EPA/600/A-96/074
Can Selected RADM Simulations Be Aggregated To Estimate
Annual Concentrations Of Fine Particulate Matter?
Brian K. Eder* and Sharon K. LeDuc*
Atmospheric Sciences Modeling Division
Air Resources Laboratory
National Oceanic and Atmospheric Administration
Research Triangle Park, NC 27711
ABSTRACT
Ambient air concentrations of fine particulate matter are an issue of increasing concern for the
U.S. Environmental Protection Agency. Accordingly, the Clean Air Act and the Amendments of
1990, call for an assessment of past and future regulations to protect both health and visibility.
Unfortunately, our most reliable tools for assessing long-term air quality change, Eulerian models,
challenge the practical limits of current computer resources and require extensive input data. To
reduce the resource requirement, an aggregation method, initially developed for RADM (Regional
Acid Deposition Model) acid-deposition applications, is currently being applied to a limited number
(thirty) of RADM simulations in order to provide estimates of long-term (annual) ambient air
concentrations of fine particulate matter. This paper briefly examines this aggregation technique, its
application to fine particulate matter, and the suitability of the original thirty RADM simulations.
INTRODUCTION
Ambient air concentrations of fine particulate matter (diameter < 2.5 p) are an issue of
increasing concern for the U.S. Environmental Protection Agency. Recent epidemiological studies
link an increase in mortality and other detrimental health effects, especially to the young, elderly and
others with respiratory problems, to fine particulate matter1. Fine particulate matter also contributes
to the deterioration of visibility, especially in the eastern United States. The Clean Air Act and the
Amendments of 1990 call for assessment of past and future regulations to protect both health and
visibility. Such an assessment will not only require estimates of changes in air quality attributable to
regulatory policy (as opposed to those attributable to changes in meteorology), but estimates of the
potential effects to health and visibility as well.
The most reliable tool for estimating air quality change for large regions, both past and future,
are regional air quality models such as the Regional Acid Deposition Model (RADM)2. This model,
and to an even greater degree, future models require massive resources, both human and computer,
for each policy and/or meteorological scenario. The benefits analyses proposed for the Clean Air
Act Amendments of 1990 require annual timescales. Unfortunately, most Eulerian models, like
RADM, challenge the practical limits of current computer resources as well as our ability to collect
the pertinent input data on annual scales. As a result, application of such models to determine the
long-term relationship between changing emissions patterns and ambient air concentrations is limited.
To circumvent this problem, results from an aggregation method, initially developed for acid-
deposition applications3,4 are currently being applied to a limited number (thirty) of RADM
simulations in order to provide estimates of long-term (annual) ambient air concentrations of fine
particulate matter. The aggregation method is based on the premise that at any given location,
* On assignment to the National Exposure Research Laboratory, U.S. Environmental Protection
Agency, Research Triangle Park, NC 27711

-------
ambient air concentrations of fine particulate matter are governed by a finite number of different,
though recurring meteorological regimes. If a collection of concentration patterns representative of
these different meteorological regimes can be identified, they can be aggregated, using appropriate
weights, to produce reasonable estimates of annual averages.
The purpose of this study is to determine whether or not the thirty original RADM
simulations selected for aggregation in the acid deposition applications are equally representative for
ambient air concentrations of fine particulate matter. Unfortunately, there exists a dearth of fine
particulate data, therefore this analysis will employ, as a surrogate, an extinction coefficient (bext)
estimated from midday human observations of visual range at airports5.
DATA
Meteorological
The cluster analysis, described in the methodology section, utilized eastern North American
zonal u and meridional v 850 mb wind components (ms"1) for 0000 UTC. Winds at 850 mb were
selected because of their proximity to the boundary layer, where a majority of pollutant transport
occurs. The data, which had a 5° latitude by 5° longitude resolution were extracted from the NMC
global analyses for the period 1979-1990. The NMC data were selected in part because they
provide information over the Atlantic Ocean with the same resolution as is provided over the
continent. To allow compatibility with RADM simulations, which generally have been on the order
of three days, the NMC data were rearranged into overlapping three day records (i.e record 1 - days
1, 2, 3; record 2 - days 2, 3, 4, etc.)
Extinction Coefficient
The aggregation results were applied to daily extinction coefficients [bext (km"1)] obtained
from Washington University's CAPITA (Center for Air Pollution Impact and Trend Analysis) located
in St. Louis, MO. The data were observed at 64 locations (Figure 1) through the eastern two-thirds
of the U.S for the period 1979-1990. The light-extinction coefficient is often used to characterize
visibility, although, in general, it has limited ability to predict human visibility. The visual range can
be estimated from the b,„. by using the Koschmieder equation:
visual range (km) = 3.91/bext	(1)
with the assumptions that a black target is viewed against the horizon in daylight, and that the
atmosphere and the illumination over the sight path are uniform. Only days that were free of
precipitation, with relative humidity less than 90% were utilized in this study. The bext was also
corrected for relative humidity.
METHODOLOGY
Clustering
The purpose of objectively defining meteorological categories is to identify recurring
atmospheric transport patterns associated with varying concentration patterns of fine particulate
matter. Identification of these patterns was necessary to facilitate selection of time periods for
simulation by RADM and in the development of the aggregation technique.
Utilizing Ward's method of cluster analysis in an agglomerative, hierarchical mode, each of
the 3-day sequences of 850 mb wind flow patterns from 1979 to 1990 were assigned into one of 19
statistically significant meteorological categories3. The analysis consisted of 288 data values
recorded at 0000 UTC (i.e. 48 grid points x 2 variables (« and v) x 3 day overlapping records) for
each of the 4383 days in the study period. For further information concerning this procedure, consult
the original article3.

-------
Aggregation
The aggregation procedure estimates mean annual extinction coefficients using a set of thirty
simulation periods selected from the 19 meteorological categories as described in the original article4.
The number of simulation periods exceeded the number of meteorological categories because in the
original study, the categories were further divided into "wet" and "dry" portions. The resulting 38
categories were reduced to 30 to keep the number of RADM simulations to a manageable number.
Calculation of the mean annual extinction coefficients makes use of weighting/scaling factors that are
based on the frequency of occurrence and the expected extinction coefficient for each of the
categories associated with the events selected for aggregation.
RESULTS
Clustering
Found in Figures 2(a) and 2(b) are two of the 19 meteorological transport patterns associated
with Clusters 9 and 12. The length of the vectors is proportional to the wind speed at 850 mb, with
the arrows pointing in the direction of the wind flow. Each vector is centered at the location where
the wind data were used in the cluster analysis.
Cluster 9, which was the most frequently observed flow pattern, is depicted in Figure 2(a).
This 850 mb flow pattern is dominated by a stationary anticyclone situated over the southeastern
United States. This pattern occurs when the Bermuda High, normally centered over Bermuda,
retrogrades and stalls, resulting in subsidence and a weak anticyclonic circulation over most of the
entire domain. Winds tend to be very light over most areas east of the Mississippi River and south
of the Great Lakes. Examination of extinction coefficients associated with this cluster (Figure 3(a)
indicate large values (generally greater than 0.16 km"1) over much of the eastern U.S. These large
coefficients, which again are indicative of low visibilities and thereby hypothesized high fine
particulate matter concentrations, are common with such patterns where stagnation, subsidence and
little or no precipitation occur6.
Cluster 12, which occurred less frequently than Cluster 9, is shown in Figure 2(b) and depicts
a well-developed low pressure system centered over New England. This 850 mb flow regime results
in a strong cyclonic circulation pattern. Transport is strong and from the northwest over most of the
domain. Note the vector's lengths as compared to those associated with Cluster 9. This strong
northwesterly flow allows migratory anticyclones that originate in relatively "clean" areas of western
Canada to traverse the eastern United States in areas north of 35° latitude7. As a result, extinction
coefficients associated with this cluster (Figure 3(b)) are much smaller, with most of the domain
reported mean extinction coefficients less than 0.12 km"1). The main exception is the Gulf Coast
area where the flow is not generally influenced by the migratory anticyclone. Here the extinction
values reach 0.17 km'1.
The mean cluster extinction coefficients associated with the other 17 clusters, each defining a
common flow pattern, lie within the range established by Clusters 9 and 12 (Figure 4). Cluster
descriptions can be found in the initial article3.
Aggregation
Results comparing the observed mean annual extinction coefficients with the aggregated
estimates of the mean annual extinction coefficient are very promising as seen in Figure 5 and 6.
Aggregated mean annual extinction coefficients at roughly two-thirds of the stations (41 of 64) were
within 5% of the actual mean annual values. The correlation between the two data sets was very
high, r2 = 0.963. The intercept was nearly zero (-0.005) and the slope nearly one (1.029), indicating
little or no systematic bias across the 64 sites. There does, however, appear to be some spatial bias
as seen in Figure 6; with a general underestimation found in the northeastern quadrant of the domain,
especially in New England, where underpredictions reach 15%. Reasons for this spatial bias are not
understood at this time.

-------
CONCLUSIONS
The thirty RADM simulation periods, originally selected for acid-deposition applications,
appear to be very representative from an extinction coefficient (inferred fine particulate matter)
perspective. Whereas acid-deposition aggregate values were within 20% of the observed values at
only 13 of the 20 sites used in the original study, roughly two-thirds (41/64) of the aggregate
extinction coefficients were within 5 % of the mean observed coefficients in the current study, and
all were within 15%. The correlation between the observed and aggregate coefficients was very
high. With the exception of the spatial bias discussed above, little systematic bias was found in the
results. It should be noted that some of the increase in the representation of these thirty simulation
periods for particulate concentrations (as opposed to acid wet deposition) can be attributed to the
removal of uncertainty inherently associated with precipitation. Results of this analysis suggest that
the original thirty RADM simulations are indeed sufficient enough to derive annual estimates of fine
particulate matter.
DISCLAIMERS
The information in this document has been funded by the United States Environmental
Protection Agency. It has been subjected to Agency review and approved for publication. Mention
of trade names or commercial products does not constitute endorsement or recommendation for use.
REFERENCES
1.	Reichhardt, T. Environ. Sci. Technol. 1995 29, 360A-364A.
2.	Chang, J.S.; Brost, R.A.; Isaksen, I.S.A.; Madronich, S.; Middleton, P.; Stockell, W.R.
and Walcek, C.J. J. Geophvs. Res. 1987 92, 14,681-14,700.
3.	Brook, J.R.; Samson, P.J.; and Siliman, S. J. AppI. Meteor. 1994 34, 297-325.
4.	Brook, J.R.; Samson, P.J.; and Siliman, S. J. Appl. Meteor. 1994 34, 326-339.
5.	Husar, R.B. and Wilson, W.E. Environ. Sci. Technol. 1993 27, 13-16.
6.	Eder, B.K.; Davis, J.M. and Bloomfield, P. J. AppI. Meteor. 1994 33, 1182-1199.
7.	Ziska, K.M. and Smith P.J. Mon. Wea. Rev. 1980 108, 387-401.
Figure 1. Station location map for bext coefficient measurements.

-------
EASTERN NORTH AMEJKAM WEAK MOUS WK> fJDJ) - n ausrois
THAXCTOKt auSTER •
DAY 2
»' ¦ >
»
- >
r?
ffO »S M 73
E3
400
•5
Figure 2(a). Eastern North American 850 rab wind field associated with Cluster 9.
EASTERN NORTH AUEMCAN USAM (SOUS KN> fEUJ - » R.USTC«S
TRAJECTORY CLUSTER tj	t a. N- IS
DAY 2
—5 -55 10 Is M rs'	ro	*S
Figure 2(b). Eastern North American 850 mb wind field associated with Cluster 12,

-------
Figure 3(a). Mean b„t (km1) associated with Cluster 9.
.07'
.08
.12
.11
.09
.12-
.11
.07
.09
.07
.14
.09
Figure 3(b). Mean b„t (km1) associated with Cluster 12.

-------
0
¦*—>
if)
o
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
0
0.05
0.1
0.15
'ext
(km*1)
Figure 4. Mean network b„, (km1) associated with the 19 meteorological clusters.
0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
Observed Mean Annual
Figure 5. Comparison of the observed and aggregate mean annual be„ (km1) for the period
1979-1990.

-------
-2.9
Figure 6. Percent deviation in aggregate estimates of the mean annual beja. Deviations are relative
to the observed annual means (i.e. aggregate-observed/observed).

-------
-- ^ ' TECHNICAL REPORT DATA
1. REPORT NO.
EPA/600/A-96/074
2 .
3 . recip:
4. TITLE AND SUBTITLE
Can Selected RADM Simulations be Aggregated to
Estimate Annual Concentrations of Fine Particulate
Matter?
5.REPORT DATE
6.PERFORMING ORGANIZATION CODE
7. AUTHOR(s)
Brian K. Eder and Sharon K. LeDuc
8.PERFORMING ORGANIZATION REPORT NO.
9. PERFORMING ORGANIZATION NAME AND ADDRESS
Same as Block 12.
10.PROGRAM ELEMENT NO.
11. CONTRACT/GRANT NO.
12. SPONSORING AGENCY NAME AND ADDRESS
National Exposure Research Laboratory
Office of Research and Development
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
13.TYPE OF REPORT AND PERIOD COVERED
Proceedings, FY-96
14. SPONSORING AGENCY CODE
EPA/600/9
15. SUPPLEMENTARY NOTES
16. ABSTRACT
-'Ambient air concentrations of fine particulate matter are an issue of increasing
concern for the U.S. Environmental Protection Agency. Accordingly, the Clean Air
Act and the Amendments of 1990, call for an assessment of past and future
regulations to protect both health and visibility. Unfortunately, our most reliable
tools for assessing long-term air quality change, Eulerian models, challenge the
practical limits of current computer resources and require extensive input data. To
reduce the resource requirement, an aggregation method, initially developed for RADM
(Regional Acid Deposition Model) acid-deposition applications, is currently being
applied to a limited number (thirty) of RADM simulations in order to provide
estimates of long-term (annual) ambient air concentrations of fine particulate
matter. This paper briefly examines this aggregation technique, its application to
fine particulate matter, and the suitability of the original thirty RADM
simulations.
17. \ KEY WORDS AND DOCUMENT ANALYSIS
a. DESCRIPTORS
b.IDENTIFIERS/ OPEN ENDED
TERMS
c.COSATI



18. DISTRIBUTION STATEMENT
REkPASE TO PmiC

19. SECURITY CLASS (This
Report)
UNCLASSIFIED
21.NO. OF PAGES
8

20. SECURITY CLASS (This
Page)
UNCLASSIFIED
22. PRICE

-------