int.  J.  Environ.  Analyt.
     Chem.  7:109, 1979.

5.   Appel, B.P., S.M.  Wall, Y.  Tokiwa and M. Hunt.   Interference Effects in
     Sampling Particulate Nitrate in Ambient Air, Atmos. Environ.   13:319,

6.   Remington,  R.D. and M. Anthony  Sihork.   Statistics with Application to
     the Biological and Health  Sciences,  Prentice-Hall, Inc., Englewood
     Cliffs, N.J., 1970.

     TABLE 1.  Results of Statistical Analysis on Quartz Filter Weights
       ST. LOUIS
TSP       SSI
Shipped Filter
24-Hour versus
Folded Weight
Indist    Marg      Indist
               Indist    Marg      Indist
Shipped Filter
Folded versus
30-Day Weight
Dist      Dist      Indist
               Dist      Marg
Shipped Filter
24- Hour versus
30- Day Weight
Dist      Dist
Dist      Marg      Indist
Control Filter
24-Hour versus
30-Day Weight
Dist      Dist
Indist    Indist    Indist
Shipped versus
Control Filter      Dist
30-Day Weight Loss
          Indist    Indist
               Dist      Indist    Indist

     TABLE 2.  Results of Statistical Analysis on Teflon Filter Weights

         ST. LOUIS

          COARSE    BLANK
Shipped Filter

24-Hour versus

30-Day Weight
Indist    Indist
Control Filter

24-Hour versus

30-Day Weight
Indist    Indist    Indist
                         Indist    Marg
Shipped versus

Control Filter      Indist    Indist    Indist

30-Day Weight Loss
                                   Indist    Indist
Dist indicates a significant difference between paired data

Indist indicates no significant difference between paired data

Marg indicates that although a statistical difference is noted,  that the real

    difference is within the experimental error associated with  the weighing


Figure 1.  High Volume Sample  Handling Diagram.

                             T.  Jarv and O.T.  Melo
                             Ontario Hydro Research
                        Toronto, Ontario,  Canada  M8Z 5S4
     Ontario Hydro (OH) voluntarily initiated  a sulfate aerosol  monitoring
program in 1975/1/ in response to concerns over health effects.   The monitoring
network underwent several  changes since its inception.  In 1981, the sampler
used for sulfate aerosol sampling was changed.   Dichotomous samplers replaced
hi-vol and RAC low-vol  air samplers, to take advantage of improved knowledge
and methodologies in the study of sulfate  aerosol.  At the same  time, a shift
in emphasis from sulfate aerosol to acid precipitation required  that other acidic
pollutants, such as nitric acid, nitrate and sulfur dioxide, be  monitored also.
To accomplish this the dichotomous sampler, originally developed for aerosol
sampling, was modified slightly to allow collection of these other pollutants.
     A comparison with the Ontario Ministry of the Environment (ONE) Acidic
Precipitation in Ontario Study (APIOS) event network and the Atmospheric Environ-
ment Service (AES) of Environment Canada Air and  Precipitation Monitoring Network
(APN) was undertaken in 1981.  The comparison, undertaken at the APIOS monitoring
site at Dorset, Ontario, indicated that the OH results for total nitrate and
sulfur dioxide were lower than those obtained  by  OME and AES/2/.
     In order to explore the differences observed in the Dorset  comparison,
two additional studies were undertaken: i) a laboratory investigation to deter-
mine whether gaseous nitric acid losses occur  in  the dichotomous sampler and
ii) a second field comparison of a dichotomous sampler, a Teflon-coated dicho-
tomous sampler and an OME/AES-type sampling system under meteorological condi-
tions similar to those encountered in the  Dorset  comparison.  In this paper,
the highlights of these more recent studies are presented.  Some interpreta-
tion of the results is also provided.
     A modified dichotomous sampler, similar to the ones used in the OH network,
was tested in the laboratory.  The sampler is  composed of three  components:
1) inlet head, 2) inlet tube and 3) virtual impactor.  The inlet head was
 placed  in  a  0.022  nf Teflon  chamber  into  which a  nitric  acid  containing  atmosphere
was introduced.  A reference filter cassette,  attached to the Teflon chamber, was
used to sample the test atmosphere.

     Four sampling configurations were examined:  these are noted in Figure 1.   A
total of thirty-six 24-h samples were collected,  with at least six 24-h samples
for each experimental  arrangement.  The filter cassettes used in the laboratory
tests consisted of a Teflon filter to remove particulates and a single nylon  fil-
ter which was then extracted and analysed for nitrate using automated colouri-
metry.   The ratio, (C+F)/REF, total (coarse + fine)  nitric acid sampled by the
dichotomous sampler to that collected on the reference filter were determined
and are shown in Figure 1.
     The (C+F)/REF results for the complete sampler  configuration indicated
that gaseous nitric acid is lost to the interior of  the dichotomous sampler.
The addition of a Teflon liner to the inlet tube reduced the nitric acid loss.
Each sampler component was found to contribute significantly to the loss of
nitric acid, with the virtual impactor appearing to  be the largest sink.
     An OME/AES-type sampler (1), a modified dichotomous sampler (2), and a
Teflon-coated modified dichotomous sampler (3) were  compared.  The following
sampler pairings were made: Pair A - 1 versus 2; Pair B - 1 versus 3 and Pair C -
2 versus 3.
     The OME/AES-type sampling system consisted of a multi-stage filter cassette
mounted in a Teflon holder, a protective polyethylene cone, a flow controller, a
Cast pumping system and a supporting stand.  The filter cassette was held in
the inverted polyethylene cone 2 meters above ground level.  The sampling flow
rate was identical to that of the modified dichotomous sampler, 16.7 L/min.  The
dichotomous samplers collected air at 2 meters also.  The dichotomous sampler
had an upper cut-off of 15 ym diameter.  The OME/AES-type sampler has been esti-
mated to collect particles smaller than about 30 ym, under laminar atmospheric
flow conditions.
     Identical filter cassettes were used with all samplers.  Each multi-stage
filter cassette consisted of a 37-mm diameter, 1  ym  pore-size Teflon filter for
sulfate, nonvolatile  nitrate and ammonium aerosol collection; followed by a
37-mm diameter, 1.1 ym pore-size nylon filter for gaseous nitric acid and vola-
tile nitrate collection; and terminated with a 37-mm diameter, potassium
carbonate-impregnated Whatman 41 cellulose filter for sulfur dioxide collection.
A total of 25 concurrent 24-h samples were collected with the three samplers.

The filters were extracted and analysed for sulfate, nitrate, ammonium and
sulfur dioxide by continuous flow analysis (automated colourimetry).   The
results were evaluated statistically with scattergrams,  the nonparametric sign
test and least squares linear regression.  The nonvolatile nitrate, volatile
nitrate and sulfur dioxide scattergrams are shown in Figure 2.   The sign test
and linear least squares results are presented in Table  I.
     An examination of Figure 2 and Table I indicates that sulfate, nonvolatile
nitrate and ammonium aerosol and sulfur dioxide concentrations  measured with
the two dichotomous samplers were statistically equivalent.  Volatile nitrate
concentrations measured with the Teflon-coated dichotomous sampler were larger
than those measured with the uncoated dichotomous sampler.  This is consistent
with the laboratory results.  The OME/AES-type sampler was found to collect more
nonvolatile nitrate and volatile nitrate (total nitrate)  than either  of the di-
chotomous samplers.  These nitrate concentration differences are attributed to
the additional coarse particulate nitrate sampled by the  OME/AES-type sampler
and to the partial volatilization of this material/3,4/.
     As a comparison, sulfate and ammonium aerosol,  both  predominantly associated
with submicron particles/5/, were sampled in a statistically equal fashion by
all three samplers.  If nitrate aerosol were predominantly associated with
submicron particles then the three samplers would have collected nitrate aerosol
in a statistically equal fashion.  The loss of nitrate aerosol  through volatili-
zation would have been comparable in all  samplers,  as the flow rates  were equal.
However, the OME/AES-type sampler collected more nonvolatile nitrate.  The
difference likely results from the additional  coarse aerosol that appears to
be sampled by the OME/AES-type sampler.
     Sulfur dioxide concentrations measured with the OME/AES-type sampler and
the Teflon-coated dichotomous sampler were statistically  equivalent.   A similar
result was expected for volatile nitrate.  However,  the  results were  clouded
by artifact volatile nitrate from the partial  volatilization of the additional
coarse nitrate aerosol collected by the OME/AES-type sampler.
     These results, combined with the Dorset comparison,  suggest that the
large nonvolatile and volatile nitrate and sulfur dioxide concentration dif-
ferences found at Dorset resulted from the following: (a) a small contribution
from the inlet height difference (<10%),  (b) losses  to the aluminum inner walls
of the dichotomous sampler, (c) occasional losses to moisture trapped on the OH

filter during the Dorset study, (d) nitric acid artifact contributions during

sampling, (e) quality control variations in the preparation of the potassium
carbonate-impregnated filters, (f) differences in sample flow rates and (g) the

different particle size ranges sampled.
     The authors thank Messrs. G.  Till,  B. Handy and D. Knebel for their help
in conducting the experiments.  This work was supported by the Chemical Research

Department of Ontario Hydro Research.
1. Melo, O.T., (1975).  A Proposal for Atmospheric Sulphate Monitoring in
   Southern Ontario, Ontario Hydro Research Division Report No. 75-19-K.

2. Concord Scientific Corporation, (1982).  The Dorset Intercomparison of Pre-
   cipitation and Air Sampling Methodologies, CSC Report 182-2 prepared for
   Ontario Ministry of the Environment - Air Resources Branch, Atmospheric
   Environment Service and Ontario Hydro.

3. Appel, B.R., Y. Tokiwa and M.  Haik, (1981).  Sampling of Nitrates in Ambient
   Air, Atmos. Environ. 15, pp 283.

4. Appel, B.R., S.M. Wall, Y. Tokiwa and M. Haik, (1979).  Interference Effects
   in Sampling Particulate Nitrate in Ambient Air, Atmos. Environ. 13, pp 319.

5. Kadowski, S., (1976).  Size Distribution of Atmospheric Total  Aerosols,
   Sulfate, Ammonium and Nitrate Particulates in Nagoya Area.   Atmos. Environ.
   10, pp 39.


                     Dale A. Lundgren and Brian  Hausknecht
                             University of  Florida
                            Gainesville, FL  32611
                   Robert M. Burton,  EMSL,  EPA,  R.T.P.,  N.C.

                               EXTENDED ABSTRACT

     A mobile  aerosol  size classifying  sampling system for  the  collection of
very  large  (lOO urn  diameter)  particles  was  designed  and  constructed  by
Lundgren  and Rovell-Rixx at the University of Florida.1  An  analysis  van was
outfitted  to  accompany  the  sampling  trailer.   A  specially  designed  air
sampling  inlet was  fitted  to  a very high flowrate  (~40  m^/min)  sampler,  which
greatly reduced the large particle sampling errors  due  to  inertial  effects, as
described  by Lundgren  and  Paulus.2   In a  10 km/hr  wind,  the design  criteria
predicted  a  less  than  20%  error  for sampling 100 urn particles.  Test  results
indicated  that the sampling error  was within  this design limit.

     Ambient aerosol mass distribution  were measured  in  five  cities across the
U.S.  and  compared  with  data  collected using  several  conventional  ambient
aerosol  samplers  and  size  selective inlet (SSl)  samplers.   The  five  cities
sampled were Birmingham, Alabama  (an  industrial  area); Research  Triangle  Park,
North Carolina  (a background  site);  Philadelphia,  Pennsylvania  (metropolitan
site); Phoenix,  Arizona  (high fugitive  dust  area);  and Riverside,  California
(photochemical  aerosol site).  These cities  provided  a  variety  of  sampling
conditions  and  aerosol  compositions.   The   actual  location  selected  in  each
city was  at  an  EPA  Inhalable  Particulate (IP) Network station where a  history
of data  for  the  high-volume air sampler, size selective  inlet sampler  and the
dichotomous sampler were available.

     Present ambient air quality standards  for particulate matter are  based on
measurements made by  the EPA  reference  method  (High-Volume  Method).^  Weight
gain  by  the sampler filter media, divided  by  the  volume of air  sampled,  is
defined as  total  suspended  particulate  matter, or TSP.  Health  effect  studies
have correlated this measurement  with adverse health effects.  However,  it is
generally  accepted   that  some of  the  particulate  mass  collected  by  this
reference  method  sampler  is   too  large to cause  health effects.   This  has
resulted  in proposed changes to the  primary air  quality  standard  and method of
measurement.  If  new standards are  to  be  set  and  the  method of  measurement
changed  it is  necessary to  determine   the relationship  between  the  present
reference  method  (High Volume   Method)  measurements   and   a  size  selective
reference method measurement.

     Large  particle size-distribution  data  were   obtained   using  the mobile
aerosol-sampling  system, called  the  Wide Range  Aerosol  Classifier  (WRAC).   A
schematic diagram of the WRAC  is shown  in Figure 1.   The  large (60cm)  diameter
aerosol  inlet  tube  leads to a cluster  of   five individual sampler  units,  each
of which  operates  at  an actual sampling  rate of 1.56 nK/min (55  acfm).   The
center sampler  collects  what   is considered to  be  a  total aerosol  mass  sample
onto  a  standard 20.3  by 25-4  cm (8"  by   10")  glass or  quartz  fiber  filter


media.   Four  other  samplers,   placed   at  90°  intervals  around  the  center
sampler,  are  single  stage,  rectangular  slot  impactors.   These  single  stage
impactors  collect  size  fractionated  samples  of  the  large ambient  particles
onto  grease  coated  impaction plates.   Remaining  particles are collected  by  a
standard filter which follows each single stage impactor.   Each  impactor has  a
different  particle  collection efficiency.  The particle cutoff  diameter  (for
50$ collection  efficiency)  for Impactors 1,  2,  3,  and  4,  are  47,  34,  18.5  &
9.3 urn, respectively.  These  impactor nozzles  were  carefully calibrated at the
University of Minnesota,  Particle Technology Laboratory and  the  University  of
Florida to determine  their exact  outpoints, as described by Vanderpool.^

     Particle  size-fractionating  samplers  were  operated   simultaneously  with
the WRAC at each site.   These samplers  included instruments typically found  at
an Inhalable  Particulate (IP) network  site  such  as:  high-volume  air sampler
(HIVOL),  a 15  urn   type  size selective inlet sampler  (SSl), and  dichotomous
sampler (Moot).  These  instruments were normally located  at the  site and were
operated by  the WRAC  sampling team during the special  sampling.   At each site
was at  least  one high-volume ambient cascade  impactor -was  also  used.  Most  of
these samplers were run  with  a duplicate unit  at  one or more location to check
for repeatability of  results.

      At each  site,  samples collected under  similar  conditions were  averaged  to
determine  a representative distribution.

Aerosol Mass  Fraction and Particle Size
Collected  by  the High-Volume  Sampler
      The  total  atmospheric  aerosol mass fraction and particle size collected
by  the  standard High-Volume Sampler  (Hi-Vol) can  be  inferred  by comparison
with  the  aerosol  mass and size  distribution measurements  made using the Wide
Range Aerosol Classifier (WRAC).

      The  grand distribution for all  41 usable  WRAC  runs produce  a  total
aerosol  concentration of 134.0  ug/riK  with 91.0$  of the  aerosol mass < 34  urn
diameter.   Most  single  city  average  distribution  and  the  41  day  grand
distribution  average  suggests that the  standard High-Volume  Sampler collected
all particles less  than~30  urn diameter (on  the average).   Calibration data  of
McFarland,  Ortiz  and Rodes^  also suggests a  Hi-Vol sampler 50$ cut  size  of
about 30  urn.   These data were also  presented in an article by  Watson,  Chow,
Shah  and Pace  which  discusses the Hi-Vol aerosol collection.

Atmospheric Aerosol Large Particle Mass Distribution
      Plots of the  total  large particle  grand  average  distribution  and various
city  average distributions  suggest  that  the  large  particle distribution  is
approximately log-normal.   These  data  also suggests  a minimum  value between
the large  and small particle mass modes  at about 3 urn (aerodynamic diameter).
If one  assumes  the  large particle mass  mode  is log-normal and  that there is a
minimum point at 3  urn, several features of  the distribution can be determined.
A  best  fitting  curve  was   drawn through  the actual  mass  measurement  data
plotted  as a cumulative distribution  curve  on  log-normal  probability paper.
Several of these curves  were  then drawn as  histograms in Figure 2.

     There has been much discussion recently about incorporating  an  upper  size
limit  for  the regulation of  particulate  matter.  A  size limit  of  10 jim  has
been   suggested.    The   WRAC   measurements  reveals   that   the  fraction   of
particulate matter  in ambient air  associated  with particles  less   than 10 urn
diameter can  vary  between  about 50% and  90%  for  single  run percentages  (for
Birmingham and Riverside respectively).  The average distribution data  for the
high concentration  days  in  Birmingham  and  Riverside  suggest that a  10  urn  size
selective sampler would collect 68% and 89% respectively, of what the standard
High Volume  Air Sampler would collect.   An  average  distribution  for  all  41
test days  from  5  cities suggest the 10 urn sampler would collect 14%  of  that
collected by the Hi-Vol.

     Ambient aerosol  distributions  display a  large particle mass mode  under a
variety of situations.   The situations include:   relatively clean   areas  like
Research Triangle  Park, areas with high small-particle  concentrations  like
Riverside,  areas  with  high large-particle  concentrations   like  Phoenix,   and
areas  with high  concentrations of large and  small particles like Birmingham.
Each of these areas will be affected  differently with the implementation  of a
new  health-related  particulate  matter standard.  This  will  relieve  certain
areas  which have  historically been in a  non-attainment  status because of the
presence of a high mass fraction of large  particles.

     This  investigation  was supported  by  cooperative  agreement  CR808606  from
the  Environmental  Monitoring Systems Laboratory,  Environmental   Protection
Agency, Research Triangle Park, North Carolina.

1.  Lundgren,  Dale  A.  and  David  C.  Rovell-Rixx,  1982.   Wide  Range  Aerosol
    Classifier,  EPA-600/4-82-040, PB82-256264 N.T.I.S.

2.  Lundgren,  Dale  A.  and  H.J. Paulus, 1975-   The Mass  Distribution of Large
    Atmospheric Particles,  JAPCA 25 (12):1227.

3.  "Reference Method  for  the Determination of  Suspended Particulates  in  the
    Atmosphere (High  Volume Method)",  40  CFR  50,  Appendix  B,  U.S.   Government
    Printing Office.

4.  Vanderpool,   Robert,  1983.   Particle  Collection   Characteristics   of  High
    Flow-Rate Single Stage  Impactors, M.S. Thesis, University of  Florida.

5.  McFarland,  A.P.,  C.A.  Ortiz  and  C.E.  Rodes, 1979-   Characteristics  of
    Aerosol  Samplers  Used  in  Ambient  Air   Monitoring,   Presented  at   86th
    National Meeting  of the American Institute of  Chemical Engineers, Houston,

6.  Watson, J.G.,  J.C.  Chow,  J.J.  Shah and T.G.  Pace,  1983.   The  Effect  of
    Sampling Inlets on  the  PM-10  and PM-15 to TSP Concentration  Ratios, JAPCA

                                    WIND SHROUD
Figure 1. Schematic diagram of mobile sampling system.

                                                         NB-HIGH-205 M9/m3

                                   GRAND AVERAGE -134 Wm3
                           10                  30                     100

                                      PARTICLE DIAMETER (Dp),Mm

                              Figure 2.  Large particle mode distributions.


                    MASS AND CHEMICAL ANALYSIS
                         Kenneth E. Noll
             Department of Environmental Engineering
                 Illinois Institute of Technology
                        Chicago, IL  60616
                           Yaacov Mamane
                      NRC Research Associate
            Environmental Sciences Research Laboratory
               U.S. Environmental Protection Agency
                Research Triangle Park, NC  27711
     A unique combination of an effective sampler and analysis of individual
particles has been used in studying large particle (> 5 Um) at a rural
site in Eastern United States.  The sampler is a modified "high volume"
rotary inertial impactor, which consists of four collectors of different
widths, rotating at high speed and collecting particles by impaction.
The collector surfaces were mylar films coated with apiezon to ensure
retention.  After sampling, the collection surfaces were weighted to
obtain the mass-size distribution.  A section of the mylar sample was
transferred to a scanning electron microscope to study in detail the
morphology and elemental content of individual particles.

     The following features characterize the impactor: (a) Particles 6
to 100 are collected effectively on four stages.  Stages A, B, C and
D — collect particles larger than 6 urn, 11 ym, 20 ym, and 29 ym,
respectively.  (b) The sampler operates at high velocities, therefore
sampling a "large volume" of air — a necessary requirement because of
the low concentration of large particles; (c) Due to the special collection
technique, no losses "to the walls" or "bouncing off" are expected.  To
insure a high degree of retention the collector faces were coated with a
thin film of apiezon; (d) No problems associated with isokinetic sampling
at variable wind speed are expected, since the collectors operate at
velocities considerably higher than the average wind speed, and the
instrument has a wind vane to point it into the wind; (e) The various
stages allow the collection of particles over restricted ranges of the

size distribution without interference from particles outside of the
range.  This eliminates errors in counting and x-ray analysis of indi-
vidual particles due to the excessive covering of the collector surfaces
by numerous small particles.

     Samples were taken during the month of August 1983 in a rural site
in western Maryland as part of the Deep Creek Lake (DCL)  Experiment.
The objective of the DCL study was to collect air quality data base and
source signatures to determine the impact of primary emissions and
secondary pollutants from combustion sources on a remote  site.  The
sampling site is located in a rural area surrounded by over fifteen
coal-fired power plants which are big enough (> 1000 MW)  and close
enough (50 to 300 km) to have a significant impact on the site.

     For the electron microscopy analysis, the mylar films were observed
in an optical microscopy to verify a homogeneous collection of particles
on the collector surface.  The analysis includes particle size, shape
and special surface features, and elemental content of the particle.

     Out of the samples collected at the DCL site, two samples which
were chosen for SEM analysis  representing two different atmospheric
conditions — low versus high wind speed.  Both samples were represen-
tative of midday summertime conditions.

     Information on a few hundred individual coarse particle has been
obtained, including their heterogeneity and surface properties.  Based
on the elemental content and  mosphology, particles were assigned to
several category groups such  as clay minerals, quartz, calcite, gypsum,
coal and oil fly ash, biological (pollen, spores, plant debris) particles,
and special particles — mostly of anthropogenic sources  with high metal
content — rich in Fe, Pb, Zn.

     The main results are summarized as follows:

(a)   In the rural  area  studied  here  the  aerosol mass  distribution peaks
     in the 10  to  20  ym range with  fairly  significant mass  in  the 20 to
     60 ym.   During windy conditions mass  concentration  is  higher for
     most parts of the  size  range, but not in  the below  10  ym  range.
     The wind speed may have two  effects on aerosol concentration:
     higher wind speeds cause resuspension of  particles, while low wind
     speeds are associated with less dispersion and higher  concentration
     of the smaller size fraction.

(b)   Electron microscopy analysis of individual large particles revealed
     the overwhelming presence  of natural  contributions  in  the whole
     range, namely minerals  (clay minerals,  calcite and  quartz — about
     50 percent),  and biological  particles such as pollen and  spores.

(c)   Contribution of  anthropogenic  sources to  large particles  was
     limited to a few percents  and  mainly  to particles smaller than 10
     ym.  Most of these were fly  ash transported  from coal-fired power
     plants situated  50 to 300  km upwind of the sampling site.

(d)   Pollen particles represent a large  fraction  of the  large  particles
     collected at the DCL site.  Different types  have been  observed even
     on the calm day  indicating a fairly long  residence  time  in the air.
     The pollens contained large  amounts of sulfur, either  as  small
     sulfate particles deposited  on the  pollen surfaces, or as absorption
     of SO  through the wet  surfaces.

(e)   Mineral particles were  found to be  enriched  in sulfate.   As with
     the pollen the sulfate  may have accumulated  on the  particle surfaces
     while being airborne.  The sulfate  was found to  be  associated with
     calcite and clay minerals  in significant  amounts, about  1.5 to 3
     percent of the particle mass,  or an average  of 0.02 g  SO, . g


                    Eliezer Ganor* and Rudolf F. Pueschel**

*   Research  Institute  for Environmental  Health
    Ministry  of Health  and Tel-Aviv  University,  Israel

** NUAA, Environmental  Research  Laboratories,  Air  Resources  Laboratory
Boulder, CO 803U3, USA

     A quantitative method for the analysis  of  individual  micrometer-size  dry

and wet aerosols is developed.   It is  based  on  mineralogical  and  microchemical

analysis.  An aerosol compound is analyzed for  crystallography  by  petrographic

microscopy, and for anions and cations by electron microscopy.  Analysis  of

the anions NO^' and S0^"~ is based on  their  microchemical  reactions  with

nitron and Ba Cl^, respectively.  The  microchemical  analysis  is identified  by

transmission electron microscopy and the anion  Cl and cations such as  Ca,  Mg,

K and Na are determined with a scanning electron microscope  interfaced with  an

X-ray energy spectrometer.  The methods were tested  in several  locations:

(a) At the Boulder Atmospheric Observatory,  a 300 m  tower  located  in  a rural

area, during Chinook conditions and within clouds,   (b) At Tel-Aviv

University, during air pollution and Sharav  conditions,  (c)   At Masada, Dead

Sea, 10 cm above sea level, during winter conditions.  The aerosols  are

classified as mineral, soot containing, water containing and  electrolite

(mixed dry and wet aerosols).


     The study of individual micrometer-size  aerosols  provides  a  great  deal  of

new information on the characteristics of aerosols, which  otherwise  cannot  be

obtained.  The aerosols in the atmosphere are  not  stable;  they  change  during

transport due to chemical  reactions within the  particles  and  of particles with

gases, coagulation of particles  and alterations  of  relative  humidity (Hanel

and Zankl, 1979; Mamane et al.,  198U).   These  changes  can  be  noticed by

analysis of individual aerosols  from different  sources.

     There are several sources of  aerosols, which  can-  be  grouped  in  two

categories:   (1) Natural  (soi1-derived aerosols, sea-spray aerosols, volcanic-

derived aerosols and  organic particles)  and  (2)  Anthropogenic.  The  basic

aerosol components have been classically defined as water-soluble  particles,

dust like particles,  oceanic particles,  soot  particles  and ash  particles  (WCP,


     In our work we classify the aerosols into  four types: mineral  and  dust

like particles, soot  particles,  electrolite particles  and  mixed particles.

The electrolite particles are sea  salt particles generated at the  sea  surface

by the action of the  wind: such  particles are  halite  (NaCl),  sylvite (KC1),

carnallite (K MgCl2-6H^O) and those containing  sulfate  and nitrate.   The  mixed

particles consist of  several compounds and are  water containing,  such  as  soot

coated with Ho S0« and dust  like particles coated  with  electrolites.


     Aerosols were collected on electron microscope  (EH) grids,  on  plain  glass

and on blue gelatine, that were mounted on three stages of a four-stage

Casella impactor.  The aerosols were collected for chemical and  mineralogical

analysis by microchemical spot test (Mamane and Pueschel,  1980),  X-ray energy

dispersion analyzer, electron microprobe analyzer and petrographic  microscope

(Ganor et al., 1982).


     The methods were tested in several locations:   (1) At the  Boulder

Atmospheric Observatory  (BAO), USA.  The BAO is a 300 m tower located in  a

rural area 20 km east of Boulder and 25 km north of  Denver.  Aerosol  samples

were collected at 10, 22, bO, 100, 150, 200, 250 and 300 m levels  on  the  tower

(Ganor and Van Valin, 1982).  (2)  In Israel, at three places: (a)  Tel-Aviv

University (TAU), a residential  area in north Tel-Aviv, on the  roof of a  15 m

building, located 2 km from the Mediterranean Sea;  (b) Tel-Aviv  Marina shore

(TAM), 0.5 m above sea level; and  (c) Masada shore  (DSM),  10 cm  above the Dead

Sea level.

     The aerosols were collected during different meteorological  conditions:

at BAO, during chinook conditions  and within a cloud; at Tel-Aviv,  in winter,

during air pollution and Saharan dust storms; at Masada shore,  in  winter.



     A treated an a non-treated marked  screen  were  observed  by  a  petrographic

microscope, a transmission electron microscope  (TEN)  and  a  scanning  electron

microscope  (SEN) for  shape,  size  distribution  and chemical  composition.

     Figures 1-3 are  photomicrographs of  particles  collected  at TAU  on  a  dusty

day-November 19, 1983.  Figure  1  is a petrographic  photomicrograph  of  Saharan

particles on a nontreated screen.  The  particles were  identified  as  quartz,

calcite, dolomite, feldspar,  clay minerals,  fossile fragments  and  oil  soot.

Figure 2 is an SEM photomicrograph of the  same  sample  on  a  non  treated

screen.  The particles were  analyzed for  their  elements with  an X-ray  energy

analyzer.  The major  elements found in  the particles were Al,  Si  and Ca.

Figure 3 is an SEM photomicrograph of the  same  sample  on  a  treated  Bad 2

screen.  The figure shows that  the particles are mixed sulfates.   The  circular

spots indicate the presence  of  sulfates.   Most  of the  Saharan  particles  are

coated with a thin layer of  sulfate.  In  the figure there are  three  particles,

identified as (a) clay, (b)  textularia  fossil  fragment and  (c)  calcium

carbonate aggreate.

     Reaction spots of individual droplets were tested in the  laboratory  on

blue gelatine, BaC^  and nitron pre-coated screens  for shape,  size

distribution and microchemical  composition by  a petrographic  microscope  and a

scanning transmission electron  microscope  (STEM).   Later, the methods were

tested at TAM and TAU.

     Figures 4 and 5  are typical  photomicrographs of  aerosols  collected  on

February 21, 1994,  at TAM and on  October  24, 1983,  at  TAU.  Figure  4a  is  a


typical ThH photomicrograph of sea drops  collected  at  JAM  on  February  21,

1984.  The reaction spots indicate the impaction of  the  drops  onto  a  HaCIo

pre-coated EM screen.  Figure 4b is  an SEN  photomicrograph  of  the  sample  shown

in Figure 4a.  The cubic particles inside the  reaction spot are  salt

particles, containing Mg, Na, S, Cl,  K and  Ca.  The  use  of  the TEH  and  SEM

with an X-ray energy analyzer in this case  gives us  more information,  which

otherwise cannot be obtained.  Figure 5  is  a typical  photomicrograph  of air

pollution particles collected onto gelatine pre-coated glass  at  TAU on  October

24,  1983.  Some of the rounded particles  show  drop  replicas,  and  are  therefore

identified as water-containing particles.

      The size distribution of particles  collected  on  stages  3 and  4 of the

Case! la impactor is shown in Figure  6 (0.8  
collected  on  a  pre-coated  carbon  screen.   The particles inside the drop are
NaCl crystals.
     Figure  9 shows  typical  TEM  and  SEM photomicrographs of nitrate,  and the
X-ray energy  dispersion  spectrum  (XEDS).   The nitrate,  identified by  the
fingerlike microreaction,  contains  Al,  Si,  Ca and  a trace of K and Fe.

     The particles,  which  have  been  tested  in  several  locations,  were
classified as dust  like  and  minerals,  soot  containing,  water containing,
droplets, nitrate and  sulfate.   Tables  1-3  summarize the microchemical
analyses:  Table 1,  the  frequency  of  elements  present  in the aerosol
particles, in percentage collected at  BAG,  TAD and  DSM;  Table 2,  the
percentage of the elements contained  in  the particles  at TAU, TAM and DSM;  and
Table 3, the percentage  of aerosol  compounds  at  BAO,  TAM,  TAU and Dead  Sea,
based on sulfate and nitrate  identification using the  Bad- and  nitron
techniques, by  petrographic  microscopy  and  by  SEM-XEDS.

     The size spectra of  cloud  droplets  and  aerosols  and  of  the  aerosol
compounds were measured using multi-microchemical  techniques,  in  different
meteorological conditions.  The techniques provided  an  assessment  of  water-
containing aerosols, and  most particles  were  identified as electrolite
aerosols.  It was also found that  because of  the  high  relative humidity  at TAU
the electrolite particles became water-containing  droplets.  Simultaneously,
various techniques were used to obtain relevant data  on individual  aerosol

compounds.  For Instance, it was found out that a considerable portion  of

particles contain dust  like nuclei, as in the Dead Sea droplets.  On  a

polluted day at TAD, about 42% of the micrometer-size particles were  found  to

be electrolites.  25%, soot; and 33%, dust like particles.  The mixed sulfate

and nitrate particles were probably formed through a heterogeneous  nucleation

of S02 and N02 on the surface of insoluble dust like particles.

1.  Ganor, E. and C. C. Van Valin, 1982.  Vertical profiles of  gases  and
    particles in the nonurban atmosphere.  Proceedings, 2nd symposium on  the
    Composition of the NonurbanTroposphere, Wi11iamsburg, VA.,  May  1982.  pp.

2.  Ganor, E., R. F. Pueschel, and C. T. Nagamoto.  Sulfates  and  nitrates:
    Concentration as function of particle size in eastern Colorado,   (in
    preparati on).

3.  Hanel, G. and B. Zankl , 1979.  Aerosol and relative humidity:   Water
    uptake by mixtures of  salts.  Tel 1 us, 31, 478-486.

4.  Mamane, Y. and R. F.  Pueschel, 1980.  A method for the detection  of
    individual nitrate particles.  Atmospheric Environment, 14,  pp.  629-639.

5.  Mamane, Y., E. Ganor,  and A. E. Donagi, 1980.  Aerosol composition of
    urban and desert origin in the eastern Mediterranean.  I:   Individual
    particle analysis.  Water, Air, and  Soil  Pollution, 14:   pp.  29-42.

6.  WCP  (World Climate Programme) 1983.  Aerosols and Their Climatic  Effects,
    Williamsburg, Va.


Height (m)
Diameter (p.m)
No. of aerosols


Ele ment




Height (m)
23 Jan 84

• • •
21 Feb 84 ly
* • •
• • *
• • •
• • •
• • •
• • •
Nov 82

• • •
• • •
• • •
• • •
21 Dec 83

• • •
• • •
• • •
• • •
* • •

Height (m) 300
Date 16 Apr 81
Conditions Chinook
Sulfate 95
Dust-like b
Water Containing ...
Soot containing ...
# 42% of the minerals
* The droplets contai
** .The droplets contai
## The droplets contai
27 Jul 81

Within Cloud

• • •
• • •
• • •
contain sul
ned Na, Al ,
24 Oct 83


Si , S, Cl , Ca
ned Na Cl and elements as
ned Mg, Al ,
Si, S, Cl , K,
19 Nov 82


• • •
• • •
• • •

, Fe.
Mg, S, K, Ca.
Ca, Fe.
21 Feb 84


• • •
* • •
86 **
• • •
» • *

21 Dec


• • •
* • •
68 ##
• • •
• • •

Fig. 1 - Petrographic microscope  photo-
         graph of Saharan particles  on
         a non-treated EM screen
Fig. 2 - Typical  non-treated screen SEM
         photomicrograph of Saharan par-
         ticles shown in Fig. 1
      Fig. 3 - Photomicrograph of the BaClp treated sample  shown  in  Fig.  2.
               The circular reaction spots indicate the presence  of  sulfates.
               (a)  clay,  (b)  textularia  fossil  fragment, and  (c)  calcium
               carbonate  aggregate.

   Fig.  4a - TEM photomicrograph  of  the  pre-coated BaCl2
             Mediterranean  drops  collected  at  TAM
Fig. 4b - SEM photomicrograph  of
          The cubic particles  in
          containing Mg,  Na,  S,  Cl,  and  Ca,
the drops shown in Fig.  4a.
reaction are salt particles,

Fig. 5 - SEM microphotograph of particles collected onto gelatine at TAU during
         a polluted  day  -  October 24, 1983. The circular replica indicate the
         presence  of water.

                   I                               TAU
                   |                            24.10.83
10 '   I


                                     2      3    456789 10

                                 R AD IU S  ,
       Fig.  6  -  Particle size distribution of the sample collected at TAU
                on  October 24, 1983

Fig. 7 - Typical  TEM microphotograph of Dead  Sea  aerosols  collected  on  pre-coated
         (a) BaC^ and (b)  nitron.  The circular reaction spots  on  the BaClp  indicate
         the presence of drops.  The circular  spot reaction on the  pre-coated nitron
         shows the presence of drops.

Fig. 8 - SEM photomicrograph  of cloud  droplets  collected  onto
         pre-coated carbon  screen  at BAO.  The particles inside
         the drops are NaCl  crystals.

Fig. 9 - TEM and SEM with X-ray energy  spectra  of
         nitrate collected at BAO.  The  nitrate  is
         identified by the finger-like  micro-reaction,
         It is mixed with Al, Si,  Ca,  K,  and  Fe.

                               Wayne  R. Ott
                   U.S. Environmental Protection Agency
                    Office of Research and Development
                          Washington, DC  20460

     Determining the risk of environmental pollution to public health
requires a knowledge of five fundamental components:  (1) the sources of
pollutants, (2) the transport of these pollutants from sources to humans,
(3) the distribution of exposures of  humans to these pollutants,  (4) the
doses received by people who are thereby exposed, and (5) the adverse
health effects resulting from these doses.  These five components may be
viewed as links in a chain — from source to effect — comprising the
full risk model (Figure 1):



      Figure 1.  Major components of conceptual risk model relating
      the sources of environmental pollution to the ultimate effects
      of pollutants on the population.
     Despite the importance of each of the five components for estimating
the public health risk associated with environmental pollution, our
scientific knowledge about each component is not balanced.  Usually,
environmental pollution comes to the attention of public officials be-
cause pollutant sources, such as a smoke stack plumes or leaking toxic
waste drums, provide obvious evidence of a disturbing environmental
condition.  Thus, a great deal is known about the sources of pollution,
and source abatement and control has received considerable research
attention.  Once a source of environmental pollution is known and iden-
tified, interest often focuses on the manner in which the pollutant
moves through the environment — its fate and transport — ultimately
becoming assimilated by ecosystems or transported to humans.  As with the
source component of this risk model, the fate and transport component

likewise has received considerable research attention.  The field of
meteorology has developed a great number of atmospheric dispersion models,
and other fields have developed models for the movement of pollutants
through streams, soil, and the food chain1.
     As with the first two components, the last component — the effects
of pollutants on humans — also has received considerable research atten-
tion.  Numerous studies have been undertaken relating various exposures
and doses to identifiable effects on animals and humans, as can be seen
in any of the published air quality criteria documents'^ 3.  However,
our knowledge of two important components of the risk model — exposure
and dose — is very limited for most pollutants.  Accurate exposure data
unfortunately are lacking for most of the air pollutants that EPA regulates.
     The environmental risk model is serial in structure:  the output of
each component serves as the input to the next component.  Thus, the
absence of valid information on any component seriously impairs our abil-
ity to assess public health risk, and the absence of human exposure data
has serious implications for regulatory policies.  If, for example, human
exposures to a criteria air pollutant were found to be negligible, then
the public health risk of the pollutant may be exaggerated, and concern
about controlling this pollutant could be reduced.  Conversely, if human
exposures to a pollutant were found to be higher than previously sus-
pected, then additional control actions might be warranted.  In all
cases, the important information needed is the frequency distribution of
exposures of the population.

     Two alternative conceptual approaches have been proposed for obtain-
ing information on the frequency distribution of exposures of the popula-
tion to environmental pollutants:'*'^
     Approach A.  An obvious solution, called the "direct approach" by
Duan^, is to measure the concentration at the boundary of the person
by monitoring the air he breathes, the water he drinks, and the food he
eats.  Several recent field studies have implemented this conceptual
approach.  The Total Exposure Assessment Methodology (TEAM) study measured

the concentration of an important class of chemicals, volatile organic
compounds (VOC's), in the air, drinking water, and breath of respondents
using personal monitoring techniques"" .   In order to generalize to a
larger population than the number actually surveyed, a multi-stage statis-
tical sampling design was used.  In Elizabeth and Bayonne, New Jersey,
365 respondents carried a Tenax portable personal exposure monitor, and
the levels of VOC's in their breath and drinking water also were deter-
mined.  A major finding was that personal exposures (and indoor levels)
were much greater than outdoor exposures for at least 1 1 important car-
     More recently, the Denver-Washington, DC, carbon monoxide (CO) human
exposure field survey was carried out10.   Because CO is associated only
with air pollutant exposures, it was not necessary to monitor food or
drinking water.  A specially designed personal exposure monitor for CO
was developed which could measure and record concentrations with a time
resolution of one minute of less11.  An interviewer delivered the monitor
to each respondent and picked it up 24 hours later.  Each respondent
carried the monitor with them while engaging in their normal daily act-
ivities.  By interpreting their diaries listing the times that each
activity began, it was possible to obtain 712 24-hour exposure profiles
in Washington, DC, and 450 48-hour profiles in Denver, CO.  Although
these data currently are being analyzed,  many new findings are emerg-
     Approach B.  An alternative approach, called the "indirect approach"
by Duan^, is to measure and fully characterize pollutant concentrations
in the locations (called "microenvironments" )  people normally visit.
Then, by combining this information with data from activity pattern and
time budget studies, it is possible to compute an estimated exposure
profile for each person.  This approach initially was suggested by Fugas1^
and is discussed by Duan^.  Computer models such as SHAPE1 ^ and NEM1 ?
have been developed for combining the activity data with the microenviron-
mental concentration data, but human exposure activity modeling is in
its infancy and needs further development and field testing.


     A new methodology, human exposure assessment, is emerging for

determining the frequency distribution of exposures of the population
to environmental pollutants.  Two approaches, direct measurement and

indirect estimation through models, have been developed.  Several field

studies have been undertaken demonstrating the feasibility of the direct

approach, yielding a wealth of new exposure data and many important new

findings.  The indirect approach has not been fully developed and needs

further work, but it, too, may yield much important exposure informa-

tion.  Initial studies have dealt with volatile organics and CO, and the

same methodology now needs to be extended to N02, inhaled particles, and

other important pollutants.


1.  Ott, Wayne R., ed., "Proceedings of the EPA Conference on Environ-
    mental Modeling and Simulation," U.S. Environmental Protection Agency,
    Report No. EPA-600/9-76-016, Washington, DC, July 1976.

2.  "Air Quality Criteria for Particulate Matter and Sulfur Oxides," U.S.
    Environmental Protection Agency, Vol. I, No. EPA-600/8-82-029a; Vol.
    II, No. EPA-600/8-82-029b; Vol. Ill, No. EPA-600/8-82-029c;  Research
    Triangle Park, NC, December 1982.

3.  "Air Quality Criteria for Oxides of Nitrogen," U.S. Environmental
    Protection Agency, Report No. EPA-600/8-82-026F,  December 1982.

4.  Ott, Wayne R., "Concepts of Human Exposure to Air Pollution," Environ-
    ment International, Vol. 7, pp. 179-196, 1982.

5.  Duan, Naihua, "Models for Human Exposure to Air Pollution,"  Environment
    International, Vol. 8, pp. 305-309, 1982.

6.  Wallace, Lance,  Ruth Zweidinger, Mitch Erickson,  S. Cooper,  Don
    Whitaker, and Edo Pellizzari, "Monitoring Individual Exposure Measure-
    ments of Volatile Organic Compounds in Breathing-Zone Air,  Drinking
    Water, and Exhaled Breath," Environment Internatinal, Vol.   8,  pp.
    269-282, 1982.

7.  Zweidinger,  Ruth, Mitch Erickson, S. Cooper, Don  Whittaker,  Edo
    Pellizzari,  and  Lance Wallace,  "Direct Measurement of Volatile
    Organic Compounds in Breathing-Zone Air, Drinking Water,  Breath,
    Blood, and Urine," U.S. Environmental Protection  Agency,  Report No.
    EPA-600/4-82-015, Washington, DC, June 1983.

 8.  Pellizzari,  E.D., T.D. Hartwell,  C.M.  Sparacino,  L.S.  Sheldon,  R.
     Whitmore, C. Leininger, H.  Zelon, and  L.  Wallace, "Total Exposure
     Assessment Methodology (TEAM)  Study:   First Season - Northern New
     Jersey," Research Triangle  Institute,  Report No.  RTI/2392/03-03S,
     Research Triangle Park, NC,  June  1984.

 9.  Wallace, L., E.  Pellizzari,  T.  Hartwell,  M. Rosenzweig,  M.  Erickson,
     C.  Sparacino, and H.  Zelon,  "Personal  Exposure to Volatile  Organic
     Compounds:  I.  Direct Measurements in Breathing-Zone  Air,  Drinking
     Water, Food, and Exhaled Breath," in press, Environmental Research,
     1 984.

10.  Akland, Gerald G.,  Wayne R.  Ott,  and Lance A.  Wallace,  "Human Exposure
     Assessment:   Background, Concepts,  Purpose, and Overview of the
     Washington,  DC-Denver, Colorado Field  Studies," Paper  No. 84-121.1
     presented at the 77th Annual Meeting of the Air Pollution Control
     Association, San Francisco,  CA, June 24-29, 1984.

11.  Ott, W.R., C. Williams, C.  Rhodes,  R.  Drago,  and F.  Burmann,  "Application
     of  Microprocessors  to Data  Logging Problems in Air Pollution  Exposure
     Field Studies,"  Paper No. 84-121.2 presented at the  77th Annual
     Meeting of the Air  Pollution Control Association, San  Francisco,  CA,
     June 24-29,  1984.

12.  Johnson, Ted, "A Study of Personal Exposure to Carbon  Monoxide in
     Denver, Colorado,"  Paper No. 84-121.3  presented at the 77th Annual
     Meeting of the Air  Pollution Control Association, San  Francisco,  CA,
     June 24-29,  1984.

13.  Hartwell, Tyler  D., Carlisle A. Clayton,  Raymond Michie, Jr., Roy W.
     Whitmore, Harvey S. Zelon,  and  Deborah A. Whitehurst,  "Study  of Carbon
     Monoxide Exposure of Residents  of Washington,  DC," Paper No.  84-121.4
     presented at the 77th Annual Meeting of the Air Pollution Control
     Association, San Francisco,  CA, June 24-29, 1984.

14.  Wallace, Lance A.,  David T.  Mage, and  Jacob Thomas,  "Alveolar
     Measurements of  1,000 Residents of Denver and  Washington, DC  — A
     Comparison with  Preceding Personal Exposures," Paper No. 121.5
     presented at the 77th Annual Meeting of the Air Pollution Control
     Association, San Francisco,  CA, June 24-29, 1984.

15.  Fugas, Mirka, "Assessment of Total Exposure to Air Pollution,"
     Proceedings  of the  International  Conference on Environmental  Sensing
     and Assessment,  Paper No. 3R-5, Vol.  2, IEEE #75-CH  1004-1, Las Vegas,

16.  Ott, W.R., "Exposure Estimates  Based on Computer Generated  Activity
     Patterns," Paper No.  81-51.6 presented at the  74th Annual Meeting of
     the Air Pollution Control Association, Philadelphia, PA, June 21-26,
     1981 .

17.  Johnson, T., and R.A. Paul,  "The  NAAQS Exposure Model  (NEM) Applied
     to  Carbon Monoxide," U.S. Environmental Protection Agency,  Office of
     Air Quality Planning and Standards, Strategies and Air Standards
     Division, Research Triangle Park, NC,  April 1982.

                        IN THE WINTER OF 1982-83
     During the  winter of 1982-83,  the  U.  S.  Environmental Protection
Agency conducted  a large-scale urban  field study  to  develop  and test
methodologies for  determining,  with known  accuracy,  the  exposures  to
carbon monoxide  (CO)  of the population  of  a  city.   The primary study
objective was to  develop  and evaluate a methodology  for measuring the
distribution of CO exposures of a representative population of an urban
area.  Two  urban areas were chosen for study; Denver,  Colorado,  and
Washington, D.C.  These areas were selected because they differ in ele-
vation, relative CO levels based on  historical  fixed site data, diversi-
ty of land use characteristics  and commuter  patterns.  Approximate dates
of field monitoring  were  November 1,  1982,  through-February 28, 1983.

     Participants in the study were chosen using a 3-stage design.  Ap-
proximately 3200 households in Denver and 5800  households in Washington,
D.C., were  screened  by telephoning  a  representative  random  sample  of
the population.  During the  screening process,  respondents  were asked
about their smoking habits, commute  times,  and  other factors which might
influence CO exposures.  Data from the screener survey made it possible
to subsequently  create a  stratified random sample  of individuals with
particular characteristics of  interest.   For example, only* non-smokers
were selected in  the  final sample,  and  persons  who commuted long dis-
tances were more  heavily   sampled  than  were  those who  commuted short

     The exposure measurements were made with a specially designed per-
sonal exposure  monitors  (PEM)  with  a  built-in  data  logger.   The data
logger was developed to provide an integrated  value  expressed in ppm-
minutes which was determined by change of activity pattern or automati-
cally on  the  clock-hour.   The clock-hour  values  were necessary  for
comparing PEM  results  with  the fixed  site results.   The interviewer
visited the respondent  and left a  calibrated PEM with instructions  for
its use and  a  diary.   The respondent carried the PEM  recording each
change of location  and activity  in the  diary  and  depressing  the data
logging button  when  they  changed activity.   A questionnaire  also  was
administered to obtain detailed information  about the respondent's home,
workplace and commute  habits.   Details of  the  survey  are  presented  by
Whitemore, et_ al_. 1  Other details are  reported in the  final  reports  by
Johnson^ and Hartwell.^

                             Study Results

1.   Quality Assurance

     (a)  Precision of PEM Values

          The assessment of  PEM precision was  determined by  having a
     member of  the  project  field  staff  carry  two  or more  randomly
     assigned monitors for a 24-hour period.   The monitors were exposed
     to typical sampling conditions  of  changing temperature, humidity,
     elevation, etc. , as well  as  to the  vibrations and physical shocks
     inherent in transporting the instruments.  In Washington, the mean
     relative of the  standard  deviation  of  the measurement  pairs  was
     30.6%.  In Denver, where  the average concentrations  were higher,
     the mean relative standard deviation was 14.2%.


          Two independent  audits  of the  PEM's  were  conducted  by  the
     Quality Assurance Division, Environmental Monitoring Systems Labor
     atory, EPA.  The first  audit  was conducted at  the start of the pro-
     ject and the second  near  the end of the study.  Results  of both
     audits indicated that  the audited  PEM's were within ±10%  of  the
     audit gases in both cities.

2.   Fixed Site Concentrations

     One goal of these studies  was to compare exposure results obtained
from fixed monitors  with  directly  measured  personal exposure  for  CO.
It should  be  noted that  the  National  Ambient Air  Quality  Standard
(NAAQS) levels  for  CO are 35  ppm for  1-hour concentrations  and 9  ppm
for 8-hour concentrations.   During  the  study period the  35  ppm level
for 1  hour  was never  reached  in  Washington, but  it  was exceeded  in
Denver on one day (12/16/82) at one site «1%).   The  8-hour level  was
exceeded at two Washington sites -  one  site  had one exceedance and the
other site had five exceedances (0.5% site-days).  The 8-hour level was
exceeded at 11 Denver sites (8.7% site-days).

3.   Personal Exposures

     The field study yielded 712 24-hour exposure profiles in Washington
and 900 24-hour exposure profiles  (450 persons @  2  days each) in Denver.
The 8-hour maximum results in Denver were approximately twice as high as
the levels found in Washington.   (The fixed  site  CO  concentrations  at
Denver were also about  twice that observed  in Washington.)  The Denver
personal exposure distribution  indicates 10.7%  of the  8-hour maximum
daily CO exposures  were above 9 ppm.  This compares to 3.9% above 9 ppm
observed in Denver  at the  fixed sites.

Other results include:

     (a)  The distribution  obtained  from the  concentrations  measured at
     a combination of fixed site monitors can generally provide a reason-
     able measure of exposure to  CO  for the  study population  over  the
     distribution except for the upper 10 percent of exposure.

     (b)  Personal  CO  exposures were  higher in  microenvironments  asso-
     ciated with motor  vehicles, such as while commuting.

     (c)  Personal  CO  exposures were  also  higher  for persons  in  high-
     exposure occupations,  e.g.,  truck  drivers,  construction  workers
     and garage/service station workers.

     (d)  In Denver, indoor  concentrations  were  higher  than  correspond-
     ing fixed site  concentrations  during the time period 0900-1600 hours.

     (e)  In Denver, the  first  day and the  second  day personal exposure
     profiles are approximately equivalent for workdays.


     From these  studies we can conclude that  the methodology exists  for
conducting exposure studies  for  CO for an urban  area.   The  studies have
provided an extensive data base from which statistical comparisons can be
performed between population subgroups, between fixed site concentrations
and personal exposure,  and between indoor and outdoor concentrations.   In
addition, factors associated with exposures can be estimated and modeled.
It is  clear  that an extension  of this  concept to other  pollutants  and
other areas  over differing  time  periods  (for  temporal  resolution)  is


1.  R. W.  Whitmore,  Jones, S. M., and Rosenzweig, M. S.  "Final Sampling
    Report for the Study of Personal CO Exposure."   Report by Research
    Triangle Institute to  the  U.  S.  Environmental  Protection Agency,
    Research Triangle Park,  N.  C.,  January 1984.

2.  T. Johnson.   "A Study of  Personal Exposure to  Carbon Monoxide in
    Denver, Colorado."  Report by PEDCo  Environmental,  Inc.,  to the U.
    S. Environmental Protection Agency,  Research Triangle  Park,  N. C.,
    December 1983.

3.  T. D.  Hartwell,  et_ al.   "Study  of Carbon Monoxide   Exposure  of
    Residents of  Washington,  D.C.,  and  Denver, Colorado."   Report  by
    Research Triangle Institute  to  the U.  S.  Environmental Protection
    Agency, Research Triangle Park,  N.  C., January 1984, Parts  I and II.

                 A.  R. Hawthorne, T. G.  Matthews,  R.  B. Gammage,
                           C. S.  Dudney, and T.  Vo-Dinh
                       Health and Safety Research  Division
                          Oak Ridge National Laboratory
                                                         By acceptance of this article, the
                                                         publisher or recipient acknowledges
                                                         the U.S. Government's right to
                                                         retain » nonexecutive, royalty-free
                                                         license in and to any copyright
                                                         covering the article.
* Research sponsored  by   the  Tennessee   Valley   Authority   under  Interagency
  Agreement IAG-40-1406-83 with the Martin Marietta  Energy Systems, Inc.   under
  Contract DE-AC05-840R21400  with the U.S.  Department of Energy.

                 A. R. Hawthorne, T.  G. Matthews, R.  B. Gammage,
                          C.  S.  Dudney,  and T. Vo-Dinh
                       Health and Safety Research Division
                          Oak Ridge National Laboratory

     Indoor air pollutants are increasingly recognized as important  contributors
to  the  total  public  exposure  to  pollutants.   Radon,  formaldehyde, volatile
organic  compounds,  combustion  gases,  and  particulates  are  among  the  more
important  indoor  air pollutants.   Indoor levels may in fact be comparable  to or
greater than levels that have caused concern  outdoors.   Sources  identified  as
contributing  to  reduced  indoor  air  quality  include  construction  products,
consumer,  products,  combustion  appliances,  and  lifestyle  habits.   When  the
potential  for elevated concentrations is considered in conjunction with the fact
that many people spend a large fraction  of  their  time  indoors,   the  need  to
understand  better  the  indoor  component  of the population's total exposure is
evident.  In addition to assessing the direct  impact of indoor air quality,  there
is  a need to determine the impact  of indoor exposures on conclusions drawn  about
outdoor air quality.  Much health effects information is based on the  assumption
that  outdoor air quality is the dominant determinant of observed health effects.
A better assessment of pollutant exposures from indoor air  relative  to  outdoor
air is necessary to test this assumption.
     For approximately five years.  Oak  Ridge   National  Laboratory  has  had  an
active   indoor   air  quality  research  program.   Areas  of  activity  include
instrumentation and methods development, source characterization, field  studies,
modeling,  remedial  measures,  and  impact  assessment.  This paper will briefly
review the following components of our research:   (1)  measurement  developments,
(2) source characterization,  and (3)  field studies.

     There  is  a  need   for   relatively   low-cost,   easy-to-use   monitoring
instrumentation  that  is sensitive enough to meet the requirements for measuring
indoor air quality.  Much of the available industrial hygiene instrumentation  is
not  suitable for monitoring indoor air quality.  Similarly, much of the equipment
used in assessing outdoor air quality is either too large,   noisy,   or  expensive
for  practical  use in indoor air quality research.   Methods to address this need

have been developed as part of oar research program.
     Both active  and  passive  methods  have  been  developed  for  formaldehyde
monitoring.   A  pumped  molecular  sieve  sampling  technique  was developed and
reported by Matthews, T.  G., and T. C. Howell, 1982a.  This procedure  addresses
the  problem  of water vapor collection and presents a procedure that allows low-
level formaldehyde monitoring even with relatively high  humidities.  The  method
uses  a  simple  water  rinse  desorption followed by pararosaniline colorimetric
analysis.  For a 30-L sample taken at 1-2 L/min a detection limit of about 25 ppb
is  achieved.   A  second  improvement in active formaldehyde monitoring involved
modifications to a commercially available CEA-555 formaldehyde monitor (Matthews,
T.   G.,  1982b).   Using  the  reported  protocol, this instrument has monitored
formaldehyde vapor as low as 10 ppb in a controlled laboratory environment.
     Passive formaldehyde monitoring techniques  have  also  been  developed.   A
dimethylsilicone  membrane sampler containing water sorbent is exposed for a 24-h
period and analyzed using the pararosaniline procedure (Matthews, T. G. , et  al.,
1982c).   The  detection limit for this method is about 25 ppb.  This sampler was
used extensively in our field studies.
     A surface monitor has been developed to measure the formaldehyde  flux  rate
from  a  solid  material  such  as pressed-wood products or a wall insulated with
urea-formaldehyde foam insulation.  The  formaldehyde  surface  emission  monitor
(FSEM)  is  a  device  approximately  20  cm  in  diameter which holds a layer of
molecular sieve parallel to the emitting surface and provides a means to  measure
the  emission  rate nondestructively (Matthews, T.  G., et al., 1984).  For a 2-h
measurement period, a detection limit of about 0.025 mg/cm^-h is achieved.
     Although not developed  specifically  for  indoor  air  quality  monitoring,
recent  advancements  in  screening  methodology  for  PNAs  by  room temperature
phosphorescence and synchronous luminescence by Vo-Dinh, T., 1983, offer  a  low-
cost  means  of  screening  indoor  air  sample  extracts  for PNA content.  This
approach was found to be particularly useful in evaluating indoor air quality  in
homes  with  wood stoves (Vo-Dinh, T., et al., 1984a) .  Another attractive device
that is currently being further evaluated for indoor air  quality  monitoring  is
the  passive  BJA  dosimeter  developed  by Vo-Dinh, T., 1984b.  The monitor is a
diffusion device using a heavy-atom-treated filter paper as the sorbent and  room
temperature  phosphorescence  as the analytical method.  The unit is particularly
attractive in that it does not require sample treatment  after  exposure  and  is
placed directly into a spectrometer for readout.

     Our activities in source characterization have  emphasized  resin-containing
materials  that  emit  formaldehyde.  Laboratory measurements have been conducted
using both small chambers and the surface emission monitor.   Current  activities
also include experiments in a room-sized environmental chamber.
     An early  example  of  formaldehyde  source  characterization  involved  the
measurement  of  emission  rates  from  simulated  wall  panels  containing urea-
formaldehyde foam insulation (UFFI) (Hawthorne,  A.   R.,   and  R.  B.   Gammage,
1982).   The  results  of  this  work  indicated that UFFI could be a significant
source of formaldehyde and that  the  levels  measured  in  the  laboratory  were
similar to levels observed in homes with recently installed UFFI.
     Characterization of formaldehyde release rates  from   fiberglass  insulation
has  recently  been  completed  (Matthews,   T.  G., et al., 1983).  These results
indicate that formaldehyde release from  fiberglass  insulation  is  expected  to
produce a minimal impact on indoor air quality.
     The most extensive source characterization activity  involves  a  continuing
study  of  the  formaldehyde  emission  characteristics of pressed-wood products.
This work includes the measurement of emission rates  of  pressed-wood  materials
from   a   product   survey  (measured  at   standard  environmental  conditions),
measurement of emission decay rates, emission rate   dependence  on  environmental
conditions, and a study of permeation barriers and  potential sinks such as gypsum

     Field studies of indoor air quality in occupied residences are an  important
component  of  our  research  activities.   The major study that we have conducted
involved 40 homes in the Oak Ridge/Knoxville area of East   Tennessee  (Hawthorne,
A.   R. , et al., 1984).  This study measured the levels of formaldehyde,  volatile
organics, particulates, and combustion gases  during  a  one-day  visit  to  each
house.   Formaldehyde concentrations were also monitored once a month with a 24-h
passive sampler for about nine months.   Radon levels  were  measured  in  all  40
homes  using  passive  track  etch  monitors  exposed  for  three months.  Hourly
readings of radon were obtained in a subset of the  homes for periods of up  to  a
week.   Air  exchange rates and meteorological data were obtained during the one-
day visits.  Air leakage was also measured  in a subset  of  the  houses  using  a
blower door (Gammage, R. B. , et al., 1984).

     A continuing investigation of volatile organic compounds in a subset of  the
40  homes  is currently underway.  Compounds more volatile than toluene are being
emphasized using a mixed sorbent bed  collection  tube  and  high-resolution  gas
chromatography.   A portable photoionization gas chromatograph is also being used
to locate sources.
     A preliminary study to measure combustion gases produced from the  operation
of unvented gas space heaters was conducted this spring in six houses.  Levels of
carbon oxides,  nitrogen  oxides,  and  oxygen  depletion  were  monitored.   Air
exchange rate measurements were also performed.
     Monitoring of radon and radon daughter levels in 60 homes in  the  Tennessee
valley is planned to begin this summer.  Quarterly measurements will be conducted
to evaluate the variability of radon in both basements and living  areas  of  the
homes.  Air exchange rates will also be determined.

 1.  Gammage, R. B. ,  A.  R.  Hawthorne,  and  D.  A.  White,  1984.   Parameters
     affecting  air  infiltration  and air tightness in thirty-one east Tennessee
     homes.  ASTM Sysposium on Measured Air  Leakage  Performance  of  Buildings,
     Philadelphia, Penn.

 2.  Hawthorne, A. R., and  R.  B.  Gammage,  1982.   Formaldehyde  release  from
     simulated  wall panels insulated with urea-formaldehyde foam insulation.  J.
     Air Pollut. Cont. Assoc. 32, p.1126.

 3.  Hawthorne, A. R., et al., 1984.  An indoor air quality study of  forty  East
     Tennessee homes.  ORNL-5965, Oak Ridge National Laboratory.

 4.  Matthews, T. G. , and T. C. Howell, 1982a.   Solid  sorbent  methodology  for
     formaldehyde monitoring.  Anal. Chem. 54, p!495.

 5.  Matthews, T. G. , 1982b.  Evaluation of  a  modified  CEA  Instruments,  Inc.
     Model  555  analyzer  for  the  monitoring of formaldehyde vapor in domestic
     environments.  Am. Ind. Hyy. Assoc. J. 43, p547.

 6.  Matthews, T. G., A. R. Hawthorne, T. C. Howell, C. E. Metcalfe,  and  R.  B.
     Gammage,  1982c.  Evaluation of selected monitoring methods for formaldehyde
     in domestic environments.  Environ. Int. 8, p!43.

 7.  Matthews, T. G., et  al.,  1983.   Determination  of  formaldehyde  emission
     levels  from  ceiling  tiles and fibrous glass insulation products.  Project
     report to the U.S. Consumer Product Safety Commission.

 8.  Matthews, T. G., A. R. Hawthorne, C. R. Daffron, M. D. Corey,  T.  J.  Reed,
     and  J.  H. Schrimsher, 1984.  Formaldehyde surface emission monitor.  Anal.
     Chem. 56, p448.

 9.  Vo-Dinh, T.,  1983.   Rapid  screening  luminescence  techniques  for  trace
     organic   analysis.    New   Directions  .in  Molecular  Luminescence.   ASTM
     Publications,  pp5-16.

10.  Vo-Dinh, T., T. B. Bruewer, G.  Colovos, T. J.  Wagner, and  R.  H.   lungers,
     1984a.   Field  evaluation  of   a cost effective screening procedure for PNA
     pollutants in ambient air.  Environ. Sci. Tech., 18, p477.

11.  Vo-Dinh, T., 1984b.  Air pollution:   Applications  of  simple  luminescence
     techniques.   Identification  and  Analysis  ol  Organic  Pollutants in Air.
     Butterworth Publishers, Best, pp.257-269.

                             Robert W.  Coutant
                       Battelle  Columbus  Laboratories
                           Columbus, Ohio 43201
                    Robert G.  Lewis and James  D.  Mulik
                   Advanced Analysis Techniques  Branch,
               Environmental  Monitoring Systems  Laboratory,
                   U.S.  Environmental  Protection  Agency
               Research  Triangle Park,  North Carolina  27711
     Most commercially available passive sampling devices employ activated
carbon as the sorbent.  With such devices, the sorption process is not
thermally reversible, and solvent desorption must be used to recover the
sample.  Consequently, the use of these devices to sample ambient concen-
trations (0.1-lOppbv) of volatile organic compounds (VOC) can impose
severe restrictions on the analytical  techniques (1).   On the other hand,
passive devices using reversible adsorption offer several advantages speci-
fically suited to sampling of ambient concentrations of VOC's.  These
          1.  Independence from solvent contamination
          2.  Increased sensitivity because of the availability
              of the whole sample for analysis
and       3.  more rapid sample turnaround
However, the sampling behavior of these devices differs from the ideal
case normally assumed for activated carbon, and failure to recognize the
differences can lead to biases in sampling and interpretation.
     This paper discusses the mechanics of sampling with reversible
adsorption, and presents a simple model for calculating sampling rates.
This model provides guidelines for proper design and application of passive
monitors employing reversible adsorption, and the performance of the EPA
personal exposure monitor (PEM) is used to illustrate the consequences
of proper and improper application of the fundamental principles.
     There are currently two designs of PEMs that use reversible adsorption.
Both of these use Tenax GC, and their major difference in the thickness of

the sorbent bed.  The EPA PEM is a large face area system having a thin bed,
while the device developed by Brown (2) is a thick-bed system.  The funda-
mental mechanics of sorption are the same for both devices, but the thin
bed system is subject to simplifications that more readily obviate the
significance of key physical parameters.  For the thin bed system, the time
averaged sampling rate can be written as
where RQ is the sampling rate at zero time and is given by R  = DA/£.
(D is the gas phase diffusion coefficient of the sorbate, A is the
effective area of the diffusion barrier, and £ is the effective length
of the diffusion path.)  k is the ratio of R  to the bed capacity, WV.;
where W is the weight of sorbent and V,  is the GC retention volume for
the sorbate.  Equation 1 indicates that for sorbates having relatively
low retention volumes, the sampling rates will be strongly dependent on
sampling time, but this effect can be offset to some extent by design
of the device to yield lower values of R .
     The thin bed model was evaluated through exposure of the EPA PEM
to various mixtures of VOC's in the Battelle dosimeter test facility.
Concentrations were in the range of 1-10 ppbv, and exposure times were
varies from 15 minutes to 24 hours.  Figure 1 shows a comparison of
experimentally determined one hour average sampling rates with values
predicted by Equation 1 for 17 common VOC's.   Agreement is good for all
but 3 compounds.   For acrylonitrile, literature values of the retention
volume vary widely and good agreement could be obtained by choosing a
retention volume near the upper limit of those cited in the literature.
     Figure 2 shows examples of long term behavior typical  of compounds
having high and low retention volumes, with the curves having been calculated
using Equation 1.   In general, we found excellent agreement between Equa-
tion 1 and measured sampling rates for sampling times between 15 minutes
and 24 hours, and for compounds having retention volumes  ranging from 0.5
L/g (trichlorotrifluoroethane) to over 2000L/g (o-xylene).

     An alternative illustration of the applicability of Equation 1 can
be gained by using the experimentally measured sampling rates to calculate
apparent retention volumes.  Table 1 shows calculated retention volumes for
4 VOC's in comparison with literature values for the same compounds.
     Passive monitors utilizing reversible adsorption can be used for
monitoring of ambient concentrations of VOC's, but strict attention must
be paid to device design and bed capacity to avoid severely time sensitive
sampling rates.  The thin bed model, which assumes that all  of the bed
capacity is available, is applicable to the EPA PEM.  However, with thick
bed systems only a small volume of sorbent near the face of the device
will be utilized and sampling rates can be even more time sensitive than
illustrated with the EPA PEM, depending on the face area to volume ratio.
With thick bed systems, the thin bed model is not applicable, and one
must resort to a more complex treatment involving a series  solution
to the problem.

R 'R  = 0.96 + 0.08
 m  p        ~
-j  R  Measured,
0  cc/min
                                             R  Predicted, cc/min.
                                             FIGURE 1.   ONE HOUR SAMPLING  RATES




2        80



 R,       50

 cc/min  40




            0   1
  Ro = 81.2 cc/min.
  Vb = 18 + 4  L/q
SDEV =3.3 cc/min.
                                                    Ro = 42 cc/min.
                                                    Vb = 324 L/q
                                                  SDEV = 1.3 cc/min.
         5   6   7   8   9  10   11  12  13  14  15  16  17  18  19  20  21  22  23  24

                           SAMPLING TIME, HR


                           TABLE 1.   RETENTION VOLUMES

Acryloni trile
Freon 113
1 ,2-Dichloro-
« u v<-a ' ; »w y
4.9 + 0.8
1.5 + 0.1
0.5 + 0.08
18 + 4
v, VL i uy ,i_/ y
0.3 - 7.0
2 - 6
0.23 - 0.47

1.   Coutant, R.  W.,  and Scott,  D.  R.,  1982.   "Applicability of  Passive
    Dosimeters for Ambient Air  Monitoring  of  Toxic  Organic Compounds",
    Environ. Sci. Techno!., 16, pp.  410-413.

2.   Brown, r.  H., and Walkin,  K.  T.,  "Performance of  a Tube-Type  Diffusive
    Sampler for Orqanic Vapors  in  Air",  Proc.  Fifth Int. SAC Conference, pp.
    205-208, May, 1981.                                              ~~


               J.R.  Stetter, S.  Zaromb, W.R.  Penrose,  and T.  Otagawa
                           Argonne  National  Laboratory
                             J. Sinclair and J. Stall
                 United States Coast Guard, Washington, DC  20593
     A  portable instrument  for detecting,  identifying,  and monitoring  chemical
 hazards  is  described  by  Stetter  et  al.  (1984a,  1984b).   The  instrument  was
 developed  at Argonne National  Laboratory  for  the purpose of alerting  U.S.  Coast
 Guard  personnel  to  the  presence of hazardous  vapors  during cleanup of  chemical
 spills  or  during  inspection of chemical shipments.  Instruments of  the same  type
 may  be  used  as   personal  monitors  for  employees  in   hazardous  waste  cleanup
 operations  and in various  industrial  environments, especially  in  the chemical,
 pharmaceutical, petroleum,  mining,  and metallurgical  industries.   They  may  also
 serve  as  inexpensive substitutes for,  or  supplements  to,  the instrumentation now
 used  to  monitor  hazardous  emissions  from  smokestacks  and   other  stationary

     The recently  completed prototype  instrument,  which  uses  the array  shown  in
 Fig.  1,  comprises four  electrochemical  sensors that  respond to toxic gases  and
 two  heated  noble-metal  filaments  that  cause  many  compounds  to  be  partially
 pyrolyzed  or oxidized in  air  (Stetter,  Zaromb, and  Findlay,   1984).   The  four
 sensors can  be rapidly switched to  one of several operating modes.   In practice,
 four modes and four  sensors yield 16 measured  parameters,  that is, 16 independent
data channels.
     The prototype  instrument  fits  into a  camera bag (Fig. 1),  weighs about  15
pounds, and  can operate for  at least  four  hours on  self-contained  rechargeable
batteries.     The   user  interface  consists  of  five  keys and  a   32-character
display.   The user  selects detection,  identification,  or  calibration modes  by
pressing the  appropriate keys.   Extensive  training  of   personnel  is  avoided  by
having the  instrument  provide menus of choices  for each operating mode  desired.

The menus are controlled by a microprocessor  that  has 12 kilobytes of memory and
extensive self-test capabilities.
     When monitoring for the presence  of  any  unknown air contaminant, the  sensor
array  is  connected  directly  to  a  sampling  probe.    A signal  from any  of the
sensors  indicates  the  presence of  a  possibly  hazardous species  near  the  probe
intake.  To  identify the detected species,  the  user  first  draws a sample through
the  probe  intake  into  a  sampling  bag.    The  collected  sample  is  then  passed
through the  sensor  array,  with  the  sensors being  switched  into four differently
selective modes  at appropriate  intervals.   The response of each  sensor at the end
of  each  interval  is recorded  in  one of  16  independent  data channels.   The
relative magnitudes of  these  response  signals provide  the  information needed  to
identify the particular  species  giving rise  to  the  observed signals.   Once the
microprocessor  identifies a compound based on the recorded data, it then sets the
sensor array  for maximum  sensitivity  to  that  compound in the  monitoring  mode.
The microprocessor  can  also  set  the alarm  level to  correspond  to an appropriate
fraction of  the  short-term exposure  limit  or immediately-dangerous-to-life-and-
health concentration of  the identified  compound.
     Of some 30  different  compounds tested  with the  array  shown in Fig. 1,  each
yielded a distinctive response pattern, as  illustrated  by the histograms of Figs.
2 and 3.  These  histograms, as well  as the  response versus concentration plots  of
Fig. 4, obtained as described in Stetter et al.  (1984a), show how  specificity and
quantitative determinations derive from the measured values.   To  demonstrate the
differences  in response  patterns, a  "fingerprint index" can be  derived from each
histogram by:
     1.  Assigning  a two-digit  index  (01 through   16)  to  each  of  the
         channels 1-16;
     2.  Listing  the  indices of  the  strongest  channels in order of de-
         creasing channel strength;  and
     3.  If the  signal  in one of the three  strongest  channels is negative,
         drawing  a  bar  over  the  corresponding  channel   index  (Stetter
         et al., 1984b).
Thus,  for  carbon tetrachloride and tetrachloroethylene,  the fingerprint  indices
become  020307 and 070602,   respectively  (cf.  Fig.   2).    Figure  4  shows the
proportionality  of  the signals in the  strongest  channels to the  concentrations  of
four different compounds.

     Table  1  lists 19  of  the tested  compounds  that  are of  concern  to the  U.S.
Environmental  Protection  Agency  (Federal  Register,  1980).   All but  one of  the
fingerprint indices listed  in  the second  column  of Table 1 can be seen to differ
from each other.   The one exception  is the  identical fingerprint  index   (070603)
for  chloroform  and  pyridine.    However,  it  is clear from  Fig.  3  that  the
histograms  of  these  two  compounds  differ  substantially  in  the ratios  of  the
normalized  responses   in  the   three  strongest  channels  --  1.00:0.51:0.18  for
chloroform  as   compared with  1.00:0.77:0.17  for  pyridine.   Moreover,  if  our
fingerprinting   procedure  is extended  to the  five (instead  of three) strongest
channels,  then the new indices  for chloroform  and pyridine become  070603 0205
and 070603 0208, respectively.   Thus,  these two  compounds  are also distinguish-
able from each  other with  the array used.
     The  last  two  columns  in Table  1  list  the ELD (estimated  lowest  detectable)
concentrations  determined as  twice  the noise  levels  (Stetter et al.,  1984b)  and
the TWA  (time-weighted  average)  threshold  limit values  of  the 19 compounds.   In
16 out of  19 cases, the ELD concentrations  fall  below the TWA values.   Only  three
nitrogen-containing  compounds   (hydrazine,  methylhydrazine,   and  nitrobenzene)
present  detection   problems  below   the  TWA  threshold   levels  with  the   sensor-
filament  array  used in  this work.   Since  this  particular  array did not  include
some of  the  most  sensitive  sensors,  such as those used  in the  parts-per-bil lion-
level   hydrazine  detector  (Rogers  et  al.,  1980),   we  expect  that  low-level
detection  should usually  be achievable with arrays tailored to the compounds  of

     The  numbers   of  differently selective  sensors,  S,  and  their  differently
selective  sensing  modes,  M,   can  be   varied   to  yield   P  =  MS   independent
parameters.  The   number P  required  to  perform  a specified  task depends on  the
number N  of  different  compounds  that  may be encountered  in a given environment
and also on the number  A of significant components that  may  be  encountered at  one
time,  in accordance with the following inequality (Zaromb and Stetter, 1984):

                               2p 1  >  V     N!
                               i  L
     Thus, P = 16 can  serve  to identify a maximum of N = 74 different  species  in
mixtures containing up to three different  air  contaminants or  a maximum of  N  =  30

different  species  in  mixtures  of  up to  four  different  contaminants.   Alter-

natively, P = 24  (obtainable  with  S = 4  and M = 6 or  S  =  6  and M = 4) may  serve

to identify up to 100 different species in mixtures of up  to four contaminants or

up to 30  species  in  mixtures of up  to eight  contaminants.  A  computer algorithm

developed  for  estimating  measurement  errors  can  be   used  to  evaluate  the

appropriateness   of   any  given  sensor array   and  selected  operating  modes for

identifying and monitoring  any  group of compounds  that  may be encountered  in a

given environment.


1.  American  Conference  of  Government  Industrial  Hygienists,  1983.   Threshold
    Limit Values for Chemical Substances  in the Work Environment Adopted by ACGIH
    for 1983-84, Cincinnati, Ohio.

2.  Federal  Register,  1980.   Rules and Regulations -- Appendix VII -- Hazardous
    Constituents,  pp33132-33133.

3.  Rogers,  Perry  M.,  et  al.,  1980.   Instrument development  of  a  toxic  level
    hypergolic vapor detector, Proceedings  of the JANNAF Safety and Environmental
    Protection  Subcommittee,   Chemical  Propulsion  Information Agency,   Laurel,
    Maryland,  ppl-22.

4.  Stetter,  Joseph  R.,  et  al.,   1984a.    Portable  device  for  detecting and
    identifying  hazardous  vapors,  Proceedings  of  the Hazardous  Material  Spills
    Conference,   J.   Ludwigson,   ed.,  Government  Institutes,  Inc.,  Rockville,
    Maryland,  pp!83-190.

5.  Stetter,  Joseph  R.,  et  al.,   1984b.     Selective  monitoring  of  hazardous
    chemicals  in  emergency  situations,  Proceedings   of   the  JANNAF  Safety and
    Environmental   Protection   Subcommittee,   Chemical   Propulsion   Information
    Agency,  Laurel,  Maryland,  in  press.

6.  Stetter,  Joseph  R.,  Solomon  Zaromb,   and  Melvin  W.  Findlay,   Jr.,   1984.
    Monitoring of electrochemical ly  inactive compounds by  amperometric toxic gas
    sensors,  Extended  Abstracts  of  1984   Pittsburgh Conference   on  Analytical
    Chemistry  and  Applied  Spectroscopy, Atlantic  City,   New Jersey,  March  5-9,

7.  Zaromb,   Solomon,  and  Joseph   R.  Stetter,   1984.     Theoretical  basis  for
    identification and  measurement  of  air  contaminants  using  selective  sensor
    arrays,,submitted for publication to Sensors and Actuators.


Simple Hydrocarbon
Ring Compounds
Carbon-Oxygen Compounds
Carbon monoxide
Chlorinated Aliphatics
Carbon tetrachloride
Tetrach 1 oroethy 1 ene
Nitrogen Compounds
Hydrogen cyanide
1 , 1-Dimethyl hydraz i ne
Nitric oxide
Nitrogen dioxide
Sulfur Compounds
Hydrogen sulfide
Sulfur dioxide





ELD Concentration!










 *See text.

 #Estimated lowest detectable concentration.

##Time-weighted average threshold exposure level (American Institute
  of Government Industrial Hygienists, 1983).


                                   Polluted Gas
        Figure 1.  Prototype  Instrument' for Detection,  Identification,
              and Monitoring of Chemical  Hazards:   A.  Photograph
                            and  B.  Basic Components.

Normalized Response
Normalized Response
Normalized Response
Normalized Response
n> -••
"T| f> lQ r
O T3 C
C. O ~1
-1 3 (D
o o> ro
—i. I/) • N) —
-h -•• w_
tt> 3 :r
o> m to ^~
3 P» ft"
<-f o o ^

T~ -«J O Cft —
o* o P> 2
N -h 3 9 vi-
tu v> £
1 r* o oo-i
Q. 3- O —

CT> 3" &• °

o o ~< =i

O 3 T _^
13 ro cu
Q- — ' — i •£_
i/> to — '• ^
. IM _^
c-l- fD f
O Q. _.
D In c

p .
5 In c






-^ _
3 C

O <" -

CD 00 -
Z jo.


M ~


o _
3 In c






A _
3 C
N3 — <
C/l — i

O o»-|
Q -vj _ i
CD 00 —
^- to — i

cr on

hO ~|
- p
D Ln


_^ o
b In c
i_ j_
O o>-
Q vl-
(D OO-
3 g-

                                             Normalized Response
Normalized Response
_i. C
ua f.
~tl _1 -H
3 ^
* • N>-

o'g *--
O ^ C/l -
o H o o>-
3 to ?
3 0 § >i-
^ 0 ^ °°~
^-t. Z  ® _»_
31 __k
I.' — i. SJ ~
^ t/,
c~^~ tx ~

CO j^_




3 fo *. O> 00 C
1 1 1 1






3 r~


O O>-
3 ^1-
 bo c
i | i i




  I  71
   D   6-
   D>  5
                       NO (later run, average
                       sensor nolse=0.0072
            Formaldehyde (average
            sensor noise=0.18 tiA)
          NO (earlier run, average
                      20      30       40
                    Concentration (ppm)
                 CCI4 (Ave.
                                    Tetrachloraethylene (Ave.
 20         40        60         80
       Concentration (ppm)
'Figure  4.   Proportionality of the Response  Signals  in the Strongest
   Channels to the Concentration in  Air of the Sampled Compounds.

              Problems and Pitfalls of Trace Ambient Organic Vapor
                 Sampling at Uncontrolled Hazardous Waste Sites

                              Michael S. Zachowski
                              Stephen A. Borgianini
                New Jersey Department of Environmental Protection
                    Hazardous Site Mitigation Administration
                                     CN 028
                           Trenton, New Jersey  08625

     Vectors of pollutant migration from uncontrolled hazardous waste sites have

been  under  investigation by  regulatory agencies for  many years.   The  current

data  base has  largely  been limited  to direct  impact  upon surface  and  ground

water via leachate  discharge  and ground water  infiltration.   Leachate sampling

and  characterization protocols  are  well  established.   Subsequently,  current

engineering practices are adequate to address  and mitigate  environmental  impact

from  leachate  streams.   Potential  ground  water  impacts  from  uncontrolled

hazardous waste sites present a more complex problem to environmental scientists

than  the  leachate  route.  Earlier ground  water investigations in  the previous

decade, though  ambitious,  fell short of  accurately defining ground  water  flow

networks.    Refinement   of  analytical  techniques   and   mathematical  modeling

approaches in the  late  1970's demanded a  more sophisticated approach to  field

techniques   and   monitoring   network   design.    Subsequent   refinement   of

environmental   measurement   system   design  and   collection   procedures   are

approaching  a   convergence   with  state-of-the-art  analytical   and  modeling


     The  airborne route  of  contaminant  migration,  though always a  concern,  has

not received comparable attention as other  more  defined  routes, i.e.  ground  and

surface waters.   Both liquids & gases,  in  their natural  physical  state, take on

the shape of their container.  In the biosphere, leachate &  surface waters  take

on  the shape   of  their  stream  channel   or  local ground  water  basin.    The

atmosphere, however,  is  relatively  dimensionless.  Water  is  of  much  greater

relative density  than air,  and is  constricted by  it's container;  minimizing

dilution effects relative to air.  In the past,  regulatory  agencies  have placed

their emphasis on point source  air emissions  due to  high concentrations emitted

and  ability  to control  said  point  sources.   Efforts  to  control  these  point

sources have  largely  been concerned with potential  health effects  to  impacted

individuals & communities.

     In New  Jersey  Department  of  Environmental Protection -  (NJDEP)  Hazardous

Site  Mitigation  Administration's   (HSMA)   efforts  to   perform  environmental

evaluations and risk  assessments  at  uncontrolled hazardous waste  sites,  it  has

become  clear  to these investigators that non-point source emissions  from  the

subject  sites  were   being  largely  ignored  by  federal  and  state  regulatory

agencies.   These investigators  believe  organic vapor and particulate emissions

from  uncontrolled  hazardous waste  sites represents a  significant  pathway  for

offsite migration of  contaminants.   In  order to provide  toxicologists  with  the

data  necessary  to  perform  a  risk assessment,  emissions  must be  qualified  and

quantified.   In order  to  develop  the  data  base,  a  literature  search  was

conducted,  revealing  inadequacies &  difficulties  in  sampling  network design  and

a wide  divergence  of  analytical methodologies.   It  became apparent  that  these

problems that  were encountered were not  unlike  those  faced by  investigators

performing water  quality  studies  in   the  early  1970's.  In many  ways,  the

problems faced by investigators attempting  to design comprehensive sampling  and

analytical  plans   to  delineate airborne  pollution  migration  from  non-point

sources are complexed by  the relative instability  of the  air  matrix  in relation

to the water matrix,  with regard to diluation effects,  variability  of direction,

and speed of transport.


     In attempts to  approximate  environmental conditions in the air  matrix the

investigator must  correlate  the  physical nature  of the  environment with  the

design  of  the  monitoring  network.    One  of  the  most   critical  physical

considerations  is  the  availability  of  adequate  meterological  data  such  as

air temperature, wind  speed,  wind direction, humidity and barometric pressure.

Though these metrological data may be readily available  at regional airports or

meterological stations,  these conditions may  be indicative  of  the  region  but

lack  site • specificity.   On  site  conditions,  such  as  landfill  elevations

considerably  above  grade,   can   significantly  disrupt  "normal"  meterological

conditions.  Localized  anthropogenic  disturbences  of natural  geomorphology  can

lead to very localized atmospheric conditions both effecting  the  site as a whole

or causing a broad heterogeniety of meterological conditions  within the site.

     The   design   of   an   environmental  sampling   network  must   take   into

consideration not  only  the regional,  but  also  localized  physical  conditions

outlined above.  In  order  to adequately design a monitoring network,  such that

mathematical  modeling  of  site  conditions  &  contaminant   transport  can  be

addressed,  accurate  meterological  data  must  be  available   for  each  sampling

station for the duration of the program.  Only by such a strategy  can localized

site conditions be defined.

     Design &  implementation of the actual  sampling network  is dependent upon

accurate definition  of  regional and localized effects catalogued  above.   These

conditions will bear significantly upon  selection of sampling  locations.   Major

problems in selection of sampling  locations  are  the  number of sampling stations

required  to adequately define  the site  situation  and  selection  of  a  true

statistical  background.    Selection  of  a  background  location  is  extremely
critical in order to determine  if  ambient  site  concentrations are significantly
different  than  background  concentrations.   Establishment  of  variability  of
background concentrations  is  required to statistically  evaluate  background vs.
site concentrations.

     The lack of a standardized data base from uncontrolled hazardous waste site
emissions  causes  difficulties in  design of  a adequate  monitoring  network with
respect  to  suspected  contaminants  &  concentrations.   Uncontrolled  hazardous
waste sites are  characterized by a broad heterogeniety of compounds disposed as
well as uneven distribution throughout the site.  These facts must be considered
in  selection  of  the analytical  scheme as well as number & location of sampling

     Classically, air  quality standards have been  directed  toward occupational
exposures.  The relationship  of these air quality standards to data derived from
ambient  air  monitoring  is  nebulous.   The paucity  of  ambient  air  quality
standards  drastically effects sampling network design since the design should be
dependent  upon compounds  &  concentrations of  contaminents  which are  known to
adversely  effect  the  environment or public health.   In  light  of  the absence of
well defined  action levels,   it becomes unclear to  the investigator the goal or
targets of his program,


     As mentioned earlier, much of the  focus of ambient  air evaluation has been
focused  upon industrial  or  occupational  exposure.   Subsequently,  much  of the
sampling equipment  available  to the investigator  is limited in it's usefulness.

Specifically, most sampling pumps are  designed  to  monitor  eight  hour exposures.

NJDEP-HSMA's investigations were more concerned with long term chronic exposure.

Based on a literature review,  no standard sampling periods were established.  In

order to calculate  daily  exposures  and to minimize man-power  considerations,  a

twenty-four  hour  sampling period  at  low  flow velocities  (10-15 ml/min.)  was

chosen.  This immediately presented a problem,  as  to  the best  of our knowledge,

no such pump was  readily  commerciable  available.   For this study, only volatile

organic constituents were considered, due to budgetary & logistical constraints.

The  sampling device that  was chosen  was  fibricated,  consisting of  a  battery

powered Gillian -  10020  pump  which draws  10-15 ml/minute  of  air  through  a

collection trap packed  with Tenax -  GC.   Laboratory calibrated  rotameters are

adjusted in  the field  to  assure that  the  proper  flow rates were achieved.  To

prevent airborne  particulates  from  entering the traps, a  glass  fiber filter is

placed before each trap.  These filters were impregnated with sodium thiosulfate

to  avoid  any oxidation of the  Tenax -  GC and  to  minimize  the formation of


     In  conclusion,  sampling  efforts  of  trace  ambient  organic  vapors  at

uncontrolled  hazardous  waste  sites  was  hindered   by  the  lack  of  standard

procedures   and   readily    available   commercial    samplers,    making   data

representativeness & comparibility among investigations poor.


     There  are a wide variety  of  analtyical methods  for specific pollutants as

well as scanning  procedures  for chemically  related  pollutants.   None of  these

analytical  methodologies is all inclusive.  Many of these  methodologies  include

their own specific  sample  collection technique.  If  an investigator  chooses to


 examine  a wide variety  of contaminents  in ambient air,  he is  faced  with the

 possiblity  of having  to  use  several different  sampling apparatus  to collect

 samples  as  the specific method requires.   It is  obvious  that  this can be quite

 cumbersome  &  require a high  degree  of labor to  maintain the  sampling network.

 In an  attempt  to maximize  the information gained & minimize the physical effort

 and  money,   involved;   investigations  should   be  narrowed  in  scope.   This

 analytical  targeting  of  compounds   could  be  selected  by toxicity,  mode  of

 transport  or  pervasiveness  in  the  environment.   Analytical  targeting  can  be

 accomplished  by  either fine  tuning  sample  collection methodologies  to cover  a

 broad  spectrum of compounds or adapting current analytical techniques to enhance

 the  detection of  specific  chemical  species.   From  a  regulatory  standpoint,

 standard  analytical  methods must  be employed  that are  legally  defensible  and

 scientifically valid.  All of the HSMA.sites carry the  potential of litigation

 therefore all  analytical methods performed must be  of demonstrated quality.

     A rigid  well  defined Quality  Assurance/Quality Control  program must  be

 established  to consistantly  demonstrate  the validity,  representativeness,  and

 comparibility  of  the  data  generated from standardized  analytical and  field

 methodologies.  At  present the  type  & frequency of Quality Control procedures

 vary   greatly.   Therefore,  Standard  Procedures   such  as  duplicate/replicate

 analysis, blank  methodologies,  and  field  practices  must be  established.   The

 problem  of  establishment  of background  for an  ambient air  study is a  case  in

 point.    The  background can change day by  day  if  not hour  by  hour,  therefore,

without measuring the variability of  your background sample, data interperation

 can be complicated.   All background data should  at a minimum be  replicated by

 collecting and analysing at  least  two background samples  from  the same station

 collected in  the  same  manner.  Utilizing  more than  one sampling  station  for

background  calculations   can  lead   to   errors  caused   by pseudoreplication.


Selection  of   background   location  and   a   knowledge  of   regional   ambient

concentration of  the compounds  of  interest are  the most  critical factors  in

obtaining useful data.   It  becomes  evident that design of  the  sampling  network

determines the usefulness of the data to be used in environmental evaluations.


     The  problems  &  pitfalls  of   trace  ambient  orgaic   vapor  sampling  at

uncontrolled hazardous waste  sites  outlined throughout this paper  provides the

investigator  with  a multitude of  complexities  in  the  evaluation  of  data

generated from  this  study.   In the process of  fine tuning &  standardizing the

environmental measurement  system the  information  gained  will  be  of  superior

quality.  This increase  in data quality  does not,  however,  lead to  an immediate

corresponding increase in  the amount of  information  gained by  the study.   The

evaluation of these  data as they relate  to public  health,  environmental impact

and environmental  fate  of trace ambient  contaminants eminating  from  non-point

sources such as uncontrolled hazardous waste sites thrusts  the  investigator into

areas of  environmental evaluation and risk assessment not  previously  addressed

at the  same  level  of sophistification as  other vectors of  pollution  migration.

The  lack  of  trace  ambient   air   standards   creates  problems  toxicological

assessents  of  potential  health  effects.   Improperly  designed  sampling  and

analytical strategies may  overlook  high,  short term  exposures,  potentially due

to specialized or localized meterological conditions that  would not  be accounted

for in mathematical modeling  for risk assessment.   One  area of promise recently

commercially available to the investigator is  real time, real world, multi point

analysis  of  ambient  air.   Advantages  of a real time, analytical system  include

large  number  of  analyses  over a  short period  of  time  which  will  yield

information  about  the variability of  the contaminent load  in  the atmosphere as


well as allowing for the correction of any apparent analytical problems can take

place almost immediately without lose of valuable  data.   Such  a system,  used in

conjunction with an ambient monitoring network  should  prove more effective than

either methodology used separately.

     The problems presented in this paper provide a substantial challenge to the

environmental  scientist,  but  are  not  insurmountable.   Accurate &  definative

characterization of trace ambient  organic vapors  emitted  from non-point  sources

is  currently  limited  to  only  an  evaluation  process.   Point  source  emissions,

once a  source  of  gross environmental pollution have been significantly  reduced

due to regulatory & technical advances.   Non-point sources, such as uncontrolled

hazardous waste sites, still pose  a  threat  to the environment  and public health

& will  require greatly increased  efforts  and state-of-the-art  technologies  to

evaluate & mitigate.


1.   Barras,  R.C.  ed.,  Instrumentation  for  Monitoring  Air  Quality,  American
     Society for Testing and Materials, STP 555,  1974.

2.   J. Bozzelli,  B.  Kebbekus,  "Collection  and Analysis of  Selected  Volatile
     Compounds in Ambient  Air",  paper no. 82-65.2,  in  Proceedings of  the  75th
     Annual Meeting,  Air Pollution Control Association,  Pittsburgh, 1982.

3.   R.  Harkov,  Toxic  Air  Pollutants  in  New  Jersey,  N.J.   Department  of
     Environmental Protection, Trenton, 49 pp.,  1983.

4.   R. Harkov, R. Katz, J.  Bozzelli,  B.  Kebbekus, "Toxic and Carcinogenic Air
     Pollutants in New Jersey - Volatile Organic  Substances",  Proceedings,  Toxic
     Air Contaminants, VIP -  I,  Air Pollution Control Association,  Pittsburgh,

5.   R. Harkov,  B.  Kebbekus,  J.  Bozzelli,  P.  Lioy,  "Measurement of  Selected
     Volatile  Organic  Compounds at  Three Locations  in  New  Jersey during  the
     Summer Season",  JAPCA 33:1177 (1983).

6.   Himmelsbach,  B.F. ed., Toxic Materials In the Atmosphere, American Society
     for Testing and  Materials, STP  786, 1981.

7.   D. Lillian, H. B.  Singh, A. Appleby, L. Lobban,  R.  Arnts,  R. Gumpert,  R.
     Hague,  J.  Tosney,  Jr.,  Kazzazis,  M.  Antell,  D.  Hansen,   B.   Scott,
     "Atmospheric  fates of  halogenated compounds",  Environ.  Sci.  Technol.  9:1042

8.   E. Pellizzari, "Analysis for organic vapor  emissions  near industrial and
     chemical waste disposal sites",  Environ.  Sci.  Technol.  16:781  (1982).

9.   E.D.  Pellizzari,  and John Bunch,  "Ambient Air Carcinogenic  Vapors  Improved
     Sampling  and  Analytical Techniques  and  Field  Studies",   United  States
     Environmental  Protection Agency,  EPA-600/2-79-081,  1979.

10.   E.D.   Pellizzari,  M.  D.   Erickson,  R.  A.   Zweidinger,  "Formulation  of  a
     Preliminary  Assessment  of  Halogenated  Organic  Compounds   in   Man   and
     Environmental  Media",  EPA-560/13-79-010,   U.S.   Environmental  Protection
     Agency, 1979.

11.   H. B.  Singh,  L.  J.  Salas,  A.  J.  Smith,  H.  Shigeishi, "Measurement of  some
     potentially hazardous organic  chemicals  in  urban  atmospheres",  Atmos.
     Environ. 15:601,  (1981).

12.   United States Environmental  Protection  Agency,  Quality Assurance  Handbook
     for Air Pollution Measurement Systems,  EPA-600/9-76-005,  1976.

13.   United  States   Environmental   Protection   Agency,   Technical   Assistance
     Document for Sampling  and Analysis of Toxic  Organic Compounds in Ambient
     Air.  EPA-600/4-83-027, June 1983.

14.   Verner,  S.  S.   ed.,  Sampling  and  Analysis  of  Toxic  Organics   In   the
     Atmosphere, American Society for Testing  and Materials, STP  721,  1979.


                J.N.  Driscoll, A.G.  Wilshire, J.W. Bodenrader
                          HNU Systems, Incorporated
                            160 Charlemont Street
                         Newton, Massachusetts 02161
     Over the past few years, the HNU PI-101 photoionizer has been one of the
primary instruments used for hazardous waste site entry program by the FIT
and TAT environmental response teams (1).  The PI-101 is used for preliminary
screening of the area including the extent of the problem.  The 101 can also
be used to determine where barrels are located and/or where problematic
barrels are stored (underground) by simply breaking the ground with the heel
or shovel and looking for an increase in hydrocarbon concentration with the
analyzer.  The PI-101 can also be used during Phase II to check soil samples
or water samples which are collected and stored for subsequent laboratory
analysis.  One gas chromatograph which can be used for analysis in the field
is our Model 301P, a gas chromatograph which will operate on batteries in the
field and provides complete GC capabilities in the laboratory with tempera-
tures to 300°C and packed/capillary capability.
     Once in the laboratory, the PI-101 can be used as a qualitative tool
(2,3) to screen  (via headspace) samples and ensure the integrity prior to GC-
MS or other analysis.  Of course a simpler, less costly, and more useful
sample screening technique involves using GC - PID/FID for analysis.
     Several years ago, a technique was described (7) that utilized the dif-
ference in response ratios for organic compounds on a photoionization detector
(PID/10.2 eV) and a flame ionization detector  (FID) as a means for hydrocarbon
     Typical ratios calculated by: PID attenuation x peak height/FT.D attenua-
tion x peak height were as follows:  Type of Hydrocarbon   PID/FID Ratio
                                     Saturated              10 or less
                                     Unsaturated            10 - 25
                                     Aromatic               >25
     The HNU Gas Chromatographs (GC 301 or GC 421) were designed to incorpor-
ate an integral PID, (non-destructive detector) with an FID or another detec-
tor in series.
     The PID response (10.2) eV was found to be a function of the electronic
structure (pi vs sigma electrons) of the solute; as a consequence, it provided

a discriminative response.  This characteristic (discriminative response)
can be turned into a very powerful qualitative tool, by comparing (ratioing)
the PID signal to that of another detector.  The preferred detector for com-
parison was one which had a homogeneous response (i.e., the FID has a homo-
geneous response in the sense that its response for organic molecules depends
mainly on the number of carbon atoms in the molecule).   In this way, general
classes of compounds could be identified, since the FID response provides the
normalizing factor.  We have been able to differentiate the presence of aroma-
tic, unsaturated, and small molecular weight chlorinated compounds in a haz-
ardous waste dump site by calculating the PID/FID ratios (5).   The compounds
present in each group can be further clarified by using retention indices.
Retention indices (RI) can be very useful to aid solute identification from
the quantitative and qualitative point of view.  Quantitatively it provides
structural information, (i.e. two adjacent homologs in a series differ by 100
units).  Several of the unknown compounds in the dump site were "identified"
by matching up its retention index.  In Table IV we have detailed an instru-
ment selection guide.
     If a hazardous waste site is particularly problematic e.g., very large
(clean up cost >$1 M), it is quite possible that Phase III will be reduced
(Table I and II), and some types of continuous monitoring equipment will be
necessary for Air Quality Monitoring.
     We have developed a very powerful and flexible series of  analyzer systems
for continuous monitoring measurement of pollutants.  The first is a dedicated
microprocessor controlled chromatograph (501) for fixed installations of spec-
ific pollutants (up to 10); the other (301) is a small transportable system
which is also microprocessor controlled but can be utilized for a variety of
     The microprocessor controlled environmental chromatograph, the HNU Model
501, which utilizes the high sensitivity Photoionization Detector (PID) or Far
UV Absorbance Detector (FUV) (6) for inorganics or hydrocarbons monitoring at
percentage or sub-ppm levels.  The Model 501 consists of four  sections, the
oven which contains the columns and PID, the fluidics bay (hardware and valves
need to inject the sample), the electronics bay where all the  electronics
(including microprocessor) for the system are located and the  multipoint
sequencer capable of sampling up to ten points (7,8).

     The Z-80 microprocessor controls all functions of the Model 501 under
direction of the operating system in ROM.  This includes temperature control,
valve control, diagnostic tests, as well as data collection and manipulation.
As a result, the Model 501 is designed for unattended operation 24 hours a
day.  Automated calibration with a known level of a hydrocarbon occurs every
eight hours as part of the program.
     The chromatographic system used consists of a precolumn, an analytical
column and a detector (PID or FUV).  The precolumn is used to remove compounds
more strongly retained from the air sample during injection.  All materials
trapped on the precolumn after injection are back flushed to vent during the
remainder of the analytical cycle.  The analytical column separates the com-
ponents for quantitation using the PID or FUV.
     The use of a properly chosen precolumn as part of the chromatographic
system allows the development of a specific analyzer.
     The Model 301 system weighs about 30 pounds and consists of two packages;
the chromatographic oven, waiving, detector, analog electronics; the micropro-
cessor controller/programmer which can be used for temperature control, valve
timing, and data reduction.  The detectors on this system can be any one or
two of the following: PID, FUV, or flame ionization (FID).  Virtually any
possible combination of contaminants can be analyzed with these detectors.  A
very flexible automatic analyzer can be constructed with the 301 components.
     In conclusion, we have shown  that HNU Systems has a variety of instruments
which can be utilized for preliminary site entry, screening, analysis and/or
continuous monitoring of hazardous waste sites.
1)  Driscoll, J.N., & Hewitt, G.F. , Instrumentation for "On Site" Survey &
    Identification of Hazardous Waste, Ind. Hyg. News (May 1982).
2)  Becker, J.H., Driscoll, J.N. & Higgins, M., A Sensitive Portable Instru-
    ment for Arson Detection. Pitt. Conf. Paper #617, Atlantic City, N.J.
    (May 1980).
3)  Driscoll, J.N., Becker, J.H.,  Click, A., & Renaud, D., Rapid Screening
    Techniques for Determination of Residual Organics in Foods, Polymers &
    Soils. Pitt. Conf. Paper #603, Atlantic City, N.J.  (March 1981).
4)  Driscoll, J.N., Ford, J., Jaramillo, L.F. & Gruber, E.T., J.Chrom., 158,
    171  (1978).

5)  Jaramillo,  L.F.,  Driscoll,  J.N.,  & Conron,  D.W.,  Identification of  Hazard-
    ous Waste Compounds,  Using  Retention Indexes & Response Ratios of PIP &
    FID in Series,  Pitts. Conf.  Paper #864,  Atlantic  City,  N.J.  (March  1981).

6)  Driscoll, J.N.  Ferioli,  P.,  & Towns, B., A  New Sensitive Universal  Detect-
    or for Gas Chromatography:   Far UV Absorbance, Research & Development
    (in press).

7)  Hewitt, G.F.  &  Driscoll, J.N., A  New Concept in Env.  Chromatography,
    Anal. Inst.,  19,  5 (1981).

8)  Driscoll, J.N., Atwood,  E.S., Hewitt, G.F., PID-Automatic GC Combination
    Detects Toxic Chemicals  at  ppb Levels, Research & Development (Feb. 1982)

         Fig. 1
             What Type of Instrumentation is Needed
                                                              PHASE I
                                                                                                             Table II
                Tasks for Hazardous Waste
                Site Evaluation
                PHASE I
                • Preliminary Field Screening
                • Site Evaluation
                • Location of Drums

                PHASE II
                • Sample  Collection
                  (soil, water, air, hazardous waste)
                • Sample  Screening
                • Analysis and Identification

                PHASE III
                • Continuous Monitoring
                • TLV Data
                • Drilling
       Table I  • "Air Quality Monitoring"
             • Initial Survey - Field
             • Portable instruments for initial survey to measure
                hydrocarbons and inorganics
                HNU PI-101 Battery operated photoionization based analyzer
                HNU GC-301 P Battery operated gas chromatograph

             PHASE II
             • Screening - Field or Laboratory
                Example: HNU PI 1O1  - Headspace
             • Identification & Analysis
                Example: GC - MS (?)
                Alternative: GC - PID/(?) other detective

             PHASE III
             • Continuous Monitoring/Air Quality Monitoring - Field
                Example: GC 301 Automatic  EC 501 /511
Hazardous Waste Instrument Selection
Skill Level of Operator
Ability to Operate in a Van
Lab. Screening (gas samples)
Liquid Samples
Capillary Capability
Auto Sampler
(301 P)
EC 501

EC 511
fOnly if 501 is not used with 511
tOnly headspace
Table III

                        FOR PASSIVE POLLUTANT SAMPLING

                        Rene Surgi and Jimmie Hodgeson
                            Department  of  Chemistry
                          Louisiana State University
                             Baton  Rouge,  LA 70803


     Silicone polymer films have previously been used  as diffusion barriers

placed over reagent substrates for the passive collection of pollutants  from

the atmosphere (Reiszner and West, 1973; Nelms, 1976).  In the present

application the collection device  is the polymer film  itself,  in which a

reagent or trapping site, has been dissolved and homogeneously dispersed.

Prior to this experimental  work, a mathematical  diffusion model was developed

to predict the response of such a device to a given pollutant dose (Rubin,

1980).  The model  predicted a much improved sensitivity over comparable

polymer diffusion barrier sensors and a response which varies as the square

root of pollutant dose.  Prior to this effort, experimental  verification of

the model  has been lacking.

     Briefly, we sought to design and evaluate a personal sampler for ozone

using a reagent, 10,10-Dimethyl-9,9'-Biacridylidene (DBA) dissolved

homogeneously in a gas-permeable, silicone-polycarbonate copolymer film  badge

(General Electric, 1978).  This system was chosen because the simplicity and

rapidity of the O-j-DBA reaction provided an ideal means for testing the

behaviour of this general kind of sensor and for model validation.  DBA  has a

reactive double bond that is expected to undergo ozonolysis at a diffusion

controlled rate (Kearns, 1969; Turro, 1970).  The rate constant of the reaction
                                                         O   1     1
between ozone and DBA was determined to be 5.2 ± 0.6  x 10  M   sec  .

The disappearance of DBA upon ozonolysis can be monitored spectrophoto-

metrically at 430 nm.  Unlike other reagents for ozone determination  (Pryor  ;and

Collard, 1978; Hodgeson and Surgi, 1984),  this reagent  is  specific  for

ozone.  Interferences from NC^ and SC^ are less than 1%.  Currently,  there  is

no  inexpensive, convenient personal monitor for ambient  concentrations of

ozone, no experimental evidence supporting Rubin's mathematical models,  nor  is

there an experimental measurement  of  the permeability  of ozone  in  silicone.


     Twenty-six  independent exposures  using ozone  concentrations from 0.036  to

1.29 ppm ranging  over times of 15  to  400 minutes were  run.   Each data point

(see figure 1)  is the average of three  independent  readings.  Various

concentrations of ozone were generated  by  photolysis of  air or  oxygen by a

mercury arc lamp  housed  in a retractable sleeve.   The  flow  rate was  maintained

at  12 1/min  (.63  M.P.H.) by bearing  flow meters.   The  ozone concentration,

corrected for  temperature  and pressure, was continuously monitored by a  Dasibi

Model 1008-AH  ozone  photometer.


     Rubin  (1980) has solved Picks laws of diffusion  to  obtain  two equations

which can be  interfaced to relate  the amount  of  reagent  (DBA)  depleted  to the

dose  (atm-min)  of ozone:
                           R(t)   =  (2SP0DTN0F2                      (1)

 R(t)  =  total  amount of gas per  unit area which has reacted up to time


                                                   FIGURE 1
                                              0.083  PPM  OZONE
<   «
2    o
111    *
o    £
N    CC
                  2 1-
                                              0.5       3
                                         (N0T)     x 10

                                                                     0.5       3
                      (NO. OF  INITIAL  SITES x  EXPOSURE TIME)     x  10

       t, moles cm
S = solubility coefficient of gas  in membrane, moles  cm    atm   .
PQ = partial pressure of gas, atm.
                                                7      1
D = diffusion coefficient of gas  in membrane, cm  min
T = time, min
NQ = initial concentration of trapping  sites, moles  cm"
(SD)^ = permeabil ity at a given thickness, b.
P(t) = the  instantaneous partial  pressure  at  a time  t,  (atm).
Since the average partial pressure of pollutant, PQ,  is  simply  the  time
averaged  instantaneous partial  pressure:
                           P0 =  1   J   P(t)dt.
                                 T   0
Thus the following expression  can  be written:
                      R(t)  =

                       [  R(t)  ]
                               2   _
                         2SDNQ       0'
     The  quantity  of  gas  reacted,  R(t),  is  directly proportional  to the sensor
response  as  measured  by  the  decrease  in  absorbance  at 430 nm.   The
proportionality  constant  includes  the molar absorptivity of DBA and the film

 thickness,  both  of  which  have been  determined experimentally.  Thus the

 response  of the  sensor  should be  proportional  to the square root of the time-

 averaged  dose.   Furthermore an experimental  determination of the

 proportionality  constant  yields a measurement of the pollutant permeability in

 the  polymer film.

     The  following  quantities have  been  determined experimentally:

     (2SD)b  = 62.3  +_ 5.8 x 10~8   (Note:  Units given by above expressions)
     N0    =  (Ai/b)  8.59  x  1(T8
     R(t) =  (AA) 8.59  x  10"8

where:  8.59 x 10"8 =    (density  of  membrane  213,  1.156 g/cm )
                       ["molar  absorptivity"]  [molecular weight    1
                       [of  DBA, 34800  cm  j  [of DBA,  386.5 9/moleJ

        AA =  initial  absorbance,  A^  -  final  absorbance, Af

Membrane thickness  (b) =  36.5  +_1.3  x  10"4cm

Substituting these  values into the above equation  yields:


                      (AA)2   =   J  P(t)dt                          (3)

                    1990  Ai     0

     By the use of  this  expression,  the  integrated  dose may  be  determined

directly from the measured  response.   A  unique  advantage of  this  device is

that the proportionality  constant given  here  remains  constant as  long  as the

film thickness is controlled.  The uncertainty  limits  given  above for  membrane

thickness reflect the  degree of control  which has  been  attained for  film



     The results of one typical  run are  shown  in Figure 1.   In this  figure  the

y-axis is proportional to the square root of the dose.  All  the  runs  performed

showed the same linear relationship.  Thus the unique dose-response

relationship predicted by the Rubin model has  been  verified.  For  each  run  the

doses may be calculated from equation (3) and  compared to  the actual  doses  as

determined from the measurement  of Og concentration  and time.  The  results  of

one such typical comparison are  shown below:



Dose: j P(t)dt

(atm x min) x 10

Actual Dose
(atm x min) x 10

% Error
Average of % Error =  10%

Average Deviation = 5%

     Since reagent  is depleted  from the  silicone  portion  of the copolymer 300

times  faster than the polycarbonate portion  (Hwang  et al,  1974a).  it is also

possible  to calculate the  intrinsic permeability  of ozone in siloxane.   A plot

of  (2SD),    against b,  extrapolated to  zero  thickness, yields a y-intercept

which  is  taken  as the  intrinsic permeability.   A  comparison of the

experimentally  derived  intrinsic permeability  to  that predicted by modeling

the permeability of ozone  in siloxane (General  Electric,  1978; Lange, 1952) is

given  below.


        A)   Experimental  - Extrapolation to Zero Thickness:
             (SD)b=u = 153 ± 20 x ID'9 	(cm3 gas) (cm thick)	
                                       (sec) (cm  polymer) (cm Hg, AP)
        B)   Model:   Boiling Point of Gas, Molecular Diameter of Ozone
                    S = 1.18 x 10"2  	3	Ncm  9as/	
                                     (cm  polymer) (cm Hg,

                               -6    (cm*
                    D = 13 x 10
            SD = 150 ± 25 x 10'9         (cm  gas) (cm thick)
                                   (sec) (cm  polymer) (cm Hg, AP)

     Although the use of this approach to determine the intrinsic permeabl ity

of ozone in siloxane is novel, such an approach is not entirely without

precedent.   Hwang et al . (1971, 19746) has investigated permeabilities of

carbon dioxide, oxygen and water through acetyl cellulose acetate membranes

and discovered the following relationship.  A plot of (b)   against  (SD)   can

be extrapolated to a y-intercept which  is equal to the reciprocal of the

intrinsic permeability.


  1. General Electric Permselective Membranes. 1978 Membrane  Products
     Operation - Medical Systems Business Operations; Schenectady, N.Y.

  2. J.A. Hodgeson,  and M.R. Surgi, 1984. The Air Qua! ity Criteria Document
     for Ozone and Other Photochemical  Oxidants; Chapter 4.4.4.  1984  (to  be
     published.  U.S. Environmental Protection Agency, Research  Triangle  Park,

  3. S.T. Hwang, C.K.  Choi  and K.  Kammermeyer, 1974a.  Gaseous Transfer
     Coefficients  in Membranes, Separation Science, 9, pp 461-478.

 4.  S.T. Hwang, and K.  Kammermeyer, 1974b.  Effects of Thickness on
    Permeability,  Polymer Science and Technology,  6, pp 197-205.

 5.  S.T. Hwang, T.E. Tang, and K. Kammermeyer, 1971.  Transport of Dissolved
    Oxygen through Silicone Rubber Membrane, Journal of Macromolecular
    Science and Physics, B5, pp 1-10.

 6.  D.R. Kearns, 1969.   Selection Rules for Singlet-Oxygen Reactions.
    Concerted Addition  Reacitons, Journal  of The American Chemical  Society,
    91, pp 6554-6563.

 7.  N.A. Lange, 1952.  Handbook of Chemistry, 8th  edition, Handbook
    Publishers, Inc., Sandusky, Ohio, pp 264-265.

 8.  P.O. Mollere,  K.N.  Houk, D.S. Bomse, and T.H.  Morton, 1976.
    Photoelectron  Spectra of Sterically Congested  Alkenes and Dienes.
    Journal  of the American Chemical  Society, 98,  pp 4732-4736.

 9.  L.H. Helms, 1976.  The Development of  a Personal Dosimeter for Vinyl
    Chloride Utilizing  the Permeation Approach.  Ph.D. Dissertation,
    Louisiana State University, 93pp.

10.  W.A. Pryor, and R.S. Collard, 1981.  Measurement of Ozone in the Presence
    of Sulfur Dioxide and Nitrogen Oxides.  Journal  of Environmental Science
    and Health, A16, pp 73-86.

11.  K.D. Reiszner  and P.W. West, 1973.  Collection and Determination of
    Sulfur Dioxide Incorporating Permeation and West-Gaeke Procedure,
    Environmental  Science and Technology,  7, pp 526-532.

12.  R.J. Rubin, 1980.  Analysis of Mathematical Models of Integrating
    Monitoring Devices. NBSIR 80-1975, National Bureau of Standards,
    Washington, D.C., 31pp.

13.  N.J. Turro, 1978.  Modern Molelcular Photochemistry.  Benjamin-Cummlngs
    Publishing Co., Menlo Park, CA.,  p 246.

                           FOR MEASURING AMBIENT NMOC
                     Frank F.  McElroy and Vinson L.  Thompson
                   Environmental  Monitoring Systems Laboratory
                       U.S.  Environmental Protection Agency
                   Research Triangle Park,  North Carolina 27711

     A  variety  of  photochemical   dispersion  models  have  been  developed  to
describe  the  quantitative  relationships   between  ambient  concentrations  of
precursor   organic   compounds   and  subsequent   downwind   concentrations   of
ozone.1   An   important   application  of  such  models  is   to   determine   the
degree  of  control  of  such  organic compounds that  is  necessary  in a  par-
ticular  area   to   achieve   compliance   with  applicable  ambient  air  quality
standards  for  ozone.1'2    For  this  purpose,  the  models  require  input  of
data on ambient concentrations of nonmethane organic compounds (NMOC).
     The   more  elaborate   theoretical   models   generally   require   detailed
organic  species data.2   Such  species   data  must   be  obtained  by analysis  of
air   samples   with  a   sophisticated,   tnulticomponent  gas   chromatographic
(GC)  analysis  system.2'3   Simpler  empirical  models  such  as  the  Empirical
Kinetic  Modeling   Approach   (EKMA)1   require  only  total  NMOC  concentration
data,  specifically  the  average  total  NMOC  concentrations  from  6 a.m.   to
9 a.m. daily.2
     Commercial,  continuous   NMOC   analyzers  have  been   used to  obtain  urban
NMOC  concentrations,2   but  these  methods  have proved  to  be only  marginally
adequate4  because  of  limitations  from  variability,  zero  and  span  drift,
lack   of   sensitivity,   non-uniform  response  characteristics,   and   the   in-
direct  nature  of   the   measurement.    Moreover,   these  methods  are  clearly
inadequate   for  determining  the   low,   upwind   NMOC   concentrations  needed
when  transport of  precursors into  an  area  is to be  considered in  the  EKMA
     NMOC  GC  species  measurements  can be  used  by summing  the  various  com-
ponents  to   obtain  a   total  NMOC  concentration.2   These  measurements   are
much  more accurate  than  continuous  NMOC  analyzer  data,  but species  data  are
not  needed  for EKMA,  and  the procedure is  therefore unnecessarily expensive and

     The   cryogenic  Preconcentration-Direct  Flame  lonization   (PDFID)  method
can  be  used to  obtain the  requisite  upwind, as  well  as  urban,  NMOC measure-
ments.6'7'8   This method  is  based  on  a simplification  of the  GC speciation
technique.   It  combines the  cryogenic concentration technique  used in  the GC
method  for  high  sensitivity with the simple flame  ionization detector (FID) for
total NMOC  measurements without  the complex  GC  columns necessary  for  species
separation.   And  because of  the use  of  helium carrier  gas,  the  FID  has less
response variation to various organic compounds than a conventional NMOC analyz-
er with air carrier or direct sample injection into the FID.
     This method  can  be used either for direct, in situ ambient measurements or
for  analysis of  integrated  samples contained in metal canisters.  Making direct
measurements at the monitoring site avoids the potential sample loss or contami-
nation  problems possible with the use of  canisters.  However,  the analyst must
be  present  during the  6  to 9 a.m. period,  and  repeated measurements (approxi-
mately  six  per  hour)  must  be taken  to  obtain the  6 to 9 a.m.  average NMOC con-
centration.   A separate  analytical  system and analyst  is  needed for each moni-
toring  site.    (Further development  of  the  method may   allow  for  automatic
operation for on-line semi-continuous analysis in the future.)
     The use of sample canisters allows the collection of integrated air samples
over the 6 to 9 a.m.  period by automated samplers at unattended monitoring sites.
One centralized system can  then analyze the samples from several  sites.   Degrada-
tion or contamination of the  air  samples  by the canister  could  be  a potential
problem,  but  tests indicate  that  the  use of  properly  fabricated,  treated,  and
cleaned stainless  steel  canisters,  as  described in the  procedure,  is practical
and adds relatively little  additional variability to the method.8

     An air sample is taken either directly from the ambient air  at the monitor-
ing  site,  where  the  analytical  system  is  located, or  from a  sample  canister
filled  previously  at a  remote  monitoring site.   A fixed-volume  portion  of  the
sample is drawn at a  low flow rate through a  glass  beaded trap  cooled to  approxi-
mately  -186° C.   This  temperature is  such  that all organic  compounds  in  the
sample other than  methane  are collected (either via condensation or adsorption)
in the trap, while methane,  nitrogen, oxygen, etc.,  pass through.   The system is
dynamically calibrated  so  that  the  volume of  sample  passing  through the  trap

does  not  have to  be  quantitatively  measured,  but must be  precisely repeatable
between the calibration and analytical  phases.
     After  the  fixed volume  air sample  has  been drawn  through  the  trap,  the
helium carrier gas  is  diverted  to pass through the trap in a direction opposite
to  that  of the  sample  flow and  into  a flame ionization  detector  (FID).   When
the residual  air  and  methane  have been cleared from  the  trap and the FID base-
line becomes  steady,  the  cryogen is  removed and  the  temperature  of the trap is
raised to  approximately  90°  C.   The  organic  compounds  collected  in  the  trap
revolatilize and are carried into the FID, resulting in a  response peak or peaks
from the FID.  The area of the peak or peaks  is  integrated, and  the integrated
value is  translated to concentration  units via  a  previously obtained calibration
curve relating integrated areas  with  known concentrations  of propane.
     The  cryogenic trap  simultaneously  concentrates  the nonmethane  organic com-
pounds while  separating and  removing  the methane  from air  samples.   Thus  the
technique is  direct  reading for NMOC and, because of the  concentration step, is
more  sensitive  than  conventional NMOC  analyzers.   Quantitative trapping  has
been  shown  for  most  compounds  tested.6   A complete description  of  the method,
including the collection of integrated  air samples in stainless steel  canisters,
is provided  in Reference  8.   Figure  1 is  a  schematic diagram of  the analytical

     The  overall  precision estimate for  the  method,  including  the  effect of
collecting  and  storing  the ambient  samples in stainless  steel  canisters,  has
been found to be 4.5%.9
     Because  of  the  number and  variety  of organic  compounds included  in  the
NMOC  measurement,  determination of  absolute  accuracy  is  not  practical.   Based
on  comparison with manual  GC  speciation  analysis—regarded as the  best avail-
able for measurement  of organic compounds—the proportional  bias  was determined
to be +5.7%, with a negligible fixed  bias.9  Although the  5.7% bias  was statisti-
cally significant, no correction  factor is proposed for the method  because this
bias is modest,  and the speciation techique is  not an absolute standard.
     Experimental  tests  indicate some  degree  of  FID baseline  shift  from water
vapor in ambient  samples,  which could result  in  positive  bias, variability, or
both.   These  problems can  be  adequately  minimized by careful  selection  of  the
integration termination point  and appropriate baseline corrections.9

     1.    Uses,  Limitations  and  Technical Basis  of Procedures  for  Quantifying
          Relationships   Between  Photochemical  Oxidants  and  Precursors.   EPA-
          450/2-77-021a,  U.S.  Environmental  Protection  Agency,  Research Triangle
          Park,  NC,  November 1977.

     2.    Guidance  for Collection of Ambient  Non-Methane Organic Compound (NMOC)
          Data for  Use  in  1982  Ozone SIP Development,  and Network  Design  and
          Siting Criteria for the NMOC and NO  Monitors.   EPA-450/4-80-011,  U.S.
          Environmental  Protection  Agency,  Research Triangle Park,  North  Caro-
          lina,  June 1980.

     3.    Guidance   for  the Collection  and  Use of  Ambient Hydrocarbon  Species
          Data  in  Development of Ozone  Control  Strategies.   EPA-450/4-80-008,
          U.S.   Environmental  Protection  Agency,  Research  Triangle Park,  North
          Carolina,  April  1980.

     4.    Richter,  Harold G.  Analysis of Organic Compound Data Gathered During
          1980  in  Northeast  Corridor Cities.   EPA-450/4-83-017,  Environmental
          Protection Agency,  Research Triangle  Park,  NC,  August  1983.

     5.    Sexton, F.W.,  F.F.  McElroy,  R.A. Michie, Jr.,  and V.L.  Thompson.  A  Com-
          parative   Evaluation  of Seven  Automatic  Ambient Non-Methane  Organic
          Compound   Analyzers.    EPA-600/S4-82-046,   Environmental   Monitoring
          Systems  Laboratory,  U.S.   Environmental  Protection  Agency,  Research
          Triangle  Park,  NC,  August  1982.

     6.    Jayanty,   R.K.M.,   A.   Blackard,  F.F.  McElroy,   and   W.A.   McClenny.
          Laboratory  Evaluation  of  Nonmethane  Organic   Carbon  Determination
          in   Ambient  Air  by  Cryogenic  Preconcentration  and  Flame   lonization
          Detection.   EPA-600/54-82-019,  July 1982.

     7.    Cox, R.D.,  M.A.  McDevitt,  K.W.  Lee, andG.K. Tannahill.   1982.  Deter-
          mination of  Low  Levels  of  Total  Nonmethane  Hydrocarbon  Content  in Ambi-
          ent  Air.   Environ.  Sci. Technol. 16,  57-61.

     8.    Determination of  Atmospheric  Nonmethane Organic Compounds  (NMOC) by
          Cryogenic   Preconcentration  and  Direct  Flame  lonization   Detection.
          Method  description  available  from  the  Methods Standardization Branch
          (MD-77), Quality Assurance Division,  Environmental Monitoring  Systems
          Laboratory,  U.S.  Environmental  Protection Agency,  Research Triangle
          Park,  NC 27711,  September  1983.

     9.    McElroy,  F.F.,  V.L.  Thompson,  D.  Holland,  W.A. Lonneman,  and R.L.
          Seila.   Cryogenic Preconcentration-Direct  FID  Method  for Measurement
          of  Ambient NMOC: Refinement and Comparison with GC  Speciation.  Sub-
          mitted for publication, February, 1984.

 VALVE      \,S  VALVE
            1.7 LITER
                                                               SAMPLE TRAP
                                                               (LIQUID ARGON)


          Figure 1. Schematic of analysis system showing two sampling modes.


           Bruce A. Thomson, John E. Fulford and William R. Davidson,
  SCIEX®, 55 Glen Cameron Road, Unit 202, Thornhill, Ontario, Canada, L3T 1P2

     In this paper we report on a series of experiments undertaken to evaluate a
mobile Tandan Quadrupole  Spectrometer System  (the TAGA®  6000)  in performing
qualitative and quantitative real-time air analysis.  The work was a cooperative
venture  between SCIEX®  and the  Environmental  Monitoring  Systems  Laboratory,
Environmental Protection Agency at  Research Triange Park.   A mobile TAGA® 6000
located at RTP was  used  to analyze  both synthetic  gas mixtures  and ambient air
in an  industrial environment.  The  results of the  tests  were intended to indi-
cate to  both SCIEX® and  EPA what  the  current  strengths and weaknesses  of the
system are, to  show what  areas require further  development  in order to streng-
then the capabilities, and to suggest hew such a system can best be used as part
of an overall approach to air monitoring.

     A series of controlled experiments was undertaken  in  order to evaluate a)
the ability to identify unknown organics at ppm and ppb levels in a mixture;  b)
the ability to  rapidly measure the concentrations  in  a  mixture with sufficient
accuracy for a  field program;  c)  the possible presence  of  matrix effects;   d)
ease  of  operation of the  system and  data manipulation  and interpretation
facilities and e) the reliability of the system in a mobile mode.
     During the two week  program the TAGA® 6000 was used to analyze seven pre-
pared  gas  cylinder mixtures.  Two  of  the cylinders were mixtures of compounds
selected from a target list of 32,  and were prepared and  certified by an outside
supplier under the  direction of  EPA.   The  actual  components and concentrations
in each were unknown to SCIEX®.  Two of the cylinders were uncertified mixtures
also supplied to EPA, but with the  components completely  unknown to SCIEX® (i.e.
not necessarily selected from the target  list).   The other three cylinders were
two component mixtures with differing  relative  concentrations  designed to show
whether matrix or interference effects were present.

     The challenge mixtures were  analyzed using both an APCI  (Atmospheric Pres-
sure Chemical  lonization)  and a more conventional CI  (with a discharge  ioniza-
tion process)  source.  The APCI  source  is  very sensitive  to polar compounds;
the  CI  source,  using  charge transfer  from  N2+/  C>2+   and NO+,  is  sensi-
tive  towards  the  chlorinated  hydrocarbons  and  aromatics.   Both  sources  are
designed to allow air to be sampled directly into the source  (with no pre-separ-
ation or concentration) so that analysis  is performed continuously in real-time.
     The cylinder mixtures were admitted either directly into the sources or by
diluting with  clean bottled air.   Compounds  were identified by first performing
scans with  a single mass spectrometer  (Ql)  to reveal the  parent  ions  from the
source.  Each  parent  ion  noted in the  spectrum was  then collisionally dissoci-
ated to produce a daughter  ion spectrum and  then compared  (using computer-match
procedures) with  standard CAD library  spectra.  Where  library spectra existed,
identification was thus accomplished  in a few  seconds.  Where no library spec-
trum existed,  CAD spectra were manually interpreted to  identify the compound.
Quantitation was  performed using  a headspace  injection technique  to produce a
five point calibration curve.   This technique  requires  that the vapor pressure
of the compound be known, and takes about 5 minutes per compound to perform.

     The two certified cylinders  were seven  component mixtures;  the components
in each were identical, but the levels  were  approximately twenty times lower in
one cylinder than the other.   The components  in each were correctly identified
by MS/MS, and  the concentrations  measured in a separate  experiment.   The other
two multicomponent mixtures consisted of 16 and 7 components respectively.   In
total, among the  four cylinders,   13  compounds  were  correctly and unambiguously
identified, 7 were identified as either one or both of a pair,  1 was incorrectly
identified and 2 were missed.   Table 1  summarizes the results  of the qualitative
     The compounds  which  were identified as   either/or  could not  be  resolved
because each pair of compounds forms  the  same parent  ion in the  source  (for
example, methylene  chloride  forms (M-H)+ and  chloroform forms  (M-C1)+,  both
at m/z 83).  Appearance potentials of these  ions are such that it is difficult


Conpounds Present in Mixture                  Identification by MS/MS
Carbon Tetrachloride
Benzyl Chloride
Vinyl Chloride
Methylene Chloride
Vinylidene Chloride

Methyl Chloroform

Ethylene Oxide
Dichloropropene (cis-1,3 and trans 1,3)
                 Carbon Tetrachloride
                 Benzyl Chloride
                 Vinyl Chloride or Dichloroethene
                 Dichloroethane or Vinyl Chloride
                 Methylene Chloride or Chloroform
                 Chloroform or Methylene Chloride
                 Vinylidene Chloride or
                 Methylene Chloride
                 Methyl Chloroform or
                 Vinylidene Chloride
                 Ethylene Oxide or Acetaldehyde
                 Allyl Chloride
                 Not Identified
                 Not Identified

              	Cylinder 9558	   	Cylinder 11745
Compound      SCIEX»(ppn)

Vinyl Choride     1.05

Benzene           1.68
Toluene           1.03
Chlorcbenzene     1.07
Carbon Tetra-
chloride          0.63
ethylene          0.78
ethylene          0.56
Certified (ppm)






  Detected but
not quantitated



Certified (ppm)






to generate characteristic molecular ions under any ionization condition.  Since
identical parent ions are formed  in  the  source,  CAD spectra are also identical.
Some preseparation will likely be required in order to resolve these ambiguities
in identification.  Evidence existed in  the  parent ion spectra for the presence
of propane and trichloro-trifluoroethane, but this was only observed after their
presence  was known.    Dichloropropene was  mis-identified  initially  as  allyl
chloride;   later experiments revealed that  these  two compounds also  form the
same ions in the source, and so cannot be resolved in real-time.
     The results of the quantitative experiments are summarized in Table 2.  The
average  deviation  from the manufacturers  certified values was  17%  in the ppm
cylinder and  36% in the ppb cylinder.   Both of these are  based  on the primary
calibration technique  using introduction of headspace vapor.   The ppb cylinder
was also calibrated by using the  other cylinders as a standard;   this  procedure
gave better  agreement,  yielding  an  average  deviation from the  manufacturer of
only 13%.  The matrix experiments, in  which  benzene and toluene were present in
mixtures  at  ratios  of approximately  1:10,  1:1  and  10:1,. revealed  that each
component  could  be quantitatively  measured  in  the  presence  of  the  other.
However with the APCI  source,  reagent ion depletion  resulted in matrix effects
above a  concentration of  about  500 ppb.   No matrix effects were  noted with the
CI source at concentrations of up to 10 ppm.

     The mobile MS/MS system proved capable of identifying unknown components in
a mixture at  the ppm and  ppb level in real-time,  using  computer library search
procedures.   Some  problems were encountered in  resolving  pairs  of  compounds
which form the  same ion in the CI source;  a  fast preseparation technique will
likely be required  where  unambiguous identification is necessary.  Quantitation
could  be performed in-situ, with sufficient  accuracy  for  a  real-time  field
monitoring program.   More effort needs to  be devoted  to  increasing the CAD
library  size, and to characterizing  single MS spectra of environmentally inter-
esting compounds so that their possible presence can be recognized in a mixture.

     The cooperation and  assistance  of the Advanced Analysis Techniques Branch,
Environmental Monitoring Division of EPA, Research Triangle Park, North Carolina
is gratefully acknowleged.

                    T. Vo-Dinh, P. D. Enlow, T. L. Ferrell ,   -,
            T. A. Callcott, E. T. Arakawa, and J. P.  Goudonnet
                     Health and Safety Research Division
                        Oak Ridge National Laboratory
                             Oak Ridge, TN 37831
     Raman spectroscopy has proved its usefulness as a practical tool for
                 M ?\
organic analysis.    '  One limitation of this spectroscopic technique,
however, is its low sensitivity due to the small Raman cross section, which
often requires the use of powerful and costly laser sources for excitation.
Recently a renewed interest has developed in Raman spectroscopy as a result
of observations indicating enhancement in the Raman scattering efficiency
by factors of 10  to 10  when a compound of interest is adsorbed onto special
metal surfaces.  '  These spectacular enhancement factors for the weak
conventional Raman scattering process help overcome the normally low
sensitivity of Raman spectroscopy.  This new Raman technique, known as
surface-enhanced Raman spectroscopy (SERS), could open new horizons for
trace organic analysis.
     This paper presents the analytical usefulness of a novel technique based
on SERS for monitoring toxic organic pollutants.  A new method for preparing
SERS-active substrate using submicron silver-coated spheres deposited on filter
paper substrates is described in detail.  The analytical advantages and
limitations of the technique are discussed.   Figure 1 shows a typical
SERS signal of 3.6 ng benzoic acid on a SERS active cellulosic surface having
    o                          o
910 A spheres coated with 2000 A film of silver.   The detection limits for
several organic compounds such as carbazole, 1-aminopyrene, 1-nitropyrene, and
benzoic acid are at the nanogram and subnanogram levels.  '  The results
of this study indicate that SERS shows great promise as a useful analytical
tool for monitoring various important air pollutants, such as the nitro-
polyaromatic species, that cannot be detected by other spectroscopic techniques.
      Research sponsored jointly by the Department of the Army under Inter-
agency Agreement Numbers DOE 40-1294-82 and ARMY 2211-1450, and the Office of
Health and Environmental Research, U.S. Department of Energy, under contract
DE-AC05-840R21400 with Martin Marietta Energy Systems, Incorporated.
      Present address:  University de Dijon, Faculte" des Sciences, Dijon,


   I  7783
   -  7302

          950      970      990      1010      1030
                               WAVENUMBER (cm'1)
               Figure 1.  Surface-Enhanced Raman Signal  of 3.6 ng of Benzoic Acid.
                       (Laser excitation wavelength = 632 nm)

1.  Lord, R. C., 1977, Applied Spectroscopy. 31, p. 187.

2.  Harvey, A. B., (Editor), 1981, Chemical Applications of Non-Linear  Raman
    Spectroscopy, Academic Press, New York.

3.  Chang, R. K. and T. E. Furtak, (Editors), 1982, Surface-Enhanced  Raman
    Scattering, Plenum Press, New York, New York.

4.  Vo-Dinh, T., M.Y.K. Hiromoto, G. M. Begun, and R. L. Moody,  1984, Surface-
    Enhanced Raman Sepctroscopy For Trace Organic Analysis.  Analytical
    Chemistry, in press.


                             Stanley L. Kopczynski

                     U. S. Environmental Protection Agency
                      Research Triangle Park, N. C.  27711

    Simple screening techniques for polycyclic aromatic hydrocarbons  (PAHs) can
facilitate the characterization of  ambient  air  quality  by providing a  quick rou-
tine means of  identifying those particulate samples which should be subjected
to a rigorous  detailed  compositional analysis.  Lengthy  and laborious solvent
extraction and fractionation procedures  commonly  employed in analyses for PAHs
may be circumvented by thermal desorption of PAH directly from particulate sam-
ples.  Studies reported by other investigators indicate that PAHs  can be effec-
tively extracted  from particulate  matter  by  sublimation at  both atmospheric
pressure and  at  reduced  pressure.^"')   Vacuum-sublimation  has  been  used to
reduce extraction time,  improve extraction yields,  and  avoid thermal degrada-
tion. (1»*,5)  in this  study  a vacuum-sublimation system was  designed and con-
structed for direct analysis of volatilized PAHs with  a capillary  gas chromato-
graph.  Early results obtained with test mixtures are  reported.
    Analyses were performed  with  a Varian Model  3700 gas  chromatograph (GC)
equipped with a Durabond,  DB-5,  fused silica capillary column  (30m x 0.32mm x
0.1 u film thickness)  and  an HNU photoionization  detector  (PID) (Model 52-02,
9.5 eV lamp).  The sample injector  of  the  GC  was  removed and the  GC column was
extended from the oven through the injector heating block to a connecting union
on an external  sample oven  (Figure 1).  The  sample  trap  (borosilicate glass
tubing) was  connected to the  carrier gas  line  and the  end of  the  capillary
column by means  of  stainless  steel tube fittings  using  graphite  or  40% gra-
phite/vespel ferrules.   The  exit  end  of   the  column emerged  from the  oven
through an auxiliary  injector heating  block and was inserted directly into the


ionization chamber of the detector (Figure 2).  The vacuum-sublimation system
is shown in  Figure  3.   The particulate  sample  is  loaded into a borosilicate
glass tube (6.4 mm O.D. x  11  cm),  which is connected to the sample trap  (3.2
mm O.D. x  11 cm) by means  of a tube  fitting using 40% graphite/vespel  fer-
rules.  The  tube  fitting is  bored  through so that  the arm of the trap can  be
extended into the sample tube.
    Sample tubes and sample traps were cleaned before  use with a 20 min helium
purge at 260°C.   PAHs  were  vacuum-sublimed  for a  period  of 30  min  at   260-
280°C and  0.04  to 0.05  Torr.  The sample trap  containing  the  extracted ma-
terial was transferred to the external oven of the GC and purged with carrier
gas (helium) at room temperature before analysis.  The extracted material was
then volatilized  at  260°C  and  concentrated  on  the cooled  section  of  the
external GC  column  (28-30°C).   After  10 min at  260°C the  oven  cover  was
removed.  After 2 min more,  the external  column was  rapidly heated to 260°C
and the temperature programmed analysis was begun at a carrier flow rate  of  6
ml/min.  The  column  was held  initially  at 55°C  for 10 min, then raised  to
225°C at 10°C/min, and  held at  the  final temperature for  30 min.  Measure-
ments were made  with a photoionization  detector to minimize chromatographic
interferences from non-aromatic species co-desorbed with the PAHs.
    Standard reference material  (SRM)  1647 from the  National Bureau of Stan-
dards (NBS) was used to test the transfer of PAHs from the sample trap to the
GC column.    A 30 pi  sample  which had been diluted  10:1 was  injected   onto
the glass wool plug of the trap and evaporated to dryness at room temperature
with a stream of  helium.  All of  the  PAHs, ranging in volatility from acenaph-
thylene to  benzo(g,h,i)perylene,  were  successfully  transferred  to   the  GC
column and eluted within 35 min.
    A previously  desorbed  urban  dust  sample  (SRM  1649 from  the NBS)  was
spiked with  a  5-component  PAH solution,  dried,  and then extracted by thermal
purging and by vacuum-sublimation.  In both cases 4  of the components, phenan-
threne, fluoranthene,  benz(a)anthracene,  and   benzo(a)pyrene  were  desorbed
with approximately 90%  or  greater efficiency.   However, the  least volatile
component,  benzo(g,h,i) perylene was not  desorbed sufficiently to be detected
(Figure 4).

    Vacuum sublimations conducted with  virgin  SRM  1649  produced  chromatograms
dominated by a strong envelope  of material  peaking at an  elution time  of  28.0
min and  containing a strong  peak  at 22.9  min  (Figure  5a).   The presence  of
most PAH  compounds  was  rather  obscure  although  the  test  sample  contained
25-70 ng of  several PAHs.   Weak peaks were found  at elution times  consistent
with those for phenanthrene and benz(a)anthracene.  Fluoranthene plus pyrene
and benzo(a)pyrene  were  overlapped by the  strong  peaks at 22.9  and 28.0  min,
respectively.  A  repeat   sublimation test  with  the spent  SRM  1649  sample
indicated that the major  portion of  desorbable  compounds detectable by  the PID
can be  extracted in a  reasonable  time  (Figure  5b).   However,   the  PAHs  of
interest constitute only a minor portion  of these  compounds.   Similar  results
were obtained with a 60 min  thermal  purge  of  SRM 1649 at  280°C and  100  ml/
min, although the extraction was somewhat less efficient.
    Although PAHs deposited  from solution  on  borosilicate  glass sample  hol-
ders may be readily  volatilized for analysis by  capillary column gas  chroma-
tography, urban  dust samples  are  much more  resistant  to volatilization  of
adsorbed PAHs.  At low PAH concentrations and in the absence of a sufficient-
ly selective detector or a clean-up step  the presence of  desorbed PAHs  is  ob-
scured by  other  co-desorbed  species.   A multidimensional  gas chromatograph
with a  mass  selective detector would  offer  improved  selectivity  for  PAHs
thermally desorbed from urban particulate matter.   A more selective PID  (8.3
eV lamp) may also be helpful.


1.  Ball, W.L., G.E. Moore,  J.L.  Monkman,  and Morris Katz, 1962.  An Evalua-
    tion of Micro-Vacuum  Sublimation Separation  of  Atmospheric Polycyclics.
    American Industrial Hygiene Association Journal,  23, pp 222-227.

2.  Burchfield, H.P.,  Ernest  E.  Green,  Ralph  J.  Wheeler,  and  Stanley  M.
    Billedeau, 1974.  Recent Advances in the Gas and Liquid Chromatography of
    Fluorescent Compounds l.A Direct Gas-Phase Isolation and Injection System
    for the Analysis of Polynuclear Arences in Air Particulates by Gas-Liquid
    Chromatography.  Journal of Chromatography, 99, pp 697-708.

3.  Monkman, J.L., L. Dubois, and C.J. Baker, 1970.  The Rapid Measurement of
    Polycyclic Hydrocarbons  in Air   by  Microsublimation.   Pure and Applied
    Chemistry, 24, pp 731-738.

4.  Schultz, Michael J., Robert M. Orheim, and Harley H. Bovee,  1973.  Simpli-
    fied Method for the  Determination  of   Benzo(a)Pyrene  in  Ambient  Air.
    American Industrial Hygiene Association Journal,  34, pp 404-408.

5.  Stenberg,  Ulf   R. and  Thomas  E.  Alsberg,  1981.   Vacuum Sublimination and
    Solvent Extraction of  Polycyclic Aromatic Compounds Absorbed  on Carbon-
    aceous Materials.  Analytical Chemistry, 53, pp 2067-2072.

6.  Thomas, Jerome F., Eldon N. Sanborn, Mitsugi Mukai, and Bernard D. Tebbens,
    1958.  A  Fractional   Sublimation  Technique   for  Separating  Atmospheric
    Pollutants. Analytical Chemistry, 30,  pp 1954-1958.

7.  Weschler,   Charles  J.,  1983.   Indoor-Outdoor Relationships  for Selected
    Organic Constituents   of  Aerosol  Particles  Collected at  Wichita,  Kansas
    and Lubbock,  Texas.   187th National Meeting.   American  Chemical Society,
    Seattle, Washington.


, r. 0 ,-, ^ ",
II 1
: : iiy
i ! i in
_-V - "l %' U
                                  ^XNICHROME WIRE
                  SAMPLE OVEN
                                       COLD N2 GAS LINE
                                       FOR COLUMN
                                               HEATING TAPE
                                            & WRAPPED OVER
                                               V*" O.D. COPPER
                                            ] | TUBING SHEATH
                                            C = ?
                                            : '  x^/ I IU.AA I MMVJ
                                            3=/   BLOCK (250 °C)
                         Figure 1.  Sample Injection System

     VENT TO
           DETECTOR (230 °C)
                               AUXILIARY INJECTOR HEATER BLOCK
          Figure 2.  Detection System Configuration



r ~~>
o r } -\ { | p s
i 1 ; ; i ! i
' i i iiii*
\ O L' v-' ^ i

lt»*x yyxxx* ysoeA—lF
u V X
i />/
TUBE vxjx^



.— —



s^/ /?
VS >^
". ''
\ /



=fl(ft — O
"MJ •> — /


rx /^
S\ ft-
• f'
x"*" _""* x
\x ,'/

i i

^ \ / A r*i 1 1 i (\ n


                                  DEWAR FLASK
                                         FigureG.  Vacuum-Sublimation System


     Figure 4.  Chromatogram of Vacuum-Sublimed Material from Spiked SRM 1649.

                                           -UNKNOWN + PYRENE + FLUORANTHENE
                                         — BENZ(a)ANTHRACENE


TIME (min)
          Figure 5.  (a) Chromatogram of Vacuum-Sublimed Material from Virgin SRM 1649.
                   (b) Chromatogram of Vacuum-Sublimed Residual Material from SRM 1649.

                               Philip K.  Hopke
                      Institute for Environmental Studies
                  University of Illinois at Urbana-Champaign

    The objective of receptor modeling is to deduce information regarding  the
origins of observed concentrations of ambient species from those measured
concentrations.  This approach is in contrast to source or dispersion modeling
that predict the ambient concentrations from emission source data and the
meteorology of the region.  In general receptor modeling has focussed on the
identification of and apportionment of aerosol mass to particulate emission
sources .

    Several previous articles have reviewed the development and implementation
of these  models upto several years ago (Cooper and Watson, 1980; Gordon, 1980).
Prior work has primarily been empirical explorations of the application of these
approaches to specific urban air quality problems.  Recently, there has been the
beginnings of some more fundamental studies examining the basic mathematical
methods being used and the ways in which the limts of these methods can be
rigorously defined.

    The fundamental assumption of receptor models has been that a mass balance
can be applied; that is, the amount of any particular observed species is  a sum
of contributions from the independent sources of that species.  For example, the
total lead concentration observed in a parcel of air can be considered to  be the
sum of the lead in that parcel from motor vehicles burning leaded gasoline plus
smelter emissions plus refuse incinerators, etc.

         Pbt - Pbmotor + Pbrefuse + Pb smelter + '"
Total aerosol mass would represent a similar mass balance.  However, the
airborne  lead is not the only species in motor-vehicle-generated particles.
Thus, ^b   t   can be considered to be the product of the concentration of  lead
in the particles, ap^ motor times the mass of motor vehicle particles in the
sampled air parcel, r  .
   ^        v        motor

           motor ~ aPb, motor motor

Generalizing this approach to multiple variables measured in multiple samples
                    ikfkj             i=1'm' J=1'n
               K.~~ _L
where x
the conc
        .  is the amount of the ith species measured in the jth sample, a-^  is
        entration of species i in material from source k at the receptor  site,
and f^. is the amount of mass contributed to sample j by source k.

    There are two commonly used multivariate methods  to  solve  for  the  desired
parameters; multiple  linear regression and  factor  analysis.  In  the  multiple
linear regression approach, commonly  called the  Chemical  Mass  Balance  method, it
is assumed that the number of  sources, p, and  their compositions,  ajif's,  are
known.  The mass contributions, f, .'s, are  then  calculated  as  the  regression
parameters.  It is usually assumed  that  the compositions  do  not  change from
source to receptor and source-measured values  are  employed.  Efforts have been
made to examine reactions of polynuclear aromatic  hydrocarbons using first order
reaction kinetics (Duval and Friedlander, 1972).   These methods  have been widely
applied to the study  of a variety of  air quality problems with particularly good
results in the Portland Aerosol Characterization Study (Core et_  al_., 1983) and
in Washington, B.C. (Kowalczyck et_  al..,  1983).

    The other approach is factor analysis (Hopke,  1981).  In this  method  only
the ambient data is employed and the  analysis  is used to  deduce  the  number of
identifiable sources, their composition, ahd the mass contributions.  These
methods have been primarily applied to data from St.  Louis,  MO (Alpert and
Hopke, 1981; Liu et_ a_L., 1982; Sever in et_ al., 1983)  where  results have appeared
to be quite good.

    Both approaches have limitations  that are  only now being studied at a more
fundamental level.  Inherent in the regression approach are  problems in
correctly calculating the mass contributions when  two or  more  of the sources
have similar compositions even when those compositions are  perfectly known.  In
real situations where there are fluctuating compositions  measured  with sampling
and analytical errors additional problems arise.   In  many cases  in the
literature, the number of sources reported  to  be resolved accurately is
considerably overestimated and the  results  are much more  uncertain than they are
reported to be (Henry, 1983).

    A similar problem exists for the  factor analysis  in that it  also will not be
able to separately identify sources of similar composition.  However,  it  can be
used to find appropriate linear combinations of  the sources  that can be
accurately fit.  Furthermore,  factor  analysis  is driven by  variations  in  the
system that can come  from both varying source  emission rates and from
meteorology.  If the  latter dominates, two  sources with quite  different
composition may be found to be inseparable by  factor  analysis.   However,  factor
analysis is not unlikely to overestimate the resolvable number of  sources nor
overlook an unsuspected source.  Thus, a combination  of these  methods  is  the
best current approach to obtain results upon which air quality management
decisions can be confidently made.

Alpert, D.J. and P.K. Hopke, 1981. A Determination of the Sources of Airborne
Particles Collected During the Regional Air Pollution Study, Atmospheric
Environ. 15, pp675-687.

Cooper, J.A. and J.G. Watson, 1980. Receptor-Oriented Methods of Air Particulate
Source Apportionment, J. Air Pollut. Control Assoc. 30, ppl!16-1125.

Core, J.E., J.A. Cooper, P.L. Hanrahan, and W.M. Cox, 1982. Particulate
Dispersion Model Evaluation: A New Approach Using Receptor Models, J. Air
Pollut. Control Assoc. 32, ppl!42-1147.

Duval, M.M. and S.K. Friedlander, 1982. Source Resolution of Polycyclic Aromatic
Hydrocarbons in the Los Angeles Atmosphere: Application of a Chemical Species
Balance Method with First Order Chemical Decay, U.S. Environmental Protection
Agency Report No. EPA-600/S2-81-161,  January 1982.

Gordon, G.E., 1980. Receptor Models,  Environ. Sci. Technol. 14, pp792-800.

Henry, R.C., 1983. Stability Analysis of Receptor Models that Use Least-Squares
Fitting, Receptor Models Applied to Contemporary Pollution Problems, S.L.
Dattner and P.K. Hopke, eds, Air Pollution Control Association, Pittsburgh, PA,

Hopke, P.K., 1981. The Application of Factor Analysis to Urban Aerosol Source
Resolution, Atmospheric Aerosol: Source/Air Quality Relationships. E.S. Macias
and P.K. Hopke, eds, American Chemical Society, Washington, D.C., pp21-49.
Kowalczyk, G.S., G.E. Gordon, and S.W. Rheingrover, 1982. Identification of
Atmospheric Particulate Sources in Washington, D.C. Using Chemical Element
Balances, Environ. Sci. Technol. 16,  pp79-90.

Liu, C.K., B.A. Roscoe, K.G. Severin, and P.K. Hopke, 1982. The Application of
Factor Analysis to Source Apportionment of Aerosol Mass, Am. Ind. Hyg. Assoc. J.
43, pp314-318.

Severin, K.G., B.A. Roscoe, and P.K.  Hopke, 1983. The Use of Factor Analysis in
Source Determination of Particulate Emissions, Particulate Sci. Technol. 1,

                          FOR RECEPTOR MODELS
                by John G.  Watson and Norman F. Robinson
    Desert Research Institute, University of Nevada, Reno, NV  89506

    Numerous chemical measurements have become available  in  recent years
which  can  be used  in receptor  models to differentiate  between  and  to
quantify the contributions of source emissions to  ambient pollutant con-
centrations.      Instrumental   neutron   activation   analysis.    x-ray
fluorescence, ion chromatoqraphy.  and step-wise thermal  combustion have
been  used   individually  and  in  combinations  to  supply  chemical  con-
centration input data  to  receptor models  for a large number  of  samples.
X-ray  diffraction,   computer  automated  microscopy,  mass spectrometry,
electron capture  and flame ionization gas chromatography, and  isotopic
enrichment  analysis  have  recently been proposed  as analytical  methods
which  would  provide more  characterization of pollutant sources.   Since
these  analytical methods  can  be  very expensive,  and since most  receptor
model  studies are performed on  a fixed badget,  some objective procedure
of selecting the observables  to  be used in the models, and  the  measure-
ment methods required  to obtain values for  those  observables.   needs  to
be applied at the study design stage.  This paper  proposes such  a proce-
dure.    The  paper's  objectives  are  to  1)   describe  a  receptor  model
measurement specification  methodology. 2)  to  illustrate the  methodology.
and  3)  to  provide measurement uncertainty specifications for a  specific
application of tiie mass balance  receptor model.


    A  measurement  posesses   the  following attributes  (Watson  et  al.,
1983):   1) an  observable specification,   2)  a  value.  3)  a  lower  quan-
tifiable limit.  4)  precision.  5)  accuracy, and  6) validity.  The model
imposes certain requirements on  the  tolerances assigned to each  of these
attributes.  Measurement methods  must then   be selected  such that they
meet or exceed tiiese tolerances.   The  most important questions regarding
measurements used in receptor models are:

    •    Does a proposed observable  differentiate  between sources?

    •    How many  observables or  measurements  of the same  observables
         are required?

    •    How low must each detection limit be?

    •    What precision must each  measurement have?


    The following steps can be followed  to  provide  a quantitative answer
to these questions.

    1.   List all p likely sources and expected  contributions.

    2.   Obtain likely  emissions  compositions for  all quantifiable spe-

    3.   Set  the  true contributions  (S-i) and true  compositions  (a- • )  at
         expected values.

    4.   Generate n  true concentrations  using  a model which  physically
         represents the situation under  study.   The linear model
                  C~i =      a^ Sj, i = 1, n                       (1)

    5.   Simulate the measurement  process to obtain  measured values
         and a^^)  k=1  to m  times (Watson. 1979)  using random  numbers
         (e^ and  e^-;^)  drawn  from  a normal  distribution  of mean  zero
         and unity standard deviation, and  measurement uncertainties a,
         and o

                           Cik =    + eik
                           aij = *ij + eijk

    6.   Apply the receptor model to answer  the questions.

         •    To  determine  whether  or not  a new  observable  differen-
              tiates between  sources,  apply the model  to simulated data
              sets with and without  the observable.

         •    To  determine  the  number  of  different observables  or
              measurements,  apply  the  model  to  selected subsets  of the
              simulated data.

         •    To determine required detection limits  (D^) vary the ratio
              of Cj/O.^

         •    To determine required precisions, vary  

    To illustrate  this methodology,  it is applied  to the mass  balance
receptor  model  using  the  effective  variance  least  squares  solution
(Watson  et  al.,  1984).    \ desert  environment  is  likely  to  receive
ambient  parti dilate  contributions  from  secondary  nitrate.   secondary
sulfate,  soil, burning, and  motor  vehicle  sources.   Typical compositions
of these sources have  been drawn from Watson  (1979).   The chemical spe-
cies included  are  organic carbon,  elemental  carbon, NH^  NO3'  SO4'  A^'
Si,  Cl,  K, Ca,  Ti,  V, Cr,  Mn.  Fe,  Ni.  Cu,  Zn.  Br,  and Pb.    In this
application of the procedure, the  measurement precision requirements are
evaluated,  so a,-.  and a,.  . are  set to 10%,  20%,  and  30% of their respec-
                i       i T
tive values.

    The  results  of  this  application  appear  in  Table  1.   These results
indicate that  measurement methods  must have  precision less than 20% in
order  to assure  model calculations which  are within a factor  of two of
reality and  to have  the majority of  the calculated  source contributions
fall within one calculated standard deviation of the true contributions.
More detailed  observations  about Table 1  are  presented in Watson et al.


    This method can  be  applied  to  factor  analysis and  linear  regression
receptor models  in  order  to determine  the value of measurement methods
to the modeling  process.   Future  research plans include  these applica-
tions  along with  creation  of  a  model /measurement  evaluation computer
package which  can be  used to design  future receptor modeling  studies in
an optimal manner.


1.  Watson,  J. G. ,   1979.     "Chemical  Element  Balance  Receptor  Model
    Methodology  for  Assessing  the Sources of Fine  and  Total  Suspended
    Parti culate  Matter in  Portland,   OR, "   Ph.D.   Dissertation,  Oregon
    Graduate Center,  Beaverton,  OR.

2.  Watson.  J.G.,  P.J.  Lioy, and  P. K.  Mueller,  1983.   "The  Measurement
    Process:   Precision.   Accuracy.   and   Validity"   in   Air  JSampling
    Instruments for Evaluation  of  Atmospheric Contaminants, frth Edition,
    American   Conference   of    Governmental    Industrial   Hygienist.
    Cincinnati, OH.

3.  Watson,  J.G.,  N.F.   Robinson,  A.. P.  Waggoner.   R.E.  Weiss,  and  J.
    Trijonis,  1984a.   "Error   Analysis   of  Mass Balance and Particle
    Scattering Budget for RESOLVE"  Desert Research  Institute Document
    6660. 1D1, Reno, Nevada.

4.  Watson,  J.G.,   J. A.  Cooper,   and  J. J.  Hun thicker,   1984b.    "The
    Effective  Variance  Weighting for Least Squares  Calculations Applied
    to   the   Mass  Balance  Receptor   Model"  accepted  by

       TABLE  1.   Averages.  Standard Deviations
    and Ranges of Source Contributions and Their
Uncertainties as a Function of Uncertainty Level.
                 (Units are yg/m  )
Source Type
1. NH NO

2. (NH4)2S04

3. Soil

4. Burning

5. Motor Vehicle

Average S-
Range of S-
Std. Oev S_
Average as .
Std. Oev osl
Range of Og.
Average S-
Range of Sj
Std. Oev Sj
Average og .
Std. Oev og.
Range of ag^
Average Sj
Range of Sj
Std. Dev Sj
Average Og .
Std. Oev 
                    TO COMPLEX CHEMICAL DATA

               W.  3. Dunn III and Michael Koehler
                Department of Medicinal Chemistry
              The  University  of  Illinois at Chicago
                      833 South Wood  Street
                     Chicago,  Illinois 60680

                           Svante Wold
                 Research Group  for Chemometrics
                         Unea Un i ve r s i ty
                      S 901  87 Umea,  Sweden

Introduct ion

    A number  of analytical  techniques  are used  in air  quality
monitoring depending on the  nature  of  the agents  being monitored.
For   low molecular  volatile  organic   chemicals  such  as
hydrocarbons, aliphatic  halides,  etc., the me thod of  choice is
gas chromatography/mass  spectroscopy (GC/MS).   Figure  1  is an
illustration  of  an output  fr om GC/MS  analysis of  a  c om p1e x
mixture.  The  output consist of two  parts,  the gas chromatogram
and  the mass spectrum.  The  GC data  are  2 -dimensiona 1  in
concentration VS.   retention  time.   The  mass  spectrum  is  3-
dimensional  with ion intensity as a function of  retention time
and mass (m/e).  The GC data contains information regarding the
number of components in the sample and their concentration.  The
ma ss  spectral data,  if  in terms of  relative ion  intensities,
contains information which can be used to  identify  the  chemical
species  present.   By applying methods  of  classification  or
pattern  recognition  to such data it  is possible  to classify and
identify the  components present.  It is  this aspect of the data
analytic problem that will be discussed  in  this  report.
Figure 1.  Output  from a  GC/MS  analysis  of  a  complex mixture,


Hi storical

    Before a discussion of the specific methods  of  data  analysis
is  presented,  it  is  worthwhile  to  review the  levels  of
information which  can  be obtained from  the application  of
classification methods to GC/MS  data  (Albano, et  al.,  1978).
There are  two  major  objectives  of such  an  analysis:  1)
classification  and 2) quantification.  With these  two objectives
in mind,  three levels  of  information can be obtained  from the
data  analytic  method.    At  the  first  or  lowest   level,
classification  is  into one  or  the other  of  a group  of well
def i ned classes.

    At the  next  level  it would  be possible,  considering  the
example  above, to  classify a compound  into one or the  other  of
the defined classes, with the possibility  that  the  sample  may  be
neither.  At  the highest  level  of  classification, it  is  the
objective to quantify the amount  of a classified compound.   This
level includes,   in  addition  to a classification  step,  a
calibration step.

    A number of pattern  recognition methods are  available.   These
methods  have  differing  potential  with  regard  to  the  above
mentioned levels of classification so at is necessary to know  in
advance  the  desired  level  of  classification.   The  methods
corrmon ly  used are:

1.  the hyperplane  or  class discrimination methods  such  as
    the linear  1 earning machine (LLM) and linear discriminant
    analysis (LDA),

2.  distance  based  methods  such  as  the  k-nearest neighbor
    (KNN) and

3.  the class modeling methods such as SIMCA.

    All methods  of pattern recognition  operate  at the first
level.  However,  LLM and LDA operate only at this level.  KNN and
SIMCA both operate at the second  level while SIMCA can operate  at
level three.

Theory of SIMCA Pattern  Recognition

    The objective  of  methods of classification is  to identify
objects,   in this case  lowmolecular weight volatile organics.
For  the purpose of  this report,  the identification will  be based
on information obtained in the mass  spectra  of the compounds.
From  information obtained on compounds similar to  those whose
identity  is  to  be determined, rules are  developed which allow the
unknowns to  be  classified.   The compounds  of  known  class
assignment are  the tjiajjljjl£. £ej^s_ while those compounds of unknown
classification  are  the t£sjt_ j^e t_ compounds.

    In order to apply mathematical methods  to the data in the


 analysis, the mass spectra for  each  compound  is  represented as an
 object  vector as  in Figure 2.  The elements of the vector are ion
 intensities  at  each mass  in  the  interval observed  for  the
 classes. The object vectors  are  tabulated  in matrix form  in which
 the  elements  of  the matrix are ion intensities,   with  j<  the
 compound index and _i_ the mass  index.                       ~

                   x= (i..i..i

              sample   20  21  22       i     p

               o                           class 1
                                   I, .     class 2

Figure 2. Objiect vector  and matrix employed in SIMCA analysis.

    If the compounds in  the  classes  (training sets)  are  similar,
the data for each class  can  be modeled by a principal  components
model  in few terms  (Wold,  1976).  This  is shown  in equation  1
where  I is the mean of  ion, _i_, A is the number of product terms
or principal components in  themodel  and e^j  is the residual of
the observed and  predicted  ion  intensities.   The  product  terms,
which model  the  systematic variation in the  data,  are  composed of
the loading  term,  ba  and the  principal component, t k.  For A=0
the class is represented by a point in space (class members are
identical);  for A=i the  class data structure is approximated by a
line  and  for A ^ 2 the class is approximated by a   plane or
hyper plane.

    Classification  of  the unknowns is  based  on information  in the
residuals,  ek-,  for  the test  compounds.   From  the  fit of the
training  set data to  their respective class mode Is,  a residual


standard deviation (RSD)  for each object and  for each class  can
be calculated.  Fr om the fit of the unknowns to the class mode 1 s ,
a classification result can  then be  obtained.  Since the RSD is
approximately F-distributed  a  confidence  interval  for  the
classification result  can  be established.

    A geometric interpretation of  the  SIMCA  classification result
is shown in Figure 3.   Here the result is shown in 3-dimensions
for convenience while  in reality  the data space is  much higher  in
d imen s i ona 1i t y.
                                 class 1
t i on .
3.  A 3-dimensiona1  representation  of a SIMCA classifica-
    In this example,  the  classes are represented  by  a  line.   The
projection of the objects onto  the  line  gives the  relative
position  of each  compound  in  the class  structure.   This  is
important   if classification is desired at the third level.   The
RSD about  the  line  defines a  volume  element in  space and
classification is based on where the unknown falls  with  respect
to the defined classes.

    The class modeling philosophy has a number of  advantages  when
considered in terms  of  the  levels of  pattern  recognition  as
discussed  earlier.   If the possibility exists  that a compound  is
a mem be r of none of the defined classes it wil 1  be observed as an
outlier to the defined classes.  This result is not  possible  with
the methods of LLM or  LDA.  Another advantage of this approach  is
that  if  other information,  such  as  health effects  data,  are
available for the membe rs of the defined classes,  correlation
methods can be used to relate  levels of these effects to chemical

Pretreatment of Data

    In order to enhance the various types of information in mass
spectral  data, the spectral data can be scaled or  transformed.
This  is  especially  critical with ma ss  spectral data  since in
normalized form it may  not  be appropriate  for classification
purposes.   Mass spectra are interpreted by attempting to identify
fragmentation patterns within the spectrum  of a compound.   These
patterns  result from the  loss of a common  fragment(s) to  give
sequences of peaks  that are  related or correlated.   This  is
illustrated in Figure 4 in which ion of  ma ss j_ results fr om the
fragmentation of ion jj_+j_ by loss  of  fragment  j_.

Figure 4.  Example  of  fragmentation patterns within mass  spectrum.

    The molecular weight  of  compounds which  contain  a common
functional  group  will  vary  depending  on  the masses  of  the
residual molecular  fragments  attached  to  the  functional  group.
The loss of a common  fragment j_ within the series can occur  in
each  spectrum to  varying  extents,   but will  be  shifted  to
different  masses  _i_.   Classification  should  be based  on  the
relative extent  of this  fragmentation and this  information  can  be
amplified  by  applying the autocorrelation transform (Wold,  et
al.,  1984)  to  the  normalized mass  spectrum.  The  autocorrelation
transform  is  given in equation 2  (Box,  Hunter  and Hunter,  1978).
                       Ed; - T)2

   and Ij+i  are the intensities of the respective ions i  and i+j
     I is the spectral mean.   r:  is called the autocorrelation
coefficient  for lag j.   Its  range is   -1   r:   1 and measures
                            the MS which  result  from loss
correlation between ions  in
c orrmo n f r a gme n t j .
    The  MS of  1,2-dichlorobutane  and its  autocorrelation
transform  are  given   in   Figure  5 .    Very  pronounced
autocorrelations are  observed between ions 2,  36  and  38 mass
units apart.  These result  from  loss  of fragments which  contain
the isotopes  35C1  and  37C1 and from the loss of HC 1  containing
the se isotopes.
            1 2
                                  38 31
Figure 5.
MS of  1,2-dichlorobutane
(bot torn) .
                                   (top)  and its autocorrelation
Application of SIMCA to MS  Data

    In order to  illustrate  the utility of SIMCA to classification
of volatile  organic compounds,  the MS of 9 dich1 orobutanes and 8
chloromethyl butenes were obtained from  the Finnigan catalogue of
mass spectra.  These two classes of  halogenated  compounds were
used as training sets.   The data matrix consisted  of relative ion
intensities in  the interval  39  to  118 m/e  with each  spectrum
consisting of the  16 most intense ions.

    Each spectrum  was transformed  to  the autocorrelation spectrum
and SI MCA wa s applied  to both  types of data  for  the  purpose of
comparison.   The  data for  each class were  modeled  by  a  2-
component (A=2  in  equation 1) model.   This  is  equivalent  to
fitting the  data  to a plane.  This is somewhat arbitrary but is
done in this case  as to give the  analyst a view of the structure
of the classes.  By  generating principal components  plots  of the
data,   the approximate structure  of the data can be observed.

    Figure 6 is such a  plot of  the  classes when the MS data are
modeled.   When  compared  to  the  results  of  applying  SIMCA to the
autocorrelation transformed spectra (Figure 7) it  appears that
the  two  classes  overlap  with very  little  structure  in  the
dich1orobutanes from either treatment.   A more  revealing method
of displaying  the results  is  the  Cooman's  plot  (Wold, et al.,
1984).  This  plot  is  obtained by fitting all of the compounds to
the class 1 and class 2models,  respectively, and calculating the
distance  of  each compound  to each class model.  When these class
distances are plotted from the  analysis of the MS  data  (Figure 8)
and  the autocorrelation transformed data   (Figure  9)  it is seen
that  the latter  result in a  much better  description  of  the
chloromethyl butenes  while the dich1orobutanes are  not well
described by either method.  It is  possible,  then,  to classify
the class 2 compounds with a high level of  certainty and that the
classification results from the autocorrelation data are much
bet te r .
                 A  A
               • di-CI2 butanes

              A Cl butenes
Figure  6. Principal
data are modeled.
components plot of the classes when the MS


CO     (/)
OO     _
                                           distance to class:


                                                              D    T
                                                              cr    Q

                                                              5    f
                                                                                                    o  -•
                                                                                                          2  -g
                                                                                                          C/1  .

                                                                                                          O.  -T3
                                                                                                          (a  -i




                               • di-CI2 butanes

                              /\ Cl butenes
                          class 2
                                         c la ssi
                             distance to class 2
Figure 9. Cooman's  plot of  the  SI MCA analyzed autocorrelation
transformed  MS data.

The authors wish  to  acknowledge discussions  with Dr.  Donald Scott
of  the  EPA Environmental Monitoring and  Surveilance  Laboratory
regarding  the  treatment of mass spectral  data.



Albano,  C.,  W.  3.  Dunn  III,   U.  Edlund,  B. Norden,  M.
Sjostrom and  S.  Wold,   1978.   Four  Levels  of  Pattern
Recognition.  An a 1 . Chbn^ Com£Ujt_._  Te_c_h_._ O_£t_irru_,  121> 429-443.

Box, G.  E.  P.,  W.  G.  Hunter  and  3.  S.  Hunter,   1978,
     lilies. 12JL ^^^llUl^ni6-!!» Wiley and  Sons, New York.
Wold,  S.,  1976.  Pattern Recognition  by Disjoint Principal
Components Models.,  Pa rt_e r_n_ Re c_O£nji_ tj_0jn ,  ^,  127-139.

Wold,  S., C. Albano, W. 3.  Dunn  III, U.  Edlund, K. Esbensen,
P. Geladi,  S. Hellberg, W.  Lindberg  and M. Sjrostrom, 1984.
Multivariate Data Analysis  in Chemistry, Proceedings  of
    N-A.!0. ASj_ 011  Ch_emome^£_i_c _s_, B.  R.  Kowlaski,  Ed.,  Feidel
    Publishing Co.,  Dordrecht,  Holland.

                             DESCRIPTION OF A CONTINUOUS
                            SULFURIC ACID/SULFATE MONITOR
        George A. Allen, William A. Turner, Jack M. Wolfson, John D. Spengler
                  Department of Environmental Science & Physiology
                          Harvard School of Public Health
                              665 Huntington Avenue
                           Boston, Massachusetts   02115
       A flame photometric/thermal speciation system for continuous measurement
of ambient total sulfate, sulfuric acid, and two other sulfate fractions is
described.  The instrumentation is suitable for long-term ground-based installa-
tions, and has a limit of detection for sulfate as sulfuric acid of 2 yg/m^
for an integrated sample period of one hour.

       An example of episodic ambient data from the system is presented.  These
data are compared to water soluble sulfate data from a co-located dichotomous
sampler, and particle scattering extinction coefficient (b  ) data from an inte-
grating nepholometer.

       Limitations of the thermal speciation technique with regard to the measure-
ment of total strong acidity of sulfates are discussed.
Presented at the Fourth Annual National Symposium on Recent Advances in Pollutant
Monitoring of Ambient Air and Stationary Sources in Raleigh, North Carolina,
May 1984.

I. Introduction
Acid Aerosols
     There is growing evidence that atmospheric aerosols in the lower troposphere  at times can
be acidic.  Unfortunately, the measurements have  been neither  extensive nor systematic.  The
temporal variation and the geographic extent of acid aerosol events have not been  documented.
While nitrous, hydrochloric, or organic gases can lead to acidified particles by absorption or  con-
densation, sulfur gases are believed to be the dominant source of acid species in the atmosphere.
Health Evidence
     Some studies of human  populations have linked sulfur dioxide and ambient particulate sul-
fates to increased respiratory diseases, but have been unable to identify the specific pollutants
responsible. It remains to be established which species of sulfates are physiochemically important,
since  they can occur within  a variety  of metal  cations  as  well as  with the more  common
ammonium and hydrogen ions.  As has been illustrated in  a recent review of the toxic effects on
pulmonary macrophages, these sulfate species have  widely varying effects.  Of the variety of sul-
fate species that exist in the atmosphere, the strong  acid sulfates (ammonium bisulfate and hydro-
gen sulfide) residing in the submicron size range  are more likely to induce responses in the human
respiratory tract.  Evidence exists for alterations  in epithelial secretory cells in the lower bronchial
airways in rabbits exposed to 250 /zg/m3 sulfuric acid after relatively  short duration exposures
(8h/d,  5d/w, 1m).  In his work, Lippman points  out that these studies provide further support for
the role of acid sulfate species in the pathogenesis of chronic bronchitis via effects on the mucocili-
ary clearance system.
     As a component of Harvard's Air Pollution Health Study,  we have developed an operational
monitoring system for real-time  sulfate/sulfuric  acid aerosol measurements.  This system will be
deployed in each of the six cities participating in the Harvard Air  Pollution Health Study in  order
to more fully characterize the nature of sulfate particulate exposure. Peak, hourly, and daily con-
centrations of sulfate particle and sulfuric acid fractions will be incorporated into our aerometric
database for the  analysis of daily pulmonary function data,  or to  more fully understand the simi-
larities or differences among our cities.
     As we assemble additional  units, we will characterize the  nature of atmospheric aerosol aci-
dity throughout the year in each city.  A plan being considered  involves the  use of the continuous
sulfate monitor to trigger a dichotomous particle sampler during 'episodic' conditions.  The  parti-
culate  filter samples would be used to characterize  the elemental composition (by XRF) or  ion
composition of aerosols, including total strong acid (H+).

Instrument Description
     Based on work done  by  Dr. Roger Tanner  at Brookhaven National Laboratories and earlier
cooperative work with Dr. Rudy Husar and Geoff Cobourn at Washington University, we at HSPH
have developed a low temperature volatilization flame  photometric detection (FPD)  method  for
continuous measurement  of  ambient  sulfate aerosol suitable  for long-term ground  based field
operations.  This method allows for some discrimination of the acidic component of the aerosol
due to the higher rate of volatilization of the r^SO^ species at lower temperature, as well as  those
sulfates that are stable at 300 °C (Sulfates of calcium, sodium, lead, zinc, iron, and copper).
     The FPD method has been used  routinely  to measure total gaseous sulfur in ambient air by
excluding  particles with a membrane  filter. The sulfur is  detected in a hydrogen  flame which
results in  an optical emission  (fluorescence)  from the electronically  excited sulfur dimer (S2*).  To
determine particulate sulfur,  a lead oxide  (PbO) coated tube  is  used as a diffusion  denuder to
remove gaseous sulfur species. Under  appropriate conditions of laminar flow,  the particles, with
much  higher   momentum,  pass  through  the  denuder  tube,  while  the  sulfur-containing gas


molecules collide and  react with the walls and remain deposited on them. SF6 doped hydrogen is
used to improve the FPD's sensitivity and stability and reduce interferences.
II. Thermal Analysis
     Thermal analysis allows discrimination between the different sulfate species. With this tech-
nique,  the ambient air is  heated  initially  to about ~120°C.  At this point most  of the H2SO4
(sulfuric  acid) is  vaporized.  The vaporized material is subsequently scrubbed by  the denuder.
Consequently, if a decrease is observed in the  FPD signal when the temperature  is elevated  to
120 ° C, the size of this decrease may be correlated with the amount of H2SO4 in the ambient air
sample.  The next step is to heat the  sample air to about 300 ° C.  At this point  the NH4HSO4
(ammonium bisulfate)  and (NH4)2SO4  (ammonium  sulfate) in  the particles is  completely vapor-
ized. Again,  the amount of the sum of these two species can be correlated with the change in the
FPD signal.  Any  non-volatile sulfur particulate [i.e.,  Na^O,, (sodium sulfate) from marine air
masses] which is  present  will give  a residual  FPD signal.  Interference by any potential  non-
denuded  gaseous sulfur species and changes in the FPD's  zero will be accounted for by determin-
ing the analyzer's baseline signal with particle-free air.
     The cycle of measurements (12 min./cycle)  is controlled by a timer that is synchronized
with the data acquisition system,  and includes:  1) ambient temperature; 2) 120 ° C;  3) 300 ° C;  4)
Instrument Baseline. See Section V for a detailed example of the thermal analysis technique.
     The temperature for the second part of the cycle  (~120°C) is chosen  to maximize the  frac-
tion of H2SO4 that is volatilized without any significant loss of (NH4)2SO4.  With  the apparatus
we  are currently  using, about 5% of the H2SO4 remains unvolatilized at  the temperature  that
~1% of the (NH4)2SO4 is lost.  The Limits of Detection for the system using SFg doped hydrogen
are presented in Table I.  The system's  flow diagram can be found in Figure  1.
in. Limitations
     It is important to note that this method generally underestimates the total amount of strong
acidity of ambient sulfate containing particles because both ammonium bisulfate (NH4HSO4) and
ammonium sulfate |(NH4)2SO4] vaporize at about the same temperature.  The NH4HSO4 is almost
as strong an acid as H2SO4, whereas (NH4)2SO4 is a relatively weak acid.
     If and when there are occasions in which  there are aerosol particles that  are  only pure
H2SO4,  and  others that  are only pure (NH4)2SO4,  the TA-FPD method will give a good quantita-
tive determination  of total  strong acidity.  In general, however,  we expect that  whenever any
H2SO  is present in ambient air, there is also some NH4HSO  present.  Under these  conditions the
TA-FPD method will underestimate  the amount of total strong acidity.  The overall situation is
even more complicated  since: 1)  individual  aerosol  particles generally  have  a  mixture which
includes sulfate, ammonium, and hydrogen ions; and 2) different aerosol particles within the same
sample of ambient air may have varying amounts of each  of these three ions.

IV.  Calibration
     The FPD method requires a calibration to determine the relation between the  aerosol sulfate
concentration  and  the size of the emission signal as measured by  the flame photometer. Dynami-
cally generated  SO2/air mixtures between 2 and 15 ppb  are used  as the  principle  means of cali-
brating the  flame  photometer.  A portable aerosol generator is  used  in situ to  determine the
system's response to H2SO4  and (NH4)2SO4 at different temperatures.  A diagram  of the  aerosol
generator can  be found in Figure 2.  In addition, we have semi-continuous calibration by measur-
ing the water  soluble sulfate of simultaneous samples  (4 to 24-hour collection period) from mem-
brane filters.  This method  will compare the time-integrated FPD signal  for  total sulfate with
chemically determined water-soluble  sulfate data.   Figures 3 and  4 show the relationship between


the FPD sulfate and a dichotomous sampler and integrating nephelometer.
V. Explanation of Continuous Sulfate Data Reduction
     Figure 5 is an example of the system's output during an episode in St.  Louis, Missouri  on
December 20, 1983, showing both the Meloy analyzer's output and the temperature of the sample
heater tube.  Four values are taken from the particulate sulfate analyzer every 12 minute  meas-
urement  cycle.   Point #1 represents the analyzer's output  when the sample air has  not been
heated. Point #2 is the output then the  sample air is heated  to ~120 °C to volatilize most  of the
H2SO4. Point #3  is  the output when  the  sample air is heated to 300 °C to volatilize H2SO4,
ammonium sulfate  and ammonium bisulfate.  Point #4 is the output when the sample air has
been filtered to remove all particles (instrument baseline).
     The first step in reducing the data is  to calculate sulfate concentrations for the  first three
points, using Point  #4 as the baseline. In this example, data  are reduced as follows:
                                               (net chart div x __ o.™
                          ppb  bu 2                              chart div

     The scaling factor of 3.06  and the exponent of 0.929 are derived from the most recent cali-
bration of the analyzer's net voltage output against multiple SO2 concentration inputs.  The first
point is total sulfate in jtg/m3. The third point  is sulfate that does not volatilize at  300 ° C
(sodium sulfate, etc.) in //g/m3.  To calculate sulfate as sulfuric acid, subtract the sulfate reading
for Point #2 from Point #1 and multiply that result by 1.08 (a factor that corrects for the frac-
tion  of sulfuric acid that is not volatilized  at the mid-point temperature,  determined by in-situ
testing for this specific  unit).   To calculate sulfate  as  ammonium sulfate plus ammonium bi-
sulfate  (this system cannot distinguish  between these two species of sulfate), subtract the non-
volatile sulfate and sulfate as sulfuric acid from the total sulfate.
For this example,
               Point #1 = 52.0 /ig/m3 Total Sulfate
               Point #3 = 2.7 //g/m3 Non-Volatile Sulfate
               Point #2 = 14.2 //g/m3 Sulfate, so:
                       Sulfate as sulfuric acid = (52.0 - 14.2) x 1.08 = 40.8 /ig/m3
                       Sulfate as ammonium sulfate
                        plus ammonium bisulfate = 52.0 - 40.8 - 2.7 = 8.5
     The signals from the sulfate analyzer are sampled by a data logger and stored on a cassette
tape. The tape and strip charts are changed  after 7 days and returned to HSPH for validation,
processing, and  Quality Assurance  checks.  Figure 6 is a plot of sulfate and particle scattering
extinction coefficient data from St. Louis for December 16 to December 23, 1983.

     The cooperation and help of Dr. Roger Tanner (Brookhaven National Laboratories) contin-
ues to be invaluable.  We are also indebted to Andrew English for our original prototype develop-
ment, which he began at HSPH in 1981; to Anthony Majahad, Stephen Bertolino, and Craig Nor-
berg of  the HSPH staff for assistance in construction, development, and operation of this system;
to Steve Fick and John  Chao for data processing efforts; and to Allison Maskell for typing and
editorial assistance.  This work is funded by grants from the National Institute of Environmental
Health Sciences (ESP-1108), and the Electric Power Research Institute (RP-1001).



Acid Aerosols
1.   R.L.  Tanner, B.P. Leaderer, J.D. Spengler, "Acidity of atmospheric aerosols," Env. Sci.  8
        Tech. 15:1150 (October 1981).

2.   J.W. Waldman,  J.W.  Munger,  D.J.  Jacob,  R.C.  Flagan,  R.C.  Morgan,  M.R.  Hoffman,
        "Chemical composition of acid fog," Science 218:677 (November 1982).

3.   P.J.  Lioy, "Ambient measurement of acidic sulfate species in the U.S.," Presented at the
        76th Annual Meeting of the Air Pollution Control Association, Paper No.  83-8.3, Atlanta,
        Georgia, June  1983.

4.   P.D.E.  Biggins, R.M. Harrison,  "  Characterization and classification  of atmospheric  sul-
        phates," J. Air Poll. Control Assoc. 29: 838 (1979).

Health Evidence

1.   R. Ferek. "Review of atmospheric acidity measurements," Progress  Report ~  Study  of Health
        Effects of Exposures to Airborne Particles, Spengler and  Ozkaynak, Harvard University
        DOE, HERAP  contract,.

2.   J.G.  French, G. Lowrimore, W.C. Nelson, J.F. Finklea, T. English, M. Hertz, "The effect of
        sulfur dioxide and suspended sulfates on acute respiratory disease," Arch.  Environ. Health.
        27:129 (1973).

3.   D. Levy, M.  Gent, "Relationship between acute respiratory illness and air pollution levels in
        an industrial city," World Health Organization (UNESCO) International Symposium on
        Recent Advances in the Assessment  of  the Health  Effects of  Environmental Pollution,
        Vol. Ill, Paris,  France: Commission of the European Communities, Luxembourg,  1975.

4.   "Health consequences of sulfur  oxides: a report from CHESS,  1970-1971, U.S. Environmental
        Protection Agency," EPA 650-74-004, U.S. Government Printing Office, Washington, DC,

5.   A.A.  Cohen, S.  Bromberg, R.W. Buechley, L.T. Heider-Scheit, C.M. Shy, "Asthma and air
        pollution from  a coal-fueled  power plant," Amer. J. Public  Health 62:1181  (1972).

6.   E.G.  Ferris,  l.T.  Higgins, J.M. Peters, "Sulfur  oxides  and suspended  particulates,"  Arch.
        Environ. Health 27:179 (1973).

7.   U.S.H.O. Committee on Science and Technology,  "The Environmental  Protection  Agency's
        research program with primary emphasis on the community health  and environmental
        surveillance system (CHESS): An investigative report," U.S. Government  Printing Office,
        Washington, DC, 1976.

8.   M. Lippman,  "Effects of  airborne  particles  on physiological  parameters," Study of  Health
        Effects of Exposures to Airborne Particles, Spengler and Ozkaynak, Harvard University
        DOE, HERAP  contract, January 1983.

9.   R.B.  Schleslinger,  L.C.  Chen, G. Leikauf, D. Spektor, "Alteration  of lung defenses by acid
        sulfates,"  Presented at the 76th  Annual Meeting of the Air Pollution  Control Association,
        Paper No. 83-8.4, Atlanta, Georgia, 1983.


10.  M. Lippman, "Health effects of atmospheric aerosols," Presented at the 76th Annual Meeting
        of the Air Pollution Control Association, Paper No. 83-8.7, Atlanta, Georgia, 1983.

11.  R.B. Schlesinger, M. Halpern,  R.E. Albert, M. Lippman, "Effect of chronic inhalation of sul-
        furic  acid  mist upon  mucociliary  clearance from  the lungs  of donkeys,"  J. Environ.
        Pathol.  Toxicol. 2:1351 (1979).

12.  E.G. Ferris, Jr., F.E. Speizer, J.D. Spengler, D. Dockery, Y.M.M. Bishop, M. Wolfson, and C.
        Humble, "Effects of sulfur oxides and respirable particles on human health,"  Amer. Rev.
        ofResp. Disease, 120:767-779 (1979).


1.   S.S. Brody, J.E. Chancy, "Flame photometric detector, The application of a specific detector
        of phosphorus  and for sulfuric compounds sensitive to sub-nanogram quantities," /.  Gas
        Chromatogr. 2:42 (1966).

2.   D.C.  Camp, R.K. Stevens,  W.G. Cobourn,  R.B. Husar,  J.F. .Collins, J.J. Huntzicker, J.M.
        Jaklevic, R.L.  McKenzie, R.L. Tanner,  J.W. Tesch, "Intercomparison of concentration
        results  from fine particle sulfur monitors," Aim. Env. 16:911 (1982).

3.   W.G.  Cobourn, "In-situ measurements of sulfuric acid  and  sulfate  aerosol  in  St. Louis,"
        Ph.D. Thesis, Washington University, Sever Institute of Technology, St. Louis, Mo., 1979.

4.   W.G. Cobourn, R.B. Husar, J.D. Husar, "Continuous in-situ monitoring of ambient particu-
        late sulfur using flame photometry and thermal analysis," Aim. Env. 12:89 (1978).

5.   P. Gormley, M. Kennedy, "Diffusion from a stream flowing through a cylindrical tube," Proc.
        R. Ir. Acad. (52A), 1949.

6.   J.J. Huntzicker, R.S. Hoffman, C. Ling, "Continuous measurement and speciation of sulfur-
        containing aerosols by flame photometry," Atm. Env. 12:83 (1978).

7.   D.B.  Kittelson, R. McKenzie, M. Vermeersch,  F.  Dorman,  D.  Pui,  M. Linne,  B. Liu, K.
        Whitby, "Total sulfur aerosol concentration with an electrostatically  pulsed flame pho-
        tometric detector system,"  Atm. Env. 12:105 (1978).

8.   T. Sugiyama,  S.  Yoshihito, T. Takeuchi, "Characteristics of S2 emission intensity with a
        flame photometric detector," J. Chromatogr. 77:309 (1973).

9.   R.L.  Tanner,  P.H. Daum,  T.J.  Kelley, "New  instrumentation  for  airborne  acid  rain
        research,"  Environmental  Chemistry Division, Brookhaven National Labs. BNL  31596
        Presented  at the  12th  Annual  Symposium on the Analytical Chemistry of Pollutants,
        Amsterdam, The Netherlands, April 14-16, 1982.

10.  R.L. Tanner, T. D'Ottavio, "Preparation of a gaseous sulfur  denuder," In-house  document,
        Brookhaven National Labs., Upton, N.Y.,.

11.  R.L. Tanner, T. D'Ottavio, R.  Garber, L. Newman, "Determination of ambient aerosol sulfur
        using a continuous flame photometric detection  system. I. Sampling system  for aerosol
        sulfate  and sulfuric acid.,"  Atm. Env. 14:121  (1980).

12.  T. D'Ottavio, R.  Garber, R.L. Tanner, and L. Newman, "Determination of ambient aerosol
        sulfur using a continuous flame photometric detection system. II. The measurement of low
        level sulfur concentrations  under  varying atmospheric conditions."  Atm. Env. 15:197-203
        (1981) .


                           TABLE I
            HSPH Continuous Sulfate TA FPD System
                     Limits of Detection
                   with  SFg  Doped  Hydrogen


As Sulfuric Acid

As Ammonium Sulfate Plus Bi-Sulfate

That Does Not Volatilize at 300° C
L.O.D. in yg/m3

Average Period
1 Hour    4 Hours
24 Hours

1)  L.O.D. is defined as twice the short term peak to peak noise
    of the system.

2)  A concentration of five times the L.O.D.  is necessary to insure
    data precision of 10%.

 3-PORT ,.
         ZERO CYCLE
                                                     5/im FILTER

                                                PUMP!	1	*-
                                                    , OUTSIDE
                                                     (SAME AS
                                                      SAMPLE AIR)
                                              (ACTIVATED CHARCOAL
                                                  a SILICA GEL)
                                  5/im FILTER
                                                      NH3(~18ppm(a 3mL/min) "
                                                                PERM. TUBE
                                    FIGURE I

                         HSPH AEROSOL GENERATOR
                                   r-PORT FOR FPO
                                     ZERO AIR SAMPLING
  AT 15PSI -
AIR— i
FILTER 5 	 ,
7 -'


— «J













!) t

i. I

                                      SYPHON LINE
                                   FIGURE 2

          ST.  LOUIS  DICHOT SOn  (OF)  VS.  FPD  TSOU
                       24  HR  INTEGRRTED SRMPLES
                      12-16-83 THROUGH 12-22-83
1 1
 5 -

             FPD TSOu = (DICHDT SOU)0.982 - 0.45
                    R2 = 0.931
                        5   6   7  8   9  10  11  12  13  14  15
                        DICHOT SOu (OF) „
                          FIGURE 3
                   ST.  LOUIS  FPD TSOu VS  BSP

                       24 HR INTEGRflTED  SflMPLES
                       12-16-83  THROUGH  12-22-83

0.0677 (FPD TSOU)
   RJ = 0. B91

   . 7


   .5 -

                                                   5  2
 a.   U
 tn  • ^


   . 1


                            FPD TSOu,  wG/M3
                                   10   11  12
                          FIGURE 4

           FIGURE 5

              ST.  LOUIS    (ONE-HOUR  flVERflGE  VflLUES)

          FROM  350:00:00  (12/16/83)   TO  357:00:00  (12/23/83)
                   -t-t-1 1-1-
                   -f-f i-1-1-
                   .}.; 1 11-.

                   •t-t-1 1-1-
      "t"f~t-! !-
      fr-f-f H"
      -t-t-4-f I-
                       •i H4 i-
                       -; t M-I-
                                  -i-f- . .
                                  -1 (- .-(--I-
• t-i-fi-
•t-t ',-', I-
             •4-4  .
      44 4- ! -i

      T'TT t T
 T I i i 1  i,
• t-t-ri-i- • «•
-t-i i-fr

              i-f-i- -t-44-i--i-
              I-4-4- -+4-I-1 t-
           •t 4-4
                                                  1 -I -I -r4-
                                                  •4 -4 -i 44-
                                        3S3          35U


                       TOTRL  SULFflTE.    SULFflTE flS  SULFURIC  flCID
 a  20
                                        353         3SU          355


 x  i
                                        353         3511


                                       FIGURE 6

                        AUTOMATED  SAMPLING  AND ANALYSIS  OF
                         FLUE  GASES  FROM PLASMA PYROLIZER
                        Marek  E. Krzymien and  Lome  Elias,
                      National Resesarch Council of Canada,
                       National Aeronautical Establishment
     A Canadian company, Plasma Research, Inc., of Kingston, Ontario, is
currently constructing a plasma torch incinerator for the purpose of disposing
of toxic waste chemicals on a commercial scale.  Hazardous materials, such as
PCBs, when subject to the intense, electrically-produced  plasma of the facility,
are expected to undergo a thorough chemical degradation to form innocuous
products such as C02 and water, or other products which can be readily
neutralized and released safely into the environment.
     The breakdown process occurring in  the plasma is highly complex and not
completely understood.  It is possible  that highly reactive molecular fragments
(free radicals, atoms, ions) produced in the plasma might recombine in a cooler
region of the torch to form environmentally undesirable products.  At the same
time, many of the hazardous chemicals to be destroyed are very stable.  For
these reasons, it is essential that the  gaseous products  vented to the
atmosphere from the exhaust stack of the facility be closely monitored to ensure
proper operation of the incinerator.
     The Unsteady Aerodynamics Laboratory (UAL), following a request from the
company, has undertaken to assist in the development of a trace gas analysis
system suitable for monitoring the concentration levels in the flue stack of the
plasma torch.  In this paper a design of the monitoring system is outlined, and
some preliminary work is described on the collection and  analysis of PCBs ,
2,3,7,8-tetrachlorodibenzodioxin  (TCDD)  and 2,3,7,8-tetrachlorodibenzofuran

2.1  Requirements
     Among the prerequisites for  a suitable analyzer in the plasma torch
scenario are the following:
(1)  high sensitivity, to allow the detection  of ppt-concentration levels;
(2)  qualitative reliability, or  specificity,  to ensure unambiguous
     identification of the target compounds;

(3)  quantitative accuracy, to meet environmental control standards;
(4)  versatility, to cover a broad range of vapours and gases;
(5)  computer-based, for unattended analysis as well as for possible feedback
     control of incinerator operation;
(6)  economical, in terms of capital outlay and operating cost.

2.2  Gas Chromatography/ECD/TFD/FID
     The detection levels mentioned in (1) dictate the use of a preconcentration
technique.  Preconcentration of a large air sample results unavoidably in the
collection of many vapours and gases in addition to those of interest and,
therefore, suggests the use of a separation technique, e.g. gas chromatography
(GC), to permit sensing of the target chemicals.  The GC sensor could take the
form of a class-specific detector, such as the electron-capture (ECD),
thermionic flame (TFD), or flame ionization (FID) detector.  However, although
highly responsive to certain types of compunds, these detectors may lack the
specificity and versatility to meet the requirements of (2) and (4).

2.3  Gas Chromatography/MS
     A more universal and positive detection technique can be achieved by the
use of a mass spectrometer (MS) as the GC sensor.  Compact, simplified MS
systems have recently become available for use in capillary GC.  As analytical
instrumentation this equipment has similar specifications to those of more
sophisticated mass spectrometers while being lower priced and is a priori the
preferred choice of GC detector over the ECD, TFD and FID with respect to
fulfilling the above prerequisites, especially items (2) and (4).

2.4  Preconcentration
     The method of trace vapour preconcentration developed previously at UAL is
considered to be adaptable to the present case.  In that approach, air to be
tested is drawn through an adsorbent-packed tube which collects or
preconcentrates the vapours of interest; after sampling, the vapours trapped in
the tube are thermally desorbed and transferred by a carrier stream to a second,
smaller adsorber tube from which they are subsequently desorbed and injected
into the GC column.
     The two-stage adsorber concept has been used with success to quantify ppt

levels of airborne vapours, including organophosphorus and carbamate
insecticides, chlorophenoxy and chlorobenzoic acid herbicides, as well as
organonitrate explosives  (1-5).  It has recently been extended for use with
capillary-column GC, and  tested in headspace sampling of high-molecular weight
hydrocarbons and fenitrothion (6).
     As currently implemented, the technique was designed to permit the
(detachable) first-stage  adsorber to be used for sampling at remote locations,
then returned and manually reinstalled in the analyzer unit.  To render the
technique suitable for automated sampling and analysis, as required for the  type
of fixed-installation monitoring of the plasma incinerator products, some
modification in instrumentation is required.

3.1  Capillary-column GC  Analysis
     Two GC instruments have been utilized, a Varian 1600/Vista U01 and an HP
5790A, each fitted for capillary column operation.
     The varian GC was equipped with an adsorber tube injector port, illustrated
in Fig. 1, as well as a regular septum inlet.  The column used in this
instrument was 30 m x 0-32 mm I.D. SPB-5 fused silica.  The column oven was
temperature programmed as follows:  initial temperature 80°, hold 10 min; 80° to
150° at 20°/min; 150° to  260° at Wmin, hold 2 min.  Helium carrier gas flow
velocity was 39 cm/sec.   Under these conditions good separation of the
individual PCBs in Aroclors 1242, 1251* and 1260 and of 2,3,7,8-TCDD and 2,3,7,8-
TCDF was achieved in approximately 30 min.  Both FID and ECD were used in the
PCB analysis.  The sensitivity of the FID allowed the analysis of microgram
quantities of the Aroclors, while with the ECD sub-nanogram samples were
     The HP 5790A GC was  fitted with a 12.5 m x 0.2 mm cross-linked dimethyl
silicone WCOT fused silica column.  The column oven was operated with the same
temperature programme as  the Varian GC.  Helium carrier gas velocity was 20
cm/sec.  Under these conditions chromatograms were similar to those obtained
with the Varian GC.  Splitless injection mode was used to inject samples.  The
GC was coupled with an HP 5970A Mass Selective Detector.  The detector was
operated in the Peakfinder programme to identify the peaks and in the Selective
Ion Monitor programme to  determine trace quantities of analytes.
     The GCs were calibrated by means of standard solutions of the Aroclors  in

iso-octane having concentrations ranging from 10~^ to 10~10 g/uL.  The
concentration of TCDD and TCDF standard solutions was 10~9 g/pL.  Calibrations
were made both through direct liquid injection and, in the case of the Varian
instrument, through deposition of the solution in the adsorber tube followed by
the appropriate desorption/injection procedure.

3.2  Adsorber Tube Sampling
     Pyrex adsorber tubes were 7.5 cm x 6.3 mm O.D. containing a 1 cm column of
Tenax GC 45/60 mesh adsorbent.  Tenax has been reported to be superior to
polyurethane foam, XAD-2 resin and Florisil as a sorbent  for collecting PCBs in
air sampling (7).  The thermal stability, hydrophobia properties and high
retention capacity of Tenax make it suitable for trapping PCBs and dioxins  from
large sample volumes of moisture-laden air, and subsequent recovery of the
target vapours through thermal desorption.
     The breakthrough volume of the sorbent plug was estimated by placing a
backup adsorber in series with the first and sampling a spiked air stream;  the
presence of PCB vapours in the backup adsorber for a measured volume of air
sample signifies breakthrough from the first tube.

3.3  PCB Vapour Source
     A continuous stream of PCB vapours in air was generated by passage of  a low
flow of N2 through a U-tube containing glass beads wet with Aroclors 1254 and
1260.  This vapour stream was mixed with a larger flow of air (8) to achieve a
controlled dilution ratio of the equilibrium vapour pressure of the PCBs in the
test stream.  With the U-tube thermostated at 0°C and a dilution ratio of 1/500,
PCB concentrations of the order of 100 ng/m3 were obtained.
     In sampling the test stream, adsorbers were maintained at room temperature,
or heated to 80°C to simulate the plasma stack temperatures.  At room
temperature (22°C) it is estimated that less than 5% of the total Arochlors in
30 L of air sampled at 0.5 L/min. escaped the first adsorber; when kept at  80°C,
the first adsorber trapped over 90% of the PCBs from a 20 L volume.
Chromatograms from some of these tests are shown in Fig.  2, obtained with the
Varian GC/ECD.
     In Fig. 2 differences between the vapour and liquid  signatures are evident,
and attest to the fact that the partial pressure of a particular component  in
the vapour phase may far exceed the mole fraction in solution.

     Using the ECD, the smallest mass of Aroclor mixture that can be measured
with S/N ^ 5 is about 0.5 ng.  With the MSD operated in the Selected Ion
Monitoring (SIM) mode and the electron multiplier voltage set at 1600V, the
smallest quantity of Aroclor that can be measured is about 1 ng.  Assuming a
breakthrough volume in sampling of not less than 20 L, the minimum detectable
concentration of PCBs measurable with the adsorber tubes is about 25-50 ng/m3.
By way of comparison, in a recent survey the atmospheric PCB background level  in
the province of Ontario was found to range from 0.01 to 1.4 ng/m3, averaging
about 0.20 ng/m3.
     The sampling system proposed in monitoring the plasma flue gases  is based
on a first-stage adsorbent bed of comparable dimensions to that tested above.

     A module has been designed and fabricated which interfaces with the Hewlett
Packard 5790A/5970A MSD system, and is presently being tested.
     The module is essentially an auxiliary oven supporting the first  and
second-stage adsorbers, and housing two six-port switching valves and  associated
plumbing.  Connection from the second adsorber to the GC is made through heated
capillary tubing.  The air sampling line is a length of heated stainless steel
tubing, 6.3 mm O.D., which is provided with an injection port for
calibration/test purposes, and with a (replaceable) filter to remove particulates
from the air sample.  Care has been taken to avoid cold spots in all vapour
transfer lines.  A schematic view of the sampler configuration is given in
Fig. 3.
     The twin-valve design shown is sufficiently flexible to allow for purging
of Ads 1 before transfer of the PCB vapours to Ads 2, by operating the valves
independently.  Valve actuators, adsorber heaters and the air pump are

     From the initial study carried out to date it is felt that a viable
monitoring system based on the sampler configuration and GC/MS approach outlined
in this report is feasible.
     In principle, the proposed system is useful for any vapour or gas that  is
amenable to GC analysis.  The preconcentrator component of the system, involving
the two-adsorber concept, is of proven efficacy in trace vapour detection, and

can, moreover, be tailored to the gases of interest through selection of

suitable adsorbent packings.  At the same time, the MS detector provides

complete versatility of detection.


     The authors thank Dr. Andre Lawrence of this laboratory for his valuable

assistance in the initial stage of the work.


1.   M. McCooeye, C. Cooke and L. Elias, March 1984. GC Analysis of Post-Spray
     Air Samples in Priceville Forest Field Study. NRC NAE LTR-UA-72.

2.   R.S. Crabbe, L. Elias and S.J. Davie, January 1983. Field Study of Effect
     of Atmospheric Stability on Target Deposition and Effective Swath Widths
     for Aerial Forest Sprays in New Brunswick. Part II.  NRC NAE LTR-UA-65.

3.   R.S. Crabbe, M. McCooeye and L. Elias, January 1984.  Effect of Atmospheric
     Stability on Wind Drift in Aerial Forest Spray Trials.  Neutral to Stable
     Conditions.  NRC NAE LTR-UA-73.

4.   R.S. Crabbe and M. McCooeye, March 1984.  Field Measurement of Ground
     Deposit and Windborne Drift from Herbicide Sprays in New Brunswick.  NRC
     NAE LTR-UA-72.

5.   L. Elias, January 1981.  Development of Portable GC Explosives Detector.
     NRC NAE LTR-UA-57.

6.   M.E. Krzymien, November 1983-  Dual Adsorber-Capillary Column System for
     Gas Chromatographic Analysis of Air Samples.  NAE-AN-20, NRC No. 22889;
     also unpublished data.

7.   W.N. Billing and T.F. Bidleman, 1981.  High Volume Collection of
     Chlorinated Hydrocarbons in Urban Air Using Three Solid Adsorbents.  Atmos.
     Env., Vol. 17 (1981), pp 383-391.

8.   M.E. Krzymien and L. Elias, 1976.  A Continuous-Flow Trace Vapour Source.
     J. Phys. E: Scient, Inst., Vol. 9 (1976), pp 584-586.

9.   E. Singer, T. Jarv and M. Sage, 1983-  Survey of Polychlorinated Biphenyls
     in Ambient Air Across the Province of Ontario. Physical Behaviour of PCBs
     in the Great Lakes (Papers Presented at a Meeting, 1981), pp 367-383
     (1983), Ann Arbor Sci.

                                                                  1 - SEPTUM RETAINING NUT
                                                                  2 - SEPTUM
                                                                  3- INJECTOR CAP
                                                                  4 - SILICONE RUBBER O-RING
                                                                  5 - BAYONET COUPLING
                                                                  6- FIRST ADSORBER GLASS
                                                                  7- GLASS WOOL PLUG
                                                                  8 - SOLID SO R BE NT
                                                                  9 - CARRIER GAS (He) INLET
                                                                  10 - 1/4 INCH SWAGELOK NUT
                                                                  11 - GRAPHITE FILLED VESPEL
                                                                      REDUCING FERRULE (1/4
                                                                      TO 1/8 INCH)
                                                                  12 - SOLENOID VALVE
                                                                  13- SOLENOID VALVE
                                                                  14- SPLIT VALVE
                                                                  15 - 1/16 SWAGELOK FITTING
                                                                      WHERE THE CAPILLARY
                                                                      COLUMN IS ATTACHED
                                                                  16- STAINLESS STEEL TUBES
                                                                      HOUSING CARTRIDGE
                                                                      HEATERS AND PLATINUM
                                                                      TEMPERATURE SENSOR
                                                                  17 - SECOND ADSORBER NICKEL
                                                                  18 - BAKELITE INSULATOR
                                     FIG. 1:  DUAL TRAP

                           (•) BACK-UP ADSORBER. HI ADSORBER
                             AT 22°C. 30L VAPOUR SAMPLE
                                           (b) BACK-UP ADSORBER, HI ADSORBER
                                             AT aO°C. 20L VAPOUR SAMPLE
                  (c) VAPOUR COLLECTED ON
                    TENAX QC ADSORBER AT
                    SO°C. 20L VAPOUR SAMPLE
                          LIQUID INJECTION
                           1.5 ng OF MIXTURE:
                           20% AROCHLOR '254
                           20% AROCHLOR 1260
                           «0% TRICHLOROBENZENE

               AIR SAMPLE
             ADS I
             ADS I
                   SAMPLE TRANSFER
             ADS I                   ADS 2
          C1, C2 - CARRIER GAS; ADS 1, ADS 2 - FIRST- AND

           Jerry D. White, Charles K. McMahon, and Hilliard L. Gibbs
                        Southern Forest Fire Laboratory
                    Southeastern Forest Experiment Station
                               Route 1, Box 182A
                           Dry Branch, Georgia 31020
     Although forest burning is prescribed widely across the United States, it
is most commonly practiced in the Northwestern and the Southern United
       1 2
States. '   In 1978, approximately 37 million metric tons of forest fuels on
all forest ownerships were burned by prescription; approximately  12.5 million
metric tons were burned in the South.   This burning produces an  estimated 0.6
million metric tons of total suspended particulate matter (TSP) annually in the
United States.  Of that total, about 0.2 million metric tons of TSP originate
in the South.
     Considerable uncertainty exists over the estimation of benzo(a)pyrene
(BaP) produced by prescribed burning.  Forest and agricultural burning were
estimated by the National Academy of Sciences to emit  127 metric  tons per year
in 1968, but that figure was reduced to 9.5 metric tons per year  in 1972.   In
a 1977 report, EPA estimated BaP emissions from prescribed burning to be 4.5
metric tons nationally, which was 0.5 percent of the BaP from all sources.
     Early data gathered by the Southern Forest Fire Laboratory suggested that
the amount of BaP emitted might also vary with the fuel condition and method of
burning.  In a series of experimental fires conducted  by McMahon  and Tsoukalas,
the ratio of BaP to TSP was found to be much higher among simulated backing
fires than among simulated heading fires in pine needle fuel  (Table 1).   The
measurements were made in a special combustion chamber at the Southern Forest
Fire Laboratory (Figure 1).  Backing fires are spreading fires that progress
into the wind and heading fires are those that progress with  the  wind.  Both
types of fires are commonly prescribed.

     A serious limitation in these results was that they represented only one
fuel type burned by prescription in the South.  Perhaps more important,  they
represented a fire environment in which pine needles were isolated from  all
natural variations in conditions of duff, soil, moisture, and wind.  Questions
were raised.  Are the order of magnitude differences between BaP/TSP ratios
from backing and heading fires seen in laboratory fires also characteristic of
fires in natural forest settings?  What is the range of BaP/TSP ratios for some
other fuels commonly prescribed burned in the South?  The study described here
was directed toward these questions.
                              Experimental Design
     In the forest, several factors, which can be selected or measured prior to
a prescribed burn, are believed to influence BaP and TSP production.  These
factors fall into two broad categories: fuel conditions and weather
conditions.  The fuel conditions are fuel type, fuel load, and moisture
content.  Weather conditions are fire type (or wind direction), wind velocity,
and relative humidity.
     In this field experiment, we examined the effect of fire type and fuel
type only.  For comparison with the laboratory experiment, we incorporated
three levels of fuel loading for one fuel type—pine needle litter.  The
statistical design chosen was a factorial experiment (2 fire types x 4 fuel
types) with an unbalanced incidence matrix.  The two fire types examined were
backing and heading fires.  The four fuel types examined were pine needles
(litter of pure slash pine needles), hardwood litter (mixed hardwood leaves and
pine needles), broomsedge (pine-needle litter with broomsedge understory), and
palmetto (pine-needle litter with palmetto understory).
     The plots burned in each fuel type were approximately 5m by 25m.  With the
exception of the pine-needle fuel, 6 plots were burned in each fuel type—3
replicate backing fires and 3 replicate heading fires.  In the case of the
pine-needle fuel, each of 3 levels of loading was treated separately, giving a
subtotal of 18 pine needle fires.  The statistical analysis was appropriately
adjusted for this unbalanced incidence of fuel types.  In all, 36 individual
plots were burned and sampled.  Results were subjected to statistical analysis
of variance.

                                  The Sampler
     A light, portable sampler was designed to collect, simultaneously, samples
of total suspended particulate matter, benzo(a)pyrene, and combustion gases.
The sampling train consisted of five units: a glass-fiber filter holder, a
polyurethane foam (PUF) trap, two personal pumps, and a gas bag (Figure 2).
The sampler, assembled from components and attached to a long aluminum pole,
was designed for portability and safety.  On one end was an air-intake probe
which could be extended into the smoke plume directly over the flaming zone.
The probe was positioned by the operator so that its temperature rarely
exceeded 60 C by raising or lowering the probe above the flames.  On the other
end of the pole were the pumps and other electrical components which were less
resistant to heat.  The person who carried the sampler could walk near the
advancing fire line holding the probe within the plume of emissions directly
above the flames.
     The sampling probe (Figure 3), consisted of four main parts: filter
holder, PUF trap, thermocouple and anemometer.  The open-faced aluminum filter
holder contained a 47-mm glass-fiber filter.  The exit of the holder fed
directly into a PUF trap constructed of PVC pipe and end caps.  The trapping
material was polyurethane foam in 3 cylindrical plugs (30-mm diameter by 35-mm
length") prepared in advance by soxhlet extraction with methylene chloride.  We
expected most, if not  all, BaP to be trapped on the glass-fiber filter; but to
be safe, we placed the PUF plug into the sampling train to trap any BaP in the
vapor phase as well.  '   The filter holder and PUF trap were attached by a
"quick-disconnect" to  an extension tube running to the pumps.  A thermocouple
was placed on the probe very near the entrance to the filter holder with a
temperature readout in sight of the person using the  sampler.  The thermocouple
was used to monitor the temperature of gases entering the sampling probe.  A
Biram anemometer, not  used in this study, was located adjacent to the sampling
probe and could be used to determine average windspeed flowing by the sampler.
     On the other end  of the sampler (Figure 4), a Dupont P-4000 pump, powered
the probe's air flow with a flow rate of 4.0 liters per minute.  A smaller
pump, a Dupont P-200,  pulled a constant proportion of the exhaust gases from
the main pump into a 2.5 liter aluminized gas bag at  a flow rate of 0.12 liter
per minute.  Thermocouple and time readouts were also located here.  About  1  to
5 mg of TSP and 1 to 2 liters of gases were collected for subsequent analysis.

     TSP was determined gravimetrically while the concentrations of carbon
monoxide (CO) and carbon dioxide (C0_) were determined by a nondispersive
infrared technique.  CO and CCL values, while not used in this study,  could be
used to estimate emission factors by the carbon balance technique as  reported
by Ward, et al.   BaP trapped in the TSP and PUF was determined by a  routine
method validated for wood smoke at the Southern Forest Fire Laboratory.   In
this method, BaP was recovered from the TSP and PUF by soxhlet extraction with
methylene chloride and quantified via high performance liquid chromatography on
a bonded octadecyl column.  The limit of detection was about 1.34 ng  BaP and
the limit of quantitation was about 2.02 ng.  A precision of better than 10%
was typical at the BaP levels determined.
     Benzo(a)pyrene appeared to be trapped completely by the sampler's
glass-fiber filter.  In only one PUF analysis out of 12 did BaP exceed the
limit of quantitation (2.02 ng).  And this one case was thought to be  due to
leakage of TSP rather than breakthrough of BaP.  In separate tests, samples of
TSP were held for BaP analysis for at least 4 months under refrigeration
without significant degradation.
     For the pine-needle fuels, the trends of the ratios of BaP to TSP in the
field were similar to trends reported in the laboratory (Table 1).  For
example, backing fires produced higher ratios than heading fires, except for
backing fires with heavy fuel loads.  Also, ratios decreased with increasing
loading of needles.  In the field, however, the ranges in observed values (7 to
45 yg per gram) were far less than in the laboratory (2 to 274 ug per  gram).
The unusually high values of ratios for laboratory backing fires were, we
believe, because conditions for pyrosynthesis of BaP were more favorable in
these fires.
     What are the conditions that influence formation of BaP during prescribed
burning?  Strong experimental evidence suggests that BaP pyrosynthesis within
the flame envelope is governed by temperature, oxygen concentration,  and length
of time BaP precursors remain inside the flame.  '    If the flames are too hot
(above 1000 C) and turbulent, BaP levels are low because oxidation is  favored
over pyrosynthesis.  On the other hand, if temperatures are low (below 600°C),
as often occurs in smoldering combustion, the BaP precursors do not cyclize to
the 5-ring BaP structure.  The optimum temperature for BaP pyrosynthesis

is near 800 C.   '    In prescribed burning, our evidence  suggests  that  the
conditions that  favor BaP pyrosynthesis are low-intensity fires  in which
flaming combustion predominates over  smoldering combustion.  These conditions
are produced in  light fuel loadings that burn with relatively nonturbulent
     When BaP/TSP ratios were listed  by fire type and  fuel type, the  range  of
values was less  than an order of magnitude (Table 2).  Over  the  four  fuel
types, no significant difference was  found between ratios from backing  and
heading fires.   A mean of 23 yg per gram for backing fires and 25  yg  per gram
for heading fires showed this clearly.  However, there was a significant
difference among mean values by fuel  type.  We cannot  explain the  variation
among fuel types at this time.  However, we believe that  it  is caused by a
combination of fuel characteristics such as fraction of green fuels,  fuel bed
porosity, fuel loading, and chemical  composition, which contribute to fire
behavior factors such as reaction intensity and fire line intensity.
Additional work  is planned.
     The mean BaP/TSP ratio of all 36 fires in the experiment was  24 yg per
gram with a relative standard deviation of 0.47.   In  another study by Ward, et
al.   currently  in progress in the Pacific Northwest,  a BaP/TSP  ratio of
15 yg per gram with a relative standard deviation of 0.49  has been determined
from 27 TSP samples.  These samples were obtained from burning unpiled  forest
residues (slash  burning).  Applying these new ratios (24  and 15) to the fuel
and TSP data available from Chi, et al. , we calculate a  new annual BaP
production of 11 metric tons for prescribed burning in the United  States.
Although this new value is still an approximation, we  believe it to be accurate
within a factor  of two and a significant improvement over  previous  estimates
because of the new information available on BaP/TSP ratios.
1.  The ratio of BaP/TSP averaged 24  yg per gram with  a relative standard
deviation of 0.47 in four forest fuels common to the Southeast.
2.  Significant  differences were not  found between heading and backing fire
types, but were  found amongst the fuel types.
3.  BaP production from prescribed burning is estimated to be 11 metric tons


1.  Southern Forest Fire Laboratory Personnel, 1976. Southern Forest Smoke
    Management Guidebook.  Gen. Tech. Rep. SE-10.  U.S. Department of
    Agriculture, Forest Service, Southeastern Forest Experiment Station,
    Asheville, NC, 140pp.

2.  Johnson, V. J., 1984.  Prescribed burning: Requiem or renaissance?
    J. For. 82:2,  pp82-90.

3.  Chi, C.; D. Horn; R. Reznik; D. Zanders; R. Opferkuch; J.Nyers;
    J. Pierovich;  L. Lavdas; C. McMahon; R. Nelson; R. Johansen; P. Ryan,
    1979. Source Assessment: Prescribed Burning, State of the Art. EPA (U.S.)
    Report EPA-600/l-79-019h, Research Triangle Park, NC, 107pp.

4.  McMahon, C. K.; S. N. Tsoukalas, 1978. Polynuclear aromatic hydrocarbons
    in forest fire smoke. In: Jones, P. W. and R. I. Freudenthal, eds.
    Carcinogenesis, Vol. 3: Polynuclear Aromatic Hydrocarbons.  Raven Press,
    New York, NY,  pp61-73.

5.  Eimutis, E. C.; R. P. Quill, 1977.  Source Assessment: Noncriteria
    Pollutant Emissions.  EPA (U. S.) Report EPA-600/2-77-107e, Research
    Triangle Park, NC, 99pp.

6.  Thrane, K. E.; A. Mikalsen, 1981.  High volume sampling of airborne
    polycyclic aromatic hydrocarbons using glass fibre filters and
    polyurethane foam.  Atmospheric Environment 15:6, pp909-918.

7.  Yamasaki, H.; K. Kuwata; H. Miyamoto, 1982.  Effects of ambient temperature
    on aspects of  airborne polycyclic aromatic hydrocarbons.  Environ. Sci.
    Technol. 16:4, pp!89-194.

8.  Ward, D. E.; D. V. Sandberg; R. D. Ottmar; J. A. Anderson; G. C. Hofer;
    C. K. Fitzsimmons, 1982.  Measurement of smoke from two prescribed fires in
    the Pacific Northwest.  Presented at the 75th Annual Meeting of the Air
    Pollution Control Association,  New Orleans, LA.

9.  White, J. D.,  1984.  A simplified determination of benzo(a)pyrene in
    particulate matter from prescribed burning.  (Submitted to Am. Ind. Hyg.
    Assoc. J.)
10. Badger, G. M.;  R. W. L. Kimber;  J. Novotny, 1964.
    aromatic hydrocarbons at high temperatures. XXI.
    M-butylbenzene  over a range of temperatures from
    intervals.  Aust. J. Chem. 17, pp778-786.
  The formation of
 The pyrolysis of
300 to 900 C at 50°C
11. Crittenden, B. D.;  R. Long, 1976. The mechanisms of formation of
    polynuclear aromatic compounds in combustion systems. In: Freudenthal,
    R. I.; P. W. Jones, eds. Carcinogenesis, Vol. I, Polynuclear Aromatic
    Hydrocarbons.  Raven Press, New York, NY, pp209-223.

12. Schmeltz, I.;  D. Hoffman, 1976.  Formation of polynuclear aromatic
    hydrocarbons from combustion of organic matter. In: Freudenthal, R. I.;
    P. W. Jones, eds. Carcinogenesis,  Vol. I, Polynuclear Aromatic
    Hydrocarbons.   Raven Press, New York, NY, pp225-239.

13. Commins, B. T.,  1969.  Formation of polycyclic aromatic hydrocarbons
    during pyrolysis and combustion of hydrocarbons. Atmos. Environ. 3, pp565.

14. Ward, Darold E.; Colin C. Hardy, 1984. Advances in the characterization and
    control of emissions from prescribed fires.  Presented at the 77th Annual
    Meeting of the Air Pollution Control Association, San Francisco, CA.

Figure 1.   Bed of pine needles
           burning in combustion
           chamber at Southern Forest
           Fire Laboratory.
Figure 2.   Portable smoke sampler.
Figure 3.  Portable smoke sampler,
           probe unit.
Figure 4.  Portable smoke sampler,
           control unit.

Fire type and                 Laboratory  fires                   Field  fires
fuel loading                 Fuel load*   Ratio               Fuel  load    Ratio
	kg/m2	yg/g	kg/m2      ug/g

Backing fires
   Light load                  0.5          274**                 0.3         45
   Median load                 1.5          135                   1.3         14
   Heavy load                  2.4          98                   1.6         _1_

   Mean value                  1.5          169                   1.1         22
Heading fires
   Light load                   0.5            3                   0.3         24
   Median load                  1.5            2                   1.3         11
   Heavy load                   2.4            2                   1.5         11

   Mean value                   1.5            2                   1.0         15
*   In the laboratory,  fuel  load  referred  to  kilograms  per  square  meter  of  pine
    needles placed on  the  burning rack;  however,  in  the field,  fuel  load
    referred  to  the difference  in the  average kilograms per square meter of
    pine needle  litter  (6  replicates)  before  and  after  the  fire.

**  Ratio values  for laboratory fires  which were  taken  from reference  4  were
    recalculated.  The  correct  ratio for the  backing fire with  light load should
    be 318 ug per gram  and not  274 pg  per  gram.

Fuel types
Fire Types
Backing fires

Pine needle
Heading fires
Ratio Number
Mg/g of
15 9
13 3
13 3
60 3
*  Ratios are reported in micrograms of benzo(a)pyrene per gram of total
   suspended particulate matter.

** Relative standard deviation (coefficient of variation) is the ratio of  the
   standard deviation of the replicates to the mean.

                               DEVELOPMENT AT MRI
                                 Fred J.  Bergman
                           Midwest Research Institute
     The purpose of  this  presentation is to describe volatile organic sampling
train (VOST) technology presently in use at  Midwest  Research  Institute  (MRI).
We  hope  this information  will  help you avoid  many  of  the problems MRI  en-
countered when it first sampled for volatile organics using Tenax traps.
     Early in 1983 MRI received a task on an EPA Office of Toxic Substances pro-
gram to  evaluate  the emissions from  hazardous  waste  incinerators.  We started
the  program  using a commercial conditioner,  desorber,  and Tenax traps.    The
Tenax in the trap (Figure 1) was held in place with plugs of glass wool,  and the
traps were  stored  in test tubes with Teflon-lined  caps.   The traps were con-
nected to the sampling system using Swagelok fittings with Teflon front and back
ferrules.  A  laboratory evaluation  for  compound retention  was  performed  and  the
system appeared  to  be working.   After our  first field  test,  however, we  found
that the  blanks  contained almost the same  levels  of volatile organics as our
samples.  A frantic  search was initiated to eliminate the problem while the test
program  continued.   We employed  a multi-approach  attack,  obtained very  low
blanks, and adopted  a procedure which is still  being  used.
     To  our  knowledge,  MRI has had more experience with the VOST as applied to
source measurements  than  any other organization.   Because  this  experience has
yielded  additional  information,  we have developed  what we believe  is  a  better
understanding of  the VOST's problems.  We  now  know,  for example,  that many  of
the  steps,  originally  incorporated in the  procedure  to eliminate the  high
blanks,  are  unnecessary.   The major  difficulties with  using the VOST will be
getting  the cartridges clean, knowing  when  they are clean, and  keeping  them
     The  successful  use of  the VOST  requires  good cartridge  design.  We designed
the  MRI  double-walled cartridge (Figure 2) to  minimize the  contamination  which
we  felt  was coming  from  outside  the  tube.   Tests show however  that  even the

double-walled cartridge does not completely protect the cartridge from contami-
nation.   With this knowledge,  we are investigating a new simplified design (Fig-
ure 3).
     Our first VOST runs were made on incinerators with wet scrubbers.  We were
having difficulty during our analysis because there was high water retention  in
the cartridges.   We found that most of the water was contained in the glass wool
used to  retain the Tenax.   To  solve this problem,  a system of  C clips and stain-
less steel  screens was developed to hold the Tenax in the  traps.   This retention
system kept  the  Tenax  compressed,  with  a surprising but now understandable im-
provement  in  cartridge performance.   Keeping the  Tenax compressed  eliminated
voids and  channeling  in  the resin bed.   The  result  was  cartridges which were
more uniform  in  their  compound  retention and which were significantly improved
in their retention capacity.
     Another way we modified  the  original Tenax tube system was the  method of
connecting the tubes  to  the system.   We  found that the Swagelok fittings with
Teflon front  and  back  ferrules  we originally used  frequently  broke  the tubes
when tightened sufficiently to obtain a  leak-free system.   We  therefore used the
end plates  in the  double-walled design  to obtain a seal  with  the tubes and at-
tached the end plates  to the system using VCO fittings.
     In  the  new  tube design, we are  attempting to  use Ultra-Torr fittings which
use an 0 ring.  While  we were  trying to  reduce the blanks, we  placed a cartridge
filled with  the  Viton  0 rings and desorbed  them  into the mass  spectrometer.
High levels  of hydrocarbons were detected.  We decided to condition  the 0  rings
in a vacuum  oven to remove the volatile  material.   When  the  first batch of 0
rings was  removed from  the  oven, about  half  had turned to black  glass,  indicat-
ing that the  vendor had mixed Viton  and  rubber 0 rings together.  We  changed  to
a  vendor that color-coded  its 0 rings (Viton is tan).  Because we have not yet
checked volatiles run  with  the  new 0 rings,  to  be safe we are  continuing to
treat the  new rings.  It appears that we may be able to employ Teflon 0 rings in
the Ultra-Torr fittings in place of Viton for even better performance.
     Let us  summarize  the  advantages and disadvantages of the two MRI tube de-
signs and  the original  all-glass commercial  tube design.  The  MRI double-walled
design gives reproducible results, provides low water retention, is rugged, pro-
tects the  exterior  of  the  tube from  contamination, goes  in the  system  one way

only, and does work.   It  is,  however,  heavy,  large, expensive  to  construct,  and
requires more time  to  recycle.   (The  cartridges  are  disassembled and  only  the
inside tube  is  placed  in the desorber,  so  the  cartridge  must  be  reassembled.)
The  new  simplified  MRI tube design will  have  the same advantages as those of
the  double-walled  design but  have the  added  advantages of  being smaller,
lighter, and  lower  in  cost.   It will  probably  require some  type  of outside  pro-
tection in the field such as being wrapped  in aluminum foil.  A system to assure
proper orientation  of  inlet to inlet  will  also be  desirable.   It is currently
untried.  The original  commercial   all-glass tube design using C clips and stain-
less  stell  screen  in place of the glass wool should  perform as well as the new
simplified MRI  design, and  the different sized ends will  assure  proper orienta-
tion.   Replacement  glass  tubes will  be  more  expensive because one end must be
drawn down,  but  the sampling train (metal  parts) will cost less  because of the
smaller fittings on one tube end.   The small tube end  will also be more fragile.
      A fourth tube design (Figure  4) has been proposed in the VOST protocol.  We
believe  this  new commercial tube  design  is unacceptable.   Because the tube is
necked  down  to  1/4 in. on  each end, it  is  necessary  to hold the  Tenax  in place
with  glass wool  plugs, with the attendant disadvantages.
      It  has  been our  experience that  for  good conditioning,  especially in  the
case  of  used cartridges,  it is essential that the gas be forced to pass through
the  cartridges.   In the commercial unit that  is presently available  for car-
tridge  conditioning,  the gas does not normally pass  through the  cartridges  but
flows around the tube.  We  understand that the manufacturer  has developed a
modification  which  forces the gas  through the cartridges.   If you decide  to use
this  conditioner,  we strongly recommend  that you use  the cartridge flow-through
modification.  We have found  that  many cartridges that have been  used  still  fail
the  purity  check after two 8-hr conditioning  periods using the  unmodified  sys-
tem.  These  same cartridges will  clean  up, however,  after only 4 to 8 hr when
the  purge gas goes  through  the cartridge.   We  have  been  informed  that  some  users
have solved  the cleanup problem by  discarding the approximately  $10  worth of
Tenax in- each cartridge  after  it has  been  used  once,  an  approach  you may wish to
      We  recommend  adopting the one-step  conditioning and monitoring technique
(Figure  5)  regardless  of  which cartridge design is  used.  The  one-step  procedure

consists of passing hydrocarbon-free nitrogen at 30 mL/min through the cartridge
while it  is  heated  at  200°C.  The exit  gas  stream  from  every  tube  is  checked  at
regular intervals using  a  flame  ionization  detector  (FID)  until  the hydrocarbon
level approaches the lower detection limit (LDL).   Using the one-step technique,
you  know  when  the cartridges  are clean.   This permits stopping  the conditioning
when most  of the cartridges pass the purity  check.   In  addition, all  components
of the  cartridge can  be  cleaned (conditioned)  at  the same  time.   This  has  the
added advantage  of  not  requiring stringently  clean  facilities  for cartridge
assembly.   If  you elect  to condition and perform  the purity  check separately,
you will find  it necessary to recycle cartridges through the desorber until they
pass the purity check.
     A manifold for conditioning the new tube design  is  shown in Figure 6.  Each
manifold  holds 10 tubes  in the  conditioning oven, so that  with  four  manifolds
40 tubes can be conditioned at a time.
     We do not agree  with the introduction  of  a chromatographic column during
the  cleanup  procedure, as  proposed  in the VOST protocol.   We do not see how it
improves  the method, and it has  a number  of  disadvantages.  Use  of a  column in-
creases the  time required  for each  purity measurement from  the  present  2  min  to
at least  30  min.  Use  of a column also  significantly reduces  the sensitivity  of
the  measurement.  When using  an FID, the hydrocarbon response  is  additive so
that the  lower detection limit  is   the  sum  of  all  components eluted.    If  the
cartridge contains,  for example, 15  components just below the 0.2 ng level, they
would pass using the  column  technique.   With the  FID only, the  response  would
equal 3 ng and would fail to pass.   Another disadvantage concerns column contam-
ination.  If you use a column when  checking cartridges  that have collected  field
samples, you will find that high boiling  hydrocarbons are  slowly accumulated  in
the  column.  This will  cause  an increase in the background hydrocarbon level.
It will,  therefore, be necessary to stop and bake out  or replace the column at
frequent intervals.
     The  only  method we  have  found  that  protects  clean cartridges is to  store
them over activated charcoal  or under water.  Tests have demonstrated repeatedly
that neither the double-walled MRI  cartridges nor the all-glass cartridges  stored
in a Teflon-capped  test  tube  will   remain uncontaminated.   After conditioning,

the cartridges should  be  placed under water or  over charcoal  as soon as they
reach room temperature.   As  an  added  precaution, we  maintain the  purge  gas  flow
on the cartridges until they are cool.
     As indicated  in  the  VOST protocol, and as  has  been  done  by  users  of all-
glass tubes  in the  past,  the tubes  are  capped  and  placed  in Teflon-lined screw-
capped test  tubes  after sampling for shipping and storage.   We question this
procedure.   Any  contamination on the  exterior  of the Tenax tubes,  if  carried  to
the inside of the test tubes, will  migrate to the Tenax.
     Since we found that  volatile organics diffused  through the 0  ring  seals  on
the double-walled  design,  we also believe that storing  the tubes  after  sampling
in screw-capped test tubes over charcoal should be avoi'ded.
     The original MRI train  is described in the VOST protocol.   The train is be-
ing improved and modified to  use the  new MRI cartridge and the type of lubricant
free valve required by EPA  (Figure 7).   The valve manufacturer  recommends  the
use of a  small  amount of  lubricant to maintain  leak-free operation.  We deter-
mined  modest amounts of  Apiezon grease in  hydrocarbon sampling  trains, will
neither add  to  nor remove measurable quantities of  hydrocarbons from the  gas
stream.  However, the use of  greases  has been  forbidden by EPA.
     The addition of a third  valve  to the train will permit carrying  out all the
required operations without  having  to remove or  replace various components.  The
valves may be arranged in  such a manner that only  one valve is in the system dur-
ing the leak check.
     In summary,  MRI  believes the  VOST procedure  can be simple and  straight-
forward, as  shown  in the  following  steps:

          Clean  new metal  parts with  suitable  solvent.

          Sonicate  all components in  hot detergent solution.

          Rinse with water and  oven-dry.

          Assemble  cartridges in clean  area.

          Condition at 200°C with 30  mL/min of hydrocarbon free gas.

          Check  exit  gas  of  each  tube  at  intervals  using  FID.

          Stop  conditioning  when hydrocarbon level  approaches  lower detection

          For cartridges  that do not pass after 8 hr, fill with fresh Tenax and

          Cool with gas flowing,  cap,  and store over activated  charcoal.

          Protect outside of tube while  sampling with aluminum  foil.

          After  sampling, cap and store  tubes  under water until  analyzed.

If benzene or toluene is  to  be measured,  the cartridges  should  be conditioned as
close as possible  to the sampling time and analyzed as soon as possible.  Cart-
ridges for benzene or toluene should be  stored  under ice water after condition-
ing and until analyzed.  For samples not requiring benzene or toluene analysis,
storage over charcoal after conditioning and under water after sampling should
be sufficient.    However,  it would probably be  a good  idea to  keep cartridges
cold when possible.
     Our presentation has been  limited  to the  sampling  train.   Our analytical
procedure remains  basically the  same as  reported to  the  contractor who prepared
the protocol.   The one  exception is that we  have  discontinued using the com-
mercial desorber.   In  its place we are  connecting inlet and outlet fittings
directly to  the Tenax  tubes (Figure 8)  so that the purge gas must pass through
the tube.  The  tube  is  then heated by a small  resistance  heater  placed  around
the glass tube.   Using this system  we have  decreased the repeatability of stan-
dardization  from 10 to 2%.
     In conclusion,  I would like to acknowledge the  following  MRI  personnel  who
made  significant contribution  during  this work:   Paul Gorman, Greg Jungclaus,
Gil Radolovich,  George Scheil, Bob  Stultz, George Vaughn, and Ken Wilcox.

                      Gloss Wool
                      Glass Wool
                                                          2 Layers ~*~
                                                          S.S. Screen
                                                                             VCO Fittings
                                                                            VCO Fittings
Figure 1.   Original  commercial trap.      Figure  2.   MRI double-walled cartridge.
                                                    '"C" Clip
        2 Layers
        S.S. Screen
                          1/2 of Ultra-
                          Torr Fitting
"C" Clip
                       Tenax TA
                         1/2 of Ulfra-
                         Torr Fitting
                                                                          Gloss Wool
                                                                          Glass Wool
Figure 3.   New simplified MRI  trap.           Figure 4.   New commercial  trap.

N2 Gas from
Liquid  N2
            Conditioning |
            Oven 200 °C I
                                           • Flow Restrictor in
                                           Place of Column
                                             10'of 1/16 OD
                                           S.S. Tubing
                                     Heated Gas


1 3 Detec
[ Oven 200 -°C

tor |

                                                  Gas Chromatograoh with FID
  Figure 5.   One-step  conditioning  and monitoring technique.
                                Brass Tube
           1/2 Ulrra-
           Torr Fitting
                                                     Gas Outlet
                                                     to FID
          Figure 6.   New MRI  conditioning  manifold.


            Ultra-Torr Union j	j


            Ultra-Torr Union
                                                                     I	'—"-To Pump
                 Figure 7.   New VOST  sampling  train.
                                  Resistance Heater Coil
                                  Power Supply
                            Figure  8.    New  desorber

                                George W.  Scheil
                           Midwest Research Institute
                          Kansas City, Missouri   64110
     This presentation describes a project conducted by Midwest Research Insti-
tute (MRI) and  funded  by the Environmental Protection Agency (EPA) to provide
background information  for the development of performance  specifications  for
continuous analyzers for hazardous organic pollutants.   The project has two main
purposes:  to assess the state of the art in  continuous  monitoring for vinyl
chloride (VC) and  to measure the actual  performance of two different analyzers
over a period of 6 months.
     The test design (Figure 1) has the analyzers connected with a calibrator to
substitute calibration gases of 0, 5, and 9 ppm VC  for the sample  gas  once each
day and  a  digital  data logger designed by MRI to provide  a phone modem link to
transmit data back  to  MRI on request, with backup records on a printed log and
magnetic tape.   The system  operates  unattended;  only a twice-monthly supply
visit is made  unless  the phone link checks indicate the need for a repair trip
to the site.

     One of  the  analyzers  to be tested is an existing  EPA  furnished process  gas
chromatograph (GC) (Applied Automation).   This unit has a  30-cm backflush column
of 1/16-in.  stainless  steel  tubing packed with  n-octane  Durapak  for removing
heavy organics,  followed by a  30-cm  analytical  column of 1/16-in. stainless
steel tubing packed  with Porasil  C to separate VC from other light organics; a
flame ionization  detector (FID);  and an analog  preset time window integrator.
The system  also  has  a  sample conditioning  section,  gas sampling valve, and con-
trol system for automatic, repetitive sampling and analysis.

     A review of  currently  available instruments was conducted  to  select the
second analyzer.  Several techniques are  commercially used  for continuous moni-
toring of VC such as gas chromatography,  infrared, and electrochemical  sensors.
Only gas chromatography  has  the necessary sensitivity and selectivity for mea-
suring VC near  the  current  10 ppm standard  in process streams in the presence
of significant  amounts  of  ethylene dichloride and other  organics.   A growing
number of VC analyzers  are using photoionization detectors  (PID)  in  place of  an
FID with some use  of electron capture detectors.  Digital integrators are also
gradually replacing analog systems for peak analysis with occasional  use of mea-
surements for peak height instead of peak area.
     A process  GC with  a PID and a digital integrator was selected as the best
choice for  the  second  analyzer.   The PID has potentially better sensitivity to
VC than  an  FID  and  little response  to  potential  interferences  such as ethyl
chloride.  The  PID,  as  opposed  to  an FID,  requires no special supply gases, and
the PID  is  not  as easily poisoned  as an electron capture  detector.   Peak  height
measurements  suffer  from nonlinearity  problems  whereas  digital  integration
matches the growing use of microprocessor controlled analyzers.
     More than  half  the cost of purchasing  a second process GC is  needed  to
duplicate the valves, column, and other basic hardware to support a detector and
integrator.   Since the PID is a nondestructive detector,  an in-line PID could be
added to the existing system.  A simple switching relay allows either integrator
to be selected.   Adding a PID and digital  integrator to the existing system thus
saves considerable money  and has the added  advantage of  allowing better dis-
crimination of any differences between the types  of detectors and integrators by
having the control system alternate each integrator with each detector.   Thus, a
revised  analyzer  system was assembled  (Figure 2).   The  added modules were a
Model PI52 PID  with  a 10.2 eV lamp (HNU Systems, Inc.) and  a  Model BC-2 instru-
ment control computer (Action Instruments).

     Before proceeding  to the field  test,  a  series of experiments was completed
in the laboratory to determine  optimum  operating conditions  and possible  inter-
ferences, and a matrix  test  of the  analyzer  control  variables  was conducted.
The typical  VC reactor product gas contains significant concentrations of ethyl-
ene and  ethylene  dichloride,  as well as VC,  and  smaller amounts of other light

chlorinated hydrocarbons.  Small amounts  of  chlorine and hydrochloric acid are
also present which required the replacement of all  stainless steel  in the sample
conditioning system with  Monel,  Teflon,  or Knyar parts.  Test mixtures of the
compounds (Table 1) were  prepared  with  permeation  tubes at about  10 ppm with
similar concentrations of VC and were sampled by the process GC.  The backflush
column rejected  most  of  the  compounds,  and  the positive interferences had
shorter retention times  than VC with at least partial separation.  Although the
FID response to  chloromethane  was  similar to that to VC, the PID response was
less than 1% of the equivalent VC response.
     During preliminary  testing the BC-2 digital  integrator  system  developed
severe problems  due to a  lack of isolation from electrical  noise.  The computer
was having  nonrecoverable  system  crashes  about once an  hour  and the probable
cost of  remedial  action  was  excessive.   Fortunately the MRI data logger, based
upon an  Epson  HX-20  briefcase computer matched with a Wintek MCS analog inter-
face,  was functioning well  in the same environment.   The excellent line isola-
tion of  a  Nicad-powered  computer together with the optical  isolation and reset
capabilities of  the Wintek  system allowed reliable  recovery  from  noise.   The
logger system  could measure  the detector signals with the addition of a simple
amplifier and  had  sufficient  idle  time during each  analysis  cycle to perform
the necessary  peak integration.  Therefore, the logger was  reprogrammed  to per-
form the digital  integration task  as well as  to log  the  results  and  communicate
with MRI.
     The reliability of  this modified system  proved  satisfactory and  the matrix
test of  control  variables was conducted.   The variables can be  separated  into
variables affecting the  entire  analyzer  (Table 2)  and variables affecting only
part of  the system (Table 3).   Each variable  was tested  by  measuring  a VC  stan-
dard gas at the  optimum  condition  and then at reasonable steps  higher and  lower
than the optimum.  A  strong  effect showed a change of more than three standard
deviations  from  the  average  concentration at the optimum condition,  and no ef-
fect showed a change of less than one standard deviation.  Since the  entire ana-
lyzer  is pneumatically driven,  the  air pressure effects  occurring only at  lower
pressure are not surprising.
     The analog  integrator measures the baseline at a fixed time just before the
VC peak and integrates any signal above that baseline until  the fixed stop time.

Since the two detectors measure the peaks at slightly different times, a change
in the integration windows or the retention time will affect each detector dif-
ferently.   The digital  integrator  program first scans the chromatogram for the
peak maximum nearest  the  expected  time of the VC peak,  jumps forward by a pre-
set offset to begin  searching for a  level  baseline  at  the start of the peak,
performs  a similar operation  to find the end  of  the peak, and subtracts the
average baseline from  the  area  under the peak.  The program is insensitive to
changes in the offsets or the factors used to set the minimum width and flatness
of the baselines.  Although the PID has no  readily controlled unique variables,
the FID is sensitive  to changes in the flame hydrogen supply pressure and thus
to its flow rate.

     While the complete  instrument  system was undergoing a 1-month reliability
test in the  laboratory  under  simulated field conditions,  the final  arrangements
for the field installation  were completed.   The test design required  that the
analyzers  be operated  for  a period of 6  months at a VC monomer production fa-
cility with  three  5-day periods of equivalence tests comparing  the analyzers
with EPA Method  106.   The  primary difficulty in selecting a test site was that
all the plants contacted had  gas streams  which were  either much less than 1 ppm
or at concentrations  of at least 1,000 ppm VC.  The analyzer  was finally in-
stalled on a reactor  offgas stream with a nitrogen dilution tee to  bring the  VC
concentration within  the 1 to  10 ppm range needed  for  equivalency testing.
     Only  partial  data  are  available  since the test period does  not end until
late May.   Figure  3  shows  linear  regression lines from the 20, 1-hour equiva-
lence tests  in the initial  test series.   The  analyzer  abbreviations shown in
these figures refer  to  PID (P), FID  (F), digital  integration  (D),  and analog
integration  (A).   All four detector-integrator  channels  read  lower than the
reference  method at  low concentrations.   As a further check of equivalence the
Method 106 integrated bag was also connected to the analyzer sample inlet.   This
detects differences  caused by variations in  the  Method  106  sampling  rate or
sudden changes missed  during  the  150-second analyzer cycle.  The results from
the  integrated  bag measurements  (Figure  4)  indicate the  same  bias pattern.

     After a series of tests the problem was isolated to the Porasil C analyti-
cal column.   The column has a  short  retention  time for VC but  the uncoated
silica was causing nonuniform adsorption.   After replacing the  analytical  column
with a Porapak  Q-S  column  the nonlinearity disappeared.  After  resetting the
system timing,  the analyzer operated for about  1 month before  the second equiva-
lence test.
     During the  second  test  series  the direct  monitor readings (Figure 5) show
some scatter but  random bias.  The  integrated bag readings (Figure  6) have less
scatter.   An examination of  the individual  test runs indicates that the  inte-
grated bag  concentrations  are  biased toward the initial  sample  concentration
caused by  a  higher  than normal  flow rate during the  first few minutes as the
pressure within the bag enclosure stabilizes.
     Data recovery efficiency is  shown in Figure 7.  The daily  data  sets are
checked for outliers by measuring the  standard deviation of the differences be-
tween the  simultaneous pairs and rejecting any pair which exceeds four standard
deviations.  The  over-range  readings were caused by  changes  within the  host
monomer plant which upset the sample dilution ratio.  Data recovery for November
and December was  affected  by bad weather which produced repeated flameouts when
the plant  instrument air  supply failed.   During February the  sample gas  input
overloads were  so  severe  that  the integrator skipped  cycles.   During April the
data logger malfunctioned  when  a power supply  in the  data logger analog inter-
face failed.  The power supply was successfully replaced.
     Finally, Figure 8 shows the  daily bias and precision results  for the test
period following the analytical  column replacement.   The number shown below each
set of error bars is the average VC concentration for  that month.   The different
detector-integrator combinations show  little overall bias with reference  to the
analyzer's original FID with analog integration.  A more detailed analysis will
be made after the test period is completed.

              I  SPAN
      Figure 1.  Original  schematic of analyzer  system.



      Figure 2.  Revised  schematic of analyzer  system.

                 +30 %r
                <   0
                                          FIRST SERIES — Monitor Readings
• PD
• FD
                                                        PPM FROM M106
                     Figure  3.   Bias of analyzer direct readings  for the first equivalence test.

                                          FIRST SERIES — from Integrated Bag
                                                            • PD
                                                            • FD
                                                            O PA
                                                            A FA
                                                         PPM FROM M106
             Figure 4.  Analyzer  bias  when sampling from  the integrated bags  for the first equivalence test.

                   <  o -
                    -30 7c
                                           SECOND SERIES — Monitor Readings

                                                         PPM FROM M106
                        Figure  5.   Bias  of analyzer direct readings for the  second  equivalence test.

              <  0
                                     SECOND SERIES — from Integrated Bag
                                                    PPM FROM M106
          Figure 6.   Analyzer  bias when  sampling  from  the  integrated  bags  for  the  second  equivalence  test.

      100 •-
                        DATA RECOVERY
                               • OUTLIERS
                               H OVER-RANGE
                               til USABLE
   Figure  7.  Average monthly data recovery of complete data sets,
                     BIAS AND PRECISION
    I  0
        • PD
        O FD
        A PA
       4.6PPM    1.5PPM    0.25PPM
           NOV      DEC      JAN      FEB      MAR      APR

Figure 8.   Analyzer bias,  compared  to  the  FA channel, and  precision.


       Compound              Effect
Dichloromethane             None
Acetaldehyde                None
cis-l,2-Dichloroethane      None
1,1-Dichloroethane          None
Chloroethane                None
Chloroform                  None
Chloromethane               FID only
Isobutane                   Both
     Variable              Effect
Backflush time          Strong
Carrier pressure        Strong
Main air pressure       Strong-low
Valve air pressure      Moderate-low
Oven air pressure       None

       Variable                  Effect
Analog integration
  Integration stop time       Moderate-PID
  Integration start time      Weak

Digital integration
  Baseline scan forward       Weak
  Noise factor                None
  Leading edge offset         None
  Trailing edge offset        None

FID detector
  Fuel pressure               Strong
  Flame air pressure          None

                                   R. H. Krueger
                                    J.  M.  Fildes
                          Roy C.  Ingerso'll  Research  Center
                              Borg-Warner Corporation
                               Wolf  & Algonquin Roads
                               Des Plaines,  IL 60018
      The measurement of low concentrations of gases in air and of chemicals in
water is becoming of increasing importance.  For this reason, there is a need to
develop less expensive equipment to make these measurements.  Low cost, quick and
easy-to-use sensors with sufficient selectivity,  sensitivity and durability are
needed.  Now, more and more attempts are being made to satisfy these requirements
through the use of sensors made from semiconductors and transistors.  These
sensors have several potential advantages such as, miniaturization, speed and
long service life.
      We believe a new type of gas detector, an  integrated silicon sensor or
sensor-on-a-chip, will be introduced into many new applications in the next few
years.  The purpose of this paper is to give an overview of some of the present
work and predict future developments.
      The object of the work on integrated sensors is to take advantage of the
advances made in recent years in the microfabrication of various potential
sensors, such as ion-selective field-effect transistors (ISFET) and
metal-oxide-semiconductor field effect  transistors (MOSFET).  Bergveld^ ' was
first to propose a sensor based on a modification of a MOSFET where the gate
metal was replaced by an aqueous solution.  This  resulted in a device in which
the channel conductance appeared to be  a function of the ionic concentration of
the solution.  Bergveld called this  device a CHEMFET or an ion sensitive field
effect transistor.  Since this work, other researchers - Zemer ',
Lundstrorrr ', Senturia^ ', Krey, et.al.' ' - have prepared and tested these
devices as sensors.
      A metal-oxide-semiconductor field effect transistor (MOSFET) is shown in
Figure 1.  The substrate material is p-type silicon with a source and drain of
n-silicon.  The gate is a metal film evaporated over a thin insulating layer of

SiC>2-  With no voltage on the gate, the source and drain are insulated from
each other.  When a positive voltage is applied to the gate, electrons are
attracted to the surface of the silicon.  This produces a thin conductive surface
layer of induced n-type material (electrons) which now forms a channel connecting
the source and drain.  The number of electrons is directly proportional to the
gate voltage so that the conductivity of the channel  increases with gate voltage.
      How does the MOSFET act as a sensor for a gas?   One example is the
mechanism proposed by Lundstronr ' for the detection  of hydrogen.  It is known
that a number of metals, palladium and platinum, adsorb and dissolve hydrogen.
This occurs at the gate surface.  Lundstrom explains:  "Some of the hydrogen
atoms diffuse through the thin metal film and are adsorbed onto the metal -
SiOp interface.  An equilibrium develops between the  number of adsorbed
hydrogen atoms on the surface and those at the interface.  The number of adsorbed
hydrogen atoms on the surface depends not only on the hydrogen present in the
atmosphere, but also on the other gases present.  The hydrogen atoms at the
interface are polarized and this gives rise to a dipole layer which corresponds
to a voltage drop ( AV) which is added to the external voltage Vg."  This is
shown in Figure 2 for a palladium MOSFET, and in Figure 3 for a palladium MOS
      Figure 4 illustrates the voltage necessary to keep a small, constant drain
current on an n-channel Pd-MOSFET.  This is called the threshold voltage and it
depends on the hydrogen pressure and temperature.  Structures such as those shown
in Figures 2 and 3 have been used as sensors to detect gases in air as low as 5
PPB for H2, 50 PPB for H2S and 100 PPB for ammonia.
      Non-hydrogenous gases cannot diffuse through the gate and, therefore, they
cannot be detected by the sensor developed by Lundstrom.  To detect carbon
monoxide, for example, Krey and co-workersU) modified the transistor to have a
palladium gate with holes.  In this way, carbon monoxide was able to reach the
metal-oxide interface.  When carbon monoxide is adsorbed at the Pal ladiurn-SiOo
interface, a rise in the dipole layer occurred just as with hydrogen.  Again, the
threshold voltage increased with carbon monoxide pressure.
      Although these palladium gate MOSFETs look very promising as sensors for
hydrogen and carbon monoxide, there are some remaining problems.  Lundstrom has
reported that storage in oxygen gives a slow response the first time this sensor

is exposed to hydrogen.  In some instances, a drift in threshold voltage due to
so-called negative bias stress instability or hole trapping occurs.  Another
problem is poor palladium adhesion caused by phase changes in the palladium under
high hydrogen pressure, even at low temperatures.
      In addition to metal gate MOSFETs, Senturia^ and researchers at Siemens
in Germany^ '  have worked with polymer and organic semiconductor coatings
deposited on the gate region.  Some minor success was achieved by the Siemen
researchers but the objective of producing a usable gas sensor for gases such as
CO, C0£, SC>2 and NO with sufficient sensitivity, reproducibility and
stability was not achieved.
      The relative value of organic and inorganic sensing materials is yet to be
defined.  Many variables are important in this work - composition, film
thickness, surface chemistry, and topography; therefore, much more work needs to
be done to develop better gas sensors.  Lundstrom, Zemel and others have outlined
several areas of research needed to improve the sensitivity and selectivity of
gas sensors:
         1.  Metals and polymer coatings on the gate
         2.  Doping of gate metals
         3.  Porous gates
         4.  Insulator thickness
         5.  Insulator composition:  Si02, Si'3N4, A^Oo
         6.  Bulk composition
         7.  Device packaging.
      The sensor work is interdisciplinary and now it appears that there is an
increasing emphasis being placed on bringing together workers in the fields of
chemistry, physics and electronics.  The performance of chemical sensitive
devices is only limited by the selection of the proper chemistry.  This is a
relatively new area for research and development.  It has a high risk, but the
potential payoff should also be high.


1.  Bergveld, P., 1970.  Development of an ion-sensitive solid-state device for
    neurophysiological  measurements.  IEEE Trans.  Biomed. Eng., 17, pp. 70-71.

2.  Krey, D., K. Dobos, and G.  Zimmer,  1982/83.   An integrated CO-sensitive MOS
    transistor.  Sensors and Actuators, 3, pp. 169-177.

3.  Lundstrom, Ingemar, 1981.  Hydrogen sensitive  MOS-structures Part I:
    Principles and applications.  Sensors and Actuators, 1, pp. 403-426.

4.  Plihal, Manfred, Hans Pink, Ludwig  Treitinger, and Peter Tischer, 1980.  Gas
    sensitive semiconductor field effect sensors.   NTIS Report No. BMFT-FB-T
    80-091, 78 pp.

5.  Senturia, Stephen D., 1980.  Studies of conduction mechanisms in
    gas-sensitive polymer films.  Naval Research Report AD-A100995, 8 pp.

6.  Zemel, J. N., 1975.  Ion-sensitive  field effect transistors and related
    devices.  Analytical Chemistry, 47, pp. 255A-266A.

                     , DRAIN
                          -Si 02
Fig. 1.   Cross Section
          of a MOSFET.
     Fig.  2.   Hydrogen Sensitivity of a
               Pd-MOS Transistor (fron Ref.  3).
                                T-  H2 ot 150°
                                                                  TIME (sec)
Fig.  3.   Hydrogen Sensitivity of  a
          Pd-MOS Capacitor (from Ref.  3).
                   Fig.  4.   Response  of  Pd-fOSFET
                             to Hydrogen
                             (from Ref. 3)

                               by James B.  Flanagan

                              Rockwell International
                   Environmental Monitoring and Services Center
                                 Chapel Hill,  NC

     The Clinical Environmental Laboratory   (CEL)   is   an  EPA  research  facility
located on the campus of the University of  North Carolina in Chapel Hill.   CEL has
been extablished for the study of the effects  of exposure  to  priority  pollutant
gases on human subjects.  Standard air monitoring  equipment is used to monitor and
control the pollutant gas levels during exposure sessions.   These instruments  are
calibrated  every  few days using automated 3-point calibrations.  Old calibration
constants are superseded whenever a new calibration is performed.
     This paper examines the  hypothesis that  by  increasing  the  size   of  the
ensemble  of  calibration  data points, the net accuracy and precision of exposure
session averages will improve as a result of the corresponding  reduction   in  the
calibration  confidence  interval.   Ozone   instrument calibrations provide actual
data with which the theory is tested.  It will be  shown that drift and  instrument
nonlinearity  complicate  predictions  made  on  the   basis  of simple statistical

     The  following  linear  regression  model  is  assumed  for  the   instrument
        yi = B xt + C                                                          (1)
     During calibration, the paired y. and  x.  values  are assumed known, and the  B
and  C  constants  are  calculated from the ensemble  of data pairs.  The following
assumptions are the basis for elementary linear regression applications:
          1.  Instrument response is linear in concentration.
          2.  The error variance of the output is  independent of level.
          3-  Random errors are small compared to  the total variation in levels.
          4.  The error in y is Normally and Independently Distributed.
          5.  Error in x is negligible relative to error in y.


     The confidence interval is a statistical estimate of  the  region  about  the
calibration  line  in which a specified percentage (e.g., 95?) of additional (x,y)

points taken under specified conditions would lie.   Points  taken  subsequent  to
calibration  are  referred  to  in this paper as "probe" points.  A "probe" of the

calibration precision may be either a single data point or an average of a  number
of  data  points.   In  order  to correctly assess the confidence interval about a

regression line, the following additional conditions must be specified:

          1.  Estimating the Population Variance.  If an instrument is  calibrated
     without  any  knowledge  of  previous  precision  history, it is necessary to
     estimate the instrument's variance from the calibration data itself.   In this
     case,  It  is  necessary  to  use  Student's t  statistics  to  estimate  the
     confidence region.
          If (1) the instrument has a known history from which a prediction of its
     variance  can  be  made,  and  (2)  it  can  be  statistically shown  that the
     calibration is representative of  the  same  variance,   then  the  confidence
     interval can be derived using Gaussian probability tables.

          2.  "Probing" the Confidence  Interval.   Calibration  curves  are  made
     using  individual  data  points,  such  as 2-minute averages.  The confidence
     interval for subsequently acquired  "probe"  points  is  dependent  upon  the
     method  used  to  acquire  these  points.   A For example, when the confidence
     interval for an hourly average is required.  This confidence interval will be
     substantially narrower than the confidence interval for a 2-minute average.
     The following four equations represent the different cases:

Population Variance Unknown;  Single Sample "Probe" of Precision:

   p = tn_2 •  Sy •  (1  + l/n + (x- xm)2/Sxx)1/2                                (2)

Population Variance Unknown;  q-sample Mean "Probe" of Precision:

   p = fcn-2 '  sy '  (1/q + 1/n + (x - Xm)2/Sxx)V2                              (3)
Population Variance Controlled;   Single Sample  "Probe"  of Precision:

   p = z '  sigma  •  (1  + 1/n + (x - xm)2/Svv)1/2                               (4)
                y                    ui    x x

Population Variance Controlled;   q-sample  Mean  "Probe"  of Precision:

   p = z '  sigma  •  (1/q +  1/n + (x - x )2/S  )1/2                             (5)
                j                      cn    xx

     s       - sample standard deviation of y about  the  regression  line,
     slgma   - population standard deviation of y  variate,
     z       - Gaussian probability statistic for  specified  probability  level,
     t       - student's t value for n-2 degrees of  freedom,
     x       - mean value of x used in regression  line data  set,
     x       - any particular observation of x,
     Sxx     - sum of squares of (x-xm)  for all  x  values,
     q       - number of points averaged for the "probe",
     n       - number of calibration points used in  the  regression  line,
     p       - confidence interval half-width.

     Calibrations of CEL gas analyzers are done  approximately every other  day  and
results  are  archived.   This  provides  a data set which can be used  to  test  the
theory of the single-point "probe" of calibration  precision.   First,  a  data set is
designated  as  the  calibration  ensemble;   data points from the  next  succeeding
calibration  are  designated  as  "probe"  data  points.   Next,  differences  are
calculated  between  the  calibration  curve and the "probe"  points.  Finally,  the
confidence interval of the method can be investigated by accumulating ensembles of
these  differences  and evaluating these by appropriate  statistical means.   If  the
basic assumptions listed above hold exactly, the  statistical  characteristics   of
the ensembles of differences should be exactly described by  equations (2)  - (5).
     The actual ensembles of experimental data are constructed by taking a "moving
window"  j  calibrations wide to calculate the B and C constants.   The  data points
for the calibration immediately following are taken  as   "probe" points,   and  the
differences  added  to  the  respective  ensembles  for  each concentration/voltage
level.  The window is then moved one calibration forward in  time, and a new set of
j  calibrations  consisting of j-1 of the previously used calibrations  and one  new
point is used to derive a second set of  calibration constants.    The   window   is
moved in this manner until N differences are collected.
     In the absence of bias, the mean squared error  should equal the variance,  and
the  ensemble  mean would be zero.  In the presence  of bias,  the mean square error
exceeds the variance and the mean error is nonzero.
     The standard deviation of the ensemble of differences should   be  related   to
the  confidence  interval.  For a single-point "probe,"  such as the case here,  the
"theoretical" standard deviation would be:

     E.S.D.= sigma '  (1  + 1/n + (x.-x )2/S  )1/2                                (6)
                                  i  ra    xx
     sigma   - the true  (unknown)  standard deviation  of the  y  variate,
     E.S.D   - expected  standard deviation of an  ensemble  of n
               deviations at concentration level  x. ,
     n       - number of individual points used  in  the  regression:
               if j three-point calibrations are  combined, then  n=3j,
     x.      - x-value:   zero, midpoint,  span, etc.

     Figure 1 shows a plot of mean and  standard  deviation  of   the  ensemble  of
differences  computed  as  a function of number  of  combined  calibration  curves, j,

used to derive B and  C.   The data are for a  Bendix   Ozone   analyzer  Model   8002,

0.1 ppm  range,  using  the   j  most  recent  calibrations to  form  the ensemble of
calibration points for the regression.   Each point  plotted  is  based  upon  an
ensemble  size  of  N=12  difference  values.  Also shown  is the theoretical  curve
(E.S.D.) for the standard deviation given by eqn.  (6).   The  solid  curves  in the
figure  refer  to  the  expected standard deviation of  eqn.  (6), with the  value of
"sigma" being obtained by forcing the line through  the  first  point.   Examination
of Figure 1 yields the following observations:

          1.  Agreement   between  the  theoretical    and  experimental    standard
     deviation  curves  as  a  function  of  j,  the number of  3-point calibrations
     forming the regression set, is moderately good.  This gives some   confidence
     in  the applicability of the theory represented  by equations (2) -  (5) to the
     CEL data.

          2.  The  mean   value  of  the  differences  increases   as   number  of
     calibrations  used   to  derive the regression  coefficients  B and C  increases.
     Furthermore, for the midpoint data,  the mean difference starts out  negative,
     goes  through  zero  and  becomes  positive  at  larger  j  values.  Two effects
     appear to cause  this behavior:
               a.  Systematic drift leads to  increasing mean  error  when   older
          calibrations are included in the calculation  of  the  regression lines.
               b.  Nonlinearity of the instrument biases the  mean  deviation for
          the midpoint data.

          3.  The assumption of independence of  variance  from  the concentration
     level is clearly violated.  Physically, this probably arises from random flow
     and pressure errors which are simply "amplified" more at  higher concentration

     The work above used only a single point "probe" of the calibration precision;
we  have not yet addressed tne precision for session averages,  which is the actual
quantity of interest.  By using equations (2) - (5)  above,  we may make an estimate
of  the  confidence  interval when the "probe" is a  mean of many 2-minute averages
which comprise a session average.  The correction factor  for  converting  from  a
confidence  interval  for  2-minute  averages to a confidence interval for session
averages can be expressed as the ratio of eqn. (5) to eqn.  (4), assuming that  the
session average is composed of q data points:
                            (1/q + 1/n + (x - x )2/S  )
                                               m    xx'
   Ratio of Improvement =	——	                    (7)
                            (1 + 1/n + (x - x )2/S  )1/2
                                             m    xx'
     This ratio of improvement  is  applied  directly   to  the  ensemble  Standard
Deviations  shown  in  Figure 1  to obtain the corresponding Standard Deviations in
Figure 2.  Since the bias may be assumed to be unaffected,  the Mean Errors plotted
in  Figures  1   and  2  are identical.  Total Error in Figure 2 is obtained as the
square root of the sum of the squares of the corrected Standard Deviation and  the
Mean Error.
     Since the Standard Deviation in Figure 2 is decreased,  the  bias  error  has
more  relative  importance in the Total Error.  Thus,  the relative effect of drift
and nonlinearity on precision for session averages is  significantly  greater  than
for individual 2-minute averages.
     In summary, the data from the CEL gas data base illustrate the improvement in
net  precision  and  accuracy  attainable  by  using larger calibration ensembles.
Although combining historical calibration data points  reduces  the  width  of  the
calibration   confidence  interval,  the  systematic  bias  caused  by  drift  and
nonlinearity reduces the amount  of improvement in total accuracy attainable.

1.   Quality Assurance Handbook for Air Pollution Measurement Systems,
    Vol. I. Principles. U.S.E.P.A., SPA-600/9-76-005.
2.   Acton, Forman, Analysis of Straight-Line Data, John Wiley and
    Sons, Inc., New York, 1959.

  o CD rn
  dJ CO ~<

  _i. cu o
  cr  ~s
  -S -h co
-i. O CO
O c-h CO
3 -"• O
co o O
  3 -"•
_i.  OJ
3 O rt-
o -h ro
—•  Q.
C 3
Q- c S
n> 3 -••
CL cr r+
  ro zs-
-1- -5
3  OO
  o ro

CQ 13- -••
  ro co

  co o ^
  ->• -s ro
  o ->• cu
  3 o 3
   OJ CO
  ro —>










• A
• * A
• • A
• o A
• • £
/^^•^••A • * A .0
-° A A A • • • • • A ° o
^AAAA O A ° o «
X o o o
A°°o« OOQ^OO °
0 ° o o o A

                   13579      13579      13579
                      WINDOW  SIZE (CALIBRATIONS)

•       <    /  o


» '•If O
• <0
• / o  ^^ ^™
cr uj uj
uj o or Q:
h- Q Q^ LU
LU 1— Q£ X
I 1 1 1 .
• o < -S
1 1 1
P 10 o
CM _j -:
2 :