-------
and is adjustable from one minute to one week per tube. After each tube has
collected the specified volume of air, the PFTs are recovered from the
adsorbent by heating and are subsequently passed from the tube through an
automated electron capture detector-gas chromatograph (EDC-GC) system, which
is capable of analyzing the 23-tube array in about three hours. This sampling
system has been used in several intermediate- and long-range tracer
experiments with successful results (e.g., Ferber et al., 1981; Fowler and
Barr, 1983; Clark et al., 1984).
Another type of adsorbent sampler collects the tracer by Fickian diffusion
toward the adsorbent material. This passive sampling device, developed
originally for indoor tracer studies, has also been used in atmospheric tracer
studies (Dietz et al., 1983). The sampler, known as the Capillary Adsorption
Tube Sampler (CATS), is best suited to sampling tracers over extended periods
of time. The sampling rate for the passive collector was recently determined
from a comparison of samplers to be equivalent to 232 ml air/day for PMCH and
217 mil air/day for PDCH (Dietz et al., 1983).
Perfluocarbon sampling devices have also been developed to sample
concentrations aloft. The main sampling methods consist of either airborne
sampling using whole-air or adsorbent samplers, or sampling through a group of
tubes suspended by a balloon. PFT samples are currently capable of being
analyzed in the laboratory and in situ, by semi-continuous and continuous
analyzers. Analytical techniques are similar for both bag samples and those
collected by adsorption. The samples are first processed to concentrate the
tracer and subsequently passed through a gas chromatograph system.
Detectability limits and measurement precision depend on the collection and
analysis methods.
The range of detection for laboratory analysis spans six orders of
magnitude, i.e., a minimum of 0.5 - 5 fL/L up (femtoliters per liter) to 5000
-13-
-------
pL/L (Ferber et al., 1981; Dietz et al., 1983), whereas semi-continuous and
continuous analyzers in situ have greater minimum detectable limits (Dietz and
Dabberdt, 1983). Continuous analyzers used to detect PFTs in recent
local/mesoscale studies revealed a practical detection limit of approximately
10 pL/L, clearly not sensitive enough to detect PFTs from long-range tracer
experiments, where the emphasis is on sub pL/L concentration levels. A
semi-continuous dual-trap analyzer, which periodically processes samples on a
four-minute basis, currently exhibits a 1 fL/L detection limit (Dietz and
Senum, 1984), suitable for airborne sampling on a regional scale.
The programmable BATS sampler has detected background concentration levels
of PDCB (0.35 pL/L) with a precision of + 10 percent, PMCH (3.6 pL/L) with a
precision of + 3 percent, and PDCH (26 fL/L) with a precision of + 5 percent
from 25-liter samples. With adjustable sampling rates of 0.5 - 40 ml/min,
these precision measures correspond roughly to a 10-hour minimum sample.
The CATS passive PFT sampler has a demonstrated detection precision for
background PMCH and PDCH concentrations of + 10 percent. However, these
precision measurements correspond to 30-day sampling periods due to the slow
sampling rate (0.14 ml/min) of the passive sampling method. Both CATS and
BATS samplers have measured nearly identical background PMCH and PDCH levels,
indicating the low variability of background levels and high accuracy of the
gas chromatography detection procedure.
The long-range tracer component of the combined experiment will use the
semi-continuous dual trap analyzer for aircraft measuremencs. Ground level
sampling will be performed using the BATS adsorbent samplers with analysis of
the samples performed in a central automated laboratory.
2.2 Sulfur Hexafluoride (SF5) Tracer
Use of sulfur hexafluoride (SFs), the first electron-attaching compound
used for atmospheric tracing, dates back to the mid-1960s. Its continued
-14-
-------
widespread use is due to its ease of detectability in GC analyzers and its
availability in the liquified gas state, which simplifies the release
procedure. The physical properties of SF6 and relative costs are shown in
Table 2-1. The high background levels of SF6 restrict its use to
local/mesoscale (i.e., <100 km) field experiments.
Sampling of SF6 is typically accomplished using a variety of whole-air
or adsorbent samplers, with both ground-level and airborne sampling
methodologies currently well established. Portable gas chromatographs are
commercially available for analyzing whole air samples with detection limits
of 5 pL/L. Processing the sample prior to analysis, i.e., using a precut
column technique (Dietz and Cote, 1971) and concentrator (Dietz et al.,
1976a), have yielded lower detectable limits and higher precision. For
example, Dietz and co-workers (1976b) demonstrated detectable limits of 0.5
pL/L with + 3 percent precision from 40 ml whole-air samples passed through a
molecular sieve trap cooled to dry ice temperatures.
Detection limits of about 7 pL/L have been produced by a semi-continuous
SFS monitor sampling at downwind distances of 90 km, whereas the detection
limits of truly continuous analyzers, used in short range (~10 km) field
experiments, are typically 10 to 30 pL/L (Dietz and Dabberdt, 1983).
During the recent short-range SFs tracer experiments associated with the
EPRI PMV&D Kincaid field measurement phase, the SFS measurement
uncertainties were assessed by analyzing numerous colocated samples,
performing analyses on duplicate samples, and auditing the performance of the
sampling and analysis procedures. Results from these QA procedures indicate
that for ground-level SFS concentrations less than 100 pL/L (i.e., 100 ppt) ,
the overall measurement uncertainty is within + 10 pL/L 90 percent of the
time. Measured concentration values exceeding 100 pL/L were generally within
+ 10 percent of the mean concentration, 90 percent of the time (Smith et al.,
1983).
-15-
-------
2.3 Isotopic Sulfur (34S) Tracer
With the development of the Isotope Ratio Tracer Method (Manowitz et al.,
1970; Newman et al., 1971), the use of sulfur isotope ratios as tracers has
become feasible. The two most predominant isotopes of sulfur, 32S and
34S, occur approximately in the ratio 34S/32S = 4.502 x 10"2 in
meteoritic sulfur. This ratio has been accepted as the standard against which
measured ratios in the environment are compared. Newman and others (1975a and
b) present data suggesting that the average isotopic ratio in atmospheric
sulfur is 4.515 x 10"2, with a range of 4.500 x 10"2 to 4.534 x 10"2 and
a standard deviation of 9.18 x 10"3. In terms of the percentage abundance
of 34S relative to total sulfur, the average, range, and standard deviation
are 4.319 percent, 4.3060-4.3373 percent, and 0.0091 percent, respectively.
The very low background variability suggests that only small guantities of
enriched 34S, enough to exceed the natural variability, are required to
serve as an atmospheric tracer.
The process by which isotopic tracers are monitored in the field involves
collection of SOz and sulfate particulates on suitable filters and
performing isotopic mass spectrography. The basic collection technique,
described by Forrest and Newman (1973), has been used in several programs
(e.g., Newman et al., 1975a and b; Hitchcock and Black, 1984). To obtain a
precision of + 0.02 percent from the mass spectrometer, a minimum quantity of
sulfur oxide is required. The sensitivity of the spectrometer used during the
mid-1970 studies required a minimum sample size of 1 mg of sulfur oxides
(Forrest and Newman, 1973) to achieve this precision.
Precision sulfur dioxide samples are typically collected on a series of
alkaline (carbonate) impregnated filters mounted back-to-back behind glass
fiber pre-filters in a hi-vol sampler. The filters are subsequently processed
to isolate the sulfur sample, which is then analyzed for isotope fractions.
-16-
-------
The precision achieved for the ratio measurement procedure is approximately
0.01 percent of the isotopic ratio (Hitchcock and Black, 1984).
Available evidence suggests that sulfur oxides produced by biological
processes may exhibit different isotope signatures than the sulfur from
geological sources (Kaplan and Rittenberg, 1964). There is also some
empirical evidence suggesting that the isotopic ratio is altered by
atmospheric chemical reactions, particularly aqueous phase S02 oxidation
(Newman et al., 1975). This evidence would imply that the sulfur isotopic
ratio distribution within aerosols may be variable and dependent on the
available ambient oxidants. Furthermore, if the fractionation varies with
particle size, for example due to a relationship between oxidation rate and
droplet pH, then dry deposition processes could influence the isotope ratio of
ambient airborne sulfur. However, the differences in isotope ratios, which
are thought to be associated with these processes, are of the order of 2
percent of the 34S/32S ratio (Hitchcock and Black, 1984), and thus could
conceivably result in a slightly higher variability than the data of Newman
and co-workers (1975) and Forrest and Newman (1973) would indicate. Use of
the isotope tracer in the combined experiments will be over limited distances
and will minimize these effects. Further studies are required to quantify the
significance of variable fractionation accompanying the atmospheric transport
of an isotopic sulfur tracer. However, even the larger estimates of the ratio
variability indicate a nearly constant proportionality, and hence, determining
suitable 34S emission rates based on detection above the background
variability, as suggested by Hicks (1984) and appear entirely adequate.
2.4 Sulfur Dioxide
Methods specified for the combined experiments combine hi-vol sampling of
SOz on alkaline impregnated cellulose filters with subsequent extraction and
-17-
-------
analysis. A review of the varied techniques used to sample ambient SO:
concentrations and their associated uncertainties is not attempted here.
Instead, SOz sampling methods employed in the recent SURE program (Mueller
and Hidy, 1983) and ERAQS program (Mueller and Watson, 1982)* are described.
Uncertainties estimated from these programs can be considered upper limits for
those specified in the COMPEX program.
Under the previous SURE and ERAQS field experiments, hourly averaged S02
measurements were taken with a commercially available flame-photometric sulfur
analyzer (Meloy Labs, Models 185 and SA 285). Since the flame-photometric
measurement principle responds to any sulfur-containing species reaching the
flame, particulate sulfate is filtered at the analyzer inlet. Other
sulfur-containing species, such as hydrogen sulfide, carbonyl sulfide, and
organic sulfur compounds, typically occur at concentrations below the level of
parts per billion and hence do not significantly affect the accuracy of
background SOj levels.
The relative uncertainty was determined for a limited number of
instruments over a limited time period during the ERAQS program (Mueller and
Watson, 1982). Results of this audit suggested that S02 measurement
uncertainty was approximately + 15 percent for 85 percent of the data and + 10
percent for 71 percent of the data.
Under the SURE program, a more comprehensive uncertainty analysis was
performed. When considering the SOz measurements from all ground-level
stations continuously operated from August 1977 through October 1979 (i.e.,
the Class I stations), the relative uncertainty at the 90th percentile of
*The ERAQS (Eastern Regional Air Quality Study) conducted between 1 January
1979 and 4 March 1980 served to extend some of the regional air quality
measurements of the SURE (Sulfate Regional Experiment) program conducted
between 1 August 1977 and 31 December 1978.
-18-
-------
concentrations (i.e., 17 ppb) was 18 percent and that at the 50th percentile
(3.5 ppb) was 86 percent. These large uncertainties at lower concentrations
were attributable to numerous SOa measurements at or below the quantifiable
limit. These large uncertainties may not be a problem in the COMPEX program
where filter sampling techniques are expected to provide a limit of detection
at 10 ppt.
Under the SURE measurement program, airborne S02 measurements were also
obtained over selected sites during intensive measurement periods (Blumenthal
et al., 1981). The S02 monitors aboard the two aircraft used in the study
were Meloy Lab flame-photometric analyzers (Model 285).
Continuous measurements were made in several spiral flight paths upwind
and downwind of the particular ground-level station. These data were then
averaged over 15 m vertical segments below 1500 m (above mean sea level) and
over 30 m segments above 1500 m. Ascent and descent rates were nominally 60
m/min below 1600 m and 120 m/min above 1600 m. Measurement data were thus
representative of approximately 15-second averages. An instrument time
response of 90 seconds (to 90 percent of concentration) (Blumenthal et al.,
1981) indicates that the concentration profiles are considerably smoothed over
the higher frequencies. Thus, uncertainties in continuous airborne S02
sampling have an additional response-time component. The uncertainties in
S02 concentration measurements from aircraft therefore require a very
detailed analysis of the accuracy of the high-resolution concentration field.
During the EPRI PMV&D Kincaid field measurement program, a large number of
quantitative audits of all the ambient air quality measurements were
performed. S02 measurements were obtained routinely over 5-minute and
hourly averaging times. Since the monitoring network was established within
20 km of a power plant, the measured SOz concentration levels and
uncertainty estimates are more applicable to the proposed COMPEX short-range
-19-
-------
experiments than are the ambient S02 measurements of the SURE and ERAQS
programs.
A summary of the audit information and estimated measurement uncertainties
for hourly SOz measurements are shown in Table 2-2, as reported by Smith and
co-workers (1983). Although the uncertainty estimates indicate improved
precision with increasing concentration, a representative maximum precision at
the highest observed levels (in terms of the coefficient of variation,. Cv)
is roughly 5 to 6 percent. Overall, a representative measurement precision is
approximately 10 percent.
2.5 Sulfate
As with SOz, the sulfate monitoring methods and associated uncertainties
can be summarized with respect to the measurements achieved during the recent
field measurement programs. Under the SURE and ERAQS programs, SOI
concentrations were analyzed from hi-vol particulate samples on a
daily-average basis and from sequential filter samplers (SFS) during limited
periods on a 2-hour basis (ERAQS) and a 3-hour basis (SURE). The hi-vol
samplers collect total suspended particulate matter with an aerodynamic
diameter less than approximately 30 um. The SFS collects particles in the
inhalable size range «11 urn) and, when used with a cyclone preseparator,
collects refined particulate matter less than 2 urn. Filters used with the
hi-vol and SFS instruments during the SURE and ERAQS programs consisted of
Teflon-coated glass fiber filters, which met stringent flow rate, ion content,
collection efficiency, and appearance criteria (Mueller and Hidy, 1983).
The precision of the SOI measurements is obtained by propagating the
uncertainties associated with the measured sulfate concentration on the
collection filters, and the measured flow rate. Volumetric flow rate
-20-
-------
TABLE 2-2
SUMMARY OF UNCERTAINTY ESTIMATES FOR
- AMBIENT S02 MEASUREMENTS
Variable
S02,
PMV&D
Network
S02,
D&M
Network
Concentration
(ppb)
50
>100
50
100
>300
LDL Bias
(ppb) U)
9 -6
0
16
+2
-2
-3.5
CV
<%)
10
7.5
18
8.8
5.8
90 Percent CI
<*>
-22
-13
-29
-17
-13.
to
to
to
to
.5
4- 10
+13
+33
+13
to +6.5
Key:
PMV&D Plume Model Validation and Development (EPRI)
D&M Dames and Moore
LDL Lower Detectable Limit
CV Coefficient of Variation
CI Confidence Interval
Source: Smith et al., 1983
-21-
-------
uncertainties of the Hi-vols pertaining to individual SURE sites ranged from
less than 1 percent to about 20 percent, with a representative value over the
entire network (Class I and II stations) of approximately 8.7 percent.
Slightly higher precision was achieved for the nine sites under the ERAQS
program. The precision of the flow rate measurements for the Class I SFS
instruments averaged about 3 percent during the SURE and ERAQS programs.
The uncertainty in sulfate measurements from hi-vol filter samples
typically ranaed from about 8 percent for ambient concentrations of
approximately 4 ug/m3 to about 1 percent for ambient concentrations of
approximately 40 ug/m3. Similar precision values were derived for both
sampling programs. Sulfate measurements from sequential filter samples ranged
from 6 percent for low ambient sulfate levels to about 1 percent for higher
ambient levels (Mueller and Watson, 1982).
Considering the sulfate variability in blank filters along with the
volumetric flow rates, the hi-vol sulfate measurement precision was 8.4
percent for the median concentration values (6.8 ug/m3) during the SURE
measurement program. SFS sulfate measurement precision for the median
concentration level of 5.4 ug/m3 was 9.7 percent (Mueller and Hidy,
1983). Under the ERAQS program, typical precision considering all measurement
and laboratory uncertainties was 8 percent for hi-vol samples, 22 percent for
SFS samples in the inhalable particle sizes, and 6 percent for SFS in refined
(<2um) size range. These values correspond to typical SO*
concentration levels of 10 ug/m3 (Mueller and Watson, 1982).
2.6 Implications to the Experimental Design
Precision of tracer and airborne S02 and SO* concentration
measurements can only serve as a rough guide in estimating the "measurement
uncertainty." For perfluorocarbon tracers, the information is incomplete,
-22-
-------
consisting only of a limited number of sample intercomparisons, and only at
background levels. Data for PMCP are not available. The high precision of
34S/32S ratio detection is based on the precision of the mass spectro-
graphic analysis technique, provided the sample size of sulfur oxides is
sufficient for analysis.
Finally, S02 and sulfate measurement precision is a function of ambient
concentration levels and the individual sampling instruments and analytic
procedures, and exhibits some variability. Concentration measurements under
the proposed COMPEX program should be performed using detailed QA/QC
procedures. Analyses of measurement precision should also be performed to
quantify the uncertainties associated with the various measurements, as was
performed under the SURE and the PMV&D programs.
-23-
-------
SECTION 3
LOCAL DATA ANALYSIS
The Electric Power Research Institute's (EPRI) Plume Model Validation and
Development (PMV&D) field study at the Kincaid power plant provides a data
base with which to examine the behavior of inert tracers for distances within
20 km of a source. Observations of several trace gases made using 18-20
collocated instruments offer a good opportunity to explore the statistical
characteristics of concentration time series for several different chemical
species in order to:
1) Determine if the proportions of the chemical concentrations remain
constant during travel from the stack to monitors within 20 km,
2) Examine and quantify the effects of measurement uncertainty and
background concentrations in the statistics and conclusions drawn
from the observations,
3) Determine if physical processes such as surface deposition, or
differences in the stack gas concentration fluctuations introduce
noticeable effects near {< 20 km) the stack, and
4) Analyze the statistical nature of concentration fluctuations that
arise in the concentration time series at and near « 20 km) of
the stack.
Analysis of these topics provides an evaluation of how well tracers represent
emission sources and how variations in sampling and averaging times affect
these simulations.
The Kincaid power plant is a typical large (>1200 MW) elevated, buoyant
point source of a variety that is thought to contribute a major portion of the
sulfur in the sulfate deposited to the surface over large regions such as the
northeast United States. The Kincaid PMV&D observations used in this study
consist of:
a) Short-term average (5 minute) concentrations of both SOz and
NOX at 18 sites for 34 weeks at distances ranging from 5 to
20 km from the source, and
-24-
-------
b) Over 100 hours of hourly averaged collocated SFS and S02
observations from 20 co-located sites (excluding mobile
observations).
The usefulness of the data base was established with careful quality
assurance procedures followed in the field, and quantitative estimates of
observational uncertainties. Such numerical analyses are valuable in
determining whether or not specific conclusions can be drawn from the data.
The present analyses was built on a study of collocated SF6, SOz, and
NOX, data reported by Bowne (1982). Bowne's findings were implemented and
additional analysis of the effects of measurement uncertainty was performed
using comparisons of five-minute observations of SOz and NO*
concentrations. The effects of background on conclusions are discussed along
with the question of the usefulness of inert tracer releases for studying near
source S02 and NOX dispersion.
3.1 Characteristics of the Kincaid Data Base
The data base selected for analysis is a subset of the Kincaid experiment
data base (Bowne, 1982) and includes emissions and stack parameters, including
the release rates of tracers, and ambient concentrations of S02 and NO*.
Processing of the data included the development of analysis statistics and
relative concentrations.
The SOz and NOX stack observations were gas concentrations drawn from
the stack. Observations were reported as five-minute averages. When the
monitoring instruments were not operating, the SOz and NOX were computed
for hourly averages using the plant load data. SFs releases were recorded
as mass flow rates based on instantaneous flow readings taken from a
rotormeter at least once every hour. The stack gas velocity measuring device
did not operate properly so hourly averages were obtained using plant input
data. The uncertainties for the stack parameters are presented in Table 3-1.
-25-
-------
TABLE 3-1
UNCERTAINTY IN THE HOURLY AVERAGES OF THE VARIOUS
STACK PARAMETERS FOR THE KINCAID SITE DURING 1981
(Source: Smith et al., 1983)
Parameter
SOz (measured)
NO*
vs
QSF
Units
ppb
ppb
m/sec
g/sec
Uncertainty Interval (90
S02 (true) = 1
NOX (true) = 1
Vs (true) = 1.
QSF (true) =
.04 SOz (obs) + 0.
.10 NOX (obs) + 0.
03 Vs (obs) + 0.17
QSF (obs) +0.05
Percent)
1 S02 (obs)
15 NOX (obs)
Vs (obs)
QSF (obs)
S02 (calculated) g/sec
Qso
g/sec
S02 (true) = 1.06 SOz (calc) + 0.18 S02 (calc)a
Qso (true) = Qso (calc) + 0.27 Qso (calc)b
2 2 Z
QNO
g/sec
QNO (true) = QNO (calc) + 0.32 QNO (calc)1
*Using measured stack SOz values.
bComputed by using bias-corrected SOz, NOX, and Vs, otherwise bias is
+0.07 S02 and + 0.13 NOX.
-26-
-------
Ambient SOz and NOX samples were drawn from a single sampling manifold
at a height of 3 m. The data polling rate was once every 10 sec for both
species. The major differences between SOj and NOX instruments is that
the SOz instrument rise time was much larger than that of the NO* device.
Rise time was exponential and took over 20 minutes to reach 95 percent of an
input step (189 ppb) in concentration. This effect causes an underestimate in
the average SOz concentration of about 16 percent for a concentration spike
lasting for five minutes or less. In analyses, care should be taken to avoid
selecting periods when the elevated concentrations persist for only a single
five-minute averaging period. The observations of both S02 and NOX were
routinely stored as five-minute average observations.
The SFs observations were made from the same air volume as the SOz and
NOX observations. The sample consisted of a one-hour integrated sample of
two-second air samples made every 20 seconds. The sampling technique
introduces some artificial sampling "diffusion" where peaks are reduced and
some zero concentrations are made slightly non-zero. The hourly average
uncertainties are summarized in Table 3-2.
3.2 Conservation of Tracer/Pollutant Concentration Ratios
3.2.1 Comparison of SOz and NOX Concentration Time Series
The underlying hypothesis in any dispersion analysis for conservative
pollutants is that a puff of air being sampled at a downwind receptor contains
the same proportion of constituents as it did when it left the stack. Under
normal plant operations, the time series of SOz and NOX concentrations at
the stack varies slightly over the course of 6 to 10 hours as demonstrated by
Figure 3-1. The concentration time series at a downwind monitoring site is
substantially more variable and intermittent at averaging periods of 5
minutes. Two questions were addressed: 1) does the proportion of S02 to
-27-
-------
TABLE 3-2
UNCERTAINTY IN THE HOURLY AVERAGES OF SF6, SO-, and NOX
CONCENTRATIONS OBSERVED AT KINCAID MONITORS DURING 1981
(Source: Smith et al.,. 1983)
Parameter* Uncertainty Interval (90 Percent)
SFS SF6 (true) = SFS (obs) + 10 ppt, SFS (obs) < 100 ppt
SFS (true) = SF6 (obs) + 0.10 SFS (obs), SFS (obs) > 100 ppt
S02 SOZ (true) = 0.94 S02 (obs) + 0.16 S02 (obs), at 50 ppb
S02 (true) = S02 (obs) +0.13 SO; (obs), S02 (obs) > 100 ppb
S02 (true) = 1.33 S02 + 0.63 S02 (obs), at 9 ppb
NOX NOX (true) =0.98 NOX (obs) +0.29 NO* (obs), at 50 ppb
NOX (true) =0.98 NOX (obs) +0.19 NOX (obs), at 100 ppb
NOX (true) =0.98 NO* (obs) +0.15 NOK (obs), NQX (obs) > 300 ppb
*Lowest detectable limit for S02 is 9 ppb and for NOX 11 ppb. TRC
suggested a value for SF6 of 2 ppt.
-28-
-------
TIKE
STRRTING DRTE: eszsei
STRRTING TIKE: 1400
ENDING DRTE: ussssi
ENDING TIKE: 1900
-------
NO* present at the stack change between the stack and a nearby monitoring
site, and 2} do the SO- and NOX time series correlate very closely with
one another? Answers to these questions aid in understanding over what
distances pollutants can be simulated by tracers.
Preliminary evidence to answer the questions was gathered by visually
inspecting time series co-plots of S02 and NOX at various monitoring
sites. The travel time for plume material is less than an hour so chemical
transformations and surface deposition are not expected to be significant.
Under such conditions, S02 and NOX time series should agree closely.
Figure 3-2 demonstrates that a close agreement does in fact occur between the
two time series at a site. In fact, over 30 time series plots at various
sites and times all demonstrate a close correspondence between S02 and NOX
concentrations when they are significantly above the background concentrations.
The background concentrations arise from several sources. For NOX, the
background concentration varied widely among the 18 monitoring sites. The
background concentration fluctuations are more constant with time
(Figure 3-3). Background can easily be removed by visual inspection.
However, an objective method of removal is somewhat more difficult,
particularly as the averaging times become longer and the ratio of the
concentration standard deviation to the average background (ac/cb)
approaches one. In the present analysis, either a station where background
was not significant was selected for analysis, or else background was removed
as a long-term average when the concentration averaging time was relatively
short, e.g., <1 hour. The S02 and NOX background as the median
concentration at each station is compiled in Table 3-3.
The proportion of S02 to NOX at the stack in Figure 3-1 was found to
be 3:1. when a background of 10 ppb of NOX was subtracted from the MOX in
Figure 3-2, a constant ratio between S02 and NO* at the downwind monitor
-30-
-------
ta.
LT
CD
O_
O_
-63
CO
-S3
-c
CD
CL.
o.
TIME
STRRT1NG DRTE. 052551
STflRTING TIME: i1E0
ENDING DflTE- 052581
ENDING TIME: 1900
O S02
A MOX
FIGURE 3-2. Time series of five-minute average S0? and NO concentrations
for station 1422 on 25 May 1981. L x
-31-
-------
laae.
1100.
1200.
nee
1400.
TIME
1500.
16BE.
STflRTING DOTE" 052881
STORTING TIME' 1000
ENDING DfiTE' 052861
ENDING TIME' 1800
O 502
A NIK
X SF6 COLQC
1700.
HRLY
FIGURE 3-3. Time series of hourly concentrations at station 1422, 28 May 1981.
-32-
-------
TABLE 3-3
THE MEDIAN FIVE-MINUTE AVERAGE S02 AND NOX DATA AT THE
ROCKWELL SITES, MAY 11, 1981 THROUGH MAY 31, 1981
Station Number S02 NO,
0424 0.0 5.0
1118 0.0 8.0
1160 0.0 7.0
1244 0.0 1.0
1335 0.0 9.0
1422 0.0 6.0
1650 0.0 13.0
1713 0.0 1.0
2019 0.0 0.0
2744 0.0 12.0
2832 0.0 5.0
3829 0.0 4.0
5146 0.0 8.0
5318 0.0 7.0
5623 0.0 1.0
5745 3.0 4.0
6052 0.0 4.0
7134 0.0 0.0
-33-
-------
was also 3:1. Figure 3-4, showing five-minute concentrations for a different
day and station (28 May 1981 Station 1118), indicates similar results when
observations are compared with the stack concentrations in Figure 3-5.
Results show a one-to-one correspondence between NOX and SOz concentration
maxima and minima, and an SOz-to-NOx ratio of 4:1 at both stations and the
stack.
The strong agreement of the five-minute averages is apparent when
statistics are performed on the five-minute averages for tracer tests and
SOz and NO, are available at the same times and locations. Table 3-4
presents a summary of these statistics. The most notable features are as
follows:
The correlations, particularly for monitoring stations unaffected
significantly by the NO* background, are in excess of 0.90.
The higher-order statistical moments of the concentration
probability density function (pdf), such as skewness and kurtosis,
are remarkably similar for SOz and NOX.
The SOz and NOX concentration variations correlate closely
(Figure 3-6} despite the fact that the proportion of S02 to NOX from the
source can change. This agreement suggests that for short travel times NO*
is a good tracer for SOz, and that the proportion of SO; to NOX does not
change greatly at Kincaid.
The close tracking of SOz and MOX is not as apparent in hourly
averages. Figure 3-7 shows a time series comparison of hourly SOz, NO*,
and SFs for the two days previously discussed using five-minute average
data. The SOz and NO* curves are roughly parallel, but the concentrations
have been reduced by averaging so that they do not show as great a difference
with the background. Table 3-5 shows several sets of station statistics that
are not significantly affected by the NOX background. In this case, the SO
-34-
-------
(D CD O O O,
1400.
TIME
STARTING DRTE- 052581
STARTING TIME: 1100
ENDING DOTE" 852581
ENDING TIME: 1900
O S02
A N0<
0 and NO concentrations
FIGURE 3-4, Time series of five-ninute average SO
for station 1118 on 25 May 1981. Background concentration* is denoted by
the dashed line.
-35-
-------
1400.
TIME
3TRRTING DRTE: 852581
3TRRTING TIME: 1033
EHDING DRTE: aszsai
ENDING TIME: 1800
2STRCK 302 HRLY
STRCK NO HRLY
X STRCK 3F6 CRLC
FIGURE 3-5. Time series of hourly concentrations at the stack on 28 May 1981.
-36-
-------
z
o
IH
'HI
CO Z
Z U
O -!
t-l VO
< t
> o
OC Z
Ul
a <
02
xz
o o
K-
o <
C o:
IB t-
Z
IMUI
O U
00 Z
o
Ul O
o
< 0
a z
U =3
U a.
ui it a.
r> < «»
Z rt"i c
CD »«
H4 (Q
X x
i o x
m z o
2?
XX-
r- h- N
C9 Ul CO
Z CC
t-l Ul t-
X 00 I/)
oz ->
uo
OO V- 3
Ul <
3 00
< t-
UJ <
X
a
CJ
00 00
H-» Ul
1- Qc
CO Ul
o
u. <
O Qe
t
oc u
X ^
=3 <-)
CO Ul
a.
< oo
O
z
ae
i
X
*i
Ul
o
AJ
5
(N
o
CO
X
If Z
CO
CO
IB
a
co CM CM ^ o rx in
o in o - CM o
>- ii i
c^ co in CN ^- m rx
O CO O t CM O
i i i i
CO lO CO ^ V 00 30
r- CO O O O
33 CM CO CM CM 33 r.
O CO O O O
M 33 fx fx ^ (J^ Q\
O O O O O O O
O O O O O O O
cjt »^ in CT* **i ^» m
^ in MO CM ^- *o f**
o o o o o o o
10 ai o m »- CM ci
M CM cr. r> »o CM
o o o o o o o
CM in m ^ m
r " t *» «» CM
O O O O O O O
o» o *r CM ^ ao «
co CM in
r* f«» o CM o en
CM in « CM 33 C-I Cx
00 33 33 CM 00 00 33
CO CO CO O "^ *~ ^
in m in 10 * *-
^X ^. \ Vx Xx x^ Xx
CM CO CM CO 00 in 33
1 f * CM CM CM Csl
Xx ^ X, ^ -x, Xx >x.
m m m m m in *o
O en o Cx 33 oo co
o *^ o o * o o
1 1 1
O C\ ^- CM fx If
o o o o o
I II II
fs. ^ cj% 10 in CM *
0 - 0 0 - 0 *-
co » ao co ao in r*
O O O O O 0
O IO 30 U? CM 00 *~
O O O O O O O
o o o o o o o
w *o CM 10 CM o r*»
m ^ \o CM in ro co
o o o o o o o
in co ~- ^- in 33 -
in CM a* o « o 33
o o o o o o o
cj^ vO ^ ^o m ^ m
^ (^ F r^ « r
0000000
33 in rx in CM o
in CM m ^- in v CM
in r^ 9t o1^ 10 iO ^o
CM CM in « CM m co
en CM CM cj* m v CM
CM CM ^ CO ^ CO
r^ w «- o co CM 33
^- ^- . CM * CM
33 33 oa ao aa ao 33
00 in o 00 00 33 33
CM CM CM CM CM CM CM
m in tn in in in in
-------
A. Station 5318 (N = 194)
B. Station 2019 (N = 61)
1C.0 £2.0 122.0 1;2.0
05£E?A'ED 532 CPP2)
3.0 £2.0 92.3 12:. 3
OBSERVED 532 rpP5)
C. Station 1118 (N = 219)
D. Station 1422 (N » 114)
60.0 12?.3 152.0 2-.2.S
OBSERVED 522 CPP5)
I I I I | I I I I 1 I i I i l ; i 1 ; , . :
0.0 63.0 123.D IE2.3
OBSERVED SG2 CPPB)
FIGURE 3-6. Scatter plots of five-minute average SCL concentrations versus N0x
concentrations for various EPRI PMV&D monitoring sites for tracer tests
conducted 12 May to 1 June 1981.
-38-
-------
1000
STRRTINC DflTE: 052881
STflRTING TIME" 1000
ENDING DBTE- 052881
ENDING TIME: 1800
x sre COLOC
FIGURE 3-7a. Hourly average S0
1422: '28 May 1981.
SF,, and NO concentrations at station
b X
-39-
-------
s>4_
fsj
"in..
x
o
1100.
1500.
1600.
TIME
1700.
1800.
b
STflRTING ORTE" 052581
STRRTING TIMEt 1400
ENDING OflTE: 052531
ENDING TIME: 1900
O S02
A N0<
X SF6 COLOC
HRLY
FIGURE.3-7b. 25 May 1981
-40-
-------
1-
^f
IH
W
Z
O
oa
o -o
O 0
U. Z .C
O V)
Q4 CU
Z X 1-
O 1 £
10 i
IH LU *-*
a: a:
< UI
a. x
§ ^
U CO
<0
IH
ae
0
^
x
o
z
o c
f
«/»
o
c
o
u
Jj
CO
^i
03
00
n
r*
CO
>fl
00
CM
«*
at
CM
03
00
O
O
00
tjO
o
CM
^f
CM
O
O
v~
o
,J
to
in
^
m
P*
at o to CM at
^ 1^ ^ CM to
o o
iii i
at in
~- m CM 03
<-^ 0 0 0 0
^ iO OQ ^ *~
*- tO 03 O 0
*
^- O O O
1
(T» ^* O 00 00
O O O O O
* «-~ ui ro
a« *cr wi CM ao
«9* (NI Csi ro
GO m « ao co
«T r*> <*- «»t «
v MD « «P~ *~-
U3 QO f** UI vO
rx u) » ^ ^
m CM N CM CM
O O^* UI ^* ff\
"
* * *
ao oo v ra a*
* ^ ^ (M ^
W CM
-------
2/NOx correlations are in excess of 0.90 and the higher moments of the
concentration distribution agree closely. The scatter plot in Figure 3-8
shows similar agreement as that obtained for the five-minute averages. The
major difference between hourly and five-minute averages for SOz and NOX
is that the influence of background concentrations becomes more significant
for the hourly averages. In low background areas, the NOX could still act
as a suitable tracer for S02.
3.2.2 Comparison of SF$ and S02 Concentrations
Analyses have compared normalized concentrations (X/Q) of S02 and
SF« at the same times and locations (collocated observations). The main
finding of the analysis is that a large scatter occurred in the co-plot of
X/Q (for S02) even when questionable data points were screened
(correlation coefficients of 0.72 were estimated). The report did not resolve
all of the questions about the suitability of SFS as a surrogate for S02.
Clearly, if the proportion of an inert tracer to an inert pollutant (SOz for
the short travel times) cannot be maintained in a parcel of air, then the
whole experimental rationale for using a tracer becomes questionable. The
question of the usefulness of the SFS tracer at Kincaid is reevaluated in
this section.
The approach used in this analysis is to factor in the influence of
measurement uncertainty in both X and Q. The 90 percent confidence
intervals were used to plot confidence regions around each relative SF«
concentration/relative SOz concentration pair. Additional screening removed
observations:
Where large gradients of tracer concentration (>50 ppt/km)
occurred (timing errors become significant).
Where the ratio of emission rate of SFs to SOz changed rapidly
(e.g., 1300-1500 of May 13, 1981).
-42-
-------
A. Station 1422 (N = 14)
er-.c
to
CU
£ 60.0
x
o
LU
IX.
u
13.E
S2C.C
"" I I I I I I I I f I I ! I I ' I I ! I I I I I
23.0 S2.3 £2.3 £2.3
OSSEPAcD £22 Cr = 3)
B. Station 5318 (N « 17)
£.T
U. . I
CD
£ 62.E
x
o
u
>
u
« ->-
co 2t
o
I I I I I I I I I I I I I I I I I I I I I I I
20.0 40.0 6S.B 63.0
035Er,l'fD 502 TPPS)
FIGURE 3-8. Scatter plots of hourly average concentrations of S0« and NO for
two EPRI PMV&D monitoring sites at Kincaid, Illinois, where the background NO
was not significant during tracer tests conducted between 12 May and 1 June 1981,
-43-
-------
Where S02 observations contain position biases.
Where background concentrations were appreciable.
Where the SOz peaks were temporary (<10 min) (the S02
instrument may not.have responded properly).
Removal of these X/Q observations resulted in a set of 51 observations,
which are plotted with their 90 percent confidence intervals in Figure 3-9.
Despite the scatter, very few observations have confidence regions that do not
intercept the one-to-one correspondence line. The confidence regions are so
large as to still allow an appreciable amount of scatter (Figure 3-9).
The X/Q statistics for both SF6 and S02 are compared in Table 3-6.
This table shows that the bias is not significant at the 95 percent level and
that the correlation coefficient is 0.85. The Kolmogorov-Smirnov test reveals
that the two distributions of X/Q are similar at the 96 percent confidence.
The statistics all suggest that within 20 km of the source, on the average,
the proportion of S02 to SF6 in a parcel of air remains constant and SF6
acts as a suitable tracer.
3.3 Statistical Nature of Concentration Fluctuations
A power spectrum of fluctuation of both S02 and NOX was computed from
three weeks of five-minute averages during the third intensive (May 11, 1981
through May 31, 1981) of the PMV&D study. The spectra of the two species are
shown in Figure 3-10 for one station. They do not exhibit a substantial
background contribution. The most important feature that emerges is that the
two spectra are parallel to one another over essentially the entire range, of
frequencies (periods of 10 minutes to 3 weeks).
The natural uncertainty in estimating hourly or longer averages was
examined using S02 averaging periods up to 24 hours. The approach used
ignored serial correlation and nonstationary effects and estimated a lower
-44-
-------
4.00
gyiiSvmi 11 ii 111 ii 111 ii i M 11111
r.00 0.50 1.00 1.50 2.00 2.50
NORMflLIZED S02 CONCENTRHT
3.00
ION
3.50 3.00
5CRTTER PLOT OF NORMflLIZED 502 MS SF6 CONCENTRRTI ON
FIGURE 3-9. Scatter plot of normalized hourly averaged 502 con~
centrations U-0~7 s/nr) versus hourly averaged SFs concentrations
(lO'7 s/m3). Sample size is 51 cases, and the shaded region
indicates the 90 percent confidence level.
-45-
-------
TABLE 3-6
COMPARISON OF S02 AND SFS NORMALIZED
CONCENTRATIONS (X/Q), IN 10"7 s/m3
95 Percent
Statistic S02 SFS Confidence Interval
Sample size 51 51
Average 0.52 0.63
Standard deviation 0.33 0.39
Skewness 1.61 1.31
Kurtosis 3.54 2.16
Correlation Coefficient (0.74 < 0.85 < 0.91)
Average residual (-0.23 < -0.11 < 0.01)
-46-
-------
JE-»
lE-3
FREQUENCY (HZ)
FIGURE 3-10. Concentration spectra recorded at station 1422, 11-31 May 1981,
-47-
-------
limit on the confidence interval of the mean using Student's t statistics.
The coefficients of variation for the averages are tabulated for selected
monitoring sites in Table 3-7. There seems to be no systematic change in the
natural uncertainty with downwind distance. The natural uncertainty remains
relatively constant for all averaging periods.
The probability density functions (pdf) of the five-minute averaged SOz
concentrations at a near-source site and a far-source site were fitted to an
intermittent exponential distribution like that suggested by Barry (1974).
The corresponding best fit for three weeks of data indicates an intermittency
factor of 0.05 for the near-source site, and 0.05 for the far-source site.
The pdfs and the fitted model are shown in Figure 3-11. The fit of the pdfs
suggests that the intermittency and the peak concentrations do not
systematically decrease with increasing downwind distance between 7 to 20 km.
3.4 Implications for Experimental Design
A local data analysis was performed to evaluate the interrelationships of
pollutants and tracers and their spatial and temporal characteristics. The
analyses were directed at determining the ability to use tracers in
experimental programs such as COMPEX. From an analysis of portions of the
EPRI PMVS.D data base, consisting of SF6, S02, and NOX concentrations
distributions within 20 km of the Kincaid Generating Station, the following
results have been obtained:
1) The 5-minute and 1-hour average SOa concentrations and those
NOX concentrations in excess of the 5-15 ppb WOX background
correlate quite closely and show no systematic phase lag.
2) When data from biased monitoring sites, low concentrations,
spatially isolated concentrations, and other justifiably
untrustworthy data are removed, the normalized concentrations
(X/Q) of the SOi and SF8 measured at the same site and time
agree indicating no measurable depletion on a local scale.
-48-
-------
TABLE 3-7
THE AVERAGE AND STANDARD DEVIATION OF THE COEFFICIENT OF VARIATION
AS A FUNCTION OF AVERAGING PERIOD. ALL COEFFICIENTS OF VARIATION
STATISTICS ARE ESTIMATED FROM SETS OF 5-MINUTE AVERAGES.
Station
0424
0118
1160
1422
1650
Average
1 hour
2.1
1.3
1.7
1.6
1.6
1.7
(1.7)
3 hour
3.4
2.0
2.6
2.9
3.0
2.8
(2.4)*
Average
6 hour
5.3
3.2
3.9
4.3
5.1
4.4
(4.2)
Standard Deviation
12 hour
7.6
4.5
5.7
6.3
7.1
6.2
(5.9)
1 hour
1.2
1.0
1.0
1.0
1.0
1.0
3 hour
2.0
1.6
1.7
2.1
1.8
1.9
6 hour
2.7
2.5
2.3
2.6
2.4
2.5
12 hour
3.5
3.7
3.2
3.7
3.3
3.5
Coefficient of variation assumes uncertainty is constant with averaging
period.
-49-
-------
* Intermittent exponential CDF
Observed CDF
80
98 99 99.8 99.9
(a) Station 1713
99.99
FIGURE 3-11. Cumulative distributions of 5 minute
averaged S02 for two EPRI PMV&D sites at Kincaid.
The intermittency factor, A, is approximately 0.05
for both sites. The intermittent exponential CDF
function was fitted to both sites.
-50-
-------
-- Intermittent exponential CDF
Observed CDF
80
70
60
50
O.
-40
CM
O
30
20
10
90 95 98 99 99.8 99.9 99.99
(b) Station 1160
FIGURE 3-11.(concluded).
-51-
B4210
-------
3) There is a large degree of uncertainty in the estimates of X/Q
due to measurement uncertainty.
4) There seems to be no significant systematic decrease in S02
concentrations due to dry deposition with downwind distances
ranging from 7-20- km, nor does the statistical intermittency of
surface concentration time series vary substantially with downwind
distances. Additionally, the analysis indicates that the
proportion of NO* to SOz remains constant between the stack
and downwind monitors.
5) The increase in averaging time does not significantly reduce the
90 percent confidence interval about the estimated mean. The
presence of many zeroes in the concentration time series seems to
contribute toward this result.
On the basis of these results, the strategy of tagging S02 emissions
with inert or reactive tracers appears to be appropriate. However, it is
important to carefully control the tracer emission rate so that its
variability accurately reflects the variability in S02 emissions.
Additionally, thorough quality assurance and quality control measures,
including frequent instrument audits and replicate sampling should be
implemented during the monitoring activities in order to accurately
characterize the measurement uncertainties.
-52-
-------
SECTION 4
REGIONAL DATA ANALYSIS
The analysis of the Kincaid data set discussed in the previous section was
beneficial in describing the local scale spatial and temporal variability of
pollutants and the feasibility of tracer use in small scale experiments. To
establish a suitable sampling network on a regional scale it is important to
understand the important temporal and spatial scales for SOz and SOS.
The most notable regional pollutants study was the EPRI Sulfate Regional
Experiment (SURE). This section describes the SURE data set and analyses of
the data to provide information on:
The spatial scale of concentration data.
The temporal characteristics of concentration data.
The spatial representativeness of single station concentration
measurements.
Uncertainties in network design.
Implications of the findings on experimental design are also discussed.
4.1 Description of the EPRI SURE Data Base
The EPRI/SURE monitoring network consisted of 54 monitoring sites
distributed over the eastern United States as shown by Figure 4-1. Monitoring
started in August 1977 and it ended on October 31, 1978. Of the 54 sites, the
first nine were designated as Class I (primary) sites where additional
aerometric and meteorological variables were observed over the remainder or
Class II sites. A summary of the number of observations for selected stations
is presented in Table 4-1. Generally, the Class I stations have about three
times as many observations as the remaining stations. Most of the EPRI/SURE
sites were located in a position that on the average is either upwind, or
-53-
-------
FIGURE 4-1. The EPRI SURE network of monitors. Sites numbered 1-9 are
class I stations.
-54-
-------
TABLE 4-1
THE NUMBER OF HOURLY S02 OBSERVATIONS AVAILABLE
FOR SELECTED EPRI/SURE SITES
Station Class Number of Observations
1 I 9545
2 I 7031
9 I 9558 8660 observations average per station
12 I 8512
13 II 3640
15 II 3651
18 II 3789
20 II 4431
22 II 3699
28 II 4031 3334 observations average per station
37 II 1455
40 II 2627
42 II 2308
43 II 4005
51 II 2099
-55-
-------
remote from large point sources of S02. The individual sites are described
in some detail by Mueller and Hidy (1982). The distances between sites range
from 28 km to 2371 km. The number of pairs of stations that fall within
certain ranges of distance are presented in Table 4-2.
The accuracies of the various measurements of interest in the present
study are summarized in Table 4-3. The 24-hour particulate SC>4
concentrations were slightly biased since only particles exceeding a certain
diameter were collected efficiently. The external data quality audits and
comparisons of the data with data from other programs are reported by Mueller
and Hidy (1983).
4.2 Spatial Scale of Concentration Data
Hourly SQz concentrations and 24-hourly SOz and sulfate concentrations
are provided for the 54 stations of the SURE network. Correlation and
autocorrelation techniques are used in the following three subsections to
examine the spatial scale of concentration data and the representativeness of
concentration measurements.
4.2.1 Analysis of One-Hour SOz Concentrations
The one-hour averaged S02 concentrations at the nine Class I sites were
examined for their statistical characteristics. Large concentrations were
rarely encountered, which confirms that their placement was far away from
large point sources. The probability density function of 1-hour average SOa
concentrations at many of the sites seems to follow an intermittent lognormal
distribution like that described by Netterville (1979). The intermittency
factor ranges typically from 0.5 to ~1.0 for the Class I stations.
Figure 4-2 shows the frequency distribution for site 4 as an example.
-56-
-------
TABLE 4-2
THE NUMBER OF STATION PAIRS AS A FUNCTION
OF THE DISTANCE OF SEPARATION
Range Number Percent
0-200 km
200-400 km
400-600 km
600-800 km
800-1000 km
1000-1200 km
>1200 km
81
219
250
235
182
140
324
5.6
15.3
17.4
16.4
12.7
9.8
22.6
Total 1431 100
-57-
-------
Station
TABLE 4-3
A SUMMARY OF THE ACCURACY OF S02 AND S04 OBSERVATIONS
MADE AT THE CLASS I EPRI/SURE NETWORK
(Excerpted from Mueller and Hidy, 1983)
1-Hour Average SOz
(LQL = 3ppb)a
Bias (%)*
90% Confidence
Interval (%)
24-Hour Average Hivol S04
(LQL =8.4 uq/m3)b
Bias (%)*
90% Confidence
Interval (%)
1
2
3
4
5
6
7
8
9
8.0
5.7
1.7
4.3
0.4
8.4
1.4
-2.9
-9.6
4.3
3.4
4.6
9.7
2.3
5.7
4.4
2.3
4.8
-3.6
-3.0
-5.3
-4.6
-8.4
-15.2
-8.1
-3.0
-2.7
5.4
6.4
3.3
6.2
10.0
1.9
7.7
2.2
4.6
* Bias {%) = 100 (ERT-EPA)XEPA
a Concentrations range from 30-500 ppb
b Assume accuracy is determined by volume flow rate rather than lab analysis.
-58-
-------
90
70
60
50
40
30
20
2 10
CM 9
O g
I/I O
7
6
5
0.1 0.2 12 5 10 20 30 40 50 60 70 80 90 95 98 99 99.5 99.9
Cumulative probability (S)
FIGURE-4-2. Cumulative distribution of 1-hr averaged S02 at EPRI/SURE
station 4 on lognormal probability paper.
-59-
8^210
-------
Some of the EPRI/SURE stations exhibit signs of cycle in the SOa
concentrations, as shown in the autocovariance plot for the Scranton,
Pennsylvania site (Figure 4-3). The autocovariance indicates that the
fluctuation is a 12-hour cycle. Such variations were discovered at several of
the class I sites. Presence of the cycles could possibly mask spatial
correlations in S02 concentrations over large distance scales and should be
considered in analyses.
The correlations of the one-hour average SOa concentrations between
pairs of S02 monitoring sites were estimated for all possible station pairs
of the 54 stations. Correlations averaged over 200 Jon station separation
categories are shown in Figure 4-4 as a function of the site separation
distance. The maximum correlation coefficients are approximately 0.6. It
appears that the SURE network resolution is insufficient to fully describe, the
relationship of correlation with distance. Visual examination of the spatial
correlations such as those shown in Figure 4-5 leads to the speculation that
the scale of SOz variations changes with location in the EPRI/SURE network.
An analysis to group stations by geographic region was pursued, but no
statistically significant systematic changes were found in the shape of the
autocorrelation function for various station groupings.
An analysis was done to examine how accurately a number of SO2
observations made at several points within the region can be used as an
estimator of the true spatial mean within the region. This is another means
of estimating the spatial representativeness. For this analysis two arbitrary
regions were selected in order to do the estimates. These two regions consist
of the Ohio River valley region and a region covering New York and western New
England (Figure 4-6). Eight stations within each of the regions were used in
the analysis. The average concentration as well as the coefficient of
-60-
-------
1200 "
g: 600
OJ
u
5 300
a o.o
-300
-600
4-
12
15
18
21 24 27
Log (days)
30
33
36
39 42
45
48
FIGURE 4-3. Autocovarience of hourly averaged SOp concentration observed at
ribUKt 4-0. Huzocovanence or nouriy averagea 5Up
the Scranton, Pennsylvania site (EPRI/SURE site #2).
-61-
81+210
-------
0.6
0.6
0.4
- 0.2
tr
ce
0-0.0
-0.2
-0.6
-0.8
-1.1
1 T
EPR1 SURE HOURLY S02
J I I
400. 600. 1200. 1600. 2000. 2400.
DISTflNCE (KM)
FIGURE 4-4. Observed spatial correlations of hourly S(L versus distance.
-62-
8<4210
-------
- -o.oi
/ -0.05 ,-' ' J
13 -- '
'1J - O.OB ....__-Oc01
~
0.70
o.is .
0., ' o.oo
, o-36 : o.oa
. 0.08
0.1B 0.24 Q.020.19 / _______ --- -
0.19-
0.19
*.""- 0. 15 .-V^ '
. ,..' 0.29 .r
: -< - - ..._.." nil
0.17
?^ ,--' °-10
0.05 ---- -- ' ...... ""'
'
'v- 0.04
0.08 \
.X
FIGURE 4-5a. Spatial distribution of correlations between station pairs:
western region (the reference station is indicated by^).
-63-
-------
i-^.---" "-x
\.. ^ . '~'
"l /-'^'"" ."I''
0.16
0.29
0.27 .
0.3B
vO.ll
\ 0.13
0. T^
0.30
0.
0.30
r
t~-0v
0.14
0.27 ./'
"" OV09
0.03
0.29
-0.05
'r i -0.04
*\
0.14 0.1?0-H~'
0.03 \0.22 __^.t
0.04
0.03
0.09
0.1B
"o.ii
FIGURE 4-5b. Mid region (the reference station is indicated
-64-
-------
^v^--
0.11
0.28
0.23
0.12
r~~
0x13
0.11
°-28
0.04
0.00'
°-17 0.090.14 /
o..io' -ore
0.03
. % ,.; _O._QO
, r_ -J1..05
0.24
0.36
0.17 X,<
-0.05
'*'!%
-Of '02 -0.02 ,i'
r ' _0.-o,-
-0.03
0.04
~ ~~ j.
0.00
I s
0.06
'0.14
0.49
FIGURE 4-5c. Eastern region (the reference station is indicated
-65-
-------
47
I-*"
V
49
.J
-£
>
f
!
1
i
»e
1
\
/
/
"t
\r
X
FIGURE 4-6. A map of the EPRI/SURE site locations occurring within the
two study areas.
-66-
64210
-------
variation (cov), estimated as oc, were computed for all eight stations.
The true average over the region is given by a weighted sum of the
observations, where the weights could be dependent on the area where each
point measurement is assumed to hold. Determination of the weights can
introduce one source of uncertainty into the analysis. Another source of
uncertainty arises from the degree of confidence that one can make in
determining the true mean from the eight observations on either regular or
irregular spaced grids. If the observations have no persistence (serial
correlation) then the 95 percent confidence interval for the estimate of the
mean can be computed using Student's t statistic as:
obs [1 - 0.72*COV] < true < obs U + 0.72*COv]. (4-1)
Serial correlation increases the confidence interval (Morris and Ebey,
1984) so the range given in Equation 4-1 represents a lowest limit on the
uncertainty. The average value of 0.72*cov is 0.62 for the New York region
and 0.66 for the Ohio River valley region. The standard deviation of 0.72*cov
is large, typically 1.5.
Several averaging periods are being suggested for the various trace gases
of interest in COMPEX. For example, hourly averaged S02 concentrations and
six-hour averaged tracer observations are design possibilities. As the
averaging period increases, the short-term variations are smoothed out,
leading to a reduction in the variance. An obvious design question is how
rapidly the variance changes with the averaging rate. In order to study this
question, the SOz data were averaged over 6-, 12-, and 24-hour periods. The
results indicate that in general the SOa variance follows as relationship of
the form:'
-67-
-------
Var(S02 N-hour average) = Var(S02 1-hour average)/!*!" (4-2)
where a ranges from 0.1 to 0.2. A summary of the variance for the averaging
periods and stations is presented in Table 4-4.
4.2.2 Analysis of 24-Hour SOz Concentrations
Averaging of hourly SOz data for 24-hour periods provides a data base in
which diurnal cycles are removed and which can be used for correlation
analysis and comparison to 24-hour average sulfate concentrations. The
correlations between all possible pairs of the 54 EPRI/SURE stations were
computed. A scatter plot of these correlations as a function of distance
between station pairs shows much scatter and only a slight variation of the
correlation coefficient with downwind distance (Figure 4-7). As in the hourly
case, grouping stations by geographical region did not provide any significant
reduction of scatter. A model for the correlation coefficient of the form:
p(d) = exp(-d/d0),
where d is the distance between stations and do is some characteristic
distance, did not provide a satisfactory fit. Estimates of the mean squared
error (MSE) are relatively constant at all ranges of station pair separation
distance. The results suggest that for 24-hour averaging the spatial
variation in SOz concentrations is quite large even for separations of the
order of 150 km. Data for smaller spatial scales are not available from SURE.
4.2.3 Analysis of 24-Hour Sulfate Concentrations
Although both 3- and 24-hour average S0« observations were made, the
3-hour average S04 data only was collected for six months at the class I
sites and the analyses were performed only on the 24-hour averaged S04 data.
-68-
-------
TABLE 4-4
THE VARIANCE OF S02 (ppb) FOR VARIOUS
EPRI/SURE STATIONS AS A FUNCTION OF AVERAGING TIME
Station
1
2
4
9
12
13
15
18
20
22
28
37
40
42
43
51
1 Hour
5.3
30.9
17.7
7.6
11.6
6.8
31.7
13.5
18.9
12.5
8.7
2.7
7.4
11.0
17.8
3.4
6 Hours
5.1
23.7
15.2
7.2
9.1
5.3
23.5
13.0
15.3
11.7
7.6
2.6
6.5
10.2
10.6
3.1
12 Hours
4.8
18.1
13.4
6.7
7.7
4.8
19.5
12.5
13.3
10.9
6.9
2.5
5.9
9.6
8.9
3.0
24 Hours
4.6
15.0
11.6
6.3
6.8
4.3
17.1
11.5
11.2
10.0
6.2
2.2
5.6
7.8
7.8
2.6
-69-
-------
1.0
0.8
0.6
r 0.2
-0.0
-0.2 -
-0.4 -
-0.6
-0.8-
-1.1
EPR1 SURE 24 HOUR S02
I
I
400.
600. 1200.
DISTRNCE (KM)
1600.
2000.
2400,
FIGURE 4-7. Scatter plot of 24-hour average S00 correlations as a function of
the distance between stations. *'
-70-
-------
The correlation coefficients were estimated for all possible station pairs
and are presented as a function of distance separating the pairs
(Figure 4-8). Despite the relatively large scatter, the average correlation
coefficients seem to decrease systematically as a function of station
separation. The average correlation coefficient was computed for ranges of
station separation. The resulting curve with its confidence interval is shown
in Figure 4-9. The curve suggests that a model of the form:
p(d) = exp(-d/d0), do = 660 km (4-3)
is a good fit of the curve in Figure 4-9. This result suggests that the
spatial persistence of the S04 concentration patterns is much greater and
extends over much larger scales than for SOz concentrations.
One parameter of interest to network design is the station separation that
is necessary to keep the MSE of interpolation below some arbitrary level. The
MSE of extending an observation at location X over a distance AX can be
estimated directly from the pair station statistics. Figure 4-10 shows the
rate of increase in the MSE with increasing AS. The MSE was fitted to a
model of the form:
MSE(Ax) = 0.45 AX°'4S (AX is in km) (4-4)
-71-
-------
l.B
0.8
0.6
0.4
2 0.2
1-0.0
-0.2
-0.6
-0.6
-1.1
I T
EPR1 SURE 24 HOUR S04
I
I
400.
800.
1200.
DISTRNCE (KM)
1600.
2000.
2400,,
FIGURE 4-8. Observed spatial correlations of 24-hour SO. versus distance.
-72-
81*210
-------
1.0
100 200 300 400 500 600 700 800 900 1000 1100 1200
Distance Between Station Pairs (km)
FIGURE 4-9. The variation in the correlation coefficient of 24-hour
average $04 as a function of station pair separation. The error bars
are for the 95% confidence interval.
8»t210
-73-
-------
400 600 800 1000 1200
Station Pair Separation Distance (km)
1400
FIGURE 4-10. The increase in 24-hour average SO. MSE as a function of
separation between station pairs.
-74-
-------
The MSE can be used to define a signal to noise ratio (s/n) in terms of the
reduction of variance, e.g.,
S/N(d) = 1.0 - MSE(d)/2 Var(c) (4-5)
where c is the S04 concentration. At a distance d= 660 km, the S/N(d) is of
the order 0.375.
According to the variogram model for the MSE that is discussed by
Huijbregts (1975), the MSE can be related to the spatial autocorrelation
function p(d) by
MSE(d) = 2 Var(c)[l - p(d)] + e (4-6)
if the S04 statistics over the EPRI/SURE network are stationary, then
e = 0. Visual inspection of average SCU concentrations such as those
displayed in Figure 4-11 indicates that the concentration averages vary with
distance. One hypothesis advanced was that the term c=(cx - cx+ax)2
is related to spatially and temporally averaged concentrations and rapidly
approaches a constant value with increasing distance. This hypothesis was
examined by computing the MSE(d) from Equation 4-6 and comparing it with the
observed MSE(d). The results are summarized in Table 4-5 and show that the
(GX - cx + Jix)2 term is relatively both constant and small. The term
can be fitted to a model of the form
(cx - cx+AX)2 = K[l - exp (-d/do)] (4-7)
where do is less than 50 km and K is a constant of order unity.
-75-
-------
r ' '-'~\
V.. '
6 :'
8
,9 /: ___ , _____ - -*'
*"' - r
' "' "
--- ^ ----- B
X 8
'B' ^-~.9 -s~ __
X ' ' '« 6 ''
t
I : ' \. * >""
2
FIGURE 4-11. Geographic distribution of the mean sulfate concentrations
for each of the 54 EPRI SURE monitoring sites.
-76-
-------
TABLE 4-5
COMPARISON OF THE OBSERVED MEAN SQUARE ERROR (MSB) IN 24-HOUR AVERAGE S04
CONCENTRATIONS AS A FUNCTION OF STATION SEPARATION VERSUS THE MSE COMPUTED
USING A SIMPLE VARIOGRAM MODEL DESCRIBED BY EQUATION 4-6. VARIANCE FOR ALL
STATIONS IS 36 }ig/m3.
Observed MSE
Predicted MSE
Distance (km)
Observed Minus
Predicted (E)
140
306
496
695
903
1090
1551
19.8
31.6
43.1
46.5
52.9
54.8
60.3
19.1
29.7
41.5
45.0
51.6
53.0
59.8
0.7
1.9
1.6
1.5
1.3
1.8
0.5
-77-
-------
4.3 Temporal Characteristics of Concentration Data
The temporal characteristics of the concentration data are important
particularly in evaluating the feasibility of the source modulation component
of the planned short-range experiments. The planned experimental program
calls for the modulation of several SC>2 sources in a region. The modulated
signal will be modified by the atmosphere and hopefully part of the signal
will be detected at target receptor locations. This signal will have to
compete with the SOa fluctuations caused by other sources. The signal takes
a significant time to reach the receptors and is transported in a
directionally biased fashion. Furthermore, the loss of signal can be highly
nonlinear with distance due to processes such as rainout.
These considerations make it important to perform a frequency analysis of
the relationships between daily average S02 and sulfate fluctuations. In
the analyses, the 24-hour average ground-level SOa concentrations in the
major source region is used as a surrogate for the S02 emitted into the air
aloft. The reasoning is that if the ambient ground-level S02 concentrations
are dominated by only the major point sources operating within about 100 km,
then there is no effective time lag between emission rates and 24-hour average
concentrations.
For the analysis, it would be good to know if there is a window of
frequencies for which the source signal would propagate to the desired
receptors with a minimum loss. This question can be estimated through the use
of the frequency gain function, G(u). The gain function is a measure of the
fraction of a unit amplitude sine wave of frequency w that reaches the
receptor site. The function is generally bounded between 0 and 1 and is often
expressed as a logarithm.
-78-
-------
The gain function, G(u), is defined as the cross power spectrum function
of the transmitted signal (the 24-hour averaged SOZ concentrations) derived
by the power spectrum of the transmitted signal (the 24-hour average SCU
concentrations), e.g..
(4-8)
(S04 - S02) S02
The gain function is useful when the emission from point sources in other
regions are either white noise, or else their spectrum resembles that of the
24-hour average S02 concentrations from the areas of interest. The S02
spectrums are not constant with frequency and do not resemble that of a white
noise signal. However, an analysis of the S02 sources from other regions
are compared in Figure 4-12, with the finding that in general, the SOz power
spectra are qualitatively similar.
Only the S02 and SCU observations at the nine class I sites could be
analyzed since only these stations have enough data (>80 percent possible) to
avoid significant bias due to missing data. All of the spectral analysis was
done on time series where the missing data was replaced with an average over
the whole time series. The gain function peaks at several frequencies as
Table 4-6 shows. The most consistently strong peak occurs for periods between
4-5 days at the station pairs analyzed. A second peak occurs for a period
ranging between 3-4 days. The results suggest that a source modulation
frequency of 3 to 5 days would be appropriate.
-79-
-------
0.04
(25.0)
0.08
(12.5)
Frequency/Period ( )
days-1 days
Key:
Station 4
Station 5
1.6
FIGURE 4-12. Inter-comparisons of power spectrums of 24-hr averaged S02 at the
EPRI/SURE stations.
84210
-80-
-------
TABLE 4-6
A SUMMARY OF THE MAXIMA IN THE S02-S04 GAIN FUNCTION, G(co),
FUNCTION OF PERIOD AND STATION S02/S04 PAIRING (BAND WIDTH IS 0.0675).
AS
S02 S04 Period
Station Station (Days)
5 2 4.8
9.0
22.0
9 1 5.3
3.3
26.0
5 1 3.5
26.0
4.8
7.6
4 1 4.8
7.6
11.8
3.5
- log G(co)
0.77
1.15
1.34
0.35
1.27
1.52
1.06
1.20
1.21
1.38
1.35
1.36
1.46
1.75
-81-
-------
4.4 Spatial Representativeness of Single Station Concentration Measurements
The question of how representative the measured concentrations at
monitoring sites are of the surrounding areas can be approached by looking at
the expected errors of extrapolating the values at the site to neighboring
locations. The station spacing required to provide adequate spatial coverage
of a region is determined by examining the mean square error of predicting
values at neighboring locations by the value at the single station, as a
function of distance from the station. This is expressed by the mean square
error:
n
MSE
xy
- £
n ^-<
xi
(4-9)
var (Cy) + (Cx - Cy)2 - 2 cov(Cx,Cy),
where
CXi, Cyi, i=l, . .., n
Cx, Cy
var{Cx), var(Cy),
cov(Cx,Cy)
= the concentrations at locations x and y, at n
times,
= the mean concentrations at x and y,
= the concentration variances at x and y,
= the covariance of the concentrations at x and y.
This can be simplified by assuming that the concentration variance does not
vary greatly within the region of interest as in equation 4-6. In this case:
MSExy ~ 2 var(C) (l-rxy) + (Cx - Cy)
(4-10)
where
var(C)
xy
a typical value of the concentration variance within the region,
and
the correlation of the concentrations at x and y.
-82-
-------
rxy and (Cx - Cy)2 are modeled as functions of the distance between x
and y, and then Equation 4-10 is used to express the mean square error as a
function of distance. This approach assumes that the concentration field is
approximately second-order stationary after the mean concentrations, or
spatial trends, have been removed. The first term of Equation 4-10 is the
contribution of the concentration fluctuations to the uncertainty associated
with using the valued measured at x as a surrogate for the value at y; the
second term is a measure of the uncertainty that results from spatial trends
in the concentration field.
It should be noted that Equation 4-10 will overestimate the mean square
error of interpolation from a network. Typically, more than one station is
used for interpolation, and this can substantially reduce the error term due
to spatial trends. However, as the station spacing is increased, the error of
interpolation from multiple stations will approach the value of Equation 4-10.
This approach has been used for the EPRI/SURE hourly SC>2 and 24-hour
SO* data collected at 54 stations in the eastern United States. This data
base is described in Section 4-1. Using Equation 4-10, var(C) is calculated
as the mean of the concentration variances at the 54 stations, and rxy and
(Cx-Cy) are given by:
rxy = exp(-d/d0) (4-11)
and
-------
TABLE 4-7
PARAMETERS FITTED TO HOURLY SOz AND 24-HOUR S04 DATA.
Parameter S02 S04
var(C) 216 (ppb)2 35.5 (iag/m3)2
do 137 (km) 680 (km)
di 0 (km) 985 (km)
a 56 (ppb)2 9.7
-------
less than the SURE network spacing. As a result, fitting of the models to the
data was unsuccessful. The value of di from the data was approximately zero
and the correlation coefficients rapidly approached zero as the station
separation distance increased. This was consistent with distribution of the
observed correlations, mean differences squared, and root mean square errors
versus distance.
Table 4-8 compares the S04 predictions of Equation 4-10 with the average
values within each of seven distance classes of the observed mean square
errors. The predicted values are calculated at a representative distance for
each class. These distances are the midpoints of each class, except for the
first and last classes, in which cases they are the mean distances within the
classes. The second term of Equation 4-10 is also tabulated, so that the
relative contributions of the two terms can be seen.
Figures 4-8 and 4-13 display the S04 sample correlations rxy and the
sample values of (Cx - Cy)2, plotted against distance. Figure 4-14
shows the sample mean square errors (MSExy), plotted against distance. It
can be seen that there is considerable scatter in all of these plots, and that
Equation 4-10 is only estimating expected or mean values. This analysis shows
that the spatial representativeness of 24-hour S04 at a station is of the
order of 100 to 200 km. The spacing of the SURE network stations is not dense
enough to adequately estimate the scale of representativeness of hourly SOa,
except to say that it is less than 100 km.
4.5 Analysis of Network Uncertainties
This section describes a method for quantifying the uncertainties
associated with monitoring networks, where the objective of the monitoring is
to estimate spatial and/or temporal averages or to obtain point estimates of
-85-
-------
TABLE 4-8
COMPARISON OF OBSERVED 24-HOUR S04 WITH PREDICTIONS OF EQUATION 4-10
Distance Average Representative Predicted Predicted
Class (km) Observed MSE Distance (km) MSE (Cx - Cy)2
<200 19.4 140 14.5 1.3
200-400 31.4 300 27.9 2.6
400-600 43.6 500 40.8 3.9
600-800 46.2 700 50.6 4.9
800-1000 53.3 900 57.9 5.8
1000-1200 54.8 1100 63.4 6.5
>1200 60.8 1550 71.4 7.7
-86-
-------
100.
90.
tn 80.
LU
O
CE
LU
LU
31
a:
o
cc.
60,
50.
40.
ce
CE
i 30.
tn
20.
10.
EPRI SURE 24 HOUR S04
I"
800.
1200.
DISTflNCE (KM)
1600.
2000.
2400.
FIGURE 4-13. Observed 24-hour SO. (C - C )^ versus distance.
*t x y
-87-
8^210
-------
12.
11.
10.
9.
8.
7.
6.
5.
3.
2.
1 1 T
EPR1 SURE 24 HOUR S04
.vt *-_
.|r r
_L
'B.
400.
800.
1200.
DISTRNCE (KM)
1600.
2000.
FIGURE 4-14. Observed 24-hour SO. root mean square errors versus distance.
-88-
84210
-------
the concentration by interpolation. First, the uncertainties inherent in
making spatial averages from point estimates are discussed. A second analysis
shows these uncertainties can be reduced by averaging over time.
Consider the problem of estimating the average concentration over a
region, at a given time, from a discrete set of measured values. The
uncertainty of this estimation can be addressed by looking at the expected
mean square error of the estimate:
estimated mean square error = E(A - A)2
where:
E = the expectation operator (over time)
A = the true spatial average, and
A = the estimate of A.
" |a| J
A = I C(x) dx, (4-13)
Q
and
n
4. = I Yt C(xi), (4-14)
where:
Q = the region of interest,
|n| = the area of the region,
C(x) = the concentration at location x,
n = the number of monitoring sites used to form the average,
xi, ...,xn = are the locations of the sites, and
Yi, . . . ,Yn = weights.
Measurement errors are neglected. For the special case where Q is a single
location x0, then A = C(x0), and the problem is one of interpolation.
The weights Yj can be chosen to minimize the expected errors, by
various methods, including the multivariate linear regression (Gandin, 1965;
-89-
-------
Eddy, 1967) and Kriging (Huijbregts, 1975; Delhomme, 1978). If the spacing of
the stations is fairly uniform in Q and the covariance structure of the
concentration field is fairly homogeneous, then taking the weights to be equal
will be near optimal for estimating a spatial average. This will be the case
for the combined experiments for which it is assumed that A is formed using
equal weights Yi = 1/n.
This gives:
/
E(A -'A)2 = E
C(x) dx J C(Xi)
n i=l
1 n n
= II V(Xi,Xj> - _2_
n
l f
i=l J
V(x,xx) dx
V(x,x') dx dx'
Q
1 1 n
C(x) dx I C(Xi)
n i=l
Q
(4-15)
where V(x,x') = E [C(x) C(x')j is the covariance of the concentrations at x
and x', and C(x) = E [C(x)J is the mean concentration at x.
This equation is analogous to Equation 4-10. The first three terms give
the contribution to the error variance from the variance-covariance structure
of the concentration field. The last term results from the structure of the
mean concentration field.
In order to estimate the terms in Equation 4-15, the approach of the
previous section is followed by modeling the spatial correlation as a
-90-
-------
distance-dependent exponential and using a mean or typical value for the
station variances to get:
V(x,x') = var(c) exp (-d/do) (4-16)
where:
d = the distance between x and x',
var(c) = the mean variance, and
do = a parameter to be determined.
In order to estimate the last term, it is assumed that the n stations are
fairly uniformly distributed within the region of interest, in which case it
is reasonable to suppose that 1/n £ Ctx*) is an unbiased estimator of the
r - i=1
spatial average 1/|Q| J C(x) dx. The last term in Eguation 4-15 is
n
estimated by s2/n, where s/Vn~is the standard error of:
n _
1/n I C(x»)
as an estimate of the mean,
1
s2 =
n-1
n n
I C(xi) - I C(xj)
(4-17)
Var(c), do, and s are estimated using hourly S02 and 24-hour SC>4 data;
the values are given in Table 4-7.
The mean square error of the estimate A will be reduced by averaging the
observations over time. In this case, the spatial-temporal averages are
estimated by:
1 T 1
AT = f f C(x, t) dx dt
T J |Q| J
0 Q (4-18)
-91-
-------
1 m 1 n
1T = _ £ £ CiJ (4-19)
m j=l n i=l
where GIJ is the observed concentration at site i at time j. It is assumed
that the times of measurement are equally spaced throughout the interval
[0,T], although not necessarily continuous. The error variance is reduced
according to the following relationship (Papoulis, 1965; Thiebaux and Zwiers,
1984):
1 m-1
E(AT - 1-r)2 = E{A -I)2 I (l-lTfm)p(T A)
m T=-(m-l) (4-20)
where:
m = the number of measurement times,
p(t) = the autocorrelation of A at lag t, and
A = the time interval between successive measurements.
The amplitude of a signal of cyclic nature is sought, then A can be
taken to be equal to the period of the signal, and averages can be formed for
different portions of the cycle; m would then be equal to the number of cycles
available for averaging. p(t) is modeled by:
p(T) = exp (- |T|/TO) (4-21)
where TO is a parameter, determined by fitting this function to the
autocorrelations of the observed area-wide mean values (Table 4-7).
So far, any errors due to individual measurement inaccuracies have been
neglected. It is reasonable to assume that these errors are independent, and
therefore, to take this into account, the measurement error variance can be
added to Equation 4-15. This error variance term is given by:
1
a2
n»m (4-22)
-92-
-------
where n*m is the number of measurements, and a2 is the error variance of a
single measurement.
Combining Equations 4-15 and 4-20 with the models given by Equations 4-16,
4-17, and 4-18 gives the expression for the expected mean square error of
estimating At by At:
E(AT - AT)2 =
var(c)
n
2 n
1 f
i=1 J
Q
exp
do
dx
exp
dx dx'
n n
1 n
+ I
n (n-1) i=l
n
IT I
m
exp (- |T|/TO) +
nm
(4-23)
This procedure was demonstrated by applying these results to the EPRI SURE
1-hour SOz and 24-hour S04 data. The expected mean square error is
modeled by Equation 4-22, using the fitted parameter values given in Table
4-7. The values of measurement error variance used are 0.18 c for hourly
SOa, 0.084 c for 24-hour S04, with c taken to be the average concentration
over the region, 8.74 ppb for SOa and 7.98 ug/m3 for S04. Results
were calculated for a square region 240 x 240 km with monitoring sites located
in a rectangular grid. This area approximates the fine grid area of the
combined experiments. These results are tabulated in Tables 4-9 and 4-10 for
varying station configur tions and averaging times.
-------
TABLE 4-9
THE EXPECTED ROOT MEAN SQUARE ERROR (RMSE) OF ESTIMATING A SPATIAL MEAN OVER A
57,600 km2 REGION, VARYING THE STATION SPACING
Number of
Monitoring Spacing of the Estimated RMSE
Stations Within Monitoring 1-Hour SOz 24-hour SOz
the Region Stations (km) (ppb) (ug/m3)
4 120 4.4 1.1
9 80 2.6 0.70
16 60 1.9 0.51
25 48 1.4 0.40
64 20 0.8 0.23
-94-
-------
TABLE 4-10
THE EXPECTED ROOT MEAN SQUARE ERROR (RMSE) OF ESTIMATING A SPATIAL MEAN OVER A
2
57,600 (240 x 240 km) km REGION WITH A STATION SPACING OF 60 km, FOR
INCREASED AVERAGING TIMES
Species Averaging Time RMSE
SO 2
SO 2
SO 2
S04
S04
1 hour
24 hours
1 week
24 hours
1 week
1.9 ppb
1.3
0.65
0.51 (ug/m3)
0.41
-95-
-------
4.6 Implications for Experimental Design
An analysis of portions of the EPRI/SURE data base, consisting of hourly
SOz and 24-hour sulfate concentration measurements throughout the eastern
third of the United States, reveal that the spatial coherence of sulfate is
much greater than S02. Additionally, the monitoring station spacing
required to confine interpolation uncertainties to less than half of the
variance in the observations is less than 200 km for S02 and greater than
approximately 600 km for sulfate. SOz data available indicate the existence
of spatial concentration patterns of a scale intermediate to the Kincaid and
SURE experiments. This limits the ability to specify a characteristic scale
of measurement to guide selection of sample resolution.
As would be expected, increasing the averaging time of SOz concentration
results in reduced variance in the measurements. However, even for 24-hour
averaged concentrations, the spatial structure is characterized by rapidly
decaying spatial correlation fields. Although the EPRI/SURE network was
designed to detect "rural" SOz concentration patterns, the data analysis
leads to a recommendation of greater spatial resolution for SOz monitoring
than that characteristic of the SURE program.
The spatial "structure" of 24-hour sulfate concentration patterns appears
fairly well determined with a spatial resolution of approximately 500 km.
The 24-hour average S02 and S04 concentration time series for two
stations were compared using spectral methods. The S02 concentration time
series were used as input signals to the acidic deposition generation
processes, while the S04 concentration time series at a remote site
sufficiently downwind of the sources were considered the output signal. The
amplitude response in the SO* concentrations due to variations in SOz
concentrations, as expressed by a gain function, was found to produce maximum
responses in S04 for a three- to five-day variation in the input S02
-96-
-------
concentration time series. However, the implications of this result are not
straightforward with regard to an appropriate modulation frequency. The
three- to five-day cycle is likely dominated by synoptic weather fluctuations.
A more rigorous analysis of the input and output time series is warranted.
A methodology has been presented for determining the typical spatial
representativeness of monitoring stations and estimating the average mean
square errors of calculating spatially averaged concentrations from data
collected at a monitoring network. This is useful for two purposes. First,
it allows one to determine the spacing of the monitors so that the objective
of the study can be met in a cost-effective manner. Second, once the network
is operational, this method will provide improved estimates of the
uncertainties associated with estimates of spatial means. The methodology was
used to investigate the monitoring requirements of hourly SOz and 24-hour
SO*, using data collected at 54 EPRI/SURE stations to estimate the
parameters involved.
The application involves calculating the average RMS error associated with
forming a spatial mean over a 240 x 240 km region (comparable in size to the
Adirondacks area) by varying the density of monitoring stations within the
region. The results of this analysis indicate that spatially representative
RMS errors are greater for S02 than sulfate. The magnitude of the S02
errors range from less than 1 ppb with a 20 km resolution to approximately
4.4 ppb with a 120 km resolution. Sulfate RMS errors range from approximately
0.2 ug/m3 to about 1.1 pig/m3 with a similar decrease in spatial
resolution, i.e., 20 km to 120 km.
These errors may be too large for the successful detection of the effects
of source modulation. The number of time periods averaged together may be
increased to reduce the errors, both by contiguous sampling periods and, if a
-97-
-------
cyclic signal amplitude is anticipated, by forming noncontiguous composite
samples of time periods matched by the phase of the modulation. This has
implications for the duration of a modulation experiment since a longer
experiment will allow for more data values to be averaged. Once the proposed
monitoring network is operational, it will be important to refine the
estimates of the correlational structure of the monitored pollutants using the
new data in order to improve the above estimates of the monitoring uncertainty.
Considering planned elements of the combined experiments, the following
can be stated:
1. Sulfate monitoring can be successfully performed with the sample
resolution suggested in the COMPEX plan.
2. SOz sampling may or may not be successful in describing spatial
variability. A data base for analysis of sample resolution is
apparently not available.
3. Inert tracers, like SOz are primary pollutants and analyses
presented did not provide information on required resolution.
4. Analyses of the spatial representativeness of spatial averages
suggests that errors may be too high to successfully detect
effects of local source modulation.
-98-
-------
SECTION 5
MODEL SIMULATION ANALYSIS
The uncertainties associated with a multi-faceted large scale field
experiment, such as the proposed COMPEX, may be condensed into a series of
statements expressing the accuracy with which various signals can be isolated
from background noise. In this context, the term "signal" is loosely defined,
and may represent a concentration or deposition value at a specific receptor,
or a sum, product, or other combination of several "signals," such as a
deposition rate, or area averaged deposition.
Regardless of the type of signal under scrutiny, the underlying assumption
of the proposed field experiment is that the particular signal is associated
with a specific source, and this signal is embedded within fluctuating
background noise. The noise is associated physically with pollutants or
tracers from multiple sources. An additional component of noise arises due to
measurement uncertainties. The monitoring network design (i.e., spatial and
temporal sampling intervals, averaging times, etc.) introduces noise as well,
but this noise is related to stochastic natural variability.
An air quality or acid deposition model provides a very useful tool for
addressing certain aspects of the uncertainty issue, primarily the
detectability of signals against background variability arising from multiple
sources. The acceptability of model results depends on the number of degrees
of freedom (i.e., complexity of modeled processes and spatial and temporal
resolution) incorporated into the model and the validity of the model
formulation. However, it is important to emphasize that despite the
complexity and sophistication of a model, its skill is ultimately limited by
the irresolvable, and inherently stochastic processes of nature.
The objective of the modeling analysis is to examine some of the key
uncertainty issues associated with various components of the proposed COMPEX
-99-
-------
study. These issues are related to the detectability of a particular signal
embedded in background noise arising due to multiple signal sources. The
detectability is influenced by the signal's magnitude in relation to the noise
and the precision with which the signal can be measured.
Pour major experimental components of the COMPEX are examinedtwo
pertaining to regional tracer experiments and two pertaining to short range
tracer experiments. The regional-scale experimental uncertainty issues
include examining the required inert tracer (perfluorocarbon) emissions rates
necessary for tracer detection over regional scales and the degree to which
single source emissions characterize diffuse emissions from an entire source
region. The short-range experimental uncertainty issues include examining the
nature of the fluctuating signal and noise resulting from a local and
mesoscale source modulation experiment, and examining the detectability of
sulfur-34 deposition by measuring the difference of inert and reactive tracer
fluxes across an array of receptors.
5.1 Mode1ing Approach
In order to investigate the uncertainties associated with field
experiments proposed under the comprehensive field plan, a modified regional
transport model is used to simulate pollutant concentration and deposition
patterns resulting from hypothetical inert and reactive tracer releases. The
model formulation is described by Durran et al. (1979) and Liu et al. (1982),
and an analysis of model performance is described by Stewart et al. (1983 a,
b). The model simulations are designed to investigate the detectability of
perfluorocarbon tracers over regional scales, and the detectability of SF6
and isotopic sulfur (34S) emissions over local scales and mesoscales.
Within this section are presented discussions on the model configuration, the
-100-
-------
hypothetical tracer emission distribution, and the simulation plans for the
long-range and local/mesoscale detectability analysis.
Modeling Region Configuration
The modeling domain chosen for these analyses is subdivided into three
regions (Figure 5-1). An outer region encompasses much of the northeast
defined on the basis of a preliminary version of the comprehensive field study
plan. Surrounding the Adirondack Mountains of New York State is an additional
fine-resolution grid.
The point sources displayed in Figure 5-1 represent 21 of the largest
SOz point sources within the modeling region and a group of three point
sources within New York State, which represent candidate sources for
local/mesoscale tracer experiments. The point sources external to New York
are grouped into three categories. Seven point sources (denoted hereafter as
the "Ohio" emissions) are clustered within approximately 700 km of the
Adirondack receptor region, a cluster of nine point sources (denoted as
"Kentucky" emissions) are situated roughly 1300 km away from the receptor
region, and the remaining five sources are situated off the Ohio River-upstate
New York axis. The shaded symbols for point sources within each of the two
source clusters represent the largest SOz emitters within their respective
clusters. The cluster names "Ohio" and "Kentucky" are derived from the
locations of these major point sources.
The S02 emissions associated with the 21 large point sources identified
in Figure 5-1 represent 7.1 million tons per year, or 35 percent of the
emissions within the full modeling region. Within the "Ohio" point source
cluster, 2.4 million tons of S02 are emitted per year. This represents
-101-
-------
I +->
Ol C
!-
S- O
to a,
o
U T5
0.1
o +*
01 as
+J i
S. C
o
c -o <-
r- .p- *J
re l- re
E cnr-
O 3
O C £
O °«-
0) O !-
c in p-
o ai °«-
E v. ^2
ai
ai o
-tJ !- «J
«4- QJ
o aj T3
C 4-> L,
O CD
t- « O
4J C re
ro o i=>
CO OJ i-
3 S- O
t i .r- C
1- 0)
*-) W:l
LU 3 Hi
ce r- u
rs o s.
g^ ^^ S
"- V O
Ll_ S. Ml
-102-
-------
about 75 percent of the total S02 emissions within the area encompassing the
point source cluster. The largest point source in the "Ohio" cluster
contributes 22 percent of the cluster's SOz emissions.
The area enclosing the "Kentucky" point source cluster emits roughly 3.6
million tons of SOz per year, of which 74 percent emanates from the cluster
of 9 large point sources. The largest of this cluster emits 14 percent of the
point source cluster's SOz emissions.
Processing of Model Simulation Results
Among the proposed experiments under the comprehensive field study plan
are the release of tracers from major emission regions in the Midwest. At
issue is the guestion of whether a single-point or a multiple-point tracer
release provides the most information on pollutant transport between a
specific emission region and the sensitive receptor. Also requiring
investigation are the questions of what quantity of tracer and what release
characteristics are necessary to ensure adequate detectability over regional
scales.
To address these questions, the regional model is exercised in a
Lagrangian mode over two one-month time spans (January and July, 1978).
Output from the model consists of a time history of plume segment locations
and ages, which can readily be combined with prescribed time varying tracer
emission rates to produce time varying concentration predictions over an array
of receptor locations (specifically, the coarse grid).
A similar plume segment location history pertaining to the three New York
point sources is combined with specified SF6 and 34S emission rates to
yield concentration histories over a series of high-resolution receptor
coordinates situated within the Adirondack Region.
-103-
-------
In addition to the Lagrangian simulations, the model is exercised over the
entire modeling domain with a complete S02 emission inventory in order to
examine the detectability of modulated S02 emissions from the three
candidate New York point sources. Major point sources other than those
illustrated in Figure 5-1 are treated in a manner consistent with routine
application of the regional model. That is, their emissions are incorporated
into the appropriate model layer (i.e., mixed layer or layer aloft) as
determined by a plume rise algorithm. Minor point sources and area source
emissions are incorporated within the mixed layer. S02 emissions from the
24 highlighted point sources undergo chemical transformation and wet and dry
deposition consistent with the remaining SOa emissions treated within the
grid framework. Deposition values from these sources and all others are
accumulated within the appropriate coarse-resolution grid element.
The use of model simulation results in the analysis of uncertainty and
detectability is illustrated in Figure 5-2. Enclosed within the thickline box
are the specific analyses of perfluorocarbon detectability, modulated emission
detectability, and uncertainties associated with mass balance calculations.
Direct input to these analyses from the model simulation consist of
concentration time series of SOz, SCU, SFe, PFT, and 34S over either
the coarse grid or fine grid. These time series are generated from both the
Lagrangian plume segment output and the full Eulerian model output. The time
series pertaining to tracer concentrations are due either to continuous or
modulated emission rates (1 day on, 2 days off). Similarly, the S02 and
S04 time series due to the three New York point sources are modified by a
selected modulation frequency.
-104-
-------
Multi-week Lagrangian
plume segment simulation
for the 24 point sources
illustrated in Fig. 5-1.
Point Source Emissions Specification
Unit inert
tracer emissions
Unit reactive
tracer emissions
1 day on/
2 day off
PFT emissions
Continuous
S02 emissions
(all pt. sources)
L
Modulated
S02 emissions
(3 NY sources)
Continuous
S-34 emissions
(3 NY sources)
Continuous
SF5 emissions
(3 NY sources)
Time history of cone.
per point source or cluster
= coarse grid
= fine grid
Analysis of
detectability of
PFT tracer
(single point
vs. cluster)
Analysis of
detectability of
modulated SC^
emissions
Analysis of uncertainties
in mass balance
experiments
Background PFT
concentration
information
Multi-week model
simulation with all
S0£ emissions except
the 24 point sources
Background
S and SF6
information
Analytical
approach
Time history
of S02, S04
deposited S
over coarse
resolution grid
FIGURE 5-2. Diagram illustrating the use of model information and
ancillary information in the analysis of detectability and uncertainties
-105-
8U210
-------
The ability to synthesize time series from model output relies on a linear
superposition assumption, which is consistent with the transformation and
deposition parameterization utilized within the model. While undoubtedly a
simplification for S02 and S04 concentrations, the superposition
assumption is valid for the inert tracer. The linearity assumption also
permits the analysis of signal detectability directly from the statistics
derived from the time series. Hence, minimal emission rates necessary to
ensure a detectable signal are easily calculated.
Additional details on the procedures used to adapt modeling results to the
analysis are presented in the following sections. Although the modeling
analyses of long-range and short-range experiments have been performed using
both the January and July simulations output, unless otherwise noted the
results are discussed with reference to the July simulation only. During July
1978, the meteorological conditions resulted in a frequency of transport from
the Midwest toward the Adirondack region that was slightly greater than in
January. Hence, the frequency of detection of long-range tracer
concentrations within the Adirondacks may represent slightly better than
average conditions.
5.2 Uncertainties in Long-Range Experiments
Long-range tracer components of the combined experiments are designed to
provide upper-bound estimates of pollutant impacts from major emissions
sources, indicate the frequency of time that the receptor impacts are not
attributable to the source region, and finally, provide the means by which
deposition experiments are integrated into the source-receptor relationships.
This section describes studies of uncertainty issues related to the long range
component.
-106-
-------
One aspect of uncertainty in the long range tracer experiments involves
the determination of suitable tracer emission strengths to ensure a high
frequency of detectability at selected monitoring sites over various transport
conditions. Another issue related to detectability of tracer is that of
tracer release strategies. The goal of the COMPEX experiment is to relate
emissions from specific source regions to deposition over selected receptors.
As pointed out by McNaughton and Bowne (1984), concentration patterns
resulting from individual point source tracer releases may not be
representative of those arising from a cluster of point sources. The larger
spatial distribution of a cluster of tracer releases may require additional
tracer quantities to produce the same frequency of detectability at a given
monitoring station. Because of the increased logistics required from
clustered tracer releases, a trade-off exists between various release
strategies. While a comprehensive analysis of the cost-benefits of various
strategies is beyond the scope of this study, an indication of the difference
in concentration patterns resulting from single- and multiple-point releases
and implications with regard to tracer quantity requirements are determined
from a modeling exercise.
To investigate these aspects, normalized tracer concentration time series
(i.e., X/Q) are extracted on an hourly basis from an 80 km resolution model
grid for the Adirondack receptor. The proposed PFT sampling duration is six
hours. Accordingly, the time series were processed into non-overlapping
six-hour average values. Both hourly and six-hour averages are analyzed to
investigate detectability issues.
For each location in the coarse-grid region, the normalized concentration
time histories are processed into a frequency distribution of X/Q exceeding
a given value as a function of that value, as illustrated in Figure 5-3a.
-107-
-------
IS1
u
c
QJ
3
CT
QJ
(a)
>>
O
c
0)
3
CT
V
t
I
(b)
FIGURE 5-3. Schematic illustration of
the frequency of exceedence of (a) the
normalized concentration, x/Q> as a
function Z, and (b) a detectable concen-
tration, XH> as a function of tracer
emission strength, Q. In the illustra-
tion A < B < C.
-108-
-------
From this distribution, the frequency with which the concentration exceeds a
given detectable limit, Xd is also calculated as a function of the tracer
emission rate (Figure 5-3b). Plots of this type are used to estimate emission
rates as a function of data recovery rate for time varying meteorological
transport and dispersion conditions, and thus are a refinement over the
steady-state dispersion method of estimating minimum emission rates.
Time series of PFT normalized concentrations resulting from a single point
source and a cluster of sources over the entire coarse-resolution grid are
processed in a similar fashion to determine the degree to which the signals
compare. Additionally, the two sets of concentration patterns are compared
via residual and correlation analyses to determine the representativeness of a
single-point approximation to a clustered-release strategy. This analysis is
performed for two clusters of point sources - those approximately 700 km and
those approximately 1300 km away from the Adirondacks.
5.2.1 Estimation of Regional Tracer Release Rates
Figures 5-4 and 5-5 illustrate the frequency with which the relative
tracer concentration (X/Q) exceeds a given value, z, as a function of that
value.* The frequency distributions, denoted F(z), are displayed for the
largest single sources in the clusters and the clusters of sources in "Ohio"
and "Kentucky" both for one-hour average and six-hour average measurement
times. Continuous tracer releases are assumed from all sources. The total
tracer mass released from the single source equals the total mass released
from the cluster. The concentration frequency represents the average of nine
receptor sites in the Adirondack receptor region.
* The figure displays F(z) = 1 - f(z), where f(z) is the cumulative frequency
distribution of z.
-109-
-------
s. s-
a; OJ U *-"
CO S. K1
=3 =
i O
u «/><>
QJ D QJ
U r U
i- W t.
= c 5
o - °
to f) ">
-
333
O O O
m
x
X
ON
I
X
X
O
k O
u
O
to
O
O
OJ
IS)
(J
UJ O)
o: o
r) s_
CJ 3
« o
U, (/)
ooi
POI
ro
(X
-------
Qj O) 0) O)
i 0 r- O
CM J- Ol J-
C 3 C 3
i- O O
to in
O
O
l/l
(/)
TO
c
c
ro
CD
U
c
tn
O)
en
o
3
C C
^ O
= n.
0)
C.
E CU
O O
V CD
M- S-
X ro
o
H- C
o o
s-
C T-
o -o
3 QJ
_f) f
r» »r-
s- c
4->
1/1 r
o ro
s-
UJ O)
C3
oot
-ill-
-------
Comparing the X/Q frequency distributions arising from an "Ohio"
clustered tracer release configuration to that from a single point release
suggests that the more concentrated tracer pattern from the single source
yields higher x/Q values with a frequency of only 2 percent within the
Adirondacks. A comparison of the "Kentucky" tracer impacts (Figure 5-5)
suggests that a larger relative concentration impact from the single source
occurs with a 20 percent frequency. This higher frequency is likely due to
the more dispersed character of the "Kentucky" point source cluster relative
to the "Ohio" cluster.
For both clustered and single source tracer releases from the "Ohio"
region, only the highest 1 percent of hourly averaged X/Q values exceed
those produced from 6-hour averaging. For lower x/Q values, the frequency
of X/Q exceedance is larger when concentrations are averaged over 6-hour
periods. This result is qualitatively what one would expect from an
intermittent but highly peaked time series. The temporal smoothing by
six-hour averaging tends to increase the frequency of low x/Q values by
flattening the sharp concentration gradients at the "edge" of a concentrated
plume. This smoothing also "clips" the peak concentrations, giving a lower
frequency of exceeding the highest X/Q values.
Figure 5-5 illustrates the 1-hour and 6-hour x/Q exceedance frequencies
for the single "Kentucky" point source and a cluster of nine sources.
Although these point sources are nearly twice as far away from the receptor as
the "Ohio" point source and associated source cluster, the frequency
distributions show a similar 40-50 percent plume impact frequency for the
month-long scenario. This is principally due to the well-organized
large-scale transport episodes occurring during July 1978. The highest
1 percent of the x/Q impacts from both the "Kentucky" point source and
source cluster are predicted to be half the magnitude of the impacts from the
single "Ohio" point source and source cluster.
-112-
-------
The frequency of X/Q values resulting from a modulated tracer release is
compared with that resulting from continuous releases. The suggested release
strategy of a one-day release followed by two days of no release is designed
to conserve tracer and to assure detectability of discrete events so that an
experiment of long duration is possible. Additionally, modulated emissions
permit calculation of tracer transport times. Figure 5-6 illustrates the
frequency exceeding values of X/Q for continuous tracer releases versus
modulated (i.e., 1-day on, 2-days off) releases from the "Ohio" and "Kentucky"
clusters of point sources. As expected, the two-day period of no emissions
reduces the frequency with which the Adirondack receptor is exposed to the
entire range of x/Q values. For high X/Q values the frequency of
exceedance is reduced by a factor of 6, whereas for low relative concentration
impacts the reduction factor is about 3.5.
Frequency distributions of exceeding x/Q can be transformed into
diagrams of the frequency of tracer concentration detectability by
determining, for each tracer emission rate (Q), the frequency with which x
exceeds Xd/ the detectable concentration. Because the exceeding x/Q
frequencies of 6-hour average concentrations are greater than the 1-hour
values, results of the tracer detectability analysis are confined to the
6-hour average concentration impacts. Results also pertain to tracer emission
released from the "Ohio" and "Kentucky" point source clusters rather than the
single point sources.
Figures 5-7 and 5-8 illustrate the percentages of six-hour PMCP*
concentrations that are detectable above a background concentration, Xd,
as a function of PMCP emission rate for the "Ohio" and "Kentucky" point source
*Two perfluorocarbon tracers are considered - PMCP and PMCH.
-113-
-------
continuous
modulated
continuous
modulated
0.01
(a)
(b)
0.1
X I10MH-9 S/H««3)
FIGURE 5-6. Distribution of x/Q for continuous and modulated (one day on, two
days off) tracer emissions for (a) the "Ohio" cluster of point sources and (b)
the "Kentucky" cluster of point sources.
-114-
-------
0.1
1.
Emission Rate Ikg/hr)
10.
(a)
o.i
1000.
Etiission Rate Ikg/hr)
JB
S
U
I)
a.
I I I I
o.i
i.
ID.
Enicsion Rate Ikg/hr)
1 T ' "I
100.
o
o
u
O *J
o
u
c.
c
tl
u
k.
1000.
FIGURE 5-7. Detectability of 6-hour PMCP concentrations over the Adirondacks
as a function of (a) continuous and (b) modulated tracer emission rates from
the "Ohio" cluster of point sources.
-115-
-------
0.1
o.
o
f>
JO
o
So-
o »
o J
0.1
u
L,
I
0.1
1.
Emission Rate (fcg/hr)
10.
100.
1000.
5x / 10x/
5 Ox;' lOOx"
I II -I I TTJ I-
10.
Ewission Rate tkg/hrJ
i i , , i , 11 i
100.
o
o
.o
10
u
(a)
1000.
5x
ii i
-rii i i i i
10.
Emssion Rate Ikg/hr)
-TTTTTJ
100.
c
+J
u
L.
II
a.
(b)
1000.
FIGURE 5-8. Detectability of 6-hour PMCP concentrations over the Adirondacks
as a function of (a) continuous and (b) modulated tracer emission rates from
the "Kentucky" cluster of point sources.
-116-
-------
clusters. Part (a) of the figure illustrates to a continuous emission
strategy, whereas part (b) refers to a modulated release strategy. Shown in
the figure are four distribution functions pertaining to detectable PMCP
concentrations (Xa) of 5, 10, 50, and 100 times background concentrations
(2.7 fl/1).
With continuous tracer emission rates, the two figures suggest that there
are two distinct detectability regimes per unit emission increase. For
example, for the "Ohio" cluster emissions, a doubling of the detectable
concentration frequency accompanies a doubling of tracer emissions for
detectability frequencies of less than 10 percent. For detectability
frequencies exceeding 20 percent, a 15-fold increase in emissions is required
to double the frequency of tracer concentration detection. For a tracer
experiment focusing on the more distant "Kentucky" source region, the largest
gain in detectability per unit emissions increase occurs at detection
frequencies below approximately 20 percent. These results therefore imply
that an upper bound on recommended tracer emission rates exists for the
continuous tracer release strategy.
For a modulated tracer release strategy, the gain in frequency of
detection per unit of emission increase is far lower throughout the range of
emission release rates, averaging for the "Ohio" tracer experiment about the
five- to 10-fold increase in emission for a doubling of detection frequency.
Figures 5-9 and 5-10 illustrate the frequency of PMCH concentration
detection as a function of emission rates for the "Ohio" and "Kentucky" point
source clusters, respectively.* These figures likewise illustrate the lower
frequency of detection per unit of tracer emission under a modulated release
strategy, as compared with a continuous release strategy.
*The background concentration of PMCH is assumed equal to 3.6 fl/1.
-117-
-------
0.)
o.
o
o
f
*>
u
u -
I)
£ -
u
L.
U
0.1
c
o
O " -
I)
0.1
1.
' '
Emission Rate Ikg/hr]
10. 100.
1ii i i i 11 1 1i' i i 11
1000._
' I g
5x/ 10x/
50x ; lOOx
I I I 7 I 7-rp 7
10.
Em ssi on Rate Ikq/hr)
i i i , , , , I ,
u
-O *J
c
II
u
(a)
100.
1000.
)
z
5x
SOx/100)/
'I
1.
lllp I
10.
Rate Ikg/hr)
.o
o
(b)
100.
1000.
FIGURE 5-9. Detectability of 6-hour PMCH concentrations over the Adirondacks
as a function of (a) continuous and (b) modulated emission rates from the
"Ohio" cluster of point sources.
-118-
-------
0.1
o
o
0.1
a
a
c
41
U
L
I
a.
0.1
1.
Emission Rate Ikg/hr)
10.
100.
1000.
I I I I I I I
Ii 1
-------
Figures such as these should provide useful guides for selecting
appropriate emission rates for planning a long-range tracer study. Selecting
an appropriate release strategy (continuous versus modulated) requires a
careful analysis of the gain in information resulting from a modulated release
since the required tracer amounts increase rapidly with increasing demands on
the desired frequency of detectable concentration.
It should be emphasized that the tracer detection frequencies discussed in
this section are derived from a one-month model simulation characterized by
several episodes of fairly well organized transport from the Ohio River valley
region toward the Adirondack receptor region. Before a long-term (i.e.,
year-long) tracer experiment is designed, the simulation should be extended to
cover longer periods of time so that more representative detection frequencies
can be deduced.
5.2.2 Tracer Release Configuration
The frequency distribution of x/Q presented in the previous subsection
indicates that the Adirondack receptor region would receive more frequent low
and moderate concentration impacts and less frequent high impacts when the
tracer is released in the clustered configuration than when released from a
single point source. The x/Q frequency distributions also show that when
the tracer is released in a modulated fashion, concentration impacts of all
magnitudes are less frequent than when released in a continuous fashion.
This section examines whether the release configuration affects the
spatial signature of the relative concentration, particularly over the
Adirondack region. If the spatial pattern of inert concentration impacts
differs widely between the cluster and single-source emission configurations.
-120-
-------
a single-source release will not provide sufficient information to
characterize the dispersion of the entire emission region.
Figures 5-11 through 5-13 illustrate the distributions of, respectively,
the monthly mean X/Q» X/Q bias, and x/Q correlation coefficients
pertaining to continuous tracer emissions from the "Ohio" source region. The
X/Q bias is defined as the relative concentration from a clustered release
configuration minus the relative concentration from a single point source
emission. Locations of the cluster of point sources and the individual point
source are indicated.
Figure 5-11 indicates that central Pennsylvania receives the greatest
impact of monthly mean tracer concentration resulting from a clustered
release. Within the Adirondack receptor region, a northwest-to-southwest
factor of 6 gradient of concentration impact occurs. With an identical mass
of tracer released continuously from a single source in southern Ohio, the
relative concentration impact in the southern portion of the Adirondacks is
about 25 percent lower than the impacts arising from a clustered release
configuration (Figure 5-12). In the northern portion of the Adirondacks, the
effects of tracer release configuration are smaller. Correlating the one-hour
relative concentration impacts arising from the clustered release and those
from the single point release indicates that the single point release is a
rather poor surrogate for multiple point source emissions (in Figure 5-12, the
correlation coefficient across the Adirondacks is roughly 0.4).
Considering next the distinction between concentration signals received
from multiple point and single point emissions from a greater upwind distance
(i.e., the "Kentucky" emissions). Figures 5-14 through 5-16 illustrate a
similar lack of agreement. The x/Q bias across the Adirondacks ranges from
approximately +40 percent to -40 percent (Figure 5-15), whereas the
-121-
-------
''x^
O
3
C
c E
O "-
u u;
§£>
V. ^-<
X t/«
Of
C O
1- w>
-t-> 3
3 .
-O U
r
s_ =
4-> O
«fl «f-
i- -C
T3 O
O "
r- 0)
t- E
CD O
o s-
o>
to
r-l 01
r-1 ,
I O)
Lrt S.
LU
01
U
-122-
-------
C C
i- O
o -
Q. in
10
S- -i-
O E
r-j Q;
re
"o
01 .r-
3 £.
C O
O)
(O
re
in O)
3 f
^- OJ
u s_
s_
CO OJ
re u
i- re
-Q S-
+J
c-
v. I/I
X 3
o
»*- 3
O C
r~
C +->
o c
r- O
4J U
3
J2 re
4J O E
in l- ^
u O-H
I- !- 1
.C -M O
O.^- i
re 3
s_ in o>
en a) s-
o t- re
O)
ca ^~> w
oi »-
(/) -r-
re c
C-J O) ~
.-( ---
i o>
m s.
uj oi c
a: o c
=> s~ -^
ts 3 e
> o c
LJ_ CO i
-123-
-------
t.
""r
.,*'
O
J=
O
+> 0)
in
C (B
O) O)
OJ
X QJ
4-J S-
OJ
O)
O
C
O)
<*- 3
<*- O
O) 3
O C
(j
O O
i- O
O) i-
i- o
s_ **-
o
U C
o
*». I/I
X i/i
M- 'i
O 0)
C
tO X3
r-l C
tn
LU 01
rv +J
t? 3
»i r
U. U
-124-
-------
O £
<- 1/1
B3 t I
E 5.
X 3
O
C CO
ro
a; -M
£=
c: o
o
O "1
3
C i
O O
3 >,
JD ^:
r- O
S_ 3
M 4->
to C
r- 0)
CO
s- E
en o
o s_
en
re
i OJ
-
UJ S-
c; a>
=) u
O
-------
\
\
\
-.L-«~
O v
\
\ \
\ " \ '
f'*
n
i
. o
,LD
\ I
o ;'--
o
"-.:--,..-...- -;-.-._
-126-
-------
.0
HI 0)
.c u
U 03
&_
C -(->
0)
3 3
»» o
a- 3
-O C
C C
OJ O
i- O
o
r- ro
y-
4- S_
O) O
o «*-
t- C
O '!-
o
C 0.
o
4J O
3 "-5
XI ro
i- E
i- CD
T3 C
O (/I
-C -O
Q. C
en s-
O OJ
ai -*-1
u a
UJ 3 CO
CK (-» TO
=3 C CU
O a> i
»-» ^
-------
correlation coefficient (Figure 5-16) averages about 0.4. The similar lack of
agreement resulting from the more remote emissions is probably due to the
greater angular spread of the point source cluster characteristic of the
"Kentucky" emissions.
A similar analysis of tracer concentrations signals resulting from
modulated emissions (one day on, two days off) was performed. Figures 5-17
through 5-19 illustrate the distributions of, respectively, monthly X/Q
mean, bias, and correlation coefficients pertaining to the modulated "Ohio"
tracer emissions. With the modulated tracer release strategy, the monthly
mean concentration over the Adirondacks has decreased by roughly a factor of
3. After normalization by the mean X/Q, the bias is similar in magnitude to
the continuous emissions case, whereas the correlation coefficient has fallen
to approximately 0.2, despite the strong on-off nature of the emissions
signal. The emissions from the more remote source region also result in
largely different concentration signals depending on tracer release
configuration.
From these results it appears as if single point tracer release
experiments will not adequately represent the transport and dispersion
associated with area-distributed emissions. While the limited simulation time
may be a factor in influencing the statistics of the signals, the transport
scenarios in July 1978 were favorable for strong tracer signal transmittance
between the "Ohio" and "Kentucky" emission regions and the Adirondacks
receptor region. Therefore, this simulation should represent a rather
stringent test of alternative tracer release configurations. The analysis of
this month-long simulation should be supplemented with additional simulation
time to confirm the results before a final decision is made regarding tracer
release strategies.
-128-
-------
o-
X «
O)
a 3
E O
C O
C C-
w- o
o
s_
C QJ
O 4-1
J2 O
r-
S- ;
4J O
IO -r
- JZ
o p
u
O. 4->
(O
S- E
en o
o s-
O) H-
ra
t~- OJ
i a>
Ln s-
ui s-
cz. a)
Z3
-------
"*>,
: A 9 """> :
i- C
o o
O. !-
I/)
S_ in
O -
01 ^:
«5 -t-l
O)
*oi 6
a
o
in i-
- (J
& 03
x-o
01
o -a
<- o
.a 03
4J O E
10 S_ "^
^ **- tfl
T3
CXV
U C f-<
f- -p- I
^ +j o
CLr .r--l
fO 3
i- in
oo a> =_
I Oi
LT5 S_
uj ai c
er u o
rs s- >-
O 3 en
i i O 0.1
U- in i,.
-130-
-------
-"x-- L.
lv, i.
O)
0)
C in
OJ re
Ol O)
+J Ol
QJ S-
+J Ol
c o
OJ fO
I- 1-
O -i-J
>*- ft)
d; 4->
o o
O i
rj
c T:
C q
o s_
s~ o
L. W-
o
u c
C
O O
c a>
o u
3 O
-C ul
i- 4->
o
"O C.
O i-
- C
-c i
O. 03
-a E
en Q)
O i
QJ CD
c; c
rH C
I Its
UJ OJ
a: 4->
=) in
tD 3
* » »
U_ U
-131-
-------
5.3 Uncertainties in Short-Range Experiments
The COMPEX design requires short-range experiments of two types. The
first type is emission modulation experiments performed to isolate the
concentration and deposition contributions from point sources situated within
the mesoscale distances (less than 200 to 300 km from the Adirondack receptor
region). The second type will investigate the sulfur deposition losses over a
variety of surface and meteorological conditions.
Within the scope of this study it is only possible to examine a few of the
key uncertainty issues associated with the design. The next two subsections
focus on the issues of signal detectability for both source modulation and
reactive tracer (34S) experiments.
5.3.1 Local Source Modulation Experiments
The primary objective of a source modulation experiment is to evaluate the
hypothesis that reductions in ambient sulfur oxide concentrations and
deposition amounts will result from reduced precursor emissions. While this
type of experiment is conceivable over a variety of spatial scales,
preliminary modeling analyses (Morris et al., 1984) suggested that over
regional scales the magnitude of the emissions modulation has to be very large
to cause significant concentration differences. Other studies confirm this
observation. Limited capabilities of transferring power among electrical
systems and numerous socio-economic problems that would result from widespread
emissions modulation require shifting the focus toward local and mesoscale
studies.
As proposed under the COMPEX field study plan, a local/mesoscale source
modulation field study appears to be more feasible. However, the following
issues must be investigated to determine feasibility:
-132-
-------
Are the emissions from west-central New York State of sufficient
magnitude to produce detectable S02 and sulfate concentration and
deposition signals in the Adirondacks region when modulated?
Are modulations of a detectable level possible among the emission
sources, given the constraints imposed by maintaining electrical
services at reasonable costs?
What type of modulation signatures are required for unambiguous data
interpretation?
A preliminary feasibility analysis suggests that the west-central New York
utilities appear well suited for source modulation experiments in terms of
electrical power supply trade-off capabilities. However, a rigorous
assessment of these capabilities is required before the experiment can be
considered feasible.
Described within this section are the results of a model simulation
analysis aimed at investigating the issues of the source modulation
experiment, namely, the detectability of a response to continuous and
modulated emissions within the Adirondack receptor region.
As currently envisioned, the local/mesoscale source modulation experiment
would include the release of an inert tracer, such as SFs, to determine the
plume locations at all times. Therefore, measurements of SOz and sulfate
concentration levels within the region where tracer is detectable provides the
pertinent information for the New York point source impact analysis. The
success of the emission modulation experiment depends on collecting sufficient
concentration data within the tracer impacted region both when emissions are
at full strength and when they are modulated.
For the modeling analysis, the modulation experiment is considered a pilot
experiment of a one-month duration. Of key interest are the frequency with
which S02 and sulfate concentration associated with inert tracer are
detectable, and the nature of the concentration impact during full emission
-133-
-------
and modulated emission time periods. Since the model predicts the
concentration patterns from the New York point sources separately, it is not
necessary to consider inert tracer dispersion. As a conservative assumption,
this analysis tacitly presumes that the tracer measurements are of sufficient
detectability to continuously monitor the location of the plumes from the
three New York point sources.
First, the time series of S02 and 30$ concentrations due to the three
candidate New York sources within the Adirondacks receptor region are
examined. Figure 5-20 shows the S02 and S04 concentration time series
over the southwestern portion of the receptor region for the June 1978 model
simulation. The figure shows the contribution from the three New York point
sources (dark shading), the contribution from the largest 21 point sources
treated with the Gaussian plume segment approach (unshaded), and the
contribution from the area sources and remaining point sources within the
modeling region (lightly shaded). Visual inspection reveals a variable but
generally small S02 and S04 concentration impacts arising from the
continuous S02 emission of the New York point: sources. The absolute S02
and sulfate concentration contributions from these three sources are displayed
in Figure 5-21. The shaded portion of these time series represent those
concentration values that would occur if the emissions were fully modulated
(i.e., on and off) on a weekly basis. From the figures it is estimated that
for this particular receptor point and time period, the New York point source
emissions contribute to the SQ2 and sulfate concentration burden about 55
percent of the time. If the emissions are modulated, the concentration impact
from these sources decrease to about 30 percent of the time.
Of course, the uncertainty associated with detecting these incremental
impacts will result in a lower frequency of detection. Using 18 percent as a
representative value of the precision in measuring hourly ground-level S02 in
-134-
-------
300
16 17 16 15 20 21 22 23 24 25 26 27 26 29 30
(a)
3CC
300-
250[
20C*
i
150
JOG
50
0
10 11 12 13 14 15
;3CC
100
16 17 18 19 20 21 22 23 24 25 26 27 26 29 30 31
(b)
FIGURE 5-20. Time series of predicted hourly
(b) $04 concentrations at the centroid of the
grid cell within the Adirondack region. Light
the contribution from area sources and all poi
modeled with the plume segment approach. The
represents the contribution from the 21 large
ed with the plume segment modeling component.
shading represents the contributions from the
sources assuming continuous emission.
July (a) SOp and
southwest 80 km
shading indicates
nt sources not
unshaded portion
point sources treat-
Finally the dark
3 New York point
-------
ou
50
40
5 30
§
20
10
0
£fl
O U
50
40
*3C
= __
n
u
Hfl
. 0
12.0
10.0
6.0
6 . 0
4. C
2.0
C«
. 0
0
Hn
u
12.0
10.0
B.O
6.0
4.0
2.0
o n
-
:: l\ ^
1 -
/ ^
" .N I fi /A, i\ A y \ / ~
" , . i j iiLj*. tun i , 1 1 , i A . , , , , ,!l J,V, , , ,i\4_iT , , , , u/V, 1 1 1 M 1 1 1 1 1 1 > 1 1 rt , /i i i\v A .j/li i . 1 1 1 \J i r
1 2 3 4 5 6' 7 8 9 10 11 12 13 14 15
-
-i
- -
NA n
I\ n i
it A d 1 1
16 17 16 19 20 21 22 23 24 25 26 27 2B 29 30 3
(a) '
_
-
-
]- t^
1 h -
i\ 1 \ ~
7 ^A J Ww^. ^L^ /J U
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
-
-
-
-
1 .1
' \/LaiMaJVlWfc1^Kiln..nu,,,,,,n,i,H.n,.inMiMA*.,iM.,«,.. ,m..i>iii,..Miinii n-m
DU
5G
30
20
10
f O
DU
50
40
3C
2C
' *
14 0
JG.O
B.C
6.0
4. C
2.0
r; r\
U t U
i A n
i * U
12.0
10. 0
8.0
e.c
4.0
:.o
n. n
16 17 16 19 20 21 22 23 24 25 26 27 26 29 30 31
(b)
FIGURE 5-21. Predicted (a) S02 and (b) $04 concentration time
series due to the three New York point source emissions only
over the sane receptor point as in Figure 3-23. Shaded portions
refer to concentration predicted for the nodulated enission
configuration.
-136-
-------
rural areas (Section 2.4), the fraction of time continuous SOz emissions are
detectable decreases to 18 percent whereas the fraction of time modulated
emissions are detected decreases to 15 percent. Similarly, if 9.7 percent is
taken to be representative precision for measuring 3-hour sulfate
concentrations using a sequential filter sampler, the detection frequencies
decrease to 8 percent for continuous and 7 percent for modulated emissions.
These estimates of detection frequencies pertain only to the particular
receptor point selected for this analysis, which is situated approximately
80 km downwind from the centroid of the three New York point sources. An
indication of the spatial distribution of detectable SQz and sulfate
concentrations resulting from the three New York point sources is provided in
Figures 5-22 and 5-23. Figure 5-22 shows the percent of time (during July
1978) that SOz concentrations from continuous and modulated emissions are
detectable above the uncertainty levels associated with the measurement
technique. The maximum frequency of S02 detection is located at
approximately the same position as that for which the time series discussed
above were computed. Therefore, according to model predictions, the fraction
of time the S02 concentration contribution from the New York point sources
are detectable is roughly one third of the time that these sources contribute
to the S02 concentration burden, assuming continuous emissions.
It is interesting to note (Figure 5-22) that for modulated emissions the
detection frequency decreases by only a few percent. Upon close inspection of
the SOz time series in Figures 5-20a and 5-21a, this fortuitous result
occurs because of the concurrence of the most prominent concentration
contribution from the New York point sources with the "on" phase of the
emission modulation. During the remaining time, SOz concentrations from
other sources are large enough and the New York point source contribution is
small enough to preclude detection of the New York SOz concentration. Since
-137-
-------
(a)
(b)
FIGURE 5-22. Percentage of time during July 1978 that $03
concentrations due to (a) continuous and (b) modulated S02
emissions from the three New York point sources are detectable,
-138-
-------
(a)
(b)
FIGURE 5-23. Percentage of time during July 1978 that SO/]
concentrations due to (a) continuous ar.d (b) modulated S02
emissions from the three New York point sources are detectable,
-------
this period of time corresponds to the "off" phase of the modulation cycle,
the lack of detectability had little effect on the detection frequency.
Figure 5-22 also suggests that the detectable S02 impacts arising from
the New York point sources decreases considerably throughout the northeastern
portion of the Adirondack receptor region during the July simulation. This
tendency can be interpreted in a broad sense in terms of trajectories from the
various high SOz emission regions. Under large-scale southwesterly flow,
the background S02 concentration due to midwestern emissions (particularly
the "Ohio" major point source cluster) are relatively high over the
Adirondacks. Because we have assumed the detectability threshold is a
constant fraction of the total S02 burden, the relatively small SOz impact
from the New York sources becomes "lost in the noise." Clearly, the
specification of instrument uncertainty plays an important role in these
detectability calculations, and these results must be interpreted accordingly.
Figure 5-23 illustrates the spatial distribution of the sulfate detection
frequency from the New York point sources under conditions of continuous and
modulated emissions. The distribution of detection frequencies is broader
than the S02 distribution because sulfate is a product species only* and
hence its distribution is of a more regional nature. The magnitudes of
sulfate detection frequencies are similar to the S02 detection frequencies
away from the region of maximum S02 plume impact. Since sulfate
concentrations are low close to the sources, there is no pronounced spatial
maximum in the frequency of sulfate detection. The sulfate detection
frequency corresponding to modulated emissions is generally 1-2 percent lower
than the continuous emission configuration for reasons similar to those
discussed with respect to S02 concentrations.
*Primary sulfate emissions were not considered in the model simulations.
-140-
-------
The relevance of these model simulations results to the proposed
short-range source modulation experiment can be summarized in the following
manner. Predicted S02 concentration levels due to the three New York point
sources are highly variable both spatially and temporally. For the month of
January 1978, there is a tendency for the maximum SOz detectability
frequency to be located east of the point sources, indicating that the amount
of SOz concentration data useful for the local source attribution analysis
is likely to be weighted by wind direction. During south-westerly flow, for
example, the influence of remote S02 emissions may adversely affect the
acquisition of pertinent data.
From the limited model simulation time period, the frequency with which
weekly modulated emissions are detectable in the ground-level concentration
distribution is only a few percent lower than that corresponding to continuous
emissions. Assuming that the "zero SOz" plume is identifiable using a
continuously emitted inert tracer (as proposed in the preliminary experimental
design), the period of time for which a detectable difference in the local
source impact could be ascertained from the data is roughly one day for the
entire month.
Since this short period of time results from the coincidental high
correlation of the time periods of good detectability and the "on" phase of
the emission modulation, it is evident that the amount of useful data acquired
during a month-long experiment will be extremely sensitive to the timing of
the modulation in relation to the prevailing meteorological transport.
To test this hypothesis, similar model output processing for the January
1978 simulation has been performed. The S02 and sulfate time series over
the Adirondack receptor region (not shown here) indicate that the three New
York point sources contributed less frequently to the concentration burden in
January than during July. The spatial distribution of the SOz concentration
-141-
-------
detectability due to continuous and modulated emissions (analogous to
Figure 5-22) is shown in Figure 5-24.
This figure reveals that the different meteorological transport, chemical
transformation, and deposition conditions of the January scenario result in a
different location of maximum S02 detectability of the three point sources.
Additionally, the decrease in frequency of detection accompanying a modulated
emission configuration is proportional to the decrease in time that SOz is
emitted.
The sensitivity of pertinent data acquisition to meteorological conditions
could be decreased by conducting the modulation experiment for an extended
period of time (i.e., one year or longer). The concentration contribution
from the three New York sources could potentially be determined by performing
a spectral analysis of the measured concentration time series, focusing on the
variance associated with the modulation frequency. Extended model simulations
would also be useful for determining the appropriate modulation frequency
(i.e., where a maximum variance differential between continuous and modulated
emissions exist).
Alternatively, the sensitivity might be decreased by increasing the
modulation frequency within the one-month experimental period. Further tests
would be required to address this issue.
To this point, concern has been focused on the magnitude of the local
source contribution and its temporal variation with and without a hypothetical
emission modulation. Results from the model simulation indicate that the
meteorological variability may strongly interfere with the ability to detect
the modulated emissions contribution. This is examined in a different manner
below, where concentration variability due to the meteorological fluctuations
is more directly compared with the concentration variability due to the
modulated emissions.
-142-
-------
FIGURE 5-24. Percentage of time during January 1978 that
S0£ concentrations due to (a) continuous and (b) modulated
S02 emissions from the three New York point sources are
detectable.
-143-
-------
Within the mesoscale region surrounding three New York point sources of
interest, the 1-hour SOa and 3-hour sulfate ground-level concentration time
series predicted by the model are analyzed in terms of time averaged mean and
root-mean squared (RMS) concentration amplitudes. The four-week January
simulation and four-week July simulation are subdivided into alternating
weekly periods. Mean concentrations and concentration fluctuations are
averaged over all odd weeks (i.e., first and third weeks of January and July)
and all even weeks (i.e., second and fourth weeks of January and July). The
mean and root-mean-squared SOz and sulfate concentrations due to continuous
New York point source emissions averaged over the odd weeks are compared with
those averaged over the even weeks to assess the variability in concentration
mean and fluctuations due to meteorological variability over time periods
consistent with the proposed modulation frequency. Figures 5-25 and 5-26
illustrate, respectively, the mean and RMS SOz concentration across the New
York state region. Part a) of each figure illustrates the average over the
odd-week; part b) the average over even-week time periods; and part c) the
difference (i.e., odd-week average minus even-week average).
The mean SOa concentration for these two time-averaged samples show
similar north-to-south concentration gradients (Figure 5-25) but a
representative difference in these average concentration values near the New
York point sources is several ug/m3, and varies across the region from 0
to 7 ug/m3. The average RMS SOz concentration distributions for these
sample time periods is approximately 20 to 60 ug/m3 and exhibits a similar
north-to-south gradient (Figure 5-26). Typical weekly RMS concentration
differences (Figure 5-26c) range from 0 to 16 ug/m!.
For subsequent comparisons with concentration variability due to modulated
emissions, the information contained in this figure can be generalized as
follows:
-144-
-------
(b)
(C)
FIGURE S-25. Predicted average $C>2 concentration distribution during the (a)
odd-week and (b) even-week periods. Part "c" illustrates the difference in
average SC^ concentrations over these two samples (i.e., odd-week average
minus even-week average). Units are
-145-
-------
(b)
(c)
FIGURE 5-26. Predicted RMS S02 concentration distribution during the
(a) odd-week and (b) even-week periods. Part "c" illustrates the
difference in RMS $62 concentrations over these two samples (i.e., odd-
week average minus even-week average). Units are
-146-
-------
According to the model predictions, a representative variability in
mean and RMS SOz concentrations due to meteorological processes
alone is several percent of the corresponding mean value around the
central New York State region.
Next, the differences in the average and RMS S02 concentrations due to a
weekly modulation of S02 emissions from the three New York point sources are
examined. Two sets of calculations are performed. For comparison with the
odd-week mean and RMS concentrations with continuous New York emissions, a
parallel time series analysis is performed excluding the point source
emissions. A similar comparison of mean and RMS concentration differences
between the even-week sample with and without the point source emissions is
also performed. From the calculations of the even-week and odd-week samples,
the sample showing the maximum contribution toward average and RMS S02
concentrations from the New York point sources is retained for comparison with
Figures 5-25 and 5-26.
Figures 5-27 and 5-28 illustrate the modulated-emission induced change in
mean and RMS concentrations, respectively. Exclusion of the point source
emissions results in a mean S02 concentration difference of, at most,
2.2 ug/m3, according to the model calculations. Throughout the region of
the maximum point source contribution, the natural variability in mean SOz
is of a greater magnitude (compare Figures 5-25 with 5-27). A similar
comparison of the RMS concentration distribution (Figures 5-26 and 5-28)
reveals that the effect on hourly concentration fluctuations of modulating the
point source emissions is small relative to the natural variability of hourly
concentration fluctuations.
The implication of these model prediction analyses is that the weekly
average and variability of the SOz concentration signal resulting from a
-147-
-------
(c)
FIGURE 5-27. Predicted average S02 concentration distribution over the odd-
week period (a) with, and (b) without the contribution from the three New
York point sources. Part "c" illustrates the S0£ concentration deficit
resulting from the modulated emissions. Units are
-148-
-------
r^
V
\ "x\ X>'
(c)
FIGURE 5-28. Predicted RMS S02 concentration distribution over the odd-
week period (a) with, and (b) without the contribution from the three
New York point sources. Part "c" illustrates the RMS S02 concentration
deficit resulting from the modulated emission. Units are
-149-
-------
weekly emission modulation may be no greater magnitude than the week-to-week
differences in SOZ amplitude and variability caused by meteorological
factors.
A similar analysis was also performed on the predicted 3-hour sulfate time
series which indicates that the week-to-week differences in average and RMS
sulfate concentrations due to meteorological factors exceed those due to
emission modulations. The difference in odd-week and even-week average
sulfate concentration levels are of the order of 1 ug/m3. Several
ug/m3 difference in RMS sulfate concentrations occur due to meteorological
variability. Imposing a weekly emission modulation on the three sources
results in mean sulfate differences of a few tenths of a ug/m3 and a
difference of similar magnitude in RMS sulfate concentrations.
Both the S02 and sulfate concentration analyses do not directly imply
that local source contributions cannot be obtained from a source modulation
experiment, but do illustrate that the signal of interest (i.e., the
concentration differences between emission and no emissions) is strongly
imbedded within noise of the same or larger magnitude. Furthermore, the
results of the analysis of the frequency of signal detection suggest that one
or two month-long modulation experiments will only give rise to a small sample
of data from which to deduce the source contributions.
5.3.2 Local Reactive Tracer Experiments
The ultimate goal of the proposed reactive tracer experiments is to
determine the depletion and dry deposition of S02 over distance scales of
from 10 to 50 km. Ideally, these experiments should be performed over a
variety of meteorological and surface conditions so that the transmittance
factor derived from the data can be combined with the results from the
long-range inert tracer experiments to determine source-receptor relationships.
-150-
-------
The proposed design of the reactive experiment is based in part on the
concepts of a plume depletion and a surface depletion approach described by
Horst (1977) and Horst and co-workers (1983). The experimental design
requires elevated releases of SF6 and 34S (as S02) with a concurrent
ground-level release of a fluorocarbon tracer. Ground-level tracer monitoring
is performed over an array of samplers arranged in concentric arcs with high
resolution within the first 10 km of the source and lower resolution at
downwind distances of from 20 to 50 km. Within 10 km of the source the
concentration measurements of the two inert and one reactive tracer are used
to assess the dry deposition of 34S via the surface depletion approach.
Beyond 10 km, depletion of the reactive tracer is determined by
calculating the reactive tracer depletion budget, making use of the downwind
difference in the concentration ratios of reactive and inert tracers released
from the same elevated position. The advantage of this method over a simple
reactive tracer source balance approach is that measurements of the tracers
need only be made below the level at which the ratio of normalized tracer
concentrations differ from unity. Under non-uniform vertical tracer
distributions, this level is likely to increase downwind as reactive tracer is
deposited, since the resultant vertical gradient in the reactive tracer
concentration near the ground will promote a downward flux of tracer from
higher levels. Thus, the further downwind the tracer is sampled, the greater
the importance of airborne sampling.
The uncertainties associated with measuring the y-z distribution of tracer
concentrations at local distances from the point source are likely to be
greater than the uncertainties in measuring the distribution farther downwind,
where the pollutant distribution is spread over larger distances and is more
uniform. This is due to the necessity of airborne sampling within regions of
high concentration variability and intermittency. Characterizing these
-151-
-------
uncertainties would require an extensive analysis of suitable field data, such
as the EPRI PMV&D data bases and might require the use of a concentration
fluctuation model as well. Such an uncertainty analysis is beyond the scope
of the present effort, but is certainly of importance for planning the
short-term field experiments. Within the remainder of this section, some
simplifying assumptions are involved in order to examine what might be
characterized as a "lowest-order" uncertainty issue; namely, to examine the
relationships between the pollutant depletion over various downwind-distance
increments and the uncertainties associated with measuring the mass flux of
34S. The detectability of the deposition calculated by differencing the
ratios of reactive to inert tracer concentration can be examined analytically
if certain assumptions are made. If, for example, it is assumed that all of
the tracer mass is measured at each downwind arc of receptors and, further,
that the concentrations of these tracers are uniformly distributed in the
vertical, then the deposition occurring between various downwind distances can
be calculated by solving a simple system of mass balance equations for SOz
and sulfate.
With the first assumption, the necessity of considering the ratio of
tracers is removed so that only the sulfur species need be considered. The
second assumption is a useful simplification because it permits the
application of deposition and oxidation rate constants directly to the entire
pollutant mass within the mixed layer, allowing the determination of simple
analytic solution.
The mass conservation equation for airborne and deposited sulfur may be
expressed as:
d_ (Mi) = - (kt + kdi) Mi (5-1)
dt
-152-
-------
d_ (M2) =
dt
- kd2M2
(5-2)
d_
dt
= kdlMi
(5-3)
d_ (R2) = kd2M2
dt
(5-4)
where Mi and M2 refer to the airborne sulfur mass as S02 and sulfate
respectively, and Ri and R2 refer to the deposited sulfur mass as,
respectively, S02 and sulfate. The S02 oxidation rate is given by kt,
whereas the dry deposition rates for S02 and sulfate is represented by kdl
and kdz, respectively. Wet deposition has not been considered.
Using initial conditions specified by Mi = 1, M2 = 0, Ri = 0,
R2 = 0, the solutions take the form:
M2(t) =
Mi(t) = exp [-(kt + kdl)t]
kt + kdi - k
d2
|exp(-kd2t) - exp[-(kt + kdj) t] J
Ri(t) = kdl [ 1 - exp [-(kt+ kdi]t)
kt + kd !
R2(t) =
kt kd2
t + kd ! - ka i
d2
-
l-exp -ka2
kt + k
dl
l-exp - kt + kd , t
(5-5)
(5-6)
(5-7)
(5-8)
Under a uniform wind speed, u, the deposition of total sulfur occurring
between downwind distances x0 - d/2 and x0 + d/2 is given by:
-153-
-------
D =
/ x0-d/2\+
\u /
x0+d/2\+ R2 / Xo+d/2
u
(5-9)
Substituting the solution (Equations 5 through 8) into Equation 9 gives an
analytic expression for the deposited sulfur as a function of downwind
distance x0/ and the separation distance d.
D(x0,d) =
2 exp
-(
Xo+d/2
-(
- exp -(kt+kai) / x0-d/2
u
k.
kt+kdi-
x0+d/2\ - exp -kd2
-TT jj [
i£o±d/2\
--
(5-10)
Figure 5-29a illustrates the percentage of initial sulfur mass removed
through dry deposition as a function travel time x0/u, and normalized
separation distance, d/x0. For these calculations, the following oxidation
and deposition rate constants were selected.
kt = 0.01 h"1
kdi = 0.036 h"1
kd2 = 0.0036 h"
The dry deposition rate constants correspond to S02 and sulfate
deposition velocities of 1 cm/s and 0.1 cm/s, respectively, and a mixing depth
of 1 km. The application of these rate constants over the entire pollutant
mass in the mixed layer is justified by the assumption of vertical uniformity
in concentration distribution.
-154-
-------
2.0
1.5
1.0
0.5
c
o
2.0
ID
O.
OJ
o
O)
M
"S 1.5
1.0
0.5
0.0
I I
I I
I I
30
(a)
1.0
2.0 3.0
4.0
5.0
6.0
7.0
8.0
Travel Time (X /u) in Hours
FIGURE 5-29. (a) percentage of initial sulfur mass removed through
SO? and sulfate dry deposition; (b) Difference in percentage of initial
sulfur mass removed through S0£ and sulfate dry deposition considering
the uncertainties in oxidation and dry deposition rate constant speci-
*----j..-__ c^n 4-ov+ fnr furthpr details. _ic;^_
-------
Recent reviews of the dry deposition and SOz oxidation processes (e.g.,
NCAR, 1983; NRC, 1983) indicate that the effective rate constants may vary
over a considerable range of values. To determine the effect of different
rate constants on D(x0,d), several simple sensitivity analyses were
performed. Analytic solutions to Equation 5-10 were obtained with high and
low values of the three rate constants as shown below:
kt = 0.005 h"1
= 0.02 h"1
kdj = 0.018 h"1
0.054 h"1
kd2 = 0.000 h"1
0.018 h"1
The ranges of S02 and sulfate dry deposition rates correspond to
deposition velocity ranges of 0.5-1.5 cm/s and 0.0-0.5 cm/s, respectively;
again, a 1 km mixing height has been assumed.
The maximum sensitivity of D(x0,d) occurs with certain combinations of
oxidation rate and deposition rates. Rapid oxidation in conjunction with low
deposition gives the lowest values of D(x0,d), while slow oxidation and high
deposition yields high values. The difference in the percentage of initial
sulfur deposited as a function of x0 and d, i.e., AD(x0,d),
corresponding to the combinations of rate constants shown in Figure 5-29b.
The fact that AD(x0/ d) and D(x0,d) are of comparable value throughout
the range of x0/u and d/x0 suggests that, under these simplified
assumptions, the uncertainties in oxidation and deposition rates do exert a
strong influence on the optimum values of x0 and d required to detect the
deposition, as long as the rate constants are constant in time.
-156-
-------
In fact, the mesoscale dimensions of the proposed field experiment ensure
that the deposition of S02 dominates that of S04, and hence the optimal
experimental parameters, x0 and d, are the intuitively obvious ones, namely,
the largest separation distances at the farthest downwind position. For the
larger-scale experiments, one might expect that the eventual SOz oxidation
to the more slowly deposited sulfate aerosol would result in an optimal x0
of intermediate dimensions. Several analyses of the sulfur mass deposition
and deposition sensitivity to different rate constants over larger space and
time scales suggest an optimal value of x0 corresponding to travel times of
about 30 hours, well beyond the range of the proposed 50 km experiment.
The simple mass balance approach is useful for establishing approximate
upper limits on the measurement uncertainties (or minimum precision) required
to detect sulfur deposition for various x0 and d. In order to detect the
deposition of 34S from the different of upwind and downwind mass flux
measurements, the true deposition must be larger than a minimum value, which
is a function of the mass flux measurement uncertainties.
In finite difference form, the measured deposition D is related to the
mass flux convergence by:
A
D/At = -(Fd - Fu)/Ax,
where Fu and Fd are the measured upwind and downwind total mass fluxes,
respectively.
* Here, the assumption is made that the measured mass flux lies within the
interval F + aF with approximately a 67 percent probability.
-157-
-------
Representing the uncertainties in mass flux measurements* by ar, and
noting u = Ax/At, yields:
f> = 1/U (Fu - Ofu) - (Fd + CTFd) (5-11)
where Fu and Fd are the expected (or true) values of the mass fluxes, and
OFU and ap-d are the corresponding measurement uncertainties, which, in
the context of the proposed experiment, include individual concentration
measurement uncertainties and uncertainties associated with integrating these
measurements across each y-z plane. In Equation 5-11 the uncertainties have
been combined with the flux measurements in such a manner as to represent the
maximum likely error in the calculated deposition. Assuming no bias in the
flux measurements, the true deposition is D = 1/u (Fu - Fd). Assuming
that the normalized uncertainties associated with all mass flux measurements
are equal, then:
D = D - 1/u (Fu + Fd) Cv,
where Cv is the coefficient of variation of the flux measurements. A
measured deposition exceeding zero therefore requires an upper limit on the
flux measurement uncertainties, given by:
Cupper . i m .. l >
DO
(Fu + Fd) (5-12)
Since the true mass fluxes at the upwind and downwind y-z planes are,
under assumptions stated, given by:
-158-
-------
Fu =- a
FH =-
I Mi / XQ - d/2 \ + M2/ XQ - d/2 \1
L \ u / V u /J
u [MI / XQ + d/2 \ + Hz I XQ - d/2 \1
L \ u / \ u /J
Equations 5-5, 5-6, and 5-10 can be substituted into Equation 12 to yield
the upper limit on mass flux uncertainty (as a function x0 and d) necessary
to detect the 34S tracer. Figure 5-30a illustrates this uncertainty limit
in terms of the downwind travel time, x0/U, and normalized separation
distance between flux measurements, d/x0. The uncertainties in rate
constants used in Equation 5-10 may be incorporated into the calculation of
upper limits on Cv by specifying the resulting deposition uncertainty by*:
<7d = AD(x0,d)
3
A more stringent upper limit on Cv results from this consideration, i.e.,
Cv
-------
(a)
(b)
1.0
4.0
5.0
2.0 3.0
Travel Time (XQ/u) in Hours
6.0
7.0
8.0
81*210
FIGURE 5-30. Minimum mass flux measurement precision (i.e., Cv.upper
limit) required "for sulfur deposition detection (a) without, and
(b) with consideration of the uncertainties in oxidation and dry SO?
and sulfur deposition rates. These values, expressed as a percentage,
are calculated from the simple mass balance approach. See text for
further details. .,.
16Q~
-------
spatial separation distance equivalent to 16 hours of travel time
(i.e., Xo/u = 8 hr, d/x0 = 2). If uncertainties in the oxidation and
deposition rate constants are factored into the analysis, the minimum
precision decreases to below 20 percent (Figure 5-30b).
Over a hypothetical monitoring network configuration consisting of
concentric arcs of S02 and sulfate monitors spaced 10 km apart at distances
ranging from 10 km to 50 km, an estimate of the minimum mass flux measurement
precision for all combinations of flux difference calculations can be obtained
(as a function of wind speed) from Figure 5-30. For example, the minimum
precision of mass fluxes determined from concentration measurements at 10 kin
and 50 km lies along the horizontal line corresponding to d/x0 = 1.33.
Since x0 = 30 km, the travel time (in hours) is given by T = 8.33/u, where u
is in m/s. The minimum mass flux measurement precision falls below 5 percent
at wind speeds exceeding ~2.8 m/s.
As stated previously, the assumptions necessary to calculate
Cv
-------
The proposed use of a conservative tracer (SF6) in conjunction with
34S is designed to circumvent the need for total mass flux determination.
Measurements of the flux of the 34S to SF6 ratios at different downwind
locations allows the calculation of 34S depletion normalized by the mass of
an inert species. The analysis presented above is now applicable with the
understanding that the precision in mass flux measurements pertains to the
flux of the reactive-to-inert tracer concentration ratios.
The assumptions of vertical uniformity in 3
-------
The analysis presented in this section can be modified, albeit with a loss
in simplification, to reflect different stabilities and non-uniform
concentration distribution but the results of Horst (1977) suggests that the
minimum mass flux measurement precision would become more stringent. Thus,
the analysis performed here can be viewed as a liberal estimate of the
uncertainties associated with calculating deposition via mass flux
differencing.
The minimum measurement precision values displayed in Figure 5-30 are
dependent on the constant oxidation and deposition rates selected. In
reality, these rates are spatially and temporally variable. A simplified
attempt to introduce the uncertainties resulting from uncertain rate constant
values has been performed. The reduction in minimum allowable mass flux
measurement precision due to rate constant uncertainty (compare Figures 5-30b
with 5-30a) must be viewed with a full appreciation of the assumptions
involved. Certainty, a more rigorous analysis, using a time-varying model
with spatially and temporally variable transport, transformation, and
deposition, would serve to refine the estimates of minimum required precision.
In an attempt to refine the analysis and to confirm the assertation that
the simple mass balance approach yields liberal Cv estimates,
results of the model prediction of X/Q values for 34S and SF6 at ground
level receptors arranged in concentric arcs of 10 km separation are examined.
Since the proposed short-range reactive tracer experiment is designed as an
intensive, short-term experiment, the model results from a one-hour period
characterized by well-defined transport (1500-1600, 6 July 1978) are discussed.
Figures 5-31 and 5-32 illustrate the x/Q distributions for SFs and
34S, respectively, out to distances of 100 km. The distribution of the
34S to SF6 relative dispersion ratios are shown in Figure 5-33. The
-163-
-------
FIGURE 5-31. Spatial distribution of hourly average SFs x/Q for 1500-1600
EST.July 6, 1978 (Milliken Power Plant, central New York State).
-164-
-------
FIGURE 5-32. Spatial distribution of hourly average sulfur 34 x/Q for
1500-1600 EST.July 6, 1978 (Milliken Power Plant, central New York State)
-165-
-------
-98.2
FIGURE 5-33. Spatial distribution of hourly average ratio of 34S x/Q to
SFe x/Q f°r 1500-1600 EST, July 6, 1978 (Milliken Power Plant, central
New York State).
-166-
-------
decreasing ratios with downwind distances are indicative of the
model-calculated deposition of 34S in the form of S02 and sulfate.
Table 5-1 lists the percentage change in the X/Q ratio (normalized by the
average x/Q ratio) across the distance interval separating the arcs where
hypothetical concentration ratio measurements would be obtained.
These results suggest that less sulfur mass is deposited over these
distance intervals than was calculated by the simplistic mass balance
approach. The likely cause of this is the fact that in the model dry
deposition is calculated from the surface layer portion of the elevated plume,
not the entire plume mass. This is in accord with the surface depletion
approximation.
The results of these deposition detectability estimates suggest that high
precision is required in measuring the ratios of 34S to SFe. Combining
the uncertainties in measuring 34S with those of measuring SFs, the
uncertainty in measuring the ratio, R, is given by:
1/2
/ \2 / V
/ * d. __ \ / - \
OR
R
where a represents the measurement uncertainty and [ ] represents the
measured concentration.
Provided sufficient S02 and sulfate mass is collected at each sampling
location, the uncertainty in measuring 34S concentrations (Section 2.3) is
negligible compared with SFs measurement uncertainties. Hence, the relative
uncertainty in the ratio measurement is approximately egual to that associated
with the SF6 measurement. The relative fraction of airborne 34S deposited
between arcs therefore must exceed the relative uncertainty in measuring
SF6. If we assume, as in the previous analysis, that the coefficient of
variation of SF6 mass flux measurements is egual to that of a single
-167-
-------
TABLE 5-1
34,
DIFFERENCE IN RATIOS OF NORMALIZED J"S TO NORMALIZED SF6 CONCENTRATIONS AS
A FUNCTION OF THE DISTANCE BETWEEN SAMPLING ARCS. DIFFERENCES ARE EXPRESSED
AS THE PERCENTAGE OF THE AVERAGE RATIO ACROSS THE INTERVAL BETWEEN ARCS.
Arc
Distance
(km)
20
30
40
50
60
70
80
90
100
Distance Between Sampling Arcs (km)*
Ratio
0.9868
0.9863
0.9857
0.9852
0.9834
0.9799
0.9865
0.9733
0.9725
10
0.05
0.06
0.05
0.18
0.36
0.35
0.33
0.08
20
__
0.11
0.11
0.23
0.54
0.70
0.68
0.41
30
0.16
0.29
0.59
0.88
1.03
0.76
40
__
0.35
0.65
0.94
1.22
1.11
50
__
.
0.70
1.00
1.27
1.30
60 70 80
__
1.05
1.33 1.38
1.34 1.41 1.46
* The distance between arcs is measured upwind from the arc distance listed in
the left-most column.
-168-
-------
concentration measurement (neglecting other uncertainties in integrating the
concentration across a y-z plane), and further assume that this uncertainty is
independent of downwind distance (i.e., SF6 magnitude), the model-predicted
relative 34S deposition values listed in Table 5-1 represent the minimum
SFe measurement precision necessary to detect 34S deposition. Each of the
assumptions stated above are liberal, indicating that the minimum SFs mass
flux precision is likely to be overestimated.
If the model results are more indicative of the magnitude of 34S
deposited than the mass balance approach, the precision of SFe measurements
in the field (i.e., ~10 percent) does not appear to yield detectable sulfur
deposition for the proposed scale of. the experiment.
From these tentative conclusions based on modeling analysis and simplified
mass balance considerations, the following recommendations emerge:
1. Long-range tracer experiments designed to determine potential
regional source contributions to air quality and deposition over a
receptor should consider spatially distributed tracer releases
rather than a single point release. Obviously, a trade-off exists
between realistic simulation of source region emissions and
logistical concern with coordinated tracer releases. Additional
analyses are required for testing final tracer release
configurations.
2. Local-source modulation experiments will probably yield more
definitive estimates of source attribution if the experiments are
conducted for long periods of time (longer than one month). The
magnitude of the detectable signals, when coupled with limited
measurement precision, is strongly embedded in noise of comparable
or greater magnitude. Hence, statistical analyses of extended
concentration time series might be the most appropriate method for
the modulated signal isolation. Long-term model simulations would
be required to further investigate this issue.
3. Because of the small magnitude of sulfur dry deposition predicted
by the model and by a simplified application of the mass balance
approach, an experiment designed to calculate the deposition by
measuring sulfur mass fluxes over local scales appears to be a
difficult undertaking. Emphasis should be on a thorough
characterization of the reactive tracer mass flux, since its
measurement precision is excellent. Measurements of reactive-to-
inert tracer concentrations will be highly affected by the inert
-169-
-------
tracer measurement uncertainty. A more sophisticated dry
deposition analysis would be useful for verifying the modeling
results presented. A detailed analysis of additional uncertainty
components in mass flux characterization is warranted.
5.4 Implications in Experimental Design
A modified regional transport model was used to examine the detectability
of inert and reactive tracer concentration signals and the signals resulting
from local SOz source modulations. The uncertainties inherent in the
modeling approach have not been extensively analyzed because of the monumental
task of such an undertaking and the limited scope of this uncertainty survey.
In most cases, however, the detectability analyses have relied on differences
of model predictions rather than the absolute model predictions. While this
strategy does not eliminate modeling uncertainties, it allows one to examine
the relative sensitivity of detectability without being overwhelmed by the
uncertainty inherent in absolute model predictions. Nevertheless, the
conclusions should be regarded only as approximate indications of signal
detectability.
For the proposed long-range inert tracer experiment, the regional model
was exercised to provide an indication of the tracer detectability frequency,
as a function of tracer emission rates, and to provide an indication of how
adequately a single point tracer release serves as a surrogate for emissions
from a source region. The main conclusions are as follows:
1. Qualitatively, the continuous release of perfluorocarbon tracers
from either a single point source or a cluster of point sources
over month-long periods yields a spatial/temporal concentration
distribution over the Adirondacks that suggests the existence of
an optimum emission rate. This optimum emission rate is expressed
in terms of gain in detection frequency per unit increase in
emissions rate. For realistic tracer emission rates, this
characteristic is not found when tracers are released in an
intermittent (one day on, two days off) manner. (Quantitative
indications of required tracer emission rates for selected
detection frequencies are presented.) Both PMCP and PMCH
perfluorocarbon species were considered and release locations in
the upper and lower Ohio River Valley regions were examined.
-170-
-------
2. Simulations of inert tracer concentration distributions over the
Adirondack region arising from single point and multiple point
releases show sufficient bias and lack of correlation to suggest
that a single point tracer release configuration is not an
adequate surrogate for a spatially distributed emission region.
The lack of similarity in concentration distribution is enhanced
when tracers are released in an intermittent manner.
Two issues of signal detectability pertaining to the proposed short-range
experiments were examined as well. First, characteristics of the S02 and
sulfate concentration distributions attributable to continuous and modulated
emissions (one week on, one week off) of three New York point sources were
examined. Second, the detectability of sulfur deposition calculated by
differencing local-scale mass flux measurements was analyzed. Results of
these analyses are summarized below.
1. The S02 and sulfate concentration signals attributable to the
three New York point sources are generally small relative to
background concentration levels. When measurement uncertainties
are considered, the frequency of 1-hour S02 signal detection
during two one-month periods is less than 20 percent for the
continuous emissions case. The frequency of detection of
three-hour sulfate concentration is lower. If modulated emissions
are considered, the decrease in detection frequency is highly
dependent on meteorological factors, indicating that for
experiments of one-month duration there is a good chance sulfur
concentration data from the "non-plume" (as deduced from the
inert-tracer-tagged plume) will be insufficient for source
attribution analysis.
2. The above finding prompted a comparative analysis of week-to-week
S02 and sulfate variability due to (a) natural meteorological
variability and (b) weekly S02 emission modulation. Results
indicate that the natural week-to-week variability is larger in
magnitude than the modulation-induced variability. While these
results should not be interpreted as suggesting that modulation-
induced signals are not extractable from data, they do suggest
that such a small data sample (from a one-month experiment) may be
insufficient for a conclusive source attribution analysis.
3. An analytic mass balance approach was used to estimate sulfur dry
deposition (as S02 and sulfate) as a function of the downwind
distance from a source and the distance interval separating
hypothetical mass flux measurement locations. Results suggest
that the minimum mass flux measurement precision (expressed as the
coefficient of variation) necessary to ensure a detectable
-171-
-------
deposition is on the order of 20 percent for a distance interval
equivalent to the initial 16 hours of plume travel time. (Results
pertaining to other separation and downwind distances are
presented graphically.) If the total sulfur mass flux is measured
at each downwind distance, the measurement precision of the
34S/32S ratio is sufficiently high to offer encouragement to
this type of experiment, under conditions appropriate for the
validity of the mass balance approach (i.e., unstable, well-mixed
conditions, high deposition velocity). If the mass flux is not
completely accounted for in the measurements, the necessity of
calculating sulfur flux differences by differencing the reactive-
to-inert tracer concentration ratios introduces the measurement
uncertainty of the inert tracer into the analysis. Since the
precision of measuring SF6 is considerably lower than the of
measuring 34S/32S, the experiment appears less feasible.
4. The model was exercised to examine the consequences of removing
some of the assumptions invoked in the mass balance analysis. The
results indicate that, conditions in which dry deposition depletes
only the lower portion of the plume (e.g., high aerodynamic
resistance associated with stable conditions), the minimum mass
flux precision requirement decreases to 1 percent for spatial
scales considered in the proposed reactive tracer experiment.
-172-
-------
SECTION 6
ADDITIONAL UNCERTAINTY ANALYSES
Portions of the COMPEX design were evaluated through analyses of an
assortment of data sets currently available. This section reviews data sets
currently available from two long-range transport experiments, the SURE
program, and precipitation chemistry networks. The topics considered include:
the potential characterization of long range transport by
ground-level concentration measurements;
the climatological categorization of data for empirical analyses;
and
the frequency of source/receptor interactions.
6.1 Analysis of Long Range Pollutant Transport Using Tracer Data
The Cross-Appalachian Tracer Experiment (CAPTEX) was designed to provide
data on the long range transport and dispersion of pollutants for use in
evaluating long range transport models. The experiment was held in the fall
of 1983 and consisted of limited releases and measurements of perfluorocarbon
(PFC) tracers over the northeast. Data available from the experiment include
short-term samples of the tracer on a ground level sampling network and
aircraft. Meteorological data from an enhanced rawinsonde network are also
available. The CAPTEX data are of interest for analysis because they show
transport and dispersion patterns which are similar to those that will be
studied in the combined experiment. The objectives of this review are to
qualitatively examine the patterns of ground level perfluorocarbon
concentrations to evaluate the likelihood of determining trajectories from
tracer data and to evaluate the sampling network resolution. The combined
experiments depend on both PFC sampling (on similar spatial and temporal
scales) and the use of ground level samples to determine trajectories for
transmittance calculations.
-173-
-------
6.1.1 Summary of the CAPTEX Program
The CAPTEX program is described in a revised work plan written by Ferber
and Heffter (1983). During September and October, 1983, a PFC tracer was
released in seven experiments, five of which were made from Dayton, Ohio, and
two were made from Sudbury, Ontario. The releases were generally of three
hours duration over which 200 kg were released. Releases were made in the
afternoon to correspond to the maximum possible convective mixing and the
largest distribution of tracer in the mixed layer.
Sampling was performed by seven aircraft and a ground sampling network of
approximately 100 samplers distributed over northeastern United States and
Southern Ontario. The samples used for this analysis were three-and six-hour
samples from the ground level network.
The aircraft data and the final CAPTEX data base were not available at the
time of the analysis. As a result, the data used in the analysis did not
represent a complete data set nor were they satisfactorily calibrated. This
limited the usefulness of the data by an inability to specify absolute
concentrations and data gaps due to sites with missing data. The problems
with the data were sufficient to limit the use of the data to a qualitative
analysis of the potential for determining trajectories, evaluating sampler
spacing, and examining the gross features of the concentrations patterns.
Sample spacing of the ground sampling network was established by different
criteria in the downwind and cross-wind directions. In the downwind
direction, samplers were placed at approximately 100 km intervals starting at
300 km and continuing 500 to 1000 km. The cross-wind spacing on sampling arcs
was established by estimating the expected width of tracer plumes as a
function of travel time. Since two release sites were used, the final
sampling grid was adjusted to provide the required resolution from both sites
without overlapping sites.
-174-
-------
6.1.2 Analysis Results
The analysis results are summarized as qualitative observations on
transport and dispersion:
Ground level tracer concentrations tend to lag the transport of
the center of mass of the tracer plume or puff. In some cases,
substantial delays were encountered where tracer concentrations
remained in an area after the main tracer cloud was transported
downwind on the sampling grid.
# In all cases studied, the progression of the tracer cloud could be
observed in time sequences of spatial plots. Plotting maximum
concentrations for each release episode provided a tracer
trajectory. These data indicated promise in estimating trajectory
positions using tracer data.
Tracer plumes appeared to be elongated and narrow relative to the
sampler spacing in a number of release events. Missing
observations on the network appeared to be the significant cause
of difficulty in describing the transport and dispersion of the
tracer.
Tracer data in some of the experiments indicate that tracer plumes
can be split by wind shear with parts of the release being
transported on significantly different paths.
6.1.3 General Observations on the CAPTEX Experiments
In addition to the questions of uncertainty and the feasibility of the
techniques in the combined experiments, the CAPTEX program provided some
information on the operational aspects of the experiment. Discussions with
the CAPTEX participants provided the following observations on experimental
design:
CAPTEX suffered data losses due to non-functioning tracer
samplers. In some tests, up to a third of the samplers were not
in operation due primarily to mechanical and electronic problems.
Some of the problems appeared to be weather-related and indicated
a deficiency in environmental testing of the samplers. Such high
data loss rates would not be acceptable in the combined
experiments due to a very high reliance on the tracer data for
determining transport paths.
The combined experiments use aircraft sampling in measurements to
support mass balance calculations. CAPTEX aircraft operations
-175-
-------
were hindered by flight limitations imposed in some areas of the
country and the reluctance of temporary pilots to fly in some
areas of complex terrain including over the Great Lakes. One
aspect of operational planning for the study is the definition and
approval of strict flight patterns. In addition, it is suggested
that project pilots be dedicated to the project.
Sounding times for the standard rawinsonde network typically
represent only transition times in the diurnal cycle and are,
therefore, not the most representative for air quality studies.
6.2 Climatological Analysis of the Experimental Design
The main COMPEX experiments will take place over a one year period as
specified by EPA. The objective of the program is to derive empirical
source/receptor relationships which will require a sufficient number of
experimental events to generate acceptable confidence limits on the results.
This section summarizes some of the studies performed to investigate the
adequacy of the experimental program. Issues considered in the analyses
include:
Characteristic durations of wet deposition events.
Climatological categorization of data in previous studies in terms
of deposition event characteristics.
Adequacy of a one year experimental program for empirical
deposition studies.
6.2.1 Duration of Wet Deposition Events
Precipitation and corresponding wet deposition events vary in duration and
timing which result in difficulty selecting an adequate sampling time. This
subsection briefly examines the duration of precipitation events relative to
the sampling frequency proposed as part of COMPEX. The COMPEX precipitation
chemistry sampling protocol is for event sampling with a daily maximum
duration of sampling. Ambient pollutant concentrations are to be collected on
a six hour basis.
-176-
-------
Table 6-1 summarizes the duration of precipitation and deposition events
reported by Thorp and Scott (1982). The preponderance of short-lived
«6-hour) storms in the summer would seem to indicate that precipitation
chemistry samples should be collected and analyzed on a 6-hour basis in the
summer. However, if most storms are separated by 40 to 60 dry hours in the
summer, then a 24-hour sample would generally cover the whole storm but not
more than one storm. Nonetheless, it may still might be important to collect
6-hour precipitation chemistry samples for direct comparison to the 6-hour
ambient pollution concentration obtained in the summer months. Winter data
showed an infrequency of short-lived (<6 hours) storms.
6.2.2 Climatological Characterization of Deposition Events
To select schemes for statistically analyzing data from acidic deposition
studies it is useful to examine data from previous experiments to first
understand, to the degree possible, concentration and deposition events and
then to review possible schemes for analysis. Variations in sulfur deposition
and concentrations are significantly influenced by variations in
meteorological conditions. Raynor and Hayes (1982), Henderson and
Weinggartner (1982), Niemann (1982), Mueller and Hidy (1983), and others have
used several data sets to demonstrate the seasonality of airborne sulfate
concentration and deposition. This dependence takes the form of summer peaks
in concentrations even though precursor SOz emissions may not be a maximum
during this period. On a shorter time scale, Tong and Batchfelder (1978)
demonstrated with time/distance transects of daily concentrations in the
northeast that sulfate concentrations exhibit a wave-like pattern related to
cyclonic migrations.
-177-
-------
TABLE 6-1
DURATION OF PRECIPITATION/DEPOSITION EVENTS
PERCENT OF STORMS AND PERCENT OF PRECIPITATION
VERSUS STORM DURATION PERIODS, BY SEASON
Percent Storms
Percent Precipitation
Storm Durations*
Summer (J,J,A)
< 6 hrs
82
58
< 24 hrs
100
98
Winter (D,J,F)
< 6 hrs
44
7
< 24 hrs
86
52
Storm Frequencies
Storms occur every 40 to 60 hours in summer
Storms occur every 35 to 90 hours in winter
* Consecutive precipitation events were deemed different storms if separated
by 3 dry hours in summer and 6 dry hours in winter.
-178-
-------
Ambient Sulfate Episodes
Mueller and Hidy (1983) and Tong and Batchfelder (1978) analyzed data from
the SURE program (Section 4.1) and described synoptic conditions for a total
of 14 events resulting in high or low regional sulfate concentrations. Most
significant peak episodes are characterized as follows:
1) Conditions favorable to an accumulation of SOz emissions in
various source regions.
2) Conditions must be favorable for the conversion of sulfur dioxide
to sulfate. High incoming solar radiation and high atmospheric
moisture content have been demonstrated to greatly enhance the
conversion of sulfur dioxide to sulfate in the atmosphere. Slow
moving anticyclonic systems in the summer months provide the
mechanism for conversion of sulfur dioxide to sulfate with clear
skies and ample solar insolation. Further, as anticyclones
migrate across the eastern United States, linking with the
semi-permanent Burmuda high pressure system, southwesterly winds
on the back side of the high pressure system advect moisture from
the Gulf of Mexico into the region. Such conditions are
associated with the maritime tropical air mass frequently
associated with elevated sulfate events in the eastern United
States during summer months.
3) Conditions including southwesterly winds on the back side of
anticyclones. The transport wind provides a link between high
emissions areas and critical receptor regions.
4) Conditions where enroute precipitation scavenging is small.
A useful scheme for understanding the relationship between synoptic
conditions and elevated sulfate concentrations was presented by Mueller and
Hidy (1983). The method is basically a classification system for describing
air mass types and is reproduced in Figure 6-1. Depending on the position of
the high pressure system in relation to a particular source or receptor
region, vastly different air mass characteristics may be experienced. In this
scheme, a high pressure system, originating in Central Canada, is centered
over the Great Lakes. The air mass maintains the basic characteristics of its
source region a continental, polar location (hence the designation "cP").
The leading edge of the high pressure system is characterized by northerly
-179-
-------
Figure 6-1. Air mass classification scheme used in SURE data analysis
SOURCE: Mueller and Hidy (1983)
-------
wind components which advect colder air into a region (cPk - "k" for
"kolder"), while southerly winds behind the center of the high advect
typically warmer air (or "cPw"). The "cP2" designation is used to identify
the center of the high pressure system. Two other air mass types are
described in this approach. The transitional (Tr) air masses essentially
describe cyclonic systems which are not stratified in terms of temperature or
moisture, but rather are well mixed. All frontal systems are grouped in this
category. The final category is the maritime tropical air mass (mT).
Although not shown in this configuration in Figure 6-1, mT air masses
frequently extend into the midwest and northeast, especially in the summer
months in association with Bermuda High. The primary difference between cPw
air masses result from advection of warm, continental air and are therefore
drier than air advected from the Gulf as represented by the mT category.
Mueller and Hidy (1983) computed mean values for a number of parameters,
including sulfate concentrations, at a number of observing stations for each
air mass category (Figure 6-2). In addition, the number of sulfate events
encompassing increasingly large geographic areas were compiled for each air
mass category (Table 6-2).
From Figure 6-2, average ambient sulfate concentrations in the northeast
were found to be highest during mT air masses, followed by cPw and cP2 air
masses. In contrast, cPk air masses, which typically occur directly behind
cold fronts and are characterized by northerly winds, exhibit the lowest
ambient sulfate concentrations. Transitional air masses, those associated
with cyclonic systems and frontal zones, also show relatively low sulfate
concentrations.
-181-
-------
20-
S04 15
(ug/m3) c<
5
.8
.6
A'
ABS. "
HUM. l5'
(g/m3)
5-
PPT
(cm)
7.5-
50-
2.5-
1020-
.
Imb)
WS
(m/s)
6
4
2-
c?2 Tr
cPk cPwl mT
cP2 Tr
cPk c.°w ml
L/
cP2 Tr
cPkIcPw mT
\:
c°2 Tr
cPk IcPwl mT
t i t t i
i i I r
Montague, MA Scranton, PA Indian River, DL Res. Tri. Pk., NC
Figure 6-2. Variations of arithmetic mean values for individual
areometric parameters for Class I stations in the
northeast coast region. SOURCE: Mueller and Hidy (1983)
-182-
-------
TABLE 6-2
ANNUAL PERCENTAGE OP EVENT DAYS BY AIR-MASS CATEGORY
Air Mass Category Annual Percentage of Days
Event Group cPk cP2 cPw Tr mT in each Event Group
Regional 0 12 7 29
Subregional 04443
Nonregional 2 10 2 18 0.5
No Event 6 4 0.5 12 0
Annual Percentage of 8 30 13.5 36 12.5
Days in each air mass
30
15
32.
22.
100
5
5
Key:
Regional Event - more than 15% of sulfate recording stations in
Northeast recorded concentrations greater than 15 ug/m3.
Subregional Event - 5-15% of stations greater than 15 ug/m3.
Nonregional Event - less than 5% of stations greater than 15 ug/m3.
No Event - no stations greater than 15 ug/m3.
Source: Mueller and Hidy (1983)
-183-
-------
These findings are supported somewhat by the results shown in Table 6-2.
In terms of the duration and the areal extent of each sulfate event, cP2 and
mT air masses are the most frequent. In other words, elevated sulfate levels
covering a large geographic area occur most frequently with cP2 and mT air
masses, and do not occur at all with cPk air masses. Conversely,
sulfate-clear days occur most frequently with Tr and cPk masses, and not at
all during mT conditions.
Also shown in Table 6-2 is the frequency at which various air mass types
occur in the SURE experimental region the midwest and the northeast. For
example synoptic conditions dominated by cyclonic systems (Tr category) occur
36% of the time, whereas only 12.5% of the days in a year are characterized as
being under maritime tropical (mT) conditions. Such climatological data are
important considerations in the design of any acid deposition measurement
program.
Wet Deposition Episodes
As mentioned, the amount of acid-forming substances deposited on the
ground by wet processes is dependent upon both the presence of sulfates in the
atmosphere and the occurrence of precipitation. Precipitation throughout the
midwest and northeast can be associated with several types of synoptic
systems. The passage of cyclonic systems through the regions are responsible
for most of the precipitation, whether from warm, cold, or occluded fronts, or
from the cyclone itself. Other rain producing events include pre-frontal
squall-lines and convective thunderstorms.
Rain is much more frequently associated with the deposition of
acid-forming substances than is snow. This could be due to reduced
photochemical conversion of sulfur dioxide to sulfates in the winter and
because snow has been found to be much less efficient than rain in scavenging
-184-
-------
sulfate particles (Niemann, 1983). Hence, the peaks deposition events of acid
forming substances through wet processes occurs primarily with precipitation
in liquid form.
The amount of acid-forming substances deposited during precipitation
events largely depends on the amount of precipitation occurring during the
event and the trajectory of the air mass en route to the receptor region.
Wilson, et al. (1982) illustrates in Figure 6-3 the amount of precipitation
received at Whiteface Mountain, New York in 1978 by trajectory sector, and the
resulting sulfate ion concentrations in the precipitation. Most of the
precipitation (56%) received at Whiteface Mountain results from air masses
that originate south and/or west of New York, and similarly, most of the
sulfate ion concentration (64%) contained in the total annual precipitation
amount is from the same sector. In fact, a disproportionate amount of the
sulfate ions contained in the total annual precipitation at Whiteface Mountain
results from the southerly through westerly sector, indicating the importance
of source regions in those directions.
Raynor and Hayes (1982) have provided a description of wet deposition
events at Brookhaven National Laboratory in New York as a function of synoptic
conditions and types of precipitation amounts. Most of the precipitation
occurring at Brookhaven occurs during warm and cold front passages, and the
deposition of the various chemical species occurs during the same conditions.
They note that the amount of sulfates deposited during sguall line
precipitation events exceeds the proportion of rainfall received in such
systems implying a dependence on precipitation. They present plots of the
amount of precipitation and chemical species deposited at Brookhaven as a
function of precipitation type. A distinction is also made in the data
-185-
-------
OHIO VALLEY/MIDWEST-
CANADIAN/GREAT LAKES-
TOTAL PRECIPITATION-
17.3 LITER
7.9 LITER
30.6 LITER
A
c
M
O
H
h
<
M
0.
U
£
ft.
CANADZAN/
GREAT UAKCS
26 X Or TOTAL.
I 20
tea i ea 21 a 2««Q 27Q saa
TRAJECTORY SECTOR
aoa
ee
OHIO VALLEY/HIDUEST-
CANADIAN/GREAT LAKES-
TOTAL DEPOSITION-
31* OP" TOTAI.
1087.9 MG/M2
S31 .6 MG/IM2
1708.5 MG/M2
ea
aa
T T
t za i ea t ea 21 a
TRAJECTORY
T T
2-4B 270
SECTOR
saa ssa 300
Figure 6-3. Precipitation (top) and sulfate ion concentration in precipitation
(bottom) as a function of the directional sector through which the
air parcel passed to reach Whiteface Mountain, New York in 1978.
SOURCE. Wilson, et al, (1982)
-186-
-------
between deposition amount and concentration which is a function of
precipitation type. For example, the concentration of sulfates in convective
rains is typically higher than that of "general" rains which are presumed to
mean warm frontal rains. Total deposition from convective storms was shown
typically to be less because the duration of precipitation in combination with
the higher concentrations in rain water were insufficient to exceed the total
deposition of long steady rain events.
Niemann (1982) has provided an analysis of "exceptional episodes" of wet
deposition at Whiteface Mountain and Ithaca, both in New York, based on data
collected during the MAP3S program. His review indicated that 3 deposition
events at Whiteface Mountain in 1978 contributed 30% of the total annual
amount of sulfates deposited. At Ithaca in 1980, 3 events contributed 14% of
the annual sulfate total. It is clear that acid deposition due to wet
processes can be of an episodic nature. Niemann goes on to point out that
such episodes tend to be localized, being on the scale of thunderstorms (about
104 km2).
Henderson (1982) also sheds some additional light on the episodic nature
of wet acid deposition events using data collected during the MAP3S program.
The average sulfate loading per precipitation event occurring with southwest
trajectories at Ithaca, New York during the summers of 1978 and 1979 was found
to be 720 micro-moles per square meter. However, the peak sulfate loading
observed during the period was 2698 micro-moles per square meter almost a
factor of four times higher than the mean value.
Synoptic Classification Schemes
Few classification schemes for statistical analyses have been developed
due to the lack of an appropriate data base. The SURE analyses developed a
classification scheme for sulfate concentrations which could be generalized to
-187-
-------
dry deposition episodes. The scheme is based on identifying synoptic types. A
limitation of the scheme is the large number of cases which fall into the
transitional classification which includes most precipitation deposition
events.
For meteorological and climatological studies, various other schemes have
been developed but in general their orientation is directed toward large scale
forecasting applications. For example, a scheme by Krick (1943) which
initially appeared interesting for this study proved too specific and
extensive for use. The Krick scheme included seven basic synoptic types based
on location of a Pacific high. These classes were further distributed over 36
types which included seasonal effects and 10 phases or subclasses.
Perhaps the best climatological scheme for analysis may be a modification
of the SURE scheme to allow allocation of transitional cases among additional
cases. The transitional cases are critical in the characterization of wet
deposition events. Expansion of the precipitation events as a subclass of the
transitional cases could possibly be in the following categories:
Precipitation events associated with cold fronts.
Non-frontal precipitation.
Warm frontal of overruning precipitation.
Precipitation in a cyclone without organized or distinguishable fronts.
Precipitation in occluded systems.
In addition to the synoptic schemes several categorization schemes are
possible such as by direction of trajectories. A problem in developing
schemes is allocation of the limited number of events and limited data
available from the field program to demonstrate this concern. Data analyses
by Thorp and Scott (1982) can be considered.
-188-
-------
The authors studied precipitation records from 77 locations covering 5
summers and 5 winter months over a 3-year period. Statistics were compiled
for a total of 4516 summer and 3870 winter precipitation events. These data
show that the average number of precipitation events in a given summer month
is 12. In winter the average monthly number of such events is 10. Thus, in a
year the total number of precipitation events at any single location is likely
to be approximately 132.
During the proposed one-year monitoring program, tracer releases will be
made during the first day of 120 3-day periods. No tracer will be released on
the second or third days of the 3-day cycle. Since roughly one-third of the
days in a year record precipitation, then only one-third of the tracer
releases can be expected to be affected by precipitation. Therefore, the
number of tracer releases likely to coincide with precipitation events at any
single location is limited to 40. Only 15 to 20 precipitation events are
likely to occur during the summer months. This is a somewhat limited data
base upon which to establish relationships among measured parameters. Any
additional sub-grouping of events will further reduce the number of events in
any group and further decrease the significance of the statistical
relationships obtained among the measured parameters.
6.2.3 Analysis of the Adequacy of a One-Year Monitoring Program
It has been noted in previous sections that the frequency of
source/receptor interactions is small. Remedial modifications were made to
the COMPEX design based on these observations. In the previous subsection it
was noted that the combination of a low source/receptor interaction frequency
of the large array of potential meteorological conditions require caution to
limit the number of categories used in analyzing the data. This section
-189-
-------
describes further the frequency of deposition events in an attempt t
determine the adequacy of a one year experimental program in developir.
source/receptor relationships.
Tables 6-3 and 6-4 summarize an analysis of MAP3S precipitation chemistr
data performed to examine the year to year variability of wet depositio
data. The data show that both precipitation and deposition vary by ;
considerable amount from year to year. Precipitation varies by up to ±21
percent from the 3- to 4-year mean values at four of the sites shown, and u{
to ±30 to 40 percent at the Virginia site. Year to year variations ii
precipitation can thus range up to ±40 percent. Meanwhile, depositior
varies by up to ±20 percent also, but the change in deposition amount does
not necessarily follow the change in precipitation amount. For example, at
the Brookhaven site, the June 1976 to May 1977 precipitation amount was 21
percent below the 4-year mean, while the deposition amount was 15 percent
above the 4-year mean for the same period. In some instances there is a
tendency for deposition to increase or decrease along with precipitation, but
in others (e.g., 1978 at the 4 MAP3S sites), the two parameters exhibit
opposite changes from year to year. Also noteworthy is the relatively
constant deposition rate at the Perm State MAP3S site, which occurred despite
the relatively large year to year changes in precipitation. Of the MAP3S
sites, Perm State had the most complete set of valid precipitation and
deposition data for the 3-year period of study.
A possible explanation for the lack of correlation between annual average
precipitation and annual average deposition is the year-to-year variability in
seasonal precipitation amounts coupled with the seasonal variability in the
total sulfur concentrations in the precipitation. An example of this
phenomenon is shown in Table 6-5 which results from analysis of the data
obtained from the MAP3S study (MAP3S/RAINE Research Community, 1982).
-190-
-------
TABLE 6-3
YEARLY PERCENT DEVIATIONS FROM 4-YEAR MEAN
PRECIPITATION AND SULFATE DEPOSITION AMOUNTS AT
BROOKHAVEN NATIONAL LABORATORY, UPTON, LI, NY
Year
6/76-5/77
6/77-5/78
6/78-5/79
6/79-5/80
Percent Change from
4-Year Mean Precipitation -21
Percent Change from
4-Year Mean Deposition +15
+18
+10
+17
-1
-14
-24
TABLE 6-4
YEARLY PERCENT DEVIATION FROM 3-YEAR MEAN
PRECIPITATION AND TOTAL SULFUR DEPOSITION AMOUNTS AT
4 MAP3S LOCATIONS*
Location
Whiteface
Mountain, NY
Ithaca,
NY
Perm State,
PA
Charlotte,
VA
Parameter
Precipitation
Deposition
Precipitation
Deposition
Precipitation
Deposition
Precipitation
Deposition
1977
+13
+6
+15
[+5]
-7
-2
-28
[-20]
1978
-16
[+2]
-21
[+2]
-11
+3
-10
t+5]
1979
+3
-8
+6
_y
+18
-1
+38
[+15]
* Brackets indicate more than 2 months missing data.
-191-
-------
TABLE 6-5
ANNUAL MEAN AND MONTHLY PERCENT DEVIATIONS FROM ANNUAL
MEAN CONCENTRATION, PRECIPITATION AND DEPOSITION VALUES FOR
CHARLOTTE, VIRGINIA, 1977-1978
Annual Mean Percent Deviations from the Annual
of the Monthly Mean of the Monthly Averages
Averages JFMAMJJASO_N
1977
Total Sulfur
Concentration 44.9 - -67 -49 -51 +36 +20 +145 -4 +20 -
(um/1)
Precipitation 4.5 - -76 +20 +15 +44 -34 -56 -5 -16 - - ^
(cm)
Total Sulfur
Deposition 1808 - -91 -31 -37 +120 -10 +21 +2 +13 -
(pm/m2)
1978
Total Sulfur
Concentration 32.1 - - -25 0 - - +21 -4 +25 +28 +15 -
(pm/1)
Precipitation 7.7 - - +29 -28 - - +55 +42 -28 -72 -7
(cm)
Total Sulfur
Deposition 2398 - - +4 -26 - - +89 +46 -7 -63 +11 -5
(um/m3)
-192-
-------
The above data and discussion show that the total annual deposition found
in a one-year monitoring program may not be representative of the total annual
deposition to be found in other years. Since a longer monitoring program is
not contemplated, methods must be developed and used to relate the data
gathered within a one year program to other years. Such methods will be
discussed in the paragraphs that follow.
Since increases and decreases in precipitation amounts do not necessarily
produce corresponding increases and decreases in deposition amounts,
precipitation alone cannot be used to extrapolate deposition amounts found in
one year to deposition amounts likely to be found in other years. Deposition
amount is a function of both precipitation amount and the pollutant
concentration in the precipitation (Raynor and Hayes, 1984). Year to year
(and other time periods) changes in precipitation are a function of
meteorological and climatological factors.
The preceding data show a number of common features of seasonal
variability in both the total sulfur (or sulfate) concentration in the
precipitation, and in the total sulfur deposition. Both parameters generally
show their maxima in the summer and their minima in winter.
The most prominent feature of the Charlotte, Virginia data is the
disparity between the 1977 and 1978 summer precipitation amounts, In 1977,
the June, July, and August precipitation amounts were well below the mean for
the year while the precipitation amounts for July and August in 1978 were well
above the average for that year. As a result, the summer (July and August)
and annual deposition amounts in 1978 were much greater than the summer (June
through August) and annual deposition amounts in 1977. However, it is
noteworthy that the summer 1977 rainfall deficiency did not produce an
equivalent deficiency in deposition. The total sulfur concentrations were
-193-
-------
greater in the precipitation in the summer of 1977 than in the summer of 1978,
so this partially offset the lack of precipitation in 1977.
The above analysis of seasonal variability in precipitation and deposition
amounts show that two years with equal precipitation amounts may not produce
equal deposition totals. If the precipitation in different years is evenly
distributed over all the months in the year, then the deposition patterns and
totals for each year will have a greater tendency to be similar. If the
precipitation in a given year occurs predominantly in the winter, then the
year with the heavy summer rains will tend to have greater deposition totals
than the other years.
These factors, plus variations in emissions, produce year-to-year changes
in the pollutant concentrations in precipitation. Therefore, the key to
relating the results of a one-year monitoring program is to first determine
the categories of meteorological-climatological conditions that are associated
with different types of short-term precipitation-concentration events. This
link has been attempted through the integration of previous and current field
study data into the COMPEX design by locating sampling sites at current or
former monitoring sites.
6.3 Potential for Pollutant Transport as Indicated by Source/Receptor Pair Data
The long range tracer component of the combined experiment requires that
releases be distributed within source areas and that releases from different
source areas will be scheduled on the basis of trajectory predictions. The
objective of this plan is to increase the chances for a tracer to affect the
receptor region. An analysis of the Atlantic Coast Unique Regional
Atmospheric Tracer Experiment (ACURATE), Heffter, et al. (1984), provides
additional evidence on the probability of a tracer from a specific source area
-194-
-------
influencing a specific receptor location. This analysis supports the need to
devise methods to increase tracer impact frequency because the climatological
probability of impact is low and decreases rapidly with distance. The ACURATE
data suggest that the number of days annually available for analyses from a
single source receptor pair separated by 1000 Jon is expected to be 47 days for
continuous releases but only 15 days for the third day release schedule
described for COMPEX.
The ACURATE program used Krypton 85 released from the Savannah River
Project (SRP) as a tracer of opportunity. Five monitors were place northeast
of SRP at distances of 325 km to 1050 km. The four closer monitors provided
12 hour integrated samples. The fifth monitor sampled over 24-hour periods.
Data were collected from March 9, 1982 through September 30, 1983 resulting in
1158 possible 12-hour sampling periods or 579 days.
Release events were identified as periods of at least 24 hours with Kr-85
releases. The .events had to be separated by at least 24-hours when Kr-85 was
not vented to the atmosphere. A total of 102 events were identified which
contained 768 half-day periods corresponding to the 12-hour sampling period.
Kr-85 releases occurred 66 percent of the time.
Kr-85 concentrations above background were counted for each sampling
period and the frequency of occurrence was calculated for each monitor. The
concentrations were pooled into categories of 100 times the detection limit,
10 times the detection limit and all samples above the detection limit. Data
from Heffter, et al. (1984) in Figure 6-4 show the frequency of occurrence of
each concentration category. These are plotted in Figure 6-4 as a function of
distance from the source. The data are well behaved and straight lines were
drawn on the probability graph.
-195-
-------
Q-
UJ
Of.
o
1/1
o
Qi
1000
900
800
700
600
500
400
300
1Q BUL
I I
> BUL
0.01 0.1 0.5 1 5 10 20 30 40 50 60 70
PROBABILITY OF OCCURRENCE - PERCENT
Figure 6-4. Frequency of occurence of Krypton 85 concentrations as a
function of distance for concentration levels ^background upper
limit (BUL), > 10 x BUL and > 100 x BUL.
-196-
-------
The release rates of inert tracer for the combined experiment are
calculated for 10 times background concentration although 100 times background
*as considered. The data from Figure 6-4 provide estimates of the frequency
>f occurrence of impact as a function of source receptor separation.
'oncentrations were observed at 10 times the detection limit of the system 10
ercent of the time at 300 km and 2.5 percent of the time at 1000 km.
oncentrations 100 times the system's detection limit were observed 2.5
ircent of the time at 300 km and about 0.2 percent of the time at 1000 km.
: course, it must be kept in mind that the source rate was not controlled and
ie total frequency of occurrence is a function of source term, meteorology,
d distance. Tracer release on a full time basis should increase the
equency of occurrence by one third while tracer release on an every third
? basis as planned for the combined experiments would decrease the frequency
occurrence to half of those shown.
An examination of the events concurrently affecting the two closest
itors was performed. The first monitor was 325 km from the source. The
ond monitor was 475 km from the source. Both reported concentrations above
detection limit for the same, or subsequent 12 hour periods, 99 times.
3 number of events constituted 58 percent of the impact periods at the
>nd monitor. A more complete analysis is required to determine if this
erence in number of impacts is due to trajectories that did not pass from
tor 1 to monitor 2 or to a narrow plume that missed the second location.
ACURATE data could be used to establish uncertainties and analysis
jdures for trajectories, concentrations at receptor locations, and source
ibutions at receptor points.
he significance of the ACURATE data is to reinforce the observation that
requency of single point sources affecting single receptors is very low.
-197-
-------
The long range tracer component of the combined experiment is desigr
minimize this effect in three ways:
1) Tracer emissions are to be distributed over a relatively larc
area (appropriate scale: 100 km).
2) Sampling grid resolution was increased over initial designs.
3) Tracer emissions will be made from one of three pairs of trac
release sites determined by forecasts to provide the best da
collection rate.
These techniques will increase the frequency of observing tracer i
substantially but the final expected frequency cannot be determined v
data currently available. Some additional analyses of the ACURATE da
be beneficial. These are:
1) A study of the reduction of frequencies due to concentrat:
below background or detectable limits versus reductions due so.
to transport.
2) A case study of data to characterize favorable and unfavor
transport episodes and to determine if distinguishable
categories exist.
3) An additional case study to determine if a one year experime
period is sufficient to develop an empirical source/reef
relationship.
-198-
-------
REFERENCES
AIHA. 1972. Air Pollution Manual Part I, Evaluation, 2nd edition. American
Industrial Hygiene Association, Detroit, Michigan
Altshuller, A. P. 1976. Regional Transport and Transformation of Sulfur
Dioxide and Sulfates in U.S.. Journal of the Air Pollution Control
Association, 26:318-324.
Barry, P. J. 1974. Stochastic Properties of Atmospheric Diffusivity. Atomic
Energy of Canada Ltd., Chalk River Nuclear Laboratories, Chalk River,
Ontario. Published in The Effects of Sulfur in Canada, NRC of Canada.
Ottawa, Ontario.
Bhargava R P. 1980. "Selection of a Subset of Pollution Stations in the
Bav Area' of" California on the Basis of the Characteristic, 24-hour Suspended
Particulate Concentrations, From the Viewpoint of Variation." Technical
37 Study on Statistics and Environmental Factors in Health,
of Statistics, Stanford University.
palo
,, c Tfpifer and J. A. McDonald. 1981. "Aircraft
Blumenthal D^ ^, J^ aS Met'eoroloaical Parameters During the Sulfate
ill*^ . i«rmt>\ program."
-------
REFERENCES (Continued)
Dietz, R. N. , E. A. Cote, R. W. Goodrich. 1976a. "Development
Application of Sulfur Hexaflouride Measurement Capabilities at Brookhav
Bookhaven National Laboratory, Upton, New York (BNL 21087).
Dietz, R. N. , and W. F. Dabberdt. 1983. "Gaseous Tracer Technology
Applications." Brookhaven National Laboratory, Upton, New York {BNL 335851
Dietz, R. N., R. W. Goodrich, J. D. Smith and W.
Report of the Brookhaven Explosive Tagging Pror
Laboratory, Upton, New York (BNL 21041).
Dietz, R. N., R. W. Goodrich, E. A. Cote, G.
1983. "Passive and Programmable Samplers for
with Perfluorocarbon Tracers." In
Laboratory, Upton, New York.
prer
Dietz, R. N. , and G. I. Senum.
Iracers-"
1984
1976b.
0) £>
& ffi
"Sun
Nati
-------
REFERENCES
AIHA. 1972. Air Pollution Manual Part I, Evaluation, 2nd edition. American
Industrial Hygiene Association, Detroit, Michigan
Altshuller, A. P. 1976. Regional Transport and Transformation of Sulfur
Dioxide and Sulfates in U.S.. Journal of the Air Pollution Control
Association, 26:318-324.
Barry, P. J. 1974. Stochastic Properties of Atmospheric Diffusivity. Atomic
Energy of Canada Ltd., Chalk River Nuclear Laboratories, Chalk River,
Ontario. Published in The Effects of Sulfur in Canada, NRC of Canada.
Ottawa, Ontario.
Bhargava, R. P. 1980. "Selection of a Subset of Pollution Stations in the
Bay Area of California on the Basis of the Characteristic, 24-hour Suspended
Particulate Concentrations, From the Viewpoint of Variation." Technical
Report No. 37, Study on Statistics and Environmental Factors in Health,
Department of Statistics, Stanford University.
Blumenthal, D. L. , W. S. Keifer, and J. A. McDonald. 1981. "Aircraft
Measurements of Pollutants and Meteorological Parameters During the Sulfate
Regional Experiment (SURE) Program." Electric Power Research Institute, Palo
Alto, CA. (EA-1909).
Bowne, N. E. 1982. "A Comparison of Tracer and Sulfur Dioxide Measurements."
TRC Environmental Consultants, Inc., East Hartford, Connecticut.
Bowne, N. E. , and R. J. Londergan. 1983. "Overview, Results, and Conclusions
for the EPRI Plume Model Validation and Development Project: Plains Site"
Electric Power Research Institute, Palo Alto, CA (EA-3074).
Brady, P. J. 1978. Optimal sampling and analysis using two variables and
modeled cross-covariance functions. J. Appl. Meteorol. 17;12-21.
Brillinger, D. R. 1975. Time Series Data Analysis and Theory. Holt Rinehart
and Winston, Inc.
Chung, Y. S. The Distribution of Atmospheric Sulfates in Canada and Its
Relationship to Long Range Transport of Air Pollutants, Atmospheric
Environment, 12:1474-1480.
Clark, T. L. , G. J. Ferber, J. L. Heffter, R. R. Draxler. 1984. "Cross-
Appalachian Tracer Experiment (CAPTEX183)." Fourth Joint Conference on
Applications of Air Pollution Meteorology, American Meteorological Society,
Boston, MA.
Delhomme, J. P. 1978. Kriging in the hydrosciences. Advances in Water
Resources, 1:251-266.
Dietz, R: N., and E. A. Cote. 1971. GC determination of sulfur hexafluoride
for tracing air pollutants. Div. Water, Air, and Waste Chem., 11:208-215.
-199-
-------
REFERENCES (Continued)
Dietz, R. N., E. A. Cote, R. W. Goodrich. 1976a. "Development and
Application of Sulfur Hexaflouride Measurement Capabilities at Brookhaven."
Bookhaven National Laboratory, Upton, New York (BNL 21087).
Dietz, R. N., and W. F. Dabberdt. 1983. "Gaseous Tracer Technology and
Applications." Brookhaven National Laboratory, Upton, New York (BNL 33585).
Dietz, R. N., R. W. Goodrich, J. D. Smith and W. Vogel. 1976b. "Summary
Report of the Brookhaven Explosive Tagging Program." Brookhaven National
Laboratory, Upton, New York (BNL 21041).
Dietz, R. N. , R. W. Goodrich, E. A. Cote, G. I. Senum and G. S. Raynor.
1983. "Passive and Programmable Samplers for Atmospheric Transport Studies
with Perfluorocarbon Tracers." In preparation, Brookhaven National
Laboratory, Upton, New York.
Dietz, R. N., and G. I. Senum. 1984. "Capabilities, Needs, and Applications
of Gaseous Tracers." Brookhaven National Laboratory, Upton, New York
(BNL 35108).
Durran, D. , M. J. Meldgin, and M. K. Liu. 1979. A study of long-range air
pollution problems related to coal development in the Northern Great Plains.
Atmos. Environ., 13:1021-1037.
EPA. 1971. Guidelines: Air Quality Surveillance Networks. U.S.
Environmental Protection Agency, Research Triangle Park, North Carolina
(AP-98).
Eddy, A. 1974. An approach to the design of meteorological field
experiments. Mon. Wea. Rev., 102:702-707.
Eddy, A. 1967. The statistical objective analysis of solar data fields. J^
Appl. Meteor., 6:597-609.
Eddy, A. 1974. An approach to the design of meteorological field
experiments. Mon. Wea. Rev., 102:702-707.
Eddy, A. 1976. Optimal raingage densities and accumulation times: A
decision-making procedure. J. Appl. Meteorol., 15:962-971.
Evans, L. S. , G. S. Raynor, and D. M. Jones. 1984. Frequency Distributions
for Durations and Volumes of Rainfalls in the Eastern United States in
Relation to Acidic Precipitation. Water, Air and Soil Pollution, 23;187-195.
Ferber, G. and J. Heffter. 1983. CAPTEX '83 Cross Appalachian Tracer
Experiments: Revised Plan, NOAA ARL, Rockville, MD.
Ferber, G. F., K. Telegadas, J. L. Heffter, C. R. Dickson, R. N. Dietz, P. W.
Krey. 1981. "Demonstration of a Long-range Atmospheric Tracer System Using
Perfluorocarbons." Air Resources Laboratories, Silver Spring, Maryland (ERL
ARL-101).
Forrest, J. and L. Newman. 1973. Sampling and analysis of atmospheric sulfur
compounds for isotope ratio studies. Atmos. Environ., 7:561-573.
-200-
-------
REFERENCES (Continued)
Fowler, M. M. and S. Barr. 1983. A long-range atmostpheric tracer field
test. Atmos. Environ., 17(9):1677-1686.
Freeman, D. L. , N. R. Robinson, J. G. Watson, and R. T. Egami. 1984.
"Propagation of Measurement Uncertainties in Gaussian Plume Models: An
Application to Experimental Data." Fourth Joint Conference on Applications of
Air Pollution Meteorology. American Meteorological Society, Boston, Mass.
Gandin, L. S. 1965. "Objective Analysis of Meteorological Fields." Israel
Program for Scientific Translations, Jerusalem (translated from Russian;
originally published 1963, Leningrad).
Geogapoulos, P. G., and J. H. Scenfield. 1982. Statistical Distributions of
Air Pollutant Concentrations, Environmental Science and Technology,
16:401A-416A.
Hanna, S. R. 1984. Concentration fluctuations in a smoke plume. Atmos.
Environ., 18(6):1091-1106.
Heffter, J. L. , J. F. Schubert, and G. A. Mead. 1984. Atlantic Coast Unique
Regional Atmospheric Tracer Experiment (ACURATE). NOAA Technical Memorandum
(ERL ARL-130).
Henderson, R. G., and K. Weingartner. 1982. Trajectory Analysis of MAP3S
Precip. Chemistry Data at Ithaca, NY, Atmospheric Environment, 16:1647-1666.
Hicks, B. B. 1984. "Background Notes on the Use of Sulfur Isotopes in
Long-Range Tracer Experiments." National Oceanic and Atmospheric
Administration, Oak Ridge, Tennessee.
Hidy, G. M., P. Mueller, and E. Tong. 1978. Spatial and Temporal
Distribution of Airborne Sulfate in Parts of the U.S. Atmospheric Environment,
12:735-752.
Hitchcock, D. R., and M. S. Black. 1984. 34S/32S evidence of biogenic
sulfur oxides in a salt marsh atmosphere. Atmos. Environ., 18(1):1-17.
Horst, T. W. 1977. A surface depletion model for deposition from a Gaussian
Plume. Atmos. Envi ron., 11:41-46.
Horst, T. W. , J. C. Doran, and P. W. Nickola. 1983. "Dual Tracer
Measurements of Plume Depletion." In Precipitation Scavenging, Dry
Deposition, and Resuspension. Elsevier Science, The Netherlands.
Huijbregts, C. J. 1975. "Regionalized Variables and Quantitative Analyses of
Spatial Data. In Display and Analysis of Spatial Data, J. Davis and M.
McCullagh, eds., Wiley, London.
Kaplan, I. R., and S. C. Rittenberg. 1964. Mircrobiological fractionation of
sulphur isotopes. J. Gen. Microbiol., 24:195-212.
-201-
-------
REFERENCES (Continued)
Klein, W. H. 1957. Principal Tracks and Mean Frequencies of Cyclones and
Anticyclone in the Northern Hemisphere. Research Paper No. 40, U.S. Weather
Bureau, Wash. DC, 60 pages.
Korshover, J. 1976. Climatology of Stagnating Anticyclones East of the Rocky
Mountains 1936-1975. ERL-ARL-55, NOAA Technical Memorandum Silver Springs, MD.
Krick, I. P. 1943. Synoptic Weather Types of North America. NOAA Technical
Memorandum.
Lamb, R. G. 1984. Air pollution models as descriptors of cause-effect
relationships. Attnos. Environ., 18(3) :591-606.
Langstaff, J., C. Seigneur, T. Myers, G. Whitten, and M. K. Liu. 1984.
"Design of an Optimum Air Monitoring Network to Support Toxic Exposure
Assessment." Systems Applications, Inc., San Rafael, California
(SYSAPP-84/175).
Leaderer, B. P., and Stalwyk, J. A. 1981. Seasonal Visibility and Pollutant
Sources in Northeaster U.S., Environmental Science and Technology, 15;305-309.
Liu, M. K., D. A. Stewart, and D. Henderson. 1982. A mathematical model for
the analysis of acid deposition. J. Appl. Meteorol., 21:859-873.
Manowitz, B. , L. Newman, W. D. Tucker, and others. 1970. "The Atmospheric
Diagnostics Program at Brookhaven National Laboratory, Third Status Report."
Brookhaven National Laboratory, Upton, New York (BNL 50280).
MAP3S/RAINE Research Community. 1982. The MAP3S/RAINE Precipitation
Chemistry Network: Statistical Overview for the Period 1976-1980. Atmos.
Environ. , 16(7)-.1603-1631.
McNaughton, D. J. , and N. E. Bowne et al. 1984, "Comprehensive Field Study
Plan to Relate Pollutant Sources to Acidic Deposition." TRC Environmental
Consultants, Inc., East Hartford, CT.
Morris, M. D. , and S. F. Ebey. 1984. An interesting property of the sample
mean under a first-order autoregressive model. Amer. Statistician,
38(2):127-128.
Morris, R. E. , D. A. Stewart, S. M. Greenfield, and M. K. Liu. 1984. "A
Preliminary Test of the Ability of Emission Modulation to Establish Empirical
Source-Receptor Relationships." Systems Applications, Inc., San Rafael, CA
(SYSAPP-84/096).
Mueller, P. K., and G. M. Hidy. 1979. "Implementation and Coordination of
the Sulfate Regional Experiment (SURE) and Related Research Programs."
Electric Power Research Institute, Palo Alto, CA (EA-1066).
Mueller, P. K., and G. M. Hidy. 1982. "The Sulfate Regional Experiment:
Documentation of SURE Sampling Sites." Electric Power Research Institute,
Palo Alto, CA (EA-1902).
-202-
-------
REFERENCES (Continued)
Mueller, P. K., and G. M. Hidy. 1983. "The Sulfate Regional Experiment:
Reporting of Findings," 3 vols. Electric Power Research Insititute, Palo
Alto, CA (EA-1901).
Mueller, P. K. and J. G. Watson. 1982. "Eastern Regional Air Quality
Measurements," Volume 1. Electric Power Research Institute, Palo Alto, CA
(EA-1914).
NCAR. 1983. "Regional Acid Deposition: Models and Physical Processes."
National Center for Atmospheric Research, Boulder, CO (Technical Note
NCAR/TN-214+STR).
NRC. 1983. Acid Deposition: Atmospheric Processes in Eastern North America.
National Research Council. National Academy Press, Washington, DC.
Nakamori, Y. , S. Ikeda, and Y. Sawaragi. 1979. Design of air pollutant
monitoring system by spatial sample stratification. Atmos. Envi ron.,
13:97-103.
Netterville, D, D. J. 1979. "Concentration Fluctuations in Plumes."
Syncrude Canada Ltd (Environmental Research Monograph 1979-4).
Newman, L. , et al. 1971. A tracer method using stable isotopes of sulfur.
Proc. American Nuclear Society Topical Meeting on Nuclear Methods in
Environmental Research.
Newman, L. , J. Forrest, and B. Manowitz. 1975a. The application of an
isotopic ratio technique to a study of the atmospheric oxidation of sulfur
dioxide in the plume from an oil-fired power plant. Atmos. Environ.,
9:959-968.
Newman, L. , J. Forrest, and B. Manowitz. 1975b. The application of an
isotopic ratio technique to a study of the atmospheric oxidation of sulfur
dioxide in the plume from a coal-fired power plant. Atmos. Envi ron.,
9:969-977.
Niemann, B. L. 1982. An Analysis of Regional Acid Deposition Episodes and
the Feasibility of Their Deterministic Simulation. U.S. Environmental
Protection Agency, Washington, DC. 19 pages.
Niemann, B. L. 1983. Further Analysis of Relationships Between Scavenging
Ratios for Exceptional Episodes and Air Quality and Storm Parameters. U.S.
Environmental Protection Agency, Washington, DC, 14 pages.
Meteorological Aspects of Acid Rain, C. Bhumralkar (ed.) Ann Arbor Science.
Niemann, B. L. 1982. Analysis of Wind and Precipitation Data for Assessments
of Transboundary Transport and Acid Deposition between Canada and the United
States.
Niemann, B. L. and P. M. Mayerhofer. 1978. "A Climatology and Meteorological
Parameters for Short Range Dispersion and Long Range Transport," in
Proceedings of the 9th Int'l Technical Meeting on Air Pollution Modeling and
Its Applications (Brussels, Belgium NATO 1978).
-203-
-------
REFERENCES (Continued)
O'Conner, J. F. 1961. Mean Circulation Patterns based on 12 years of recent
North. Hemis. data. Mon. Wenther. Review, 89:211-227.
Pack, D. H., et al, 1978. Meteorology of Long Range Transport, Atmospheric
Environment, 12:425-444.
Papoulis, A. 1965. Probability, Random Variables, and Stochastic Processes.
McGraw-Hill, New York.
Pickett, E. E., and R. G. Whiting. 1981. The design of cost-effective air
quality monitoring networks. Env. Monitor. Assess., 1:59-74.
Raynor, G. S., and J. V. Hayes. 1982. Variation in Chemical Wet Depoistion
with Meteorological Conditions. Atmospheric Environment, 16(7):1647-1656.
Sloane, C. S. 1983. Summertime Visibility Declines: Meteorological
Influences. Atmospheric Environment, 17:763-774.
Smith, F., C. E. Decker, R. B. Strong, J. H. White, F. K. Arey, and W. D,
Bach. 1983. "Estimates of Uncertainty for the Plume Model Validation and
Development Project Field Measurements Plains Site." Electric Power
Research Institute, Palo Alto, CA (EA-3081).
Smith, F. B. 1981. The Significance of Wet and Dry Synoptic Regions on Long
Range Transport of Pollution and its Deposition, Atmospheric Environment,
15(5):863-873.
Stern, A. C. 1968. Air Pollution, 2nd edition. Academic Press, New York.
Stewart, D. A., R. E. Morris, A. B. Hudischewskj, M. K. Liu, and D.
Henderson. 1983a. "Evaluation of Episodic Regional Transport Models of
Interest to the National Park Service." Systems Applications, Inc., San
Rafael, CA (SYSAPP-83/020).
Stewart, D. A., R. E. Morris, M. K. Liu. 1983b. "Evaluation of Long-Term
Regional Transport Models of Interest to the National Park Service." Systems
Applications, Inc., San Rafael, CA (SYSAPP-83/215).
Sullivan, P. J. 1984. "Whence the Fluctuations in Measured Values of
Mean-Square Fluctuations?" Fourth Joint Conference on Applications of Air
Pollution Meteorology. American Meteorological Society, Boston, Mass.
Thiebaux, H. J., and F. W. Zwiers. 1984. The interpretation and estimation
of effective sample size. J. Climate Appl. Meteor., 23:800-811.
Thorp, J. M., and B. C. Scott. 1982. Preliminary Calculations of Average
Storm Duration and Seasonal Precipitation Rates for the Northeast Sector of
the United States. Atmos. Environ., 16(7):1763-1774.
-204-
-------
REFERENCES (Continued)
Tong, E. Y., M. T. Mills, and B. L. Niemann. 1979. Characterization of
Regional Episodes of Particulate Sulfates and Ozone Over the Eastern United
States and Canada. Presented at the WMO Symposium on the Long Range Transport
of Pollutants and Its Relation to General Circulation Including
Stratospheric/Tropospheric Exchange Processes.
Tong, E. Y., and R. B. Batchelder. 1978. Aerometric Data Compilation and
Analysis for Regional Sulfate Modeling. Teknekron, Inc., Berkeley,
California, 135 pages.
Tijonis, J. and K. Yuan. 1978. Visibility in the Northeast United States,
U.S. Environmental Protection Agency, (EPA-600/3-78-075).
Venkatram, A. 1983. Uncertainty in predictions from air quality models.
Boundary-Layer Meteor., 27:185-196.
Wilson, J. W., V. A. Mohnen, and J. Kadlecek. 1982. Wet Deposition
Variability as Observed by MAP3S. Atmospheric Environment, 16:1657-1666.
Yerg, M. C. 1973. "An Optimal Sampling and Analysis Methodology."
University of Oklahoma, Department of Meteorology Report.
-205-
-------