OCR error (C:\Conversion\JobRoot\00000CE0\tiff\20013KYV.tif): Unspecified error
-------
CONTENTS
1. INTRODUCTION 1
1.1 PURPOSE AND SCOPE 1
2. EPISODE MONITORING 3
2.1 STAGGERED SAMPLING 3
2.2 SHORT-TIME INTERVAL SAMPLING 5
3. ALTERNATIVES TO THE REFERENCE METHOD 7
3.1 ESTABLISHING ACCEPTABILITY 7
3.2 ALTERNATIVE METHODS 8
3.2.1 Soiling Index 8
3.2.2 Photometers and Transmissometers 9
3.2.3 Beta Gauge 9
3.2.4 Piezoelectric Principle 10
3.2.5 TEOM-Tapered Element Oscillating Microbalance . 11
3.3 SITE- AND SEASON-SPECIF 1C RELATIONSHIP 12
3.4 OPERABLE SIZE RANGES FOR PARTICULATE INSTRUMENTS ... 12
4. REFERENCES 15
APPENDICES
A. REFERENCE METHOD FOR THE DETERMINATION OF PARTICULATE
MATTER AS PM^ IN THE ATMOSPHERE 16
1.0 APPLICABILITY 16
2.0 PRINCIPLE 16
3.0 RANGE 17
4.0 PRECISION 18
5.0 ACCURACY 18
6.0 POTENTIAL SOURCES OF ERROR 19
7.0 APPARATUS 20
8.0 CALIBRATION 24
9.0 PROCEDURE 25
10.0 CALCULATIONS 28
11.0 REFERENCES 29
B. ESTABLISHING RELATIONSHIP OF ALTERNATIVE SAMPLING METHODS
TO THE REFERENCE METHOD 32
1.0 OCCASIONS WHEN RELATIONSHIP MUST BE ESTABLISHED .... 32
2.0 TEST CONDITIONS FOR TAKING MEASUREMENTS 34
3.0 METHODS FOR CALCULATING AND DISPLAYING THE RELATIONSHIP 36
4.0 METHODS FOR ASSESSING THE RELATIONSHIP 41
-m-
-------
1. INTRODUCTION
The EPA requirements which are prescribed in Title 40 Code of Federal
Regulations (40 CFR) Part 58 for ambient air quality monitoring for purposes
of State Implementation Plans (SIPs) are being revised, and will include
revisions to the emergency episode monitoring requirements now described in
Appendix C of 40 CFR Part 58. The emergency episode guidance provided in
Federal Regulation Part 51.16 and Appendix L will be amended to reflect the
new requirements. The new regulations would permit the use of two modified
versions of Appendix B of 40 CFR, Part 50 - Reference Method for the
Determination of Particulate Matter in the Atmosphere. The regulations
will also permit the use of non-reference methods provided that specific
site relationships to the reference method have been determined and docu-
mented and that these methods provide for short-term measurements.
1.1 PURPOSE AND SCOPE
The procedures specified in the reference method are for a 24-hour
sample collection period and a 24-hour equilibrium period before weighing.
Therefore, the use of the reference method is not responsive to emergency
episodes because of the 48-hour delay in determining the concentrations.
During episode monitoring conditions, the concentrations must be available
on a more current basis than provided for in the reference method for
pollutant index calculations and/or corrective actions where necessary.
There are two purposes for this guideline. The first is to explain
the priciples of operation of the two episode particulate monitoring methods
which are modifications of the reference method: (1) sampling over short
-------
sampling time intervals; and (2) staggered sampling. These methods may be
used by an agency in their participate matter episode network without
further testing in most cases.
Secondly, a procedure is described for establishing a site- and season-
specific relationship between the reference method and particulate methods
other than the two mentioned above.
-------
2. EPISODE MONITORING
The reference method for measurement of PM^Q requires the use of an
air sampler designed to pull air through a filter media in order to collect
the entrained particles. The average concentration of PM^Q collected over
a 24-hour period is calculated by determining the mass of the particles
collected and dividing the mass by the volume of air sampled. Appendix A
provides information on the key topics that are included in the text of the
PM^g reference method.
Two approaches recommended for episode monitoring require modifica-
tions to the reference method in order to reduce the time from the start
of the sampling period to the reporting of the concentration results and/or
to reduce the time between subsequent measurements at the same sampling
location. One approach uses short sampling intervals, and the other uses
staggered sampling intervals. Both methods rely on greatly reduced
equilibration time for the filter moisture content.
2.1 STAGGERED SAMPLING
Staggered high-volume sampling is a method that provides the opportunity
to update the information system at intervals more frequently than every 24
hours, but involves a short period for equilibration of the high-volume
filters. The method requires the use of multiple monitors at the site.
Operationally, the method involves staggered starting of the high- volume
samplers at a constant time interval, running the sampler for 24 hours,
then allowing the filter to equilibrate for 2 hours before determining the
final weight of the filter. As an example, if it is determined that an
-------
update of the PM^g concentration is required every 4 hours, six high- volume
samplers would be necessary. Samplers would be started at 4-hour intervals
after the first one is started. Twenty-four hours after the first sampler
is started, the filter from that sampler is removed, equilibrated for 2
hours, then weighed and the PM^Q concentration determined; hence, the first
concentration is determined approximately 26 hours after the start of
sampling, then an updated concentration is available and the facilities
available to accommodate multiple samplers, the PM^o concentration can be
updated at any desired time increment. More than six samplers at one site
are probably not practical; consequently, increments of 4 hours are the
shortest practical interval .
Some obvious precautions to be made when employing multiple samplers
include arrangement of the samplers so they sample the same atmosphere but
do not interfere with each other when operating simultaneously. Operation
of a number of samplers will also require an electrical power supply capable
of operating the samplers simultaneously without appreciably affecting the
voltage as samplers turn on and turn off.
The principal advantages of staggered PM^Q sampling are obvious:
§ Rigorous performance testing is not required for implementation,
because it is primarily based on the reference method.
• The method is compatible with analytical facilities dedicated
to the reference method.
• Relatively rapid data reporting promotes a near real-time
basis for contingency plan response and public information.
However, staggered PM^Q sampling is not without disadvantages when
compared with the reference method. For discussion purposes, these
disadvantages are separated into (1) predictable shortcomings and (2)
uncertainties. Predictable shortcomings are those problems that can be
-------
assessed prior to implementation. They include the following:
• Increased demands of equipment, expendable inventory, electrical
power, and labor may place an unacceptable strain on monitoring
resources.
t The physical aspects of the monitoring site may preclude acceptable
siting of multiple samplers.
• Data from the initial sample will not be available for 26 hours
after the sampling start time.
The principal area of uncertainty that cannot be properly assessed
prior to implementation lies in the shortened filter equilibration period.
Equilibration is necessary to standardize the moisture condition of the
filter/sample matrix so that inferred sample mass is not unduly biased by
retained moisture. Moisture retention arises from two processes:
hygroscopicity (differential absorption of water vapor onto hydrophilic
materials) and retention of unbound moisture (i.e., water droplets sampled
as if they were PM^o).
Problems associated with hygroscopicity can be suppressed by using
nonhygroscopic filtration material, but in many cases chemical and physical
properties of the particles must be accounted for.
2.2 SHORT-TIME INTERVAL SAMPLING
With this method, an abbreviated sampling period and filter equili-
bration period are used. A 4-hour sample period followed by a 2-hour
filter equilibration is suggested for a reasonable sampling duration. The
shorttime interval sampling method requires that a clean tared filter be
installed on the sampler after every 4-hour sampling period. After removal
from the sampler, the filter is conditioned in a dessicator for 2 hours,
then weighed. The PM^o concentration can be determined 6 hours after
starting the sample. Trends of PM^o concentrations can be determined if
-------
at least four successive sample values indicate changes in the same direc-
tion.
The obvious advantage to using the short- time interval is the oppor-
tunity for quicker results from the start of sampling; hence, it is more
representative of the most recent conditions and only one sampler is required
(compared with six samplers needed for the staggered- sampl ing scheme).
The filter and sample may be less subject to errors introduced, due to the
abbreviated equilibration time when compared with the staggered-sampl ing
method.
Some disadvantages that should be considered in conjunction with the
short-time samples include the following:
• A larger bias in the PM^Q concentration may occur due to artifact
particle formation on the filters. (Bias, due to artifact particle
formation, will be reduced with proper selection of filter material ,
e.g., by use of quartz or Teflon filters.)
• The possiblity for human error is greater due to the increased number
of filter changes required.
• The average of six 4-hour concentration values is only an estimate
of the true 24-hour continuous sample value.
-------
3. ALTERNATIVE TO THE REFERENCE METHOD
Atmospheric aerosols may be monitored or measured with instruments
that use several principles for detection other than the reference method.
The advantage of these other instruments is primarily that they provide
real-time readout of atmospheric aerosol concentrations.
A disadvantage of using the nonreference technique is the necessity
of developing a site-specific correlation of an alternative monitoring
technique with the reference method.
3.1 ESTABLISHING ACCEPTABILITY
Alternative methods of PM^Q sampling may be acceptable if a site- and
season-specific relationship is established between the alternate method
and the reference method. The guidance provided for establishing the
relationship is adapted from similar requirements for TSP monitoring. The
evaluation procedure requires simultaneous side-by-side sampling using the
reference method and the candidate method. A comparison of at least 10
simultaneous samples (or sample periods) is required. The data resulting
from the simultaneous samples are analyzed to determine if a useful
statistical relationship exists that will allow the use of the alternate
method. Details of the procedure for evaluating the acceptability of
candidate methods are provided in Appendix B.
The coefficient of correlation between the reference method and
alternative method being compared should have a value of 0.85 or greater
in order to accept the alternative method.
-------
Several factors should be considered when collecting samples or analyzing
data to determine the correlation of the measurement technique. Among the
factors to consider are the following:
• Influences of seasonal variations
• Influences from local sources
t Wind direction
Siting factors and similar considerations are discussed in a companion
guideline document--"Network Design and Optimum Site Exposure Criteria for
Particul ate Matter".
3.2 ALTERNATIVE METHODS
The leading methods that may be considered as alternatives to the
reference method are discussed in the remainder of this section.
3.2.1 Soiling Index is a sampling method that has been used in air
monitoring stations for many years and has been relied upon to determine
the onset of TSP episodes. Soiling index is determined by using an auto-
mated tape sampler that draws sample air through a filter tape and then by
evaluating the change in transmittance or reflectance of light through the
spot on the filter where the sample was collected. The transmittance or
reflectance of light is highly variable depending on the type of aerosol
collected. The optical properties of the aerosol sampled must remain
constant in order to develop a good correlation between mass concentration
and light transmittance. Site-specific studies would have to conducted
to show the uniformity of aerosols sampled before an adequate correlation
with the reference method could be developed.
-------
Regan et al . (1979) have reported finding excellent correlation
between fine participate concentrations and COM values. Regan's study
included correlation of COH values (average of six 2-hour samples) compared
against the concentration of fine particules collected on 12-hour samples
using a high-volme sampler equipped with an Andersen size-fractionating
sample head. Total particle concentrations from the high-volume samplers
were corrected to represent particles less than 15 ym. Correlation coeffi-
cients for the concentrations from five stages of the size-fractionated
samples ranged from 0.947 to 0.971. A regression analysis was run to
establish a quantitative relationship between COH and fine particulate
data. A good correlation was found in the instance reported, but the
results cannot be legitimately transferred to other sites without establish-
ing the relationship at each site.
3.2.2 Photometers and Transmissometers are included in this text
primarily because they have been used experimentally to attempt to determine
suspended particle concentrations. Aerosol measurement with optical tech-
niques is most effective for particles that are about 2 urn diameter or
smaller, where the light effect is most pronounced. In transmissometers,
particles in the light path attenuate a light beam and a signal responsive
to the changes in light at the detector. Integrating nephelometers also
depend on changes in light intensity as the detection principle; however,
in the integrating nephelometer, scattered light is detected, rather an
attenuated light beam. Some success has been reported with correlating
atmospheric mass concentration the response of the integrating nephelometer
(Butcher 1972).
3.2.3 The Beta Gauge, produced in various models by the GCA
Corporation, operates on the principle of collecting particles on a
-------
filter media that is located over a beta particle emitter. As the particles
are collected on the filter material, fewer beta particles penetrate the
accumulated mass. A correlation between the attenuation of beta particles
and the collected particle mass allows the instrument to provide real-time
indication of particle concentration. The typical sampling period for the
beta gauge is 4 minutes; hence, a series of sample must be averaged to
obtain a value corresponding to a long-term samples using a PM^o reference
method. A continuously cycling measurement mode is available with a printer
to record each sample result. Jaklevic et al . (1981) have reported the
beta gauge method is equivalent in accuracy to gravimetric methods when
proper attention is paid to instrument design and calibration procedures.
The conclusion is based on a study that included comparison of data using
the beta gauge, a dichotomous sampler, and a high volume sampler.
3.2.4 The Piezoelectric principle has been used in other instruments
designed to sample airborne particles and automatically display or record
a mass concentration, such as the piezobalance produced by TSI, Incorporated.
The standard piezopbalance is available with a 3.5 urn particle diameter
cutoff option, but the instrument can be ordered with particle selectors
with cutpoint up to 10 pm. The piezobalance operates using a piezoelectric
quartz crystal that oscillates at a constant frequency under static conditions.
As particles are collected on the detecting crystal, the frequency of the
oscillations changes in respect to the mass of particles collected. The
change in frequency is detected and compared with the frequency at the
initiation of the sampling period, and a concentration is calculated at
the end of a 2-minute sampling period. The concentration is displayed at
the end of the sample period until the beginning of the next period. The
10
-------
piezobalance is available in a portable, manually operated version and a
fixed-location, fully automated version that is capable of continuously
cycling (i.e., sampling, data recording, and automated cystal cleaning).
The beta attenuation and the piezoelectric quartz crystal instruments
provide the short-term sampling options that are desirable for episode
monitoring. Some caveats must be recognized when considering the use of
these instruments. The effect of the moisture collected in the sample is
not compensated for by either instrument. The effect of moisture collection
is likely to have a more significant effect on the piezoelectric instrument.
Particle collection efficiencies of the instruments must also be evaluated.
Although the instruments are considered to be useful for collecting particles
up to 10 ym aerodynamic diameter, ambient conditions may considerably alter
the actual size of particles collected. Fairchild (1980) has evaluated
both types of instruments and compared results with those from gravimetric
sampling. The report indicates the instruments evaluated accurately sample
particles with +^25 percent of the gravimetric concentration when four or
more instrument readings were averaged. The report also stated that the
piezoelectric quartz crystal collected particles greater than 6 m aerodynamic
diameter with poor efficiency due to poor mechanical coupling with the
quartz crystal.
3.2.5 TEOM-Tapered Element Oscillating Microbalance
A report of a recently designed instrument to detect fine particles
and determine the concentration after a 30-minute sampling period indicates
the instrument may be useful for short-term measurement of fine particles.
The instrument must undergo further development, but is to be commercially
available. The TEOM works by collecting particles on a filter cartridge
that is attached to the narrow end of hollow tapered tube. The wide end
11
-------
of the tube is attached to the instrument. The tapered tube oscillates
during the sampling period, with frequency of the oscillation being accurately
monitored. Loading of particles in the filter cartridge causes a shift in
the oscillation frequency, which is interpreted and translated into a
concentration by a microprocessor.
The system that has been developed and reported on is equipped with a
2.5 ym cutpoint cyclone preseparator, so the measurment relates to the fine
aerosol portion of airborne particle mass. The TEOM can be operated for a
long time period with printer output programmed on a 1/2- to 1-hour schedule.
A heater element in the intake maintains a constant temperature of the
incoming aerosol to provide frequency stability and minimize the effect of
humidity on the measurement.
The TEOM is a new device that has not been proven in field use. The
device reportedly has resolved difficulties that exist with the quartz
crystal micrebalance instruments.
3.3 SITE- AND SEASON-SPECIF 1C RELATIONSHIP
Any of the automated aerosol measurement units must be compared with
the PMiQ reference method to obtain a site- and season-specific relationship
before the measurements from the nonreference method can be accepted for
episode monitoring. Guidance for developing the site- and season-specific
relationship is provided in Appendix B.
3.4 OPERABLE SIZE RANGES FOR PARTICIPATE INSTRUMENTS
The Lawrence Berkeley Laboratories (LBL) has compiled an inventory of
the specifications and capabilities for a variety of air pollutant monitoring
instruments. The LBL document describing particle-collecting or detecting
instruments provides the following table, Table 1, to indicate possible
combinations of instruments that will detect particles over a wide range of
12
-------
in
c
o
C
<_>
•0
u
•— *
e -
o »
II
u •*
rtt (ft
S_ i.
C- C
(/>
W- C
c o
(-0
I/)
*-
oo-^
(/t
(U u)
F— (O
^il—
* E
•- C
O « i.
«
-c o
n> -^
• *-A
c a
r i.
c c
M
c-
5
u
01
1—
•
"
-n
1A
JJ
u
to
e
a
k-
e
u
e
o
u
u m
<-> o
i?
o o
— o
1°'
1
""•SCO
u c c o
—* a — —
o u e i
in s s in
O w 0 O
*• c"— •
UU<0
u
1
' ~.~ o
^
a
u
5"
in
•"•
jf
HI *• '
""" ** IM
£ °'
to
i
5
|d
^m
Jo
•C — ,
oc = i
11 5 i-»
Jo
— 5 0
u £"
— •— • '
1 1! =
— s °
^C-
§
«•" £
^ 2 •
-j O
« *J •
«£ o
3*
u
V
m
in
a
U
a
In
«i
u
u
£
III 1
000 0
• ""
PM in o O
CO— —
o o o o
f> ^1 m in
iii i
r*» in o O
do — —
COO O
*•*> i«n in in
i
•M Ifl O O
o" d — —
0 C O O
f"> i-> O m
1 , , ,
PM in o O
do— —
o o o o
P»* fs» d4 P»4
III 1
« « « «
<•>» m o o
do— —
o o o o
r~> (O o m
III t
« « « «
r« in o O
0 0 — -»
s
' 0s
«
« 1
<>1 — — C
• in 2 co o
'— O CUO O -• /—
u «— ._oo— so
E o u « ^o °> — o
-p O. u as- tn -3 •
o F >. w — — wo —
^ "
m
a
d
*
o
o
d
d
i
in
O
o
d
I
o
d
1
o
o
d
1
1
1
1
1
1
u ^
—
u
e:
E
s 21
5-2
«*. "° •s >.
1 ^ £ S
u u; B.^
o u = C "s
M BO U 25
— C > o
in ct) O <•• -
k. u >,
w c. • f
in a "S 2
I-s I S S
•• u
O as
in » £1
— C
j: w
= = X
— — s
in w
••* E 4}
- J 3 0 *
—' "I -i «)
— "> ^ U
tf y o , S
2 ""* C» — to
5 §• s ~ 2
< JS B i- —
13
-------
particle sizes. The instruments are described in generic terms, but the
listing can give valuable guidance to possible instrument types that may
be feasible for PM^Q episode monitoring.
14
-------
4. REFERENCES
1. Acton, F.S, 1959. Analysis of straight-line data. New York: John
Wiley and Sons.
2. Butcher, S.S., and Charlston, R.J. 1972. An Introduction to air
chemistry. New York: Academic Press.
3. Fairchild, C.I.; Tillery, M.I.; and Ettinger, H.J. 1980. An evaluation
of fast response aerosol mass monitors. LA 8220 , Los Alamos Scientific
Laboratory, Los Alamos, N. Mex.
4. Kramer, D.N., and Mitchel, P.W. 1967. Eavluation of filters for
high-volume sampling of atmospheric particulates. J. Am. Ind. Hyg. Assoc.
(28)3:224-28.
5. Lewis, C.W. 1981. The tapered element oscillating micrebalance: a
monitor for short-term measurement of fine aerosol mass measurement.
EPA Report 600/S2-81-146. Environmental Sciences Research Laboratory,
Research Triangle Park, NC 27711.
6. Lippmann, M.; Kleinman, M.T.; and Bernstein, D.M. 1979. Size-mass
distributions of the New York summer aerosol. Ann. N.Y. Acad. Sci.
322:29-44.
7. Natrella, M.G., ed. 1963. Experimental statistics, handbook 91.
National Bureau of Standards, Washington, D.C.
8. Regan, G.F.; Goranson, S.K.; and Larson, L.L. 1979. Use of tape
samplers as fine particulate monitors. J. Air Pol. Control Assoc.
(29) 1158-1160.
9. U.S. Environmental Protection Agency. 1979. Guidance for selecting
TSP episode monitoring methods. EPA Report 450/4-79-U07, Office of
Air Quality Planning and Standards, U.S. Environmental Protection Agency,
Research Triangle Park, MC 27711.
15
-------
APPENDIX A
REFERENCE METHOD FOR THE DETERMINATION OF PARTICULATE
MATTER AS PM10 IN THE ATMOSPHERE - Appendix J 40 CFR 50
(Reproduced as proposed in the Federal Register)
1.0 APPLICABILITY
This method provides for the measurement of the mass concentration
of particulate matter with an aerodynamic diameter less than or equal to
a nomimal 10 micrometers (PM^Q) in ambient air over a 24-hour period for
purposes of determining attainment and maintenance of the primary and
secondary national ambient air quality standards for particulate matter
specified in §50.6 and §50.7 of this chapter. The measurement process is
nondestructive, and the PM^g sample can be subjected to subsequent physical
and chemical analyses. Quality assurance procedures and guidance are
provided in Part 58,.Appendices A and B, of this chapter and in References
1 and 2.
2.0 PRINCIPLE
2.1 An air sampler draws a measured quantity of ambient air at a constant
flow rate into a specially shaped inlet where the suspended particulate
matter is inertially separated into one or more size fractions within the
PM]_Q size range. Each size fraction in the PM^Q size range is then
collected on a separate filter over the specified sampling period. The
particle size discrimination characteristics (sampling effectiveness and
50 percent cutpoint) of the sampler inlet over the PM^Q size range are
functional specifications described in Part 53 of this chapter.
2.2 Each filter is weighed (after moisture equilibration) before and
after use to determine the net weight (mass) gain due to collected "PM^o-
The total volume of air sampled, corrected to EPA reference conditions (25°
16
-------
C, 101.3 kPa), is determined from the measured volumetric flow rate and the
sampling time. The concentration of PM10 in the ambient air is computed as
the total mass of collected particles in the PM10 size range divided by the
volume of air sampled, corrected to reference conditions, and is expressed
in micrograms per standard cubic meter (ug/std m3). For samples collected
at temperatures and pressures significantly different from EPA reference
conditions, these corrected concentrations can differ substantially from
actual concentrations (in micrograms per actual cubic meter), particularly
at high elevations. The actual PM10 concentration can be calculated from
the corrected concentration, using the actual temperature and pressure
during the sampling period.
2.3 A method based on this principle will be considered a reference
method only if (a) the associated sampler meets the requirements specified
•
in this appendix and those in Part 53 of this chapter and (b) the method
has been designated as a reference method in accordance with Part 53 of
this chapter.
3.0 RANGE
The lower limit of the mass concentration range can be estimated from
the repeatability of filter tare weights, assuming the nominal air sample
volume for the sampler. The upper limit of the concentration range cannot
be specified. For samplers having a filter-changing mechanism, there may
be no upper limit. For samplers that do not have a filter-changing
mechanism, the upper limit is determined by the point at which the sampler
no longer maintains the specified operating flow rate due to increased
pressure drop across the loaded filter(s). This limit cannot be specified
because it is a complex and undetermined function of particle size dis-
tribution and type, humidity, filter type, and perhaps other factors.
17
-------
4.0 PRECISION
The reproducibil ity of PM10 samplers must be 15 percent or better as
required by Part 53 of this chapter, which prescribes a reproducibility
test procedure that determines the variation .in the PM10 concentration
i
measurements of identical samplers under typical sampling conditions.
Other specifications are provided in Part S3 for the particle size dis
crimination characteristics and the flow rate stability of the sampler.
Continual assessment of the precision via collocated samplers is required
by Part 58 of this chapter for PM10 samplers used in certain monitoring
networks.
5.0 ACCURACY
Because the sizes of the particles making up ambient particulate
matter vary over a wide range and the concentration of particles varies
with particle size, it is difficult to define the absolute accuracy of PM10
samplers. Part 53 of this chapter provides a specification for the sampling
effectiveness of PM10 samplers. This sampling effectiveness specification
requires that the expected mass concentration measurement calculated for a
candidate PM10 sampler, when sampling a specified typical ambient particle
distribution, be within ± 10 percent of that calculated for an ideal sampler
whose sampling effectiveness is explicitly specified. Also, the particle
size for 50 percent sampling effectiveness is required to be 10 ± 1 micro-
meters (urn). Other specifications related to accuracy apply to flow
measurement and calibration, filter media, analytical (weighing) procedures,
loss of volatiles, and artifact and nonsampled particulate matter. The
flow rate accuracy of PM10 samplers used in certain monitoring networks is
required by Part 58 of this chapter to be assessed periodically via flow
rate audits.
18
-------
6.0 POTENTIAL SOURCES OF ERROR
6.1 Loss of Volatile Particles: Volatile particles collected on filters
can be lost during shipment and/or storage of the filters prior to the
postsampling weighing3. Although shipment and storage of loaded filters
are sometimes unavoidable, filters should be reweighed as soon as practical
to minimize these losses.
6.2 Artifact Particulate Matter: Positive errors in particle mass measure-
ments can result from retention of gaseous species on filters4' 5. Such
errors include the retention of sulfur dioxide and nitric acid. Retention
of sulfur dioxide on filters, followed by oxidation to sulfate is referred
to as artifact particulate sulfate formation, a phenomenon which increases
with increasing filter alkalinity6. Artifact particulate nitrate, result-
ing primarily from retention of nitric acid, occurs to varying degrees on
many filter types, including glass fiber, cellulose ester, and many quartz
fiber filters5' 7> 8> 9> 10. Filters that meet the alkalinity specifica-
tion (section 7.2.4) should show little or no artifact sulfate. Negative
artifact is the loss of collected particulate matter during sampling by
volatilization or chemical reaction11. Loss of true atmospheric particu-
IS\
late nitrate has been observed on Teflon filters8 and inferred for quartz
fiber filters11. The significance of this problem for PM10 mass measure-
ments will vary with location and ambient temperature. However, for most
sampling locations, PM10 mass concentration errors due to nitrate artifact
are expected to be small.
6.3 Nonsampled Particulate Matter: Particulate matter can be deposited
on filters during periods when the sampler is inoperative12. Timely
installation and retrieval of filters prior to and following the sampling
period should help to minimize this problem.
19
-------
6.4 Humidity: The effects of ambient humidity on the sample are unavoid-
able. The moisture conditioning procedure in section 9.0 is designed to
minimize the effects of moisture on the filter medium.
6.5 Filter Handling: Careful handling of filters between presampling and
postsampling weighing is necessary to avoid errors due to damaged filters
or loss of particles from the filters. Use of a filter cartridge or
cassette may reduce the magnitude of these errors.
6.6 Flow Rate Variation: Variations in the sampler's operating flow rate
can alter the particle size discrimination characteristics of the sampler
inlet. The magnitude of this error will depend on the sensitivity of the
inlet to variations in flow rate and on the particle distribution in the
atmosphere during the sampling period. The use of an automatic flow
controller (section 7.1.4) is required to minimize this error.
6.7 Air Volume Determination: Errors in the air volume determination
can result from errors in the flow rate and/or sampling time measurements.
The automatic flow controller also serves to minimize errors in the average
flow rate determination. The use of an elapsed time meter (section 7.1.5)
is required to minimize the error in sampling time.
7.0 APPARATUS
7.1 PM10 Sampler
7.1.1 The sampler shall be designed to:
a. Draw the air sample, via reduced internal pressure, into the
sampler inlet and through the filter(s) at a uniform face
velocity.
b. Hold and seal the filter(s) in a horizontal position so that
sample air is drawn downward through the filter(s).
20
-------
c. Allow the filter(s) to be installed and removed conveniently.
d. Protect the filter(s) and sampler from 'precipitation and
prevent insects and other debris from being sampled.
e. Minimize leaks that would cause error in the measurement of
the air volume passing through the filter(s).
f. Discharge exhaust air at a sufficient distance from the
sampler inlet to minimize the sampling of exhaust air.
g. Minimize the collection of dust from the supporting surface.
7.1.2 The sampler shall operate at a controlled flow rate specified by
its designer or manufacturer, and it shall have an inlet system that
provides particle size discrimination characteristics meeting all of the
applicable performance specifications prescribed in Part 53 of this chapter.
The sampler inlet shall show no significant wind direction dependence.
This requirement can generally be satisfied by an inlet shape that is
circularly symmetrical about a vertical axis.
7.1.3 The sampler shall provide a means to measure the total flow rate
during the sampling period. A continuous flow recorder is recommended.
The sampler may be equipped with additional flow measurement devices if it
is designed to collect more than one particle size fraction.
7.1.4 The sampler shall have an automatic flow control device capable
of adjusting and maintaining the sample flow rate within the limits
specified for the sampler inlet over normal variations in line voltage and
filter pressure drop. A convenient means must be provided to temporarily
disable the automatic flow control device to allow calibration of the
sampler's flow measurement device.
7.1.5 A timing/control device capable of starting and stopping the
sampler shall be used to obtain an elapsed run-time of 24 ± 1 hr (1,440
21
-------
± 60 nrin). An elapsed time meter, accurate to within 15 minutes, shall be
used to measure sampling time. This meter is optional for samplers with
continuous flow recorders if the sampling time measurement obtained by
means of the recorder meets the ± 15 minute accuracy specification.
7.1.6 The sampler shall have an associated operation or instruction
manual as required by §53.4 of this chapter and which includes either the
text or a reproduction of this appendix.
7.2 Filters
7.2.1 Filter Medium: No commercially available filter medium is ideal
in all respects for all samplers. The user's goals in sampling determine
the relative importance of various filter evaluation criteria (e.g., cost,
ease of handling, physical and chemical characteristics, etc.) and, con-
sequently, determine the choice among acceptable filters. Furthermore,
certain types of filters may not be suitable for use with some samplers,
particularly under heavy loading conditions (high mass concentrations),
because of high or rapid increase in the filter flow resistance that would
exceed the capability of the sampler's automatic flow controller. However,
samplers equipped with automatic filter-changing mechanisms may allow use
of these types of filters. The specifications given below are minimum
requirements to insure acceptability of the filter medium for measurement
of PM10 mass concentrations. Other filter evaluation criteria should be
considered to meet individual sampling and analysis objectives.
7.2.2 Collection Efficiency: >_ 99 percent as measured by OOP test
(ASTM-2986) with 0.3 urn particles at the sampler's operating face velocity.
7.2.3 Integrity: ± 5 ug/m3 (assuming sampler's nominal 24-hour air
sample volume), measured as the concentration equivalent corresponding to
the difference between the initial and final weights of the filter when
weighed and handled under simulated sampling conditions (equilibration,
22
-------
initial weighing, placement on inoperative sampler, removal from sampler,
reequalibration, and final weighing).
7.2.4 Alkalinity: <0.005 milliequivalents/gram of filter as measured
by ASTM-D202 following at least two months storage at ambient temperature
and relative humidity.
7.3 Flow Rate Transfer Standard
7.3.1 A flow rate transfer standard, suitable for the flow rate of the
sampler and calibrated against a primary standard that is traceable to NBS,
must be used to calibrate the sampler's flow measurement device.
7.3.2 The reproducibility and resolution of the transfer standard must
be 2 percent or less of the sampler's operating flow rate.
7.3.3 The flow rate transfer standard must include a means to vary the
sampler flow rate during calibration of the sampler's flow measurement
device.
7.4 Filter Conditioning Environment
7.4.1 Temperature range: 15 to 30° C.
7.4.2 Temperature control: ± 3° C.
7.4.3 Humidity: 45 ± 5 percent relative humidity.
7.5 Analytical Balance
7.5.1 The analytical balance must be suitable for weighing the type and
size of filters required by the sampler. The range and sensitivity required
will depend on the filter tare weight and mass loading. Typically, an
analytical balance with a sensitivity of 0.1 mg is required for high volume
samplers (flow rates > 0.5 mVmin). Lower volume samplers (flow rates <
0.5 mVmin) will require a more sensitive balance.
23
-------
8.0 CALIBRATION
8.1 General Requirements
8.1.1 Calibration of the sampler's flow measurement device is required
to establish traceability of the flow measurement to a primary standard. A
flow rate transfer standard calibrated against a primary flow or volume
standard shall be used to calibrate the sampler's flow measurement device
at the field site.
8.1.2 The particle size separation characteristics of PM10 samplers
usually require that specific air velocities be maintained in the separa-
tion system. Therefore, the sampler must be set to operate at and maintain
the specified volumetric flow rate, measured under the actual ambient
conditions of use (Qa). In contrast, the mass concentration of PM10 must
cl
be computed using the flow rate based on the standard volume at EPA refer-
ence conditions (Qstd).
8.2 Flow Rate Calibration Procedures
8.2.1 The calibration procedure given here is based on flow rates at
ambient conditions (Q ) and serves to illustrate the steps involved in the
3
calibration process. Alternative procedures based on other measures of
flow rate (e.g., Qstcf) "iay be used provided the requirements of section
8.1 are met. Consult the sampler manufacturer's instruction manual for
specific guidance on calibration. Reference 13 provides additional infor-
mation on the use of the commonly used measures of flow rate and their
interrelationships.
8.2.2 Calibrate the flow rate transfer standard against a primary flow
or volume standard traceable to NBS. Establish a calibration relationship
(e.g., an equation or family of curves) such that traceability to the
primary standard is accurate over the expected range of ambient conditions
24
-------
(i.e., temperatures and pressures) under which the transfer standard will
be used. Recalibrate the transfer standard periodically (minimum of once
per year).
8.2.3 Disable the sampler's flow controller during calibration of the
sampler's flow measurement device.
8.2.4 Install a clean filter (or filters) in the sampler. Remove the
sampler inlet and connect the transfer standard to the sampler such that
the transfer standard accurately measures the sampler's flow rate. Make
sure there are no leaks between the transfer standard and the sampler.
8.2.5 Choose three flow rates evenly spaced over a range of ± 10 per-
cent of the sampler's specified operating flow rate (actual mVmin), and
by suitable adjustment of the sampler flow rate, obtain a calibration curve
of flow rate (actual mVmin) versus the sampler's flow indicator reading.
Record the barometric pressure and ambient temperature. Daily or seasonal
temperature and daily or average pressure corrections for subsequent flow
indicator readings may be required for certain types of flow measurement
devices (see NOTE following step 9.6).
8.2.6 Re-enable the flow controller, adjust the flow rate (actual
m3/min) to the manufacturer's specified operating set point, and use the
transfer standard to verify that the flow rate is correct with a clean
filter (or filters) in place.
8.2.7 Replace the sampler inlet.
9.0 PROCEDURE
9.1 The sampler shall be operated in accordance with the general
instructions given here and with the specific instructions provided in the
sampler manufacturer's instruction manual. NOTE: This procedure assumes
25
-------
that the sampler's flow rate calibration was performed using flow rates at
ambient conditions (Q ).
a
9.2 Inspect each filter for pinholes, particles, and other imperfections;
establish a filter information record and assign an identification number
to each filter.
9.3 Equilibrate each filter in the conditioning environment for at least
24-hours.
9.4 Following equilibration, weigh each filter, and record the presampling
weight with the filter identification number.
9.5 Install a preweighed filter (or filters) in the sampler following the
instructions provided in the sampler manufacturer's instructional manual.
9.6 Turn on the sampler and adjust (if necessary) the automatic flow
controller to the manufacturer's specified operating set point. Run the
sampler for at least 5 minutes to establish run-temperature conditions.
Record the flow indicator reading and, if needed, the barometric pressure
and ambient temperature. Determine the sampler flow rate (in actual
m3/min) using the sampler's flow rate calibration curve.
NOTE: No onsite pressure or temperature measurements are necessary if
the sampler flow indicator does not require pressure or temperature
corrections or if average barometric pressure and seasonal average tempera-
ture for the site are incorporated into the sampler calibration (see step
8.2.5). For individual pressure and temperature corrections, the ambient
pressure and temperature can be obtained by onsite measurements or from a
nearby weather station. Barometric pressure readings obtained from air-
ports must be station pressure, not corrected to sea level, and may need
to be corrected for differences in elevation between the sampler site and
the airport.
26
-------
9.7 If the sampler flow rate (actual mVmin) is outside the acceptable
range specified by the sampler manufacturer, check the sampler for leaks
and, if necessary, adjust the automatic flow controller set point. Stop
the sampler.
9.8 For samplers without continuous flow recorders, record the initial
flow rate (in actual mVmin) as Qa/-*nl-* \-
9.9 Set the timer to start and stop the sampler at appropriate times. Set
the elapsed time meter to zero.
9.10 Record the sample information (filter identification number(s), site
location or identification number, sample date, and starting time).
9.11 Sample for 24 ± 1 hours.
S.12 For samplers'without continuous flow recorders, as soon as practical
following the sampling period, run the sampler for 5 minutes to again
establish run-temperature conditions. Record the flow indicator reading
and, if needed, the barometric pressure and ambient temperature. Stop the
sampler. Determine the final flow rate (in actual mVmin) using the
sampler's flow rate calibration curve and record as Qa(f-jna-i ^ (see NOTE
following step 96). If Qa/-f^naT\ is outside the sampler manufacturer's
specified operating range, the sample must be invalidated. For valid
samples, calculate the average flow rate (in actual m3/min), and record
as Qa.
9.13 For samplers with continuous flow recorders, examine the flow record.
If Q is outside the sampler manufacturer's specified operating range for
a
more than 6 hours of the 24-hour sampling period, the sample must be
invalidated. For valid samples, record the average flow recorder reading
during the sampling period. If needed, estimate the average temperature
and pressure at the site during the sampling period from weather bureau or
27
-------
other available data. Determine the average flow rate (in actual m3/min)
using the sampler's flow rate calibration curve and record as Q (see NOTE
3
following step 9.6).
\
9.14 Carefully remove the filter (or filters) from the sampler following
the sampler manufacturer's instructions. Touch only the outer edges of
the filter.
9.15 Place the filter(s) in a protective holder or container (e.g., petri
dish, glassine envelope, or manila folder).
9.16 Record the elapsed time on the filter information record and any
other factors, such as meteorological conditions, construction activity,
fires or dust storms, etc., that might be pertinent to the measurement.
If the sample is known to be defective, void it at this time.
9.17 Transport the exposed sample filter (or filters) to the filter
conditioning environment as soon as possible for equilibration and sub-
sequent weighing.
9.18 Equilibrate the exposed filter(s) in the conditioning environment for
24-hours.
9.19 Immediately after equilibration, reweigh the filter(s) and record the
weight(s) with the filter identification number(s).
10.0 CALCULATIONS
10.1 Calculate the average flow rate over the sampling period corrected to
EPA reference conditions as Q ... When the sampler's flow rate calibration
and operation is based on flow rates at ambient conditions,
calculated as:
* r * r
'a Kstd
28
-------
where: 'Qstd = avera9e ^low rate at EpA reference conditions, std m3/min;
Q = average flow rate at ambient conditions, mVmin;
a
P. = average barometric pressure for the site or average
barometric pressure during the sampling period, 1
-------
3. Clement, R. E. , and F. W. Karasek1. Sample Composition Changes in
Sampling and Analysis of Organic Compounds in Aerosols. Int. J. En-
viron. Analyt. Chem.,.7:109, 1979.
4. Lee, R. E. , Jr., and J. Wagman. A Sampling Anomaly in the Deter-
mination of Atmospheric Sulfate Concentration. Amer. Ind. Hyg. Assoc.
J., 27:266, 1966.
5. Appel, B. R., S. M. Wall, Y. Tokiwa, and M. Haik. Interference Effects
in Sampling Participate Nitrate in Ambient Air. Atmos. Environ.,
13:319, 1979.
6. Coutant, R. W. Effect of Environmental Variables on Collection of
Atmospheric Sulfate. Environ. Sci. Techno!., 11:873, 1977.
7. Spicer, C. W., and P. Schumacher. Interference in Sampling Atmospheric
Particulate Nitrate. Atmos. Environ., 11:873, 1977.
8. Appel, B. R., Y. Tokiwa, and M. Haik. Sampling of Nitrates in Ambient
Air. Atmos. Environ., 15:283, 1981.
9. Spicer, C. W., and P. M. Schumacher. Particulate Nitrate: Laboratory
and Field Studies of Major Sampling Interferences. Atmos. Environ.,
13:543, 1979.
10. Appel, B. R. Private Communication, 1982.
11. Pierson, W. R. , W. W. Brachaczek, T. J. Korniski, T. J. Truex, and
J. W. Butler. Artifact Formation of Sulfate, Nitrate, and Hydrogen
Ion on Backup Filters: Allegheny Mountain Experiment. J. Air. Pollut.
Control Assoc., 30:30, 1980.
12. Chahal, H. S., and D. J. Romano. High-Volume Sampling Effect of
Windborne Particulate Matter Deposited During Idle Periods. J. Air
Pollut. Control Assoc., 26:885, 1976.
30
-------
13. Smith, F. , P. S. Wohlschlegel, R. S. C. Rogers, and D. J. Mulligan.
Investigation of Flow Rate Calibration Procedures Associated with the
High Volume Method for Determination of Suspended Particulates.
EPA-600/4-78-047, U.S. Environmental Protection Agency, Research
Triangle Park, North Carolina 27711, 1978.
31
-------
APPENDIX B
ESTABLISHING RELATIONSHIP OF ALTERNATIVE
SAMPLING METHODS TO THE REFERENCE METHOD
1.0 OCCASIONS WHEN RELATIONSHIP MUST BE ESTABLISHED
Whenever a method other than staggered high-volume sampling or short time
interval sampling is to be used for episode monitoring, a specific site relation-
ship with the reference method must be established and documented.
The advantage of using the staggered sampling or short time interval
sampling methods is that these methods use the principle of the reference
method. Consequently, in most cases, a specific site relationship need not
be determined, thus reducing the testing time, testing cost, and the need
perhaps for skilled personnel or expensive equipment.
The advantage of other methods over the two methods suggested above may be
the advantage of automation, which could minimize the loss of time between air
samples and important strategy decisions.
Requirements (see Table B-l) that must be satisfied to determine whether
a method can be designated as equivalent to a reference method are found in
Subpart C of 40 CFR Part 53. Comparability for a candidate method is demon-
strated if all the differences between corresponding measurements made by the
candidate method and by a reference method on simultaneously collected PM-|g
samples at each of the three test sites are less than 15 percent of the
reference method value. In addition, a candidate method must exhibit perfor-
mance better than, or equal to, the requirement stated in Subpart D of 40 CFR 53.
32
-------
TABLE B-l. TEST SPECIFICATIONS FOR PM1QMETHODS
Concentration range, yg/m3 30 to 150
Minimum number of test sites: 3
Minimum number of candidate method samplers per site: 3
Minimum number of 24-hour measurements per sampler per site: 5
Minimum number of sample sets: 15
Acceptable difference, percent of reference method: +15
33
-------
These criteria define what good comparable performance is for a monitoring
method. A careful review of the full text of 40 CFR Part 53 should be made
before testing for comparability of two methods in order to ensure that the
most current rules are being applied. However, failure of a method to meet
the specifications for comparability to a reference method does not render the
data from the candidate method useless. The procedure described below may be
used to determine the correlation of the candidate method with a reference
method and consequently allow those data to be assessed with respect to
applicability to episode"monitoring.
The remainder of this Appendix* describes the procedures that should be
followed to collect test data and to calculate, display, and assess a specific
site relationship between a candidate method and the reference method.
2.0 TEST CONDITIONS FOR TAKING MEASUREMENTS
The reference method uses the filtration method described in Appendix A.
The procedures described in Appendix A should be followed when sampling is
done using the reference method. Sampling using alternate methods (candidate
methods) for which the specific site relationship is to be established should
be conducted according to the applicable manuals (e.g., manuals produced by
instrument vendors that produce the instrument to be tested).
At least 10 ambient air measurements should be made simultaneously by
the candidate and the reference method. The air samples should be taken
* It is important to note here that much of the text included in the remainder
of this chapter follows very closely the text provided in an EPA document
(EPA 450/4-79-007) referenced previously.
34
-------
simultaneously in the same area (i.e., within 2-3 meters of each other) with-
out interference between samples or instruments. It is preferable to take all
measurements during a single season, because site-specific relationships may
vary with season. If more than one season is covered by the sampling period,
then 10 or more measurements per season should be made and results analyzed on
a season-specific basis. The definition of a season is left to the discretion
of those determining a site-specific relationship. In most areas of the contig-
uous United States, December through February is considered the winter season,
with subsequent 3-month groupings comprising the remaining three seasons.
These measurements should be made on ambient air containing PM-jg con-
centrations in the range that the candidate method will be subjected to during
an episode. To accomplish this, the measurements could initially be performed
in nonepisodic (normal) atmospheric conditions to establish the relationship,
but resumed during an episode to determine if the established relationship
changes significantly. The relationship may be a function of the pollutant
source, such as coal combustion, dust storms, or fuel use. If such is the
case, then the relationship should be established under different source
strengths.
If the candidate method has a shorter measuring time interval than the
reference method, a sufficient number of sequential interval measurements
should be made to equal the time period of the reference method (24 hours).
The PM,Q concentration as determined by the reference method and the PM,Q
concentration determined by the candidate method (or the mean of sequential
determinations by the candidate method) are considered a "test pair."
35
-------
All records, test data, procedural description and details, and other
documentation obtained from (or pertinent to) tests made for the purpose of
testing a candidate method should be identified, dated, and signed by the
tester. Test data are to be compiled and forwarded to the U.S. EPA Regional
Office for approval.
3.0 METHODS FOR CALCULATING AND DISPLAYING THE RELATIONSHIP
The test pairs are used to estimate the functional relationship between
the candidate and the reference methods. For a full season of 24-hour mea-
surements taken every sixth day, 15 or 16 test pairs would be available in
the absence of missing data. Hypothetical data (16 pairs) that are used to
illustrate appropriate analytical methods appear in Table B-2. The method
suggested for estimating a functional relationship between the candidate method
(X) and the reference method (Y) is linear regression. Figure B-l shows all
calculations required to produce regression estimates and associated outputs
from the inputs in Table B-2. The method suggested for displaying a relation-
ship between values from the two methods is a scattergram, as illustrated in
Figure B-2. The latter figure also includes the estimated regression line.
The essential outputs from the regression procedure are the estimated
slope (Step 12 in Figure B-l), intercept (Step 13), and correlation coefficient
(Step 14). The slope and intercept are required in order to make subsequent
predictions based on the regression estimates (Step 15). The correlation
coefficient is useful in assessing the strength of the association between
values from the two methods. The residual mean square (Step 11) and sum of
squared deviations about T (Step 6) are necessary if one wishes to construct
36
-------
TABLE B-2. SIMULATED DATA FOR CANDIDATE AND REFERENCE METHODS
x - PMJO
ug/m-3
Response measured
by candidate method
357
392
311
281
240
287
259
233
231
237
209
161
199
152
115
112
Y = PM1Q
ug/nr
Response measured
by reference method
459
419
375
334
310
305
309
319
304
273
204
245
209
189
137
114
37
-------
X denotes Candidate Method Response, Y denotes Reference Method Response
a) 2X = Sum of X
= 3776
b) Y = 2X/n = Average of X
= 236
c) n = Number of test pairs
= 16
d) 2Y = Sum of Y
= 4505
e) T = 2y/n = Average of Y
= 281.5625
Step (1):
2XY = Sum of X times Y
= 1,170,731
Step (2):
(2X) (2Y)/n
* 1,063,180
Step (3): S = 2 (X - Y) (Y - T) or 2XY - (2X)(2Y)/n
Step (1) - Step (2}
= 107,551
Step (4): 2X2 = Sum of each X squared
= 985,740
Step (5): (2X)2/n = a2/c
= 891,136
Step (6); ' Sxx = 2 (X - IT = Step (4) - Step (5)
= 94,604
Step (7): 2Y2 = Sum of each Y squared
= 1,404,543
Figure B-l. Basic Worksheet Showing the Calculation Steps
38
-------
Step (8): CSY)2/n - d2/c
= 1,268,439
Step (9); S = 2(Y - T)2 = Step (7) - Step (8)
= 136,104
A 2
Step (10) : 2 (Y - Y) = Residual sum of squares (sum of squared
differences between predicted and actual
y values)
= S(Y - Y)2 - [(X -~X)(Y -~Y)]2/Z(X -
= Step (9) - [Step (3)]2/Step (6)
* 136,104 - (107,5512/94,604)
* 13,834
Step (11): S2 = [2(Y - Y)2]/(n - 2) = Residual mean square
= Step (10)/(c - 2)
= 13,834/14 = 988
Step (12): Z = S /S = Slope of the regression curve
= Step (3) /Step (6)
= 107,551/136,104 = 1.137
Step (13); I = T - Z T = Y intercept
= e - Step (12)b
=* 281.5625 - 1.137(236) = 13.23
Step (14):. r = Correlation coefficient
* S / VT~ VT~
xyj xx yy
» Step (3)/[Step (6)]1/2 [Step (9)]1/2
= 107,551/[136104]1/2 [94604]1/2
' 0.948
Figure B-l. Basic Worksheet Showing the Calculation Steps (Continued)
39
-------
Step (15); Prediction of Reference Method Value from Candidate
Method Value
(a) Equation of regression line
Y = I + ZX
Y = 13.23 + 1.137 X
(b) Sample prediction
Let X0 » 240
Y0 = 13.23 + 1.137(240)
= 286.1
Step (16): 95 Percent Confidence Interval for Predicted Value (Y0)
(a) Equation for interval
Interval = Y0 +_ tn_2, 0.05 (s ^ + 1/n + (xo - "^2/ i.Q6267
= 286.1 +_ 69.5
= (216.6, 355.6)
Figure B-l. Basic Worksheet Showing the Calculation Steps (Concluded)
40
-------
confidence intervals around predicted reference method values (Step 16). The
value of t, used for a 95 percent confidence interval, varies with the number
of test pairs. The appropriate number can be found in a table of t-values,
which appears in most statistical texts.
Advanced slide-rule calculators with statistical capability can be used to
quickly obtain estimates of the slope, intercept, and correlation coefficient.
With such a calculator, one needs only to input the test pairs. The estimates
are retrievable by pushing specified keys on the calculator. With some alge-
braic manipulation, the residual mean square and sum of squared deviations can
be derived from intermediate outputs, which are retrievable from selected
memory areas within the calculator.
4.0 METHODS FOR ASSESSING THE RELATIONSHIP
The regression procedure illustrated in Figure B-l should be used only
if the relationship between candidate and reference method values appears to
be reasonably linear over the range of interest. A visual assessment of the
degree of linearity can be made upon the basis of the scattergram (Figure B-2).
In cases where the relationship does not appear to be linear over the
entire range of interest, other techniques for relating the two methods may be
used. For example, some agencies have found that a piece-wise linear function
is superior to a single linear function. Tests of significance are not as
straightforward as for the linear case and require care in interpretation.
Often, it is possible to linearize a nonlinear relationship by transformation
of variables (e.g., -^x7 log X, etc.). In such cases, the procedures outlined
in the previous section" of this Appendix may be applied to the transformed
41
-------
450 -
400 t
350
•3
J
a
2
u
<*•
a
si
300 -
£ 250 i
200
ISO
100
200 250 300
Values for Candidate Method
350
400
Figure B-2. Graph of Relationship Between Candidate and Reference Method Values
42
-------
variables. In some rare cases, it may be necessary to use nonlinear regression
techniques to properly fit the measured data.
The visual assessment of linearity should precede all calculations. Once
a reasonable approximation to linearity has been achieved in the original or
transformed data, the calculations shown in Figure B-l can be performed. The
resultant regression equation can be used to draw the regression line through
the data, as in Figure B-2. This line can be drawn by predicting Y values from
X values at low and high extremes, plotting the resultant (X,Y) pairs, and
connecting these plotted points. Any apparently nonrandom pattern of residuals
about the regression line should then be readily discernible.
The strength of the association between the candidate and reference method
values can be discerned by examining the correlation coefficient. In the case
of 16 pairs, a correlation coefficient of 0.5 or greater is statistically
significant at the 5 percent level. A coefficient of this magnitude, however,
is of limited practical significance. The square of the correlation coefficient
is a measure of the proportion of the variation in the reference method values
explained by regression on the candidate method values. Thus, a correlation
coefficient of 0.5 means that only 25 percent of the variation in reference
method values is explained. A correlation coefficent of 0.7 means that approx-
imately 50 percent of the variation is explained. Consequently, agencies are
not advised to use a candidate method whenever the coefficient of correlation
is less than 0.7. A more stringent cutoff level for candidate method accept-
ability would be a correlation coefficient of 0.85, which means that nearly
75 percent of the variation is explained.
43
------- |