United States
Environmental Protection
Agency
Atmospheric Research and
Exposure Assessment Laboratory
Research Triangle Park NC 2771
Research and Development
EPA/600/S4-89/026 Sept. 1989
Project Summary
Re-Examination of Interim
Estimates of Annual Sulfur Dry
Deposition Across the Eastern
United States
Terry L. Clark, Robin L. Dennis, and Steven K. Seilkop
During the summer of 1987 annual
amounts of sulfur dry deposition
were first estimated for more than
7,000 lakes in the eastern United
States. These estimates, heretofore
termed interim estimates since they
were expected to be superceded in
the near future, were derived from
predictions of the Regional Acid
Deposition Model (RADM) adjusted
using the empirical data from two
monitoring networks. Since that time,
additional years of empirical data
have become available and a portion
of the previously available empirical
data has been superseded. Conseq-
uently, the process of estimating
annual amounts of sulfur dry deposi-
tion was repeated to determine
whether these interim estimates
should be revised, and if so, by how
much. This study concludes that the
interim estimates appeared to be too
low by 13% and recommends that the
interim estimates be systematically
increased by the same amount
A comparison of the revised esti-
mates to empirically-derived sulfur
dry deposition amounts suggests
that there is some systematic error in
the revised estimates. Adjusted
RADM predictions of dry deposition
tend to be biased low in the most
significant source regions (where at
least 200 ktonnes SO2/yr are emitted
within 80 km of the site). Conversely,
in locations farther removed from sig-
nificant sources (81-160 km) there is
evidence that the estimates are
biased high. However, in general,
sulfur dry deposition estimates from
adjusted model predictions are within
i 60% of the empirical data.
This Project Summary was devel-
oped by EPA's Atmospheric Research
and Exposure Assessment Laboratory,
Research Triangle Park, NC, to an-
nounce key findings of the research
protect that is fully documented in a
separate report of the same title (see
Project Report ordering information at
back).
Introduction
Dry deposition rates are a function of
the air concentration near the surface and
the dry deposition velocity. Since dry
deposition velocities cannot be measured
directly, they are inferred from vertical
mass flux measurements. In addition,
since they are a function of atmospheric
stability and surface attributes (e.g.,
vegetative type, roughness length, physi-
cal conditions, spatial fluctuations of
terrain and surface roughness), dry
deposition velocities can vary signif-
icantly across small areas (e.g., less than
1 km2). Because of the potential for small
spatial-scale variations in dry deposition
velocity, there is considerable uncertainty
in using a dry deposition estimate from
an individual site to represent an average
regional value.
In response to the needs of the Aquatic
Effects Program of the National Acid
Precipitation Assessment Program, inter-
im estimates of annual sulfur dry
deposition were derived in August 1987
for 7,000 lakes in the eastern United
-------
States. Since that time, several develop-
ments have occurred that necessitated a
re-examination of these interim estimates
of sulfur dry deposition. First, the
empirical data at the U.S. sites have been
revised as a consequence of improve-
ments to the dry deposition algorithm of
the inferential model used to derive the
empirical estimates. Secondly, in both
the United States and Ontario, data are
now available for additional years. The
expanded data base not only has pro-
vided us with a more statistically repre-
sentative sample, but also has enabled
network staff to identify outliers and
discard or correct erroneous values.
Thirdly, improvements to the dry depo-
sition module of RELMAP have greatly
reduced its oversmoothing problem,
thereby increasing this model's potential
as an appropriate estimator of spatial
patterns. Finally, the calculations from a
third model, the Advanced Statistical
Trajectory Regional Air Pollution
(ASTRAP) model, became available.
Procedure
Estimation of a spatial pattern of sulfur
dry deposition across the eastern United
States is strongly hampered by the fact
that there are only four U.S. sites at
which empirical estimates are available.
To circumvent this paucity of data, the
predictions of regional-scale deposition
models are used in conjunction with the
available empirical data. The following
procedure was developed for estimating
annual amounts of sulfur dry deposition
across the eastern United States:
1. Construct spatial patterns of annual
amounts of SO2 and sulfate dry
deposition from available regional
deposition models that relate emis-
sions, transport, dispersion and trans-
formation to dry deposition using dry
deposition velocities assumed to
represent the area of each model grid
cell,
2. Adjust model predictions by a con-
stant factor (based on the comparison
of model predictions to site-specific
empirical estimates) to correct for
model bias,
3. Select the spatial pattern produced
by the regional models with the
smallest mean-square error,
4. Estimate dry deposition amounts at
specific locations of interest by inter-
polating adjusted model predictions,
and
5. Assess the uncertainty of these esti-
mates by examining the correspond-
ence between the model predictions
and empirical estimates and charac-
terizing the spatial and interannual
variability of the empirical estimates.
The first step of this approach was
executed by constructing grids of annual
sulfur dry deposition from each of three
operational deposition models used by
EPA - RADM, RELMAP and ASTRAP.
As was the case described in the 1987
report, six three-day episodes of RADM
output were averaged and normalized to
construct one annual grid (Cases I, II and
IV of the April 1981 Oxidizing and
Scavenging Characteristics of April Rains
(OSCAR) Experiment, the four-dimen-
sional data assimilation run of OSCAR IV,
the August 1979 Northeast Regional
Oxidant Study (NEROS) case, and an
October 1984 case). Unlike the RADM
grid, those of RELMAP and ASTRAP
were constructed from simulations of the
entire year of 1980. The RELMAP results
presented here were derived from the
improved model version and differed
from those presented in the 1987 report.
The second step was accomplished by
first comparing interpolated model pre-
dictions with the annual means of the
empirical estimates at each of four U.S.
sites of the COre Research Establish-
ment (CORE) Network and 18 sites of the
Acidic Precipitation in Ontario Study
(APIOS) Network. Empirical estimates at
the CORE sites are expressed as the
product of the weekly-mean, inferred dry
deposition velocities and measured air
concentrations of S02 and sulfate for the
years 1985 to 1987, inclusive. In contrast,
Ontario empirical data are based on a
cruder method — the product of esti-
mated annual mean dry deposition veloc-
ities and annual means of measured air
concentrations of SO2 and sulfate for the
years 1982 to 1986, inclusive.
From the comparisons, mean-square
errors were calculated and used as a
measure of concurrence of interpolated
predictions and empirical estimates. To
minimize the bias of the annually-normal-
ized RADM and the ASTRAP predictions,
the ensembles of predictions were ad-
justed systematically by factors of 0.43
and 0.57, respectively. These factors are
the regression coefficients of a linear
regression of empirical estimates on
model predictions forced through the
origin. The RELMAP mean-square error
was already small and could not be sub-
stantially reduced by systematic adjust-
ments to the model predictions.
Based on the comparisons of ad-
justed/unadjusted predictions with the
empirical estimates, the adjusted RA(
predictions appeared to replicate best 1
characteristics exhibited by the empiri
data. That is, the bias, root-mean-squi
error, and average error for adjust
RADM predictions are lower than those
the other two models and the ovi
smoothing of the gradient aero
southern Ontario is less severe than 1
other two models. Each model tends
overpredict the sulfur dry deposition
the empirical data range of 3 to 6
S/ha/yr. This is indicative of the mode
slower rate of decreasing dry deposit!
away from the high emissions regions,
in other terms, their smoothing tender
near steep gradients.
Uncertainty Assessment
The uncertainty in sulfur dry deposit!
amounts obtained from the estimati
procedure described above is primal
related to three main factors: (1) t
accuracy with which the RADM captui
the underlying spatial pattern of c
deposition, (2) the accuracy of t
empirical dry deposition estimates tl
are used to adjust the predictions, and
the potential systematic differences t
tween the empirical dry deposition ei
mates from recent years and the peri
of interest.
Although RADM appears to repres<
an improvement over Lagrangian modi
relative to oversmoothing in areas imn
diately downwind of emission sour
regions, there is still some estimati
bias in these areas. It appears that I
pattern of over/underestimation is link
to the proximity of major source region:
The model predictions are greater tfi
the empirical estimates at sites local
within 81 to 160 km of major sour
regions. This suggests that the mo<
tends to oversmooth dry deposition g
dients near source regions. The path
of overprediction does not, howev
emerge at sites within 81 to 160 km
the major sources in Sudbury, Ontario.
fact, in this region the model performan
appears to parallel that in regions will
80 km of major source regions, w
underpredictions and a single modi
overprediction of 6%. One cause of tl
difference in model behavior might
related to the Sudbury stacks, which ;
much taller than those elsewhere in No
America. These tall stacks might depc
sulfur compounds farther downwind tr
typical sources, with much of the depc
tion occurring in the 81-160-km ran
rather than within 80 km. Consequently
would not be unreasonable to expect tl
the model's behavior within 81-160 km
-------
the Sudbury source region might be
similar to that observed in regions within
80 km of typical large source regions.
Oversmoothing of the spatial gradients
is also evident in the groups of sites near
less significant source regions. Within
160 km of source regions emitting
between 20 and 125 ktonnes of S02,
RADM estimates are considerably
greater than the empirical estimates. With
one exception, the relative error at these
sites is of the same order as that at sites
within 81 to 160 km of much larger
sources. In contrast, the model predic-
tions at the two sites removed from major
source regions (i.e., < 3 ktonnes/ yr with-
in 160 km of the site) are within 10% of
the empirical estimates.
In summary, a comparison of the re-
vised estimates to empirically-derived
amounts of sulfur dry deposition sug-
gests that there is a systematic error in
the revised estimates. Adjusted RADM
predictions of dry deposition tend to be
biased low in the most significant source
regions (where at least 200 ktonnes
S02/yr are emitted within 80 km of the
site). Conversely, in locations farther
removed from significant sources (81-160
km) there is evidence that the estimates
are biased high.
It is noteworthy that the adjusted model
predictions are generally within ± 60% of
the empirical estimates. Relative errors
are generally less than 40% (in absolute
value) at high deposition sites (where
empirical estimates exceed 9 kg S/ha/yr).
At sites with moderate deposition (where
empirical estimates are between 2 and 6
kg S/ha/yr), which are fairly close to
significant source regions (e.g., Penn
State, south-central Ontario, and north-
east of Toronto), estimation errors are
generally between 45% and 60%, with
one error approaching 100%. For other
sites with moderate deposition, errors
range ±40%. However, some of the
greater deviations (in absolute value) in
this group of sites might also .be
explained by the relationship of the sites
and their distance from the sources.
With the very limited available data, it
is impossible to separate and quantify the
three sources of error of the empirical
estimates (i.e., model bias, subgrid-scale
variability, and empirical estimation
errors). Therefore, the best that we can
do in characterizing the uncertainty in our
dry deposition estimates is to consider
the aggregate of all these errors, as
reflected in the distribution of RADM
deviations from empirical estimates.
These deviations suggest that the ad-
justed RADM predictions of dry depo-
sition generally are expected to lie within
±60% of the actual values. Although
there is some evidence that the magni-
tude and direction of the errors in model-
predicted dry deposition might be related
to distance from significant source
regions, we do not feel that the available
data allow us to further refine the ±60%
estimate of uncertainty.
Conclusions and
Recommendations
Three regional models, RADM,
RELMAP and ASTRAP, were applied to
construct grids of annual sulfur dry
deposition. Comparison of RELMAP pre-
dictions, adjusted RADM and ASTRAP
predictions with empirical data at 22 sites
indicated that RADM best replicated the
steep gradient downwind of a significant
emissions source region. Although each
model exhibited a tendency to smooth
the gradient, the degree of smoothing
appeared to be a function of the spatial
resolution of the model. Based on the
model comparisons with available empir-
ical data, the adjusted RADM predictions
appear to be the best estimates to date
of the spatial distribution of annual sulfur
dry deposition in the eastern United
States.
Since the RADM adjustment factor
here was 13% greater than that used in
the Interim Report, an identical system-
atic increase in the interpolations appear-
ing in the Interim Report is recom-
mended. The difference in adjustment
factors was a result of using as many as
five years of Canadian data and two to
three years of U.S. data, as opposed to
only one to two years of data that were
available at the time of the Interim
Report.
The comparison of the revised esti-
mates to empirically-derived amounts of
sulfur dry deposition suggests that there
is a systematic error in the revised esti-
mates. Although adjusted RADM pre-
dictions of dry deposition are generally
within ±60% of the empirical estimates,
they tend to be biased low in the most
significant source regions (where at least
200 ktonnes S02/yr are emitted within 80
km of the site). Conversely, in locations
farther removed from significant sources
(81-160 km) there is evidence that the
estimates are biased high.
Because of the anticipation of data
from additional sites and periods and
impending improvements to the algor-
ithms that calculate dry deposition
velocities, it is recommended that this
procedure be repeated at a later time.
Therefore, these estimates of the annual
sulfur dry deposition across the eastern
United States could be revised in the
future.
-------
The EPA authors, Terry L Clark and Robin L Dennis (also the EPA Project
Officers, see below), are on assignment to the Atmospheric Research and
Exposure Assessment Laboratory, Research Triangle Park, NC 27711 from the
National Oceanic and Atmospheric Administration; Steven K. Seilkop is with
Analytical Sciences, Incorporated, Research Triangle Park, NC 27713.
The complete report, entitled "Re-Examination of Interim Estimates of Annual Sulfur
Dry Deposition Across the Eastern United States," (Order No. PB 89-233 4641 AS;
Cost: $13.95, subject to change) will be available only from:
National Technical Information Service
5285 Port Royal Road
Springfield, VA22161
Telephone: 703-487-4650
The EPA Project Officers can be contacted at:
Atmospheric Research and Exposure Assessment Laboratory
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
United States Center for Environmental Research
Environmental Protection Information
Agency Cincinnati OH 45268
Official Business
Penalty for Private Use $300
EPA/600/S4-89/026
000085833
ls4!.
CHICAGO
------- |