Calculation of Trends from 1900 through 1990 for Sulfur and NO -N Deposition,
Concentrations of Sulfate and Nitrate in Precipitation, and Atmospheric Concentrations of
SOX and NOX Species over the Southern Appalachians
Jack D. Shannon
Environmental Research Division
Argonne National Laboratory
Argonne, EL 60439
10 April 1998
Modeling results have now been provided via e-mail to Dr. Pat Brewer of SAMI, to Prof. Jack Cosby at the University of
Virginia, and to Prof. Talat Odman at Georgia Tech. The results sent to SAMI consist of twelve Excel spreadsheet files
of (1) S wet deposition, (2) S dry deposition, (3) S occult deposition (i.e., from cloud and fog droplet uptake), (4) sulfate
concentration in precipitation, (5) S02 average air concentration, (6) sulfate average air concentration, (7) NOX-N wet
deposition, (8) NOx-N dry deposition, (9) NOx-N occult deposition, (10) nitrate concentration in precipitation, (11)
average NOx (combined NO and NO, as NO,) air concentration, and (12) average NO3 (combined N03 - and HN03 as
N03) air concentration. All variables are provided at five-year intervals from 1900 through 1990 for a set, selected by
SAMI, of 33 receptor locations in and near the SAMI region (see Table 1), and include both annual and seasonal totals or
averages. The Excel files for deposition and concentration in precipitation were also sent to Prof. Cosby. One dozen
corresponding sets of the same variables for a 13 by 13 subset of the grid being used in SAME modeling studies at
Georgia Tech were sent to Prof. Odman. The nested grid is shown in Figure 1. The NW corner is cell #400 of the Georgia
Tech arid and the SE comer is cell #832.
Method of calculation
A model of regional transport and deposition was exercised with SOx and NOx emission inventories defined at five-vear
intervals from 1900 through 1990. The model calculations used meteorological conditions for an I I-yr period, 1980
through 1990. Because meteorology is not specific to the year of simulation, results are more representative of a moving
average centered about the year of interest than they are of the specific deposition or concentration for that year. because
of climatological variability of regional transport for individual years.
Regional transport model (ASTRAP)
The modeling results have been calculated with the Advanced Statistical Trajectory Regional Air Pollution (ASTRAP)
model (Shannon, 1981, 1985, 1996). ASTRAP has been developed and improved to serve as a tool for modeling regional
air pollution for direct application in policy analysis and assessment; it is not a diagnostic model designed to investigate
air pollution processes in the atmosphere. Thus, ASTRAP is highly parameterized, but the parameterization rates are
based on relevant field studies of the key processes such as dry deposition or on the results of more detailed regional
models, and comparison with observations has been extensive.
One submodel of ASTRAP is a one-dimensional numerical model of the evolution of the vertical profile of pollutant
concentration associated with an initial unit emission in each of six model layers to 800 in in turn. Mixing height, vertical
eddy diffusivities (stability profiles), dry deposition velocities, rates of chemical transformation (constrained to be linear),
and rate of loss from the mixed layer to the free troposphere above are specified as functions of season and time of day,
but not as functions of the particular day or location, other than an east-west difference defined roughly along the 100th
meridian. Atmospheric behavior in the vertical is thus generic rather than for a particular date and location. Seasonal
statistics of the average concentration profile, dry deposition increment, and airborne mass, all for an initial unit emission
and defined as functions of initial emission layer and time since release (out to seven days after emission) are calculated
in the numerical modeling. To avoid calculation biases associated with the time of release (a tall stack emission during
the night would be completely decoupled from the surface for several hours by the nocturnal inversion, for instance),
statistics are averaged for simulated emission releases throughout the diurnal cycle.
A second submodel of ASTRAP calculates horizontal trajectories for a grid of virtual sources covering North America.
The calculations use wind fields at 1000 mb and 850 mb (roughly about 100 m and 1500 m elevation), gridded with
roughly 330 km spacing in the S AMI area and updated four times daily and covering North America and adjacent waters,
and precipitation fields gridded with roughly 110 km spacing and updated twice daily. (Horizontal spacing is approximate
because the meteorological analyses are for polar stereographic map projections whose horizontal spacing varies with
latitude.) Virtual source horizontal spacing is the same as that of the wind field. The meteorological data sets cover the
period 1980-1990 and were provided by Atmospheric Environment Service of Canada. The precipitation fields are
necessary for the trajectory calculations because the initial unit emission mass assumed in the calculation is depleted by
wet removal as a function of the half-power of the precipitation total in the corresponding grid cell during that time step.
For each virtual source, a seasonal ensemble trajectory is produced from individual simulated trajectories initiated at 6-h
intervals by fitting bivariant normal distributions (puffs) to the ensemble of trajectory end points at plume-age (time since
release) intervals of 6 h out to seven days travel time. That it, a puff is fitted to the distribution of all trajectory locations
six hours after release, another puff is fitted to the trajectory locations 12 hours after release, etc. Fitting of the
distribution functions take into account wet removal, i.e., the endpoint of a trajectory that has not encountered wet
removal is given greater weight (because it has more remaining airborne mass) than the endpoint of a trajectory whose
simulated pollutant mass has experienced depletion by precipitation. Superposition of the puffs, which overlap and
expand with time, forms a mean horizontal plume for that particular virtual source. The mean horizontal plume is not
constrained to be a straight line. Individual sets of trajectory statistics are produced for each season of the 11 -yr period,
and the 11 statistical realizations are then averaged for each of the four seasons. Similar sets of statistics are produced for
the occurrences of wet deposition as a function of plume age. Typically the horizontal plumes and the deposition
"plumes" have different orientations, because precipitation is more frequent for some wind directions than for others.
Ultimately, that can be a significant factor in causing source apportionment for wet deposition at a site to be quite
different from the source apportionment for dry deposition.
A third submodel of ASTRAP combines the statistics from the first two submodels with an emission field to estimate
concentrations and deposition at a specified set of receptor locations (a receptor grid is simply a regularly spaced set of
receptor points). The emission field for most applications is gridded horizontally with a spacing of about 115 km and
vertically in six layers to 800 m. For a cell of the emission grid, two-dimensional horizontal dispersion and wet removal
statistics, linearly interpolated from the statistics for the four surrounding trajectory virtual source locations, are
combined with statistics for the appropriate emission layer from the one-dimensional integration and multiplied by the
emission rate to produce three-dimensional concentration functions. (The deposition functions would also become
three-dimensional but are only one level deep, thus effectively two-dimensional.) The values at a receptor point of the
three-dimensional concentration or deposition functions for each 6-h time step out to seven days are additive, as are the
contributions at the receptor point from other emission layers and other emission cells.
Emission trend data
a. Emission data
Emission data provided through SAMI were the annual totals of S02 and NOX by state at five-year intervals from 1900
through 1990 for five source categories: Utility, Industrial, Transportation, Residential/Commercial, and Other.
Inspection of the data indicated that smelter emissions, at least for western states, must have been included in Other
rather than in Industrial. Inspection also indicated that railroad emissions of SOX, which would have been particularly
significant early in the period before diesel engines were introduced, must have been included in Other rather than in
Transportation. No trend information for Canadian or Mexican emissions was provided. Because ASTRAP, the transport
model applied, assumes linear chemistry and removal rates proportional to concentration or airborne mass, errors in
emissions produce a proportional error in modeled concentrations and deposition associated with the emissions of
concern. However, if the emission errors are not uniform spatially, then the effect of those errors on modeled
concentrations and deposition at a point becomes more complex.
The emission data provided lacked seasonal, vertical, and within-state horizontal resolution, so it was necessary to make
assumptions before applying the emission fields within the ASTRAP model of regional air pollution. A set of seasonal
SOX and NOx emission inventories for 1985, previously gridded vertically in six layers to 800 m. effective stack height
and horizontally within states and provinces, was used as the basic emission structure to be modified by the historic
emission inventory. The key assumptions in this study involved the following:
(1) Seasonal emission variations: It was assumed that a much greater fraction of early emissions were associated with
space heating, primarily combustion of coal, than is currently the case, and that the pattern gradually shifted to the current
(1985) aggregate conditions, i.e., roughly 25. 24, 27, and 24% winter, spring, summer, and autumn, respectively, for SOX
emissions. A seasonal distribution function was subjectively estimated (Figure 2). It is believed to be qualitatively
correct, in that it reflects an initial predominance of winter emissions from heating and a gradual shift to a summer peak
because of the growing importance of power -eneration for air conditioning Quantitatively, however, the seasonal
emission trends remain highly uncertain. The main effects on annual model results of changing from a winter emission
maximum to a summer emission maximum (for a given annual emission rate) are increases in wet deposition of sulfate
(because the effectiveness of wet removal processes for SOX is less in snow and cold conditions) and in increasing
average air concentrations of sulfate while decreasing air concentrations of SO, (because oxidation of SO, is faster in
warm, humid conditions). The seasonal emission pattern was also used for NOX emissions. The effects on annual model
results of the seasonal NO, emission pattern is minor for wet deposition of nitrate, as the wet removal processes are less
temperature dependent. The effect of the seasonal emission parameterization on annual dry deposition of NOx-N also
appears to be relatively minor, as the parameterization of the effectiveness of NOX transformation to N03_/HNO3
increases in the summer but the proportion assumed to be HN03 (large dry deposition velocity) decreases. Because of the
linear nature of ASTRAP and the manner in which the seasonal emission parameterization was applied, users of model
results could scale modeled seasonal results for a particular year by the ratio of a different assumed seasonal emission
factor to that shown in Figure 2 to get modified concentration or deposition results; the seasons could be re-aggregated to
produce new annual totals.
(2) Vertical emission variations: It was necessary to make assumptions for the vertical distribution of effective emission
heights as well. The general trend is thought to be known a significant fraction of surface emisions from combustion for
heating and relatively low industrial and electrical utility stacks initially, with a gradual shift to taller stacks for utility
and industrial sources because of shifts to gas or electrical heating and actions by point sources to reduce local
concentrations. The results of the emission-height allocation algorithm, nationally aggregated, are shown in Figures 3a
and 3b. The results for SO and NOX differ because of the differing importance of the various emission sectors. Results
for individual states vary about the nationally aggregated pattern because the contributions from the five source
categories were not constant among states. Generally similar patterns were assumed for emiss ons in Canada and Mexico.
The estimation of the trend in effective emission heights for these simulations is recognized to be arbitrary. The effect of
an increase in effective emission height is to reduce local dry deposition (because local surface concentrations are lower)
and to increase the magnitude and transport scale of wet deposition (because less pollutant mass is removed locally).
Local air concentrations are decreased, but regional concentrations, particularly of the secondary species, are increased.
(3) Within-state horizontal variations: It was assumed that the within-state emissions, for each of the six-emission layers,
was always that associated with the -ridded 1985 emissions. Because the vertical distribution of emissions for earlier
years, as parameterized for this study, differed from that in 1985, the within-state horizontal distribution of emissions for
all layers combined during other years differed from that for 1985 because the relative importance of emissions within
each layer changed with time.
The emission trend information provided by SAMI was for the contiguous United States. Trends for emissions in Canada
and northern Mexico also had to be estimated. In the absence of province-specific emission trends for years prior to 1980,
it was assumed that the SOX emissions of each province varied according to the ratio of the aggregated total Canadian
emissions for the -'ear of interest to that for 1980, as plotted in Shannon (1990). Canadian emissions of NO., and
emissions of both SOX and NOX in northern Mexico were assumed to vary from 1985 conditions in the same manner
that aggregated emissions in the United States varied. While more detail in the emission trends for Canada and northern
Mexico is clearly desirable. the effects of these crude assumptions on calculations of concentration and deposition in the
southern Appalachians is believed to be minor, compared to overall uncertainty in the study, because of the distance of
the source regions from the southern Appalachians.
Uncertainties in simulation results
All air pollution modeling studies contain uncertainty. The uncertainty arises from two general causes: limitations in the
underlying assumptions, formulation, and structure of the atmospheric model (here ASTRAP), and limitations in the data
input to the model for simulations.
Limitations in ASTRAP
ASTRAP is devised to be exercised repeatedly and efficiently for policy analysis and assessment, and thus the model
treats physical and chemical processes in a highly parameterized fashion. Although a submodel of ASTRAP calculates
trajectories at intervals of six hours, it is the seasonal ensemble statistics of those trajectories that are used in
concentration and deposition calculations rather than individual trajectories; i.e., ASTRAP calculates seasonal
concentration and deposition patterns directly rather than calculating short-term results and adding and averaging them.
Diagnostic models, on the other hand, make calculations of concentration and deposition for daily or shorter times and
produce seasonal results by averaging or weighting short-term patterns. In this simulation the trajectory and wet removal
statistics are for an I I -yr period. Thus, the trajectory results should be representative of the expected climatology of
regional transport and deposition. However, during individual years and seasons, the frequency and intensity of patterns
of wind and precipitation exhibit climatological variability about the expected mean conditions or expected frequencies
of meteorological patterns. The modeling results may be more representative of a running mean sampled at five-year
intervals than of individual years sampled at five-year intervals, because of the lack of climatological variability in the
exercise. Even if one assumes that parameterizations in ASTRAP are representative and unbiased and that the frequency
of meteorological patterns is close to what one would expect from climatology, ASTRAP results should still be
smoother than analyses of observed fields because of the departure of the distribution of actual trajectory endpoints from
the bivariant normal distribution assumed in the Generation of ensemble statistics, and because of the spatial variation of
processes in the vertical such as mixing and linear chemical transformation that are not included in ASTRAP.
The parameterizations in ASTRAP, such as for the rate of transformation of SO,, to sulfate, are based upon
generalizations of the results of field studies and the results of more detailed models, both under emission conditions
typical of the 1970s and 1980s. During the early years of the period examined here, the parameterizations are being
applied in a regime of emissions lower by a factor or two or more and thus would have a greater uncertainty then,
although the fact that the parameterizations appear to work relatively well for both the cleaner and the more polluted
portions of the current emission field increases confidence in their applicability for the earlier emission fields. The use in
ASTRAP of spatially constant parameterizations of dry deposition velocities and mixed-layer mixing is bound to increase
uncertainty in mountainous terrain such as the SAMI area. A receptor higher than the surrounding region is sometimes
above the regional mixed layer and thus experiences lower surface air concentrations and (in the absence of surface
differences other than elevation) resulting dry deposition than the surrounding region. On the other hand, direct impaction
of clouds on vegetation and terrain at elevated sites may lead to uptake of pollutants in droplets becoming a significant or
even dominant mode of deposition. For these reasons, this application of ASTRAP has used a parameterization of
local dry deposition in which the dry deposition that would be modeled to occur at that horizontal location on flat terrain
is scaled by the ratio of the modeled air concentration at the relative elevation of the receptor (height above the
approximate minimum terrain in the substate region) to the modeled concentration at the modeled surface layer.
The wind fields used in ASTRAP trajectory calculations are large-scale (horizontal spacing about 345 km). Only
large-scale elevation features affect the input wind fields, and that effect is largely implicit. The wind fields would not
capture such local detail as flows up or down valleys. They are also partly artificial, in that where the ground surface is
above the 1000-mb level (or above the 850-mb level in the west) the wind field is calculated from the pressure field, with
certain temperature assumptions.
Shannon J.D. (198 1) A model of regional long-term average sulfur atmospheric pollution, surface removal, and net
horizontal flux. Atmospheric Environment 15, 689-701.
Shannon J.D. (1985) User's Guide for the Advanced Statistical Trajectory Regional Air Pollution (ASTRAP) Model. U. S.
Environmental Protection Agency Report EPA/600/8-85/016 (NTIS PB85-236784/XAB).
Shannon J.D. (1990) Modeled trends in sulfur deposition since 1900 in North America. In: Air Pollution Modeling and
Its Applications VIII, (H. van Dop and D. G. Steyn, eds.), Plenum Press, New York and London, 61-68.
Shannon J.D. (1996) Atmospheric Pathways Module Documentation: Tracking and Analysis Framework, Lurnina
Decision Systems, Inc., Los Altos, Calif.
Shannon J.D. (1997) Scales of Sulfur Concentrations and Deposition from the Perspective of the Receptor, Atmospheric
Environment 31, 3933-3939.
Shannon J.D. and Trexler E.C. Jr. (1995) Climatological Variability in Regional Air Pollution, pp. 5 11-5 14 in
Proceedings, 6th International Meeting on Statistical Climatology, 19-23 June 1995, Galway, Ireland.
Table 1: Receptor points for concentration and deposition trend calculations (latitude, longitude, altitude (m), and name)
35.40 83.99 727 Joyce Kilmer/Slickrock Wilderness NC
35.37 82.87 922 7 Shining Rock Wilderness NC
35.93 81.91 938 Linville Gorge Wilderness NC
34.37 87.50 235 Sipsey Wilderness AL
37.61 79.43 587 James Riverface Wilderness VA
34.97 84.62 550 Cohutta Wilderness GA
39.08 79.39 750 Dolly Sods Wilderness WV
32.46 87.25 58 Black Belt Substation AL
34.29 85.97 349 Sand Mountain AL
33.18 84.41 270 Georgia Station GA
37.70 85.05 353 Mackville KY
37.08 82.99 335 Lilley Cornett Woods KY
38.12 83.55 204 Clark State Fish Hatchery KY
39.41 77.00 172 White Rock MD
38.91 76.17 6 Wye MD
35.06 83.43 686 Coweeta NC
35.70 80.62 219 Piedmont Research Station NC
35.74 82.29 1987 Mt. Mitchell NC
33.55 80.43 24 Santee National Wildlife
35.96 84.29 341 Walker Branch TN
35.66 83.59 640 Great Smoky Mts National Park - Elkmont TN
38.04 78.54 174 Charlottesville VA
37.33 80.57 963 Horton's Station VA
38.52 78.44 1074 Shenandoah National Park - Big Meadows VA
37.98 80.95 753 Babcock State Park WV
39.09 79.66 505 Parsons WV
37.17 78.31 146 Prince Edward VA
36.47 83.83 361 Speedwell TN
39.03 76.82 46 Beltsville MD
35.26 79.84 198 Candor NC
36.11 82.04 1219 Cranberry NC
36.04 85.73 302 Edgar Evins State Park TN
37.92 83.06 455 Crockett KY
02/18/98 WED 14:52 FAX 404 894 2265
Fi , I -
GA TECH ENV ENG
N ested r
16 ;I it p Jo J) J7 j] J, Ja ,I 47 .0 J4 v P P ju ill ill .1 J9 ;5 is p p jq jo ji
p J4 0 ji id JI j3 X 0 J% A 41 ;3 45
.1 Ja .11 0 A
1 x A ;Qj
A A? jet im IV ill JL
J12 Ju JU! 1 X ill 419 m ju J25 im J71 ::I ja 9 J31 in .23 i3l J.'41 J37 ~11 I
J41 J42 J43 jtM JW 46 JV j4 ;-'? J- .41 In._ M J,4 J90 -js ~A A J9 IN X 02 .10 JA J65 J66 .67 ('.v - J70 in
in J79 J30 J16 J82 Ill ISO 41
~17 ;11 ;0
;m IV 19 JU A X JW A JO ;Dl mj
ni 0~- v is
-x X JU J15 1
J-1 N4 0 04 O 'm
;M xe .31 N7 In ,o -,v ~--l 0 in M JM, JIn p .0 J14 M illJul
x m v
4 e 2 'A jo p
J~ jul ;13 p
.115 X-1 J14
J11 JfS J11 i Al M J23 rq XJ; ia xi i)o I An Jz~ X 44 J4~ 6 YI
X11 9 in ju jS A M7 JSJ J.'? X ;4l IQ. V /x im x M ju J7 1-4: in J 0A! I A x
4 ;63 ! 3 W
it, iis Ir JL4 ji617------
1 4:11i J19
Al in in 'm 'o im in ill J3 a x, fT in P A A 43 in 659 A .01 A A JIM A A ;AA 0 J30 J51 A J13 J.4 p
is .45 11 J59 .91 A is A X A 361 A: JIM in
JU A .91 A JM ix im A .0 p jn x p 0 A IV in in A
7 % 1 1
49, A .41 JA j% jql 0 .10 JM .1G .9 'W ill J12 4D J14 ill J)6 ill J19 p P A A Aop
J3 in M JN A P M in p P PA P p A N .41 x A x J;4 j" p Al IJ JS 54 Jb j% J57 _q JS X
x x v P x % x in A, im -Ir -in V`J ju J3 A P in A A A it jo 31 ill in \ ju P in 4% 01m A J% x
J% p SO 91 A 011 $1 A A X .14 ~10 In 31 JU 416 J17 Jill ill p 0 o in A 437a
ol jQ1 03 kM & h36 07 ill AD kv .0: Y~ .03 N .00 A x in AD 0 JIS A% ~17 IQ At Al .7,; KA X
I i , in
06 jwT A p 11 pn in p op .0 477 0, h7q w a x A 0 Fx in i A A im x px
;at x am p J07 p p wIl ji: jn 113 v ;15 ;16 X ill OM 74 A in dm P. X d-4 P Im .31 j-" Pi dNv
im M YA p x J41 ;e 'U X ~b J67 ;A 14 JA' ;51 12 453 JN JO J36 J37 A -79 .74 Jill X1 X ;41 V '% X Iu
It D Xv A
in in J73 P in P I'm J11 44 X P ill 4T in m in 419 A i9c JIS A T in A
06 X in Iff 10 ill in JU 114 J15 p JJ7 p I Ai- im .0 m JU x is fF Jul P J30 in in IN 315 J36 J37 A J39 A
A ill - jig
.ul in P3 J" N I* J41 J4 X1 X! In x At \x -jr, J% Ju jo x w jV jM In 172 J71 Vd M
1 A S A
P P 117 01 p 91, 04 a X6 '-m- im ill P 03 ju I% Aq A JR A A PC Jim 04 x .0 in -IN p NO
A A on P JU /q6 J17 511 JN j:1 J: In n TA 0 A x At IT) in rA 0 .0 .01 in J3 is xl p- so A 3,11
9 N x JS JM J53 p J" ft Wj JA .25 p JV jV JO p A J67 :0 M rl :U J13 M M .0 P JM P A
`~w jew ixi jo im m JO Pic jail jot: jot: jail '41!
~4 Jill jot jai joll Q; jr, i Al in .0 J2 x Q] in jot jm .1 Ju P A J A J
11 jon l jM1 ja~ ioc1 X , j,- 1 l C w A J(S IN
J~ N J07 IN: J%
9 jx I
Parameterization of SO
seasonal emission pattern
1900 1930 1960 1990
Figure 2: Parameterization of seasonal emission patterns used in trend calculations.
. * ----------- ---------------- --------------------- - .........
V --- Ik --- 0 41 40 --- 9-Ak --- 0 --- i
--- I* --- ............ ...........
C0___ e --- --- 0 ...
A .... A ------ A ----- I -------------- ............ ............
-- ------- Winter ... ......... ......... ..........
...... ...... ..
.0 Spring-Autumn ............ ............ ............
... ... ....
. ... ............ ............ ...
7 ... ....
Parameterization of trend in SO emission level x 100
1900 1930 1960 1990
Figure 3a: Parameterization of nationally aggregated effective emission levels for SO
C) lnt~ 1-17
I ................. ............ ............
............... ... -
- ----- Mid-level
. ............ .........
\ ......... ..
. ................ .... ............
................................ ......... ......... ....
. ....... . ................
............... .......... .... ... ............
Parameterization of trend in
NO emission level
-----NOx mid-level ......
........ .............................. .......
Figure 3b: Parameterization of nationally aggregated effective emission levels for NO
I 4MCD C., X,
Historical Sulfur Deposition to
Great Smoky Mountain National Park
S Deposition 1900 to 1990, normalized to 1990 values
1.60 S kg/ha
1.00.- 36 Noland
0.00 i i i i i
1900 1910 1920 19301940
1950 1960 1970 19801990
- 16 Elkrnt
x MAGIC ND - EPA Region IV -MAGIC ND - EPA Region V - - - - - ASTRAP: Elkmont
1991-1995 Noland Divide: 36 S kg/ha 1990 Elkmont: 16 S kg/ha
Historical Sulfur Deposition to
Shenandoah National Park
S Deposition 1900 to 1990, normalized to 1990 values
1900 1910 1920 19301940
*0. . . . ..
1950 1960 1970 1980 1990
- - - - - ASTRAP: Big Meadows
MAGIC WOR/NFDR - EPA Region III
1990 BigMdw: 20 S kg/ha
1991-95 ave: 13(WOR) or 15 (NFDR) S kg/ha